Dynamic scheduling method for digital twin-driven flexible job shop

Abstract

The invention belongs to the field of dynamic flexible job shop scheduling, and discloses a digital twin-driven flexible job shop dynamic scheduling method, which comprises the following steps: acquiring real-time state data of a physical shop, capturing a dynamic disturbance event, and transmitting the real-time state data and the dynamic disturbance event to a virtual shop; the virtual workshop calls the dynamic scheduling strategy to generate a new scheduling scheme and transmits the new scheduling scheme to the physical workshop, the dynamic scheduling strategy generates an initial scheduling scheme according to the static workshop parameters and judges whether rescheduling is triggered or not, if rescheduling is triggered, the static workshop parameters are updated based on the real-time state data and the dynamic disturbance event to obtain real-time workshop parameters, and the real-time workshop parameters are sent to the physical workshop. Generating a rescheduling scheme by using an improved harmony search algorithm, and taking the rescheduling scheme as a new scheduling scheme; otherwise, taking the initial scheduling scheme as a new scheduling scheme; and the physical workshop executes flexible job-shop dynamic scheduling according to the new scheduling scheme. According to the invention, the distribution uniformity, diversity and robustness of the dynamic scheduling scheme are improved.

Description

Technical Field

[0001] This invention belongs to the field of dynamic flexible job shop scheduling, specifically relating to a digital twin-driven dynamic scheduling method for flexible job shops. Background Technology

[0002] The Job Shop Scheduling Problem (JSSP) is a crucial foundation for optimizing manufacturing resource allocation and intelligent manufacturing processes. It is also a typical NP-hard combinatorial optimization problem that has long attracted widespread attention. The Dynamic Flexible Job Shop Scheduling Problem (DFJSSP), a complex extension of JSSP, can more accurately characterize the actual production environment by comprehensively considering machine failures, emergency job insertions, and other unforeseen dynamic disturbances. In recent years, academia and industry have conducted extensive research on DFJSSP, striving to explore scheduling methods that combine solution quality and real-time response capabilities. However, within the (Industry 5.0) I5.0 framework, incorporating the dynamic characteristics of workers, which have significant uncertainties and are difficult to quantify, into the scheduling modeling and decision-making process further exacerbates the complexity of the problem, significantly increasing the difficulty of solving DFJSSP.

[0003] As a core supporting technology for the digital transformation of manufacturing, Digital Twin (DT) provides a new technical path for the dynamic scheduling needs of complex manufacturing systems by achieving deep integration of physical and virtual spaces. Currently, DT-based DFJSSP modeling and solving has become one of the important research directions in the field of intelligent manufacturing. Existing research has focused on addressing machine- or order-related emergencies, which are generally considered the most significant and common dynamic events affecting the production process. However, worker factors (such as learning effects and fatigue) influence scheduling decisions in production systems, but these factors still lack systematic modeling in existing flexible workshop scheduling research. While existing research occasionally touches upon worker factors, few studies have deeply integrated them with DT technology to achieve real-time perception and closed-loop feedback of worker status.

[0004] Existing research largely focuses on dynamic events such as the insertion of new jobs, paying insufficient attention to personnel uncertainties such as worker absenteeism in flexible work workshops. Although some literature explores worker factors, few studies have taken into account both social and environmental objectives within the I5.0 framework. Furthermore, the application of DT (Digital Transformation Technology) in characterizing deeper human factors such as worker fatigue and their impact on scheduling efficiency is still lacking, and its technological potential urgently needs further exploration.

[0005] In actual production, worker factors (such as the learning effect and fatigue accumulation) are key variables affecting production efficiency. On the one hand, the learning effect shortens processing time through experience accumulation, improving production efficiency while reducing operational error rates; on the other hand, physical fatigue caused by prolonged high-intensity work weakens work capacity, thereby inhibiting productivity. Existing research shows that reasonable production scheduling can effectively alleviate worker fatigue accumulation, thereby improving production efficiency and worker satisfaction.

[0006] Existing research has gradually incorporated learning effects and fatigue factors into scheduling optimization problems. Current techniques have constructed scheduling models that consider heterogeneous worker fatigue for the DFJSSP (Dead-Leadership-Side-Side) scheduling problem with dual resource constraints. For the scheduling scenario of a two-machine flow shop, the influence mechanism of learning effects has been considered, combined with specific production time constraints, to improve production efficiency while alleviating worker fatigue. For the work-interval scheduling problem, characteristics such as worker fatigue effects and dynamic fatigue recovery have been considered. However, due to the dual constraints of the dynamic time-varying nature of the production environment and the need for real-time assessment, accurately quantifying learning effects and worker factors such as fatigue remains a significant challenge. Summary of the Invention

[0007] The purpose of this invention is to provide a dynamic scheduling method for flexible workshops driven by digital twins, which improves the distribution uniformity, diversity and robustness of dynamic scheduling schemes.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A digital twin-driven dynamic scheduling method for flexible workshops, which constructs a virtual workshop that is mapped in real time to the physical workshop using digital twin technology, includes:

[0010] Collect real-time status data of the physical workshop and capture dynamic disturbance events. Transmit the real-time status data and dynamic disturbance events to the virtual workshop. The real-time status data includes the real-time status of machines, operations and workers. The dynamic disturbance events include worker absence and machine failure.

[0011] The virtual workshop invokes a dynamic scheduling strategy to generate a new scheduling scheme and transmits it to the physical workshop. The dynamic scheduling strategy generates an initial scheduling scheme based on static workshop parameters and determines whether rescheduling is triggered. If rescheduling is triggered, the static workshop parameters are updated based on real-time status data and dynamic disturbance events to obtain real-time workshop parameters. An improved harmony search algorithm is then used to generate a rescheduling scheme, which is used as the new scheduling scheme. Otherwise, the initial scheduling scheme is used as the new scheduling scheme.

[0012] The physical workshop implements dynamic scheduling of flexible work workshops according to the new scheduling scheme.

[0013] Several alternative methods are provided below, but they are not intended as additional limitations on the overall solution above. They are merely further additions or optimizations. Provided there are no technical or logical contradictions, each alternative method can be combined individually with respect to the overall solution above, or multiple alternative methods can be combined with each other.

[0014] Preferably, the determination of whether a rescheduling is triggered includes:

[0015] Based on real-time status data and dynamic disturbance events, detect whether there are any worker absences. If so, trigger rescheduling and set the day as the starting point for the next periodic rescheduling interval; otherwise, continue to determine whether the periodic rescheduling interval has been reached.

[0016] If the periodic rescheduling interval has not been reached, rescheduling will not be triggered; otherwise, it will continue to determine whether worker absence was detected in the previous period.

