KUBAL, Nandkishor (Flat No. 602, Mandar Prasun Dham,,Opp. Biria Hospita, Chinchwad Pune, IN-411033, IN)
BHAT, Shrikant (E-l Rambag Complex, Ramakrishna NagarKhamla Road, Nagpur, IN-440015, IN)
NANDOLA, Naresh, Naranbhai (Kanthariya Village, SurendranagarChuda, Gu arat, IN-363410, IN)
BADWE, Abhijit (12 Harekrishna Apt, Model Colony, Pune, IN-411016, IN)
KUBAL, Nandkishor (Flat No. 602, Mandar Prasun Dham,,Opp. Biria Hospita, Chinchwad Pune, IN-411033, IN)
BHAT, Shrikant (E-l Rambag Complex, Ramakrishna NagarKhamla Road, Nagpur, IN-440015, IN)
NANDOLA, Naresh, Naranbhai (Kanthariya Village, SurendranagarChuda, Gu arat, IN-363410, IN)
| WE CLAIM: 1. A method for updating a model in a model predictive controller, the method comprising: assessing the deviation of the operating performance level from the desired performance level of the process plant; diagnosing the model predictive control (MPC) for the model plant mismatch (MPM), by updating the model in a model predictive controller and of the MPC thereof; wherein diagnosing the said MPC comprises determining the model prediction error in relation to MPM; quantifying the MPM; and updating the model in the said model predictive controller. 2. The method as claimed in claim 1 , wherein assessing the deviation of the operating performance level from the desired performance level of the process plant include monitoring the performance of the said MPC; and detecting the cause for the said deviation, the said deviation of the operating performance level from the desired performance level of the process plant corresponds to MPM. 3. The method as claimed in claim 1 , wherein diagnosing the said MPC includes correcting the said deviation of the operating performance level from the desired performance level of the process plant by correcting the model of the said MPC for the MPM and of the said MPC thereof. 4. The method as claimed in claim 1 , wherein determining the model prediction error in relation to MPM comprises providing set points with white noise to the controllers in the process plant and to the existing model of the MPC, after adding white noise to the set point values of the controllers in the process plant. 5. The method as claimed in claim 4, wherein determining the model prediction error in relation to MPM further comprises estimating the model prediction error by calculating the difference between the output value of the process plant and of the existing model of the MPC. 6. The method as claimed in any of claims 1 , 3, 4 or 5, wherein quantifying the said MPM using non zero lag correlation coefficients between the said model prediction errors and each of the manipulated variables of the said MPC, said manipulated variables of the MPC being the set point values of the controllers in the, process plant and including white noise. 7. The method as claimed in any one of the preceding claims, wherein the model update is performed offline or online. 8. The method as claimed in any one of the preceding claims, wherein the process involved in the process plant is open loop or closed loop. 9. The method as claimed in any one of the preceding claims, wherein the said model include entire model of the process plant and/or of the sub model thereof. 10. A system for updating a model in a model predictive controller in accordance with the method as claimed in any one of the preceding claims, the said system comprises: a performance monitor unit for monitoring the performance of the MPC; a detection unit for detecting the cause for deviation of the operating performance level from the desired performance level of the process plant; a white noise generator unit for generating white noise signals that are characteristically only sufficient to compute non-zero lag correlation coefficients; an estimator unit for estimating the model prediction error by calculating the difference between the output value of the process plant and of the existing model of the MPC; a quantifier unit for quantifying the MPM; and an updater unit for providing an updated model based on the quantified MPM and updating the model of the process plant in the model predictive controller. |
PREDICTIVE CONTROLLER
FIELD OF THE INVENTION
The invention relates to a Model Predictive Control (MPC), and more particularly to a method and a system for updating a model in a Model Predictive Controller.
BACKGROUND
In a process control industry, Advanced Process Control (APC) is employed so as to reduce operating costs, achieve high productivity, maintain quality consistently, and for other similar reasons. APC allow transition from present operating schema to an improved and more productive operating schema of the process control industry, and also accommodate operating and design constraints of the process involved in the process control industry.
Generally, a multivariable APC implements a more popular advanced multivariable control scheme called Model Predictive Control (MPC) in Multivariable Predictive Controllers. MPCs uses a mathematical model of the process involved in the process plant, in order to predict the future dynamic behavior of the process and accordingly provide optimal manipulated variables for the process and operation of the plant thereof. From this, it can be understood that accuracy of the model is a key element in effective and successful implementation of MPC.
