A method for scheduling a rhythmized AGV feeding path of a workshop based on a resident risk perception

Abstract

The present application belongs to the technical field of path scheduling, and particularly relates to a kind of rhythm workshop AGV feeding path scheduling method of residence risk perception, steps include: obtaining the feeding task data of current workshop, constructs feeding task input data set;Calculate the time state of each candidate station in the unvisited station set and the residence pressure of each candidate station;Suppose that a certain candidate station is visited first, construct the subsequent predicted access sequence corresponding to the candidate station;In turn, calculate the service completion time, completion time margin and late time of each predicted station under the condition of visiting the candidate station first, calculate the downstream propagation pressure generated by visiting the current candidate station first;Build integrated access cost, get the complete AGV feeding access sequence;Based on AGV feeding access sequence, generate AGV feeding control instruction, and send to AGV to execute workshop feeding task.The present application can reduce continuous delay and beat fluctuation, improve feeding punctuality and stability.

Description

Technical Field

[0001] This invention belongs to the field of path scheduling technology, specifically relating to a rhythmic workshop AGV feeding path scheduling method based on dwell risk perception. Background Technology

[0002] A takt-time workshop is a workshop that organizes and manages production around the production takt time. In such workshops, the material requirements, feeding times, and work sequences of each workstation need to match the production takt time to ensure continuous and stable operation of the production line. As manufacturing systems develop towards flexibility, automation, and just-in-time (JIT) production, internal workshop logistics are gradually shifting from traditional batch distribution to fixed-point, timed material delivery tailored to specific workstations and production takt times. Automated Guided Vehicles (AGVs), as crucial equipment for material delivery in the workshop, need to deliver materials to the corresponding workstations in a timely manner based on their material requirements, service time windows, and priorities.

[0003] In the material feeding process of a rhythmic workshop, AGVs not only need to complete the travel tasks between the distribution center and each workstation, but also need to perform operations such as docking, unloading, barcode scanning confirmation, status verification, and material handover at the workstations. Therefore, they will incur a certain amount of dwell time at the workstations. Existing workshop logistics path scheduling problems are usually abstracted as vehicle routing problems with time windows. An optimization model is constructed by considering factors such as distance between nodes, vehicle capacity, service time windows, and workstation service time, and AGV access sequences are generated using genetic algorithms, ant colony algorithms, particle swarm optimization algorithms, or neighborhood search algorithms. In these methods, workstation dwell time is usually only considered as a fixed time consumption item in the recursive calculation of arrival and completion times.

[0004] However, in a rhythmic material feeding scenario, the impact of workstation dwell time is not limited to the current workstation. If a workstation with a long dwell time is assigned to an inappropriate access location, the AGV's prolonged stay at that workstation will continuously consume the available execution time of subsequent tasks, compressing the completion time margin of subsequent workstations, and potentially causing the risk of lateness to propagate along the access sequence, leading to continuous lateness and fluctuations in production rhythm. Existing scheduling methods typically focus on optimizing the total path distance, number of vehicles, or task completion time, lacking explicit modeling of the downstream time margin compression effect and the risk of lateness propagation caused by dwell time, making it difficult to identify in advance the impact of long-dwelling workstations on the overall stability of the material feeding path.

[0005] Furthermore, existing constructive path scheduling methods typically rely on local judgments based on node transfer distance, service completion deadline, time window urgency, or workstation priority when selecting the next workstation to visit, lacking a forward-looking assessment of whether prioritizing the current candidate workstation will affect the timely delivery of materials to subsequent workstations. When a workstation is urgent, but prioritizing its service could significantly increase the risk of delays for subsequent workstations, existing methods struggle to comprehensively balance the necessity of the current visit with the downstream propagation impact. Summary of the Invention

[0006] In view of the shortcomings of the prior art, the purpose of this invention is to provide a rhythmic workshop AGV feeding path scheduling method based on dwell risk perception. By jointly sensing the dwell pressure at the workstation and the downstream propagation pressure, the AGV feeding access sequence is iteratively constructed, which can reduce continuous delays and rhythm fluctuations, and improve the timeliness and stability of feeding.

