最小二乘支持向量机回归的卫星钟差非线性组合预报

Nonlinear combined prediction of satellite clock error based on least squares support vector machines regression

针对应用单一方法预报卫星钟差的局限性,文章提出了基于最小二乘支持向量机回归的卫星钟差非线性组合预报方法:首先根据历史钟差数据建立二次多项式模型和灰色模型,然后利用这些模型进行钟差预报,最后采用最小二乘支持向量机回归算法对两种模型的预报结果进行非线性组合,以获得最终预报值;对比了RBF核函数、线性核函数和多项式核函数对组合预报性能的影响,并将本文组合预报方法与经典权组合方法进行比较。结果表明,本文方法优于经典权法,且线性核函数更适合组合预报。

In view of the single approach limitations, an integrated method to combine the result of single models was proposed to predict the satellite clock error in the paper. This method employed least squares support vector machines（LSSVM）regression for combination of the quadratic polynomial model and the grey model in the nonlinear manner. Because a kernel function influences on performance of the LSSVM, the RBF function, the linear function and the polynomial function were employed to find out which kernel function is most suitable for hybrid prediction. In order to verify the feasibility and validity, the method was compared with the classical weighed algorithm. The result showed that the proposed meth- od would be better classical one, and the linear kernel function could be more suitable for the proposed method.