Abstract: Energy management plays an important role in the operation optimization of integrated electricity-heat energy systems. How ever,the fluctuation of renew able energy pow er generation and the randomness of energy loads in the system make the energy management problem full of challenges. In order to solve this problem,this paper proposes an optimal energy management approach for integrated energy system considering the uncertainties of renew able energy and load demands. In this paper,the energy management problem of the system is expressed as a Markov decision process w ith unknow n transition probability,and the state space,action space and rew ard function of the process are defined. In order to solve the M arkov decision process,an optimal energy management approach based on deep Q-learning netw ork is proposed.Simulation results show that the proposed method can adaptively respond to the random fluctuations of source and loads and realize the optimal energy management.