利用随机森林回归的现货市场出清价格预测方法  

Forecasting Method of Clearing Price in Spot Market by Random Forest Regression

作　　者：魏勤 陈仕军[1] 黄炜斌[1] 马光文[1] 陶春华 WEI Qin;CHEN Shijun;HUANG Weibin;MA Guangwen;TAO Chunhua(College of Water Resource and Hydropower,Sichuan University,Chengdu 610065,Sichuan Province,China;Sichuan Dahui big data Co.,Ltd.,Chengdu 610041,Sichuan Province,China)

机构地区：[1]四川大学水利水电学院,四川省成都市610065 [2]四川大汇大数据有限公司,四川省成都市610041

出　　处：《中国电机工程学报》2021年第4期1360-1367,共8页Proceedings of the CSEE

基　　金：国家重点研发计划项目(2018YFB0905204,2016YFC0402205,2016YFC0402208);四川大学专职博士后研发基金(2018SCU12062)。

摘　　要：为得到一种实用性较强且具有较高精度的电力现货市场出清价格的预测方法,该文尝试将随机森林回归应用到现货市场出清价格预测。首先通过随机森林回归的特征重要度分析功能对历史出清价和负荷输入进行特征筛选,然后建立基于随机森林回归的市场出清价预测模型,以网格搜索和交叉验证的方法确定模型参数,最后与基于决策回归树、支持向量机回归和人工神经网络的方法在北欧现货市场公开数据的基础上进行对比试验。试验结果表明该文设计预测方法相较其他方法的平均预测精度至少提高了25%,且预测效果较为稳定,同时输入特征筛选方法的应用能够进一步提高各个模型的预测精度。In order to obtain a practical and high-precision forecasting method of spot market clearing price for electricity,we attempted to apply random forest regression(RFR)to the prediction of clearing price in spot market in this paper.Firstly,the feature importance analysis function of RFR was used to filter the input feature of historical clearing price and power load.Then we established a prediction model of market clearing price based on RFR,and the parameters of model were determined by grid search and cross validation.Finally,the experiment was carried out to compare this model with methods based on classification and regression tree,support vector machine regression and artificial neural network on the basis of the public data in Nordic spot market.The results show that the average prediction accuracy of the proposed prediction method is improved by at least 25%compared with other methods,and the prediction effect is relatively stable.Meanwhile,the application of input feature selection method can further improve the prediction accuracy.

关 键 词：电力现货市场 市场出清价 随机森林 特征筛选 网格搜索 交叉验证 

分 类 号：TM73[电气工程—电力系统及自动化] F416.61[经济管理—产业经济]