Practically, plant dynamics changes, resulting in a mismatch between the model and the plant. This mismatch is termed as Model Plant Mismatch (MPM). MPM leads to inaccurate predictions of the plant dynamics. Using APC having model impacted by the said MPM will degrade the MPC and overall control performance thereof, which consequently alters the product quality and causes economic losses. It becomes important to update the model upon detection of poor performance of the controller, in order to eliminate the performance degradation of the controller. Poor performance of the controller can be detected by well established MPC performance monitoring. For instance, a simple approach could be to analyse the prediction errors, which being the difference between the model predictions and true outputs. After the detection of poor performance of the controller, cause for the same, such as poor model (i.e. MPM), unmeasured disturbance, and constraint saturation etc. can be identified and diagnosed using established diagnosis techniques. Currently, upon identification or detection of poor model, MPM is diagnosed by and after re-identification of model. Re-identification of model involves designing the perturbation signal, deciding and/or considering the operating conditions of the plant during perturbation, choosing appropriate model and estimating model parameters. This requires high degree of expertise and is quite time consuming. Also, more importantly, it involves longer perturbation period, by which a large amount or number of product with low quality, usually termed as off spec product are produced during the perturbation period.
Thus, there is a need for an alternative method that significantly reduces perturbation period and more efficiently reduces and/or eliminates MPM.
OBJECTS OF THE INVENTION
It is an object of the invention to reduce the perturbation period during diagnosis of MPM for reducing and/or eliminating MPM.
It is another object of the invention to reduce the MPM more efficiently.
SUMMARY OF THE INVENTION The invention is aimed at providing a method and system for updating a model in a model predictive controller of a process plant, so as to reduce and/or eliminate the deviation of the operating performance level from the desired performance level of the process plant arising out of MPM.
Accordingly, the invention provides a method for model update in a model predictive controller. The method comprises the steps of assessing the deviation of the operating performance level from the desired performance level of the process plant. Diagnosing the model predictive control (MPC) for the model plant mismatch (MPM) is performed by updating the model in a model predictive controller and of the MPC thereof. Diagnosing the said MPC comprises determining the model prediction error in relation to MPM. Then, quantifying the MPM and updating the model in the said model predictive controller.
Accordingly, the invention also provides a system for updating a model in a model predictive controller in accordance with the method of the invention. The system comprises a performance monitor unit for monitoring the performance of the MPC, a detection unit for detecting the cause for deviation of the operating performance level from the desired performance level of the process plant and a white noise generator unit for generating white noise signals that are characteristically only sufficient to compute m+1 non-zero lag correlation coefficients, where m is number of manipulated variables of MPC. An estimator unit is provided for estimating the model prediction error by calculating the difference between the output value of the controllers in the process plant and of the existing model of the MPC. Also, a quantifier unit is provided for quantifying the MPM. Further, the system has an updater unit for providing an updated model based on the quantified MPM and for updating the model of the process plant in the model predictive controller.
BRIEF DESCRIPTION OF THE DRAWINGS
With reference to the accompanying drawings in which:
Fig. 1 depicts a typical model predictive control setup in a process plant; and Fig. 2 shows a framework for model update in model predictive controller, in accordance with the invention. DETAILED DESCRIPTION
The invention is described with reference to Figs. 1 and 2 by way of illustration of non exhaustive exemplary embodiment. In Fig. 1 , process plant (100) is controlled by primary level controllers such as PID controllers etc., in a distributed control system, DCS (1 10). There is provided a model predictive control ( 120) which uses a mathematical model of the process involved in the process plant, hereinafter referred to as model, in order to predict the future dynamic behavior of the process and accordingly provide optimal manipulated variables for the process and operation of the plant thereof. A performance monitor unit (130) is provided for monitoring the performance of the MPC (120). Further, the detection unit (140) detects the cause for deviation of the operating performance level from the desired performance level of the process plant. Considering, the deviation mentioned herein above has been caused by or arising due to model plant mismatch (MPM), being the mismatch between model and the process plant, this invention is further explained with relevance to open loop condition of the process plant, which consideration is not restrictive in nature rather purely exemplary for the purpose of better understanding.
