Industrial production-oriented digital twin driven multi-agent scheduling real-time reconstruction method

Abstract

The application discloses an industrial production-oriented digital twin driven multi-agent scheduling real-time reconstruction method, relates to the technical field of digital twin, and comprises the following steps: constructing a digital twin body by collecting order and workshop data, driving scheduling agents to make autonomous decisions, and distributing orders by using a target dynamic exploration algorithm, and the scheduling is optimized in real time according to data changes, the system synchronously executes the distribution strategy, and the algorithm parameters are dynamically adjusted during the production peak period, so that the adaptive synchronization of the scheduling plan and the actual production is realized. The application is driven in real time by the digital twin body and the scheduling agent, adopts a target dynamic exploration algorithm that combines exploration, yield and waiting coefficients, realizes dynamic order distribution and load balancing, reconstructs the scheduling in real time according to data changes, adjusts the parameters during the peak period, suppresses inefficient workshop distribution, and gives priority to efficient production capacity, so that the scheduling agility, anti-interference ability and order delivery reliability are significantly improved, and the resource fairness and overall production capacity optimization are considered.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to a digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production. Background Technology

[0002] By creating a real-time synchronized digital twin virtual model of the physical workshop in the digital space, and mapping all production resources and orders to intelligent agents capable of autonomous decision-making, these agents dynamically calculate the optimal scheduling plan through communication, competition, and cooperation in the virtual environment. The plan is then verified through simulation before being sent to the physical workshop for execution. If real production is disrupted by unexpected events such as equipment failure or order insertion, the intelligent agents in the digital twin will reschedule and quickly generate new solutions to cope with the changes, thereby greatly improving the agility and anti-interference capability of production scheduling.

[0003] For example, the intelligent scheduling and production planning system and method based on digital twins provided in Chinese Patent Publication No. CN119539413B arranges the set of orders to be scheduled into two sequences by using different sorting strategies. Then, it uses a resource-time matrix to simulate the production process of these two sequences separately. By simulating the production process under different scheduling methods, it generates corresponding simulation logs. Finally, based on the comparison between the simulation logs, it determines the optimal scheduling method for corresponding scheduling and management. In this way, potential problems can be identified in advance through virtual simulation in digital twin technology, allowing for timely optimization of the scheduling method and adjustment of the production plan, thus avoiding losses in actual production.

[0004] For example, Chinese Patent Publication No. CN118295332A provides a multi-objective production line scheduling method, module, and production line digital twin system, which includes the steps of constructing an objective function; minimizing the objective function by executing a simulated annealing algorithm based on multi-objective optimization; and further minimizing the objective function by executing a population optimization algorithm to obtain the optimal production plan data. This invention establishes an objective function related to the production line's cycle time and balance rate, introduces weighting coefficients, and uses a suitable optimization algorithm to simultaneously optimize the production line's cycle time and balance rate, thereby ensuring high production efficiency and load balancing across all processes on the production line.

[0005] However, existing scheduling agents are unable to respond in real time to emergencies such as equipment failures and order insertions, resulting in a disconnect between scheduling plans and actual production. This not only leads to low production efficiency but also increases the risk of untimely delivery during peak production periods. Summary of the Invention

[0006] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production. This method solves the problem that existing scheduling systems cannot respond to emergencies in real time, leading to a disconnect between scheduling plans and actual production, resulting in low efficiency and delivery risks during peak production periods.

[0007] To achieve the above objectives, this invention provides the following technical solution: a digital twin-driven multi-agent scheduling real-time reconstruction method for industrial production, comprising the following specific steps: Step 1: Collect order data and workshop equipment data from different workshops. The workshop equipment data includes the workshop's production efficiency and production stability. After preprocessing, construct a digital twin and a scheduling agent. The digital twin drives the scheduling agent to make autonomous decisions. Step 2: The scheduling agent allocates orders using a target dynamic exploration algorithm. Step 3: Changes in order data and workshop equipment data trigger the scheduling agent to update and allocate orders, and optimize the order allocation. Step 4: Based on the scheduling agent's allocation strategy, synchronously allocate orders to each workshop in reality and analyze whether it is a peak production period. If it is a peak production period, adjust the target dynamic exploration algorithm; if it is a stable production period, return to Step 1 or terminate directly.

