Micro-energy system is an important aggregation part of urban distribution network terminals, and its ability to cope with the random characteristics of source-load provides an effective support for stable operation of urban distribution network. An intelligent dynamic scheduling method considering the random fluctuation of source-load is proposed for the micro-energy system in urban industrial park. A mathematical model is constructed for the economic dispatch of micro-energy system considering multiple dispatchable resources in the industrial park. Then, the constructed economic dispatch model of micro-energy system is represented as a deep reinforcement learning model with continuous action regulation. Finally, a dual delayed deep deterministic policy gradient algorithm is used to obtain the dynamic continuous dispatch policy under the deep reinforcement learning model. The proposed method not only avoids modeling the uncertainty of random fluctuation of source-load, but also avoids the discontinuity of adjustable equipment output with discrete Q-learning. Simulation results show that the proposed dynamic scheduling method has better economy and adaptivity.