Short-Term Load Forecasting Method Based on PSO-VMD-BiLSTM-BiGRU Model

With the advancement of smart grid technologies and the emergence of novel energy systems, loads characterized by diversity and flexibility have become a critical component of these systems. Research on their prediction models is essential for the operation, maintenance, and planning of novel energy systems. This paper presents a short-term joint forecasting method for multi-variable loads based on PSO-VMD-BiLSTM-BiGRU, taking into account the coupling characteristics of multi-variable loads and the correlations with other factors. First, various correlation analysis methods are employed to study the coupling characteristics of multi-variable loads and the correlations of influencing factors, and predictive features are selected accordingly. Second, variational mode decomposition (VMD) optimized by particle swarm optimization (PSO) is utilized to decompose the multi-variable loads, thereby improving load utilization. Finally, BiLSTM and BiGRU models are employed to forecast the low-frequency and high-frequency components, respectively, and the predicted results are combined to obtain the final forecast. The effectiveness of the proposed method is validated through experiments using data from the ASHRAE-Great Energy Predictor III competition project, comparing it with conventional load forecasting methods.