Potential energy saving estimation for retrofit building with ASHRAE-Great Energy Predictor III using machine learning

Energy is material basis for social development and conflicting issue of economic development around the world. With the continuous urbanization, the energy consumption of buildings will be further increased, accounting for about 40% of the total energy consumption eventually. Thus, enhancement of energy efficiency of the buildings has become an essential issue to reduce the amount of gas emission as well as fossil fuel consumption. Delivering a high-quality built environment in an energy efficient way is the crucial key to energy conservation. Energy efficiency retrofit for buildings is considered to be a promising way to achieve energy savings. Machine learning provides the ability to learn from data using multiple computer algorithms. This paper introduces several algorithms, including random forest and Light GBM, to analyze building energy consumption based on the data from Kaggle competition, providing discussion of improvement in model efficiency and economic analysis by simulating different scenarios. In addition, sensitivity analysis is conducted to show the influence of different parameters in models and metrics to quantify the accuracy of prediction are proposed. The results of this paper can help people understand quantitative influence of different variables on energy use and energy baseline models. Future works will incorporate more data type in order to enhance the performance of prediction.