A deep hybrid prediction framework for building operational carbon emissions: Integrating enhanced extreme learning machines

To address the urgent need for reducing carbon emissions in the construction industry, this paper proposes a novel hybrid prediction model, MIV-IHHO-DELM, for forecasting carbon emissions from public building operations. Unlike traditional single prediction models, which often lack accuracy and stability, our model integrates three key innovations: (1) an Improved Harris Hawk Optimization Algorithm (IHHO) enhanced with three strategies to improve high-dimensional optimization capabilities; (2) a Deep Extreme Learning Machine (DELM) optimized by IHHO for initial weight tuning; and (3) the Mean Impact Value (MIV) method for feature selection, which effectively screens multiple parameters from three categories: building energy structure and usage habits, outdoor meteorological conditions, and historical carbon emission data.Through extensive experimental analysis, the proposed MIV-IHHO-DELM model demonstrates superior performance. In ablation experiments, the model achieves a 46.26 % reduction in MAPE and a 37.92 % reduction in RMSE compared to the IHHO-DELM model, and a 71.62 % reduction in MAPE and 73.74 % reduction in RMSE compared to the MIV-DELM model. In comparative experiments, the MIV-IHHO-DELM model achieves a MAPE of 0.704 % and RMSE of 18.0133, outperforming other state-of-the-art prediction models in both accuracy and computational efficiency. Furthermore, the model exhibits strong generalization ability, maintaining high prediction accuracy even with reduced data samples.This study contributes to the field by providing a robust and efficient tool for predicting carbon emissions in public buildings, offering valuable data support for carbon reduction strategies in the construction sector.