一种基于遗传RBF神经网络的智能容错滤波算法

An Intelligent Fault Tolerant Filter Algorithm Based on RBF Neural Network Optimized by Genetic Algorithm

针对组合导航中标准卡尔曼滤波容错性能不足的问题,提出一种基于遗传RBF神经网络的智能容错滤波算法,其基本思想是通过RBF网络自动调节滤波增益来控制不确定性噪声的影响,进而提高滤波容错性。在RBF神经网络中,隐层单元与核函数宽度的选取对网络的性能具有重要影响,进而利用自适应遗传算法对其隐层单元数及核函数宽度进行了优化,隐层单元中心和输出层连接权值分别由K-均值聚类和最小二乘算法确定,最后得到精度较高且结构优化的RBF网络。为检验方法的应用效果,以SINS/GPS组合导航系统为例进行了仿真验证,实验结果表明遗传RBF网络容错滤波算法能在满足导航精度和计算量增加较小的前提下,比标准卡尔曼滤波具有更强的容错能力,由此也说明了方法的有效性。

Aiming at the shortage of standard kalman filter for fault tolerant in integrated navigation,an intelligent method is proposed based on RBF neural network optimized by genetic algorithm for fault tolerant.In this method,which the uncertain noise effect is controlled by adjusting the filter gain in real-time so that the fault tolerant performance is improved.The design of hidden units and width of kernel function is important for the RBF neural network,so the number of hidden units and the width of kernel function are optimized by using the adaptive genetic algorithm.The centers of hidden units are calculated by using the K-mean clustering algorithm and the weights of output layer are calculated by using the least square algorithm.At last,the network structure is optimized,and the algorithm has high accuracy too.In order to test the effect of this method,the simulation based on the SINS/GPS integration navigation system is demonstrated,and it indicates this proposed method has stronger fault tolerant ability than the standard kalman filter under the condition of satisfying the navigation accuracy and adding less calculation.This also proves the availability of the proposed method.