Research On State And Parameter Estimation Algorithm Based On Neural Network

Through decade development,due to outstanding learning and mapping capabilities,high fault tolerance,robustness,and information processing methods for parallel computing,neural networks have been widely used in many fields.This paper focuses on the two directions of state estimation and system identification,and aims at combining neural network to improve the estimation accuracy and convergence speed of the algorithm.As an important achievement in the development of neural networks,linear neural networks are coupled with Kalman filtering algorithms and applied to the parameter identification and state estimation of controlled state space models.From the simulation results,compared with conventional identification algorithms,it can be seen that the parameter convergence curve of the new algorithm is smoother,parameter estimates are closer to true values,and estimation errors are smaller.Multilayer neural network is the most widely used feed-forward network.Among them,BP algorithm is the most classic weight update algorithm,which is also called BP neural network.This paper mainly uses the nonlinear mapping ability and learning ability of BP network to make up for the shortage of filtering algorithm in a certain extent.At the same time,corresponding optimization algorithms are introduced to offset the defects of the neural network itself.And through multi-sensor linear discrete system and multi-sensor nonlinear discrete system,state fusion estimation,self-tuning observation fusion estimation and nonlinear observation fusion estimation based on neural network are realized respectively,and the effectiveness of the proposed algorithm can be verified from simulation examples.When the network topology structure is complicated,the cross-covariance matrices are not easy to be calculated.In order to improve the estimation accuracy of the Kalman consensus filtering algorithm,the SCI algorithm and the BCI algorithm are introduced as fusion criteria,and a CI fusion robust consensus filtering algorithm is proposed.Compared with Kalman consensus filtering algorithm,the estimation error of the filtering algorithm with fusion criterion is smaller.To improve the performance of the algorithm,starting from the basic framework,the feedback structure is introduced into the algorithm structure,that is,state fusion and error matrix updating are performed by using the state estimation and error matrix after fusion.The simulation results show that although the structure increases the calculation in a certain extent,the complexity of the algorithm still within an acceptable range.Compared with the fusion algorithm,the feedback structure can still further improve the algorithm.Compared with the BP neural network,which represents the forward neural network,the feedback neural network has stronger computing power and better stability,associative memory,and optimized computing function.Therefore,based on the CI fusion robust Kalman consistency algorithm,this paper introduces the Elman network to modify the filter values.Thereby,a CI fusion Kalman consensus algorithm based on Elman neural network is obtained.