Production-logistics resource real-time allocation integrated method based on dt and information entropy

Abstract

The application belongs to the technical field of intelligent manufacturing and digital twin workshop, and discloses a production-logistics resource real-time deployment integrated method based on DT and information entropy, which comprises the following steps: a digital twin workshop is built by fusing production-logistics resources, process flow, production line layout and operation data, and an iterative state model is constructed; a pre-scheduling result is generated based on the state model when a task is triggered, is mapped to the digital twin workshop for deduction, and is processed by a gate control cycle unit to obtain a predicted completion time; average urgency and normalized system information entropy are calculated based on the predicted completion time, and a criterion index is constructed to trigger resource allocation; the optimal resource combination is selected by taking the individual information entropy of the task as the weight; and the state model and the digital twin workshop are iteratively updated by issuing execution and collecting feedback data. The application realizes the integration of resource allocation and real-time scheduling, and improves the operation stability of the workshop in a dynamic disturbance environment.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and digital twin workshop technology, specifically involving a method for real-time allocation and integration of production and logistics resources based on DT and information entropy. Background Technology

[0002] With the advancement of Industry 5.0, discrete manufacturing workshops are gradually transforming into a production model characterized by multiple varieties, small batches, and high flexibility. Disturbances such as dynamic order arrivals, fluctuating processing times, and difficulties in production-logistics coordination are prevalent, leading to significant time-varying and uncertainties in workshop system load and resource demands. Digital twin (DT) technology, with its high-fidelity virtual mapping and multi-source disturbance simulation capabilities, provides a new technical path for production-logistics collaborative scheduling; however, existing scheduling algorithms still struggle to adapt to the highly dynamic and highly disturbed workshop operation scenarios.

[0003] In existing research on real-time allocation of production and logistics resources, although various algorithms have been proposed to solve complex workshop scheduling problems, existing technologies often struggle to balance multiple indicators such as scheduling efficiency, energy consumption cost, and task on-time rate due to limitations in computational complexity, real-time response, and multi-objective coordination capabilities. For example, heuristic rule-based methods (such as FIFO+SPT) can optimize task execution order locally, but they struggle to balance global energy consumption and completion efficiency in high-disturbance environments; mathematical programming-based methods (such as FIFO+LINGO) possess global optimization capabilities, but their computational complexity is too high to meet the real-time response requirements of actual engineering projects; and scheduling schemes based on mobile edge computing (MEC) can assist decision-making through edge nodes, but in strongly coupled scenarios of deep integration of production and logistics, they still generally suffer from high overall costs and large-scale task delays. At their root, despite their different technical approaches, these methods are essentially limited by the following three common mechanism defects: first, the separation of production and logistics. Most existing methods treat production scheduling and logistics scheduling separately, failing to consider the coupling relationship between machine tool processing, AGV transportation, and other links. This makes scheduling results susceptible to global deviations due to fluctuations in local links. Secondly, there is a lack of uncertainty quantification. There is a lack of effective means to quantify the uncertainty of system operation. When faced with multi-source disturbances such as equipment failure, AGV obstacle avoidance, and communication delays, it is difficult to dynamically assess the robustness of scheduling schemes, leading to resource allocation delays or over-allocation. Thirdly, there is a lack of closed-loop integration mechanisms. Existing methods mostly rely on static parameters and preset plans, making it difficult to achieve a dynamic balance between indicators such as scheduling efficiency, energy consumption cost, and task on-time rate. Furthermore, they do not fully utilize the high-fidelity simulation capabilities of digital twin technology to pre-verify scheduling schemes. This results in static separation modes such as "configure first, then schedule" or "schedule only, no configuration" being unable to cope with the cumulative effects of continuous disturbances.

[0004] The scheduling requirements of discrete manufacturing workshops have shifted from single-indicator optimization to multi-dimensional collaboration. The aforementioned limitations severely restrict the operational efficiency and resource utilization of smart manufacturing workshops. Therefore, there is an urgent need for a real-time production-logistics resource allocation method that can integrate the high-fidelity simulation capabilities of digital twins, combine information entropy to quantify system uncertainties, and achieve a closed-loop integrated resource allocation and real-time scheduling mechanism. This method would break down the scheduling barriers between production and logistics, achieve dynamic balance across multiple objectives, and adapt to highly dynamic and highly disturbed workshop operation scenarios. Summary of the Invention

[0005] This invention provides a real-time production-logistics resource allocation and integration method based on DT and information entropy, aiming to solve the technical problems mentioned in the background art.

