模型融合预测在供热负荷预测中的应用  

Application of the Model Fusion Prediction in Heating Load Forecasting

作　　者：张保东 鄢烈祥 许成 陈超 张凡 ZHANG Baodong;YAN Liexiang;XU Cheng;CHEN Chao;ZHANG Fan(Jingneng Dongfeng(Shiyan)Energy Development Co.,Ltd.,Shiyan 442002,China;不详)

机构地区：[1]京能东风(十堰)能源发展有限公司,湖北十堰442002 [2]武汉理工大学化学化工与生命科学学院,湖北武汉430070 [3]汉谷云智(武汉)科技有限公司,湖北武汉430072

出　　处：《武汉理工大学学报(信息与管理工程版)》2025年第1期118-125,共8页Journal of Wuhan University of Technology:Information & Management Engineering

摘　　要：针对集中供热热力站负荷预测问题,提出一种基于模型融合的预测方法,该方法将随机森林、极度梯度提升(XGBoost)、BP神经网络模型作为基础学习模型,采用粒子群算法(PSO)进行各模型的超参数优化,将基学习器的预测结果组合作为新的特征变量,使用SVR支持向量机作为元学习器再对这些新的特征变量进行学习并预测热负荷。以湖北省某集中供热项目为研究对象,基于实测运行数据及天气数据进行模型训练及测试。测试结果表明,相较单一预测模型,基于模型融合的集成预测方法能够提供更为精确的预测结果。集成预测模型的平均绝对误差(MAE)、均方误差(MSE)、均方根误差(RMSE)和平均绝对百分比误差(MAPE)均为最低。Aiming at the load forecasting problem of central heating stations,a model fusion forecasting method was proposed.This method used random forest,extreme gradient boosting(XGBoost),and BP neural network models as basic learning models,and used the particle swarm algorithm(PSO)to perform various tasks.For the hyperparameter optimization of the model,the prediction results of the basic learning model were combined as new feature variables,and the SVR support vector machine was used as a meta-model to learn and predict the heat load.Taking an actual centralized heating project in Hubei Province as the research object,model training and testing were conducted based on measured operating data and weather data.The test results showed that compared with a single prediction model,the integrated prediction method based on model fusion can provide more accurate prediction results.The mean absolute error(MAE),mean square error(MSE),root mean square error(RMSE)and mean absolute percentage error(MAPE)of the ensemble forecast model were the lowest.

关 键 词：集中供热 负荷预测 模型融合 热力站 集成学习 

分 类 号：TK11[动力工程及工程热物理—热能工程]