Identification And Control Of Block-oriented Model Based On Extreme Learning Machine

Most of real systems are nonlinear system. Therefore, building a mathematical model for a practical system is the foundation of controller design, optimization and prediction,etc. Recently, researchers have paid attention to block-oriented model which is one of the effective models for dealing with nonlinear systems. The block-oriented model is composed of a linear dynamic subsystem and a static nonlinear function. According to the order of connection, block-oriented models mainly conclude two typical model, i.e,Hammerstein model and Wiener model. In this paper, extreme learning machine(ELM) is proposed to describe the static nonlinear part. In this thesis, the model identification and predictive control of nonlinear system are studied, and the main work includes:Firstly, we propose a new Hammerstein model based on ELM, called ELM-Hammerstein model, where the static nonlinear function is characterized by ELM neural network, while the linear dynamic subsystem is represented by autoregressive model with e Xogeneous(ARX) model. An online identification method for Hammerstein model is proposed. To improve the convergence rate and stability of recursive least square(RLS), a novel changing forgetting factor recursive least square(CFF-RLS)algorithm is proposed to estimate the parameters of ELM-Hammerstein model.Secondly, we put forward a novel Wiener model, called Laguerre-ELM model. In Laguerre-ELM model, the static nonlinearity of Wiener model is characterized by ELM neural network, and the linear dynamic part is represented by Laguerre filters. Laguerre filters can approximate a stable linear dynamical system to any degree of accuracy. In addition, Lipschitz quotients criterion is adopted to determine the appropriate number of Laguerre filters. Comparison with other methods are made in simulation, the experimental results show the effectiveness of the proposed method.Finally, a predictive control method is proposed according to identified Laguerre-ELM Wiener. The philosophy of our control approach is to transfer a complex nonlinear control problem to a simple linear one by introduce an inverse model of static nonlinearity. In simulation, the proposed method is applied to the identification and control of fluid flow control system. Results demonstrate that the proposed method can track expected output, and outperforms traditional PID control method.