[0017] If no worker absences were detected in the previous period, a rescheduling is triggered; otherwise, the periodic rescheduling interval is updated and a rescheduling is triggered.

[0018] Preferably, the step of detecting whether there is worker absence based on real-time status data and dynamic disturbance events includes:

[0019] If the dynamic disturbance event includes a worker absence event, it indicates that a worker absence has been detected; otherwise, the fatigue level of each worker who has not experienced a worker absence event is calculated based on the real-time status data, and the worker's fatigue recovery time is obtained based on the fatigue status. If the fatigue recovery time is greater than the rest threshold, it is determined that a worker is absent; otherwise, no worker absence is detected.

[0020] As a preferred embodiment, the improved harmony search algorithm uses a three-layer representation scheme for the solution, including a process sequence layer, a machine allocation layer, and a worker allocation layer.

[0021] The process sequence layer records a process sequence vector. The process sequence vector represents the execution order of each job in the order of job indices. Each number in the process sequence vector represents a job number, and the number of times the same job number appears is recorded. Indicates the first in the assignment One process;

[0022] The machine allocation layer records the machine allocation vector, and each number in the machine allocation vector represents the machine number assigned to the corresponding position in the process sequence vector.

[0023] The worker allocation layer records a worker allocation vector, where each number in the worker allocation vector represents the worker number assigned to the corresponding position in the process sequence vector.

[0024] Preferably, in the improved harmony search algorithm, the harmony memory is initialized using a best-point set initialization strategy, and the harmony memory is as follows:

[0025]

[0026]

[0027] In the formula, Represents the harmonic memory bank. Indicates the size of the harmonic memory bank. The dimensions representing the harmonic memory bank, For the first in the harmony memory bank The first harmony Components of each dimension , , Indicates the first The upper bound of each dimension, Indicates the first The lower bound of each dimension, express The set of best points in the unit cube of dimensional Euclidean space is formed by... indivual A point set composed of dimensional points.

[0028] Preferably, the improved harmony search algorithm updates the harmony memory bank through improvisation, specifically including:

[0029] Generate a first random number between [0,1]. If the first random number is less than the harmony memory consideration rate, then randomly select a harmony from the initialized harmony memory bank. And select the current optimal harmony from the harmony memory as Otherwise, a new harmony will be randomly generated as the final new harmony;

[0030] Employing priority process crossover operators for harmony and optimal harmony The process sequence vectors are crossed, and the machine allocation vectors in the crossover result are mutated to generate a preliminary new harmony;

[0031] Generate a second random number between [0,1]. If the second random number is less than the harmony adjustment rate, perform improvisational fine-tuning on the initial new harmony to obtain the final new harmony; otherwise, do not update the harmony memory and end the process.

[0032] If the objective function value of the final new harmony is better than the worst objective function value in the harmony memory, then the harmony corresponding to the worst objective function value in the harmony memory is replaced with the final new harmony, and the update ends; otherwise, the final new harmony is discarded, the harmony memory is not updated, and the update ends.

[0033] Preferably, the improved harmony search algorithm, after updating the harmony memory, employs a t-distribution-based random perturbation technique to optimize the harmony, including:

[0034] Selectively choose a portion of the harmony from the harmony memory bank as the harmony to be perturbed;

[0035] Calculate the function value of the t-distribution with the number of iterations as the degrees of freedom;

[0036] The perturbed harmony is obtained by weighting each harmony to be perturbed using the function value of the t-distribution and then superimposing the weighted harmony with the corresponding harmonic to be perturbed.

[0037] Preferably, the improved harmony search algorithm uses harmony as its objective function to minimize the maximum completion time, total carbon emissions, and maximum cumulative worker fatigue.

[0038] This invention provides a digital twin-driven dynamic scheduling method for flexible workshops. First, a dynamic multi-objective DFJSSP mathematical model oriented towards I5.0 is constructed. This model simultaneously considers real-time worker performance and machine load, treating worker absence as a key dynamic disturbance, aiming to minimize maximum completion time and carbon emissions while maximizing worker satisfaction. Second, to address the limitation of traditional scheduling lacking real-time awareness, a DT-based dynamic scheduling framework is designed, enabling real-time monitoring of worker performance and machine load status in the flexible workshop. Based on this, a hybrid scheduling strategy integrating event-driven and periodic rescheduling is proposed to achieve rapid adaptation to sudden disturbances. Furthermore, to address the challenge of high-dimensional solution space optimization, an improved harmony search algorithm (IHSA) is proposed. This algorithm introduces a three-layer encoding scheme to achieve efficient adaptation between the algorithm and the scheduling model, and combines a best-point set initialization strategy and a t-distribution perturbation mechanism to enhance the algorithm's global convergence and local exploitation capabilities. Simulation experiments, statistical analysis, and analysis of actual engineering cases show that the algorithm has significant advantages in terms of the uniformity, diversity, and robustness of the solution set distribution, effectively verifying its engineering application value in complex dynamic environments. Attached Figure Description

[0039] Figure 1 This is a diagram of the overall architecture of the digital twin of the present invention;

[0040] Figure 2 This is a schematic diagram of the digital twin dynamic scheduling process of the present invention;

[0041] Figure 3 This is a flowchart of the dynamic scheduling method for a flexible workshop driven by a digital twin according to the present invention;

[0042] Figure 4 This is a flowchart of the dynamic scheduling strategy of the present invention;

[0043] Figure 5 This is a flowchart of the improved harmony search algorithm of the present invention;

[0044] Figure 6 This is a schematic diagram of an embodiment of the three-layer representation scheme of the present invention based on process sequence, machine allocation, and worker allocation;

[0045] Figure 7 This is a schematic diagram of the priority operation crossover process used in the process sequence of the present invention;

[0046] Figure 8 The figure shows the experimental results of the influence of different algorithm parameters on the maximum completion time in the experiment of this invention;

[0047] Figure 9 This is a convergence diagram of each algorithm in the experiment of this invention when solving actual engineering examples;

[0048] Figure 10 This is a schematic diagram of a scheduling scheme that takes into account worker absenteeism in the experiment of this invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0051] Currently, how to achieve real-time perception and dynamic response to DFJSSP (Workshop-Departmental Workshop-Solution) involving worker factors based on Data Technology (DT) has become a key scientific problem that urgently needs to be solved. Therefore, this embodiment treats worker absence as an explicit dynamic disturbance event and incorporates it into the DFJSSP modeling framework, aiming to balance economic benefits with the improvement of environmental and social benefits. Furthermore, this embodiment constructs an overall DT architecture and proposes a novel hybrid scheduling strategy. Simultaneously, an improved harmony search algorithm (IHSA) is designed and introduced to solve DFJSSP in a DT workshop environment.

