Control Performance Assessment For ILC-Controlled Batch Processes Based On MPC Benchmark

Nowadays,industrial automation is very popular.How to ensure that the controller is always in the best operating state is worthy of scholars to explore,because it is closely related to energy loss,product quality,business benefits and other aspects.To this end,control performance assessment techniques are available to enable the monitoring of industrial processes without disturbing the system.The optimal achievable performance is used as a benchmark for the current performance of the system.Field force can decide whether it is necessary to take improvement measures such as setting parameters and even replacing the controller based on their comparison.This process not only does not require too much manpower and material resources,but also solves the problem of huge engineering volume in the traditional maintenance process.Although the concept of performance evaluation has been proposed for a long time,it is difficult to cover every aspect as a wide-ranging area.Especially for the performance evaluation of iterative learning control,there are very few achievements in this field.The reality is that iterative learning control is extremely effective for batch processes,and batch processes are widely used in actual industrial production.Therefore,this paper takes the performance evaluation of iterative learning control as the subject,and further promotes the improvement of the performance evaluation system.During the analysis of evaluation process,this paper found that it is especially important to select or design a benchmark that is close to reality.It can give better guidance,so that the performance of the current system gradually moves closer to the benchmark.After many comparisons,this paper selects the MPC benchmark.Before the benchmark is solved,the model of the control system should be known,but this requirement is not always met in practical applications.This paper considers the known and unknown conditions of the model.For the latter,subspace identification is used to solve the model parameters.The algorithm can be summarized as follows:Firstly,the Hankel matrix is constructed according to the measured input and output data;then the subspace matrix is calculated by means of geometric tools such as QR decomposition and SVD decomposition;finally,the system model parameters are estimated by the column subspace and the row subspace.After obtaining the model,the closed-loop system under the action of iterative learning control is converted into a two-dimensional FM-? model,so as to simplify the research.At the same time,the Roesser model is a special case of the FM-? model,and the FM-? model can be transformed into the FM-? model.Therefore,the research on the FM-?model makes the conclusions of this paper more general.Then,this paper proposes an algorithm to optimize the cost function and gives a detailed derivation proof.According to these algorithms,the optimal control law is obtained and a novel performance evaluation surface is fitted,so that the performance gap can be more intuitively reflected in the stereo coordinates.Finally,the effects of the above work were verified by simulation.