Prediction of Heating and Cooling Load Using Machine Learning Techniques

Abstract

Accurate prediction models for heating and cooling loads are essential for enhancing building energy efficiency, a key component of sustainable development. This study evaluates the effectiveness of three regression methods—Linear Regression, Random Forest Regression, and Bidirectional Encoder Repre sentations from Transformers-based Regression—in predicting these loads. The dataset comprises various building parameters, with heating and cooling loads as target variables. We aim to identify the most accurate and robust predictive model for load forecasting in buildings. Linear Regression was used as the baseline model and is a widely applied method for forecasting. Random Forest Regression is an ensemble learning approach that has been applied to a number of forecasting problems. It averages predictions from multiple decision trees and is, therefore, able to capture non-linear relationships in the most accurate ways. BERT-based Regression, though initially designed for natural language processing tasks, has also been utilized to solve some forecasting problems. Its capability to handle large datasets and complex relationships furthers its case in prediction applications. The results underscore Random Forest Regression as the most effective method for predicting building energy loads. This study utilized Python and its various machine-learning libraries to implement and compare the above methods.