[0007] To achieve the above objectives, this invention provides a method for scheduling the feeding path of AGVs in a workshop based on dwell risk perception, comprising the following steps: S1. Obtain the material feeding task data of the current workshop, abstract the distribution center and each material feeding station as path scheduling nodes, construct the material feeding task input data set, and initialize the set of unvisited stations, the current node, the current completion time and the AGV material feeding access sequence. S2. Based on the input data set of the material feeding task, calculate the time status of each candidate workstation in the set of unvisited workstations, including the estimated arrival time, service start time, service completion time, completion time margin, and late time, and calculate the self-retention pressure of each candidate workstation. S3. Assuming that a certain candidate workstation is visited first, remove the candidate workstation from the current set of unvisited workstations, and select several predicted workstations from the remaining unvisited workstations to construct the subsequent predicted visit sequence corresponding to the candidate workstation. S4. Taking the service completion time of the candidate workstation as the prediction starting point, according to the subsequent predicted access sequence corresponding to the candidate workstation, calculate the service completion time, completion time margin and late time of each predicted workstation under the condition of prioritizing access to the candidate workstation. Based on the completion time margin and late time of each predicted workstation, calculate the downstream propagation pressure generated by prioritizing access to the current candidate workstation. S5. Combining node transfer distance, downstream propagation pressure, self-residence pressure, and workstation priority, construct a comprehensive access cost for each candidate workstation. Select the candidate workstation with the lowest access cost from the current unvisited workstation set as the next access workstation, add it to the AGV feeding access sequence, and update the current node, current completion time, and unvisited workstation set. If the updated unvisited workstation set is not empty, return to S2 with the updated scheduling state to continue execution until all workstations to be fed have been accessed, resulting in a complete AGV feeding access sequence. S6. Generate AGV feeding control instructions based on the AGV feeding access sequence and send them to the AGV to execute the workshop feeding task.

[0008] As a preferred embodiment of the present invention, in S1, the material feeding task input data set Represented as: ; In the formula, This represents a set of nodes. Node 0 corresponds to the distribution center, and n is the number of workstations waiting to be fed. Nodes 1 to n each correspond to n workstations waiting to be fed. The currently unvisited workstations waiting to be fed constitute the candidate workstation set. Represents the node transition distance matrix. Represents the node transition time matrix. , These represent the transition distance and transition time from node i to node j, respectively. This represents the service time window of node k. Let k be the earliest allowed service time. The service completion deadline for node k; This indicates the dwell time of node k; This indicates the workstation priority of node k; Let the set of unvisited workstations be... Initialize the AGV feeding access sequence Q as an empty sequence. Let the current node be i and the current completion time be . Initially, i=0, and the initial completion time is... .

[0009] In a preferred embodiment of the present invention, in step S2, for any candidate workstation j in the set of unvisited workstations U, the time state of candidate workstation j is calculated: ; In the formula, This indicates the estimated time for the AGV to arrive at candidate workstation j; Indicates the service start time of candidate workstation j; Indicates the service completion time of candidate workstation j; This indicates the completion time margin for candidate workstation j; Indicates the late arrival time of candidate workstation j; Let j be the earliest allowed service time for candidate workstation j. The deadline for completing the service for candidate workstation j; This indicates the dwell time of candidate workstation j.

[0010] As a preferred embodiment of the present invention, when When the AGV arrives at candidate workstation j ahead of schedule, it needs to wait until the earliest permitted service time. Service begins afterward; when When, it means that candidate job j can complete the service no later than the service completion deadline. Complete the feeding service; when When, it indicates that candidate workstation j is late, and the lateness time is determined by... express.

[0011] As a preferred embodiment of the present invention, in S2, for candidate station j, its own residence pressure Represented as: ; In the formula, A preset positive number used to prevent the denominator from being zero, and with They have the same time dimension; This is the late amplification factor; This is the preset time scale parameter.

[0012] In a preferred embodiment of the present invention, in step S3, assuming that candidate workstation j is accessed first, candidate workstation j is removed from the current set of unaccessed workstations to obtain a set of remaining unaccessed workstations. For any candidate workstation x in this set, the service completion deadline time of candidate workstation x, the workstation priority of candidate workstation x, and the node transfer distance from candidate workstation j to candidate workstation x are performed using minimum-maximum normalization to obtain the normalized service completion deadline time. Normalized chemical level priority and normalized node transition distance ; Construct a prediction ranking index for candidate workstation x based on the normalization results. : ; In the formula, , , These are non-negative ranking weight coefficients; according to Sort the remaining unvisited workstations in ascending order, select the top h workstations as predicted workstations, and construct the subsequent predicted access sequence for candidate workstation j. : ; In the formula, This represents the u-th predicted workstation in the subsequent predicted access sequence.

