BUILDING ENERGY CONSUMPTION PREDICTION BASED ON CONTEXTUAL FEATURES BY USING ENSEMBLE LEARNING METHOD

• Main Author: Indrapraja, Rachmadi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/61739
• Institution: Institut Teknologi Bandung

Building energy-consumption is one of the greatest users in consuming electrical energy globally. According to the references, building energy-consumption reaches the percentage of 36%. Prediction of energy consumption highly impacts on efficiency and effectiveness in operating building assets. Therefore, various methods are developed to conduct prediction of electrical energy use in buildings. Prediction can be made by developing a model based on algorithm. One of the methods which are used in developing prediction model is applying Machine Learning algorithms. This research aims to show the application of several types of Machine Learning Algorithms. The algorithms being used: Support Vector Regression (SVR) and Random Forest (RF). Furthermore, this study also conducts the technique of combining both algorithms by using Ensemble Learning (EL). In developing model, we also need extraction technique and features engineering from the data of measurement result. In this context, negative cleansing is used for extraction technique and Contextual Features for features engineering. The parameters of environmental condition are also used to complete features engineering. The results are four groups of datasets. Thereafter, all datasets are combined with the three algorithms. All models generated from this study have been evaluated by measuring the value of MSE (mean square error) and MAE (mean absolute error). All models show satisfactory evaluation results with the value of MSE < 0,1 and the value of MAE < 0,1 by using Ensemble Learning. Results for predictions also shows the smallest value RMSE = 0.0283 and MAE = 0.0210 during pre-pandemic. Smallest value during pandemic are RMSE = 0.0282 and MAE = 0.0235.