Improved ensemble paradigms and multi-objective optimization for modelling energy consumptions and carbon emissions of residential buildings under different Chinese climates

This study uses improved particle swarm optimization (IPSO)-optimized ensemble learning paradigms (ELPs) with multi-objective optimization (MOO) for forecasting energy consumption and carbon emission of residential buildings, considering different meteorological conditions. Specifically, five hybrid ELPs, including AdaBoost, decision tree, extreme gradient boosting, gradient boosting, and random forest regressor, were constructed using IPSO. Five typical cities of China were considered for building energy modelling. The outcomes of the hybrid ELPs were compared with three widely used neural network-based paradigms. Five building energy parameters, viz. heating, cooling, lighting, equipment, and energy consumption, along with global cost, were estimated based on seven distinct influential parameters. According to the results, the employed ELPs outperformed the neural network-based paradigm. The employed hybrid paradigm of random forest regressor and IPSO, RFR-IPSO, achieved the most desired accuracy between 99.92% and 100% based on the R2 index. Contrarily, the outcomes of MOO exhibit that the parameters heating and cooling consumptions are the most influential variables and exhibit a higher impact on the building energy and global cost. The proposed approach can be used as an additional scientific foundation for energy conservation and emission reduction, while also offering decision-making approaches for the advancement of old societies in China.