基于主从神经网络的短期电力负荷预测研究  

Study on Short Term Power Load Forecasting Based on Master-Slave Neural Network

作　　者：陈展[1] 张利民[2] 赵海成 张晓华[1] 沈勇[1] 阮亮[4] 

机构地区：[1]广东电网公司佛山供电局,广东佛山528000 [2]国网河北省电力公司保定供电局,河北保定071000 [3]山西耀光煤电有限责任公司,山西晋中031100 [4]华北电力大学,北京102206

出　　处：《陕西电力》2014年第7期57-60,共4页Shanxi Electric Power

基　　金：国家自然科学基金项目资助(70901025)

摘　　要：传统的BP神经网络在对负荷样本进行训练时容易陷入局部最小点,致使算法不收敛,网络训练失败。主从神经网络基于Hopfield神经网络的良好动态特性,2个Hopfield主网络和1个BP从网络构成,可以有效改善BP神经网络训练时易陷入局部最小的问题。由此,提出了一种基于主从神经网络的短期电力负荷预测方法,用主从神经网络对短期负荷样本进行训练。通过对某地实际负荷的训练并与实际符合对比验证,表明该方法具有更快的收敛速度、更小的训练误差和很好的预测效果,可进一步提高短期负荷预测效率。Traditional BP neural network easily gets into part minimum,which leads to non convergence of algorithm and fail training. Based on the good dynamic evolution performance of Hopfield network,the master slave neural network is consisted with two Hopfield networks as master network and one BP network as slave network. It can solve the problem well. At last,the new neural network is applied to train the load sample and predict the power load. The result shows that the master slave neural network has a quicker asymptotic convergence rate and smaller network system errors,and predicts the short term power load well. Therefore, the new measure based on the master slave neural network can improve the prediction effect of short term power load better.

关 键 词：短期负荷预测 BP神经网络 HOPFIELD神经网络 主从神经网络 

分 类 号：TM715[电气工程—电力系统及自动化]