Method of association of user equipment in a cellular network according to a transferable association policy

Method for associating user equipment with base stations of a cellular network, said association method implementing a multi-agent reinforcement learning method, each user equipment being represented by an agent, characterised in that:

• when a user equipment if enters the network, the agent representing this user equipment downloads an instance of a meta model (πw0) to initialise an association strategy model (πw,j), said meta model having an architecture independent of the user;
• the agent representing the user equipment uj constructs (311, 313) a local observation vector (olj(t) comprising observables relating to this equipment and a global observation vector (ogj(t)) comprising observables relating to an environment of this equipment;
• the agent associated with the user equipment uj projects the local observation vector and the global observation vector (312,314) into the same reference space and combines (320) the two vectors thus projected to provide a code (cj(t)) of the state (Sj(t)) of the agent;
• the agent associated with the user equipment uj updates (335) the association strategy model by means of a policy gradient method;
• the agent associated with the user equipment uj decides (337) on the base station with which to associate and receives in return a common reward calculated from a utility function of the network.