The invention provides a method that uses a fuzzy sets-based mathematic method to calculate the fuzzy probability of generating accidents, thus assisting to analyze influences by various uncertain artificial factors on the rules and features of power accidents, mal-operation accidents and unprofessional operation accidents, and fully utilizing data; in addition, considering mass linear non-fractional data exist in the power accident, mal-operation accident and unprofessional operation accident data, a kernel k-Means Clustering method is employed to firstly map original space sample data to a high dimension space, i.e., the kernel space, thus enabling the sample data to be linear fractional (or nearly linear fractional), and classifying data in the kernel space. The classification results can help users to determine high risk behaviors that cause power accidents, mal-operation accidents and unprofessional operation accidents, thus deriving weak links of the accidents, making targeted prevention measures and control strategies, and reducing generation probabilities or preventing repeated formations of power accidents, mal-operation accidents and unprofessional operation accidents.