基于神经网络的垂直地埋管换热量预测  

THERMAL PREDICTION OF A VERTICAL GROUND HEAT EXCHANGER USING ARTIFICIAL NEURAL NETWORKS

作　　者：陈尚沅[1] 茅靳丰[1] 韩旭[1] 

机构地区：[1]解放军理工大学军环教研中心,南京210007

出　　处：《太阳能学报》2017年第11期3077-3084,共8页Acta Energiae Solaris Sinica

摘　　要：在研究地埋管结构和物性对换热的影响时,往往难以整体考虑所有因素,该文采用人工神经网络模型预测地埋管换热器在一系列运行工况下的换热量。通过建立三维地埋管数值模型及实验验证,获取网络需要的训练数据和测试数据。以岩土导热系数、回填土导热系数、进水流量、地下水渗流速度、进水温度和钻孔深度为输入,换热量为输出,建立一个三层基于反向传播算法的神经网络BP模型,探讨不同的训练函数和隐含层节点数对网络精度的影响。通过实验和模拟检验表明,最优网络模型换热量预测值与检验值的相对误差最大为0.11,线性拟合度为0.894,具有较好的精度。When study how the geometries and characteristics influence on the performance of ground heat exchanger（GHE）,it is difficult to take all factors into consideration. In this paper,we presented an artificial neural network topredict the heat flux of vertical GHE according to different design parameters. A three-dimensional model is establishedto get the training and testing data. With the soil thermal conductivity,grout thermal conductivity,inlet flow,inlet watertemperature,Darcy＇s velocity and borehole depth as inputs,heat transfer as outputs,a 3-layer artificial neural networkbased on back propagation algorithm is established,the influence of different training functions and neuron numbers isdiscussed. The verification result,tested by experiment and simulation,showed that the optimal model can be used topredict the thermal performance of GHE with the relative error of 0.11 and R-squared of 0.894.

关 键 词：地埋管换热器 数值模拟 因素分析 人工神经网络 

分 类 号：TU831.2[建筑科学—供热、供燃气、通风及空调工程]