Forecasting the total building energy based on its architectural features using a combination of CatBoost and meta-heuristic algorithms

Abstract

This research examines the overall energy usage in residential buildings, focusing on architectural characteristics. The study utilizes a combination of the CatBoost method and meta-heuristic algorithms for analysis. The main approach of this research is based on the accuracy defects of individual models, which leads to the employment of CatBoost as a group model. Due to the lack of enough examinations while utilizing CatBoost method, this model and its hyperparameters are optimized using various meta-heuristic methods, including Phasor Particle Swarm Optimization (PPSO), Slime Mould Algorithm (SMA), Sparrow Search Algorithm (SSA), Ant Lion Optimizer (ALO), Artificial Bee Colony (ABC), and Grey Wolf Optimizer (GWO). Eventually, the performance of all models is compared by conduction of a case study, using diverse statistical examination indexes divided by the dwelling types i.e., (1) Standard efficiency upgraded dwellings (D1), (2) High efficiency upgraded dwellings (D2), and (3) Ultra high efficiency upgraded dwellings (D3). The results show that the hybrid proposed method has a proper ability to investigate the total site energy. The results show that for the D1 dwelling and according to the test dataset, the integrated CatBoost-SMA model indicates the most desired performance in predicting the total site energy. But for D2 and D3 dwellings and referring to the test dataset, the statistical evaluation indexes emphasize that the integrated CatBoost-PPSO method shows the most reliable performance.