CROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims the benefit of and priority to U.S. provisional patent application no. 63/397,537 filed on August 12, 2022, the entire content of which is herein incorporated by reference.

FIELD

[0002] This disclosure relates generally to building management or building automation. More specifically, it relates to systems and methods for control of a heating, ventilation and air conditioning system in a building using neural network.

BACKGROUND

[0003] Building management systems (BMS) or building automation systems (BAS) are systems used in buildings to manage heating, ventilation and air conditioning (HVAC), lighting, power, security, elevators and other building systems.

[0004] HVAC constitutes around anywhere from 30% to 60% (e.g., 40%) of a commercial building's total energy and it is now becoming a priority to achieve HVAC energy reduction, but energy efficient HVAC is not simple to achieve or sustain.

[0005] Traditionally, HVAC systems are considered as a group of independent mechanical equipment. Each pump, chiller, tower and air handling unit are designed to be turned on, run at a fixed speed and turned off. This way of thinking leads to a logic of equipment components that are designed to operate efficiently in isolation, and a BMS that controls the equipment by turning it on and off automatically. Operating data that typically resides in the typical BMS is not easily accessible by building operators. If operating data is available, it may be in the form of unformatted streams of data points, a format that is incompatible with performance measurement or problem diagnosis. [0006] Management of HVAC by BMS or BAS generally involves the use of thermostats or other sensors provided at various locations in the building, with each thermostat acting as a sensor measuring the temperature at its specific location. Typically, thermostats include a target temperature range, (i.e., a target temperature, accompanied by a value that is added to or subtracted from the target temperature and thus defining what is called the acceptable temperature band). This temperature range can be changed by room occupants or building operators. When the measured temperature in the room of the thermostat is out of the target temperature range, the thermostat sends an instruction to the HVAC equipment to start heating or cooling the room. This responsive management of room temperature may be inefficient and not cost effective. Improved efficiency and cost effectiveness can be obtained if occupancy of the room or other environmental variables which affect the thermal targets within the building are considered in the control of the HVAC system. In addition to improving efficiency and reducing costs, HVAC control that would be predictive as opposed to responsive would improve comfort and provide additional energy savings.

[0007] In order to link temperature control with room occupancy, many existing BMS or BAS require replacing existing thermostats with thermostats with occupancy or vacancy sensors or adding occupancy or vacancy sensors to specific rooms and linking these additional sensors to HVAC control to use this data. In large buildings, there can be hundreds of thermostats. The cost and effort of replacing hundreds of simple thermostats with thermostats with occupancy sensors can be significant and is a major hurdle for the adoption of this technology in large buildings.

[0008] Temperature predictions for HVAC systems are often based on physics-based Resistance-Capacitance (RC) circuit models (or simply RC models), which can be formulated as differential equations to simulate the thermal behavior of the HVAC systems. RC models are typically used with for temperature predictions in buildings with a small number of zones, typically less than ten zones.

[0009] As an alternative to RC models, temperature predictions for HVAC systems may also use black-box models such as neural networks. However, when using neural network to simulate behavior of large HVAC systems, a significant amount of computing resource may be required. For example, it may require a significant amount of input data, which may take days or months to gather, for training the neural networks in order to implement the neural networks for HVAC system control of large buildings. SUMMARY

[0010] In accordance with an aspect, there is provided a computer-implemented system for controlling a heating, ventilation, or air conditioning (HVAC) system for a building having a plurality of zones, the system may include: a processor; and a non-transient memory device storing: a temperature model for computing a rate of temperature change for at least one of zone of the building, the model including an explicitly-defined model portion and an approximated model portion; and a set of computer-readable instructions that when executed, causes the processor to: receive sensor data from a plurality of temperature sensors within the building; receive outdoor temperature data; compute the rate of temperature change for the at least one zone by providing the sensor data and the outdoor temperature data to the temperature model, wherein the approximated model portion is solved using a function approximator configured to generate an output reflecting a quantity of heat added or removed to the at least one zone; determine a predicted temperature based on the rate of temperature change; and in response to the predicted temperature, generate command signals for controlling at least one set point of the HVAC system.

[0011] In some embodiments, the function approximator includes a neural network trained to generate at least a portion of the output.

[0012] In some embodiments, the output from the approximated model portion further reflects a total amount of temperature change during a given period based on a current set point.

[0013] In some embodiments, the output from the approximated model portion further reflects an environmental thermal contribution.

[0014] In some embodiments, the environmental thermal contribution includes solar radiation.

[0015] In some embodiments, the rate of temperature change for the at least one zone i is represented by: wherein T _t is a respective temperature of the zone i obtained from the sensor data, T _env is the outdoor temperature data, _t is a time-scale associated with a thermal decay of zone i, ij are the time scales describing a thermal exchange between zone i and adjacent zone j, C _HVAC is related to a thermal capacity of zone i, and q _HVAC is the quantity of heat added or removed to the at least one zone i.

[0016] In some embodiments, C _HVAC q _HVAC represents the total amount of temperature change computed by the approximated model portion.

[0017] In some embodiments, the rate of temperature change for at least one zone i is represented by: where T _t is a respective temperature of the zone i obtained from the sensor data, T _env is the outdoor temperature data, r _t is a time-scale associated with a thermal decay of zone are the time scales describing a thermal exchange between zone i and adjacent zone j, and fi(U,SolRad, . . ) models an environmental thermal contribution in zone i.

[0018] In some embodiments, f^U.SolRad, .. ) = NN(9, U, Solrad, .. )[i], where NN represents the function approximator for the approximated model portion, U represents a vector of one or more HVAC controls of the HVAC system, SolRad is a given data reflecting solar radiation contribution, and 0 represents one or more parameters of the function approximator NN.

[0019] In some embodiments, the equation fi U,SolRad, . . ) is determined for each zone in the plurality of zones.

[0020] In accordance with another aspect, a computer-implemented method for controlling HVAC components of a building is provided, the method may include: storing or accessing a temperature model for computing a rate of temperature change for at least one of zone of the building, the model including an explicitly-defined model portion and an approximated model portion; receiving sensor data from a plurality of temperature sensors within the building; receiving outdoor temperature data; computing the rate of temperature change for the at least one zone by providing the sensor data and the outdoor temperature data to the temperature model, wherein the approximated model portion is solved using a function approximator configured to generate an output reflecting a quantity of heat added or removed to the at least one zone; determining a predicted temperature based on the rate of temperature change; and in response to the predicted temperature, generating command signals for controlling at least one set point of the HVAC system.

[0021] In some embodiments, the function approximator includes a neural network trained to generate at least a portion of the output.

[0022] In some embodiments, the output from the approximated model portion further reflects a total amount of temperature change during a given period based on a current set point.

[0023] In some embodiments, the output from the approximated model portion further reflects an environmental thermal contribution.

[0024] In some embodiments, the environmental thermal contribution includes solar radiation.

[0025] In some embodiments, the rate of temperature change for the at least one zone i is represented by: wherein T _t is a respective temperature of the zone i obtained from the sensor data, T _env is the outdoor temperature data, _t is a time-scale associated with a thermal decay of zone i, Ttj are the time scales describing a thermal exchange between zone i and adjacent zone j, C _HVAC is related to a thermal capacity of zone i, and q _HVAC is the quantity of heat added or removed to the at least one zone i.

[0026] In some embodiments, C _HVAC q _HVAC represents the total amount of temperature change computed by the approximated model portion.

[0027] In some embodiments, the rate of temperature change for the at least one zone i is represented by: wherein T _t is a respective temperature of the zone i obtained from the sensor data, T _env is the outdoor temperature data, _t is a time-scale associated with a thermal decay of zone i, Tij are the time scales describing a thermal exchange between zone i and adjacent zone j, and f^U.SolRad, .. ) models an environmental thermal contribution in zone i.

[0028] In some embodiments, ft(U,SolRad, .. ) = NN(0, U, Solrad, .. )[i], wherein:

NN represents the function approximator for the approximated model portion, U represents a vector of one or more HVAC controls of the HVAC system, SolRad is a given data reflecting solar radiation contribution, and 0 represents one or more parameters of the function approximator NN.

[0029] In some embodiments, the equation fi U,SolRad, . . ) is determined for each zone in the plurality of zones.

