基于CNN-BiLSTM的电力负荷中短期预测  

Short-term Electrical Load Forecasting Based on CNN-BiLSTM Model

作　　者：庄依洁 刘景豪 李盈 ZHUANG Yijie;LIU Jinghao;LI Ying(Department of Statistics,School of Economics,Jinan University,Guangzhou,Guangdong 510632,China)

机构地区：[1]暨南大学经济学院,广东广州510632

出　　处：《数学建模及其应用》2022年第4期62-70,共9页Mathematical Modeling and Its Applications

摘　　要：首先对某地区电力负荷数据进行重复值、缺失值和异常值的处理,再进行特征工程对特征进行挖掘,并基于Copula函数进行特征筛选,接着基于深度学习理论建立了基于CNN-BiLSTM的多变量分时负荷预测模型,通过模型融合进行了误差修正;然后对各行业日负荷最值序列进行突变点检测和分析,基于突变点分别对各行业建立了基于Prophet时间序列分解方法的日负荷最值预测模型;最后通过模型准确度评估验证了模型的有效性,结果表明融合模型能有效地应用于实际的电力系统负荷预测中.In this paper,repeated values,missing values and outliers are firstly processed for load data in a certain area,and then feature engineering is carried out to mine the features,and feature selection is carried out based on Copula function.Following a multi-variable load prediction model based on CNN-BiLSTM is established based on deep learning theory,and the error is corrected by model fusion.Then,the mutation point detection and analysis are carried out on the daily maximum load sequence of each industry,and the daily maximum load prediction model based on Prophet algorithm is established for each industry.Finally,the validity of the model is verified by model evaluation,and the proposed fusion model can be effectively applied to the actual electrical load prediction.

关 键 词：特征筛选 CNN-BiLSTM模型 模型融合 突变点检测 Prophet模型 

分 类 号：O29[理学—应用数学]