摘要
针对机器学习模型在住宅能耗预测领域的应用进行研究。首先在能耗数据集中对住宅能耗的影响因素进行分析,NSM(当前时刻距离当天零时的秒数)与家电能耗的相关性最强,相关系数为0.22,其次是照明能耗,相关系数为0.21。其次提出并讨论4种家电能耗的数据驱动预测模型:支持向量机、BP神经网络、随机森林和梯度提升机。其中,基于集成学习方法的2个模型--随机森林和梯度提升机是表现性能最好的模型,梯度提升机能耗预测模型在训练集中有最小的均方根误差RMSE(9.99),随机森林能耗预测模型在测试集中有最小的均方根误差RMSE(77.07),集成学习方法在住宅能耗预测方面具有优势。
The application of machine learning model to the prediction of residential energy consumption is studied.First of all,the influencing factors of residential energy consumption is analyzed in data set.The correlation between NSM(number of seconds from midnight for each day)and energy consumption of household appliances is the strongest,whose coefficient of correlation is 0.22;followed by energy consumption of lights,whose coefficient of correlation is 0.21.Secondly,four kinds of data-driven prediction models for energy consumption of household appliances are presented and discussed,including support vector machine(SVM),BP(Back-propagation)neural network,random forest(RF)and gradient boosting machine(GBM).Among them,random forests and gradient boosting machine,which are based on ensemble learning,behave the best performance.The energy consumption prediction model named gradient boosting machine has a minimum RMSE(9.99)in the training set;the energy consumption prediction model named random forest has the smallest RMSE(77.07)in the test set.The ensemble learning method shows the advantages in the prediction of residential energy consumption.
作者
程亚豪
陈焕新
王江宇
Cheng Yahao;Chen Huanxin;Wang Jiangyu(Huazhong University of Science and Technology)
出处
《制冷与空调》
2019年第5期35-40,共6页
Refrigeration and Air-Conditioning
基金
国家自然科学基金(51576074)
空调设备及系统运行节能重点实验室开放基金(SKLACKF201606)
关键词
住宅能耗
预测模型
机器学习
集成学习
数据挖掘
residential energy consumption
prediction model
machine learning
ensemble learning
data mining
