改进神经网络辅助的GPS/INS组合导航算法

Improved neural network aided GPS/INS integrated navigation algorithm

针对全球定位系统(GPS)和惯性导航系统(INS)组合系统中GPS中断时,导航性能会急剧降低的情况,提出了一种改进径向基神经网络结合自适应滤波辅助的组合系统导航算法.该算法探讨了遗传算法参数寻优和最近邻聚类学习算法,解决了径向基神经网络训练中参数合理选取的问题,构建了INS加速度增量、姿态增量与GPS位移增量之间的非线性映射模型.当GPS出现故障时,利用该映射模型和改进的自适应滤波实时预测出GPS伪位置与其对应的协方差,进而计算出预测位置辅助的导航解,利用实测数据对算法进行验证.结果表明:GPS发生故障情况下,改进径向基神经网络算法能够辅助组合系统解算出稳定的次优导航解,其精度明显优于纯INS导航.

Aiming at that the navigation performance of GPS/INS integrated system will be drastically reduced when GPS outages, this paper proposes an algorithm to address the problem based on improved Radial Basis Function Neural Networks （RBFNN） and adaptive filter. With the nearest neighbor clustering algorithm and genetic algorithm parameter optimization, the optimal parameter of RBF neural network training is selected to build a nonlinear mapping model between the increments of INS acceleration, attitude and the increments of GPS displacement. Real-time GPS pseudo-position is predicted with the mapping function and the corresponding covariance matrix is estimated by an improved adaptive filter. Then the aided navi- gation solution can be calculated. Finally, the field data were collected to validate the algorithm, the results show that the improved RBFNN algorithm can provide a relatively stable navigation solution once GPS is blocked, the accuracy of navigation is obviously better than that of INS-only.