考虑高比例风电波动的多注意力TCN电价预测方法  

A multi-attention TCN based electricity price forecasting method considering high proportion of wind power fluctuation

作　　者：李子凯 杨波 周忠堂 李新 陈凤伟 焦润海[2] LI Zikai;YANG Bo;ZHOU Zhongtang;LI Xin;CHEN Fengwei;JIAO Runhai(Linyi Power Supply Company,State Grid Shandong Electric Power Company,Linyi 276002,Shandong,China;School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China)

机构地区：[1]国网山东省电力公司临沂供电公司,山东临沂276002 [2]华北电力大学控制与计算机工程学院,北京102206

出　　处：《电测与仪表》2025年第3期138-146,共9页Electrical Measurement & Instrumentation

基　　金：国网山东省电力公司科技项目(520607210007)。

摘　　要：在风电占比较高的电力系统中,风电的强波动性使得短期内的供需形势发生剧烈变化,增加了电价的不确定性和预测难度。文中分析了电价的周期性特点,重点研究了风电波动性对电价波动的影响,利用负荷和风电功率构造了一种能够表征其他高成本发电方式对电价影响的新特征;然后将注意力机制与时间卷积网络结合,建立了双层多头自注意力时间卷积网络来挖掘电价时序规律以及外部因素对电价的影响作用;通过北欧电力市场真实数据进行预测效果验证,结果表明文中方法与现有电价预测方法相比将平均绝对误差(mean absolute error,MAE)值降低约45%。In power system with a high proportion of wind power,the strong volatility of wind power causes drastic variations in the short-term balance of supply and demand,which increases the uncertainty of electricity prices and the difficulty of electricity price forecasting.This paper analyzes the cyclical characteristics of electricity prices,fo-cusing on the impact of wind power fluctuations on electricity prices.Using load and wind power,a new feature is constructed to characterize the impact of variations in electricity generation capacity of other high-cost power genera-tion methods on electricity prices.By combining attention mechanism with time convolutional network,a double-layer multi-head self-attention time convolutional network is established to explore the temporal patterns of electrici-ty prices and the impact of external factors on electricity prices.The forecasting effect is validated using actual data from the Nordic electricity market,and the results indicate that the proposed method reduces the MAE value by ap-proximately 45%compared to existing electricity price forecasting methods.

关 键 词：电价预测 风电功率波动 发电成本 时间卷积网络 多头注意力机制 

分 类 号：TM614[电气工程—电力系统及自动化] TM73