基于AP聚类和CNN-BILSTM模型的建筑能耗预测
Building Energy Consumption Prediction Based on AP Clustering and CNN-BILSTM
孔耀棕 1张沛露1
作者信息
- 1. 吉林建筑大学,吉林 长春 130000
- 折叠
摘要
建筑能耗占全社会能耗总量近40%,建筑业是我国能耗最大的行业之一.为了反映建筑运行特性、减少碳排放,提出了一种基于Affinity Propagation(AP)聚类的CNN-BILSTM神经网络的混合建筑能耗预测模型.通过AP聚类把影响建筑能耗的因素进行聚类,得到主要的影响因素.在CNN-BILSTM模型中,卷积神经网络可以提取建筑物时间序列消耗数据中的深层次特征,双向的LSTM网络(BILSTM)中信息的流动是双向的,能够有效处理能耗数据之间的关系.实验结果表明,对比其他模型,CNN-BILSTM在评价指标上表现更好,预测更加准确.
Abstract
Building energy consumption accounts for nearly 40%of the total energy consumption in society,and building industry is one of the industries with the highest energy consumption in China.In order to reflect the operational characteristics of buildings and reduce carbon emissions,a hybrid building energy consumption prediction model based on Affinity Promotion(AP)clustering and CNN-BILSTM neural network is proposed.The factors that affect building energy consumption are clustered through AP clustering.In the CNN-BILSTM model,convolutional neural networks can extract deep level features from building time series consumption data.The flow of information in BILSTM is bidirectional and can effectively handle the relationships between energy consumption data.The experimental results show that compared to other models,CNN-BILSTM performs better in evaluation indicators and pre-dicts more accurately.
关键词
AP聚类/卷积神经网络/BILSTM/建筑能耗Key words
AP clustering/Convolutional Neural Networks/BILSTM/building energy consumption引用本文复制引用
出版年
2024
NETL
NSTL
万方数据