Abstract: In order to explore trigger factors of crowd stampede and evaluate the accident risk quantitatively, a static Bayesian network model was improved based on the Bayesian estimation theory. A dynamic Bayesian model, by which posterior parameters can be calculated according to the collected real-time data, was built for obtaining quantitative assessment results of dynamic risk. Finally, a certain commercial district with a large population in Beijing City was used as an example to verify the effectiveness of the proposed method. The results show that the risk of crowd jamming and that of stampede are 0.8×10^-3 and 7.6×10^-6 respectively, that with the introduction of real-time data, the risks are improved to 2.4×10^-3 and 1.63×10^-5 respectively, that slow evacuation, occupied evacuation routes, and the lack of safety logo are main causes of jam and stampede, that the probability of occurrence and weight of influence of each basic event in the instance are dynamic, which proves that the model is effective.

Key words: public safety, risk assessment, dynamic Bayesian network, probability updating, stampede