Study On Application Of Kalman Filtering Algorithm Aidded By Neural Network

The famous Kalman filtering method which rooted in methods of liner state space of dynamical system, it provides a recursive solution for problem of linear filtering. However, its inherent shortcomings limited its practical application, course the Kalman filtering method asks for specific theoretical model and known statistical characteristics of noise, and it has bad filtering accuracy in some case, or it could be diverged, too large amount of calculation and disaster of dimensions also happened. As to Artificial Neural Network, too much dependence on samples, less universality, much difficulty in the actual application and over-fitting problem would make it has many unavoidable defects. So this two algorithms of Neural Network and Kalman filter are highly complementary to each other.This paper aim to combine the two useful methods, the performance of the combined algorithm is better than that of any single method, and this combination has broaden their scope of application. Firstly, the Kalman filtering method and its derivatives in nonlinear conditions are studied. Then, the design and optimization process of BP neural network are described in detail, and the simulation results are verified by Matlab simulation. Some parameters which influence the accuracy of Kalman filter are proposed as the input of BP Neural Network for learning and training, then the trained network is used to correct the Kalman filtering process. And at last, the performance of the proposed algorithm is verified by simulation.The algorithm proposed by this paper has better performance on stability and accuracy of filtering, which are verified through the simulation, this improvement meet the design requirements and basically to achieve the desired results.