建筑设备科学与工程学院

Electric Power Systems Research

Building energy consumption forecasting Improved black kite algorithm CNN-BILSTM-Attention model Hyperparameter

Forecasting building energy consumption is essential for improving energy efficiency and enabling energy scheduling. However, existing energy consumption prediction models face two main issues: (1) Although some building energy consumption prediction models have demonstrated good predictive performance, they may have limitations when confronted with the complex nonlinear and time-dependent characteristics of building load data. (2) Existing studies mainly focus on comparing the performance of different prediction models, with limited attention given to the impact of optimizing combinations of hyperparameters within the models. Therefore, this study proposes a CNN-BiLSTM-Attention hybrid prediction model optimized by the improved Black Kite Algorithm (IBKA). The improved IBKA is then applied to optimize the hyperparameters under different combinations. Simulation results indicate that the building energy consumption prediction model optimized by IBKA algorithm outperforms models optimized by other algorithms, such as PSO or BWO. The study further demonstrates that optimizing six hyperparameters, including learning rate, the number of bidirectional long short-term (BiLSTM) hidden layers, and the maximum number of iterations, leads to optimal prediction accuracy. On Dataset 1, compared with models optimized by other algorithms, the proposed model achieves lower root mean square error (RMSE) and mean absolute error (MAE) values, with an R² of 98.74% under the same six-hyperparameter configuration. In addition, on Dataset 2, the proposed model also exhibits good prediction performance for the test days from summer, transitional, and winter seasons, indicating reliable prediction capability.

冯增喜(教师)

Journal paper

255,112837

0378-7796

no

2026-01-01