Multi-agent reinforcement learning defense deployment method

The invention relates to a multi-agent reinforcement learning defense deployment method, which comprises the following steps of: confirming each server existing in an environment and each piece of Internet of Things equipment controlled by each server, initializing each server into an Actor-Critic network, and defining the state of each server; in each time slot, an attacker is arranged to initiate an attack in an attack mode, and the attack interval is a set value; each device in the environment makes a response, and a detection interval is obtained through an Actor network of the corresponding server; inputting the state of the Internet of Things equipment into the Actor network of the corresponding server to obtain an optimal defense strategy index, and obtaining a final security strategy of each server based on the index; each server takes the state and the security policy as the input of the Critic network, and calculates the utility, the delay and the data protection level of the security policy; and st