[0006] In a first aspect, the present invention provides a method for real-time allocation and integration of production and logistics resources based on DT and information entropy, including: Step S1: Integrate workshop production-logistics resources, process flow, production line layout and real-time workshop operation data to build a digital twin (DT) workshop and construct an iterative production-logistics resource status model; Step S2: When a task triggers scheduling, the current workshop resource configuration is determined based on the state model. A pre-scheduling result is generated by combining the workshop resource configuration and task requirements. The pre-scheduling result is mapped to the high-disturbance environment of the digital twin (DT) workshop for simulation to obtain the simulation completion time. The simulated completion time is processed using the gated loop unit method to obtain the predicted completion time. The predicted completion time of the task in the set of tasks to be scheduled is updated. Step S3: Calculate the average urgency and normalized system information entropy of the set of tasks to be scheduled based on the predicted completion time, and construct resource allocation criteria. When the criteria reach a preset threshold, trigger production-logistics resource allocation, add the resources to be activated to the available resource pool, and form a resource allocation scheme. Step S4: Based on the predicted completion time, determine the individual information entropy of the task. Using the individual information entropy of the task as the weight, calculate the service duration and energy consumption of each production-logistics resource combination in the available resource pool under the resource allocation scheme to obtain the comprehensive evaluation value of each production-logistics resource combination. Select the resource combination with the smallest comprehensive evaluation value as the optimal real-time scheduling result corresponding to the current task. Step S5: Send the resource configuration scheme and the optimal real-time scheduling result to the physical workshop for execution, collect real-time status data of production-logistics resources and task execution feedback data of the physical workshop, update the status model and the digital twin (DT) workshop, and wait for the next task to trigger scheduling.

[0007] Furthermore, the digital twin (DT) workshop is built on Unity3D software and achieves bidirectional communication with the physical workshop through local area network interconnection and TCP / IP protocol; Furthermore, the state model defines static attributes, real-time state attributes, and configurable attributes for each resource; the static attributes include service type, idle energy consumption, and resource location; the real-time state attributes include service queue, estimated service time, completed service queue, completed service time, and resource energy consumption; the configurable attributes include the current start / stop status of the resource, the number of online resources, and the maximum available limit of the resource size.

[0008] Further, in step S2, when a task triggers scheduling, the current workshop resource configuration is determined based on the state model, and a pre-scheduling result is generated by combining the workshop resource configuration and task requirements. Specifically, this includes: when a task arrives, reading the real-time state attributes and configurable attributes in the state model to determine the current quantity of workshop resources, including machine tools and automated guided vehicles; constructing a resource combination set formed by combining machine tools and automated guided vehicles based on the quantity of workshop resources; calculating the expected service time and resource energy consumption for each resource combination in the resource combination set; using task urgency as a weighting factor to weight and sum the expected service time and resource energy consumption to obtain a combination evaluation value; and selecting the resource combination with the smallest combination evaluation value as the pre-scheduling result.

[0009] Furthermore, the method for calculating the task urgency is as follows: Obtain the deadline and current time of the task, and determine the remaining completion time of the task based on the difference between the deadline and the current time; obtain the standard processing time and standard transportation time of the task, and determine the estimated service time of the task based on the sum of the standard processing time and the standard transportation time; determine the urgency of the task based on the ratio of the estimated service time to the remaining completion time.

[0010] Further, in step S2, the pre-scheduling result is mapped to the high-disturbance environment of the digital twin (DT) workshop for simulation to obtain the simulated completion time. The simulated completion time is then processed using a gated loop unit method to obtain the predicted completion time. Specifically, this includes: determining the fluctuation range of manufacturing processing time based on processing records in the workshop's historical operating data; performing path planning based on the resource locations in the state model to determine the obstacle avoidance time of the automated guided vehicle (AGV) during logistics transportation; determining the communication delay of command and data interaction based on the real-time communication status of the local area network between the physical workshop and the digital twin workshop; mapping the pre-scheduling result to the digital twin workshop; and, during the simulation, incorporating the manufacturing processing... The fluctuation range of time, the obstacle avoidance time, and the communication delay are injected into the simulation as disturbance factors to obtain the simulation completion time. The simulation completion time, the number of machine tools and the number of automated guided vehicles in the current digital twin workshop are used as inputs. The time-series features of the inputs are extracted and calculated by a trained gated recurrent unit method to correct the deviation between the simulation environment and the actual operation, and the predicted completion time is output. The training process of the gated recurrent unit method includes: calculating the deviation between the predicted completion time and the historical real completion time using the mean square error as the loss function, backpropagating the deviation based on backpropagation, and iteratively updating the weight parameters of the gated recurrent unit using an adaptive moment estimation algorithm until the loss function converges.