[0030] In accordance with yet another aspect, there is provided a non-transitory computer- readable medium having stored thereon machine interpretable instructions which, when executed by a processor, cause the processor to perform the computer-implemented method claimed in any one of the above claims.

[0031] Many further features and combinations thereof concerning embodiments described herein will appear to those skilled in the art following a reading of the instant disclosure.

BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In the Figures,

[0033] FIG. 1A is a schematic diagram illustrating an example system having a controller device for control of the HVAC components of a building, according to one embodiment.

[0034] FIG. 1 B is a schematic diagram illustrating another example system having a controller device for control of the HVAC components of a building, according to one embodiment.

[0035] FIG. 2 is a schematic diagram illustrating various parameters in a building that can be controlled by a controller device, according to one embodiment. [0036] FIG. 3 is a schematic diagram illustrating an exemplary architecture of a system including a controller device for control of the HVAC components of a building, according to one embodiment.

[0037] FIG. 4 is another schematic diagram illustrating an exemplary architecture of a system including a controller device for control of the HVAC components of a building, according to one embodiment.

[0038] FIG. 5 is a schematic diagram of an example architecture of a neural network maintained by a machine learning model of the controller device, in accordance with an embodiment.

[0039] FIG. 6 is an example process performed by a controller device for dynamic management of the HVAC components of a building, according to one embodiment.

[0040] FIG. 7 is a schematic diagram of computing device which may be used to implement a controller device.

[0041] These drawings depict exemplary embodiments for illustrative purposes, and variations, alternative configurations, alternative components and modifications may be made to these exemplary embodiments.

DETAILED DESCRIPTION

[0042] Efficient management and operation of the Heating, Ventilating and Air Conditioning (HVAC) system(s) in a building or network of building can reduce energy consumption leading to a reduction in building operating costs. Predictive or more finely tuned responsive control of the HVAC systems can not only improve efficiency but also improve environmental conditions, comfort and reduce management time.

[0043] The present disclosure provides systems and methods that use machine learning to predict a future temperature in a building complex having one or more zones, facilitating proactive control. A neural networks is a type of universal function approximator that can enable scalable solutions, which can be leveraged to help determine a predicted temperature. Based on the predicted temperature, one or more command signals can be generated for controlling at least one set point of the HVAC system in order to meet a target temperature.

[0044] In some embodiments, physics, such as thermodynamics, of an environment is simulated in the form of a differential equation with machine learning for HVAC temperature predictions. These embodiments may reduce the data and computing resource requirements for training neural networks: for example, traditionally it may take months to gather sufficient amount of training data for training machine learning models of large HVAC systems without any differential equation, while in some embodiments with differential equations used in conjunction with the neural networks, the data required may only take days or mere hours to collect. The use of differential equation with a machine learning model can also improve generalization of the model, enhance performance, and provide increased human interpretability. In addition, by embedding physically meaningful parameters, this approach can provide information about the thermal characteristics of buildings

[0045] Neural Ordinary Differential Equations (Neural ODEs) generally embed a neural network into the mathematical structure of an ordinary differential equation. A “Universal Ordinary Differential Equation” is a Neural ODE in which known parts of the equation can be made explicit while the unknown components of the Neural ODE can be modelled with a neural network.

[0046] Universal Ordinary Differential Equations (UODEs) have not been applied to the domain of HVAC temperature predictions. In some embodiments, the non-HVAC components of an RC model-based differential equation may be implemented explicitly, leaving the HVAC contribution to be modelled via a neural network. Further, thermal properties of buildings can be extracted, by means of features of a neural network, to inform commercial HVAC control strategies.

[0047] As mentioned, HVAC temperature predictions are typically exclusively physics-based or exclusively neural network-based, each of which holds its own advantages and disadvantages. The embodiments described herein integrate both physics-based differential equations and neural networks, in order to improve efficiency of HVAC control systems. For example, the flexibility and expressive capacity of neural networks can be used to model a wide range of systems and to learn directly from source data (e.g., temperature data). At the same time, the embodiments leverage prior physics knowledge by modeling said physics knowledge with differential equations (e.g., RC models), which reduces data and computing resource requirements for training a neural network. The result system can provide meaningful information about the thermal properties of the buildings.

[0048] Traditionally, RC models used to simulate a HVAC system typically require the spatial relationships between zones to be known in order to design the appropriate RC circuit analogue, which can be complex and time consuming for buildings of a large number of zones. With the embodiments described herein, the system does not make any such assumptions and allows the learned parameters of the neural networks to indicate possible neighbors of a given zone.

[0049] In some embodiments, an example system disclosed herein may include a control module configured to extract and monitor all data items, i.e. "points" and considers the thermodynamic properties of both the building as a whole and zones therein to predict the value of all data items, i.e. "points" and control each component of the HVAC system to optimize HVAC performance. Components and operating parameters of the HVAC system controlled by the example system may include set points, fan speed, valve openings, damper position, temperature of the cold water, temperature of the hot water, pipe pressure, differential pressure, pump speed, evaporator rate, and compressor pressure, amongst others.

[0050] In some embodiments, an example system disclosed herein may be configured to send instructions to the BMS/BAS to adjust operating parameters of one or more or all HVAC components and operating parameters or to send instructions directly to one or more specific HVAC equipment.

[0051] In some embodiments, an example system disclosed herein may provide and enable predictive HVAC control. The example system may maintain or update a control module which can utilize advanced data processing and/or artificial intelligence including traditional linear, nonlinear regression models, supervised learning, unsupervised learning, and deep learning techniques to provide predictive HVAC control systems and modules that are capable of learning. In some embodiments the example system may integrate and interconnect devices including sensors, and HVAC components, within existing building infrastructure using wired or wireless networks. Some of the features described herein may utilize big data systems, machine learning and artificial intelligence, cloud computing technologies, and cloud services, for example.

[0052] Predictive control of the HVAC system can include control of one or more HVAC systems based on forecasted future outdoor environmental conditions and historic internal features or behaviors, which may include but not limited to: temperatures, wind speed and direction, cloud cover percent, eclipses, outdoor humidity, time of day, date, thermodynamic patterns derived from historic dataset and predicted occupancy amongst other factors.

[0053] In some embodiments, an example system disclosed herein may receive forecasts or other information from third party providers including government weather stations. Meteorological forecasts include short range, medium range and long range meteorological forecasts. The weather conditions and forecasts may be updated at regular intervals. In some embodiments, weather information and tracking are updated at fixed intervals, for example, every 1 , 5, 10, 15, 20, 30 or 60 minutes. Optionally, in some embodiments, the system may be configured to use historical data to predict impact of specific weather conditions on HVAC needs for specific parts of the building as zones or rooms with exterior walls or windows may be more impacted by outside conditions than interior zones or rooms.

[0054] In embodiments, when an example system receives meteorological forecasts, one or more HVAC component operations may be adjusted based on these forecasts. For example, in embodiments where the HVAC system comprises multiple HVAC units, individual units may be turned on or off or adjusted based on anticipated need, optionally allowing for the selection of heating or cooling units based on anticipated conditions and allowing for the operation of multiple units at part-load, if such operation is more efficient than operation of a single unit at full load. In some embodiments, where multiple chillers serve the building the systems and methods may provide for chiller staging. Optionally, chiller staging considers equipment configuration and type, refrigerating capacity, chilled water flow rates, power consumption by water condensers and water tower fans.

[0055] In some embodiments, buildings may be pre-cooled or heated in anticipation of changes in temperature thereby optionally allowing for off-peak electricity consumption.

[0056] In some embodiments, an example system may be configured to control ventilation or air flow to redistribute heat or cool air, for example, to regulate air flow to move cooler air from basement levels to upper levels. In some embodiments, the system may be configured to regulate intake of outdoor air and optionally to use outdoor air for space cooling, for example, to flush the building with cool outdoor air at night to reduce or avoid cooling during the daytime. The system may be further configured to redistribute heat or cool air only if the redistribution is cost effective or otherwise advantageous. [0057] In some embodiments, an example system may be configured to control HVAC components based on room occupancy including controlling heating, cooling and ventilation based on actual or predicted occupancy. Optionally, zones or rooms may be pre-emptively cooled or heated based on predicted occupancy rates. In some embodiments, usage of supply air fans, hood exhaust and make up fans is dependent on actual or predicted occupancy where usage is reduced in unoccupied rooms or zones. The system may be optionally figured to automatically switch on ventilation system or electrical components (e.g., lights) if occupants are detected in the room or zone. Optionally, in addition to occupancy the type of occupancy is considered, for example, occupants involved in strenuous activity will impact environmental conditions to a greater extended then occupants who are not active.

