Search

CN-121998347-A - Hybrid flow shop scheduling method based on dynamic arrival of power battery module workpiece

CN121998347ACN 121998347 ACN121998347 ACN 121998347ACN-121998347-A

Abstract

The invention relates to a mixed flow shop scheduling method based on dynamic arrival of a workpiece of a power battery module, which comprises the following steps of S1, constructing an order model and a machine model considering faults, wherein the order model is dynamically arrived according to uncertainty in a production shop of the power battery module, S2, constructing a mixed flow shop scheduling mathematical model taking minimized overdue time as an optimization target according to production constraint of the power battery module and combining the order model and the machine model considering faults, S3, designing state characteristics, scheduling rules and rewarding functions according to real-time information of the workshop based on the mixed flow shop scheduling mathematical model, S4, solving an optimal scheduling scheme based on Dueling DQN network, outputting the optimal scheduling rules according to the state characteristics of S3, arranging the module workpiece to the machine, and updating scheduling states. The invention realizes the efficient dispatching of the power battery module workshops in a dynamic environment.

Inventors

  • HUANG JIE
  • CHEN CHENG
  • HUANG JINGLI
  • LIU SHANGKUN

Assignees

  • 福州大学

Dates

Publication Date
20260508
Application Date
20260127

Claims (10)

