Research And Application On Performance Evaluation And Self-healing Of Model Predictive Controller

Model Predictive Control(MPC)as a type of advanced control method has a wide range of applications in aerospace,petrochemical,power generation and other industrial fields.MPC controllers operate normally and maintain good controller performance has very important practical significance.In industrial practice,MPC controllers are more accurate at the initial stage of commissioning and generally meet the performance requirements.After a period of operation,due to various factors,the MPC controller performance may decline and deviate from the system performance benchmark.In view of the performance degradation of the model predictive controller in actual operation,this paper proposes two self-healing algorithms of the model predictive controller.The contents of the main academic research are as follows:(1)A model prediction controller performance evaluation and self-healing method based on ISE-TSV index is proposed.First,based on the integral squared error(ISE)and total squared variation(TSV)indicators,the model prediction controller is evaluated in current time.Then,based on the inverse characteristics of the infinite time domain MPC and based on the analysis of the ISE-TSV indicators,an MPC robust control self-healing method is proposed.The model predictive control simulation and experimental results on the double inverted pendulum give eloquent proof of the validity and practicability of the proposed self-healing method.(2)A self-healing algorithm of model predictive controller based on new neural dynamics is proposed.When the model mismatch is serious and the MPC controller performance is greatly reduced,the new neural dynamics algorithm is used to start the self-healing module of the prediction bias model,correct the model prediction output,and then use the corrected model prediction output to use the new neural dynamics.The algorithm recognizes the parameters of the object model and updates the controlled object model to realize the self-healing of the controller performance.And the simulation results of model predictive control with a linear two-stage inverted pendulum give eloquent proof of the validity and practicability of the proposed self-healing algorithm.(3)In the VS environment,use C#design and write an APC performance monitoring software to achieve the functions of data acquisition,index calculation,multi-index evaluation and Web publishing of the controller,and achieved good industrial practical application results.