摘要： 针对用电负荷的周期性特点,将用电负荷特征学习建模为小时、天数、负荷数3个维度的回归问题,提出一种基于支持向量回归机的三维回归模型。将支持向量机的核函数设计为多个核函数的线性组合分别进行参数训练,并给出多路径逐步逼近的参数训练算法。仿真结果表明,与三层神经网络、最小二乘非线性拟合模型相比,该模型具有较好的用电负荷特征学习与预测能力。

关键词: 用电负荷特征, 支持向量回归机, 三维回归模型, 参数训练算法, 参数自适应

Abstract: According to the periodic characteristics of electric load,a Three Dimensional Regression Model studies are(TDRM) based on Support Vector Regression(SVM) is proposed.The power load characteristics studies are modeled as three dimensional regression problems,such as hours,days and loads.The kernel functions of Support Vector Machines(SVM) are designed as linear combinations of multiple kernel functions for separate parameter training,and the parameter training algorithm is designed to gradually approach the optimum from multiple different paths.Simulation results show that compared with the three layer neural network and the least square nonlinear fitting model,this model has better performance in power load characteristics learning and prediction.

Key words: power load characteristics, Support Vector Regression(SVR), Three Dimensional Regression Model(TDRM), parameter training algorithm, parameter adaptation

中图分类号: 

• TM732