Energy consumption prediction and diagnosis of public buildings based on support vector machine learning: A case study in China

Abstract

As one of the three major fields of building energy consumption, public buildings (PBs)are under pressure regarding energy saving and emission reductions, with PB energy consumption accounting for 38% of the total consumption. Thus, CO2 emissions released in PBs have become crucial for China in achieving its emission mitigation goal in the “Post Paris” period. This paper is the first to develop a support vector machine (SVM) method to predict and diagnose PB energy consumption based on 11 input parameters, including historical energy consumption data, climatic factors and time-cycle factors. Months with air-conditioning energy consumption in Wuhan were considered the study period, and we used June and July data for model prediction training, August data as the test set, and September data to diagnose the air conditioner energy consumption anomaly. The results show that air conditioning energy consumption was abnormal for four days in September. Relevant policies and suggestions are proposed based on the causal analysis. This research is expected to provide theoretical guidance and a practical data reference for building operations management.