[0011] Further, step S3, calculating the average urgency and normalized system information entropy of the set of tasks to be scheduled, specifically includes: for each task in the set of tasks to be scheduled, if the task urgency of the task is greater than a preset truncation upper limit, the preset truncation upper limit is used as the truncation urgency of the task; otherwise, the task urgency of the task is used as the truncation urgency of the task; wherein, the preset truncation upper limit is less than 1; the effective urgency is obtained by power transformation of the obtained truncation urgency of each task, and the average urgency is obtained by averaging the effective urgency of each task in the set of tasks to be scheduled; the proportion of the effective urgency of each task in the set of tasks to the sum of all effective urgency is used as the probability of the task, and the system information entropy is calculated from the probability of each task according to the Shannon information entropy definition; the normalized system information entropy is obtained by dividing the system information entropy by the logarithm of the total number of tasks to be scheduled.

[0012] Further, in step S3, a resource allocation criterion index is constructed. When the criterion index reaches a preset threshold, production-logistics resource allocation is triggered, and the resources to be activated are added to the available resource pool to form a resource allocation scheme. Specifically, this includes: obtaining the resource allocation criterion index by linearly weighting and summing the average urgency and the normalized system information entropy; triggering production-logistics resource allocation when the criterion index reaches a preset threshold, adding the resources to be activated to the available resource pool to form a resource allocation scheme; the resources to be activated are machine tools or automated guided vehicles of the same type that are pre-deployed in the workshop but not yet started. When they are triggered to be added to the available resource pool, they are started and enter a waiting execution state.

[0013] Furthermore, the real-time resource status data in step S5 includes the resource's service queue, estimated service time, completed service queue, completed service time, and resource energy consumption; the task execution feedback data includes the actual task start time, actual processing time, actual transportation time, and task completion status; the iterative update refers to using the collected real-time resource status data and task execution feedback data to refresh the real-time status attributes and configurable attributes in the production-logistics resource status model.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) Deep coupling of production and logistics results in superior overall performance of scheduling. This invention integrates production and logistics into a unified resource set for joint optimization. It uses the urgency of the task as the weight to perform a weighted comprehensive evaluation of service time and service energy consumption, thereby achieving deep integration of production resources and logistics resources at the scheduling level. It breaks through the limitation of the separate processing of production scheduling and logistics scheduling in the prior art, effectively avoids global deviation caused by fluctuations in local links, and significantly improves the global consistency and collaborative efficiency of the scheduling scheme.

[0015] (2) Information entropy quantifies system uncertainty and digital twin pre-verifies scheduling robustness. This invention constructs task urgency based on predicted completion time and uses Shannon information entropy to calculate system information entropy and normalized system information entropy to achieve quantitative characterization of workshop operation uncertainty. At the same time, by using multiple simulations of the digital twin workshop under high-disturbance environments such as the fluctuation range of manufacturing and processing time, AGV obstacle avoidance time and communication delay, the pre-scheduling results are pre-verified, which significantly improves the robustness of the scheduling scheme to multi-source disturbances and effectively solves the problem of resource allocation lag or excess caused by the lack of uncertainty quantification in the prior art.

[0016] (3) Resource allocation and real-time scheduling are integrated into a closed loop to achieve dynamic balance of multiple objectives. This invention constructs a resource allocation criterion index by linearly weighting the average urgency and the normalized system information entropy. At the time of task triggering, it dynamically determines whether to increase production-logistics resources and uses the allocation result as a constraint condition for real-time scheduling, forming a closed-loop integrated mechanism of "configuration-scheduling-reconfiguration-rescheduling". Compared with the static separation mode of "configuration first, then scheduling" or "schedule only, no configuration" in the prior art, this invention can achieve a dynamic balance between scheduling efficiency, energy consumption cost and task on-time rate, effectively cope with the cumulative impact of continuous disturbances, and ensure the stable operation of the system in a highly dynamic and strongly disturbed workshop.

[0017] (4) Balancing real-time response and global optimization, adapting to engineering practice. This invention utilizes the gated cyclic unit method to quickly predict simulation completion time, combined with resource allocation criteria based on information entropy and a weighted evaluation mechanism. While ensuring global multi-objective optimization capabilities, it effectively reduces computational complexity, overcomes the shortcomings of insufficient real-time response capabilities based on mathematical programming methods (such as FIFO+LINGO) and the lack of global optimization capabilities based on heuristic rules (such as FIFO+SPT), and has good engineering applicability. Attached Figure Description

[0018] Figure 1 A flowchart of the real-time allocation and integration method for production-logistics resources based on DT and information entropy provided by this invention; Figure 2 A comparison diagram showing the effects of the real-time allocation and integration method for production and logistics resources provided by this invention with existing technologies. Detailed Implementation

[0019] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the relevant invention and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0020] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figures 1-2 The real-time allocation and integration method for production-logistics resources based on DT and information entropy in this embodiment of the invention includes: Step S1: Integrate workshop production-logistics resources, process flow, production line layout and real-time workshop operation data to build a digital twin (DT) workshop and construct an iterative production-logistics resource status model.

