基于相似日筛选与组合深度学习模型的日前电价预测方法  

Day-ahead Electricity Price Prediction Method Based on Similar Day Screening and Combined Deep Learning Modeling

作　　者：艾雨 贾燕冰[1,2] 韩肖清 AI Yu;JIA Yanbing;HAN Xiaoqing(Shanxi Key Laboratory of Power System Operation and Control(Taiyuan University of Technology),Taiyuan 030024,Shanxi Province,China;Key Laboratory of Cleaner Intelligent Control on Coal&Electricity(Taiyuan University of Technology),Ministry of Education,Taiyuan 030024,Shanxi Province,China)

机构地区：[1]电力系统运行与控制山西省重点实验室(太原理工大学),山西省太原市030024 [2]煤电清洁控制教育部重点实验室(太原理工大学),山西省太原市030024

出　　处：《电网技术》2025年第1期242-251,I0088,共11页Power System Technology

基　　金：国家自然科学基金重点项目(U1910216);国家自然科学基金青年基金项目(51807129)。

摘　　要：准确的日前电价预测是市场运行和政策规划的基础,而市场披露信息是电价预测的重要依据。提出了引入Self-attention机制的CNN-GRU组合深度学习电价预测模型。首先,针对山西电力现货市场交易流程及日前电价形成机制,采用最大互信息系数法对市场披露的日前边界条件等信息数据进行特征提取,以确定电价关键影响因素及其权重系数。其次,基于加权灰色关联度的历史相似日筛选方法生成电价预测历史数据集,并挖掘电价及其特征的内部变化规律。然后,基于历史数据集,采用引入Self-attention机制的CNN-GRU模型得到预测电价。最后,通过算例验证了所提预测方法的有效性及准确性。Accurate day-ahead electricity price prediction is the basis for market operation and policy planning,while market disclosure information is essential for electricity price prediction.This paper proposes a combined CNN-GRU deep-learning electricity price prediction model introducing the Self-attention mechanism.Firstly,for the trading process of the Shanxi electricity spot market and the formation mechanism of electricity price before the day,it is proposed to adopt the maximum mutual information coefficient method to extract features from the market disclosed information data such as boundary conditions before the day,to determine the key influencing factors of electricity price and its weight coefficients;and then based on the weighted grey correlation of the history of similar day screening method to generate the historical dataset of electricity price prediction,and to excavate the internal change rule of electricity price and its features;Finally,the CNN-GRU model introducing Self-attention mechanism is used to get the predicted electricity price based on the historical data set.The effectiveness and accuracy of the prediction method proposed in this paper are verified through examples.

关 键 词：日前电价预测 边界条件 最大互信息系数 相似日筛选 Self-attention机制 

分 类 号：TM73[电气工程—电力系统及自动化]