[0013] In a preferred embodiment of the present invention, in step S4, candidate workstation j is taken as the prediction starting point. And let the service completion time of the predicted starting point be For the u-th predicted workstation in the subsequent predicted access sequence Its service completion time Completion time margin Late time They are respectively: ; ; ; In the formula, This represents the (u-1)th predicted workstation in the subsequent predicted access sequence. Service completion time; Indicates the predicted workstation To the predicted work station Transfer time; Indicates the predicted workstation The earliest permitted service time; Indicates the predicted workstation Length of stay for service; Indicates the predicted workstation The deadline for service completion; Downstream propagation pressure corresponding to candidate station j Represented as: ; In the formula, Represents the propagation attenuation coefficient and ; When the subsequent predicted access sequence corresponding to candidate workstation j is empty, let .

[0014] As a preferred embodiment of the present invention, in step S5, after calculating the self-retention pressure and downstream propagation pressure of candidate workstation j, for any candidate workstation j in the current unvisited workstation set U, the node transfer distance from the current node i to candidate workstation j, the downstream propagation pressure of candidate workstation j, the self-retention pressure, and the workstation priority are performed using minimum-maximum normalization within the current unvisited workstation set U to obtain the normalized node transfer distance. Normalized downstream transmission pressure Normalization of its own residence pressure and normalization of chemical grade priority ; Construct the access cost of candidate workstation j based on the normalization results. : ; In the formula, , , , These represent distance weight, downstream propagation pressure weight, self-retention pressure weight, and priority weight, respectively. according to Sort the candidate workstations in the currently unvisited workstation set U in ascending order, and select the candidate workstation with the lowest access cost as the next workstation to be visited: ; In the formula, Indicates the next workstation selected for this round; Will Add to the current AGV feeding access sequence and update the current node. Update the current completion time to the access time. The service completion time is then removed from the set U of currently unvisited workstations. The updated set of unvisited workstations is obtained. If it is not empty, the updated scheduling state is returned to S2 to continue execution until all workstations to be fed have been visited, and a complete AGV feeding access sequence is obtained.

[0015] The beneficial effects of this invention are: This invention, based on the calculation of the expected arrival time, service completion time, completion time margin, and lateness time of candidate workstations, further constructs the self-residence pressure of candidate workstations. Under the assumption of prioritizing access to the candidate workstation, it constructs a subsequent predicted access sequence to predict the time status and lateness of subsequent workstations. This quantifies the impact of the current access decision on the compression of the completion time margin of subsequent workstations and the risk of lateness propagation. It can identify the continuous lateness problem that may be caused by long-staying workstations in advance during the path construction stage, overcoming the shortcomings of existing methods that only perform static time recursion and lack forward-looking assessment of downstream risks.

[0016] This invention normalizes node transfer distance, downstream propagation pressure, self-residence pressure, and workstation priority, and incorporates them into the candidate workstation access cost function. This allows the AGV to consider travel cost, the urgency of accessing the current workstation, the feasibility of subsequent tasks, and the importance of the workstation when selecting the next workstation to visit. By iteratively selecting the candidate workstation with the lowest access cost and dynamically updating the scheduling status, it can reduce decision-making biases that are locally reasonable but overall unstable, reduce continuous delays and production rhythm fluctuations, and improve the timeliness of AGV material feeding tasks, path execution stability, and scheduling reliability in rhythm-based workshops. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the principle of this invention; Figure 2 This is a schematic diagram of the AGV feeding access sequence using the nearest neighbor algorithm in a typical workshop feeding example; Figure 3 This is a schematic diagram of the AGV material feeding access sequence in a typical workshop material feeding example using the earliest deadline priority algorithm. Figure 4 This is a schematic diagram of the AGV feeding access sequence of the minimum time margin first algorithm in a typical workshop feeding example; Figure 5 This is a schematic diagram of the AGV feeding access sequence in a typical workshop feeding example, which uses an ablation algorithm to remove downstream propagation pressure. Figure 6 This is a schematic diagram of the AGV feeding access sequence in Example 1 of a typical workshop feeding example. Detailed Implementation

