基于图神经网络的建筑能耗预测

Modeling of Building Energy Consumption Prediction Based on Graph Neural Networks

在城市建筑管理中,准确预测建筑能耗对实现建筑节能和构建智慧城市具有重要意义.由于能耗数据的复杂性,长期并准确地预测建筑能耗是时间序列预测中极具有挑战性的难题之一.近年来,研究人员将神经网络模型应用于能耗预测任务,并取得了优秀的预测结果,然而建筑能耗会受到多维因素的影响,为了提高预测精度,提出了一种基于图神经网络的建筑能耗预测方法.该方法使用改进的图卷积网络来捕获时间序列的空间依赖关系,通过时间卷积模块来获取时间序列的时序依赖关系,并通过时空融合,更充分地挖掘多元时间序列中的时序特征,支持在端到端的框架中联合学习,在真实的能耗数据集上的实验结果证实了模型拥有更加优异的性能表现.

In urban building management,the high proportion of building energy consumption is a huge problem at present.Accurately predicting building energy consumption is of great significance to achieve building energy conservation and the building of smart cities.Due to the complexity of energy consumption data,long-term and accurate forecasting of building energy consumption is one of the most challenging problems in time series forecasting.In recent years,researchers have applied neural network models to the task of energy consumption prediction and achieved excellent prediction results.However,building energy consumption is affected by multidimensional factors.In order to improve the prediction accuracy,this paper proposes the modeling of building energy consumption prediction based on graph neural networks.The method uses a modified graph convolutional network to capture the spatial dependencies of time series,and a temporal convolution module to obtain the temporal dependencies of time series.Through the fusion of time and space,time series features that multivariate time series can be more fully mined,and joint learning in an end-to-end framework can be supported.The experimental results on the real energy consumption dataset confirm that the model has better performance.