[0052] DFJSSP is an important extension of the classic JSSP. The DFJSSP considering worker factors studied in this embodiment can be formally described as follows: Assuming there are One assignment It is necessary to Taiwanese machine The upper processing is provided, and it is equipped with the following in the initial state: Available workers and the machine set and workers' group These are considered static shop floor parameters. For the DFJSSP problem studied, this embodiment aims to solve the following core decision-making problems: allocating suitable machines to each operation, determining the order of multiple operations on each machine, and assigning corresponding operators to each machine. Furthermore, this embodiment considers worker absences as a major dynamic disturbance event affecting the scheduling process. This embodiment also sets all operations, machines, and workers to be available at time 0; at any given time, a machine is only allowed to process one operation; at any given time, a worker is only allowed to operate one machine; at any given time, each operation is only allowed to be executed on one machine; and operations can be paused during processing.

[0053] The research objective of the DFJSSP problem in this embodiment is to maximize worker job satisfaction while minimizing the maximum completion time and total carbon emissions. These objectives consider not only economic benefits but also the comprehensive impacts on social and environmental dimensions. Given the negative correlation between worker fatigue and job satisfaction, this embodiment quantifies worker satisfaction by using maximum cumulative fatigue.

[0054] Therefore, the optimization objective function in this embodiment is defined as shown in formula (1):

[0055] (1)

[0056] Furthermore, the calculation methods for each sub-objective in formula (1) are defined as follows:

[0057] Minimize the maximum completion time, expressed as :

[0058] (2)

[0059] Minimize total carbon emissions, expressed as :

[0060] (3)

[0061] Minimize the maximum cumulative worker fatigue, expressed as :

[0062] (4)

[0063] According to existing research, fatigue accumulates exponentially over time, and its dynamic change process is shown in formula (5):

[0064] (5)

[0065] After a period of continuous rest, a worker's fatigue level is updated using the following formula:

[0066] (6)

[0067] The objective function in this embodiment is subject to the following constraints:

[0068] (7)

[0069] (8)

[0070] (9)

[0071] (10)

[0072] (11)

[0073] (12)

[0074] (13)

[0075] (14)

[0076] (15)

[0077] (16)

[0078] Among them, constraint (7) limits the completion time of any task to no more than the maximum completion time. Constraint (8) ensures the process priority order of each process within the same workpiece. Constraints (9) and (10) correspond to the non-overlapping resource constraints of machines and workers, respectively, that is, ensuring that the same resource can only process one process at a time and determining the processing order between processes. Constraints (11) and (12) respectively stipulate that any process must and is only allowed to be assigned to one machine and one worker for processing. Constraints (13)-(16) define the value range of all decision variables in the model, that is, they are all 0-1 variables.

[0079] in, For job indexing, ; For process index, ; For machine indexing, ; Index for workers, ; For the time (decision) point, ; For the first One assignment; For homework The number of processes in the process; For the first 1 machine; For the first A worker; For homework The One process; For homework Completion time; For homework The This process is done on the machine. Carbon emissions; For homework The The start time of each processing step; For homework The The processing completion time of each step; The maximum completion time for all tasks represents the maximum time to complete the task. The total carbon emissions generated by all processes and machines; The cumulative time from the decision time t to the start of the operation; For the first A worker at the decision point Cumulative fatigue over time; For the first The maximum cumulative fatigue level of an individual worker; For the first The rate at which an individual worker accumulates fatigue; For the first The fatigue recovery rate of individual workers; 0-1 decision variables: when the task The The process was assigned to the first The value is 1 when the machine is processing the data, and 0 otherwise. For 0-1 decision variables: if the task The The process was assigned to the first The value is 1 when a worker performs an operation, and 0 otherwise. For 0-1 decision variables: if the task The The first process is in the The value is 1 if the g-th process on the machine that precedes operation h is performed, and 0 otherwise. For 0-1 decision variables: if the task The The first process is the first The value is 1 if the g-th process, which precedes operation h, is performed by a worker; otherwise, it is 0.

[0080] This embodiment presents a digital twin-driven dynamic scheduling method for flexible workshops. It constructs a virtual workshop that is mapped in real-time to the physical workshop using digital twin technology and establishes a system architecture oriented towards DFJSSP. For example... Figure 1 As shown, this architecture extends the five-dimensional DT framework, which includes physical workshops, virtual workshops, DT data, visualization, and data flow. On one hand, a data management module is introduced as an intermediary layer to ensure real-time interaction between the virtual and physical workshops. On the other hand, the worker model (real-time synchronization of the dynamic changes of physical workers; state change data originates from sensors set up by workers in the physical workshop, such as wearable devices (smart bracelets monitoring heart rate / fatigue), RFID / UWB positioning (location), machine vision (posture / operational behavior), and equipment sensors (work interaction data), etc.; in this embodiment, it is mainly used to detect worker fatigue levels based on sensor data) and the scheduling module are integrated into the virtual workshop. This architecture also integrates real-time monitoring of worker performance, highlighting the human-centered nature of the DT system and its crucial supporting role in scheduling decisions. Furthermore, this architecture employs a real-time data-driven approach to execute a hybrid scheduling strategy specifically designed for disturbances such as worker absenteeism, thereby effectively enhancing the system's responsiveness.

[0081] The architecture mainly consists of three parts: physical workshop, virtual workshop, and intermediate data management module. The three parts work together to achieve seamless data flow and dynamic decision support.

[0082] The virtual workshop serves as the decision-making center of the DT (Data Technology) architecture, aggregating and integrating data from the physical workshop in real time through a data management module. The processed data continuously drives the virtual workshop's operation, supporting the simulation, monitoring, and analysis of worker performance. This embodiment embeds worker models within the virtual workshop to emphasize the critical impact of workers on the dynamic production system. Simultaneously, the virtual workshop integrates advanced scheduling models, strategies, and algorithms to support scheduling decisions.

[0083] The physical shop is the actual entity executing production activities, responsible for collecting real-time production data and executing scheduling plans. Its structure consists of three interconnected layers: a network layer, a resource layer, and an execution layer, which collaborate to coordinate and manage production tasks. The real-time data collected by the physical shop covers information such as machine performance, operational processes, and worker status (detected through sensors), providing more comprehensive data support for real-time system decision-making. Simultaneously, the scheduling plans generated by the virtual shop are ultimately implemented in the physical shop.

[0084] The data management module, by integrating three major functional components—data storage, data processing, and data visualization—plays a crucial role as a data intermediary within the system. The seamless collaboration among these components significantly enhances the overall system's adaptability and dynamic responsiveness within the perception-control cycle.

