In view of life safety, economic security, social security and ecological security, 20 indicators were selected to construct regional water security dynamic evaluation index system and grading standard. The AGPSO-MEPP is proposed to optimize the projection direction of the maximum entropy projection pursuit (MEPP), and the AGPSO-MEPP water safety evaluation model is proposed. The APSO algorithm is constructed and the inertia weight is linearly reduced. Particle Swarm Optimization (LDWPSO) algorithm and Basic Particle Swarm Optimization (PSO) - MEPP model were used to evaluate the water security of Yunnan Province from 2006 to 2015 and 2020. The results show that: The optimal, worst-case, average and standard deviation of objective function of AGPSO were better than those of APSO, LDWPSO and PSO, and had the best global optimum. The water security assessment of Yunnan Province from 2006 to 2013 was “unsafe” by the AGPSO-MEPP model, “almost safe ” from 2014 to 2015, and “safe ” in 2020. From 2006 to 2015, the water security of Yunnan Province showed an increasing trend with time, and the upgrading trend was remarkable. The water security evaluation results of APSAP-MEPP model are the same as those of APSO-MEPP model, but the rankings are different. There are differences in the evaluation results and ranking with LDWPSO-MEPP and PSO-MEPP models. Among them, the difference between PSO-MEPP model and the ranking of PSO-MEPP model is the most significant, which indicates that the extreme value optimization ability of the algorithm determines the accuracy of evaluation.