基于模糊聚类的多神经网络在短期电力负荷预测中的应用  

Multi-Neural Networks Based on Fuzzy Clustering for Short-Term Load Forecasting

作　　者：王立[1] 朱学峰[1] 

机构地区：[1]华南理工大学自动化学院,510640

出　　处：《电气应用》2007年第7期24-27,共4页Electrotechnical Application

摘　　要：根据城市短期电力负荷所呈现出的趋势性和一定的周期性,本文提出了模糊聚类和多个神经网络相结合的负荷预测方法,即首先根据一定长度时间段负荷的变化趋势相似性,将众多相同时间段长度的样本聚类,再用不同的神经网络对每个类别的数据样本进行学习,最后将待测数据所属时间段判定类别后用相应的神经网络进行预测。文中采用某市电力负荷实测数据进行了建模和计算,通过与普通的单一人工神经网络方法的比较,证明该方法具有预测平均绝对误差小、训练速度快、推广能力好的优点,有潜在的应用价值。According to the tendency and periodicity showing in the city short-term load, a forecasting method is proposed that combines the fuzzy clustering and multi-neural networks. Firstly, the samples of the same time period clustered is made, then, different networks to train the samples of different type separately is used, fi- nally, the load through the corresponding network with its type known is forecasted. With this model, the better accuracy is achieved in comparison with the single neural networks model. It is petentially worthy application.

关 键 词：模糊c均值聚类(FCM) 神经网络 负荷预测 

分 类 号：TM715[电气工程—电力系统及自动化]