[0085] like Figure 2 As shown, the dynamic scheduling process based on data processing (DT) forms a closed-loop feedback mechanism of perception, analysis, decision-making, and control. Through real-time collaboration between the virtual workshop, data management module, and physical workshop, and combined with real-time data and optimization algorithms, this process achieves rapid adaptive response to production disturbances. This process has iterative and continuous feedback characteristics, thereby empowering the system's real-time coordination capabilities and significantly enhancing its adaptability and operational efficiency in dynamic production environments.

[0086] like Figure 3 As shown in the figure, the digital twin-driven flexible job shop dynamic scheduling method of this embodiment specifically includes the following steps:

[0087] Step 1: Collect real-time status data of the physical workshop and capture dynamic disturbance events. Transmit the real-time status data and dynamic disturbance events to the virtual workshop. The real-time status data includes the real-time status of machines, operations and workers. The dynamic disturbance events include worker absences and hidden potential risks (such as machine failures caused by long-term excessive machine load) that have an explicit impact on scheduling decisions.

[0088] Dynamic disturbance events are extracted from real-time status data. In this embodiment, real-time data involving machines, workers and operations in the physical workshop, dynamic disturbance events and parameter sets of the improved harmony search algorithm are continuously transmitted to the data management module for parsing and processing, and then transmitted to the virtual workshop.

[0089] Step 2: The virtual workshop invokes a dynamic scheduling strategy to generate a new scheduling scheme and transmits it to the physical workshop. The dynamic scheduling strategy generates an initial scheduling scheme based on static workshop parameters and determines whether rescheduling is triggered. If rescheduling is triggered, the static workshop parameters are updated based on real-time status data and dynamic disturbance events to obtain real-time workshop parameters. An improved harmony search algorithm is then used to generate a rescheduling scheme, which is adopted as the new scheduling scheme. Otherwise, the initial scheduling scheme is adopted as the new scheduling scheme. This process involves the reallocation of workpieces, machines, and workers to ensure that the scheduling goal is approached while satisfying all constraints.

[0090] To effectively respond to dynamic disturbances and minimize their impact on production systems, it is crucial to develop a suitable response mechanism. For example... Figure 4As shown, this embodiment proposes a hybrid scheduling strategy for DT (Digital Transmission) workshop environments that integrates event-driven and periodic rescheduling. This strategy combines event-driven rescheduling and periodic rescheduling. When the system does not detect any sudden disturbance events and the preset rescheduling period is reached, periodic rescheduling will be triggered; once a dynamic event is detected, the system will immediately initiate event-driven rescheduling to achieve a rapid response to disturbances. The hybrid scheduling strategy of this embodiment includes the following process:

[0091] Step 2.1: Generate periodic rescheduling intervals and generate initial scheduling schemes.

[0092] The periodic rescheduling interval in this step can be either a default value or calculated based on the total processing time and maximum completion time of all machines in the previous interval. Furthermore, the initial scheduling scheme is a static scheme generated by IHSA under static shop floor parameters; that is, the initial scheduling scheme does not consider worker absences or machine malfunctions.

[0093] Step 2.2: Detect whether there are any workers absent based on real-time status data and dynamic disturbance events. If so, trigger rescheduling (event-driven rescheduling) and set the current day as the starting point of the next periodic rescheduling interval, then proceed to step 2.5; otherwise, continue to determine whether the periodic rescheduling interval has been reached.

[0094] This embodiment's event-driven rescheduling method aims to dynamically adjust the scheduling scheme to maintain system stability in the event of worker absences. To avoid frequent triggering of this method, this embodiment sets clear triggering conditions: the system will trigger rescheduling when the recovery period for worker absences is long, may delay the overall delivery schedule, or when long-term leave significantly impacts production continuity; for short-term work interruptions caused by temporary fatigue or short-term leave, the system will not trigger rescheduling. The specific steps are as follows:

[0095] If the dynamic disturbance event includes a worker absence event (obtained from attendance records), it indicates that a worker absence has been detected; otherwise, the fatigue level of each worker who has not experienced a worker absence event is calculated based on the real-time status data, and the worker's fatigue recovery time is obtained based on the fatigue status. If the fatigue recovery time is greater than the rest threshold, it is determined that a worker is absent; otherwise, no worker absence has been detected.

[0096] Worker performance can be measured by the probability of worker operational errors, which is dynamically influenced by learning effects and fatigue levels, as shown in equations (17) and (18).

[0097] (17)

[0098] (18)

[0099] in, Indicates the first A worker at a certain time (decision point) The probability of operational errors; and The weights are used to balance the effects of learning effects and fatigue levels on the probability of worker operational errors. This represents the learning effect, with values ​​ranging from 0 to 1. This indicates the cumulative number of tasks. Indicates the learning rate; and They represent the first The cumulative fatigue level and maximum cumulative fatigue level of each worker. If a worker's error probability exceeds a certain threshold (e.g., the worst historical average of all workers), then the worker needs to be rested to recover to another threshold (e.g., the best historical average).

[0100] Based on formulas (17) and (18), the fatigue level corresponding to the probability of worker operation error can be derived, as shown in formula (19).

[0101] (19)

[0102] The rest time required for the current state to return to normal can be obtained by transformation formula (6), as shown in formula (20).

[0103] (20)

[0104] in, This indicates recovery time. Furthermore, according to the World Health Organization's recommendations, the rest allowance for manufacturing workers should be no less than 15% (at least 0.15 minutes of rest for every minute of work). Therefore, in this article, recovery time exceeds... The situation is classified as a longer recovery period, which means the worker is considered absent.

[0105] Step 2.3: If the periodic rescheduling interval has not been reached, rescheduling will not be triggered, and proceed to step 2.6; otherwise, continue to determine whether worker absence was detected in the previous period.

[0106] Step 2.4: If no worker absence was detected in the previous period, trigger rescheduling (periodic rescheduling) and proceed to step 2.5; otherwise, update the periodic rescheduling interval and trigger rescheduling (periodic rescheduling) and proceed to step 2.5.

[0107] Periodic rescheduling triggers and generates new scheduling schemes based on preset time intervals to address potential problems in the system. The effectiveness of this strategy is highly dependent on the interval setting: too short an interval can lead to frequent rescheduling, interfering with the stability of system operation; too long an interval may weaken the system's response time to dynamic events, causing the scheduling scheme to lag or even become disconnected from the actual production status. In this embodiment, the current periodic rescheduling interval setting depends on the total processing time and maximum completion time of all machines in the previous periodic rescheduling interval, as shown in formula (21):

[0108] (twenty one)

[0109] in, Indicates the length of the current periodic rescheduling time interval; This indicates the maximum completion time of the previous periodic rescheduling interval; This represents the total processing time of all machines within the previous periodic rescheduling interval. If an event-driven rescheduling was triggered in the previous interval, the time when the event-driven rescheduling occurred is taken as the starting point of the new periodic rescheduling interval.

