Research on Optimization Management of Building Energy System using Random Forest with Particle Swarm Optimization

In recent days, the energy consumption in buildings has produced with the effect of environmental caused including air and noise pollution. In previous, for the consumption of building energy several methods are utilized such Artificial Neural Network (ANN). However, the existing ANN model suffers dimensionality and takes more time for training. In this paper, a Machine Learning (ML) based Random Forest with Particle Swarm Optimization (RF-PSO) is proposed for the optimizing the management of building energy system. With PSO hyper parameter tuning the proposed RFPSO models dimensionality is reduced with less training time. Initially, the data is taken from Building Data Genome Project 2 (BDGP2) dataset and then preprocessed with min-max scalar, z-score normalization. After that, the preprocessed data is further processed through RF for classification to classify residential or non-residential energy consumptions. Next, the PSO algorithm is used to select the optimum hyper parameters of RF classification to validate the testing dataset. From the results, the proposed RF-PSO model produces outstanding results by achieving 2.20 of MAE, 2.80 of RMSE, and 0.66 of $R^{2}$ respectively, when compared to existing Support Vector Machine (SVM) model.