Nonlinear Adaptive Control Method Driven By RVFLN Data Model

The implementation of"Made in China 2025"has been announced China,leading to a great change in manufacturing production.The design of modern industrial control system is no longer limited to the derivation of mechanism equilibrium law to directly detect the inherent model of the mechanism parameters of the production process.How to effectively utilize a large data of online and off-line data and knowledge,which imply the information of operating condition variation and operation disturbance,to realize the optimal control of the production process and equipment with complex dynamic characteristics and strong non-linearity has become the research focus in the control field.With the rise of the machine learning and artificial intelligence technology,the data driven modeling technique,nonlinear system running indexes online estimation and adaptive control theory,the combination of related by direct use of controlled object offline and online data to describe the operation law of nonlinear system and correlation model,combined with reflection of system parameters,structure,such as data,achieve the function of nonlinear system prediction and control of expectations,is easy to implement and control the advantages of high accuracy and good steady-state performance.However,most of the existing adaptive control methods design the controller parameters based on the estimation results of pure nonlinear artificial neural network or the estimators identified alternately by projection algorithm and neural network.In this paper,the existing control methods are considered deeply,and the problems are proposed respectively:(1)In the complex nonlinear system,there is also a certain linear mapping,so the simple nonlinear mapping network is usually difficult to describe the complex relationship,which leads to poor generalization performance of the model and affects the control accuracy.(2)The method of alternating identification requires repeated iteration of the linear model and the estimation model of high-order nonlinear terms,which leads to slow modeling speed,unguaranteed overall convergence,and influences the control performance.This article based on alternating identification model based on adaptive control method,for a class of discrete time expressed as low order linear model under the combined model and unmodeled dynamics,and strengthen the structure based on the research RVFLN model adaptive control method,makes the adaptive control system in the model to estimate the phase more quickly and higher precision.The main work and innovation achievements are as follows:(1)For the identification of alternate solutions to provide a model adaptive controller is easy to calculate loss problem,this paper proposes a RVFLN estimation model based on the structure of straight chain and strengthen method of adaptive controller is designed,using RVFLN model of straight chain and strengthen the structure characteristics of low order linear model of nonlinear system with unmodeled dynamic item identification at the same time.In addition,in the process of model parameters are updated regularization item was introduced and the output weight deviation constraint fusion(L₂),improve the model to estimate the generalization performance and in the process of slowly time-varying environment performance,to ensure the convergence of the nonlinear system identification model,and the estimated results combined with a step ahead optimal control strategy,through contrast experiment,confirmed that the proposed algorithm has high precision of the model and the control system of fast response speed,short setting time,and the advantages of small steady-state error.It has laid the theoretical and application foundation for the following chapters.(2)Modeling method for traditional RVFLN tend to rely on experimental trial and error method to determine the enhanced node and enhanced because of the random nature of the random algorithm output nodes may be invalid or redundant nodes to a greater influence on the accuracy of problem,the adaptive control model to estimate link a supervised method to construct RVFLN model of the structure of the incremental model was achieved by historical input and output data structure and model parameters of data driven self-learning,and analyzes its convergence.At the same time,the bounded property of the linear model and the estimated value of the unmodeled dynamic term is proved in the case of unmodeled dynamic global bounded,which improves the stability and convergence analysis of the adaptive control system based on the RVFLN model.Through a series of simulation experiments,the superiority of the proposed estimation algorithm and the effectiveness of the control method based on the proposed estimation algorithm are verified.(3)A series of adaptive control methods based on RVFLN model proposed in this paper are applied to an industrial example of ash density control in heavy medium coal preparation.For validation of heavy medium coal preparation process,the author of this paper ash value change and give two kinds of working condition of coal dynamic change at the same time,tracking control and stability control of ash content in the design experiments,the proposed method and the traditional adaptive control method based on alternating identification model and the control method based on linear model,the experimental results verify the validity and superiority of the method.