[0008] Furthermore, the specific method by which the scheduling agent allocates orders through the target dynamic exploration algorithm is as follows: the order volume of equipment data is counted to obtain the total number of orders. The scheduling agent first randomly allocates the total number of orders to the equipment in each workshop to obtain the number of allocated orders for each workshop. A scaling factor is set, and comprehensive calculations are performed based on the scaling factor, the total number of orders, the number of allocated orders for each workshop, and the workshop equipment data. After standardization processing, the target dynamic exploration formula is obtained, and the orders are preferentially allocated to the workshop with the largest output value of the target dynamic exploration formula.

[0009] Furthermore, the specific method for obtaining the target dynamic exploration formula is as follows: the exploration coefficient is obtained by comprehensively calculating the total number of orders, the number of orders allocated to each workshop, and the scaling factor; the revenue coefficient is obtained by comprehensively calculating the production efficiency and production stability of the workshop; the waiting coefficient is obtained by comprehensively calculating the number of orders allocated to each workshop and the production efficiency of the workshop; and the target dynamic exploration formula is obtained by comprehensively calculating the exploration coefficient, the revenue coefficient, and the waiting coefficient.

[0010] Furthermore, the specific method for obtaining the target dynamic exploration formula is as follows: ;in, This represents the formula for dynamic exploration of the target. Indicates the exploration coefficient. Represents the profit coefficient. This represents the waiting coefficient.

[0011] Furthermore, the specific method for obtaining the exploration coefficient is as follows: convert the total number of orders into the form of a logarithmic function, then calculate the ratio of the total number of orders to the number of orders allocated to each workshop to obtain the value to be explored, convert the value to be explored into the form of a square root, and multiply the value to be explored with the scaling factor to obtain the exploration coefficient.

[0012] Furthermore, the specific method for obtaining the profit coefficient is as follows: the production efficiency of the workshop is multiplied by the production stability of the workshop to obtain the profit coefficient.

[0013] Furthermore, the specific method for obtaining the waiting coefficient is as follows: Set the waiting coefficient, initially assigning it a value of zero. Calculate the production time spent by any workshop by multiplying the number of allocated orders for each workshop by the reciprocal of the workshop's production efficiency. Preset a queuing production time threshold. Compare the production time spent by any workshop with the queuing production time threshold. If the production time spent by any workshop is greater than the queuing production time threshold, calculate the difference between the production time spent by any workshop and the queuing production time threshold, and assign the result to the waiting coefficient. If the production time spent by any workshop is less than or equal to the queuing production time threshold, no further assignment is made.

[0014] Furthermore, the specific method for analyzing whether it is a peak production period is as follows: count the number of workshops to obtain the total number of workshops, sum the production time spent in any workshop according to the number of workshops to obtain the total production time, set a production time threshold, and compare the total production time with the production time threshold. If the total production time is greater than the production time threshold, it indicates a peak production period. If the total production time is less than or equal to the production time threshold, it indicates a stable production period.

[0015] Furthermore, the specific method for adjusting the target dynamic exploration algorithm is as follows: by reducing the scaling factor, the exploration coefficient is reduced, thereby reducing the probability that the scheduling agent will prioritize the allocation of a workshop with a target dynamic exploration formula that has a small benefit coefficient but a large exploration coefficient.

[0016] Beneficial effects Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: 1. By constructing a digital twin and a scheduling intelligence agent, order data and workshop equipment data are collected in real time. When order volume fluctuates or equipment status changes, the system can immediately trigger scheduling updates. Combined with a dynamic decision-making mechanism of exploration coefficient, revenue coefficient, and waiting coefficient, the scheduling intelligence agent can quickly adjust order allocation strategies, effectively cope with emergencies such as equipment failure and order insertion, and avoid the problem of delayed response in traditional scheduling systems. This significantly improves the agility and anti-interference ability of production scheduling, ensuring that the production plan and actual production are always synchronized.

[0017] 2. Through the target dynamic exploration algorithm, the scheduling agent comprehensively considers the exploration coefficient, benefit coefficient, and waiting coefficient during the order allocation process, achieving a dynamic balance between exploration and utilization. In the initial stage, the exploration coefficient dominates to learn the characteristics of the workshop; in the stable stage, the benefit coefficient takes priority to maximize efficiency; when data changes drastically, the coefficient linkage triggers strategy reconstruction. This mechanism not only improves the intelligence level of scheduling decisions, but also avoids unreasonable resource allocation caused by over-reliance on a single factor, thereby enhancing the flexibility and robustness of the system while ensuring production efficiency.

