Particle Filtering Based Identification Algorithm For Systems With Output-error Type Model Structures

The output-error type system can be divided into linear output error system and nonlinear output error system.Because of its simple structure and flexible characteristics,it has been widely concerned and applied in engineering practice.The recursive least square algorithm and the stochastic gradient algorithm are the classical identification algorithms to solve the linear output error system.The recursive least square algorithm has the advantages of high accuracy and fast convergence,and the stochastic gradient algorithm has a small amount of calculation.The research of this paper mainly focuses on the linear output error system,the nonlinear output error system and the quantized linear output error system.The main work of this article is as follows:For the linear output error system,combined with the characteristics of small calculation,slow convergence and low accuracy of the stochastic gradient algorithm,an auxiliary model based weighted multi-innovative stochastic gradient algorithm is proposed by introducing a weighted idea and multi-innovation identification method.Utilizing the idea of weight,designing a more rational way to use innovation can make more reasonable use of existing information.Through the analysis of algorithm performance,the feasibility of the algorithm is theoretically verified.For the nonlinear output error system,combined with particle filtering technology and multiinnovation identification method,a particle filtering based recursive least square algorithm and a particle filtering based multi-innovation recursive least square algorithm are proposed.In the identification algorithm of the output-error system,the process output of the output-error system is usually replaced by the output of the auxiliary model,but the auxiliary model is susceptible to the presence or absence of process noise and whether the system output is linearly affected,which has certain limitations.Therefore,the particle filtering technology is introduced to further improve the performance of the algorithm,and the multi-innovation identification method is introduced to further accelerate the convergence speed.Through the analysis of algorithm performance,the feasibility of the algorithm is theoretically verified.For the quantized linear output error system,by introducing recursive prediction error method and particle filtering technology,the standard auxiliary model based on stochastic gradient algorithm is modified to present the auxiliary model based recursive prediction error stochastic gradient algorithm and the particle filtering based recursive prediction error stochastic gradient algorithm.Through the simulation experiment,the recursive prediction error method can effectively process the quantization information,and the particle filter technology can further improve the identification performance of the proposed algorithm.