[0058] In embodiments where room occupancy is predicted, HVAC settings can be adjusted to account for number of anticipated occupants to compensate or adjust for heat or CO2 generated by the occupants of the space. Optionally, the system may control ventilation of a specific space on the basis of controlling for either actual or anticipated CO2 levels. In some embodiments, outdoor ventilation rate is minimized to the rate necessary to maintain acceptable air quality, i.e., CO2 levels. In environments with extreme weather conditions, indoor air is optionally filtered to minimize amount of outdoor ventilation required.

[0059] In some embodiments, an example system may be configured to manage building energy consumption including managing electric consumption during peak demand time thereby reducing costs.

[0060] In some embodiments, an example system may be configured to manage HVAC components in response to outside air quality and/or air quality within a specific area or zone. For example, the system may be configured to control ventilation rate based on measured pollutant level including level of carbon monoxide and/or nitrous oxides in a particular zone or area such as a car park or loading dock. Optionally, the system may be further configured to direct flow of outside or indoor air through filters to improve air quality, when necessary.

[0061] Optionally, the example system may be configured to adjust air flow or air intake in response to outside air quality. For example, in some embodiments, outside air is filtered or purified if outside air quality is low such that airborne particles, likes dust, pollen and bacteria are removed. Accordingly, in some embodiments, air intakes are selected based on actual or anticipated air quality. [0062] In some embodiments, where multiple factors are considered in predicting HVAC requirements, individual factors can be weighted. Weighting of factors may be based on historical data or anticipated impact.

[0063] In some embodiments, an example system may be configured to allow building operators to weight or prioritize one or more factors, for example, maximize energy efficiency or cost savings. In such embodiments, a user interface may be provided allowing for selection or ranking of priorities.

[0064] In some embodiments, an example system may be configured to control HVAC components to minimize energy expenditure while maintaining a minimum comfort in occupied zones and/or areas by taking into account factors that impact perceived comfort level of an occupant, including temperature and humidity. For example, set point for temperature and humidity are both adjusted to maintain comfort while reducing energy requirement.

[0065] In some embodiments, an example system may be configured for a single building or a network of buildings. In some embodiments, the two or more buildings in the network of buildings are physically connected, for example, by tunnels, enclosed walkways, bridges. In such cases, set points for temperate and humidity in a first building may affect set points for temperate and humidity in a connected, second building, and the example system may be configured to operate the HVAC components of the buildings as a whole.

[0066] In some embodiments, an example system may be configured for retrofit into existing building and be further configured to interface with existing HVAC components. In other embodiments, the system may be incorporated into new builds. In new builds, the system and methods are optionally integrated into or are a module of the BMS or BAS.

[0067] In some embodiments, an example system may be configured to interact directly with components of the HVAC system, for example in direct communication with the sensors and the actuators/controllers of the HVAC system. In alternative embodiments, the system may communicate directly with the existing master controller. Optionally, the system may be a combination of direct interact interaction with the master controller and direct communication with components of the HVAC system.

[0068] In some embodiments, an example system may generally include a controller device in communication with sensors and components of the HVAC system. The controller device may be further connected to a remote server or cloud server. In some embodiments, both the controller device and remote server or cloud server include algorithm modules configured to analyze data.

[0069] Analysis of sensor data and/or calculation of HVAC parameter changes may be completed on the on-site controller device, on the remote server or cloud server or combination thereof. In some embodiments, the system is configured such that processing is switched to the on-site controller device if there is no link to the remote server or cloud server.

[0070] In the depicted embodiments, one or more neural networks are used for function approximation of certain portions of an ordinary differential equation. In other embodiments, other types of function approximators may be used in lieu of or in combination with the one or more neural networks. In some embodiments, such function approximators may include regressionbased function approximators, polynomial-based function approximators, or the like. Generally speaking, function approximators can include a class of generic functions can be parametrized to represent different input-output mappings.

[0071] In one embodiment, referring to FIG. 1A, there is illustrated a system having a controller device 100 in communication with HVAC components 400 of the building and with a remote server 300. More specifically, there is described a system for managing HVAC components 400 of a building, the system including: a processor; a memory storage device storing a set of instructions and a temperature model for computing a rate of temperature change for at least one of zone of the building, the model including an explicitly-defined model portion and an approximated model portion, the set of instructions when executed by the processor, cause the processor to: receive sensor data from a plurality of temperature sensors within the building; receive outdoor temperature data; compute the rate of temperature change for the at least one zone by providing the sensor data and the outdoor temperature data to the temperature model, wherein the approximated model portion is solved using a neural network trained to generate an output reflecting a quantity of heat added or removed to the at least one zone; determine a predicted temperature based on the rate of temperature change; and in response to the predicted temperature, generate command signals for controlling at least one set point of one or more components 400 of the HVAC system.

[0072] In some embodiments, the plurality of forecasts may include at least one of: a predicted temperature value, a predicted water usage amount, a predicted electricity usage amount, a predicted gas usage amount, a predicted weather, and a predicted humidity level. Some forecasts, such as the predicted temperature value, predicted electricity usage or the predicted humidity level, may be determined based in part on a weather forecast from a remote server 300 or third party servers 600.

[0073] A forecast predicting a dynamic state of the building may be obtained from one or more control modules, or from the BAS/BMS if there is one. The dynamic state may include, for example, a predicted occupancy level in one more zones of the building.

[0074] In some embodiments, the plurality of current states of the building may include at least one of: a number of zones, a temperature measurement, a set point, sensor data, actuator data, occupancy schedule, and occupancy data. In some embodiments, the plurality of current states of the building is received from a Building Management System (BMS) 150.

[0075] In some embodiments, the system further has a display device for displaying the one or more operating values for the HVAC components. The display device may be, for example, display device 420 shown in FIG. 4.

[0076] In an alternative embodiment, referring to FIG. 1B, an example system may include a controller device 100 in communication with the existing BMS or BAS 150 and with a remote server 300. Forecasts can be provided directly from third party servers 600 to the controller device 100 or via the remote server 300 as shown. More specifically, there is a system for managing HVAC components 400 of a building, the system including a controller device 100 in communication with the pre-existing BMS or BAS 150. The BMS or BAS 150 communicates with sensors 200 in the building and with the HVAC components 400 and acts as an intermediary between the controller device 100 and sensors 200 and the HVAC components 400 such that the controller device 100 provides control instructions to the BMS or BAS 150 which then communicates with the necessary HVAC components 400. Similarly, sensors 200 communicate with the BMS or BAS 150 which then transmits the data to controller device 100.

[0077] In a further example embodiment, the system comprising a controller device 100 in communication with the existing BMS or BAS 150 and one or more sensors 200 or one or more HVAC components 400.

[0078] Embodiments of the system do not require changing thermostats of the building, allowing retrofitting of the system using existing thermostats. A controller device 100 can be advantageously provided and made to be in communication with the existing BMS/BAS 150 the sensors 200, the actuators, and/or the individual controllers, and the internet, for low-cost and simple retrofitting of the existing thermostats and HVAC equipment in the building. Actuators in HVAC systems control the dampers. By utilizing a low voltage signal, the actuator places the damper at any point between fully open and fully closed.

[0079] The system, as illustrated in FIGs. 1A and 1 B, can be an loT (Internet of Things) solution to individually control HVAC components 400 of a building management system (BMS) 150 in order to optimize the efficiency of the whole system. The system can advantageously make use of existing sensors 200 (e.g., thermostats which comprise a thermometer) throughout the building, and of actuators or controllers on the various HVAC components 400 to individually control each one of them, at an individual component level. A controller device 100 is in communication both with these sensors 200, the actuators, and with the internet (i.e., a remote server 300). By converting data formats, the controller device 100 is adapted to receive information from the sensors 200 and transmit to the actuators even though the format of the data is heterogeneous.

