A hybrid intelligent approach for university building energy management: Integrating GABP prediction with improved PSO scheduling

Accurate energy load forecasting is crucial for optimizing the dispatch of integrated renewable energy systems, specially building-integrated photovoltaic (BIPV) systems, and advancing the “dual carbon” goals in the building sector. A comprehensive energy system study should encompass precise load forecasting, efficient energy systems, and effective optimization scheduling. While neural network-based forecasting models are prone to falling into local optima, traditional PSO for microgrid optimization suffer from slow convergence and rigid constraint handling mechanism. Therefore, this paper proposes a GABP (Genetic Algorithm-based Back Propagation‌) load forecasting model based on energy-level classification and establishes a microgrid system comprising teaching buildings, laboratory buildings, and dormitory buildings. To minimize the daily comprehensive power supply cost, an improved PSO algorithm is adopted for low-carbon dispatch. Improvements include an adaptive inertia weight linearly decreasing from 0.9 to 0.4 to balance global and local search, and a mutation after velocity updates to expand the search range and escape local optima. The results demonstrate that the optimized load forecasting model achieves higher accuracy, with reductions in MAE, RMSE, and MSE, along with increases in R2 by 9%, 5.7%, and 1% for teaching buildings, laboratory buildings, and dormitory buildings, respectively. Following optimization, the economic benefits on each typical day increased by 503.2% (June 23rd), 450.3% (September 11th), and 96.5% (December 25th). The three typical days reduced carbon emissions by 160.0 kg, 197.2 kg, and 485.5 kg respectively. This study provides theoretical insights for the development of zero-carbon campus microgrids.