AFSA-BLSTM with deep learning enabled low-carbon buildings energy load prediction

The variation in building energy loads affects the scheduling strategies of urban smart grids, accurate prediction of these variations is crucial for reducing building carbon emissions. Numerous methods have been proposed to estimate energy load variations, with most deep learning approaches being optimized on small-scale dataset. However, when applied to unknown environments, optimizing hyperparameters becomes challenging. This study presents a novel building energy load prediction method that integrates Bi-directional Long Short-Term Memory (Bi-LSTM) networks for feature extraction and load forecasting, enhanced by an Artificial Fish Swarm Algorithm (AFSA) module for efficient hyperparameter optimization. Experimental evaluations on two public datasets demonstrate 8.1% and 9.9% accuracy improvements respectively compared to baseline models, revealing the significant potential of the AFSA-BLSTM framework in energy load prediction applications.