遗传算法对SVR风速预测模型的多参数优化

实验分析发现:通过相空间重构法求出的嵌入维数E和时间延迟τ往往不是支持向量回归机(SVR)预测模型的最优参数。针对此问题,建立了一种基于遗传算法(GA)的多参数同步优化的SVR预测方法。利用GA方法对E、τ和SVR模型中的惩罚因子C、核函数宽度γ进行同步优化获得全局最优解,建立SVR风速预测模型。对比单纯优化C、γ的模型,以2组风速数据为例进行实验研究,建立的模型预测误差约为6.56%和7.74%。而对比模型的误差为12.00%和9.30%。这一结果表明,同时对E、τ、C、γ进行优化的模型相对于单纯优化C、γ的模型,预测精确度大大提高。

Windspeed prediction method based on SVR and multi-parameter optimization of GA

Phase space reconstruction was used to apply in the prediction of wind speed time series for characteristic factors extraction.After several experiments, embedding dimension E and time delay τ, which were typical parameters of phase space reconstruction, might not be the optimum values for support vector regression model.For solving this problem, a multi-parameter optimization method based on genetic algorithm was proposed to optimize embedding dimension E, time delay τ and other support vector regression model parameters (penalty parameter C, kernel function parameter γ) synchronously.Two groups of wind speed time series were predicted by using this method and the prediction errors are 6.56% and 7.74%.The errors of the contrast method (optimize C, γ only) are 12.00% and 9.30%.The results show that the optimal selection of E, τ, C, γ is necessary.Compared with the contrast model, this method can greatly improve the prediction accuracy.