负荷预测中多模型的自动筛选方法  

MULTI-MODEL AUTOMATIC SIFTING METHODOLOGY IN LOAD FORECASTING

作　　者：高峰[1] 康重庆[1] 夏清[1] 黄永皓[1] 尚金成[2] 孟远景[2] 何南强[2] 

机构地区：[1]清华大学电机系,北京市100084 [2]河南省电力公司,河南省郑州市450052

出　　处：《电力系统自动化》2004年第6期11-13,40,共4页Automation of Electric Power Systems

基　　金：国家重点基础研究专项经费资助项目(G1998020311);清华大学基础研究基金资助项目(JC2002018)

摘　　要：预测模型的多样性一直是负荷预测中特别强调的一个问题。然而,对于多种预测模型,如何在舍弃效果较差的模型的同时选择出比较有效的模型,从而获得更加准确的预测结果,始终是一个难点。文中提出了一种新颖的多模型自动筛选算法,它应用odds-matrix方法来定量确定每种单一模型的权重,每种方法的权重代表了该方法的优劣性,通过权重分布函数判断各个预测模型的显著性。算例表明这种思路可以得到令人满意的预测结果。The diversity of models is an important issue in load forecasting. To improve the precision for forecasting, it is necessary to distinguish between better models and bad ones. But this task is very difficult. This paper proposes a novel multi-model automatic sifting methodology to solve this problem. The odds-matrix method of the new algorithm is used to calculate the weight of each model, which reflects the 'optimality' of an individual forecasting model. Thus, the efficiency of each model can be differentiated via evaluating probability distribution function of the weights. Numerical studies show that this method is satisfactory in improving forecasting precision.

关 键 词：负荷预测 odds-matrix 综合预测模型 自动筛选 

分 类 号：TM715[电气工程—电力系统及自动化]