[0018] The embodiments of the present invention will be further described below with reference to the accompanying drawings: Example 1: As Figure 1 As shown, a method for scheduling the feeding path of AGVs in a rhythmic workshop based on resident risk perception includes the following steps: S1. Obtain the material feeding task data of the current workshop, abstract the distribution center and each material feeding station as path scheduling nodes, construct the material feeding task input data set, and initialize the set of unvisited stations, the current node, the current completion time and the AGV material feeding access sequence. S2. Based on the input data set of the material feeding task, calculate the time status of each candidate workstation in the set of unvisited workstations, including the estimated arrival time, service start time, service completion time, completion time margin, and late time, and calculate the self-retention pressure of each candidate workstation. S3. Assume that a certain candidate workstation is visited first (that is, for each candidate workstation in the current unvisited workstation set, assume that the candidate workstation is visited first), remove the candidate workstation from the current unvisited workstation set, and select several predicted workstations from the remaining unvisited workstations to construct the subsequent predicted visit sequence corresponding to the candidate workstation. S4. Taking the service completion time of the candidate workstation as the prediction starting point, according to the subsequent predicted access sequence corresponding to the candidate workstation, calculate the service completion time, completion time margin and late time of each predicted workstation under the condition of prioritizing access to the candidate workstation. Based on the completion time margin and late time of each predicted workstation, calculate the downstream propagation pressure generated by prioritizing access to the current candidate workstation. S5. Combining node transfer distance, downstream propagation pressure, self-residence pressure, and workstation priority, construct a comprehensive access cost for each candidate workstation. Select the candidate workstation with the lowest access cost from the current unvisited workstation set as the next access workstation, add it to the AGV feeding access sequence, and update the current node, current completion time, and unvisited workstation set. If the updated unvisited workstation set is not empty, return to S2 with the updated scheduling state and continue execution until all workstations to be fed have been accessed (the updated unvisited workstation set is empty), thus obtaining a complete AGV feeding access sequence. S6. Generate AGV feeding control instructions based on the AGV feeding access sequence and send them to the AGV to execute the workshop feeding task.

[0019] In S1, the workshop AGV scheduling controller can be used to obtain the current material feeding task data of the workshop through the workshop manufacturing execution system, AGV scheduling system, workstation status acquisition device or production line control system. The workshop AGV scheduling controller can be implemented by an industrial control computer, server, edge computing device or scheduling module in AGV scheduling system. It is used to execute the dwell risk perception path scheduling method in this embodiment and issue material feeding access sequence or material feeding control command to AGV.

[0020] Input data set for material feeding task Represented as: ; In the formula, This represents a set of nodes. Node 0 corresponds to the distribution center, and n is the number of workstations to be fed. Nodes 1 to n correspond to n workstations to be fed. The workstations to be fed that have not yet been visited constitute a set of candidate workstations. The candidate workstations are the workstations to be fed that have not yet been visited and can be selected by this scheduling in each round of selecting the next workstation to be visited. Represents the node transition distance matrix. Represents the node transition time matrix. , These represent the transition distance and transition time from node i to node j, respectively. This represents the service time window of node k (i.e., the workstation to be fed materials corresponding to workstation k; in subsequent steps, the candidate workstation will be used for description). Let k be the earliest allowed service time. The service completion deadline for node k indicates that the AGV not only needs to arrive at node k within the specified time, but also needs to... Complete the material feeding service at this workstation beforehand; This indicates the dwell time of node k; This indicates the workstation priority of node k; Workstation priority can be determined based on the importance of the production task corresponding to the workstation, the urgency of the materials, the workstation cycle time constraint, the production plan priority, or the manual preset level, and can be obtained or set by the workshop manufacturing execution system, the production line control system, or the workshop AGV scheduling controller. The larger the value, the higher the priority.

[0021] Let the set of unvisited workstations be... Initialize the AGV feeding access sequence Q as an empty sequence. Let the current node be i, and the current completion time (the service completion time of node i) be... Initially, i=0, and the initial completion time is... .

[0022] In S2, for any candidate workstation j in the set of unvisited workstations U, calculate the time state of candidate workstation j: ; In the formula, This indicates the estimated time for the AGV to arrive at candidate workstation j; Indicates the service start time of candidate workstation j; Indicates the service completion time of candidate workstation j; This indicates the completion time margin for candidate workstation j; Indicates the late arrival time of candidate workstation j; Let j be the earliest allowed service time for candidate workstation j. The deadline for completing the service for candidate workstation j; This indicates the dwell time of candidate workstation j.

[0023] when When the AGV arrives at candidate workstation j ahead of schedule, it needs to wait until the earliest permitted service time. Service begins afterward; when When, it means that candidate job j can complete the service no later than the service completion deadline. Complete the feeding service; when When, it indicates that candidate workstation j is late, and the lateness time is determined by... express.

