基于机器学习的燃气电厂电价预测研究  

Electricity Price Forecast for Gas Power Plants Based on Machine Learning

作　　者：秦亚迪 秦锋[1] 李劼 周苏洋 康家豪 沈建羽 QIN Yadi;QIN Feng;LI Jie;ZHOU Suyang;KANG Jiahao;SHEN Jianyu(CNOOC Gas&Power Group Technology R&D Center,Beijing 100028,China;School of Electrical Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]中海石油气电集团有限责任公司技术研发中心,北京100028 [2]东南大学电气工程学院,南京210096

出　　处：《油气与新能源》2025年第2期74-81,共8页Petroleum and new energy

基　　金：国家自然科学基金面上项目“跨境综合能源系统分布式协同运行及联合仿真技术研究”(52177076);中海石油气电集团有限责任公司科技项目“燃气轮机进气冷却技术应用研究课题”(QDKJZH-2024-43)。

摘　　要：随着电力市场的放开,准确的电价预测有助于平衡电力市场供需关系。本研究旨在探讨燃气电厂在电力现货市场中的成交电价预测问题,以提升电厂生产安排的合理性和经济效益。首先,通过皮尔逊相关系数分析了多个因素对电价的影响,并结合时域分析,筛选出与电价强相关的18个核心变量。然后,应用Random Forest(随机森林)、AdaBoostRegressor(自适应增强回归器)等不同机器学习算法的5种模型,分年度、季节、工作日3种训练策略下,对电价进行按天和按小时的预测,根据预测结果,选取最优模型。最后,将模型应用到广东省的3家燃气电厂,研究发现不同训练策略的预测误差存在较大差异,全年数据一起训练的效果最佳。机器学习建模在电价预测中具备有效性,尤其是缩减变量建模可进一步优化预测结果。With the opening of the electricity market,accurate electricity price forecast is helpful to balance the supply and demand relationship in the electricity market.This paper aims to explore the problem of predicting the transaction electricity price of gas power plants in the electricity spot market in order to improve the rationality and economic efficiency of the production arrangement of power plants.It analyzes the influence of several factors on electricity price by Pearson correlation coefficient and screens out the 18 core variables strongly correlating with electricity price by combining with time domain analysis firstly,builds five models applying different machine learning algorithms including Random Forest and AdaBoostRegressor secondly,predicts the electricity prices on a daily and hourly basis under three training strategies,namely data together for the whole year,separated by seasons,and separated by types of working days and selects the optimal model according to the results thirdly,finds there are large differences in the prediction errors of different training strategies,and the best results are obtained when the year-round data are trained together by applying the model to three gas-fired power plants in Guangdong Province finally.Machine learning modeling is effective in electricity price prediction and reduced variable modeling can further optimize the prediction results.

关 键 词：燃气电厂 电价预测 影响因素 机器学习 训练策略 

分 类 号：TM73[电气工程—电力系统及自动化] F426[经济管理—产业经济]