摘要
供电公司需要常态化地预测未来一个月供电区域内全行业高压用户售电量以及外贸企业用电量,目前电量预测虽然方法较多,但却普遍存在预测精度不能满足实际业务需求、稳定性较差的问题,为提高预测的准确度,本文首先对用于建模的电量历史数据进行了数据平滑,然后用实验的方法对长短期记忆神经网络的多个参数进行了调整,随后又提出了新的春节期间电量的预测方法。在用长短期记忆神经网络模型、整合移动平均自回归模型、先知模型对供电区域内全行业高压用户和所有外贸用户的用电量做一个月的短期预测后发现,经过细致调参和改制后的LSTM长短期记忆神经网络模型的预测效果较为满意,平均预测误差在三个模型中最小,而且平均误差正好在供电公司业务允许的范围之内。
The power supply company needs to regularly forecast the electricity sales of high-voltage customers and the electricity consumption of foreign trade enterprises within the power supply area in the coming month,in order to improve the accuracy of the prediction,firstly,the historical data of electricity quantity used for modeling are smoothed,and then the model of the model is built,the experimental method was then used to adjust several parameters of the Long-short Term Memory,followed by a new method of forecasting electricity during the spring festival.The findings are based on a one month short term forecast using the Long Short-Term Memory model,the integrated moving average autoregressive model,and the prophetic model for all high voltage users and all foreign trade users in the power supply area,after careful adjustment and modification,the prediction effect of the Long Short-Term Memory model is satisfactory,the average prediction error is the smallest among the three models,and the average error is just within the business permitted range of the power supply company.
作者
张永建
周晟
沈澄泓
周长星
ZHANG Yongjian;ZHOU Sheng;SHEN Chenghong;ZHOU Changxing(State Grid Shaoxing Power Supply Company,Shaoxing 312099 Zhejiang,China;Zhejiang Fanhai Zhixing Power Technology Co.,Ltd.,Hangzhou 310052 Zhejiang,China)
出处
《电力大数据》
2021年第8期9-18,共10页
Power Systems and Big Data
关键词
电量预测
长短期记忆神经网络
时间序列
整合移动平均自回归模型
先知模型
electricity forecast
long short-term memory networks
time series
autoregressive integrated moving average model
prophet
