Abstract: Aiming at minimizing the time difference in synchronized processing of various components of a product and the makespan of all the jobs, a rolling horizon dynamic scheduling policy based on coupled transient chaotic neural network is investigated for a one-of-a-kind production system. A novel energy function expression, which includes the two objectives, the constraints of operation precedence and resource sharing, is constructed. Because there are two types of variables (i.e., the real type and the Boolean type) in the energy function, the transient chaotic neural network is decomposed into two coupled subnets to handle the two types of neurons respectively. A negative self-feedback chaotic item is included in the dynamic equation to improve the ability of searching the global optimum. A dynamic scheduling policy based on rolling horizon decomposition is proposed to adapt to the continuous changing of synchronization level during the processing. Simulation results indicate that the proposed algorithm has good global optimization capability and can improve the computational efficiency significantly.

Key words: Dynamic scheduling, Energy function, Neural network, Rolling horizon decomposition, Synchronous processing