一种遗传算法优化的卫星钟差预报

Prediction of satellite clock errors based on genetic algorithm optimized

针对导航卫星钟差预报精度不高的问题,该文引入了GM-LSSVM钟差预报模型,采用全局寻优能力较强的遗传算法对模型的参数选取过程进行优化,避免模型陷入局部最优,从而改善了组合模型中惩罚因子和核函数参数选择的盲目性。最后选取国际GPS服务组织提供的卫星钟差数据,分别建立GM(1,1)模型、LSSVM模型、GM-LSSVM模型和遗传算法优化的GM-LSSVM模型进行短期钟差预报分析和仿真实验。仿真结果表明,优化后的模型预报精度小于1.3ns,精度比前3种模型提高了45%~60%,符合钟差预报的要求。

Aiming at the low level of the performance of navigation satellite clock error＇s predication, a method of grey least square support vector machine was proposed. Besides, the genetic algorithm which has a good ability of global optimization was used to optimize the penalty and kernel bandwidth parameter of new model to prevent the model from sinking into local optimal. Lastly, four models were established using the clock data from IGS to forecast short term clock error. The results showed that the accuracy of new model was superior to other models, the error of new model was less than 1.3 ns and is the 45%- 60% of the other models. And the performance indicators meet the requirements of satellite clock error＇s short term prediction.