Research on the optimization of Rescue Action Policy in Stochastic Weather Conditions through Q-learning

Based on the stochastic weather conditions, and the reasonable assumption of the application scenario, the optimal strategy design of carrying food, oil, equipment and other materials to the designated place in the form of a team was studied. Some specific area are divided into several divisions to establish action space for an action team. The weather probability distribution is set, the state-action value function that depends on the expected cumulative reward is constructed and iteratively updated, and the Q learning optimization model of the action policy for rescuing disaster areas is established when there is a ground supply station or airdrop supply point by preset. With the help of two hypothetical maps, the simulation obtains the optimal policy of the action when the most materials remain. The solution results are reasonable and reliable, the model is flexible, and the application range is wide.