Enhanced Energy Prediction of Next-Generation Urban Buildings Optimized with Pyramidal Dilation Attention Convolutional Deep Neural Networks

Investors, urban designers, and energy policymakers increasingly rely on modeling and analysis of building energy performance to develop long-term sustainable energy strategies that reduce energy consumption and emissions. However, a gap remains between building energy modeling and conventional planning techniques due to inconsistent energy data and the lack of scalable building models. In this work, we propose the Enhanced Energy Prediction of Next-Generation Urban Buildings Optimized with Pyramidal Dilation Attention Convolutional Deep Neural Network and Bitterling Fish Optimization (EEP-UB-PDACN-BFO). The approach begins with input data obtained from the Building Energy Ratings (BER) dataset, which is preprocessed using the Bilinear Double-Order Filter (BDOF) to detect and remove inconsistencies. The preprocessed data are then passed through the Pyramidal Dilation Attention Convolutional Network (PDACN) to forecast the Energy Use Intensity (EUI) of buildings. To further improve accuracy, Bitterling Fish Optimization (BFO) is applied to optimize the weight parameters of the PDACN. The proposed EEP-UB-PDACN-BFO model is implemented in Python and evaluated using performance metrics including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Mean Square Error (MSE), and the coefficient of determination (R2). The proposed method achieves an R2 of 0.95, an RMSE of 6.2 kWh/(m2∙year), and an MAE of 3.9 kWh/(m2∙year), outperforming existing approaches. These results demonstrate that the proposed method substantially enhances building performance modelling, providing a more robust tool for the development of sustainable energy policies and industrial regulations.