Multi-target flexible job shop scheduling method and device based on deep reinforcement learning

Abstract

The invention discloses a multi-target flexible job shop scheduling method and device based on deep reinforcement learning, and relates to the technical field of dynamic scheduling. The method comprises the steps: reading a production line state feature vector at a current rescheduling moment, and inputting the production line state feature vector into a trained target strategy network of a targetintelligent agent to obtain a scheduling target; inputting the production line state feature vector and the scheduling target into the workpiece strategy network of the trained workpiece intelligentagent to obtain a workpiece assignment rule, and inputting the workpiece assignment rule into the machine strategy network of the trained machine intelligent agent to obtain a machine distribution rule; selecting a to-be-machined workpiece according to the workpiece assignment rule, selecting a machining machine according to the machine allocation rule, and machining the next procedure of the to-be-machined workpiece through the machining machine. Therefore, different optimization objectives, workpiece assignment rules and machine allocation rules are intelligently selected according to the state of the production line at different rescheduling moments, and multi-objective collaborative optimization and complete real-time, autonomous and unmanned intelligent factories are realized.

Description

technical field [0001] The invention relates to the technical field of dynamic scheduling, in particular to a multi-objective flexible job shop scheduling method and device based on deep reinforcement learning. Background technique [0002] In related technologies, most of the multi-objective flexible job shop dynamic scheduling methods are based on simple scheduling rules or meta-heuristic algorithms. Simple scheduling rules mainly include first-come-first-out (FIFO for short), shortest delivery time first ( Earliest due date, referred to as EDD), the longest remaining processing time priority (Most remaining processing time, referred to as MRPT), etc., which select a workpiece to process on a machine at each rescheduling time, which has the advantage of high real-time It can respond to uncertain events immediately, but its disadvantage is that it is short-sighted and cannot get a better scheduling plan in the long run, and a single scheduling rule is often applicable to a ...

AI Technical Summary

Problems solved by technology

Storing such a large and complex Q table will not only introduce a huge computational burden, but also introduce a large number of useless states that have never been experienced, which reduces the efficiency of the algorithm, and the existing dynamic scheduling algorithms based on reinforcement learning often only consider a single Optimizing objectives (e.g. total tardiness), unable to achieve multi-objective optimization

Images