[0018] 3. By introducing a load balancing constraint mechanism with a waiting coefficient as its core, and by monitoring the waiting time of the task queues in each workshop, overloaded workshops are dynamically de-weighted. When a workshop is assigned too many orders, causing the production time to exceed the threshold, the waiting coefficient will automatically increase, reducing the probability of priority allocation of orders to that workshop and preventing local congestion from affecting the overall system efficiency. At the same time, during peak production periods, the scaling factor is adjusted to suppress excessive exploration of inefficient workshops and prioritize support for highly stable and efficient workshops, thereby balancing fairness and maximizing overall capacity in resource allocation.

[0019] 4. By automatically analyzing peak and stable production periods and dynamically adjusting the parameters of the target dynamic exploration algorithm during peak periods, the exploration coefficient is reduced, thereby lowering the allocation priority of inefficient workshops. This adjustment ensures that orders are preferentially allocated to workshops with high production efficiency and stability, effectively reducing the risk of delivery delays caused by uneven resource allocation during peak production periods, improving the reliability of timely order delivery, and optimizing the overall system capacity.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] Figure 1 This invention presents a flowchart of a digital twin-driven real-time reconfiguration method for multi-agent scheduling in industrial production.

[0022] Figure 2 This invention is a structural diagram of a real-time simulation and decision optimization system for a digital twin-driven embodied intelligent agent for industrial production. Detailed Implementation

[0023] 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.

[0024] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0025] Example 1: like Figures 1-2 As shown, this embodiment of the invention provides a digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production, including the following specific steps: Step 1: By deploying IoT sensors, equipment control systems, and manufacturing execution systems in different workshops, IoT sensors, such as current sensors, vibration sensors, and temperature sensors, are used to sense the status of equipment and the trajectory of work-in-process in real time. Equipment control systems, such as programmable logic controllers, are used to obtain accurate equipment instructions, operating status, and alarm information. Manufacturing execution systems are responsible for managing, monitoring, and controlling the production execution process in the factory workshop. Real-time collection of order data and workshop equipment data; data cleaning to remove redundant values; workshop equipment data includes workshop production efficiency and workshop production stability; workshop production efficiency is the reciprocal of the average time required for each piece of equipment in the workshop to produce a single product; workshop production stability is the reciprocal of the average number of failures of each piece of equipment in the workshop; and then a digital virtual workshop twin is generated. The digital virtual workshop twin achieves high-fidelity modeling through deep integration of three levels: geometry, physics, and behavior. The geometric model layer accurately depicts the three-dimensional appearance, kinematic relationships, and workspace of the equipment, providing a spatial verification basis for virtual simulation. The physical model layer embeds physical rules such as processing accuracy, energy consumption, and thermal deformation, enabling the digital virtual workshop twin to predict the performance status and trends of the equipment under different operating conditions. The behavioral model layer defines the interaction logic and response rules between entities such as equipment and materials in the production process, enabling the digital virtual workshop twin to autonomously simulate the real production process. These three layers together constitute a virtual entity that can accurately map, deeply analyze, and proactively predict the dynamics of the physical workshop, providing a core support environment for the scheduling agent that combines state perception, strategy verification, and decision optimization capabilities. A scheduling agent is built, and the digital virtual workshop twin provides the scheduling agent with a real-time status perception and strategy verification environment, thereby driving it to make autonomous decisions and dynamically optimize. The scheduling agent adopts a reinforcement learning architecture based on the actor-critic. For example, when the digital virtual workshop twin perceives changes in key events such as equipment status or order data, it immediately triggers the scheduling agent to perform real-time reconstruction calculations, thereby achieving rapid autonomous response and dynamic optimization to production disturbances.

