基于RBF神经网络的多系统钟差预报算法

针对海上条件下,对于实时定位应用,实时数据流无法下载的情况,文中提出一种基于RBF神经网络的卫星钟差预报算法,给出基函数的中心、方差以及隐含层到输出层的权值的计算方法,采用滑动窗口的方法,用样本数据训练后的网络预测下一个历元的钟差值,依次往后训练网络直到预测完整个时间段,通过实验验证了算法的可用性。短期预报中,GPS预报精度在1 ns以下,BDS和GLONASS在2～3 ns左右;长期预报中,GPS预报精度在几十纳秒左右,而BDS和GLONASS在几百纳秒左右,文中给出了相应的结果分析。

Satellite Clock Bias Prediction Algorithm with Multi System Based on RBF Neural Network

For real-time location applications, real-time data streams cannot be downloaded for maritime conditions. A satellite clock bias prediction algorithm based on RBF neural network is proposed. And then this paper gave the calculation of the center of the basis function, the variance and the weight of the hidden layer to the output layer. The sliding window method is used to prediction the clock bias of next epoch with the network trained by the sample data, and then train the network backwards until the whole time period is predicted. The availability of the algorithm is verified. In the short-term prediction, the GPS prediction accuracy is below 1 ns, BDS and GLONASS are around 2 ns to 3 ns; in the long-term prediction, the GPS prediction accuracy is about tens of nanoseconds, while the BDS and GLONASS are in the hundreds of nanoseconds. And the corresponding results analysis is given.