[0110] In one instance, if the initially generated periodic rescheduling interval is 30 days, and worker absence is detected on day 5, event-driven rescheduling is triggered, and day 5 is used as the starting point for the next periodic rescheduling interval. This continues until day 30, when the periodic rescheduling interval is reached, ending the current periodic rescheduling interval. The maximum completion time from day 1 to day 30 is used as the maximum completion time and total processing time of all machines for the previous periodic rescheduling interval. The length of the new periodic rescheduling interval is calculated, and day 5 is used as the starting point for the new periodic rescheduling interval. This new periodic rescheduling interval is then triggered, and periodic rescheduling is initiated.

[0111] Step 2.5: Based on real-time status data and dynamic disturbance events, update the static workshop parameters to obtain real-time workshop parameters, and use the improved harmony search algorithm to generate a rescheduling scheme.

[0112] To adapt to real-time changes in the status of machines and workers, this embodiment uses worker absences determined from real-time status data and recorded in dynamic disturbance events to adjust the original worker set. The corresponding workers are removed to obtain the real-time available worker set. Simultaneously, the original machine set is adjusted based on machine faults recorded in the dynamic disturbance events. The corresponding machines are removed to obtain the set of machines available in real time. Based on the real-time workshop parameters of the set of available workers and machines, an improved harmony search algorithm is used to generate a rescheduling scheme.

[0113] The harmony search algorithm is a widely accepted metaheuristic algorithm, renowned for its efficient search process, concise algorithm structure, and relatively few algorithm parameters. The harmony search algorithm mainly involves the following four parameters: harmony memory size (… ), harmonic memory consideration rate ( ), pitch adjustment rate ( ) and bandwidth ( Here, the harmony memory size represents the number of solutions stored in the harmony memory; the harmony memory consideration rate represents the probability of selecting a harmony from the memory when generating a new solution; and the pitch adjustment rate represents the probability of fine-tuning the selected harmony, with the specific adjustment magnitude controlled by the bandwidth parameter. However, when this algorithm is applied to scheduling problems, several limitations remain. For example, it requires a high degree of diversity in the initial harmony memory and is prone to getting trapped in local optima. These limitations weaken its ability to find the global optimum to some extent. Therefore, to effectively address the challenges faced in solving the DFJSSP model, this embodiment proposes an IHSA, which implements the following three key enhancements on top of the basic HSA:

[0114] (1) A three-layer representation scheme is proposed to achieve efficient adaptation between the algorithm and the scheduling model.

[0115] (2) The optimal point set initialization strategy is adopted to initialize the harmony memory bank, so as to improve the uniformity and diversity of the initial population distribution and enhance the global search capability of the algorithm.

[0116] (3) A t-distribution perturbation mechanism is introduced into the basic HSA to expand the search range of harmony and prevent the algorithm from getting trapped in local optima. This mechanism not only improves the algorithm's optimization ability but also further improves the solution accuracy.

[0117] The specific procedures of IHSA are as follows: Figure 5 As shown.

[0118] Step 2.5.1: Initialize the harmony memory library.

[0119] The harmony memory, as a key storage structure, stores a set of candidate solutions (called harmony solutions), which guide the search process. Each harmony solution (called pitch) consists of multiple variables. This embodiment uses a best-point set initialization strategy for harmony memory initialization. This strategy generates a uniformly distributed set of points to cover the solution space, ensuring a uniform distribution of initial solutions and thus enhancing the algorithm's global search capability. The concept of the best-point set is as follows: Let... express The unit cube in Euclidean space express The point in the middle, It is a constant. The definition of the optimal point set is shown in formula (22):

[0120] (twenty two)

[0121] Wherein, the deviation satisfies , It is only dependent on and A constant (of any positive number). Best-case value. ,in, It is to satisfy The smallest prime number.

[0122] The initialization range of the harmony memory bank is determined using the optimal point set initialization strategy, as shown in formula (23):

[0123] (twenty three)

[0124] The first harmonies The calculation method for each component is shown in formula (24):

[0125] (twenty four)

[0126] In the formula, Represents the harmonic memory bank; Indicates the size of the harmony memory bank; The dimension representing the harmony memory is equal to the total number of decision variables; in this embodiment, this value is equal to... (I represents the total number of assignments); For the first in the harmony memory bank The first harmony Components of each dimension , , Indicates the first The upper bound of each dimension, Indicates the first The lower bound of each dimension, express The set of best points in the unit cube of dimensional Euclidean space is formed by... indivual A point set composed of dimensional points. In this embodiment... , .

[0127] Each harmony in the initial harmony memory is considered as a solution, i.e., a dynamic scheduling scheme. In this embodiment, the solution adopts a three-layer representation scheme, including the process sequence layer, the machine allocation layer, and the worker allocation layer.

[0128] The process sequence layer records the process sequence vector. The process sequence vector represents the execution order of each process in the order of the job index. Each number in the process sequence vector represents the job number and the number of times the same job number appears. Indicates the first in the assignment The machine allocation layer records the machine allocation vector, where each number represents the machine number assigned to the corresponding position in the process sequence vector; the worker allocation layer records the worker allocation vector, where each number represents the worker number assigned to the corresponding position in the process sequence vector.

[0129] To more clearly illustrate how the scheduling scheme is represented, Figure 6 An example is shown, involving 3 jobs, 9 processes, 5 machines, and 3 workers. The first and last rows represent the candidate workers and machines for the corresponding processes, pre-assigned. For example, and These constitute the candidate worker set and machine set for the first task, respectively.

[0130] The process sequence vector represents the execution order of each process in a sequence arranged by its job index. Each number in the vector represents a job number, and its position corresponds to the [number]th [position] of that job. Each process is a task. Similarly, the machine allocation vector and worker allocation vector, based on candidate machine indices and candidate worker indices, represent the resource allocation scheme for each process. For any given process, the information on the designated processing machine and operating worker can be directly obtained by reading the values ​​at the corresponding positions in the MA and WA vectors. The machine and worker information allocated to each process are the index values ​​at the corresponding positions in the corresponding machine allocation vector and worker allocation vector, respectively. The length of each vector is equal to the total number of all unscheduled processes in the system. For example, the first number "3" in the process sequence vector indicates... The first step needs to be done by workers In the machine Execute above.

[0131] Step 2.5.2: Update the harmony memory bank through improvisation.

[0132] After implementing the optimal set initialization strategy, IHSA will initiate the improvisation of new harmonies, including three improvisation methods: harmonic selection, harmonic adjustment, and random generation. The selection of these methods is controlled by a probability mechanism that compares the randomly generated numbers with the probability values ​​of the harmonic memory consideration rate and the harmonic adjustment rate.

