基于梯度提升树的写字楼月度用电量预测研究  

Research on Monthly Electricity Consumption Forecasting for Office Building Based on GBDT

作　　者：王鸿斌 武进军 吴旭 陈长清[2] 陈鹏远 张天安 WANG Hongbin;WU Jinjun;WU Xu;CHEN Changqing;CHEN Pengyuan;ZHANG Tian’an(Tianjin Anjie IOT Science and Technology Co.,Ltd.,Tianjin 300384,China;School of Software,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]天津安捷物联科技股份有限公司,天津300384 [2]华中科技大学软件学院,湖北武汉430074

出　　处：《电力科学与工程》2021年第4期30-36,共7页Electric Power Science and Engineering

摘　　要：以气象和时间作为影响写字楼用电量的主要因素,首先使用网络爬虫抓取到气象数据,包括最高温、最低温、风力、湿度、气压、天气和能见度等,同时提取日期相关的时间特征,如星期、节假日、第几周、季节和小时,再使用递归特征消除法进行特征选择,除去特征重要度低的因素,最后在模型选择上,与梯度提升树模型、差分整合移动平均自回归模型ARIMA、神经网络模型LSTM进行效果对比。实验证明,梯度提升树算法在写字楼月度用电量预测中效果最佳。Weather and time are taken as the main factors that affect the electricity consumption of office building.First,the meteorological data captured by web crawlers,including the highest temperature,lowest temperature,wind,humidity,pressure,weather and visibility.Meanwhile,the time features related to the date are extracted,such as week,holiday,week,season,and hour.Then recursive feature elimination method is used for feature selection to remove features with low feature importance.In the model selection,the gradient boosting decision tree(GBDT),the autoregressive integrated moving average model(ARIMA),and the long-short-term memory network(LSTM)are used to compare the effects.Experiments have proved that the gradient boosting decision tree algorithm can better meet the requirement of monthly electricity consumption forecasting for office building.

关 键 词：用电量预测 梯度提升树 ARIMA模型 神经网络 

分 类 号：TM734[电气工程—电力系统及自动化]