  1. 1. The hybrid flow shop scheduling method based on dynamic arrival of the power battery module workpiece is characterized by comprising the following steps: Step S1, constructing an order model for dynamic arrival of a workpiece and a machine model considering faults according to uncertainty in a power battery module production workshop; Step S2, according to the production constraint of the power battery module, combining an order model dynamically reached by a workpiece and a machine model considering faults, and establishing a mixed flow shop scheduling mathematical model taking the minimized overdue time as an optimization target; step S3, based on a mixed flow shop scheduling mathematical model, designing state characteristics, scheduling rules and rewarding functions according to real-time information of the shop; and step S4, solving an optimal scheduling scheme based on Dueling DQN networks, outputting an optimal scheduling rule according to the state characteristics of the step S3, arranging the module workpiece to a machine, and updating the scheduling state.
  2. 2. The method for scheduling hybrid flow shop based on dynamic arrival of power battery module workpieces according to claim 1, wherein the order model is constructed based on dynamic arrival characteristics of the workpieces in power battery module production, specifically, the arrival of the power battery module workpieces in the shop satisfies a poisson process, each workpiece Is a random variable Arrival time interval Obeying an exponential distribution: (1) (2) Wherein, the Is the average arrival time of the workpiece, Is the workpiece index.
  3. 3. The hybrid flow shop scheduling method based on dynamic arrival of power battery module workpieces according to claim 2, wherein the machine model is based on the failure start time and end time of the power battery module shop machine, specifically as follows: failure initiation of a machine And end time As random variable, machine Is a fault interval of (a) And repair time Obeying an exponential distribution: (3) (4) (5) (6) Wherein, the Is the machine number of the machine, and the machine number, Is the number of times of failure, Is the mean time to failure of the machine, Is the machine average repair time.
  4. 4. The hybrid flow shop scheduling method based on dynamic arrival of power battery module workpieces according to claim 1, wherein the production objective is to minimize the total overdue time of all the workpieces, in particular as follows: (7) in the formula, For the total number of the workpieces, the total delay time of all the workpieces is calculated by each workpiece Is the time of arrival of (2) And finishing time Calculating; the mixed flow shop scheduling mathematical model includes the following constraints: workpiece arrival constraint, namely, the moment that each workpiece arrives at a workshop is different, and the workpieces can be processed only after arriving at the workshop: (8) Wherein, the Is a workpiece First procedure Is a start time of (2); machine fault constraints for machines The machine cannot be in a fault state during the scheduled and machining process: (9) Wherein, the Is the working procedure In machines The processing time is longer than the processing time, Is the working procedure Is used for the start time of (1), Is a decision variable, when working procedure Is assigned to machines And is 1 when it is, and is 0 otherwise, Is a decision variable, working machine And 0 at the time of failure, and 1 otherwise.
  5. 5. The hybrid flow shop scheduling method based on dynamic arrival of power battery module workpieces according to claim 1, wherein the state space comprises seven characteristics extracted from module shop workpieces, machines and working procedures, wherein the seven characteristics comprise workpiece characteristics, average finishing rate of the workpieces, standard deviation of workpiece finishing rate and average delay rate of the workpieces, machine characteristics, average machine utilization, standard deviation of machine utilization and average failure rate of the machines, and working procedure characteristics and average delay rate of the working procedures.
  6. 6. The hybrid flow shop scheduling method based on dynamic arrival of power battery module workpieces according to claim 5, wherein the scheduling rules comprise five compound scheduling rules, which are respectively: The method comprises the steps of (1) screening a workpiece set to be scheduled, screening out predicted delay workpieces, selecting the workpiece with the largest delay time from all delay workpieces, traversing all machines to be scheduled, and selecting the machine with the shortest processing time from the current machine set; the method comprises the steps of (1) screening a workpiece set to be scheduled, screening out expected delay workpieces, selecting the workpiece with the largest delay time from all delay workpieces, traversing all machines to be scheduled, and selecting the machine with the lowest failure rate from the current machine set; the compound scheduling rule 3 is used for screening a workpiece set to be scheduled, screening out predicted delay workpieces, selecting the workpiece with the largest delay time from all delay workpieces, traversing all machines to be scheduled, and selecting the machine with the lowest utilization rate from the current machine set; the compound scheduling rule 4 is used for screening a workpiece set to be scheduled, screening out predicted delay workpieces, selecting the workpiece with the largest work rate from all delay workpieces, traversing all machines to be scheduled, and selecting the machine with the shortest processing time from the current machine set; the compound scheduling rule 5 is that firstly, a workpiece set to be scheduled is screened, predicted delay workpieces are screened, the workpiece with the largest work rate is selected from all delay workpieces, then all machines to be scheduled are traversed, and the machine with the lowest failure rate is selected from the current machine set; and a compound scheduling rule 6, firstly screening a workpiece set to be scheduled, screening out predicted delay workpieces, selecting the workpiece with the largest work rate from all delay workpieces, traversing all machines to be scheduled, and selecting the machine with the lowest use rate from the current machine set.
  7. 7. The hybrid flow shop scheduling method according to claim 6, wherein the reward function is related to a delay time, and the delay time at the current moment is lower than the forward reward at the next moment in order to achieve the minimum delay time, and otherwise, the reward is given, so that the scheduling system is encouraged to schedule the work pieces with tension in delivery period preferentially: (10) Wherein, the Is the total delay time of all workpieces in the workshop at the current moment.
  8. 8. The hybrid flow shop scheduling method based on dynamic arrival of power battery module workpieces according to claim 1, wherein the Dueling DQN network comprises a public feature extraction layer, a shunt layer and an aggregation layer which are sequentially connected, the public feature extraction layer comprises a plurality of full-connection layers and is used for extracting high-order features of state feature vectors, the shunt layer comprises a state value branch and an action advantage branch which are arranged in parallel, the state value branch is used for outputting a state value function of a current state, the action advantage branch is used for outputting an advantage function of each scheduling action, and the aggregation layer is used for conducting aggregation calculation on the state value function and the advantage function to obtain a Q value of each scheduling action, and the Q value is shown as a formula (13): (11) Wherein, the For the seven-dimensional state feature vector at the current time, In order to select the action to be scheduled, Is a parameter of the network full-connection layer, And Parameters of unique layers of the value stream and the dominant stream respectively, For the size of the motion space, Is a function of the state-value, Is an action dominance function.
  9. 9. The hybrid flow shop scheduling method based on dynamic arrival of power battery module workpieces according to claim 8, wherein the parameter update of the target network of Dueling DQN network adopts a soft update mechanism, specifically comprising the steps of obtaining the current weight parameter of the current weight parameter target network of the main network after each parameter iteration update of the main network, calculating the linear weighted sum of the weight parameter of the main network and the current weight parameter of the target network by using a preset soft update coefficient, assigning the linear weighted sum to the target network, and enabling the parameters of the target network to smoothly approximate to the parameters of the main network, wherein the specific calculation of the soft update is as follows: (12) Wherein, the In order to update the target network parameters, Is a preset soft update coefficient and has a value range of 。
  10. 10. The hybrid flow shop scheduling method based on dynamic arrival of power battery module workpieces according to claim 1, wherein the Dueling DQN network adopts an attenuation mechanism Greedy strategy training, which comprises presetting an exploration rate The method comprises the steps of (1) maintaining a higher exploration rate at the initial stage of model training, enabling a scheduling agent to randomly select a scheduling rule with a higher probability to explore a global state space, gradually reducing the exploration rate according to the attenuation step length along with the increase of training iteration times until the exploration rate reaches the minimum value, enabling the scheduling agent to select the scheduling rule with the maximum Q value with the higher probability at the later stage of model training, and carrying out scheduling decision by utilizing the learned optimal strategy.

