BP神经网络在捷联惯导初始对准中的应用研究

提出了基于多层BP神经网络的滤波器，并用于捷联惯导初始对准中．采用BP网络替代初始对准系统中的闭环卡尔曼滤波器，可以确保系统的误差状态始终为小量，实现了惯导初始对准中的滤波与校正功能。采用BP神经网络滤波的优点是：数据并行计算速度快，在滤波时不需要初始数据。仿真结果表明，这种方法简化了系统运算的代数结构，提高了系统状态估值运算的实时性，并且可以保证系统的对准精度。

Application of BP neural network to alignment of SINS

A filter based on a multilayer BP neural network is used instead of the closed-loop Kalman filter for the initial alignment of strap-down INS,which can keep the error small and realize the functions of estimation and alignment in the INS. By comparing the results of adopting different hiding layers BP neural network, the use of double hiding layers is a better choice, and the variance plot is given for training simulation. Adopting BP neural network has lots of advantages, such as fast speed in parallel calculating for the real-time data and no initial data is required for the filter. It predigests the system's algebra structure and is more attractive for real time than classical filters. Simulation results show that its precision is similar to that of a Kalman filter.