In order to reduce the number of training rounds required for agent convergence

improve the utilization efficiency of experience samples

and optimize the energy scheduling of the integrated energy system(IES)

this paper introduces deep reinforcement learning(DRL) algorithm and proposes an improved deep deterministic policy gradient(DDPG) algorithm based on multi-environment instances and data feature score experience sampling. Firstly

multiple environment instances are introduced to enable extensive interactions between the agent and the environment

thereby obtaining informative experience. Secondly

feature quantization is applied to different data and experience sampling is performed according to feature to enhance sample utilization efficiency. Finally

the improved DDPG algorithm is compared with the classical soft actor-critic (SAC) and twin delayed deep deterministic policy gradient(TD3) algorithms

validating the effectiveness of the proposed method in improving convergence speed and sample utilization efficiency. Furthermore

the performance improvement of the model through incremental learning is verified through case simulations.