一种基于改进分类和回归树的神经元网络电力负荷预测方法  

Short-term Load Forecasting with ANN Based on Improved CART

作　　者：陈芳[1] 赵剑剑[1] 张步涵[1] 

机构地区：[1]华中科技大学电力系,湖北武汉430074

出　　处：《长沙电力学院学报（自然科学版）》2003年第3期34-37,共4页JOurnal of Changsha University of electric Power:Natural Science

摘　　要：通过改进CART(分类和回归树)分类法选择训练样本,可以降低与预测日不一致负荷模式的影响,提高预测精度,并运用人工神经元网络预测下一天的96点负荷.主要包括3个部分.首先,运用CART分类法将输入空间分成若干矩形互斥区域,每一个区域对应一种负荷模式;其次,根据分类结果选取神经元网络的训练样本.最后,合理映射天气因素和日期、星期类型并进行预测.实际应用表明本方法对于大波动负荷地区能够改善预测精度,提高预测速度.A method of choosing trainning samples through improved CART is presented. It can reduce the influence of learning data with different load patterns and improve the forecasting precision. Three sections are included. Firstly, it uses improved CART to divide the input space into some paralleling rectangle ones which corresponds to one kind of load patterns. Secondly, it chooses trainning samples according to the classified results. Finally, it maps the cement of climate and day types, week types reasonably and forecast the load of next day based on it. The practical application results show that this load forecasting method is effective. It can improve the forecasting precision and accelerate the trainning process, especially for the system whose load fluctuates within a large range.

关 键 词：电力系统 负荷预测 神经元网络 分类 回归树 负荷模式 预测精度 

分 类 号：TM715[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]