【摘要】 为了准确反映建筑运行特性，本文基于卷积神经网络（CNN）构建了建筑能耗的深度学习模型，充分利用CNN的特征提取能力和全连接层的回归预测能力，通过结合建筑能耗数据特点，设计适配CNN模型的数据结构以挖掘深层特征，建立了能耗预测模型框架。以上海某办公楼的能耗数据为案例，结果表明：CNN模型相较于多层感知机（MLP）模型有更高的训练效率和更好的预测性能，其中较优的逐行扫描卷积神经网络（R-CNN）在训练集和测试集上决定系数（R~2）分别达到0.994 7和0.993 4，可用于建筑能耗的建模与预测。更多还原

【Abstract】 In order to accurately reflect the operation characteristics of buildings, a deep learning model of total building energy consumption is constructed based on convolutional neural network(CNN). Making full use of the feature extraction ability of CNN and the regression prediction ability of the dense layer, combined with the characteristics of building energy consumption data, the data structure adapted to CNN model is designed to mine the deep features, and the energy consumption prediction model framework is established. Taking the energy consumption data of an office building in Shanghai as an example, the results show that the CNN model has higher training efficiency and better prediction performance than the MLP model. The coefficient of determination(R~2) of the superior row-wised CNN in the training set and test set reaches 0.994 7 and 0.993 4respectively, which can be used for building energy consumption modeling and prediction.更多还原