摘要
公共建筑制冷机房的短时能耗预测对机房能耗优化和故障检测诊断至关重要,传统方法存在输入维度高、难以捕捉机房特性、参数拟合工作量大等问题。提出了一种基于机器学习(ML)算法的制冷机房短期能耗预测模型,该模型仅需八个变量作为输入。分别使用一次冷冻水变流量机房和一次冷冻水定流量机房采集的真实数据对比实验了7种常见的机器学习算法(ANN、BPNN、SVM、DT、RF、XGB、MPR)以及传统经验累加公式(Sum)的预测表现。实验结果表明,RF和XGB模型因其预测准确性和鲁棒性取得了最优表现,传统方法表现最差。
Chiller plants traditional energy consumption calculation methods-typically based on empirical formula summation-often suffer from issues such as high input dimensionality,laborious calibration,and limited generalizability under variable operating conditions.A machine learning(ML)approach for short-term energy consumption forecasting in the chiller plant with eight independent input variables was proposed.Seven machine learning models together with Sum were selected to predicted the energy consumption under both variable and constant primary chilled water flow conditions..The Random forest(RF)and Extreme gradient boosting trees(XGB)were determined to be the preferred model due to their better accuracy of the energy consumption forecasting than the Sum and other five machine learning models.A practical tool for engineers to evaluate energy performance and validate control strategies was provided by the novel ML-based model.
作者
袁杨
邹秋生
王懋琪
Yuan Yang;Zou Qiusheng;Wang Maoqi(Sichuan Provincial Architectural Design and Research Institute Co.,Ltd,Chengdu,610000;Chongqing University,Chongqing,400044)
出处
《制冷与空调(四川)》
2025年第5期639-651,共13页
Refrigeration and Air Conditioning
基金
既有公共建筑空调蓄冷改造技术研究(KYYN2025026)。
关键词
制冷机房
短时能耗预测
机器学习
历史数据
Chiller plant
Energy consumption prediction
Machine learning algorithms
Historical operational data
