Abstract: This article addresses the subjectivity interference and algorithmic deficiencies that exist in scenario simulation for sudden events. Using petroleum and chemical fires as the object the article first analyzes the dynamic evolution of the scenario to, demonstrate the shortcomings of previous Bayesian network nodes and probability distribution. Subsequently, frequency thinking is utilized to determine conditional probability values, and an inequality algorithm is established that takes into account the costs of information collection, errors in countermeasures, and risks of delayed response. Finally, taking a fire in a certain liquefied petroleum gas storage tank area as an example, the fire is divided into three situations: out of control, difficult to control, and controllable. Corresponding strategies of defense, attack, and annihilation are proposed. By using software deduction and probability correction, the optimal rescue strategies at different stages are calculated.

Key words:  firefighting and rescue, petrochemical fire, Bayesian network, cost risk