[0021] Specifically, via local area network and TCP / IP protocol, static attribute data is collected from the physical workshop, including production resource identifiers, service types, capacity limits, spatial locations, and idle energy consumption; and logistics resource identifiers, transportation capacity, current location, and remaining power. Real-time status attribute data is collected from machine tool current processing tasks, service queue length, estimated service time, completed service time, and real-time energy consumption; and automated guided vehicles (AGVs) current load, operating path, obstacle avoidance status, and communication latency. Configurable attribute data is collected from resource start / stop status, online quantity, and scale limit. Process flow data includes the sequence of each order's process steps, standard processing time, and inter-process dependencies. Production line layout data includes machine tool station coordinates, warehouse locations, and logistics channel topology. This data is transformed and semantically mapped in the Unity 3D platform to construct a digital twin workshop that corresponds one-to-one with the physical workshop. Based on this, static attributes, real-time status attributes, and configurable attributes are written into the production-logistics resource status model. This model uses resource identifiers as indexes to establish the association between static attributes and real-time status attributes, and establishes the mapping relationship between resource scale and task load through configurable attributes. When a task is triggered for scheduling, the state model is read to determine the available quantity and location distribution of machine tools and automated guided vehicles, providing a data foundation for constructing a resource combination set. During the simulation phase, the resource location data in the state model supports path planning and obstacle avoidance time calculation, while real-time energy consumption data supports energy consumption extrapolation during simulation. During the resource configuration phase, the online resource quantity and scale limit in the configurable attributes of the state model are invoked to determine the space for increasing the allocation of resources to be activated. During the scheduling execution phase, the real-time resource status data and task execution feedback data returned from the physical workshop are written into the state model, refreshing the real-time status attributes and configurable attributes to achieve iterative updates of the state model. Thus, the digital twin workshop and the state model form a data-driven closed-loop foundation: the state model provides input data, including resource quantity and disturbance factors, to the gated loop unit method; the predicted completion time output by the gated loop unit method drives the update of task attributes in the state model; information entropy calculation directly reads the task deadline and service time data in the state model; the triggering result of the resource configuration criterion is fed back to the state model by modifying the number of online resources in the configurable attributes; the real-time scheduling result changes the service queue and expected service time in the state model through resource combination selection. Taking a flexible workshop containing 8 machine tools, 6 automated guided vehicles, and 20 orders as an example, the initial state model records that 1 machine tool is online, 1 automated guided vehicle is online, and the remaining resources are in an inactive state; when an order arrives and triggers scheduling, the state model is read to determine that the available resources are 1 machine tool and 1 automated guided vehicle.

[0022] Step S2: When a task triggers scheduling, the current workshop resource configuration is determined based on the state model. A pre-scheduling result is generated by combining the workshop resource configuration and task requirements. The pre-scheduling result is mapped to the high-disturbance environment of the digital twin (DT) workshop for simulation to obtain the simulation completion time. The simulated completion time is processed using the gated loop unit method to obtain the predicted completion time. The predicted completion time of the task in the set of tasks to be scheduled is updated.

