大用户电力负荷的多模型模糊综合预测

Multi-model Fuzzy Synthesis Forecasting of Electric Power Loads for Larger Consumers

研究大用户的短期电力负荷预测问题,给出一种基于变权综合模糊推理的多模型综合预测方法。该方法首先引入基于质心相似度聚类的负荷模式分析算法,挖掘历史负荷数据中合群的典型负荷模式,并按相似性进行分组,同时剔除少量的离群异常记录;然后给出基于共轭梯度的RBF神经网络训练算法,分别对每类典型负荷模式建立相应的单元预测模型;最后利用基于相似度加权的多模型变权综合模糊推理策略,实现各单元模型预测结果的自适应融合。案例仿真验证了多模型模糊综合预测方法的可靠性。

A multi-model based variable weighted fuzzy synthesis forecasting method is proposed for the power load forecasting of large consumers. A clustering algorithm based on Renyi-entropy and centroid similarity is introduced to mining typical load patterns from historical load data and grouping them according to similarities as well as detecting atypical outliers. A conjugate gradient based learning algorithm for the RBF neural network is designed to construct unit forecasting model for each group of typical load patterns. Then,the forecasting results of all unit models are integrated adaptively by using variable weighted fuzzy synthesis inference. The simulation results show that the multi-model fuzzy synthesis forecasting method can raise the prediction accuracy and stability significantly.