Volume 17: Technology Innovation to Accelerate Energy Transitions

An Attention-based Seq2Seq Model for Short Term Energy Consumption Prediction Yuhang Zhang, Xiangtian Deng, Yi Zhang, Yi Zhang

Abstract

Building energy consumption prediction is important in energy system management, building operation, and energy supply planning. This study proposes a novel model with attention based seq2seq method, which is a deep learning algorithm, to improve the prediction performance. The developed model is performed with experiment on a real energy profile data of an office building in Shenzhen, China. The prediction performance of the proposed hybrid model is evaluated with indicators of MSE, RMSE, MAE and SMAPE. The results demonstrate that attention mechanism can improve the prediction performance of model whose input are time series. Compared with the metrics of prediction result of other models, the MSE, RMSE, MAE, SMAPE of prediction result of proposed model decrease by more than half percent.

Keywords nergy consumption prediction, seq2seq model, attention mechanism, LSTM