基于人工神经网络和模糊推理的短期负荷预测方法

SHORT TERM LOAD FORECASTING USING ARTIFICIAL NEURON NETWORK AND FUZZY INFERENCE

电力系统短期负荷预测是电力系统调度部门制定发电运行计划的依据,也是制定电力市场交易计划的基础。人们提出了许多种短期负荷预测方法,如时间序列法、状态空间法、人工神经网络法等,但是这些方法都无法精确地描述电力负荷模型。在对大量历史负荷数据进行统计分析的基础上,作者提出了一种基于人工神经网络和模糊推理的预测新方法,其中首先根据实际经验将负荷日类型划分为周一、工作日、周六、周日和节假日5种类型;然后根据不同的类型日建立相应的人工神经网络模型用以预测负荷归一化系数;最后通过模糊推理策略预测日最大负荷和日最小负荷。实际算例表明,所提出的方法能够提高短期负荷预测的精度。

Short-term load forecasting of the future days or weeks is not only the basis for the scheduling of generating sets, but also the basis to work out the transaction schedule in electricity market. In the past three decades many forecasting models, such as the time series method, the state-space method and artificial neural network (ANN) method, are put forward, but all of these forecasting methods can not accurately describe the load model of power system. On the basis of statistics and analysis of a large amount of historical load data, a new load forecasting method based on fuzzy inference and ANN is proposed in which firstly according to the actual experience the daily load curves are classified as five kinds, i.e., the kinds of Monday, work day, Saturday, Sunday and festivals and holidays, then for different kinds of load curves the corresponding ANN models are established to forecast the normalization coefficients, finally the maximal and minimum loads in a day can be forecasted by fuzzy inference strategy. The results of practical calculation examples show that the accuracy of forecasted short-term load can be improved by the proposed method.