[0023] Specifically, when a task triggers scheduling, this method reads the configurable attributes and real-time state attributes from the state model to determine the current number of online machine tools and online automated guided vehicles (AGVs), and constructs a resource combination set formed by combining machine tools and AGVs accordingly. For the task to be scheduled, this method calculates the service time and service energy consumption of each resource combination. The service time is obtained by adding the machine tool processing time and the AGV transportation time, and the service energy consumption is obtained by adding the machine tool processing energy consumption and the AGV transportation energy consumption. This method obtains the task's deadline and current time to determine the remaining completion time, obtains the task's standard processing time and standard transportation time to determine the estimated service time, determines the task urgency based on the ratio of the estimated service time to the remaining completion time, and uses the task urgency as a weighting factor to weight and sum the service time and service energy consumption to obtain a combination evaluation value. The resource combination with the smallest evaluation value is selected as the pre-scheduling result. This method maps pre-scheduling results to a digital twin workshop, injecting fluctuations in manufacturing processing time, obstacle avoidance time of automated guided vehicles (AGVs), and communication delays as disturbance factors during simulation. The simulated completion time is obtained through multiple iterations. The method takes the simulated completion time, along with the current number of machine tools and AGVs, as input. It uses a gated cyclic unit method to extract and calculate temporal features to correct deviations between the simulation environment and actual operation, outputting a predicted completion time. This predicted completion time is then updated in the attribute fields of the corresponding tasks in the set of tasks to be scheduled. In this process, the digital twin workshop provides simulated completion time data containing disturbance factors, and the gated cyclic unit method performs temporal correction on this data based on parameters trained using historical real completion times. It should be noted that during the initial deployment phase in the workshop, since there was no historical real completion time data to train the gated recurrent unit, the system adopted the following cold start strategy: First, at least 500 simulations were conducted using the digital twin workshop under various disturbance scenarios, and the simulated completion time was used as an approximate predicted value for scheduling. Simultaneously, as the physical workshop gradually went into operation, the system continuously collected real completion time data and accumulated it into a training set. When the training set sample size reached at least 200 samples, formal training of the gated recurrent unit was initiated. The mean squared error was used as the loss function to calculate the deviation between the predicted completion time and the actual completion time. Weight parameters were updated based on backpropagation, and thereafter, retraining was performed every preset period to continuously improve prediction accuracy. The high-disturbance simulation of the digital twin workshop and the gated recurrent unit method form a functionally mutually supportive predictive whole, together constituting the prediction means. The state model provides resource data for pre-scheduling generation, the pre-scheduling results drive the digital twin simulation, the simulation data supports the prediction of the gated recurrent unit, and the prediction results, in turn, update the task attributes in the state model, forming a data closed loop.Taking a flexible workshop containing 8 machine tools, 6 automated guided vehicles (AGVs), and 20 orders as an example, when an order triggers scheduling, the state model records that the current online resources are 1 machine tool and 1 AGV. Based on this, the method generates a pre-scheduling result for this resource combination. After extrapolating the manufacturing time fluctuation, obstacle avoidance delay, and communication delay in the digital twin workshop, the result yields a simulated completion time of 420 seconds. The gated loop unit method, combined with historical operating data, corrects this deviation and outputs a predicted completion time of 385 seconds, which is then written into the task set.

[0024] Step S3: Calculate the average urgency and normalized system information entropy of the set of tasks to be scheduled based on the predicted completion time, and construct resource allocation criteria. When the criteria reach a preset threshold, trigger production-logistics resource allocation, add the resources to be activated to the available resource pool, and form a resource allocation scheme.

[0025] Specifically, for each task in the set of tasks to be scheduled, when the task urgency of the task is greater than a preset truncation limit (e.g., 0.9), the preset truncation limit is used as the truncation urgency of the task; otherwise, the task urgency is used as the truncation urgency. Wherein, the preset truncation limit is less than 1, and the truncation urgency is subjected to a power transformation (the power value of the power transformation ranges from 1.5 to 3, preferably 2) to obtain the effective urgency. The average urgency is obtained by averaging the effective urgency of all tasks. The effective urgency of each task is then calculated. The proportion of the effective urgency to the sum of all effective urgency is used as the probability of the task. The system information entropy is calculated from the probabilities of each task according to the Shannon definition of information entropy. The system information entropy is divided by the logarithm of the total number of tasks to obtain the normalized system information entropy. The average urgency and the corresponding normalized system information entropy are weighted and summed using a linear weighting coefficient of 0.35 for the average urgency and 0.65 for the normalized system information entropy to obtain the resource allocation criterion index, i.e., criterion index = 0.35 × average urgency + 0.65 × normalized system information entropy. When this criterion index reaches a preset threshold, production-logistics resource allocation is triggered. Pre-deployed but not yet started machine tools of the same type are added to the available resource pool as resources to be activated, forming a resource allocation scheme. Taking a flexible work workshop containing 8 machine tools, 6 automated guided vehicles (AGVs), and 20 orders as an example, at a certain moment, the set of tasks to be scheduled contains 5 tasks. The predicted completion time of each task is processed as described above to obtain the effective urgency level, with an average urgency level of 0.65. The system information entropy is normalized to 0.82, and the linearly weighted criterion index is 0.7605. This value reaches the preset threshold of 0.70, triggering resource allocation. Two machine tools to be activated and one AGV to be activated are added to the available resource pool, increasing the number of online machine tools from 1 to 3 and the number of online AGVs from 1 to 2.

[0026] Step S4: Based on the predicted completion time, determine the individual information entropy of the task. Using the individual information entropy of the task as the weight, calculate the service duration and energy consumption of each production-logistics resource combination in the available resource pool under the resource allocation scheme to obtain the comprehensive evaluation value of each production-logistics resource combination. Select the resource combination with the smallest comprehensive evaluation value as the optimal real-time scheduling result corresponding to the current task.