[0080] In some embodiments, the controller device 100 may execute one or more control modules that implement one or more algorithms to predict a building’s needs in terms of heating, cooling and ventilation throughout the building. These predictions may be based on data acquired from the sensors 200 (e.g., sensors to detect air temperature, velocity, humidity and pressure, human presence) and data acquired from the internet (e.g., local weather forecast, including sunlight intensity and orientation, and temperature, wind or precipitation). By running analytics on the historic dataset of the building into the program, the controller device 100 can determine how to control each of the HVAC components 400 to optimize the system (i.e., minimize cost or energy consumption) based on one or more temperature settings.

[0081] In some embodiments, the controller device 100 can determine a predicted temperature of one or more zones in a building based on a change of temperature neural network model stored in a data storage device, and based on real time temperature data from one or more sensors 200, as further elaborated below.

[0082] Conventionally, HVAC control systems relying on strictly or predominantly neural network systems tend to require a large amount of training data, and have needed to be implemented in a centralized (e.g., rooftop) HVAC system, or implemented remotely, due to the data and training requirements. In contrast, in some embodiments, the controller device 100 is implemented entirely in a thermostat or loT device, which may be a wall-mounted device. This is made possible because of the reduced computing resources required as a result of utilizing universal ODE for predicting a temperature in a given room or building, which can use function approximators (e.g., neural networks) to reduce the amount of training data required for simulating environmental thermal factors and training the control modules to control the HVAC system. This provides more direct access to sensor data across the building, better data security, and faster training and deployment process for the control modules that can be implemented on site, and trained on site of a building 170.

[0083] Electrical equipment 500 can also be controlled in a similar fashion by the controller device 100. This applies specially to lighting, the needs of which depend strongly on occupancy, which is measured, and which affects heating or AC needs. This can also apply to other types of equipment having similar requirements and effects, such as industrial equipment, local servers, electrical loads which depend on occupancy (e.g., ornamental equipment), electricity distribution equipment, electricity storage (i.e. , battery banks and the like), alarms systems, and so on.

[0084] The control modules as well as temperature models 110 can be stored and executed on either or both of the controller device 100 and on remote servers 300 (e.g., the cloud) depending on the needs in terms of bandwidth and time reactivity.

[0085] Based on the analysis and determination from the controller device 100, each HVAC component 400 is instructed, through the controller device 100, to modulate its operation. The components can include, for example (without limitation): air system components (fresh air intake; duct, exhaust and supply fans, night purge, (de-humidification), cooling system components (chillers, compressors, condenser water delivery, head pressure, condensing pressure) and heating system components (boiler, steam pressure or delivery). A very large number of components can be managed and controlled by the controller device 100 in real time or near real time based on a variety of input including current states of the building, different forecasts, and different user objectives.

[0086] In some embodiments, the controller device 100 may connect to an electrical or thermal storage device, such as, for example, battery systems, electric vehicles, or fuel cell systems. The controller device 100 may further connect to a potential energy storage system such as a flywheel, compressed air tank, water storage tank, underground thermal, one or more storage mediums, phase-change materials, and so on. [0087] In some embodiments, the controller device 100 may connect to one or more district heating provider (e.g., hot water or steam provider) or a cooling provider (e.g., cold water provider).

[0088] Today, large buildings make heavy use of Building Management Systems (BMS) thus making equipment operating conditions more visible to the operators, and also more easily controllable. This visibility focuses all the attention on the monitoring tools and is not concerned by thermodynamic equilibrium within the building. Indeed, the focus of prior art BMS is made on the optimization of individual equipment performance while at the same time discarding more than 50% of a building's thermal energy into the environment. Instead of managing the operating conditions of individual equipment, as in typical BMS, the disclosed system and controller device 100 are configured to manage the energy flow in the building by collecting real time data from sensors 200 located throughout a building and by managing the heat flow as a whole.

[0089] The system may be configured to reduce the inefficiency of the thermal energy in buildings. A controller device 100 can be used not only to collect data from sensors 200, but also to connect with (through a BMS 150 in some cases) every chiller (or any other cooling system component), boiler (or any other heating system component), pump and an entire array of the control points of a building. With this infrastructure capability, advanced algorithms or modules can be deployed by being installed on an easy-to-deploy controller device 100 to deliver dynamic states of the building base on real time environmental conditions and the internal load of the building, thus keeping all the mechanical systems optimally positioned 24/7. This can be made by connecting the controller device 100 with existing equipment (e.g., sensors 200, controllers, actuators of HVAC components 400), sometimes through the BMS 150. By putting the intelligence in the controller device 100 and not on the equipment (sensors, controllers, actuators), the existing equipment do not need to be replaced and retrofitting is made possible. The deployment of the system would thereby eliminate drifting and reduce occupant complaints, equipment alarms and periodic commissioning (physical inspections and maintenance). HVAC systems that are producing and distributing the precise quantity of thermal energy needed in real time to provide occupant comfort will attain a thermal balance performance level.

[0090] The potential energy savings unlocked by the disclosed embodiments, when retrofitted to existing HVAC equipment within a building, goes well above 30% of the HVAC energy consumption. In some situations, by combining advanced control modules to match the weather pattern with the thermal load requirement of the building in real time or near real time, it becomes possible to deliver a lower kW/ton ratio.

[0091] In some embodiments, the controller device 100 may implement relational-control algorithms to optimize all the equipment within an all-variable flow HVAC system (chillers, fans, pumps, etc.). By doing so, the controller device 100 may use the least amount of power required to maintain occupant comfort levels. Control set points are automatically calculated based on real time building load information inputs and the weather conditions prevailing outside of the building. This approach results in a global thermal load management for the building instead of an equipment-based management strategy.

[0092] The controller device 100 may implement relational-control algorithms to monitor the different parameters of the entire fleet of equipment and sensors 200 within a building to deliver continuous, automatic adjustments to the system based on the building load, regardless of facility type. In some embodiments, an example objective of these systems and methods is to generate energy usage savings (kWh/yr.), demand savings (kW), heating/cooling load (Therms), cooling tower water usage savings (gal/yr.), carbon footprint reduction (lbs. /yr.) and Power Usage Effectiveness (PUE) Reduction. By doing so, these systems and methods save impressive amount of dollars, reduce the load on the grid and improve tenant comfort.

[0093] This global view of a building, as monitored via the controller device 100, can provide real time system adjustment recommendations, identify operational inefficiencies, and provide real time or near real time building re-commissioning at any time of the day. In some embodiments, this controller device 100 can manage energy demand by analyzing building occupancy, the building's thermodynamic patterns and outside weather to automatically adjust the heating and cooling of the building's different zones to the optimal levels in real time.

[0094] According to an embodiment, the system can include a remote server 300 connected to the controller device 100. The remote server 300 can be a single server, a plurality of servers wherein each one is dedicated to certain tasks, or a plurality of servers organized in a network to perform tasks in a distributed manner, e.g., the cloud.

[0095] According to an embodiment, referring to FIG. 3, a controller device 100 locally collects and aggregates all the data points from various sensors 200 and sub-systems within the building including, but not limited to, the following data points: fans, electric meters, elevators, occupancy counters, steam meters, internal or external air temperatures, water, and so on. To collect these data points, the controller device 100 can connect to an existing BMS 105 system over a variety of protocol (e.g., BACnet, Modbus, LonWorks, etc.). This controller device 100 is compatible with all the main BMS in the market (e.g., JOI, Schneider, Honeywell, Siemens, Automated Logic, Panasonic, Legrand, Delta, IBM, Hitachi). An exemplary cloud solution server 300 network is connected to and collects data from the controller device 100 by a data link 315 and one or more other data sources including weather stations 310. The cloud solution server 300 includes one or more databases that can store data received from the controller device 100 and the one or more other data sources. Data from edge building may be stored in a separate database. The controller device 100 is optionally configured to update information in the database at set intervals or when a change data is observed.

[0096] According to another embodiment, the cloud solution server 300 network collects data from multiple buildings, each one having their own controller device 100, 100, in communication with their respective electrical equipment, such as lighting, and with their respective HVAC controllers.

