EpisodicFutureThinkingMechanismforMulti-agent ReinforcementLearning