[0024] Based on the above calculations, it can be determined whether the AGV will encounter waiting, insufficient completion time margin, or delay if it visits candidate workstation j in the next step.

[0025] To characterize the urgency of candidate workstation j, the workshop AGV scheduling controller calculates the dwell pressure of candidate workstation j based on its dwell time, completion time margin, and lateness time. : ; In the formula, A preset positive number used to prevent the denominator from being zero, and with Having the same time dimension, it can be taken as 0.06~6s; The late arrival amplification factor (dimensionless) is set according to the importance of late arrival loss relative to the urgency of the stay, and can also be calibrated through historical scheduling data. It can usually be taken as 1 to 5. The preset time scale parameter is used to normalize the lateness time. It can be the production cycle time, the historical average lateness time, the typical stay time, or other characteristic time scales. 120s is an example.

[0026] This is a dimensionless evaluation index used to characterize the current urgency of access, formed by the candidate workstation's dwell time, completion time margin, and lateness time. When the dwell time of candidate workstation j... The longer the time, the more time leeway to completion. The smaller or late time The larger the value, the greater the self-retention pressure of candidate workstation j, indicating that the workstation needs to be prioritized for access.

[0027] In S3, to determine whether selecting candidate workstation j will compress the time margin of subsequent workstations, assuming priority access to candidate workstation j, candidate workstation j is removed from the current set of unvisited workstations, resulting in the set of remaining unvisited workstations. For any candidate workstation x in this set, the service completion deadline of candidate workstation x, its workstation priority, and the node transfer distance from candidate workstation j to candidate workstation x are subjected to minimum-maximum normalization to obtain the normalized service completion deadline. Normalized chemical level priority and normalized node transition distance ; Construct a prediction ranking index for candidate workstation x based on the normalization results. : ; In the formula, , , This is a non-negative ranking weight coefficient, which can be set according to the importance of service completion deadline, workstation priority, and node transfer distance to the predicted ranking. The values ​​can be 0.5, 0.3, and 0.2 respectively; when the maximum value in the minimum-maximum normalization is equal to the minimum value, the normalization value of the corresponding indicator is set to 0.5, and the same applies below.

[0028] Workshop AGV scheduling controller according to Sort the remaining unvisited workstations in ascending order, select the top h workstations as predicted workstations, and construct the subsequent predicted access sequence for candidate workstation j. : ; In the formula, This indicates the u-th predicted workstation in the subsequent predicted access sequence; the value of h is selected based on the number of remaining unvisited workstations. For example, when there are 5 remaining unvisited workstations, h=3 can be taken; when the number of remaining unvisited workstations is less than the preset prediction number, all remaining unvisited workstations are used as predicted workstations.

[0029] In S4, in constructing the subsequent predicted access sequence Subsequently, the workshop AGV scheduling controller further predicts the current priority access candidate workstation j and then follows the... The time status of each predicted workstation is recorded when visiting the predicted workstations sequentially. Let candidate workstation j be the prediction starting point. And let the service completion time of the predicted starting point be The superscript (j) indicates that the time state was calculated under the assumption of "priority access candidate workstation j"; for the u-th predicted workstation in the subsequent predicted access sequence... Its service completion time Completion time margin Late time They are respectively: ; ; ; In the formula, This represents the (u-1)th predicted workstation in the subsequent predicted access sequence. Service completion time; Indicates the predicted workstation To the predicted work station Transfer time; Indicates the predicted workstation The earliest permitted service time; Indicates the predicted workstation Length of stay for service; Indicates the predicted workstation The deadline for service completion; Based on the prediction results, the workshop AGV scheduling controller calculates the downstream propagation pressure corresponding to candidate workstation j. : ; In the formula, Represents the propagation attenuation coefficient and The value can be determined by analyzing historical data or conducting simulation experiments based on the rate at which the impact of lateness decays along subsequent workstations; for example, it can be taken as 0.6 to 0.9. That is u-1 power; When the subsequent predicted access sequence corresponding to candidate workstation j is empty, let .

[0030] The impact of the candidate workstation's residency service on the compression of the completion time margin of subsequent workstations and the potential risk of lateness propagation is represented by the influence of the u-th predicted workstation. The influence of a prediction station is attenuated. In the subsequent prediction access sequence, the earlier the prediction station appears, the greater its influence on the current decision; the later the prediction station appears, the less significant its influence becomes. Gradually weakening.

