Efficiency in building energy use: Pattern discovery and crisis identification in hot-water consumption data

As global challenges such as climate change and pandemics increasingly disrupt urban systems, the need for efficient and resilient management of energy resources has become critical. The energy used to prepare domestic hot water (DHW) takes a large proportion of residential buildings’ total thermal energy demand. However, it is often overlooked in research due to its stochastic nature and high dependence on user behaviour. This study explores the identification of the crisis and its severity level in the DHW consumption data and the corresponding control actions necessary to mitigate its impact. To identify crisis severity, we utilised the mobility data of retail/recreation activities and transit stations, making the results generalisable for any crisis. In addition, we used power consumption for DHW preparation data from 10 residential apartment buildings located in Kaunas city to develop a machine learning-based hybrid ensembling stacking classifier (ESC) capable of predicting the crisis and its severity level. Finally, we applied principal component analysis (PCA) and k-means clustering to categorise DHW consumption hours throughout the day for each severity level. The results showed that the developed ESC classifier significantly outperforms ($R^2=0.99$) the baseline LGBMC classifier ($R^2=0.92$). Combining the classifier with extracted daily consumption patterns and clusters allows the optimisation of control actions on the supply, distribution, and demand side of the DHW system.