[0026] Step Two: Analyze the order volume of equipment data to obtain the total number of orders. The scheduling agent first randomly allocates the total number of orders to the equipment in each workshop, obtaining the number of orders allocated to each workshop. Differences in equipment operating years and production performance across workshops can lead to variations in the number of orders allocated later. Based on historical experience, a scaling factor is set offline and statically to control the relative fairness of the scheduling agent's scheduling. A comprehensive calculation is performed based on the total number of orders, the number of orders allocated to each workshop, and the scaling factor. Standardization is then applied to eliminate dimensional differences and transform values ​​of different orders of magnitude into a unified numerical range, yielding the exploration coefficient. A comprehensive calculation is performed based on the workshop's production efficiency and stability, and standardization is applied to obtain the revenue coefficient. A comprehensive calculation is performed based on the number of orders allocated to each workshop and the workshop's production efficiency, and standardization is applied to obtain the waiting coefficient. Finally, the exploration coefficient, revenue coefficient, and waiting coefficient are comprehensively calculated and standardized to obtain the target dynamic exploration formula.

[0027] Step 3: By changing order data and workshop equipment data, such as a sudden increase or decrease in order volume, an increase or decrease in workshop equipment production efficiency, or a malfunction or increased stability of workshop equipment, the scheduling agent is triggered to update and allocate orders. At the same time, the target dynamic exploration formula is optimized so that orders are preferentially allocated to the workshop with the largest output value of the target dynamic exploration formula.

[0028] Step 4: Based on the allocation strategy of the scheduling agent, synchronously allocate orders to each workshop in reality and analyze whether it is a peak production period. If it is a peak production period, adjust the exploration coefficient. If it is a stable production period, return to Step 1 or end directly.

[0029] Example 2 differs from Example 1 in that: The specific method for obtaining the target dynamic exploration formula is as follows: ; in, This represents the formula for dynamic exploration of the target. Indicates the exploration coefficient. Represents the profit coefficient. The waiting coefficient represents the decision-making process of the scheduling agent, which begins in the initial stage. Since the production efficiency and stability of each workshop are not yet clear, i.e., the profit coefficient is uncertain, the scheduling agent first performs random allocation to start learning. Subsequently, the exploration coefficient becomes dominant. The larger the value, the larger the output value of the target dynamic exploration formula, which in turn enables the workshop to obtain more order allocations. This is done to actively collect data. In the later stage, when the efficiency and stability of each workshop tend to stabilize, the profit coefficient is determined. At this time, the decision basis becomes the sum of the profit coefficient and the exploration coefficient. The larger this sum is, the larger the output value of the target dynamic exploration formula, and the more orders are allocated to the corresponding workshop. At this stage, the influence of the exploration coefficient on allocation is weaker than that of the profit coefficient. However, if there is a sudden increase or decrease in order volume, a sudden change in equipment efficiency, or a malfunction, it will trigger a dynamic adjustment of the exploration coefficient or the revenue coefficient, thereby changing the formula output value and triggering the scheduling agent to re-explore and recalibrate. In addition, the target dynamic exploration formula introduces a key constraint: even if the sum of the revenue coefficient and the exploration coefficient of a certain workshop is the largest, if too many orders have been arranged for it, resulting in a large number of orders being passively waiting, the scheduling agent will reduce the output value of its target dynamic exploration formula through the waiting coefficient, thereby constraining the previous sum value and ensuring that this workshop is no longer continuously over-allocated, so as to achieve overall load balancing and optimal efficiency.

[0030] The specific methods for obtaining the exploration coefficient are as follows: Converting the total number of orders into a logarithmic function not only avoids excessively large exploration coefficients that would increase computer computational overhead, but also ensures that the increase or decrease of the exploration coefficient is less than the increase or decrease of the revenue coefficient. Then, the ratio of the total number of orders to the number of orders allocated to each workshop is calculated to obtain the value to be explored. Converting the value to be explored into the form of a square root further avoids excessively large exploration coefficients that would increase computer computational overhead, and further ensures that the increase or decrease of the exploration coefficient is less than the increase or decrease of the revenue coefficient. Finally, the value to be explored is multiplied by the scaling factor to obtain the exploration coefficient. ; in, Indicates the exploration coefficient. Indicates the total number of orders. Indicates the first The number of orders allocated to each workshop. This represents the scaling factor, used not only to zoom in or out on the total number of orders and the number of orders. The ratio of the number of orders allocated to each workshop ensures that even if the profit coefficient in the target dynamic exploration formula of a certain workshop is relatively small, the exploration coefficient is relatively large, and the final target dynamic exploration formula also becomes larger. This allows for priority allocation, improving the allocation fairness of weaker workshops. Moreover, it also ensures that the increase or decrease of the exploration coefficient is less than the increase or decrease of the profit coefficient. In addition, the scaling factor in the target dynamic exploration formula of each workshop is equal, meaning that if the scaling factor is adjusted, the scaling factor in the target dynamic exploration formula of each workshop will be adjusted to be consistent.

