Evaluating the impact of the COVID-19 pandemic as an unforeseen event on societal electricity consumption behavior is a research task of profound significance. This study aims to contribute new theories and methods to this field, providing a comprehensive understanding of the potential effects of the pandemic on electricity usage. Our proposed machine-learning approach demonstrates significant results and advantages in two key aspects. Firstly, by introducing an abnormality detection algorithm rooted in lunar calendar alignment, we creatively established a detection system capable of accurately identifying unusual fluctuations in electricity consumption, thereby precisely determining the onset of the pandemic. Secondly, we designed an evaluation system that integrates a CNN-LSTM predictive model and a controlled variable strategy, enabling us to reconstruct electricity usage patterns unaffected by the pandemic. Through a comparative analysis of actual electricity consumption against the reconstructed profiles, we deeply evaluated the significant impact of the pandemic on different regions and industries. The introduction of the pandemic impact percentage as a quantification method further emphasizes the innovation and practicality of our study. The method was successfully applied to electricity consumption data from six different categories in nine cities of a certain southern province in China, revealing significant effects on commercial and industrial electricity consumption due to the pandemic. Additionally, our method extends beyond the study of pandemics and can be applied to investigate the impact of other events (such as holidays, typhoons, and special production scheduling) on electricity consumption. Overall, by delving into the complex associations behind electricity consumption, this study provides a robust example for future similar research and offers crucial decision support for the electricity industry and emergency management.INDEX TERMS COVID-19 pandemic, electricity consumption, lunar calendar alignment, abnormality detection, time series prediction.