摘要
为了准确预测不同用电设施的电力需求模式,设计了一种基于长短期记忆网络(LSTM)深度学习方法的自定义电力需求预测模型。利用混合数据抽样(MIDAS)方法对自回归分布滞后(ARDL)进行改进,提出了LSTM+MIDAS模型的电力需求预测方法。通过每天以5 min的频率收集住宅、市政厅、医院、工业等4类设施的用电数据,采用短期、长期、季节性预测3种方法进行了对比试验,通过测试验证了预测模型的误差率,分析了实际用电需求监测系统中电力模式的季节性波动规律,并预测各设施的用电需求。实验结果表明,利用所提出的LSTM+MIDAS模型进行电力需求预测的总体误差率均有所降低,并且可以有效检测电力需求波动性。
In order to accurately predict the power demand patterns of different power facilities,this paper designs a user-defined power demand forecasting model based on long-term and short-term memory network(LSTM)deep learning method.The mixed data sampling(MIDAS)method is used to improve the autoregressive distributed lag(ARDL),and a power demand forecasting method based on LSTM+MIDAS model is proposed.Through collecting the electricity consumption data of four kinds of facilities,such as residence,city hall,hospital,industry,etc.,with the frequency of five minutes every day,the comparative test is carried out by using three methods of short-term,long-term and seasonal prediction.The error rate of the prediction model is verified through the test,the seasonal fluctuation law of the power mode in the actual power demand monitoring system is analyzed,and the power demand of each facility is predicted.The experimental results show that the overall error rate of power demand forecasting using the proposed LSTM+MIDAS model is reduced,and the power demand fluctuation can be effectively detected.
作者
洪小林
Hong Xiaolin(College of Science,Hohai University,Nanjing 210098,China)
出处
《能源与环保》
2021年第6期229-235,共7页
CHINA ENERGY AND ENVIRONMENTAL PROTECTION