改进BP网络的海底地形辅助导航算法

Underwater terrain-aided navigation algorithm based on improved BP networks

鉴于传统的迭代最近点算法存在着易陷入局部最优的缺陷和实时性不好的问题,提出了一种将BP神经网络引入迭代最近点算法中进行地形匹配的新方法。针对传统BP算法存在的局部极小和收敛速度慢等缺点,采用自适应学习方法、引入动量因子、可变化的学习率因子和可调激活函数等措施进行了BP算法的改进。仿真结果表明,改进后的算法可以在一定程度上克服由于局部收敛带来的匹配失效问题,能够获得很好的匹配效果,同时也解决了在实时性上存在的突出问题。

Terrain,aided navigation is one of the developing trends for the underwater vehicle＇s navigation technique, As the traditional ICP algorithm is liable to get local minimization problem and have a bad performance of real-time, a BP neural network was presented in the ICP algorithm. In view of the drawbacks of local minimization problem and slow rapidity of convergence, several improved ways were put forward. The improvement includes adaptive learning method, momentum, and variable learning rate. The simulation results indicate that the algorithm can improve the results of terrain matching, and significantly overcome the problem of matching failure caused by local convergence. It also solves the serious problem of real-time.