Dynamic scheduling and resource allocation of green supply chain using deep deterministic policy gradient algorithm

Abstract

With regard to the problem of resources miscallocation and unexpected emissions peaks in classical supply chains that operate in non-stationary environments and face sudden disturbances, this paper presents a model for dynamic scheduling and resource allocation of green supply chain using the deep deterministic policy gradient approach. The study constructs a multi-objective reward function in a continuous state space, integrating dynamic carbon footprint accounting, comprehensive logistics costs, and safety stock deviation penalties, based on Markov decision processes. A multi-agent collaborative control architecture under a centralized training and decentralized execution paradigm is designed. Quantitative evaluation in a large-scale supply chain simulation environment shows that, while maintaining a system resilience score of 94.6, the proposed method reduces the overall logistics scheduling cost by 18.5%, the total carbon emissions per unit cycle by 25.3%, achieves a resource allocation accuracy of 96.4%, compresses the end-to-end single decision latency to 12.4 ms, and controls the long-term cost volatility within 1.5%. This method balances environmental impact and economic benefits, providing an end-to-end computational model for low-carbon collaborative scheduling of high-concurrency resources.