This invention relates to control and optimisation of development of oil and gas wells, in particular removing anomalies and errors from data in the control loop.

In the field of process control in the oil and gas industry, well data is logged and recorded and then analysed by engineers. Currently data cleaning, which is the removal of anomalies and errors in oil well <BR> production data (i. e. , separating the good from the bad data) is performed by skilled analysts from the oil and gas companies. This process is both time consuming and laborious. The analysts explore the data using computer based tools (e. g. , trend analysis using spreadsheets) and well and reservoir models, and apply their considerable knowledge of oil production in order to make decisions about where and how to clean and analyse the data.

Statistical process control (SPC) is well known as a tool for real time monitoring of well data, applying rules to

data points with respect to targets and limits and signalling problems using alarms.

The problem with conventional SPC in the oil and gas industry is that the raw data that is logged is often recorded in extreme environments and at inaccessible locations.

Typical responses to out of control data points in conventional SPC either make assumptions about the normal distribution of the data, or the capability of the measurement apparatus. In the oil and gas industry, the measured data are made"dirty"by the ambient conditions of the measurement systems and the physical constraints of the downhole environment mean that the measurement systems such as downhole gauges are often less than capable.

Typical responses too conventional SPC also include shutting down equipment, often automatically, when data points go out of control. In the oil and gas industry, shutting down equipment can be very expensive, for example with costs of up to E2 million per hour in the event of shutting down a production well.

US Patent No US5,282, 261 assigned to Du Pont discloses a computer neural network process measurement and control system and method that uses real time output data from a neural network to replace a sensor or laboratory input to a control. A historical database is disclosed for providing a history of sensor and laboratory measurements to facilitate the training of the neural network.

European Patent Application No EP0881357 discloses the development of an oil or gas reservoir controlled using a neural network and genetic algorithm programs to define a neural network typology and the optimal inputs for that topology. The method disclosed utilises neural network technology to use multiple input parameters for determining corrolations with a desired output and uses genetic algorithms to define the neural network topology and corresponding optimal inputs.

The problem with both of these approaches is that the neural network filters out useful information that may be present in original source measurements. There is no mechanism for providing experts with the detail of the cleaning that has been done, as the neural networks act as a"black box". Neither is there a mechanism for feeding back the results of expert analysis into the training of the neural networks.

It would be advantageous to provide a control system that used a variety of methods of cleaning data including neural network and genetic algorithm processing, then feed the cleaned data into a separate analysis and control system.

It would be advantageous to provide data cleaning that maintains the integrity of the original data and marks up anomalies for subsequent review by skilled engineers, without losing the information.

It would be further advantageous to provide data cleaning that removes anomalies and marks up data corresponding to the removed data for subsequent review by skilled engineers.

It is an object of at least one embodiment of the present invention to aid operators in getting the best productivity out of their oil assets through a combination of data smoothing techniques, automated analysis, and skilled expert analysis.

It is an object of at least one embodiment of the present invention to remove anomalies and errors from oil and gas data logs It is a further object of at least one embodiment of the present invention to mark up anomalies and errors in oil and gas data logs.

According to a first aspect of the present invention, there is provided a system for controlling the development of an oil or gas well comprising: a data capture means for receiving well data; a data analysis means for analysing said well data, the data analysis means being responsive to a data cleaning means for cleaning said well data; and a control means for controlling further development of the oil or gas well.

According to a second aspect of the present invention, there is provided a device for controlling oil and gas production comprising: a data capture means for receiving and well data ;

a data analysis means for analysing said well data, the data analysis means being responsive to a data cleaning means for cleaning said well data; and a control means for controlling further development of the oil or gas well.

According to a third aspect of the present invention, there is provided a method for controlling oil and gas production comprising the steps of: receiving well data; cleaning said well data; analysing said well data; and controlling further development of the oil or gas well.

Preferably said data cleaning means is adapted to remove anomalous and/or erroneous data from said well data.

More preferably said data cleaning means is adapted to modify anomalous and/or erroneous well data.

Typically the system further comprises a data transmission means for transmitting said well data from said data capture means to said data cleaning means.

Typically said system further comprises a data delivery means for delivery of data from said data cleaning means to said data analysis means.

Most preferably said data cleaning means is adapted to mark up said well data.

Preferably said data cleansing means is adapted to smooth said well data.

Preferably said well data is stored as XML (Extensible Mark-up Language).

Preferably said data cleaning means is adapted to mark-up said XML document.

Typically said marking-up of said XML document comprises the step of adding an attribute to a data object in said XML document.

Preferably said data cleaning means further comprises a database of rules.

Preferably said data cleaning means further comprises a database of known cases.

Preferably said data cleaning means is responsive to feedback from said data analysis means.

Preferably said data cleaning means is adapted to remove random noise from said well data.

More preferably said data cleaning means is adapted to remove unexplained spikes or trends of said well data.

Preferably said data cleaning means further comprises models of reservoir or well processes.

Preferably said data analysis means is adapted to display said well data marked-up responsive to the changes made by the data cleaning means.

