基于动态数据及多因素相似日的日前电价预测方法研究  

Day-ahead Electricity Price Forecasting Based on Dynamic Data and Similar Day Approach with Multi-dimensional Impact Factors

作　　者：杨雪 黄思皖 史鉴恒 王宝岳 王凯 董世佛 李昊义 YANG Xue;HUANG Siwan;SHI Jianheng;WANG Baoyue;WANG Kai;DONG Shifo;LI Haoyi(Huaneng Clean Energy Research Institute,Changping 102209,Beijing,China;Huaneng Jiangxi Power Generation Co.,Ltd.,Nanchang 330000,Jiangxi,China;Huaneng Lancang River Energy Sales Co.,Ltd.,Kunming 650000,Yunnan,China;Huaneng Hebei Power Generation Co.,Ltd.,Shijiazhuang 050000,Hebei,China)

机构地区：[1]中国华能集团清洁能源技术研究院有限公司,北京102209 [2]华能江西分公司,江西南昌330000 [3]华能澜沧江能源销售有限公司,云南昆明650000 [4]华能河北分公司,河北石家庄050000

出　　处：《电力大数据》2024年第7期35-44,共10页Power Systems and Big Data

基　　金：华能集团总部科技项目(HNKJ21-H36)。

摘　　要：日前电价的准确预测对保障电力市场参与者的利益具有重要意义。该文提出了一种基于动态数据以及多维影响因素相似日的电力市场短期日前电价预测方法,旨在提高预测准确性和适应市场动态变化的能力。首先,通过综合考虑轻梯度提升机(light gradient boosting machine,LightGBM)的特征重要性和皮尔森相关系数,筛选出影响电力现货市场价格的关键特征,并构造延迟特征、历史相似日特征和综合加权特征作为衍生特征,以丰富电价预测的输入信息。其次,为了适应电力市场的动态变化并提高历史数据的适用性,引入了滚动训练机制,定期更新数据集并重新训练预测模型。采用广东电力现货市场的数据进行仿真分析,实验结果表明,通过合理的特征构造、超参数优化以及滚动训练策略的选择和应用,可以有效改善电价预测模型的预测性能。Accurate forecasting of day-ahead electricity price is of great significance to safeguard the interests of electricity market participants.This paper proposes a day-ahead electricity price forecasting method based on dynamic data and similar day approach with multi-dimensional impact factors,aiming to improve the prediction accuracy and the adaptability to market dynamics.Firstly,by comprehensively considering the feature importance of LightGBM and the Pearson correlation coefficient,key features that affect the price of electricity spot markets are screened out.Derived features such as delay features,historical similar day features,and comprehensive weighted features are constructed to enrich the input information for electricity price forecasting.Secondly,to adapt to the dynamic changes of the electricity market and improve the applicability of historical data,a rolling training mechanism is introduced to regularly update the dataset and retrain the forecasting model.Simulation analysis is carried out using data from Guangdong electricity spot market.The experimental results show that reasonable feature construction,hyperparameter optimization,and the selection and application of rolling training strategies can effectively improve the forecasting performance of the electricity price forecasting model.

关 键 词：电力市场 日前电价预测 特征构造 滚动训练 相似日 

分 类 号：TM73[电气工程—电力系统及自动化]