VLPSO-CNN-BiLSTM: Variable Length Particle Swarm Optimization-Based CNN-BiLSTM Model for Electricity Load Forecasting

Electricity load forecasting is one of the critical tasks in modern power system management. In recent years, deep learning has been widely applied in electricity load forecasting. However, it often relies on manual design of neural network architectures and hyperparameters, requiring extensive trial-and-error that increases temporal and economic costs. This study proposes an automated deep learning approach termed as VLPSO-CNN-BiLSTM for electricity load forecasting by utilizing variable length particle swarm optimization (VLPSO) for optimizing a hybrid neural network model (CNN-BiLSTM) based on convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM).The neural architecture and hyperparameters of the CNN-BiLSTM model are defined as the decision variables and a performance index evaluating forecasting accuracy, i.e., mean square error (MSE) is considered as the optimization objective. VLPSO is elaborately designed as the optimization tool to search for the high-performance architecture and hyperparameters of the CNN-BiLSTM. Experimental results on the 2018 Australian Energy Market Operator dataset demonstrate that the proposed approach outperforms-enhanced grey wolf optimization (EGWO) algorithm-based CNN in terms of root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE).