Abstract: In recent years, the SVM method in load forecasting application research has become the hot spot. This paper in view of the traditional standard support vector machine method in the prediction of time and prediction accuracy of deficiencies, firstly applies the MKL in power system short-term load forecasting field. The algorithm is realized through solving quadratic constrained programmer in the hybrid kernel space. Compared with the standard support vector regression algorithm, this method not only can improve the prediction performance, but also can reduce the number of support vectors. The practical example shows that, the method can effectively improve the prediction accuracy, shorten the prediction time, and with good generalization performance.

Key words: short-term load forecasting, multi-kernel learning, Support Vector Machines（SVM）, kernel function

摘要： 近年来，支持向量机（SVM）方法在电力系统负荷预测领域的应用研究成为了热点，鉴于传统的标准支持向量机方法在预测时间和预测精度方面的不足，首次将多重核支持向量回归方法（Multiple Kernel Learning，MKL）应用于电力系统短期负荷预测领域。通过在混合核空间求解二次约束下的二次规划问题实现多重核支持向量回归算法。该方法较标准的支持向量回归算法，不仅可以提高预测性能，而且能够减少支持向量的个数。实际算例表明，该方法能够有效地提高预测精度，缩短预测时间，具有良好的泛化性能。

关键词: 短期负荷预测, 多重核学习, 支持向量机, 核函数