Offline And Online Identification Of Nonlinear Dynamic Process Based On Simplified Wiener Structure

In the process industry,the modeling of chemical processes involves the inherent complexities of nonlinear characteristics,dynamic characteristics,and uncertainty of objects.Among them,Wiener structure,as a typical representative of modular nonlinear structure,has strong generalization ability.Therefore,it is of great practical significance to study the processes modeling and identification based on the Wiener structure.Usually,a linear dynamic block is connected in series with a nonlinear static block in the Wiener structure.This structure can describe the nonlinear dynamic characteristics of the processes completely and fully.In addition,when the nonlinearity of the actual processes is approximately monotonic(such as dead zone,saturation,and monotonic distortion within the processes),this paper further transforms the Wiener structure into a simplified Wiener structure,namely SWS,by constructing the inverse function form of the intermediate variable and the output variable.It can effectively describe the single-input-single-output(SISO)processes with approximately monotonic nonlinearity.However,the environmental noise in the actual production processes is very complex,and the error term in SWS considers the influence of environmental noise and model error at the same time,this paper needs to develop offline and online identification methods suitable for SWS under the complex interference influence.Since the Gaussian mixture model(GMM)can describe the noise terms,error terms and other data items with complex distributions,this paper introduces the expectation maximization(EM)algorithm to identify the parameters of the SWS model offline based on GMM.This offline identification algorithm based on GMM-EM has the ability to deal with complex noise,while ensuring the robustness of model parameters estimation.Moreover,this paper also studies the switching point detection and online identification method for the switching system with Wiener structure.Among them,each sub-mode of the switching system can be modeled by using SWS.The online recursive identification algorithm adopts a multi-innovative design method.On this basis,by using three identifiers with different innovation lengths,the switching point detection of the switching system is effectively realized,and the online identification algorithm can quickly converge the new model parameters to near the true value after switching.Finally,several typical simulation experiments are designed in this paper.In the experiment,several typical dynamic processes with approximately monotonic nonlinearity are selected for simulation,and these processes are modeled by SWS,then the SWS model parameters are identified by the proposed offline and online identification algorithms.The validity and reliability of the simplified Wiener structure and the proposed identification algorithms are verified by testing the predictive output ability of the model.