[0027] Specifically, the predicted completion time output in step S2 is read, and the task urgency is determined according to the method described in step S2, combined with the task deadline. A negative logarithmic transformation with base e is performed on this urgency to obtain the task's individual information entropy. Since the task urgency may exceed the (0,1) interval, before performing the negative logarithmic transformation, the task urgency is truncated to a preset truncation upper limit, ensuring that the truncated task urgency falls within the (0,1) interval, thus guaranteeing that the negative logarithmic transformation result is a non-negative real number. Subsequently, the negative logarithmic transformation result is normalized by dividing it by the sum of the negative logarithmic transformation results of all tasks in the task set to be scheduled, so that the task's individual information entropy ranges from [0,1]. This task's individual information entropy represents the task's state characteristics in the task set to be scheduled and is used as a valid weight in subsequent weighted calculations. The available resource pool determined by the resource allocation scheme formed in step S3 is read, and the available resources are obtained. The production-logistics resource combinations formed by the machine tools and automated guided vehicles (AGVs) in the pool are analyzed. For each resource combination, its service duration and service energy consumption are calculated. The service duration is obtained by adding the machine tool processing time and the AGV transportation time. The service energy consumption is obtained by adding the machine tool processing energy consumption, the AGV transportation energy consumption, and the idle energy consumption of the corresponding resource. Idle energy consumption is the maintenance energy consumption of the machine tool or AGV while waiting to execute a task, which is read from the static attributes in the state model. The service duration and service energy consumption of each resource combination are normalized to their maximum and minimum values. Specifically, for the service duration, the service duration of all resource combinations in the currently available resource pool is traversed. The service duration is calculated by taking the maximum and minimum values. The normalized service duration for each combination is calculated as (service duration of the combination - minimum value) / (maximum value - minimum value). The same maximum and minimum value normalization method is applied to the service energy consumption to obtain the normalized service energy consumption for each combination. After normalization, the normalized service duration and normalized service energy consumption of all combinations are mapped to the [0,1] interval. The normalized service duration and service energy consumption are weighted and summed using the task individual information entropy as the weight to obtain the comprehensive evaluation value of each resource combination. The resource combination with the smallest comprehensive evaluation value is selected as the optimal real-time scheduling result for the current task. In the above process, the predicted completion time data drives the generation of task individual information entropy. In this system, the individual task information entropy data drives the weight allocation in resource combination evaluation, while the service duration and service energy consumption data of the resource combination drive the generation of the comprehensive evaluation value. The three types of data are passed sequentially and mutually constrain each other. The algorithmic features of the individual task information entropy and the technical features of the production-logistics resource combination selection are mutually supportive and interactive. The individual task information entropy quantifies the urgency of the task and the system state characteristics. This quantification result directly determines the weight allocation of service duration and service energy consumption in the comprehensive evaluation value, making high-urgency tasks tend to prioritize time and low-urgency tasks tend to prioritize energy consumption. The algorithmic features and technical features together constitute the technical means to solve the multi-objective collaborative scheduling problem.Taking a flexible workshop containing 8 machine tools, 6 automated guided vehicles (AGVs), and 20 orders as an example, when a task triggers scheduling, step S3 adds 2 machine tools to be activated and 1 AGV to be activated to the available resource pool. The current available resource pool contains 3 machine tools and 2 AGVs, forming 6 production-logistics resource combinations. The task individual information entropy determined by the predicted completion time is 0.8. After the normalized values ​​of service duration and service energy consumption of each combination are weighted by this weight, the combination with the smallest evaluation value is selected as the real-time scheduling result. The machine tools corresponding to this combination perform the processing operation, and the AGVs corresponding to this combination perform the material transportation operation.

[0028] Step S5: Send the resource configuration scheme and the optimal real-time scheduling result to the physical workshop for execution, collect real-time status data of production-logistics resources and task execution feedback data of the physical workshop, update the status model and the digital twin (DT) workshop, and wait for the next task to trigger scheduling.

[0029] Specifically, the resource allocation plan and optimal real-time scheduling results are sent to the physical workshop for execution via TCP / IP protocol. Upon receiving the resource allocation command, the machine tool or automated guided vehicle (AGV) awaiting activation changes from a stopped state to a ready-to-start state. This state change data is transmitted back to the state model to update the resource start / stop status and online quantity in the configurable attributes. The optimal production-logistics resource combination that receives the real-time scheduling command adds the task to the execution queue. Based on the service queue length, it determines whether to enter the queue for execution or begin execution. During the operation of the physical workshop, data is collected through the machine tool CNC system, the AAV control system, and the sensor network. Real-time resource status data and task execution feedback data are generated. The real-time resource status data includes the service queue, estimated service time, completed service queue, completed service time, and resource energy consumption. The task execution feedback data includes the actual task start time, actual processing time, actual transportation time, and task completion status. The above data is written into the status model to update the real-time status attributes and synchronously mapped to the digital twin workshop to refresh the resource status and task progress, so that the digital twin workshop and the physical workshop maintain consistency. After the update is completed, the system waits for the next task to be triggered, which is generated by the arrival of a new order or the start of the previous process. Taking a flexible workshop containing 8 machine tools, 6 automated guided vehicles (AGVs), and 20 orders as an example, a certain task is scheduled to be executed by machine tool No. 3 and AGV No. 2. The actual processing time of the machine tool is 85 seconds, the actual transportation time of the AGV is 32 seconds, and the actual energy consumption is 1.2 kWh. The task completion status is normal completion. After the above data is transmitted back through the workshop network, the completed service time of machine tool No. 3 in the state model increases by 85 seconds, the resource energy consumption increases by 1.2 kWh, the service queue of AGV No. 2 decreases by 1 item, and the status of the corresponding objects in the digital twin workshop is updated synchronously. When the next order arrives and triggers scheduling, step S2 reads the updated state model to determine the completed service time and energy consumption data of machine tool No. 3, and generates a new pre-scheduling result accordingly.