[0097] In some embodiments, as shown in FIG. 1B, the controller device 100 communicates with the HVAC controller or BMS/BAS 150 which, in turn, communicates with the components under its control. In an embodiment, the HVAC components 400 may be controlled by a HVAC controller which communicates directly with the controller device 100. In another embodiment, the HVAC controller, which communicates directly with the controller device 100, controls both the HVAC components 400 and the sensors 200, in order to obtain all the measured values from the sensors and send them to the controller device 100, and to communicate with the HVAC components after having received instructions from the controller device 100.

[0098] According to an embodiment, the data collected by device 100 from the variety of different sensors 200 located throughout the building is translated in a universal open format from the different protocols from which it is collecting the data, and sends the data to a database on a remote server 300, e.g., a cloud-computing database distributed on remote servers communicating in a network. This communication is preferably made over a wireless connection, and further preferably over an encrypted connection since private information can be collected by the sensors 200.

[0099] According to an embodiment, the database, preferably on a remote server 300 (e.g., on the cloud), aggregates all the data in an historic dataset and keeps a fine granularity of the historic time line for each data point. This database become overtime a big data picture of the thermodynamic behavior of the building and is used to extract additional value (e.g., trends) from the data set. Specific algorithms use the dataset to build a real time thermodynamic behavior models of the building including a prediction of the optimal settings of the different HVAC devices in real time.

[00100] A collection of control modules or algorithms can be executed by the controller device 100, in different combinations and orders, to leverage the thermodynamic model derived from the dataset with real time value of the data point, the number of persons in each zone and the outside weather parameters condition in real time and the forecast of the next few hours. Based on the results of execution of these control modules, one or more control commands may be generated and sent to the different controllers in the building, which can dictate the modulation of all the HVAC devices to maintain the desire temperature and humidity level in the building at all time. These commands may include optimal settings for the next time interval (example: 5 minutes), once the time interval has elapse, a new round of execution of control modules may be initiated by the controller device 100 to produce a new series of orders. This process may continue and self-adjust base on the behavior of the HVAC equipment.

[00101] According to an embodiment, the control modules may be distributed over the controller device 100 and the remote server 300 and 310. To optimize the response time in reason of network communication delays, some of the control modules may be executed on the controller device 100 and some may be executed in the cloud.

[00102] For example, control modules executed on device 100 may include those involving real time responses such as a sudden change of occupancy in a zone, and control modules executed on the remote server 300 involve those not related to real time responses, such as detecting trends and correlations between events and power demand, or applying weather forecasts to make power demand forecast. In other words, and in some embodiments, real time operations are preferably performed locally and analytics is preferably performed remotely. Privacy of data can also be taken into account when deciding if data is communicated over the internet to a remote server 300.

[00103] In one embodiment of the invention, control modules relating to dynamic thermal equilibrium process can be divided in different subgroups related to their functions, for example, control modules can be broadly divided into modules that specifically control air components of an HVAC, modules that specifically control water side components, modules that increase energy efficiency and/or reduce energy costs, and modules that optimize control of the HVAC system. For example, in some embodiments, subgroups of modules may specifically instruct particular HVAC components with respect to one or more of:

• the air systems managing the ventilation, modulation of the air temperature and humidity of the air flowing in the building;

• the cooling systems producing and distributing the cooling thermal load in the building, serving in most cases the different air systems;

• the heating systems producing and distributing the heating thermal load in the building, serving in most cases the different air systems; and

• the control systems.

[00104] Each of these subgroups has its own set of algorithms modeling the thermal flow of the building and calculating the optimal running configuration of the different HVAC device. The following is an example list of modules or algorithms per subgroup.

[00105] Regarding air systems, the controller device 100 can configure and execute control modules to perform the following operations (without limitation):

• Modulate the fresh air intake based on the number of persons in each zone of the building.

• Modulate the fresh air intake based on the outside temperate level.

• Modulate the air flow speed based on the number of persons in each zone of the building.

• Modulate the pressurization of each zone.

• Modulate the humidity control required.

• Modulate the duct static pressure based on the fan speed and the VAV modulation.

Modulate the supply air fans, hood exhaust and make-up fans based on an occupancy level of each zone. Modulate the economy cycle (use outdoor air for space cooling) based on the exterior parameters and inside load.

• Modulate the night purge based on weather parameter and forecast demand (flushing the building with cool outdoor air at night to avoid mechanical cooling at start-up).

• Modulate the computer room air speed based on a cooling load (use air movement to remove heat load instead of supplying cold air to the room).

• Modulate the humidification and de-humidification based on the load.

[00106] Regarding cooling systems, the controller device 100 can configure and execute control modules to performing the following operations (without limitation):

• Modulate the staging of chillers and compressors based on the load balance along the optimal curve of the chillers.

• Modulate the chilled water pumps speed base on the number of persons in the building.

• Modulate the water evaporator pump of the water tower base on the number of persons in the building.

• Modulate the chilled water temperature and speed flow base on the number of persons, outside temperature and wet bulb.

• Modulate the condenser water delivery temperature to maintain the optimal coolest possible CW for cooling of refrigeration equipment.

• Modulate the head pressure control (for air cooled condensers & water tower only) to maintain the optimal condenser fans energy consumption.

• Modulate the condensing pressure based on the heat load to be extracted (for water- cooled condensers).

• Modulate the staging of chillers and compressors based on the load balance along the optimal curve of the chillers.

Modulate the chilled water pumps speed base on the number of persons in the building. Modulate the water evaporator pump of the water tower base on the number of persons in the building.

• Modulate the chilled water temperature and speed flow base on the number of persons, outside temperature and wet bulb.

• Modulate the condenser water delivery temperature to maintain the optimal coolest possible CW for cooling of refrigeration equipment.

• Modulate the head pressure control (for air cooled condensers & water tower only) to maintain the optimal condenser fans energy consumption.

• Modulate the condensing pressure based on the heat load to be extracted (for water- cooled condensers).

[00107] Regarding heating systems, the controller device 100 can configure and execute control modules to perform the following operations (without limitation):

• Modulate the hot water I steam delivery temperature based on the heating demand load.

• Modulate the boiler sequencing based on the load demand.

• Modulate losses in de-energized boilers.

• Modulate the steam pressure based on the load demand.

[00108] Regarding controls systems, the controller device 100 can configure and execute control modules to perform the following operations (without limitation):

• Modulate space temperature set points and control bands based on the optimal range trend and deadband (albeit drifting).

[00109] Referring now to FIG. 2, the controller device 100, by executing various control modules, can generate control commands to operate the following HVAC components 400 and parameters in the building 170, without requiring physical modification of the building 170:

A. Cooling production optimization 105

Variable head pressure control (air cooled condensers & water tower). ■ Adjust condensing pressure base on heat load to be extracted.

■ Variable head pressure control (water-cooled condensers).

■ Adjust head pressure of the condenser water base on heat load to be extracted.

■ Optimized secondary chilled water pumping.

■ Balance the quantity of chilled water circulated base on the Al-Ill needs (reduce circulation when low request).

■ Humidity control based on the outside humidity level.

■ Predictive cooling production based on the weather forecast (6 hours).

■ Balance the chiller temperature base on the evaporator target.

B. Distribution optimization 120

■ Reset chilled water delivery temperature.

■ Set the optimal warmest possible water for cooling.

■ Reset ting condenser water delivery temperature.

■ Set the optimal coolest possible CW for cooling of refrigeration equipment.

■ Reset ting heating hot water de livery temperature.

■ Set the optimal coolest possible water for heating.

■ Heating water Delta T modulated with Pressure Delta P.

■ Staging of chillers and compressors.

■ Load balance the chillers at the optimal efficiency curve.

C. Ventilation and air flow optimization 130

■ Supply air fans, hood exhaust and make-up fans.

■ Reduce usage when not needed (occupancy driven).

Use outdoor air for space cooling when possible. ■ Night purge.

■ Demand control ventilation base on controlling CO2 for occupied

■ space.

■ Pre-emptive cooling or heating based on a sun position.

■ Pre-emptive cooling or heating based on the number of persons in a zone.

■ Demand control ventilation base on controlling CO levels.

■ Duct static pressure rese.t

■ Reduce fan speed in proportion of VAV modulation.

■ Use air movement to remove heat load instead of supplying cold air to the room.

D. System control optimization 140

■ Occupancy control.

