Performance Assessment And Model Plant Mismatch Detection Of MPC Based On Closed-Loop Data

As the representative of the advanced control,Model Predictive Control(MPC)has been widely applied in complex industrial process,performance monitoring technology has great effect to improve the safety,product quality of process and economic benefit.In this paper,for performance assessment of constrained MPC and detection of model plant mismatch,we have studied a kind of constrained MPC performance assessment method based on multi-step prediction error approach.In addition,a data-driven method is introduced to detect the model plant mismatch.The main research of this paper is as follows:Considering various constraints in the practical industry process,an extended multi-step prediction error approach is proposed.It is based on the multi-step prediction error approach,which also introduces the offset of the system.Via the routine operating data,time series model of the system can be built.By calculating the prediction error,the offset and the covariance,a closed potential curve can be drawn.This approach not only takes the dynamic performance but also the tracking performance into consideration.Consequently it can reflect the performance of constrained MPC system more generally.The simulation example on the WoodBerry binary distillation column demonstrates the validity of the method.For the detection of model plant mismatch,a method based on Markov parameters is introduced to solve it.By the input and output data which is sufficiently motivated,Markov parameters can be estimated based on the Joint Input-Output subspace method.The Markov parameters which are mismatched can also be calculated.Finally,the mismatched input-output channel will be detected by the quantization index.The approach does not need to identify the overall system.It not only can avoid a vast number of calculations,but also has high accuracy.The simulation example on the Shell distillation column demonstrates the validity and feasibility of the method.In order to further verify the application results of the performance asessment method based on the extended multi-step prediction error approach and the detection approach of model plant mismatch based on the subspace approach,we have carried out experimental research in the actual process control experimental device in our laboratory.The experimental results show the effectiveness of the research methods.