[0031] In S5, after calculating the self-retention pressure and downstream propagation pressure of candidate workstation j, for any candidate workstation j in the current unvisited workstation set U (if the current unvisited workstation set contains only one candidate workstation, the candidate workstation is directly determined as the next visited workstation, and normalization and comprehensive access cost comparison are no longer performed), the node transfer distance from the current node i to the candidate workstation j, the downstream propagation pressure of the candidate workstation j, the self-retention pressure, and the workstation priority are subjected to minimum-maximum normalization within the current unvisited workstation set U to obtain the normalized node transfer distance. Normalized downstream transmission pressure Normalization of its own residence pressure and normalization of chemical grade priority ; The workshop AGV scheduling controller constructs the access cost of candidate workstation j based on the normalization result. Simultaneously, it characterizes the AGV's travel cost, the urgency of accessing the candidate workstation, the workstation priority, and the propagation impact of prioritizing access to the candidate workstation on subsequent workstations: ; In the formula, , , , These represent distance weight, downstream propagation pressure weight, self-residence pressure weight, and priority weight, respectively. All are non-negative weight coefficients and can be determined through experiments, expert evaluation, or manually set based on the importance of distance, downstream propagation pressure, self-residence pressure, and priority to the actual access decision. For example, in scenarios where more emphasis is placed on reducing the pressure of subsequent late arrivals and the urgency of accessing the current workstation, values ​​of 0.2, 0.35, 0.3, and 0.15 can be used respectively. The calculation formula shows that the greater the distance, the greater the access cost; the greater the downstream propagation pressure, the greater the access cost; the greater the self-retention pressure, the more the candidate workstation needs to be accessed first, so it is a factor in reducing the access cost; the higher the workstation priority, the lower the access cost.

[0032] according to Sort the candidate workstations in the currently unvisited workstation set U in ascending order, and select the candidate workstation with the lowest access cost as the next workstation to be visited: ; In the formula, Indicates the next workstation selected for this round; Will Add to the end of the current AGV feeding access sequence Q to obtain the updated access sequence. : ; At the same time, the current node Update to selected workstation and the current completion time Updated to access Service completion time : ; Delete from the collection of never visited workstations Get the updated set of unvisited workstations Subsequently, with , , , As the scheduling state for constructing the next round of material feeding access sequence, return to S2 to continue execution until all material feeding stations have been accessed, resulting in a complete AGV material feeding access sequence.

[0033] In S6, the workshop AGV scheduling controller generates AGV feeding control instructions based on the complete AGV feeding access sequence. The AGV feeding control instructions include at least the feeding station access sequence, and may further include one or more of the following: the estimated arrival time of each feeding station, the planned service start time, and the planned service completion time. The workshop AGV scheduling controller sends the AGV feeding control instructions to the AGVs, enabling the AGVs to complete the workshop feeding tasks according to the instructions.

[0034] To visually demonstrate the differences in path execution among different scheduling algorithms in specific task scenarios, a typical workshop material feeding example with 12 workstations awaiting material feeding was selected for comparative visualization. The results are as follows: Figures 2-6 As shown. Figures 2-6 In the diagram, yellow pentagrams represent distribution centers, green dots represent on-time completion stations, red dots represent late completion stations, red numbers represent the late arrival time of the corresponding station, and path lines represent the material feeding and access order of the AGVs. NN represents the nearest neighbor algorithm, EDD represents the earliest deadline first algorithm, MSM represents the minimum time margin first algorithm, NoDown represents the ablation algorithm to remove downstream propagation pressure, and DRAS represents the resident risk-aware scheduling method proposed in this embodiment.

[0035] Figure 2 This is a schematic diagram of the AGV material feeding and access sequence obtained using a Neural Network (NN). The NN primarily selects the next access station based on the distance between the current node and each candidate station. Therefore, its path is relatively continuous in space, but it lacks consideration of station dwell time and its subsequent propagation impact. The NN completed 8 stations on time, and 4 stations were late: stations 5, 6, 7, and 10, with a total delay time of 489 seconds and an on-time completion rate of 66.7%. The delay times for stations 6 and 10 were 196 seconds and 173 seconds, respectively, indicating that relying solely on distance for local greedy selection can easily consume the available execution time of subsequent stations in the early access decisions, thus causing delays in later stations.