[0133] (1) Generate a first random number between [0,1]. If the first random number is less than the harmony memory consideration rate (HMCR), then randomly select a harmony from the initialized harmony memory bank. And select the current optimal harmony from the harmony memory as Otherwise, a new harmony is randomly generated as the final new harmony.

[0134] In IHSA, if the generated random number If the harmonic consideration rate is less than that of the harmonic memory, a harmonic is selected from the initialized harmonic memory. The selected harmonic is represented as... Subsequently, regarding and Perform crossover and mutation operations, where, The goal is to achieve the optimal harmony of the objective function. The objective function is calculated with reference to formula (1), and the Pareto Dominance rule is used to compare the merits of the two objective function values.

[0135] On the other hand, when When the value is greater than or equal to the harmony memory consideration rate, IHSA will no longer select harmonies from the harmony memory bank, but will instead generate new harmonies randomly and update them directly to the harmony sequence, thereby enhancing the algorithm's global exploration capability.

[0136] (2) Using priority process crossover operator to pair harmony and optimal harmony The process sequence vectors are crossed, and the machine allocation vectors in the crossover result are mutated to generate a preliminary new harmony.

[0137] This embodiment uses a priority process crossover operator to reconstruct the process sequence, aiming to enhance the diversity of solutions and generate better scheduling results. Figure 7 The process of priority process overlap is illustrated. Specifically, firstly, a set of jobs is randomly divided into two non-empty complementary subsets. and This is to ensure a balanced representation of the original harmonic memory. and In the process sequence section, first... China belongs to Copy the assignment numbers sequentially into New Harmony, then... China belongs to The job numbers are sequentially copied into the new harmony. The order of the copied job numbers in the new harmony strictly follows their order of appearance in the original job sequence, thus ensuring the feasibility of the solution. To enhance the diversity of solutions, when two operations have the same sequence number, they are randomly arranged as adjacent operations without affecting feasibility. For machine assignment and worker assignment, they are directly copied. and The machine number and worker number corresponding to the selected operation are assigned to the machine and worker assignment sections of the new harmony, thereby maintaining the consistency of resource allocation.

[0138] Taking machine-assigned mutation operations as an example, the execution steps are as follows: First, an operation is randomly selected from the job sequence, and the currently assigned processing machine is determined according to the machine assignment vector corresponding to the operation. Then, a machine is randomly selected from the candidate processing machine set (excluding the current machine) to replace the machine at the corresponding position in the machine assignment.

[0139] (3) Generate a second random number between [0,1]. If the second random number is less than the harmony adjustment rate, perform improvisational fine-tuning on the initial new harmony to obtain the final new harmony; otherwise, do not update the harmony memory and end.

[0140] After completing the crossover and mutation operations, the algorithm will determine whether to perform improvisational fine-tuning on the harmony based on the harmonic adjustment rate. Specifically, when the generated random number is less than the harmonic adjustment rate PAR, the selected pitch will be fine-tuned by adding a small perturbation as shown in formula (25), and the fine-tuning range is calculated based on the bandwidth. If the condition is not met, the algorithm will directly use the harmony in the memory bank without making any changes.

[0141] (25)

[0142] in, By adjusting The value is then applied to generate a new harmonic value using the bandwidth parameter; The second random number; For bandwidth.

[0143] (4) If the objective function value of the final new harmony is better than the worst objective function value in the harmony memory, then replace the harmony corresponding to the worst objective function value in the harmony memory with the final new harmony and end the update; otherwise, abandon the final new harmony, do not update the harmony memory and end the update.

[0144] Step 2.5.3: After updating the harmony memory, the harmony is further optimized using a random perturbation technique based on the t-distribution. This perturbation mechanism, leveraging the probabilistic characteristics of the t-distribution, can effectively help the algorithm escape local optima and improve its global search capability. The probability density function of the t-distribution is detailed in formulas (26) and (27).

[0145] (26)

[0146] (27)

[0147] in, Let be the probability density function of the t-distribution. It is a random variable based on the t-distribution. The degree of freedom is t, and its value determines the shape of the t-distribution. The gamma function is represented by the formula (27) (where, This is the input parameter of the function, usually a positive real number.

[0148] Based on the t-distribution random perturbation technique, global random perturbation (randomly selecting a subset of individuals, such as 10%-30%) and local optimum neighborhood perturbation (targeted perturbation when some individuals are detected to be trapped in local optima) are used to select the harmonics to be perturbed. The new harmonic relationship is generated through t-distribution perturbation, as shown in Equation (28):

[0149] (28)

[0150] in, It is the new harmonic value generated after the disturbance. These are the harmonic values ​​before the perturbation. `iter` indicates the number of iterations. It is a function value of the t-distribution with iter as the degree of freedom. In the early iterations, due to the low degree of freedom, the heavy-tailed effect of the t-distribution will produce large perturbations, thus guiding IHSA to explore farther regions and enhancing the algorithm's global search capability. In the later stages of the algorithm's iterations, as the degree of freedom increases, the t-distribution tends to a normal distribution, the perturbation amplitude decreases, and it is beneficial for the algorithm to refine its exploration around the current solution.

[0151] (6) Determine whether the iteration termination condition is met. If it is met, output the optimal harmony as the optimal solution and end the iteration; otherwise, return to step 2.5.1 to continue the iteration.

[0152] Step 2.6: If rescheduling is triggered, the rescheduling scheme shall be used as the new scheduling scheme; otherwise, the initial scheduling scheme shall be used as the new scheduling scheme, and the new scheduling scheme shall be sent to the physical workshop through the data management module.

[0153] Step 3: The physical workshop executes dynamic scheduling of the flexible workshop according to the new scheduling scheme. In this embodiment, after the new scheduling scheme is generated, it is sent to the physical workshop via the data management module to drive actual production execution.

[0154] This invention verifies the effectiveness of its method through a series of simulation experiments and real-world case analyses. All experiments were conducted on a computer running Windows 11 (64-bit operating system), equipped with an Intel Core processor (2.30 GHz) and 16 GB of memory.

[0155] (1) Experimental design and algorithm performance indicators.

[0156] To verify the effectiveness of IHSA in solving the DFJSSP model, this experiment adopted the following four main steps: (1) using the Taguchi experimental design method to analyze the key parameters of IHSA and determine the optimal values; (2) using the Gurobi solver to verify the accuracy of the solution obtained by IHSA; (3) conducting algorithm comparison experiments and analyzing the statistical significance of the results based on the independent samples t test with 95% confidence; (4) introducing manufacturing engineering examples to verify the practical application performance of the algorithm.

