基于优化残差组合的北斗卫星短期钟差预报研究

Research on short-term clock bias prediction of BeiDou satellite based on optimized residual difference combination

为解决传统模型因使用卫星钟差一次差分序列而导致预报精度差的问题,进一步提升预报精度,提出一种优化残差组合对卫星钟差一次差分序列进行预报的方法.该方法首先根据北斗卫星钟差序列的特点,利用四分位法(IQR)代替中位数法对一次差分序列进行预处理,然后利用自回归滑动平均模型(ARMA)将经过预处理后的卫星钟差一次差分序列分成趋势项和残差随机项,接着利用极限学习机(ELM)模型对残差部分进行建模预测,最后将ARMA模型的预测结果和ELM神经网络的残差预测结果求和后进行差分还原.结果表明:当卫星钟差呈非线性时,组合模型的预报精度比传统模型提升了38.2%,在北斗卫星钟差短期预报中具有一定的可行性.

In order to solve the problem of poor prediction accuracy caused by the traditional model using the satellite clock bias primary difference sequence and further improve the prediction accuracy, an optimized residual combination is proposed to forecast the satellite clock bias primary difference sequence.This method firstly according to the characteristics of the beidou satellite clock bias sequence, using quarterback method instead of the median method of time difference sequence preprocessing, and then using autoregressive moving average(ARMA)model after preprocessing the satellite clock bias of a differential sequence is divided into trend item and random item residual, then using the extreme learning machine(ELM)model to simulate the residual part modeling prediction, Finally, the prediction results of ARMA model and residual prediction results of ELM neural network are summed and then differentially restored.The results show that when the satellite clock bias is nonlinear, the prediction accuracy of the combined model is 38.2% higher than that of the traditional model, which has certain feasibility in the short-term prediction of the BeiDou satellite clock bias.