[0031] The specific method for obtaining the profit coefficient is as follows: The revenue coefficient is calculated by multiplying the workshop's production efficiency and production stability. Production efficiency reflects the workshop's output capacity per unit time, while production stability reflects the workshop's reliability in maintaining high efficiency and low failure rate during long-term operation. The revenue coefficient obtained by multiplying the two can comprehensively evaluate the workshop's real-time production capacity and long-term operational reliability, avoiding focusing only on short-term efficiency while ignoring equipment failure risks, thus providing more comprehensive guidance for order allocation decisions.

[0032] The specific method for obtaining the waiting coefficient is as follows: A waiting coefficient is set, initially set to zero. The production time for any workshop is calculated by multiplying the number of orders allocated to each workshop by the reciprocal of the workshop's production efficiency. The reciprocal of the workshop's production efficiency represents the average time required for any equipment in that workshop to produce one product. A pre-set queuing production time threshold is set. The production time for any workshop is compared with the queuing production time threshold. If the production time for any workshop is greater than the queuing production time threshold, the difference between the production time for any workshop and the queuing production time threshold is calculated, and the result is assigned to the waiting coefficient. This result has been standardized and compressed to a uniform value range with the profit coefficient and the exploration coefficient. If the production time for any workshop is less than or equal to the queuing production time threshold, no further assignment is made, meaning the waiting coefficient remains zero.

[0033] The specific method for analyzing whether it is a peak production period is as follows: The number of workshops is counted to obtain the total production time. The production time spent in any workshop is summed to obtain the total production time. A production time threshold is set based on historical experience. This threshold is not a fixed empirical value, but a dynamic parameter that can be periodically calibrated based on historical production data, so that it can better adapt to the long-term changes in the workshop production rhythm. The total production time is compared with the production time threshold. If the total production time is greater than the production time threshold, it indicates a peak production period. If the total production time is less than or equal to the production time threshold, it indicates a stable production period.

[0034] The specific method for adjusting the exploration coefficient is as follows: During peak production periods, to ensure timely product delivery, scaling factors can no longer guarantee fairness in workshop production. Even if a workshop has low production efficiency and stability, a large exploration coefficient can still maximize the output value of its target dynamic exploration formula, leading to priority allocation. This is because scaling factors amplify the ratio of the total number of orders to the number of orders allocated to a particular workshop, resulting in a large exploration coefficient. Therefore, reducing the scaling factor decreases the exploration coefficient, thereby reducing the probability that the scheduling agent will prioritize workshops with low return coefficients but high exploration coefficients in their target dynamic exploration formulas. This ensures that orders are prioritized for workshops with high production efficiency and stability, ensuring timely product delivery.

[0035] Application simulation: In practical application, a factory has four production workshops, all of which deploy the digital twin-driven multi-agent scheduling real-time reconfiguration method of this application. Three of the workshops have old equipment, which are designated as workshop A, workshop B, and workshop C, respectively. The remaining workshop has new equipment, which is designated as workshop D. Experiments show that workshop D has higher production efficiency and production stability than the other three workshops. Initially, orders are randomly assigned, and the workshop with the higher exploration coefficient receives more orders. However, as time progresses, workshop D has a higher benefit coefficient, meaning its production efficiency and stability are higher than the other three workshops. Therefore, workshop D receives the most orders. But the increased number of orders leads to a higher waiting coefficient, which in turn reduces the output value of the target dynamic exploration formula in workshop D. Consequently, workshops A, B, and C have the highest target dynamic exploration formula output value, resulting in the highest order allocation.

