A Fusion Estimation Algorithm Based on Machine Learning for Nonlinear Systems

This paper is concerned with distributed estimation problem for multi-sensor nonlinear systems. Based on the approximation performance of the BP network, we uses local CKF estimations as inputs and estimations of a distributed fusion CKF weighted by matrices (MW-CKF) algorithm as the target outputs to train the network, a BP net-based MW-CKF algorithm is proposed, which realizes the fusion effect of the MW-CKF algorithm. Compared with the MW-CKF, the proposed method reduces the computational cost of calculating cross-covariance matrices, and the theoretical equivalence of the machine learning method is proved. A target tracking process in a horizontal plane simulation verifies the effectiveness of the proposed fusion algorithm.