Abstract Estimation of a building’s heating and cooling loads is an important factor taken into account implementation of energy saving measures in order to enhance energy performance of the building. In this work, the heating and cooling loads are predicted to enhance the building energy performance using different types of artificial neural networks namely, Elman network, recurrent network and back propagation network. The effect of eight input variables (relative compactness, surface area, wall area, roof area, overall height, orientation, glazing area, glazing area distribution) on two output variables (heating load and cooling load of residential buildings) is studied. The collected features are given as input to various neural networks for predicting the heating and cooling loads. The performance of the method is calculated in terms of mean absolute error, mean square error and mean relative error. Among all the networks back-propagation neural network has highest accuracy. The mean absolute error in predicting the loads is found to be 0.1 for heating load and 0.1254 for cooling load which is much better than already existing methods. The results of the work further reinforce the fact that ANN is an important tool for prediction and analysis of energy performance of a building.

中文翻译：

---

用于提高能源性能的城市建筑热冷负荷近似的改进方法

摘要 建筑物热负荷和冷负荷的估算是考虑实施节能措施以提高建筑物能源性能的重要因素。在这项工作中，使用不同类型的人工神经网络，即 Elman 网络、循环网络和反向传播网络，预测热负荷和冷负荷将提高建筑能源性能。研究了八个输入变量（相对密实度、表面积、墙体面积、屋顶面积、总高度、方向、玻璃面积、玻璃面积分布）对两个输出变量（住宅建筑的热负荷和冷负荷）的影响。收集到的特征作为各种神经网络的输入，用于预测热负荷和冷负荷。该方法的性能是根据平均绝对误差、均方误差和平均相对误差来计算的。在所有网络中，反向传播神经网络的准确率最高。发现预测负荷的平均绝对误差对于热负荷为 0.1，对于冷负荷为 0.1254，这比现有方法要好得多。工作结果进一步强化了这样一个事实，即人工神经网络是预测和分析建筑物能源性能的重要工具。