An effective MCTS-based algorithm for minimizing makespan in dynamic flexible job shop scheduling problem

Abstract

In the past several decades, most of the research methods are designed to solve the static flexible job shop scheduling problem. However, in real production environments, some inevitable dynamic events such as new jobs arrival and machine breakdown may occur frequently. In this paper, we study a dynamic flexible job shop scheduling problem (DFJSP) considering four dynamic events, which are new jobs arrival, machine breakdown, jobs cancellation and change in the processing time of operations. A rescheduling method based on Monte Carlo Tree Search algorithm (MCTS) is designed to solve the proposed DFJSP with the objective of minimizing the makespan. Several optimization techniques such as Rapid Action Value Estimates heuristic and prior knowledge are adopted to enhance the performance of the MCTS-based rescheduling method. The response time to dynamic events is critical in DFJSP but has not been solved very well. To greatly reduce the response time to dynamic events, when dynamic events occur, multiple continuous specified time windows are designed for the proposed method, according to which the corresponding subsequent partial schedule for the remaining unprocessed operations is progressively generated. Some experiments have been conducted to compare the proposed method with the commonly used completely reactive scheduling methods and the GA-based rescheduling method. The experiment results indicate that the proposed method is an efficient and promising method for dynamic scheduling both on solution quality and computation efficiency.