分区支持向量回归及其在负荷预测中的应用  

Partitioned Support Vector Regression and Its Application on Load Forecasting

作　　者：谷云东[1] 胡芳芳[1] 

机构地区：[1]华北电力大学数理学院,北京102206

出　　处：《电力系统及其自动化学报》2013年第3期20-24,共5页Proceedings of the CSU-EPSA

基　　金：国家自然科学基金项目(71171080);国家重点基础研究发展计划(973计划)项目(2009CB-320602);中央高校基本科研业务费专项资金资助项目(12MS84)

摘　　要：为解决局地短期电力负荷预测问题,给出了一种基于自适应聚类分区和支持向量机回归的多模型变结构负荷预测新方法。首先根据局地电力负荷的特点,使用模糊C均值聚类方法给出预测时间点的自适应分区;然后利用支持向量回归方法对不同分区分别构建预测子模型;最后提出一种基于仿真分析的模型评估和参数优化方法,在测试评估的基础上优化确定各子模型的参数,并适当调整各分区的结构。实验表明该预测方法比常用的神经网络预测方法具有更高的准确率和更强的鲁棒性。A multi-model based variable structure forecasting method for local short term electric load is proposed based on fuzzy clustering and support vector regression（SVR）. Firstly, the main idea of this method is introduced. Then, prediction time was partitioned into several sections adaptively by means of fuzzy C-means clustering, according to the characteristics of local power load pattern. Sub-models for each section were constructed based on support vector regression in the following. A simulation analysis based model evaluation and optimization method was proposed for the selection of model parameters and the regulation of section structure. The results of contrasting tests show that the pro- posed method has higher accuracy and stronger robust than neural network.

关 键 词：负荷预测 分区支持向量回归 模糊C均值聚类 自适应分区 变结构 模型评估 

分 类 号：TM744[电气工程—电力系统及自动化]