The electric power industry is a high-risk industry with frequent accidents. To ensure the life safety of employees and reduce the probability of accidents, it is necessary to utilize certain technical means to identify the main causative factors in electric fatal accident reports. This research proposes an accident causation identification method that combines text categorization and catastrophe association analysis. First, an improved text feature extraction method (TF-IDF-GloVe-LDA) is proposed by fusing term frequency-inverse document frequency (TF-IDF), global vectors for word representation (GloVe), and latent dirichlet allocation (LDA) algorithms. Then, an SVM classifier was used to categorize the 473 accident report texts into five categories, ACCIDENT, PROCESS, CAUSE, PROBLEM, and RESPONSE, based on the extracted text feature vectors. Second, for the 461 CAUSEs categorized from the accident report texts, the TF-IDF algorithm is utilized to mine them to obtain 62 accident causative factors. Using the word cloud map and semantic network, a visualization analysis is performed to reveal the intrinsic connection between causative factors. The human factor analysis and classification system (HFACS) is combined to construct a framework and a Boolean dataset for electric power fatal accident causative factors. Finally, the catastrophe progression method is utilized to improve the FP-growth algorithm and propose a catastrophe association analysis method. It is utilized to mine the Boolean dataset to obtain 28 association rules and construct the accident causal network. The main causative factors were ultimately identified by analyzing the critical causal chains within the network and examining the frequency and sensitivity of each node. The findings of this study indicate that a lack of effective management, supervision, and training serves as the fundamental cause of fatal accidents in the electric power industry when compared to equipment and environmental factors. This paper offered a fresh perspective on identifying primary accident causative factors in non-standardized text and exploring their intricate mechanisms of interaction, which holds significant implications for the pre-control management of accidents in the power industry.