[0036] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production, characterized by: The specific steps include the following: Step 1: Collect order data and workshop equipment data from different workshops. The workshop equipment data includes the workshop's production efficiency and production stability. After preprocessing, construct a digital twin and a scheduling intelligence agent. The digital twin drives the scheduling intelligence agent to make autonomous decisions. Step 2: The scheduling agent allocates orders through a target dynamic exploration algorithm. The target dynamic exploration algorithm calculates a target dynamic exploration formula for each workshop by coordinating and optimizing three decision objectives: capacity exploration, efficiency utilization, and load constraints during the decision-making process. Step 3: By changing order data and workshop equipment data, the scheduling agent is triggered to update and allocate orders, and the workshop order allocation is changed according to the changes in the output value of the target dynamic exploration formula. Step 4: Based on the allocation strategy of the scheduling agent, synchronously allocate orders to each workshop in reality, and analyze whether it is a peak production period. If it is a peak production period, adjust the target dynamic exploration algorithm to suppress the exploration of inefficient capacity and prioritize the guarantee of efficient capacity. If a stable production period is found, return to Step 1 or end directly.

2. The digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production according to claim 1, characterized in that: The specific method by which the scheduling agent allocates orders using the target dynamic exploration algorithm is as follows: The system calculates the total number of orders by analyzing the order volume of equipment data. The scheduling agent first randomly allocates the total number of orders to the equipment in each workshop to obtain the number of orders allocated to each workshop. It then sets a scaling factor and performs comprehensive calculations based on the scaling factor, the total number of orders, the number of orders allocated to each workshop, and the workshop equipment data. After standardization, it obtains the target dynamic exploration formula and prioritizes allocating orders to the workshop with the largest output value of the target dynamic exploration formula.

3. The digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production according to claim 2, characterized in that: The specific method for obtaining the target dynamic exploration formula is as follows: The exploration coefficient is obtained by comprehensively calculating the total number of orders, the number of orders allocated to each workshop, and the scaling factor. The revenue coefficient is obtained by comprehensively calculating the production efficiency and production stability of each workshop. The waiting coefficient is obtained by comprehensively calculating the number of orders allocated to each workshop and the production efficiency of each workshop. The exploration coefficient, revenue coefficient, and waiting coefficient are combined to obtain the target dynamic exploration formula.

4. The digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production according to claim 3, characterized in that: The specific method for obtaining the target dynamic exploration formula is as follows: ； in, This represents the formula for dynamic exploration of the target. Indicates the exploration coefficient. Represents the profit coefficient. This represents the waiting coefficient.

5. The digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production according to claim 4, characterized in that: The specific method for obtaining the exploration coefficient is as follows: The total number of orders is converted into a logarithmic function. The ratio of the total number of orders to the number of orders allocated to each workshop is then calculated to obtain the value to be explored. The value to be explored is converted into a square root form. The value to be explored is then multiplied by the scaling factor to obtain the exploration coefficient.

6. The digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production according to claim 4, characterized in that: The specific method for obtaining the profit coefficient is as follows: The profitability coefficient is obtained by multiplying the workshop's production efficiency by its production stability.

7. The digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production according to claim 4, characterized in that: The specific method for obtaining the waiting coefficient is as follows: Set a waiting coefficient, initially set to zero. Calculate the production time for any workshop by multiplying the number of allocated orders for each workshop by the reciprocal of the workshop's production efficiency. Set a preset queuing production time threshold. Compare the production time of any workshop with the queuing production time threshold. If the production time of any workshop is greater than the queuing production time threshold, calculate the difference between the two and assign the result to the waiting coefficient. If the production time of any workshop is less than or equal to the queuing production time threshold, no further assignment is made.

8. The digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production according to claim 7, characterized in that: The specific method for analyzing whether the production period is peak in step four is as follows: The number of workshops is counted. The production time spent in any workshop is summed to obtain the total production time. A production time threshold is set, and the total production time is compared with the production time threshold. If the total production time is greater than the production time threshold, it indicates a peak production period. If the total production time is less than or equal to the production time threshold, it indicates a stable production period.

9. The digital twin-driven multi-agent scheduling real-time reconfiguration method for industrial production according to claim 8, characterized in that: The specific method of adjusting the target dynamic exploration algorithm in step four is as follows: By reducing the scaling factor, the exploration coefficient is reduced, thereby decreasing the probability that the scheduling agent will prioritize the allocation of a workshop with a small return coefficient but a large exploration coefficient according to the dynamic exploration formula.

Images