Abstract: The volatility of renewable energy output, the randomness of load demand and the dynamic characteristics of hot water circulation in the power-gas-heat integrated energy system (IPGHES) have brought many challenges for the scheduling process, which makes the traditional stochastic scheduling methods unable to adapt to the load of integrated energy system and diversity of renewable energy. In response to the above issues, a typical daily scheduling method based on the improved deep deterministic policy gradient (IDDPG) algorithm is proposed to flexibly solve the randomness problems in the supply and demand process. First, the prioritized experience replay (PER) mechanism is added to the experience tank of the DDPG to distinguish the values of the different experiences. Then, the OU random noise using of the Gaussian process of decreasing variance is applied to the strategy network parameter vector to improve the exploration performance. Further, the second order oscillatory-Bayesian (SOO-Bayes) algorithm is utilized to adjust the structural parameters. By constructing the dynamic IPGHES of park model that interacts with the IDDPG data at the cost of energy exchange, equipment depreciation and imbalance between supply and demand, the status space, the scheduling action and the bonus functionare defined, and the decision-making scheduling of the working days and the weekends is analyzed and contrasted according to the IDDPG. Finally, an actual micro-grid example in a university is used to prove that the proposed scheduling method is better than the random scheduling, the CPLEX solver scheduling and the traditional DDPG scheduling.