[0036] Figure 3 This is a schematic diagram of the AGV material feeding access sequence obtained using EDD (Engineering Deployment). EDD prioritizes accessing workstations with earlier service completion deadlines, reflecting a certain degree of time urgency. However, it does not fully consider the impact of node transfer distance, workstation dwell time, and the time margin of the current access selection on subsequent workstations. Only 4 workstations completed on time using EDD, while 8 workstations were late, with a total lateness of 973 seconds, resulting in an on-time completion rate of 33.3%. Figure 3 It can be seen that there are many cross-regional transfers and path intersections in the AGV access path, indicating that simply determining the access order according to the deadline may lead to frequent back-and-forth trips between different areas for the AGV, and further compress the completion time margin of subsequent workstations.

[0037] Figure 4 This is a schematic diagram of the AGV material handling sequence obtained using MSM. MSM prioritizes workstations with smaller current completion time margins, which, compared to EDD which only considers service deadlines, can better reflect the current urgency of candidate workstations. However, MSM and EDD yielded the same results, with 8 workstations still experiencing delays, totaling 973 seconds of delay (in this example, the access order of MSM and EDD is the same, hence the same total delay time), resulting in an on-time completion rate of 33.3%. This indicates that simply sorting candidate workstations based on their current time margins is insufficient to identify the potential propagation effect of delays on subsequent workstations caused by prioritizing a particular long-staying workstation.

[0038] Figure 5 This is a schematic diagram of the AGV material feeding access sequence obtained using NoDown. NoDown removes the downstream propagation pressure term during the access cost construction process, considering only the candidate station's own state and other local factors. Only 3 stations completed on time under NoDown, while 9 stations were late, with a total lateness of 1003 seconds, resulting in an on-time completion rate of only 25.0%. Specifically, stations 8 and 7 had lateness times of 339 seconds and 278 seconds, respectively, indicating that without downstream propagation pressure evaluation, the scheduling method struggles to prevent local dwell time from continuously occupying subsequent tasks, leading to the further propagation of lateness risk along the access sequence.

[0039] Figure 6 This is a schematic diagram of the AGV material feeding access sequence obtained using DRAS. When selecting the next access station, DRAS considers node transfer distance, the candidate station's own dwell pressure, downstream propagation pressure, and station priority simultaneously. This allows it to meet the current urgent station access needs while also assessing the impact of the access selection on the completion time margin of subsequent stations. DRAS achieved on-time completion for 11 stations, with only station 12 experiencing a 58-second delay, for a total delay time of 58 seconds, resulting in an on-time completion rate of 91.7%. Compared to NN, EDD, MSM, and NoDown, DRAS significantly reduced the number of delayed stations and the total delay time.

[0040] Depend on Figures 2-6 As can be seen, compared with NN, EDD, MSM, and NoDown, DRAS can reduce the total lateness time from 489s, 973s, 973s, and 1003s to 58s, respectively, and increase the on-time completion rate to 91.7%. These results indicate that the method in this embodiment, by jointly sensing the self-sustaining pressure and downstream propagation pressure of candidate workstations during the path construction stage, can effectively reduce the risk of subsequent lateness propagation caused by long-sustaining workstations, reduce continuous lateness and production cycle fluctuations, and improve the timeliness and stability of AGV feeding path execution.

[0041] To further avoid the influence of randomness from a single case, 50 sets of workshop material feeding tasks were randomly generated for statistical verification. Each task set contained 20 workstations to be fed, and the location coordinates, service time window, dwell time, and workstation priority of each workstation were randomly generated within a preset range. Path scheduling was performed using NN, EDD, MSM, NoDown, and DRAS, and the average total travel distance, average total lateness time, average number of late workstations, and average on-time completion rate of each algorithm were statistically analyzed across the 50 sets of cases. The experimental results are shown in Table 1.

[0042] Table 1. Comparison of average performance of different scheduling algorithms under multiple randomized test cases.

[0043] As shown in Table 1, among the 50 random workshop material feeding tasks, DRAS performed better in terms of average total lateness time, average number of late workstations, and average on-time completion rate. Compared to NN, DRAS reduced the average total lateness time from 962.2 to 305.6, the average number of late workstations from 5.6 to 2.9, and the average on-time completion rate from 72.2% to 85.6%. Although DRAS has a higher average total travel distance than NN, it significantly reduces the risk of lateness and is more suitable for rhythmic workshop material feeding scenarios with high timeliness requirements.

[0044] Furthermore, compared to NoDown, DRAS reduced the average total late time from 720.4 to 305.6, the average number of late stations from 9.1 to 2.9, and the average on-time completion rate from 54.6% to 85.6%. These results demonstrate that downstream propagation pressure assessment can effectively reflect the compressing effect of the current candidate station on the time margin of subsequent stations, thereby reducing the risk of subsequent lateness propagation caused by long-stayed stations and improving the timeliness and stability of material feeding path scheduling.

