Electric power generation safety incidents can lead to severe consequences, including casualties and widespread power outages. Previous research has mainly focused on the mechanisms and causal relationships of accidents. However, these incidents result from multiple factors working together, lacking systematic analysis. This study examines 161 electric power generation safety incidents from 2015 to 2022, utilizing grounded theory for coding to construct a causal model. The derived model is used as a conditional variable for fuzzy set qualitative comparative analysis (fsQCA), with accident severity as the outcome variable. Forty-five cases are selected for assigning values, and R language and fsQCA software are integrated for univariate necessary condition analysis, followed by configurational analysis. Results show the grounded theory-derived causal model includes six factors: human unsafe behavior, equipment factors, enterprise safety management, on-site safety management, safety qualifications of personnel, and environmental factors. Necessary condition analysis indicates incidents result from multiple conditions. Configurational analysis identifies seven paths condensed into three types: management deficiency, low safety qualifications, and unsafe behavior. Recommendations are proposed for each type, discussing intrinsic connections between variables based on conditional variables in configurational paths. The aim is to reduce electric power generation safety incidents, ensure personnel safety, and guarantee continuous electricity supply.