[0030] To verify the feasibility of the production-logistics resource real-time allocation and integration method based on DT and information entropy (hereinafter referred to as RT-RCSM) of this invention, a flexible operation workshop production scenario was constructed with 20 orders in a layout of 8 machine tools and 6 AGVs. Under the integration framework of this algorithm, it was compared with typical scheduling rule combinations such as FIFO+SPT, FIFO+LING and FIFO+MEC to verify the impact of different random order arrival levels on completion time, energy consumption and delay. The experiment was run 100 times according to the experimental conditions in Table 1, and the mean and standard deviation were recorded.

[0031] Table 1 Experimental Layout

[0032] Table 2 Comparison of Indicators for the Four Methods

[0033] Experimental results show that the method of this invention maintains good overall performance under different order arrival rates, with optimal completion time and energy consumption, and achieves zero delays. This indicates that the method can ensure on-time order completion while balancing production efficiency and energy consumption control, demonstrating good stability and applicability. Initially, the system has limited resources. As more and more orders arrive at the workshop, the current production conditions cannot meet the processing demands. By integrating resources through this method, the processing energy consumption can be reduced while meeting the processing needs of all orders.

[0034] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0035] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time production-logistics resource allocation and integration method based on DT and information entropy, characterized in that, include: Step S1: Integrate workshop production-logistics resources, process flow, production line layout and real-time workshop operation data to build a digital twin (DT) workshop and construct an iterative production-logistics resource status model; Step S2: When a task triggers scheduling, the current workshop resource configuration is determined based on the state model. A pre-scheduling result is generated by combining the workshop resource configuration and task requirements. The pre-scheduling result is mapped to the high-disturbance environment of the digital twin (DT) workshop for simulation to obtain the simulation completion time. The simulated completion time is processed using the gated loop unit method to obtain the predicted completion time. The predicted completion time of the task in the set of tasks to be scheduled is updated. Step S3: Calculate the average urgency and normalized system information entropy of the set of tasks to be scheduled based on the predicted completion time, and construct resource allocation criteria. When the criteria reach a preset threshold, trigger production-logistics resource allocation, add the resources to be activated to the available resource pool, and form a resource allocation scheme. Step S4: Based on the predicted completion time, determine the individual information entropy of the task. Using the individual information entropy of the task as the weight, calculate the service duration and energy consumption of each production-logistics resource combination in the available resource pool under the resource allocation scheme to obtain the comprehensive evaluation value of each production-logistics resource combination. Select the resource combination with the smallest comprehensive evaluation value as the optimal real-time scheduling result corresponding to the current task. Step S5: Send the resource configuration scheme and the optimal real-time scheduling result to the physical workshop for execution, collect real-time status data of production-logistics resources and task execution feedback data of the physical workshop, update the status model and the digital twin (DT) workshop, and wait for the next task to trigger scheduling.

2. The real-time allocation and integration method for production-logistics resources based on DT and information entropy according to claim 1, characterized in that, The digital twin (DT) workshop is built on Unity3D software and achieves bidirectional communication with the physical workshop through local area network interconnection and TCP / IP protocol.

3. The real-time allocation and integration method for production-logistics resources based on DT and information entropy according to claim 1, characterized in that, The state model defines static attributes, real-time state attributes, and configurable attributes for each resource. The static attributes include service type, idle energy consumption, and resource location; The real-time status attributes include service queue, estimated service time, completed service queue, completed service time, and resource energy consumption; The configurable attributes include the current start / stop status of the resource, the number of online resources, and the maximum available limit of the resource size.

