Auto-tuning method for data-driven models in building energy consumption prediction: a case of cooling load prediction

Abstract

Building consumes a significant portion of energy in the world. Improving energy use efficiency in building sector is a key approach of achieving sustainable and environmental friendly development targets. Building energy consumption prediction is essential for energy planning, equipment control and system management. Traditional physics-based model is widely used but requires sophisticated input parameter. Data-driven models, on the other perspective, utilize historical data to make predictions for future scenarios. However, the accuracy of this model greatly relies on the parameters of the machine learning algorithms within. This research proposes an auto-tuning method for machine learning algorithms in building energy prediction models, and discusses the difference between global optimum and local optimum. Based on this research, the accuracy and efficiency of the data-driven models are significantly improved.