基于切换模型极限学习机的超短期负荷预测  

Ultra-short term load forecasting based on switching model extreme learning machine

作　　者：邓明丽[1] 张晶[1] 

机构地区：[1]国网四川省电力公司技能培训中心,成都610065

出　　处：《电测与仪表》2015年第13期71-76,110,共7页Electrical Measurement & Instrumentation

摘　　要：针对极限学习机算法中输出波动大与模型不稳定的问题,提出采用切换模型极限学习算法进行超短期电力负荷预测的方法。该算法通过切换模型准则,将建立的多个神经网络模型分为误差较小的保持模型和误差较大的更新模型两部分。保持模型无需进行在线更新,减低了模型输出的波动性;更新模型则需采取随机方法进行在线更新,使得训练误差达到最小,提高模型的泛化能力。通过对某地区电力负荷的预测仿真,结果表明了所提方法提高了预测速度,节省了计算时间,具有更佳的泛化能力和预测精度。In view of the output fluctuation and model of instability in the extreme learning machine algorithm,this paper raises the method of switching model limit learning algorithm for ultra short term power load forecasting. The algorithm divides a plurality of the established neural network model into two parts： keeping model of small errors and updating model of large errors by switching model guidelines. To keep model has no need for online update,so as to reduce the volatility of the output of model; updating model needs to adopt stochastic methods to update online,so that the training error reaches a minimum,and the generalization ability of the model is improved. Finally,through the simulation of power load forecasting of a certain area,the predicted results show that the proposed method can improve the prediction speed,save computing time,and has better generalization ability and prediction accuracy.

关 键 词：极限学习机 切换模型 负荷预测 更新模型 预测精度 

分 类 号：TM76[电气工程—电力系统及自动化]