Optimized Operation of Integrated Energy System in Zero-Carbon Park Based on Deep Reinforcement Learning

Aiming at the multi-energy coupling and zero-carbon emission operation challenges of the integrated energy system in the park, this study adopts the deep reinforcement learning method and optimizes its operation based on the flexible actor-critic framework. Firstly, we construct a deep reinforcement learning model framework for optimal scheduling of integrated energy system in the park. Secondly, in order to interact with the agent, we designed the environmental model of the integrated energy system in the zero-carbon park, and planned in detail the core elements of deep reinforcement learning, such as action space, state representation and reward function. Finally, we set two scenarios for analysis : one is the traditional scenario without considering the carbon neutralization constraint, and the other is the carbon neutralization scenario integrated with the joint operation mechanism of carbon capture and power-to-gas. In these two scenarios, we combine the time-of-use electricity price strategy to conduct an in-depth analysis of the optimal scheduling results of deep reinforcement learning. The research results show that the proposed model and method are significantly effective. The zero-carbon park integrated energy system using deep reinforcement learning not only improves the utilization rate of clean energy, but also reduces the overall cost of system operation, and successfully achieves the carbon neutrality goal of the park.