The "dual carbon" goal drives the rapid development of new power systems with new energy. Large-scale new energy access brings strongly random disturbance to the power system. Traditional control methods cannot cope with the issue the power system becoming unstable because of the strongly random disturbance. In this paper, from the perspective of automatic generation control (AGC), a polymorphic energy-coordinated control strategy with information relaxation is proposed to obtain the optimal cooperative control of the polymorphic energy system. In the "control" part, the proposed strategy uses lookahead-bounded Q-learning (LQ) with complete information relaxation characteristics to predict the upper and lower bounds of future Q values, so as to improve the fast convergence ability and control performance of Q learning in a strong random environment. In the "allocation" part, the proposed strategy uses a novel hierarchical double Q-learning based multi paxos (HDQMP) strategy to solve the "dimensional disaster" problem caused by the surge of units. Through the simulation of the improved IEEE standard two-area load frequency control model and the multi-state energy system model of large-scale new energy access, the effectiveness of the proposed method is verified. Compared with other methods, it has better control performance and faster convergence speed.