Abstract: To effectively prevent and control safety risks during prefabricated building construction, an intelligent safety assessment model based on GA−PSO−BP neural network was proposed. Grounded in the 4M1E theoretical framework (man, machine, Material, method, environment) and employing correlation coefficient analysis, 25 critical indicators were selected from an initial pool of 36 candidate indicators to establish a safety evaluation index system for prefabricated building construction. In terms of model development, the global search capability of genetic algorithm (GA) was integrated with the rapid convergence characteristics of particle swarm optimization (PSO) to form a hybrid GA-PSO optimization algorithm. This hybrid algorithm was utilized to determine the optimal initial weight and threshold parameter ranges for the BP neural network. The applicability of the GA−PSO−BP model was validated through comparative simulation experiments and engineering case studies. The results demonstrate that the GA−PSO−BP model exhibits significant advantages in improving accuracy and convergence efficiency. The average error rate was only 0.94%, representing a substantial reduction compared to traditional BP neural network (3.28%), GA−BP model (2.14%), and PSO−BP model (2.80%). Furthermore, the proposed model achieved optimal solutions in just 42 iterations. In practical engineering applications, the safety level assessment results (Level 4) generated by this model showed high consistency with expert evaluations and actual construction conditions, confirming its effectiveness in addressing the dynamic and nonlinear characteristics of complex construction scenarios.This research provides a high-precision and efficient intelligent assessment tool for prefabricated building construction safety. The established indicator system and methodological framework can also serve as a reference for safety management in other types of construction projects.