Integration of functional resonance analysis method and reinforcement learning for updating and optimizing emergency procedures in variable environments

•We integrated FRAM with RL for updating emergency schemes.•FRAM is utilized to model the emergency response process.•RL is introduced to update emergency schemes.•We optimized durations of activities through multi-objective optimization. Blowout accidents are prone to generate personal casualties, property losses, and even environmental disasters. To alleviate the consequences of accidents, it is essential to conduct effective emergency operations and update emergency schemes when necessary. In the update of the emergency plan, how to effectively optimize the allocation of resources is an open question. To deal with above difficulties, we propose a hybrid methodology by integrating the functional resonance analysis method (FRAM) and reinforcement learning (RL) for updating and optimizing emergency schemes. In the proposed methodology, FRAM is utilized to model the emergency response process based on function, variability, and coupling. Since the environment of emergency operations usually changes, RL is introduced to update emergency schemes that are constructed by FRAM. The selection of reward value by the agent reflects the variability of functional nodes in the FRAM model. To optimize emergency schemes, the interval analytic hierarchy process is integrated with multi-objective decision-making to analyze the duration, cost, and exposure risk of emergency operations. The installation of a capping stack, an emergency technique for deepwater blowout accidents, is employed to illustrate the applicability of the methodology. The results show that the proposed model is beneficial to determine emergency actions adapted to condition or scenario change in accidents.