A Study On Closed-loop Identification And Model Predictive Control Of Thermal Process

Nowadays,with development of supply and demand of energy resource,the energy issues have been aggravating increasingly,high efficient operation and low carbon emission for coal-fired power plant is fulfilled with large expectation.Thermal process of power plant is complex process with large nonlinear and thermal inertia suffering from the unpredictable disturbance such as coal sources variation and valve oscillation,moreover,aging of equipment and wear and tear of actuators change the dynamic property of thermal process.Besides,a large amount of working data is collected during power plant routine operating data,base on which a mathematical model is established to alleviate the issue identification experiment disturbs the system normal operation.For these reasons,the classical thermal process control strategies(PID based)cannot achieve satisfied control performance,and the model with satisfied accuracy cannot be the traditional open-loop system identification approaches.This paper studies the theories and applications of operating data-driven close-loop identification and model predictive control.The effectiveness of proposed closed-loop identification and robust model predictive control approaches are demonstrated by applications of main steam temperature system(MST)and CO₂ capture system in coal-fired power plant.The main research contents and contributions of this work are stated as follows:1.Model structure determination and parameters estimation have been involved in closed-loop identification.Firstly,in the framework of prediction error identification method,a model order criterion of total identified model error is proposed,the model order is the value which minimizes the criterion;then,the identification data is filtered with the pre-estimation model being the filter,the filtered control variable is uncorrelated with the system disturbance;finally,the unbiased model parameters estimation is achieved by applying the output error identification method.2.Focusing on model uncertainty,a zonotope-type uncertain model is estimated based on proposed closed-loop set-membership identification method.The control variable is uncorrelated with the system disturbance via the pre-estimation model,the model structure determined by the closed-loop identification method is utilized as the linear regression parameterization of feasible system set,and zonotope is employed as approximated feasible system set to be identified,whose parameters are estimated by solving optimization problem recursively.A proposed optimizing index,which is composed by model uncertainty and the nominal model accuracy,about the de-ordered zonotope is minimized to guarantee the optimality of the identified zonotope with constant order.3.An explicit robust model predictive control approach for control target tracking is developed to meet the requirement of robustness and online computation burden for thermal process controll.Zonotope identified by closed-loop set-membership is used as prediction model,and the center model of zonotope as nominal model,by limiting the ideal MPC control performance to preset level,the whole system state space has been automatically partitioned into some simplex-type subspaces and corresponding local MPC control laws is calculated.During the control law implementation,the control law is linearly weighted summation of RMPC laws of the system states at vertexes of simplex which the current state belongs to.To drive system output exactly track the control set point,a manipulated variable target observer is developed based on nominal model.For EFERMPC controller design with uncontrollable state space prediction model,the whole system state space is decomposed into controllable state space and uncontrollable state space based on nominal model,and then by partitioning the controllable state space into several simplexes and designing the RMPC control law,an EFERMPC control strategy is developed with uncontrollable state space prediction model.4.To enhance the maintainability and control performance of thermal process controller,a synthetic RMPC approach combined with control performance assessment is presented.Based on convex hull RMPC,an adaptive robust model predictive control(ARMPC)approach is presented to guarantee controlled object sufficiently excited.The proposed synthetic RMPC is a controller with essence of switching between RMPC controller and AMPC controller,RMPC is adopted during the system normally working;if the monitored control performance is larger than preset threshold,it needs to update the controller parameters:and if control output variation or controlled output deviation is too large,the weighting matrix of RMPC performance index should be adjusted;besides,if the maximum error of the prediction model is larger than model error bound,then the controller is switched to the ARMPC,the prediction model parameters and the error bound are updated by the zonotope based closed-loop set-membership identification method,after the adjustment finishing,the controller is switched back to the ARMPC.