Building energy consumption prediction based on spatiotemporal enhanced LSTM

Accurate prediction of building energy consumption (BEC) plays a pivotal role in building energy conservation management, energy system optimization, and the achievement of carbon emission reduction goals. To this end, this study proposes a BEC prediction method based on spatiotemporally enhanced LSTM (SE-TPA-LSTM): First, channel attention is employed to focus on key variables, and weights are adaptively adjusted according to the degree of influence of each variable on energy consumption. Then, LSTM is used to extract deep features by time steps, and temporal attention is applied to highlight the features of important time steps. Finally, information is aggregated through non-linear mapping to achieve accurate prediction. Experimental verification is conducted on the AEP dataset, and comparisons are made with 7 mainstream methods, which verifies the validity of our method.