Research On The Identification Method Of Hammerstein System With Time Delay Or Missing Data

The block-oriented Hammerstein system contains nonlinear static blocks followed by linear dynamic blocks.Because of the flexible choice of expressions for nonlinear and linear blocks,the block-oriented Hammerstein system has the characteristics of simple form and wide application.In this paper,the Hammerstein system is used as the mathematical model.A recursive least squares algorithm for parameter estimation of novel Hammerstein-QD nonlinear systems with time-delay is established for the nonlinear dynamic industrial process with time-delay.An auxiliary model based expectation maximization algorithm of the Hammerstein system with data loss is presented for the nonlinear dynamic industrial process with data loss.The specific research content of this paper is as follows:(1)According to the different linear blocks,the classification of Hammerstein nonlinear models is given.The least squares algorithm,radial basis neural network,expectation maximization algorithm and auxiliary model identification thoughts involved in this paper are discussed in detail,which paves the way for subsequent identification work.(2)A new Hammerstein-QD model is proposed for the nonlinear industrial process with time-delay.A time-delay radial basis function neural network is adopted as the nonlinear block in the Hammerstein-QD model.Further,the recursive least squares algorithm is investigated to estimate unknown parameters in the system.The simulation results validate the effectiveness of the proposed identification modeling method for the nonlinear time-delay system.(3)For the Hammerstein nonlinear system with output data loss,the Hammerstein system is reshaped into a bilinear identification system,and an auxiliary model based expectation maximization algorithm is established.The multiple parameter vectors in the system are estimated iteratively.This improvement avoids the limitation of the traditional expectation maximization algorithm that needs to convert the model to be identified into an autoregressive identification model with a single parameter vector.In the identification process,the parameterization of this kind of autoregressive identification model is usually difficult to achieve.A simulation example is employed to demonstrate the effectiveness of the proposed novel expectation maximization algorithm.