4. The real-time allocation and integration method for production-logistics resources based on DT and information entropy according to claim 3, characterized in that, In step S2, when a task triggers scheduling, the current workshop resource configuration is determined based on the state model, and a pre-scheduling result is generated by combining the workshop resource configuration and task requirements. Specifically, this includes: When a task arrives, the real-time status attributes and configurable attributes in the state model are read to determine the current quantity of workshop resources, which include machine tools and automated guided vehicles. Construct a resource combination set consisting of machine tools and automated guided vehicles based on the quantity of workshop resources; The estimated service time and resource energy consumption for each resource combination in the computing resource combination set to perform a task; The estimated service time and resource energy consumption are weighted and summed using task urgency as a weighting factor to obtain a combined evaluation value. The resource combination with the smallest combined evaluation value is selected as the pre-scheduling result.

5. The real-time allocation and integration method for production-logistics resources based on DT and information entropy according to claim 4, characterized in that, The method for calculating the urgency of the task is as follows: Obtain the deadline and current time of the task, and determine the remaining completion time of the task based on the difference between the deadline and the current time; Obtain the standard processing time and standard transportation time for the task, and determine the estimated service time for the task based on the sum of the standard processing time and the standard transportation time; The urgency of the task is determined based on the ratio of the estimated service time to the remaining completion time.

6. The real-time allocation and integration method for production-logistics resources based on DT and information entropy according to claim 4, characterized in that, In step S2, the pre-scheduling result is mapped to the high-disturbance environment of the digital twin (DT) workshop for simulation to obtain the simulation completion time. The simulated completion time is then processed using the gated cyclic unit method to obtain the predicted completion time. Specifically, this includes: The range of fluctuation in manufacturing processing time is determined based on processing records in the workshop's historical operation data; Path planning is performed based on the resource locations in the state model to determine the obstacle avoidance time of the automated guided vehicle during logistics transportation. Based on the real-time communication status of the local area network between the physical workshop and the digital twin workshop, the communication delay for instruction and data interaction is determined; The pre-scheduling results are mapped to the digital twin workshop, and the fluctuation range of the manufacturing and processing time, the obstacle avoidance time, and the communication delay are injected as disturbance factors into the simulation operation to obtain the simulation completion time. The simulation completion time, the number of machine tools and the number of automated guided vehicles in the current digital twin workshop are used as inputs. The time-series features of the inputs are extracted and calculated using a trained gated recurrent unit method to correct the deviation between the simulation environment and the actual operation, and the predicted completion time is output. The training process of the gated recurrent unit method includes: calculating the deviation between the predicted completion time and the historical actual completion time using the mean squared error as the loss function, backpropagating the deviation based on backpropagation, and iteratively updating the weight parameters of the gated recurrent unit using the adaptive moment estimation algorithm until the loss function converges.

7. The real-time allocation and integration method for production-logistics resources based on DT and information entropy according to claim 5, characterized in that, Step S3 calculates the average urgency and normalized system information entropy of the set of tasks to be scheduled, specifically including: For each task in the set of tasks to be scheduled, if the task urgency of the task is greater than the preset truncation limit, the preset truncation limit shall be used as the truncation urgency of the task; otherwise, the task urgency of the task shall be used as the truncation urgency of the task; wherein, the preset truncation limit is less than 1. The effective urgency is obtained by power transformation of the truncated urgency of each task, and the average urgency is obtained by averaging the effective urgency of each task in the set of tasks to be scheduled. The probability of a task is the ratio of its effective urgency to the sum of all effective urgency in the set of tasks to be scheduled. The system information entropy is then calculated from the probabilities of each task according to the Shannon definition of information entropy. The normalized system information entropy is obtained by dividing the system information entropy by the logarithm of the total number of tasks to be scheduled.

8. The real-time allocation and integration method for production-logistics resources based on DT and information entropy according to claim 7, characterized in that, Step S3 involves constructing resource allocation criteria. When these criteria reach a preset threshold, production-logistics resource allocation is triggered, adding the resources to be activated to the available resource pool to form a resource allocation scheme. Specifically, this includes: The resource allocation criterion index is obtained by linearly weighting and summing the average urgency and the normalized system information entropy. When the criterion index reaches the preset threshold, the production-logistics resource configuration is triggered, and the resources to be activated are added to the available resource pool to form a resource configuration scheme. The resources to be activated are machine tools or automated guided vehicles of the same type that are pre-deployed in the workshop but not yet turned on. When they are triggered to be added to the available resource pool, they are turned on and enter the waiting execution state.

9. The real-time allocation and integration method for production-logistics resources based on DT and information entropy according to claim 3, characterized in that, The real-time resource status data mentioned in step S5 includes the resource's service queue, estimated service time, completed service queue, completed service time, and resource energy consumption; the task execution feedback data includes the actual task start time, actual processing time, actual transportation time, and task completion status. The iterative update refers to refreshing the real-time status attributes and configurable attributes in the production-logistics resource status model using the collected real-time resource status data and task execution feedback data.

Images