基于增量余弦RBF网络的惯性导航初始对准

针对传统高斯RBF网络应用于惯性导航初始对准建模时，对处于两基函数中心点之间的值拟合效果不太理想的情况，提出了一种将增量余弦RBF网络用于惯性导航初始对准建模的方法。该方法采用增量余弦函数作为RBF网络的基函数，对多变量非线性系统有很好的拟合能力。相对于传统高斯RBF网络，增量余弦RBF网络的基函数具有更强的局域性，解算时同时参与运算的基函数数量更少，有效地降低了网络的解算时间。仿真结果表明，增量余弦RBF网络用于惯性导航的初始对准，既可获得较高的对准精度，又有效地降低了系统的解算时间，提高了系统的实时性。

Initial Alignment of Inertial Navigation System Based on the Raised-Cosine RBF Neural Network

When applying the traditional Gauss RBF neural network to the modeling of initial alignment of inertial navigation systems, it is usually not ideal to approximate the value in the midway between two grid points. In order to solve this problem, a method for modeling the initial alignment of inertial navigation systems using raised-cosine RBF neural network was presented. A raised-cosine function is used as the radial basis function in the RBF neural network. It has good ability to approximate multivariable nonlinear system. Compared to traditional gauss RBF neural networks, the radial basis function of raised-cosine RBF neural network has more compact form. When calculation the output, it needs less radial basis functions at the time, thus reducing the calculate time effectively. The simulation results show that using raised-cosine RBF neural network for initial alignment in inertial navigation systems can not only obtain relatively high alignment accuracy, but also reduce the system calculating time.