Application of multi-objective optimization algorithm in carbon management of electric-carbon coupled system

Abstract

In the context of deep integration of carbon emission regulation in the power system, the synergistic optimization of carbon cost and dispatch economy becomes the core challenge of the multi-objective scheduling problem. A distributed optimization framework integrating improved multi-objective evolutionary algorithm and deep reinforcement learning strategy is constructed, which systematically integrates Pareto sorting, adaptive perturbation, Actor-Critic decision network and master-slave scheduling mechanism to realize the joint optimization of carbon quota adjustment and real-time power generation strategy for the electric-carbon coupled system. In the IEEE-118 node grid simulation, the constructed model reduces the average scheduling cost by 2.8%, carbon emission by 10.3%, and the number of convergence rounds by 43.0% compared with the NSGA-II algorithm, which demonstrates significant solution efficiency and strategy stability. The system adopts GPU acceleration and asynchronous caching mechanism to maintain 92.6% resource utilization when the number of node concurrency is ≥8. Comparative analysis shows that the fusion algorithm has better solution set distribution and scheduling adaptability in non-convex objective space. This method can provide an engineering-deployable intelligent optimization tool for carbon scheduling for large-scale multi-energy systems, which is of value for dissemination.