Emergency plan updating method and device based on multi-agent reinforcement learning

One or more embodiments of the invention provide an emergency scheme updating method and device based on multi-agent reinforcement learning, and the method comprises the steps: constructing an accident emergency scheme, and determining a plurality of emergency links through a function resonance relation between emergency scene elements. And determining an adaptive emergency scheme by utilizing reinforcement learning based on the scene element reward value. Considering the influence of constraint variables, and selecting an interval analytic hierarchy process to calculate a control variable weight; and a multi-objective function is fused, and an emergency time, cost and exposure risk multi-objective optimization model is established. And optimizing the emergency scheme according to the multi-objective optimization result. According to the method provided by the embodiment of the invention, reinforcement learning is combined, the function resonance relationship between emergency scene elements and the constrain