Optimal Energy Consumption Prediction for University Buildings Based on Energy Classification Using GA-BP

A prediction model based on genetic algorithm-optimized back propagation neural network with energy class division is proposed in this study. The building energy consumption patterns are analyzed to perform energy class division, and the level numbers are utilized as inputs to the neural network for energy consumption prediction. The prediction results serve as fundamental data for formulating energy-saving measures. The results demonstrate that using the level numbers derived from energy class division as model inputs achieves significantly better performance compared to using actual energy consumption values. For the case studies of teaching buildings, laboratory buildings, and dormitory buildings, the mean absolute error(MAE), root mean square error(RMSE), and mean squared error(MSE) are reduced by an average of $5.38 \%, 2.22 \%$, and 4.16%, respectively. The optimized model exhibits markedly improved computational speed and prediction accuracy. Furthermore, increasing the number of energy levels is shown to have minimal impact on prediction performance, while enabling accurate energy-level forecasting for up to $\mathbf{1 5}$ days in advance.