■ Automatic switching of ventilation system and lights if the presence of occupants in the area is detected.

■ Optimum start/stop heating/cooling for each zone.

■ Space temperature set points and control bands.

■ Set the optimal range trend and deadband.

■ Master air handling unit supply air temperature signal.

■ Modulate master air feed temperature to avoid simultaneous heating and

■ Cooling.

[00110] Examples of sensors 200 provided in a building 170 include, without limitation, temperature sensors (often in a thermostat), occupancy sensors, humidity sensors, pressure sensors, and sensors found within the HVAC system such as air speed sensors.

[00111] Other data not measured by the sensors 200 can be collected and used. For example, online calendars may be queried to detect upcoming events in particular locations in the building. In another example, the remote server 300 can query third party servers 600, as shown in FIGs. 1A and 1B, to collect weather forecasts for the location of the building, and take into account sunlight, temperature, wind, humidity, pressure and precipitation of the environment around the building to feed to the thermodynamic model of the building being computed by the remote server 300. For example, a given zone in the building may receive more sunlight than another and the remote server 300 may determine that an optimal course of action would be to trigger ventilation to have the air flow from one zone to another zone to warm the other zone without having to resort to the heating system and cool down the sunlit zone without needing the air conditioning system.

[00112] FIG. 4 is schematic diagram illustrating an exemplary architecture of a system 480 including a controller device 100 for dynamic control of the HVAC components 400 of a building 170, according to at least one embodiment. In some embodiments, the controller device 100 is configured to access and use a stored temperature model 110 to determine a predicted temperature for a given zone of a building 170.

[00113] In some embodiments, a rate of change of temperature may be generated by the controller device 100 based on a Universal ODE, where some elements of the Universal ODE are modelled using a physics-based equation or representation, and other elements are modelled using a function approximator such as a neural network. In some cases, the decision to use a physics-based equation or a function approximator to represent one or more element affecting the rate of change of temperature may be based on the amount of training data required to train the function approximator to properly model said element. As an example, it can be possible that a physics-based representation (e.g., — — 7})) is used to represent the heat exchange

^T '-i between adjacent zones i and j, but after collecting a few months of data, a function approximator such as a neural network may be trained to better represent the heat exchange between adjacent zones i and j, so the physics-based representation may be replaced with NN (T;, 7}).

[00114] In some embodiments, a rate of change of temperature may be generated by the controller device 100 based on a Universal ODE, in which non-HVAC components are defined explicitly based on an RC model-based differential equation, and the HVAC contribution from the HVAC components 400 are modelled via a neural network. Depicted in FIG. 5, a neural network 500 may have an architecture 500 including an input layer, a plurality of hidden layers, and an output layer. The input layer receives input features. The hidden layers map the input layer to the output layer. The output layer provides the prediction of the neural network. In other embodiments, other suitable deep learning models and architectures can be used. [00115] For example, temperature prediction may be based on a temperature model 110 stored in a database connected to controller device 100. Alternatively, the temperature model 110 may be stored on database server 450, or as part of BMS 150. The temperature model 110 may, in some embodiments, take the form of an ODE that models a rate of change of temperature in a given zone i of a building have a plurality of zones.

[00116] In some embodiments, a modular software framework may be implemented by the system for controlling the controller device 110. For example, different portions of a temperature model 110 (which may be a universal ODE) can be implemented as discrete software modules, which may be swapped dynamically before or during operation. For example, after a certain amount of time when a discrete differential equation from the model 110 used to simulate a certain physics property (e.g., solar radiation) of the environment has accumulated sufficient amount of raw data, or reached a certain threshold, the discrete differential equation used to simulate the physics property may be swapped with a corresponding function approximator element, which may be a neural network, that can be trained rather quickly based on the accumulated raw data, to better predict the physics property at a future point in time.

[00117] The temperature model 110 may include an explicitly-defined model portion and an approximated model portion, with the approximated model portion modelled based on a neural network trained to generate an output reflecting a quantity of heat added or removed to the at least one zone.

[00118] In some embodiments, the approximated model portion may be modelled using one or more function approximators, which may include a class of generic functions parametrized to represent different input-output mappings. A function approximators may be a neural network 500 shown in FIG. 5.

[00119] In some embodiments, the output from the approximated model portion can also include a value reflecting a total amount of temperature change during a given period based on a current set point.

[00120] In some embodiments, the output from the approximated model portion can include a value reflects a solar radiation.

[00121] For example, temperature model 110 may include an equation to determine a rate of temperature change for at least one zone v. ₌ _ _(r . _ _Tmv) _ _Z . _ (T. _ _r ) _{+ Chvac} q _HVAC

14.1. t j t j [1 ] where T _t is the temperature of the I ^th zone , T _env is the outdoor air temperature, r _t is the timescale associated with the thermal decay of zone i, are the time scales describing the thermal exchange between zones i and j, C _HVAC is related to the thermal capacity of the zone, and q _HVAC is the heat injected or extracted by the HVAC system. Thermal capacity may be defined as the quantity of heat necessary to produce a unit change of temperature in the respective zone i.

[00122] CHVAC HVAC in equation [1] may be used to represent the total amount of temperature change computed by the approximated model portion, which can be applied to every zone in the building, in some cases with slight modification. The HVAC system heat contribution can be modelled in different ways depending on the details of the HVAC system and the assumptions made in the modelling. However, complex HVAC systems, solar radiation, internal gains, and other difficult-to-model terms of the equation make explicit modelling cumbersome and timeconsuming, particularly for larger buildings with hierarchical HVAC systems. HVAC systems can also be difficult to model if they are controlled by set points because the underlying control logic that maps the set points to the equipment controls varies significantly and is particular to the HVAC system under consideration. This is where it is useful to leverage the immense flexibility of a neural network, to learn the HVAC and possibly additional terms directly from data.

[00123] In another embodiment, the term C _HVAC q _HVAC can be replaced with a function approximator or neural network NN, where the rate of temperature change is determined based on: where f^U.SolRad, . . ) = NN(0, U, Solrad, . . )[j] is a function that models the unknown or hard to model terms in the differential equation for zone i.

[00124] Equation [2] is parametrized by the neural network NN(0, U, Solrad, . . ). U represents a vector of HVAC controls, SolRad is solar radiation data, 0 are the parameters of the function approximator (e.g., neural network) and other inputs to the function approximator can be included as well. fi U, SolRad, . . ) models an environmental thermal contribution in zone i. [00125] In some embodiments, U may represent a vector including one or more HVAC controls including values reflecting, without limitation, a temperature setpoint, ON/OFF switch, a state of one or more HVAC components such as a fan, a heating coil, a cooling coil, valve positions, and/or damper positions.

[00126] Solar radiation data SolRad may be obtained from a pyranometers or an external database, which may be a third party database, such as climacell, darksky.

[00127] Other inputs to the neural network may include salient feature such as: temperature schedule, solar positions, extra measurements of solar radiation data. In addition, additional inputs to the neural network may include occupancy data which can be from CO2 sensors, head counts from an access system of the building 170 or Wi-Fi system.

[00128] In some embodiments, the model portion approximated by a neural network may include other sources of environmental thermal contribution instead of or in combination with solar radiation.

[00129] For each zone in the building 170 having a plurality of zones, there may be such an equation for each zone i, i = [1, 2, ...n], and all the equations can be solved together as a system of equations. The index in square brackets on the right-hand side [j] indicates that the equations share a single neural network with multiple (e.g., n) outputs, with each output representing a rate of temperature change for each respective zone i. For greater clarity, a zone in a building 170 may refer to a room, or a group of interconnected rooms.

[00130] In equation [2] above, the following quantities are unknown parameters to be learned from input data by neural network 500: T^, 0. While 0 are the parameters of the neural network,

T; and Ttj are physically meaningful and their values can be extracted to gain insight into the thermal properties of one or more zones of the building 170.

[00131] By including a neural network 500 in the differential equation [1] or [2], the neural network 500 is tasked with solving the practical problem “given the HVAC system’s set points, what is the HVAC system’s contribution to the rate of change of the temperature of a given zone?" This is a much simpler and more structured problem than directly putting all the available data into a neural network and asking “what is the temperature at discrete times in the future?”, thereby leading to a much more efficient machine learning model that requires substantially less training time. [00132] The hybrid approach of the present embodiments, including an explicitly-defined model portion and an approximated model portion, reduces the data requirements for training neural networks (from months of data to days in some cases), improves generalization of the machine learning model, and can enhance performance, and provides increased human interpretability.

