The Modeling Quality Monitoring Of Industrial Process With Model Predictive Controller

Model predictive control(MPC)was put forward at the end of 1970 s and has been widely used in petroleum,chemical,electric power,aerospace,metallurgy and other industrial control fields.MPC is suitable for complex processes such as time delay,constraint,coupling,multivariable and so on because of its three links:predictive model,rolling optimization and feedback correction.The MPC controller has the characteristics of simple modeling,good dynamic control effect and strong robustness and so on,and usually has good control performance in the early stage of production,however,with the passage of time,due to the impact of valve stiction,sensor deviation,noise disturbance,model mismatch and other factors,the performance of MPC controller will decline gradually.Therefore,it has practical significance for ensuring the safety,high quality,high efficiency and low power consumption of the production process to conduct real-time assessment and monitoring of the performance of MPC controller,and detect the performance degradation timely and alarm,and further diagnose the performance deterioration causes.Model quality is one of the decisive factors that affect the performance of model-based controller.There is usually disturbance with drift in the process of industry,which brings difficulties to the model quality assessment.In this paper,we adopt the first-order integrated moving average(IMA(1,1))time series model to describe the disturbance model for industrial engineering process with drift disturbance,and put forward an approach to assess model quality based on the input and output data.Firstly,according to the feedback invariant principle of the disturbance in the control system,we can estimate the disturbance innovations using the routine closed-loop data,and this paper proposes a model quality index which is defined as the ratio of the quadratic form of estimated disturbance innovations and real residuals in the model to assess modeling quality.QR decomposition is applied to solve the problem that the estimated disturbance has a large deviation due to the high correlation of the closed-loop data.According to the Wood-Berry distillation column simulation case,we conduct three sets of simulation experiments which are based on the process model and disturbance model mismatch and different controller parameters to demonstrate the effectiveness of the proposed method.According to the Tennessee-Eastman industrial process,we conduct three groups of simulation experiments respectively by designing three different MPC controllers to demonstrate the effectiveness of the proposed method.The existing model quality monitoring methods are devoted to the monitoring of the overall model quality,and do not distinguish mismatch between the process model and the disturbance model.Aiming at this problem,this paper presents a new method for assessing the quality of process model based on input and output data of the system.Firstly,according to the feedback invariant principle of the disturbance in the control system,we can estimate the disturbance innovations by the external excitation and the routine closed-loop data,and this paper proposes a model quality index which is defined as the ratio of the quadratic form of estimated disturbance innovations and real model residuals to assess modeling quality.Combining the proposed index and the overall model quality index,the process model mismatch and the disturbance model mismatch can be separated successfully.According to the Wood-Berry distillation column simulation case,we conduct five sets of simulation experiments which are based on the process model and disturbance model mismatch and different controller parameters to demonstrate the effectiveness of the proposed method.According to the Tennessee-Eastman industrial process,four groups of simulation experiments are conducted respectively by designing four different MPC controllers to demonstrate the effectiveness of the proposed method.