基于公式拟合和基于机器学习的地源热泵能耗建模对比分析

Comparison of Equation Fitting Based and Machine Learning Based Ground Source Heat Pump Energy Consumption Models

本研究以北京某办公楼的地源热泵系统为例,分别采用EnergyPlus公式拟合模型、支持向量机算法和BP神经网络算法对其制冷工况和供热工况进行能耗建模。系统采集了两年的运行数据,包括热泵和水泵的电耗、地源侧和用户侧的进出水温度与流量。结果表明,EnergyPlus公式拟合模型、支持向量机算法和BP神经网络算法在制冷工况下的标准平均误差(NMBE)分别为-2.22%、-1.12%和0.43%,变异均方根误差(CVRMSE)分别为10.70%、9.45%和5.01%;供热工况下NMBE分别为-1.20%、-0.97%和0.03%,CVRMSE分别为11.06%、14.26%和7.91%。3个模型的精度都满足要求,但是EnergyPlus公式拟合模型仅适用于满足热平衡公式的35.3%的情况。基于机器学习的模型对整个运行期间预测都较准确,其中BP神经网络模型效果最好。

This paper presents a case study to model the ground source heat pump system of an office building in Beijing.There modeling methods are used,including EnergyPlus equation fitting model,support vector machine algorithm and BP neural network algorithm.The system has collected two years of operating data,including the power consumption of heat pumps and water pumps,and the temperatures and flow rates of supply and return water on the ground source side and the user side.The results show that the normalized mean bias errors(NMBE)of EnergyPlus equation fitting model,support vector machine algorithm and BP neural network algorithm under cooling conditions are-2.22%,-1.12%and 0.43%,respectively,and the coefficients of variation of the root mean square error(CVRMSE)are 10.70%,9.45%and 5.01%,respectively.The NMBE under heating conditions are-1.20%,-0.97%and 0.03%,and CVRMSE are 11.06%,14.26%and 7.91%,respectively for three models.The accuracy of three models all meets the requirements,but the EnergyPlus equation fitting model is only suitable for stabilized cases,which accounts for 35.3%.The machine learning based models can ensure accuracy during the entire operation period,and the BP neural network model has the best performance.