It becomes imperative to diagnose MPC for the MPM. Further reference is herein made to Fig. 2 in order to describe the invention. Diagnose or diagnosis referred herein also includes effecting necessary correction in the model (240) of the MPC (220) with relevance to correcting the said deviation and of the MPM thereof. Accordingly, the MPC (220) is taken offline with the set point values to the controllers being either constant with predetermined values or that evolved from the MPC (220) immediately before taking the MPC (220) offline. The MPC (220) is construed to be online in the latter case, where the set points for the controllers are the manipulated variables of the MPC (220). A white noise generator unit (230) is provided for generating white noise signals that are characteristically only sufficient to compute (m+1 , where m is number of manipulated variables of MPC) non-zero lag correlation coefficients between the model prediction errors, e and the manipulated variables, u of the MPC. The white noise signal is added to the set points as follows:
(SPi ) D cs., = SPi (t) + Wi(t)
Here SP, (t) is the i th setpoint value when the MPC is in "Offline" mode, Wj(t) is the value in the i th white noise sequence at time instant t and (SP, )ocs , t is the i th setpoint value going to the DCS (210). It should be noted here that in the event of MPC (220) being "Online", SPj (t) is computed by MPC (220) and hence is a manipulated variable of the MPC (220). Also, in the MPC (220) "Online" mode,
(SPi )DCS,, = SPj (t) = MVi(t)
The (SP; )ocs.t values are also passed to the existing Model (240), outputs of which are denoted by > , which are predicted values of controlled variables. The actual values of controlled variables from plant (200) are denoted by y. The controlled variable values from the plant (200) may be obtained through online measurements or through other suitable means such as periodic laboratory analysis. The difference between the measured value y and that predicted by the model (240), y is known as the model prediction error or model residual and is denoted by e. Note that in Figure 2, (SPi )ocs,t is denoted by u for notational simplicity. Quantifier unit (260) calculates the lag correlation coefficients between e and u. The values of these coefficients are used to calculate the "gap" between the existing model and the current plant, ie to quantify the MPM. Considering a single input and single output open loop condition of a process plant, for the purpose of simplicity in understanding, the following case for gain mismatch is explained. This is in no way restrictive and is purely exemplary and non exhaustive with regard to the invention. Similarly, the invention holds good for multi input and multi output as well as closed loop conditions also and can be applied coextensively.
For a first order time delay system, the polynomial representation is as follows: y(k) = ay(k-l) + bu(k-t dp -l) + d(k)
y (k) = a m y (k-1) + b m u(k-t dm -l) where y(k) is the output of the plant; y (k) is the output of the model; and e(k) is the model error;
a and b are the parameters of the plant; a m and b m are the parameters of the model;
k is the sampling instance;
td p is the time delay for plant; t dm is the time delay for model
Considering only gain mismatch (i.e. a= a m and td P =td m )
e(k) = y(k) - y (k) = ae(k-l) + (b-b u(k-t dp -l ) + (k) ( 1 )
It is to be noted that even if there is a mismatch in all the three parameters i.e. a,b and td , the method of the invention can be applied. For simplicity reason mismatch in only one parameter i.e. gain or b and b m is considered.
Now, the correlation coefficient at lag m between two time-series xi and ? is given
<*xl °x2
<r x j Ox2 = E fxi(k)x 2 (k-m)J
Then, from Equation 1 ,
At lag 0, r ' 0 = E [e(k)u(k)]= 0
At lag \ , r 'i= E [e(k)u(k-l)]= 0
At lag t dp + 1, r ' ldp +l = E [e(k)u(k-t dp -l)] = (b- b„J a 2
At lag t dp + 2, r ' ldp +l = E [e(k)u(k-t dp -2)J = a(b- b„J a 2
At lag t dp + n, r ' ldp +n = E [e(k)u(k-t dp -n)J =a" ~ '(b- b nx ) <r Thus, the first non-zero correlation coefficient is observed at lag t dp + l- Moreover, the correlation shows an exponential decay.
Λ
Now, suppose Y , dp +] is the correlation coefficient observed between e and U at lag t dp + 1. We can then write,
Λ
r , dp+l = (b- b a 2
Here, b- b m is the MPM quantitatively. An updater unit (270) updates the existing model (240) to an updated model (280) based on the quantified MPM and updates the model of the MPC (220) in the model predictive controller. The MPC (220) is then switched to online mode.
The invention can also be extended to updating a sub model in the similar manner as described herein before in the description. The identification of the sub model that needs to be updated due to MPM can be accomplished by known techniques such as that which applies partial correlation analysis or any other suitable techniques or method. This way, updating the model at sub model level further reduces the effort in updating the model for MPM and becomes more efficient.
Thus the invention can be applied for updating the entire model or the sub model as the case may be.
The exemplary embodiments described in this description as not restrictive in nature with regard to the scope of the invention. Other modification, changes, alteration that would cater to the same functionality and approach, and not being specifically mentioned or stated in this description are construed to be well within the scope of the invention. The usage of singular terms would include its plural form and vice versa within the meaning of the scope of the invention.
The invention finds extensive usage in chemical and petrochemical, cement, pulp and paper, and pharmaceutical industries to name a few. Some of the other applications include power generation, supply chain management and behavioural health etc.
Next Patent: TRICYCLE COMPOUNDS AS PHOSPHODIESTERASE-10 INHIBITORS