Introduction/Objective: With the rapid global energy transition and advancement of clean energy technologies, integrated energy systems (IES) must progressively evolve towards economic, low-carbon, and high-efficiency operation. Methods: To achieve this, this paper proposes a multi-objective coupled optimization strategy for IES, targeting economy, low-carbon footprint, and exergy efficiency (EE), based on deep reinforcement learning (DRL). The strategy incorporates an electricity-cooling-gas demand response (DR) model to optimize the load profile and introduces a stepped carbon trading (SCT) mechanism. Furthermore, a model for comprehensive EE is constructed using the EE coefficient method. Finally, a multi-objective optimization framework for IES is developed, integrating system economic cost, low-carbon operation, and comprehensive EE. Results: Case studies demonstrate that the proposed optimal scheduling scheme enhances the system's operational economy while simultaneously balancing low-carbon and high-efficiency performance. Discussion: The proposed DRL-based multi-objective coupled optimization model significantly reduces total operating costs (TOC), effectively cuts overall carbon emissions (CE), and improves comprehensive energy utilization efficiency, thereby achieving low-carbon and economic operation while ensuring high operational efficiency. By incorporating electricity-cooling-gas demand response models for load optimization and introducing a stepped carbon pricing mechanism, along with establishing an integrated EE model using the EE coefficient method, the study enables synergistic supply-demand optimization and substantially enhances EE. Conclusion: The adopted Deep Deterministic Policy Gradient (DDPG) algorithm effectively addresses the highly nonlinear, multi-constrained, and strongly coupled complexities of IES, successfully identifying balanced optimal solutions among the three interdependent objectives—economic cost, CE, and system efficiency—demonstrating its strong potential and effectiveness in solving such multi-objective collaborative optimization problems.
Chen et al. (Mon,) studied this question.