The present invention will now be illustrated with reference to the following figures in which: Figure 1 illustrates a schematic diagram of the overall system architecture in accordance with the present invention; and Figure 2 illustrates a schematic diagram of a device mounted at a well head in accordance with the present invention.

With reference to Figure 1 that shows the whole system, the data capture equipment 10 logs well data. The well data may be well logging data. The well data may be production logging data. For example there may be eight data sources and typically each is sampled at the rate of one data point per second. The data sources (e. g., pressure sensors or thermocouples or other downhole gauges) provide inputs analogue inputs to an analogue to digital converter. Digital inputs may also be logged.

The data is collected and transferred 20 from the oil installation to a data centre which may be at the location of the well or at a remote site, e. g. , onshore if the well is offshore.

In the prior art, data cleaning 30 is typically performed by skilled analysts using a combination of simple computer based tools, e. g. , spreadsheets and statistical

packages, and oil production knowledge and experience.

This cleaning process results in data in which anomalous and erroneous data is either removed and/or modified.

This process is very time consuming and results in a significant delay between receipt of logged data and on delivery 40 of cleaned data for analysis and control of production.

According to the present invention, the data cleaning is performed by a data cleaning module. The data is cleaned using knowledge based techniques and machine learning techniques and furthermore the data is marked up with the results of the cleaning.

The data cleaning module is a set of software components that receive uncleaned well data as input and output cleaned well data. The cleaned well data is delivered 40 to a data analysis module 50, which also comprises a set of software components.

In the preferred embodiment, the data cleaning module and data analysis module are both accessed and operated using a Data Analysis Workbench (DAW) which is a computer program with a graphical user interface. The DAW framework is extensible and flexible.

The DAW includes software components mathematical models of well and reservoir processes. These models are incorporated into the simulation component. The purpose of the simulation component is to predict the well/reservoir behaviour and to assist in the data cleaning process. Secondly, there is a filtering component, the purpose of which is to remove the noise

from the data using standard filtering techniques. The trend analysis is implemented by another DAW component that uses statistical tools identified in. Finally, the interpretation stage of the data cleaning process is carried out by a number of interacting components. One of these components is an intelligent information processing system capable of identifying and diagnosing critical conditions using both rule-based and evolutionary approaches. A data repository is provided where previously logged data is stored together with its analysis.

The end-user interface uses standard components to construct the user interface. In particular, generic manipulation and display functionality are realised through"calls"to standard application packages, e. g. spreadsheets.

The DAW is designed so that it can be used in two modes of operation: semi-automated (driven by an analyst) or completely automated. In the case of semi-automated operation, it is possible to learn from the decisions made by the analyst. Therefore, provision is made for a learning/feedback process both within the data cleaning module and between the analysis and cleaning modules. For easier mode of operation, it is possible to review the results of the data cleaning process, via the graphical user interface.

Since the output of one stage of data cleaning is used as the input for the next stage or delivery to the analysis stage 40, a uniform data format is provided for the data to be seamlessly passed around the system. An XML

(eXtensible Mark-up Language) data format is used for marking-up logged data, and specifically marking the results of data cleaning. Thus the result of applying the data cleaning process is an output data set, in which changes to the input data set are clearly marked up in XML, together with reasons for the changes.

In using mathematical models and statistical analysis for data cleaning, algorithms store electronically expected models for data and carry out data screening techniques to detect departures from models. Diagnostic information is presented to users in the form of tables and graphs that highlight suspect or cleaned data. The output, optionally combined with user input, may be used as an input into"intelligent"diagnostic algorithms for decision making.

These"intelligent"technologies are both quantitative and qualitative. For quantitative analysis, optimisation techniques are used that intelligent search and evaluate possible explanations. These include evolutionary algorithms (e. g. , genetic algorithms) and pattern recognition techniques. For qualitative analysis, approaches that interpret data on the basis of expert knowledge (e. g. , rule based systems) or infer from known<BR> cases (e. g. , cased based reasoning) are used.

Data cleaning may require both the removal of random noise, as well as the diagnosis of unexplained spikes or trend in the well data.

It will be appreciated that the data analysis 50 may be via software statistical analysis as described, or 4-D or virtual reality.

The data analysis module provides data to the control module 60 and either automatically or semi-automatically control outputs are provided to the production assets 70.

The well data from the production assets are captured by the data capture module 10.

In a further embodiment, the entire process is performed in real time. This allows the device to be located adjacent a well for immediate control of a well via the analysed data. This is shown in Figure 2. The device, generally indicated by reference numeral 100, is located adjacent the well 120. Data from downhole devices (not shown) is typically transmitted to the device 100 via control lines 140, run from the well 120.

Cleaning and analysis of the data is carried out as described hereinbefore, with reference to Figure 1.

Anomalies in the data are transmitted to a viewing unit 160 for viewing by an operator 180. The operator 180 need only adopt the control of the well if they consider it necessary. Generally, the device 100 will use the analysed data to adopt the control of the well 120 automatically. The advantage of such an adaptive control allows the well 120 to operate as a so-called "intelligent well"operated via neutral network and knowledge based management systems which optimises performance of the well by allowing automatic control in real time or near real time.

Further modifications and improvements can be made by one skilled within the art within the scope of the invention herein disclosed.