In order to quantitatively evaluate the disaster prevention suitability of urban emergency shelters, this paper construct an evaluation index system for evacuation site suitability from the three aspects of effectiveness, accessibility, and safety. Using real-coded accelerated genetic algorithm (RAGA) to optimize the projection pursuit (PP) model to analyze the suitability of emergency shelters in Yangpu District. The result shows: indicators such as effective evacuation area, distance from medical aid sites, and number of people that can be served have a greater impact on the disaster reduction capabilities of emergency evacuation sites. Among the existing emergency shelters in Yangpu District, the Public Green Space on Dalian Road is highly adaptable. the Fudan Science Park Primary School and Fifteenth Middle School are poorly adaptable, and urgently need to be optimized.