In response to the pressing issues of wasted resources in mining areas and the urgent demand for large-scale energy storage, a novel integrated energy system framework for mining areas with composite energy storage is proposed. The framework utilizes abandoned underground mine caverns as spatial resources and harnesses the storage capacity from the huge potential difference of mine water inflow along with its contained low-enthalpy geothermal energy. A multi-level energy recovery approach is adopted to minimize resource waste by coupling pumped hydro storage, compressed air energy storage, and water-source heat pump technologies. To overcome the limitations of traditional optimization modeling methods, the operational optimization problem of the integrated energy system is transformed into a markov decision process. An optimal scheduling strategy is then developed based on the deep Q-learning network reinforcement learning algorithm, aiming to maximize net operational profit, enhance wind power consumption, and reduce carbon emissions. Finally, case studies under different scenarios are conducted through simulation analysis. The results verify that the proposed DQN-based optimization strategy can effectively address system nonlinearities and uncertainties from wind power and load demand, while ensuring real-time response capability for scheduling. Moreover, the proposed system model is demonstrated to achieve significant energy savings, improved energy storage density, and considerable economic and environmental benefits.