人工神经元网络和径向基网络模型预测建筑冷热负荷的研究  

Research on artificial neural network and radial basis network model to predict building heating and cooling load

作　　者：丁治雄 吴观华 陈智刚 DING Zhixiong;WU Guanhua;CHEN Zhigang(China State Construction Lighting Co.Ltd.,Shenzhen 518000,China)

机构地区：[1]中建照明有限公司,广东深圳518000

出　　处：《新型建筑材料》2024年第6期150-155,159,共7页New Building Materials

基　　金：广东省重点研发计划项目(2023GDSF420)。

摘　　要：采用人工神经元网络(ANN)和径向基网络(RBF)模型预测了建筑冷热负荷,判断了影响建筑能耗的显著因素。通过对ANN和RBF模型隐含层神经元数量进行优化,发现8-65-1和8-97-1结构的ANN模型预测建筑热、冷负荷与数据集中的结果比较吻合,相关系数(R^(2))分别为0.962、0.953;8-28-1和8-6-1结构的RBF模型预测的结果更加吻合,R^(2)达到了0.985、0.997。RBF模型的预测精度要优于ANN模型,RBF模型预测热、冷负荷与数据集结果的R^(2)分别为0.989、0.992,而ANN的R^(2)分别为0.972、0.967。采用敏感性分析发现,影响建筑冷热负荷的8个参数中表面积的影响最显著,其次是墙面积、屋顶面积和玻璃面积,而玻璃面积分布及相对密实度的影响最不显著。Artificial neural network(ANN)and radial basis function network(RBF)were used to predict the cooling and heating loads of buildings,and significant factors affecting building energy consumption were identified.By optimizing the neuron number in the hidden layer of ANN and RBF models,it was found that the ANN model with configurations of 8-65-1 and 8-97-1 could accurately predicted building heating and cooling loads which were consistent with the results in the dataset,with correlation coefficients(R^(2))of 0.962 and 0.953,respectively.The predicted results of the RBF models with configurations of 8-28-1 and 8-6-1 were more consistent,with R^(2) of 0.985 and 0.997,respectively.The prediction accuracy of the RBF model was better than that of the ANN:the R^(2) of RBF for predicted cooling and heating loads with dataset results were 0.989 and 0.992,respectively,while the R^(2) of ANN were 0.972 and 0.967.Using sensitivity analysis,it was found that among the 8 parameters that affecting building cooling and heating loads,surface area had the most significant impact,followed by wall area,roof area,and glazing area,while glazing area distribution and relative compactness had the least significant impact.

关 键 词：径向基网络 人工神经元网络 建筑 冷热负荷 预测 

分 类 号：TU111.19[建筑科学—建筑理论]