计及特征信息的自适应短期电价预测模型  

Adaptive Short Term Electricity Price Forecasting Model Considering Feature Information

作　　者：陈奕汝 何妍妍 赵志扬 高函 徐梦佳 郭烨烨 刘子毅 魏子琳 张金良[2] CHEN Yiru;HE Yanyan;ZHAO Zhiyang;GAO Han;XU Mengjia;GUO Yeye;LIU Ziyi;WEI Zilin;ZHANG Jinliang(Marketing Service Center,State Grid Zhejiang Electric Power Co.,Ltd.,Hangzhou 310000,Zhejiang Province,China;North China Electric Power University,Changping District,Beijing 102206,China)

机构地区：[1]国网浙江省电力有限公司营销服务中心,浙江省杭州市310000 [2]华北电力大学经济与管理学院,北京市昌平区102206

出　　处：《电力信息与通信技术》2025年第1期18-27,I0001,I0002,共12页Electric Power Information and Communication Technology

基　　金：中央高校基本科研业务费专项资金资助(2023FR001)。

摘　　要：随着我国新能源装机容量历史性超过火电,电价的多样化特征更加突出,如何提高电价预测精度对于电力市场参与者变得空前重要。为此,文章提出了一种考虑特征信息的自适应短期电价混合预测模型。首先,利用改进变分模态分解-样本熵(improved variational mode decomposition-sample entropy,IVMD-SE)的组合数据预处理模型实现电价序列的分解与重构。其次,对周期性特征较为明显的子序列采用季节性差分自回归滑动平均模型(seasonal autoregressive integrated moving average,SARIMA)进行预测,对非线性特征较明显的子序列使用鲸鱼优化的长短期记忆神经网络(whale optimization algorithm-long short-term memory,WOA-LSTM)进行预测。最后,将各分量预测值加总求和得到最终的电价预测结果。并依托澳大利亚电力市场数据设置多组对比实验,与基准模型相比,所提模型的误差指标在不同地区与不同季节的条件下均能达到最低。As the installed capacity of new energy in China has historically exceeded that of thermal power,the diversified characteristics of electricity prices have become more prominent,and how to improve the accuracy of electricity price forecasting has become unprecedentedly important for electricity market participants.To this end,this paper proposes an adaptive short-term electricity price hybrid forecasting model that considers feature information.Firstly,the decomposition and reconstruction of the electricity price series are achieved by using the combined data preprocessing model of"improved variational modal decomposition-sample entropy"(IVMD-SE).Secondly,the seasonal differential autoregressive sliding average model(SARIMA)is used to predict the more obvious periodic subseries,and the whale optimization algorithm long short-term memory neural network(WOA-LSTM)is used to predict the more obvious nonlinear subseries.Finally,the predicted values of each component are summed up to get the final electricity price forecasting result.The proposed model is used to set up multiple comparative experiments based on the Australian electricity market data,and compared with the baseline model,the proposed model can achieve the lowest error index in different regions and seasons.

关 键 词：电价预测 改进变分模态分解 长短期记忆网络 混合预测模型 特征信息 

分 类 号：TM715[电气工程—电力系统及自动化]