[00133] In some embodiments, the explicitly-defined portion may be used to model a physics property or characteristics for a room or zone, when the associated phenomena or behaviour behaves in such a way that facilitates explicit definition based on physics modeling, and that at least some inputs to the explicitly-defined portion are available. Meanwhile, approximated model portions can be used where it is difficult to model the physics property or characteristics of the room or zone based on existing physics behaviours, or that when it is more efficient to model the physics property or characteristics to use a function approximator, for example, when there is a large amount of historical data for training the function approximator to simulate the physics property or characteristics.

[00134] The reduced data requirements allows implementation on devices with limited computing resources (e.g., without a GPU) for neural network training. For example, implementation in thermostats or other devices located in each zone (e.g., room) becomes possible.

[00135] In some embodiments, the temperature model 110 also provides a mechanism to extract physically meaningful parameters from sensor data to quantify thermal properties of one or more zones directly from the sensor data.

[00136] Given the rate of change of temperature generated based on the temperature model 110, the controller device 100 can determine a predicted temperature for one or more zones of the building 170.

[00137] For example, if a current temperature of the zone is currently measured at x degrees, the rate of temperate change is y degrees per minute, then the predicted temperature for the zone at two hours from now may be z = x + y * 60 minute.

[00138] In some embodiments, the rate of temperate change represented by equation [1] or [2] may be in a continuous time domain if dt is sufficiently small. In this case equation [1] or [2] can be used to estimate the temperature at an arbitrary time. [00139] The controller device 100, in response to the predicted temperature, can generate command signals for controlling at least one set point of the HVAC system.

[00140] In some embodiments, generating the command signals may include generating one or more operating values for the HVAC components 400 based on the predicted temperature; and transmitting the one or more operating values to the HVAC components.

[00141] In some embodiments, the input to generate one or more command signal may include one or more of: predicted space temperature prediction for each zone, virtual electrical or gas meter values, HVAC controls inputs and outputs (from BMS 150), occupancy schedule (from BMS 150 or a third party database), grid GHG intensity data (from an API or third party database), tariff structure (from an API or third party database), energy billing and consumption data (from an API or third party database), mechanical design information (from BMS 150 or engineer review), and occupancy (from BMS 150 or sensors 200). The output from the controller device 100 may include: roof-top unit (RTU) and/or Variable Air Volume (VAV) commands. RTU commands may include, for example: fan start/stop, fan speed, heating stage, cooling stage, and economizer damper. VAV commands may include, for example: space setpoint, fan start/stop, reheat modulation and damper modulation.

[00142] The system 480 may include a database server 450 connected to the controller device 100. The database server 450 is connected to a dashboard display device 420 and a monitoring system 460. The controller device 100 may implement relational-control algorithms to monitor, via the monitoring system 460, different parameters of the entire fleet of HVAC components 400 and sensors 200 within a building 170 to deliver continuous, automatic adjustments to the system based on the building load, regardless of facility type. The goal of these systems and methods is to generate energy usage savings (kWh/yr.), demand savings (kW), heating/cooling load (T herms), cooling tower water usage savings (gal/yr.), carbon footprint reduction (lbs. /yr.) and Power Usage Effectiveness (PUE) Reduction. By doing so, these systems and methods save impressive amount of dollars, reduce the load on the grid and improve tenant comfort.

[00143] In some embodiments, the control device 100 is configured to simulate performance of the BMS system 150 (e.g., temperature / humidity of particular spaces, demand levels, emissions, etc.) during multiple rounds of execution of one or more control modules. During each round, a different combination and order of the control modules may be executed. In this way, each “path to a desired outcome” can be evaluated, and an optimal path that brings the BMS closest to the client’s desired outcome can be selected. The simulations may be analyzed and compared by the control device 100, which may implement a powerful prediction engine that ingests inputs (e.g., weather forecasts - e.g., next few hours), occupancy rates, emissions dates, etc., to predict the performance of particular paths.

[00144] An example energy saving control module may be configured to save as much energy as possible by managing optimal start and stop times for the HVAC components 400 within a HVAC system. The energy saving control module may use temperature predictions, which may be generated by a separate temperate prediction module, as well as a current outdoor air temperature condition and building conditions, which may be sent from a BMS 150, to decide when the optimal time is to start and stop one or more HVAC components 400.

[00145] Another example control module may be configured to modulate chiller or boiler delivery temperature in accordance with the a predicted outdoor air temperature, a predicted building thermal load, which may be generated by a separate thermal load prediction module, and a present outdoor air temperature, which may be sent from a third party weather database or a sensor 200. This control module may be configured to determine one or more set points for chillers, boilers or variable-speed fan cooling towers.

[00146] The controller device 100 may be configured and installed to connect with HVAC components 400 directly, and when the controller device 100 is deployed at an existing building, the controller device 100 can connect with an existing control system, such as BMS 150, with existing control loops for HVAC components 400. In the latter situation, the existing control loops are included within the above-noted simulations. When use of an existing control loop is predicted to be part of the optimal path, then the controller device 100 may yield control of HVAC components 400 to the existing control loop of the BMS 150. In some examples, control is yielded until it is no longer optimal. In other examples, the controller device 100 may take over control of the HVAC components 400, in which case, the previous control loop of the BMS 150 may be completely overwritten or overridden.

[00147] The controller device 100 may also be configured to enable a client (e.g., an enterprise user or a building manager) to create its own custom logic. Clients can have multiple versions of BMS systems in their buildings, each with their own software, firmware or control loop. Making any changes to existing sequences of operation can be laborious, expensive and vendor dependent. The controller device 100 provides clients with a standard programming and vendor agnostic platform to write its own proprietary logic and push it down to the underlying building BMS 150.

[00148] The system may include a display device 420 such as a dashboard, which may be configured to display a plurality of data items such as current and predicted temperature, control module or temperature model being deployed for each BMS 150, and other metrics.

[00149] By leveraging the accuracy of various prediction models, such as temperature prediction model, the controller device 100 calculates a sequence of possible trajectories, and returns the optimal combination with respect to one or more specific user objectives. In some embodiments, the control device 100 can leverage machine learning models to drive controls directly, e.g., by using reinforcement learning agents.

[00150] It is common for buildings to be designed with oversized HVAC equipment and to be programmed to overcool or overheat when a high temperature or low temperature event occurs. In some embodiments, as the controller device 100 detects can deploy control modules depending on the HVAC equipment and operating parameters of the building to minimize energy consumption and cost while maintaining I improving comfort (adherence to a setpoint, e.g., 21 Celsius degrees). This can be achieved by predicting, via a prediction control module, when the HVAC system needs to start heating or chilling in the morning based on today’s weather forecast (e.g. start at 6:30 AM for a 7:00 AM occupancy start vs. the statically programmed HVAC schedule of 6 AM). However, if the predicted weather is too high or low then an Al-based control module will release the building back to the BMS 150 so it can turn on as early as possible to meet the anticipated demand for cooling or heating.

[00151] One or more input may be obtained or received from the BMS 150 or sensors 200 of the building, including for example, actual temperature, mechanical equipment values, number of zones, and heat leakage. One or more input may be forecasted values obtained from a third party, such as outside temperature forecast. One or more input, such as variance of temperature, may be forecasted value(s) from one or more control modules configured to predict such values.

[00152] The one or more machine learning based control modules may be implemented based on one or more machine learning models such as, for example, long short-term memory, gated recurrent unit, or multi-layered perception. [00153] One or more input may be obtained or received from the BMS 150 or sensors 200 of the building, including for example, outside temperature, cloud cover, position of sun, humidity, pollution data, a number of zones and equipment associations. One or more input may be forecasted values obtained from a third party, such as outside temperature forecast.

[00154] In some embodiments, the controller device 100 can send the commands directly to the HVAC BMS outputs points (e.g., binary output, analog output) to dynamically control and manage the HVAC components 400.

[00155] FIG. 6 is an example process 650 performed by a controller device 100 for dynamic management of the HVAC components 400 of a building 170, according to one embodiment.

