The application discloses a distributed
hybrid flow shop dynamic scheduling method based on a learnable iterative
greedy algorithm, relates to the
flow shop scheduling technical field, and aims at a dynamic workpiece arrival and
machine fault scene, takes minimizing the maximum
completion time as a core target, is based on three coupled subproblems of a distributed
hybrid flow shop, adopts a two-dimensional coding strategy to represent a factory distribution and a workpiece sequencing scheme, combines a first available
machine rule and a
first come first served rule to construct a decoding mechanism, designs a multi-level neighborhood operation set as an action space of an agent, deals with emergent events of the dynamic workpiece arrival and the
machine fault, extracts state features as input by using a long short-
term memory network, trains the agent by using a proximal policy optimization
algorithm and an experience replay mechanism, dynamically adjusts a
decision strategy, makes the agent self-adaptively select a rescheduling, a destruction reconstruction and a local search strategy, balances global exploration and local development, and finally improves scheduling quality.