Modeling And Control Method Of Multivariable Systems Based On Gaussian Process Model

Most modern industrial processes are multivariable systems,with characteristics of nonlinearity,strong coupling,and random uncertainty.For such systems,it is difficult to establish a precise mathematical model of the system and design the controller using traditional methods.With the rapid development of new sensors and computer technology,industrial processes collect and store a large amount of offline and online data.Therefore,it is of great significance to analyze and utilize the information inside the data to realize the modeling and control of multivariable systems.Based on the Gaussian process(GP)model,this paper proposes a new modeling method and control strategy for multivariable systems under the partial least square(PLS)framework.The main contents are as follows:Aiming at the nonlinear and strong coupling characteristics of multivariable systems,a GP-PLS model is established,and a PID control strategy in the latent variable space is proposed.First,the dynamic PLS external model is used to extract the feature information of input variables and output variables,which can eliminate the correlation between variables,reduce the dimension of variables,and obtain the feature matrix.In order to describe the nonlinear characteristics of multivariable systems and characterize the model uncertainty,GP model is used as the internal model of PLS ??to establish a dynamic GP-PLS prediction model.Then,based on the established model,the multiple input multiple output(MIMO)control problem in the original space is converted into multiple single input single output(SISO)control problems in the latent variable space.A simple PID controller is used to control multiple single loops to achieve multi-variable control with less calculation time.Finally,a numerical example and the continuous stirred tank reactor(CSTR)system are used to verify the simulation,which shows the feasibility and effectiveness of the proposed modeling and control method.Model predictive control(MPC)is a widely used multivariable feedback control technology.Based on the proposed GP-PLS model,the variance information provided by the GP model is used to design the model predictive controller in the latent variable space.And tracking control of the multivariablesystems can be eventually solved by reconstructing the obtained control law in the latent variable space back into the original space.Finally,the proposed control method is applied to the CSTR system to verify its effectiveness.The research in this paper not only enriches the control theory of multivariable systems,but also can be generalized and applied to actual industrial processes.It has certain theoretical significance and application value.