Analysis of Influencing Factors by Machine Learning to Predict Energy Consumption of Educational Institutes

Educational institutes, as innovation drivers of science and technology worldwide, are in a great position to advance energy efficiency. However, progress on building energy efficiency studies have generally been limited by (1) scarcity of data on building energy consumption, and (2) highly complex/technical traditional building energy estimation methods. The advent of new and dynamic streams of building energy data and machine learning methods provide new ways to model building energy consumption. In this paper, we start by outlining key movements in climate and sustainability that sets the context for our study. Next, we proceed to examine existing research on factors affecting building energy consumption and conduct an exploratory analysis of our dataset to identify buildings with energy consumption above the benchmark. We then proceed to employ, evaluate, and compare the effectiveness of various machine learning algorithms for building energy estimation before concluding with recommendations to reduce building energy consumption. According to the findings, LightGBM and XGBoost are moderately effective in forecasting building energy usage, with a margin of error of roughly 48 kWh per hour in our model. Through our data exploration and machine learning endeavors, we have identified several significant elements that influence building energy usage, including Floor Area, Air Temperature, and Building Age.