基于粒子群算法优化Elman神经网络的GPS快速钟差预报方法

Elman neural network based on particle swarm optimization for prediction of GPS rapid clock bias

为提升导航卫星原子钟快速钟差预报精度,提出了一种粒子群优化(PSO)算法改进的Elman神经网络钟差预报方法。首先,将动态递归类Elman神经网络引入钟差预报,通过PSO算法改进Elman神经网络的权值和阈值,以提升训练速度和预测精度;然后,将优化方法应用到钟差预报中,给出了利用该方法进行快速钟差预报的步骤;最后,与常用二次多项式模型(QPM)、灰色模型(GM)和超快速钟差产品IGU-P进行对比分析。结果表明:PSO-Elman模型对4颗不同类型的GPS卫星钟都能取得良好的预报精度与稳定度,其24 h平均预报精度和稳定度相对于QPM、GM、IGU-P产品分别提高了85.1%,74.2%,88.9%和71.3%,53.3%,28.3%。

To improve the prediction accuracy of the satellite rapid clock bias,a modified Elman neural network clock bias prediction method based on particle swarm optimization(PSO)algorithm was proposed.The Elman recurrent neural network was introduced to predict the clock bias,and its weights and thresholds were improved by PSO algorithm to improve the training speed and prediction accuracy.Then,the optimization method was applied to the rapid clock bias prediction,and the steps of using this method for the rapid clock bias prediction were given.Finally,the optimization method was compared with common quadratic polynomial model(QPM),gray model(GM)and ultra rapid clock bias product IGU-P.The results show that the PSO-Elman model achieves high accuracy and stability for four different types of GPS satellite clock,and its prediction accuracy and stability improved by 85.1%,74.2%,88.9%and 71.3%,53.3%,28.3%respectively,compared with QPM,GM and IGU-P products.