Abstract: Multicollinearity among variables can be eliminated by extracting principal components based on PLS.Nonlinear relationship among variables can be approximated by LSSVM.Combination of PLS and LSSVM is used to predict the annual electricity consumption of Sichuan province from the year of 1978 to 2007.The results show that PLS-LSSVM has higher accuracy than PLS and LSSVM.

Key words: partial least square, support vector machine, annual electricity consumption

摘要： 偏最小二乘法通过提取主成分能有效地消除变量间的多重共线性，最小二乘支持向量机能很好地逼近变量间的非线性关系。偏最小二乘与最小二乘支持向量机相结合用于年用电量的预测，充分发挥了两者的优点。对四川省1978～2007年的用电量进行了实证分析，与PLS模型和LSSV模型的预测成果进行了对比，结果表明年用电量预测的PLS-LSSVM模型有较高的精度。

关键词: 偏最小二乘, 支持向量机, 年用电量