As an important firefighting equipment, fire cylinders need to undergo regular safety evaluations during their service period. In order to efficiently and accurately evaluate the safety status of fire steel cylinders, a safety evaluation model suitable for fire steel cylinders was established based on the Analytic Hierarchy Process and Fuzzy Comprehensive Evaluation Method. The feasibility of the model was verified through case evaluation. Secondly, the BP neural network based on Multi population genetic algorithm (MPGA) is used to optimize the safety evaluation model of fire steel cylinders. This method improves the process of updating weights and thresholds of the BP neural network through multi population genetic algorithm, improving the accuracy of BP neural network prediction results and the efficiency of fire steel cylinder safety evaluation. Finally, the construction of safety evaluation models for fire steel cylinders based on BP, GA-BP, and MPGA-BP was completed using Python. By comparing and analyzing the prediction results of three models, it was found that the MPGA-BP neural network has the smallest prediction error, proving that the proposed MPGA-BP safety evaluation model has high accuracy and can more efficiently and accurately evaluate the safety of fire steel cylinders.