A two-stage imputation method for enhancing urban building energy data resilience using Bayesian inference

As advanced techniques such as artificial intelligence and digital twins become increasingly integrated into urban systems, effective management of missing data is becoming more important in urban areas. A total of 8,603 buildings were found to have one or more missing data in their 2018 monthly electricity consumption data, out of about 440,000 buildings in Seoul. There were four types of missing data, consecutive missing type, non-consecutive missing type, and mixed missing type and full missing type. This study proposes a two-stage imputation method, which integrates machine learning and Bayesian inference techniques. This method was applied to six real-world buildings in Seoul. The case study identified three key findings. First, the imputation results achieved CVRMSE values ranging from 3.88 % to 12.18 %. Second, the method demonstrated effectiveness across diverse types of missing data. Third, the proposed method can effectively handle cases with up to seven missing data points. This method not only enhances the integrity of urban data but also contributes to data-driven analysis and decision-making processes within urban systems.