Predicting the consumed heating energy at residential buildings using a combination of categorical boosting (CatBoost) and Meta heuristics algorithms

The main purpose of this study is to investigate the amount of the daily consumed heating at residential buildings. In order to improve the predictions, the Categorical Boosting (CatBoost) method combined with six other meta-heuristics algorithms, and six different hybrid models were made. During the network training, the K-Fold cross validation algorithm has been used to pre-vent overfitting. Also, characteristics of the building as well as the temperature outside the building are considered as the main inputs of the problem. The results showed that the proposed hybrid model can improve the predictions of consumed heating with acceptable accuracy. The results confirm that optimizing the hyper-parameters of Catboost can be very useful in improving the predictions. The results showed that the Catboost model which its hyper-parameters opti-mized by Artificial Bee Colony algorithm, has the best performance among all investigated hybrid models. On the other hand, the hybrid Catboost-ABC model has the weakest performance among all models. For example, based on the test dataset, the R2 values of the hybrid Catboost-AOA model and the hybrid Catboost-ABC model are respectively equal to 0.9851 and 0.9770, which are the highest and lowest values of this index.