NASH EQUILIBRIUM STRATEGY AND SOCIAL NETWORK CONSENSUS EVOLUTION MODEL IN CONTINUOUS ACTION SPACE

摘要

Provided in the present invention are a Nash equilibrium strategy and a social network consensus evolution model in a continuous action space, which belong to the field of reinforcement learning methods. The strategy of the present invention comprises the following steps: initializing parameters; randomly selecting an action x_i according to a normal distribution N(u_i,σ_j) according to a normal exploration rate; and performing execution, and then obtaining a return r_i from the environment; if the return r_i acquired by an agent i after executing an action x_i is greater than a current cumulative average return Q_i, the learning rate of u_i is α_ub, and vice versa, the learning rate is α_us; according to the selected learning rate, updating u_i, variance σ_i, and Q_i; and finally updating the cumulative average strategy (I); and if the cumulative average strategy (I) converges, outputting the cumulative average strategy (I) as the final action of the agent i. The present invention has the beneficial effects of: maximizing its own interests in the process of interacting with other agents, and finally learning the Nash equilibrium.