基于粗糙集理论和启发式径向基函数神经网络的中长期电力负荷预测模型  

A model for medium and long term load forecasting based on rough set theory and heuristic radial basic function neural network

作　　者：夏非[1] 范莉[1] 苏浩益 

机构地区：[1]南昌工程学院,江西南昌330099 [2]湘潭电业局,湖南湘潭411104

出　　处：《电力系统保护与控制》2012年第16期21-26,共6页Power System Protection and Control

摘　　要：针对中长期负荷预测的特点,提出了基于当年负荷总量、当年负荷增长量、前年负荷总量、前年负荷增长量和年国内生产总值等多个指标的启发式径向基函数神经网络中长期负荷预测模型。通过正交信号修正法处理原始数据,利用偏最小二乘法拟合出单位国内生产总值电耗。以单位国内生产总值电耗作为启发式算子,在当年历史负荷数据的基础上合理假设待预测年的负荷总量,利用启发式算子反推出该负荷值对应的年国内生产总值,结合其余指标量构建神经网络扩展训练样本训练神经网络,创造性地将神经网络外推预测转化为内插求值。利用粗糙集理论对启发式神经网络模型的预测值进行修正,使预测精度进一步提高。实际算例的结果表明,所提出的方法预测精度较高,具有较强的可行性和实用性。According to the characteristics of the medium and long term load forecasting, a heuristic radial basic function （RBF） neural network load forecasting model is proposed based on multiple targets such as the load gross quantity in this year, the load growth quantity in this year, the load gross quantity in prior year, the load growth quantity in prior year, and the gross domestic product （GDP）. Firstly, we use the orthogonal signal correction to process primary data, and use the partial least square method to fit the unit of GDP electricity consumption. Then taking unit of GDP electricity consumption as heuristic operator, we assume the load gross quantity of next year reasonably based on original load data, and use the heuristic operator to get the corresponding GDP. Combining with the other indexes, we construct neural network extended training samples, and transform the neural network extrapolation forecast into the interpolation evaluation creatively. Finally, the predicted value of heuristic neural network forecasting model is corrected on the basis of rough set theory to further improve the forecasted result. The results of practical examples show that the proposed method has higher forecast precision and the model has feasibility and validity.

关 键 词：负荷预测 径向基函数神经网络 正交信号修正法 属性约简 粗糙集理论 启发式算子 

分 类 号：TM715[电气工程—电力系统及自动化]