<p>Communities without effective sensor and warning systems mainly rely on the self-organization of their members in their early evacuation response to disasters. Based on a multilayer network system, we proposed a modeling framework to analyze the self-organized evacuation process. Based on a social network obtained from the survey of a typical Chinese community, we simulated an evacuation in this community. The results showed that (1) a loudspeaker system can eliminate the impact of the clustering coefficient and degree centrality of initial peer incident discoverer on the diffusion efficiency of warning messages; (2) compared to the time interval of the loudspeaker system, its broadcast range is a more important driver of evacuation; (3) the usage frequency of the chat app significantly affects the diffusion of warning messages. The novelty of this study lies in that (a) its use of a multilayer network to model the self-organized evacuation system and expansion of the warning diffusion mechanism and (b) its strengthening of the integration of modeling and survey and improvement of the processing of social system data in simulation. Consequently, this study provides actionable insights for self-organized emergency management. First, prioritizing broadcast range of loudspeaker yields higher returns than increasing repetition frequency. Second, leveraging online community chat groups to clarify risk boundaries effectively inhibits shadow evacuations. Furthermore, mobile broadcast strategies should aim for broad, uniform coverage to prevent information distortion. These findings can offer guidance for optimizing warning strategies and evacuation organization plans in areas with underdeveloped emergency systems.</p>