Abstract: A real supply chain system is composed of nodes with autonomous learning ability. The interactions among the nodes make the supply chain system produce complex nonlinear dynamics behaviors. In view of a supply chain system with learning ability in discrete-type manufacturing industry, the influence of learning policy on the dynamics behaviors of the system is studied. By calculating the distribution in ordering parameters space of the largest Lyapunov exponent and average cost per period, the relation between the distributions of nonlinear dynamics behavior modes and operating cost is analyzed for the manufacturer with learning ability. The study shows that the reinforcement learning policy used by the manufacturer plays a remarkable role in reducing average value of the cost distribution, and has obvious effect on the topology distribution of the complex dynamics behavior modes.

Key words: Nonlinear dynamics behavior, Reinforcement learning, Supply chain system