基于微粒群算法的LS-SVM时间序列预测

Time Series Forecasting via Least Squares Support Vector Machine Based on Particle Swarm Optimization

将微粒群算法引入到最小二乘支持向量机(LS-SVM)时间序列预测,建立预测模型。思路来自微粒群算法可以在超平面空间中实现优化搜索,因此,将微粒群算法中的微粒运动公式进行修正,从而实现对LS-SVM的训练。然后用训练过的LS-SVM进行预测,即得到最终结果。将此方法应用于销售量预测,结果表明,此模型有更高的预测精度,而且在不同的LS-SVM学习参数下模型的误差相对稳定。

The Particle Swarm Opimization（PSO） are introduced to the time series forecasting method based on Least Squares Support Vector Machine （LS- SVM）, then the forecasting model is established. The feature that PSO can optimize the procedure of searching in hyperplane space inspires the mind. By modifying the velocity equation of particle in original PSO, the training of LS - SVM is realized. Then forecasting result is obtained by using LS - SVM which is trained by PSO. This method is applied in the sales volume prediction. The result indicates that the forecasting model has higher forecasting accuracy and steadier forecasting error at different LS- SVM parameters.