Description

Hybrid flow shop scheduling method based on dynamic arrival of power battery module workpiece Technical Field The invention relates to the technical field of dynamic workshop scheduling, in particular to a hybrid flow workshop scheduling method based on dynamic arrival of power battery module workpieces. Background The production of the power battery module is a key link in the new energy automobile industry chain, and the production process involves a plurality of complex procedures. Workshop scheduling is used as a core of production management, and directly determines production efficiency and resource utilization rate. In conventional shop scheduling studies and applications, it is generally assumed that the production environment is static, i.e., all process task information is known and fixed before scheduling begins. Based on this assumption, the traditional rule-based scheduling algorithm, mathematical programming method and conventional heuristic algorithm can calculate a more ideal scheduling scheme. However, in an actual power battery module production site, the production environment has high dynamics and uncertainty. The method is mainly characterized in that the arrival of the workpieces is random, production orders are often dynamically issued, the arrival time of the workpieces to a production line is unpredictable, the machine resources are unreliable, and sudden faults can occur in production equipment. In the face of the dynamic workshop environment, the existing static scheduling method has the obvious limitations that firstly, the real-time self-adaption capability is lacked, a pre-scheduling scheme generated by the traditional method is often invalid when encountering sudden disturbance, the production priority cannot be adjusted in real time, secondly, the rescheduling is delayed, the optimization method relying on global recalculation consumes long time, and the requirement of quick response of a production field is difficult to meet. Therefore, there is a need for an efficient scheduling method for power battery module workshops that can sense environmental changes in real time, handle machine failures and dynamic workpiece arrival. Disclosure of Invention In order to solve the problems, the invention aims to provide a hybrid flow shop scheduling method based on dynamic arrival of a workpiece of a power battery module, which solves the technical problems that the scheduling rule is short, the calculation time of a heuristic algorithm is too long, the real-time performance and the overall optimization performance cannot be considered when the existing power battery module shop scheduling technology faces complex working conditions such as dynamic arrival of the workpiece and random failure of a machine. In order to achieve the above purpose, the present invention adopts the following technical scheme: A hybrid flow shop scheduling method based on dynamic arrival of power battery module workpieces comprises the following steps: Step S1, constructing an order model for dynamic arrival of a workpiece and a machine model considering faults according to uncertainty in a power battery module production workshop; Step S2, according to the production constraint of the power battery module, combining an order model dynamically reached by a workpiece and a machine model considering faults, and establishing a mixed flow shop scheduling mathematical model taking the minimized overdue time as an optimization target; step S3, based on a mixed flow shop scheduling mathematical model, designing state characteristics, scheduling rules and rewarding functions according to real-time information of the shop; and step S4, solving an optimal scheduling scheme based on Dueling DQN networks, outputting an optimal scheduling rule according to the state characteristics of the step S3, arranging the module workpiece to a machine, and updating the scheduling state. Further, the order model is constructed based on dynamic arrival characteristics of the workpieces in the production of the power battery module, and specifically comprises the steps that arrival of the workpieces of the power battery module in a workshop meets a poisson process, and each workpieceIs a random variableArrival time intervalObeying an exponential distribution: (13) (14) Wherein, the Is the average arrival time of the workpiece,Is the workpiece index. Further, the machine model is based on the fault start time and the fault end time of the power battery module workshop machine, and specifically comprises the following steps: failure initiation of a machine And end timeAs random variable, machineIs a fault interval of (a)And repair timeObeying an exponential distribution: (15) (16) (17) (18) Wherein, the Is the machine number of the machine, and the machine number,Is the number of times of failure,Is the mean time to failure of the machine,Is the machine average repair time. Further, the goal of the process is to minimize the total time