[00156] At step 602, the controller device 100 stores or access a temperature model 110 for computing a rate of temperature change for at least one of zone of the building 170, the model 110 including an explicitly-defined model portion and an approximated model portion.

[00157] In some embodiments, the approximated model portion is solved using a function approximator configured to generate an output reflecting a quantity of heat added or removed to the at least one zone.

[00158] In some embodiments, the function approximator includes a neural network 500 trained to generate at least a portion of the output.

[00159] In some embodiments, the output from the approximated model portion further reflects a total amount of temperature change during a given period based on a current set point.

[00160] In some embodiments, the output from the approximated model portion further reflects an environmental thermal contribution.

[00161] In some embodiments, the environmental thermal contribution includes a solar radiation.

[00162] At step 604, the controller device 100 receives sensor data from a plurality of temperature sensors 200 within the building 170. The sensor data may include, for example, pressure sensors, temperature sensors, humidity sensors, light sensors, movement sensors, indoor air quality sensors, occupancy sensors, and so on. [00163] At step 606, the controller device 100 receives outdoor temperature data, which may be obtained from one or more external databases, such as local weather station or weather database.

[00164] At step 608, the controller device 100 computes the rate of temperature change for the at least one zone by providing the sensor data and the outdoor temperature data to the temperature model, where the approximated model portion is solved using a function approximator, such as a neural network, trained to generate an output reflecting a quantity of heat added or removed to the at least one zone.

[00165] In some embodiments, the rate of temperature change for the at least one zone i is represented by: where T _t is a respective temperature of the zone i obtained from the sensor data, T _env is the outdoor temperature data, _t is a time-scale associated with a thermal decay of zone are the time scales describing a thermal exchange between zone i and adjacent zone j, C _HVAC is related to a thermal capacity of zone i, and q _HVAC is the quantity of heat added or removed to the at least one zone i.

[00166] In some embodiments, C _HVAC q _HVAC represents the total amount of temperature change computed by the approximated model portion.

[00167] In some embodiments, rate of temperature change for the at least one zone i is represented by: where T _t is a respective temperature of the zone i obtained from the sensor data, T _env is the outdoor temperature data, _t is a time-scale associated with a thermal decay of zone are the time scales describing a thermal exchange between zone i and adjacent zone j, and ft(U, SolRad, . . ) models an environmental thermal contribution in zone i. [00168] In some embodiments, fi(U,SolRad, .. ) = NN(0, U, Solrad, .. )[i], where NN represents the function approximator for the approximated model portion, U represents a vector of one or more HVAC controls of the HVAC system, SolRad is a given data reflecting solar radiation contribution, and 0 represents one or more parameters of the function approximator NN.

[00169] In some embodiments, the equation ft U,SolRad, . . ) is determined for each zone in the plurality of zones.

[00170] At step 610, the controller device 100 determines a predicted temperature based on the rate of temperature change.

[00171] For example, if a current temperature of the zone is currently measured at X degrees, the rate of temperate change is Y degrees per time unit (e.g., minutes), then the predicted temperature for the zone at two hours from now may be X + Y * 60 minutes.

[00172] In some embodiments, the rate of temperate change represented by equation [1] or [2] may be in a continuous time domain if dt is sufficiently small. In this case equation [1] or [2] can be used to estimate the temperature at an arbitrary time.

[00173] At step 612, the controller device 100, in response to the predicted temperature, generates command signals for controlling at least one set point of the HVAC system.

[00174] In some embodiments, generating the command signals may include generating one or more operating values for the HVAC components 400 based on the predicted temperature; and transmitting the one or more operating values to the HVAC components.

[00175] FIG. 7 is a schematic diagram of computing device 700 which may be used to implement the controller device 100, in accordance with an embodiment.

[00176] As depicted, computing device 700 includes at least one processor 702, memory 704, at least one I/O interface 706, and at least one network interface 708.

[00177] Each processor 702 may be, for example, any type of general-purpose microprocessor or microcontroller, a digital signal processing (DSP) processor, an integrated circuit, a field programmable gate array (FPGA), a reconfigurable processor, a programmable read-only memory (PROM), or any combination thereof. [00178] Memory 704 may include a suitable combination of any type of computer memory that is located either internally or externally such as, for example, random-access memory (RAM), read-only memory (ROM), compact disc read-only memory (CDROM), electro-optical memory, magneto-optical memory, erasable programmable read-only memory (EPROM), and electrically- erasable programmable read-only memory (EEPROM), Ferroelectric RAM (FRAM) or the like.

[00179] Each I/O interface 706 enables computing device 700 to interconnect with one or more input devices, such as a keyboard, mouse, camera, touch screen and a microphone, or with one or more output devices such as a display screen and a speaker.

[00180] Each network interface 708 enables computing device 700 to communicate with other components, to exchange data with other components, to access and connect to network resources, to serve applications, and perform other computing applications by connecting to a network (or multiple networks) capable of carrying data including the Internet, Ethernet, plain old telephone service (POTS) line, public switch telephone network (PSTN), integrated services digital network (ISDN), digital subscriber line (DSL), coaxial cable, fiber optics, satellite, mobile (e.g., 4G, 5G network), wireless (e.g. Wi-Fi, WiMAX), SS7 signaling network, fixed line, local area network, wide area network, and others, including any combination of these.

[00181] For simplicity only, one computing device 700 is shown but system 100 may include multiple computing devices 700. The computing devices 700 may be the same or different types of devices. The computing devices 700 may be connected in various ways including directly coupled, indirectly coupled via a network, and distributed over a wide geographic area and connected via a network (which may be referred to as “cloud computing”).

[00182] For example, and without limitation, a computing device 700 may be a server, network appliance, set-top box, embedded device, computer expansion module, personal computer, laptop, personal data assistant, cellular telephone, smartphone device, UM PC tablets, video display terminal, gaming console, or any other computing device capable of being configured to carry out the methods described herein.

[00183] In some embodiments, a computing device 700 may be adapted to function as a controller device 100.

[00184] The discussion herein provides many example embodiments of the inventive subject matter. Although each embodiment represents a single combination of inventive elements, the inventive subject matter is considered to include all possible combinations of the disclosed elements. Thus if one embodiment comprises elements A, B, and C, and a second embodiment comprises elements B and D, then the inventive subject matter is also considered to include other remaining combinations of A, B, C, or D, even if not explicitly disclosed.

[00185] The embodiments of the devices, systems and methods described herein may be implemented in a combination of both hardware and software. These embodiments may be implemented on programmable computers, each computer including at least one processor, a data storage system (including volatile memory or non-volatile memory or other data storage elements or a combination thereof), and at least one communication interface.

[00186] Program code is applied to input data to perform the functions described herein and to generate output information. The output information is applied to one or more output devices. In some embodiments, the communication interface may be a network communication interface. In embodiments in which elements may be combined, the communication interface may be a software communication interface, such as those for inter-process communication. In still other embodiments, there may be a combination of communication interfaces implemented as hardware, software, and combination thereof.

[00187] Throughout the discussion herein, numerous references will be made regarding servers, services, interfaces, portals, platforms, or other systems formed from computing devices. It should be appreciated that the use of such terms is deemed to represent one or more computing devices having at least one processor configured to execute software instructions stored on a computer readable tangible, non-transitory medium. For example, a server can include one or more computers operating as a web server, database server, or other type of computer server in a manner to fulfill described roles, responsibilities, or functions.

[00188] The technical solution of embodiments may be in the form of a software product. The software product may be stored in a non-volatile or non-transitory storage medium, which may be a compact disk read-only memory (CD-ROM), a USB flash disk, or a removable hard disk. The software product includes a number of instructions that enable a computer device (personal computer, server, or network device) to execute the methods provided by the embodiments.

[00189] The embodiments described herein are implemented by physical computer hardware, including computing devices, servers, receivers, transmitters, processors, memory, displays, and networks. The embodiments described herein provide useful physical machines and particularly configured computer hardware arrangements.

[00190] Of course, the above described embodiments are intended to be illustrative only and in no way limiting. The described embodiments are susceptible to many modifications of form, arrangement of parts, details and order of operation. The disclosure is intended to encompass all such modification within its scope, as defined by the claims.