Abstract: In view of the complexity and high risk of emergency evacuation on drilling platforms, an emergency evacuation analysis model based on Bayesian networks is constructed. Twenty influencing factors are identified from four dimensions: personnel, equipment, environment, and organization. The prior probability is determined by integrating expert experience via fuzzy numbers, and the conditional probability is estimated using the Expectation-Maximization (EM) algorithm. The results show that the success probability of emergency evacuation under static conditions is 87.34%. However, reverse reasoning indicates that incomplete evacuation procedures, insufficient equipment operation proficiency, and poor performance of personal rescue equipment are key risk factors. As time elapses, the success probability exhibits an exponential growth trend, rising from initial 0.8734 to 0.9654, which reflects the effectiveness of the emergency response mechanism. The findings provide quantitative analysis tools and risk prevention and mitigation strategies for emergency management of drilling platforms.

Key words: Bayesian network, drilling platform, emergency evacuation, expectation maximization, triangular fuzzy number