A new GA-based RBF neural network with optimal selection clustering algorithm for SINS fault diagnosis

Abstract

In this paper, a new adaptive genetic algorithm (GA)-based radial basis function (RBF) neural network with optimal selection clustering algorithm (OSCA) is proposed for the fault diagnosis of micro electro-mechanical system (MEMS) gyroscopes and accelerometers of strapdown inertial navigation system (SINS). The number of hidden layer nodes and parameters of RBF neural network are obtained by using OSCA. The connection weights are encoded to generate the chromosome, which is operated by adaptive GA. Orthogonal least square algorithm (OLS) is used to train the weights and gradient descent algorithm (GDA) with momentum term is used to estimate the parameters of Gaussian function. Adaptive GA, OLS and GDA with momentum term iterate alternately. Experimental results show that the proposed GA-based RBF neural network with OSCA quickly converges and effectively improves the diagnostic accuracy rate of SINS fault diagnosis.