[0157] To comprehensively evaluate the algorithm's performance, this experiment used the Inverted Generational Distance (IGD), Hypervolume (HV), Spacing (SP), and Pareto Front Proximity (P-PF) metrics to quantify the convergence, distribution breadth, uniformity, and overall quality of the solution set. Lower IGD and SP values, and higher HV and P-PF values, indicate superior algorithm performance.

[0158] Table 1 lists the dataset used in the experiment, which includes 18 instances and their size information. The numbers in the instance names correspond to the number of jobs, processes, machines, and workers, respectively. The first five instances (from I_5_10_2_2 to I_7_32_6_4) were constructed according to the instance generation rules proposed in existing literature (Shi, JX, Chen, MZ, Ma, YM, & Qiao, F. (2023). A newboredom-aware dual-resource constrained flexible job shop scheduling problem using a two-stage multi-objective particle swarm optimization algorithm. Information Sciences, 643, 119141.) specifically to verify the accuracy of the solution set. The remaining instances used the original data from this study, with additional randomly generated worker information. To reduce the impact of algorithm randomness, each experiment was repeated 30 times, and the final result was the average of the multiple runs.

[0159] Table 1 Dataset Size

[0160]

[0161] (2) Taguchi parameter analysis.

[0162] This experiment uses the Taguchi experiment to systematically determine the three main parameters of the improved harmony search algorithm: harmony memory size (HMS), harmony memory consideration rate (HMCR), and pitch adjustment rate (PAR). Three discrete levels were defined for each parameter in this experiment, and the corresponding values ​​are listed in Table 2.

[0163] Table 2 Parameter Levels

[0164]

[0165] Based on the number of parameters and the number of levels, an orthogonal array was selected. Taguchi's analysis results are as follows: Figure 8 As shown. The optimal parameter values ​​determined by Taguchi analysis are: HMS=200, HMCR=0.90, PAR=0.30.

[0166] (3) Comparison of algorithm performance with exact solvers.

[0167] This experiment used the Gurobi 10.0.1 solver to verify the effectiveness of the proposed IHSA algorithm. The comparative experiment covered two stages: initial scheduling and dynamic rescheduling, focusing on the objective function value and solution time.

[0168] To ensure the accuracy and reproducibility of the research results, five small instances (from I_5_10_2_2 to I_7_32_6_4) were selected for the experiment, and dynamic events were manually set for each instance. Since DFJSSP is an NP-hard problem, the exact solver struggles to obtain the optimal solution within an acceptable timeframe when the scale is large. For fairness, the maximum runtime of Gurobi was set to be consistent with the average solution time of IHSA, comparing the quality of their feasible solutions. As shown in the comparison data in Table 3, during both initial scheduling and rescheduling, the objective value obtained by IHSA was superior to that of the Gurobi solver, but its running speed was not significantly better. This indicates that, under similar computational efficiency conditions, IHSA has a significant advantage in solution quality. It is noteworthy that even after dynamic events are triggered, the algorithm maintains good performance, demonstrating strong adaptability.

[0169] Table 3 Performance Comparison Results of IHSA and Gurobi

[0170]

[0171] Note: Bold values ​​indicate that IHSA is superior to Gurobi.

[0172] (4) Comparison with baseline algorithms

[0173] To further verify the superiority of IHSA, this experiment selected the following benchmark algorithms for comparative testing: Basic Harmony Search Algorithm, Discrete Particle Swarm Optimization (DPSO), Non-Dominated Sorting Genetic Algorithm (NSGA2), and Improved Gray Wolf Optimization Algorithm (IGWO). All selected algorithms demonstrated excellent performance in solving similar problems, and their typicality and representativeness provide a reliable reference for comparative analysis. The parameters used in the benchmark algorithms are shown in Table 4.

[0174] Table 4. Parameter values ​​used in the benchmark algorithm

[0175]

[0176] Note: n is the population size; It is the inertial constant; and All are acceleration coefficients; These are parameters used in the dynamic operation of the evolutionary population.

[0177] Tables 5 and 6 list the comparison results between IHSA and the benchmark algorithm on four key evaluation metrics. Table 5 shows the performance in the initial scheduling phase, and Table 6 shows the performance in the rescheduling phase. The results show that, in both initial and rescheduling, compared with the benchmark algorithm, IHSA achieves lower IGD values, higher HV values, smaller SP values, and relatively larger P-PF values ​​in most cases. These results confirm that IHSA outperforms the benchmark algorithm in overall performance and maintains robust performance even when dealing with dynamic event disturbances.

[0178] Table 5 Performance of IHSA and benchmark algorithms in initial scheduling

[0179]

[0180] Note: Bold values ​​indicate that IHSA is superior to the benchmark algorithm; ↑ indicates a better (higher) value, and ↓ indicates a better (lower) value.

[0181] Table 6 Performance of IHSA and benchmark algorithms in rescheduling

[0182]

[0183] Note: Bold values ​​indicate that IHSA is better than the benchmark algorithm; ↑ indicates a better (higher) value, while ↓ indicates a better (lower) value.

[0184] To verify the statistical significance of the performance difference between IHSA and the benchmark algorithm, an independent samples t-test was conducted at a 95% confidence level (significance level: 0.05). Table 7 shows the specific test results for Example I_19_91_14_12.

[0185] Table 7 Independent Samples t-Test for Example I_19_91_14_12

[0186]

[0187] As shown in Table 7, the values ​​in the cells are the p-values ​​obtained from the independent samples t-test between IHSA and the benchmark algorithm. The asterisks in the upper right corner (…) The asterisk () indicates that the result is statistically significant at a 95% confidence level, meaning the performance difference between the improved algorithm and the corresponding benchmark algorithm is not due to random factors. Statistically, over 85% of the cell values ​​are marked with an asterisk. This result shows that, across the four evaluation metrics, the performance difference between IHSA and the vast majority of benchmark algorithms is statistically significant. Therefore, the comprehensive advantages exhibited by the improved algorithm are statistically significant, verifying its robustness. The above experimental analysis results collectively demonstrate that IHSA has superior performance in solving the DFJSSP model.

[0188] (5) Engineering examples.

[0189] To further verify the engineering application potential of IHSA, this experiment uses real-world data from a company for case analysis. For its remanufacturing scenario, the production process is modeled as a flexible job shop scheduling problem considering worker absenteeism disturbances. The experiment not only compares the optimization capabilities and convergence characteristics of IHSA and benchmark algorithms, but also focuses on evaluating the impact of rescheduling strategies on the maximum completion time. The data in Table 8 show that IHSA significantly outperforms all benchmark algorithms in solving real-world engineering examples, validating its optimization capabilities and application advantages.

[0190] Table 8 Target values ​​of IHSA and benchmark algorithms

[0191]

[0192] Note: Bold values ​​indicate that IHSA outperforms the benchmark algorithm.

[0193] like Figure 9 As shown, by comparing the convergence curves of each algorithm, it can be observed that IHSA has a faster convergence rate and can obtain higher quality solutions in all three objective dimensions. Figure 10 The diagram visually illustrates the initial scheduling and IHSA rescheduling scheme under worker absence disturbances. The upper and lower parts of the diagram represent the initial state and the rescheduling result after the absence of the third worker (shown by dashed boxes), respectively. The comparison shows that the maximum completion time after rescheduling (marked by the vertical dashed line) only extends slightly, with negligible deviation. This characteristic of maintaining relatively stable performance indicators under strong disturbances reflects the robustness of the scheduling strategy to a certain extent.

[0194] In summary, IHSA significantly outperforms benchmark algorithms in both optimization accuracy and convergence rate. Particularly in scenarios involving worker absenteeism, the rescheduling strategy can quickly generate high-quality adjustment plans, ensuring the stable operation of the production system. These results fully validate the effectiveness and robustness of the proposed method in addressing the DFJSSP problem, demonstrating its potential for engineering applications in dynamic and uncertain environments.

[0195] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0196] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A digital twin driven flexible job shop dynamic scheduling method, characterized in that, The method comprises the following steps: Collecting real-time state data of the physical workshop and capturing dynamic disturbance events, transmitting the real-time state data and dynamic disturbance events to the virtual workshop, wherein the real-time state data includes real-time states of machines, jobs and workers, and the dynamic disturbance events include worker absenteeism and machine failure; The virtual workshop calls a dynamic scheduling strategy to generate a new scheduling scheme and transmits it to the physical workshop, wherein the dynamic scheduling strategy generates an initial scheduling scheme according to static workshop parameters, judges whether to trigger a rescheduling, if the rescheduling is triggered, updates the static workshop parameters to obtain real-time workshop parameters based on the real-time state data and the dynamic disturbance events, and generates a rescheduling scheme by using an improved harmony search algorithm, and takes the rescheduling scheme as the new scheduling scheme; otherwise, takes the initial scheduling scheme as the new scheduling scheme; The physical workshop executes the flexible job shop dynamic scheduling according to the new scheduling scheme.

2. The digital twin driven dynamic job shop scheduling method of claim 1, wherein, The judgment whether to trigger the rescheduling comprises: Detecting whether there is worker absenteeism according to the real-time state data and the dynamic disturbance events, if there is, triggering the rescheduling and taking the day as the starting point of the next periodic rescheduling interval; otherwise, continuing to judge whether the periodic rescheduling interval is reached; If the periodic rescheduling interval is not reached, the rescheduling is not triggered; otherwise, it is judged whether worker absenteeism is detected in the last period; If worker absenteeism is not detected in the last period, the rescheduling is triggered; otherwise, the periodic rescheduling interval is updated and the rescheduling is triggered.

3. The digital twin driven dynamic job shop scheduling method of claim 2, wherein, The detection whether there is worker absenteeism according to the real-time state data and the dynamic disturbance events comprises: If the dynamic disturbance events contain worker absenteeism events, it means that worker absenteeism is detected; otherwise, the fatigue levels of workers who have not occurred worker absenteeism events are calculated according to the real-time state data, the fatigue recovery time of the workers is obtained according to the fatigue state, if the fatigue recovery time is greater than the rest threshold, it is judged that there is worker absenteeism; otherwise, it is not detected that there is worker absenteeism.

4. The digital twin driven dynamic job shop scheduling method of claim 1, wherein, The improved harmony search algorithm uses a three-layer representation scheme, including a process sequence layer, a machine allocation layer and a worker allocation layer; The process sequence layer records a process sequence vector, the process sequence vector represents the process execution sequence of each job in the arrangement of job indexes, each number in the process sequence vector represents a job number, and the same job number appears the first path process; The machine allocation layer records a machine allocation vector, and each number in the machine allocation vector represents the machine number allocated to the process in the corresponding position of the process sequence vector; The worker allocation layer records a worker allocation vector, and each number in the worker allocation vector represents the worker number allocated to the process in the corresponding position of the process sequence vector.

5. The digital twin driven dynamic job shop scheduling method of claim 1, wherein, The improved harmony search algorithm uses a good point set initialization strategy to initialize the harmony memory library, and the harmony memory library is as follows: In the formula, Represents the harmonic memory bank. Indicates the size of the harmonic memory bank. The dimensions representing the harmonic memory bank, For the first in the harmony memory bank The first harmony Components of each dimension , , Indicates the first The upper bound of each dimension, Indicates the first The lower bound of each dimension, express The set of best points in the unit cube of dimensional Euclidean space is formed by... indivual A point set composed of dimensional points.

6. The digital twin driven dynamic job shop scheduling method of claim 1, wherein, The improved harmony search algorithm updates the harmony memory library through improvisation, specifically comprising: generating a first random number between [0, 1], if the first random number is less than the harmony memory consideration rate, randomly selecting a harmony memory from the initialized harmony memory library as the harmony memory , and taking the current optimal harmony memory selected from the harmony memory library as the harmony memory ; otherwise, generating a new harmony randomly as the final new harmony; A priority process crossover operator is applied to the sequence vectors of the harmony and the optimal harmony and the sequence vectors of the harmony and the optimal harmony is crossed, and a mutation operation is performed on the machine assignment vectors in the crossing result to generate a preliminary new harmony Generating a second random number between 0 and 1, if the second random number is less than the harmony adjustment rate, performing improvisation fine-tuning on the preliminary new harmony to obtain the final new harmony; otherwise, the harmony memory library is not updated and the process is ended. If the objective function value of the final new harmony is better than the worst objective function value in the harmony memory, the final new harmony replaces the harmony corresponding to the worst objective function value in the harmony memory, and the updating ends; otherwise, the final new harmony is abandoned, the harmony memory is not updated, and the updating ends.

7. The digital twin driven dynamic job shop scheduling method of claim 1, wherein, The improved harmony search algorithm uses a random disturbance technique based on t-distribution to optimize the harmony after updating the harmony memory, comprising: selecting part of the harmonies in the harmony memory as to-be-disturbed harmonies; calculating the function value of the t-distribution with the number of iterations as the degree of freedom; weighting each to-be-disturbed harmony by the function value of the t-distribution, and superimposing the weighted value with the corresponding to-be-disturbed harmony to obtain a disturbed harmony.

8. The digital twin driven dynamic job shop scheduling method of claim 1, wherein, The objective function of the harmony in the improved harmony search algorithm is to minimize the maximum completion time, total carbon emissions and maximum cumulative worker fatigue.

Images