[0045] Example 2: A rhythmic workshop AGV feeding path scheduling device with residence risk perception, comprising: One or more processors; Memory, used to store one or more computer programs; When one or more programs are executed by one or more processors, the one or more processors execute the method in Embodiment 1.

[0046] Example 3: A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method in Example 1.

[0047] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can make equivalent substitutions or modifications based on the technical solution and concept of the present invention within the scope of the technology disclosed in the present invention, and such modifications should also be considered to fall within the scope of protection of the present invention.

Claims

1. A method for scheduling the material feeding path of AGVs in a rhythmic workshop based on residence risk perception, characterized in that, Includes the following steps: S1. Obtain the material feeding task data of the current workshop, abstract the distribution center and each material feeding station as path scheduling nodes, construct the material feeding task input data set, and initialize the set of unvisited stations, the current node, the current completion time and the AGV material feeding access sequence. S2. Based on the input data set of the material feeding task, calculate the time status of each candidate workstation in the set of unvisited workstations, including the estimated arrival time, service start time, service completion time, completion time margin, and late time, and calculate the self-retention pressure of each candidate workstation. S3. Assuming that a certain candidate workstation is visited first, remove the candidate workstation from the current set of unvisited workstations, and select several predicted workstations from the remaining unvisited workstations to construct the subsequent predicted visit sequence corresponding to the candidate workstation. S4. Taking the service completion time of the candidate workstation as the prediction starting point, according to the subsequent predicted access sequence corresponding to the candidate workstation, calculate the service completion time, completion time margin and late time of each predicted workstation under the condition of prioritizing access to the candidate workstation. Based on the completion time margin and late time of each predicted workstation, calculate the downstream propagation pressure generated by prioritizing access to the current candidate workstation. S5. Combining node transfer distance, downstream propagation pressure, self-residence pressure, and workstation priority, construct a comprehensive access cost for each candidate workstation. Select the candidate workstation with the lowest access cost from the current unvisited workstation set as the next access workstation, add it to the AGV feeding access sequence, and update the current node, current completion time, and unvisited workstation set. If the updated unvisited workstation set is not empty, return to S2 with the updated scheduling state to continue execution until all workstations to be fed have been accessed, resulting in a complete AGV feeding access sequence. S6. Generate AGV feeding control instructions based on the AGV feeding access sequence and send them to the AGV to execute the workshop feeding task.

2. The method for scheduling the feeding path of a cycle-based AGV in a workshop based on residence risk perception as described in claim 1, characterized in that, In S1, the set of input data for the feeding task Represented as: ； In the formula, This represents a set of nodes. Node 0 corresponds to the distribution center, and n is the number of workstations waiting to be fed. Nodes 1 to n each correspond to n workstations waiting to be fed. The currently unvisited workstations waiting to be fed constitute the candidate workstation set. Represents the node transition distance matrix. Represents the node transition time matrix. , These represent the transition distance and transition time from node i to node j, respectively. This represents the service time window of node k. Let k be the earliest allowed service time. The service completion deadline for node k; This indicates the dwell time of node k; This indicates the workstation priority of node k; Let the set of unvisited workstations be... Initialize the AGV feeding access sequence Q as an empty sequence. Let the current node be i and the current completion time be . Initially, i=0, and the initial completion time is... .

3. The method for scheduling the feeding path of a cycle-based AGV in a workshop based on residence risk perception, as described in claim 2, is characterized in that... In S2, for any candidate workstation j in the set of unvisited workstations U, the time state of candidate workstation j is calculated: ； In the formula, This indicates the estimated time for the AGV to arrive at candidate workstation j; Indicates the service start time of candidate workstation j; Indicates the service completion time of candidate workstation j; This indicates the completion time margin for candidate workstation j; Indicates the late arrival time of candidate workstation j; Let j be the earliest allowed service time for candidate workstation j. The deadline for completing the service for candidate workstation j; This indicates the dwell time of candidate workstation j.

4. The method for scheduling the feeding path of a cycle-based AGV in a workshop based on residence risk perception as described in claim 3, characterized in that, when When the AGV arrives at candidate workstation j ahead of schedule, it needs to wait until the earliest permitted service time. Service begins afterward; when When, it means that candidate job j can complete the service no later than the service completion deadline. Complete the feeding service; when When, it indicates that candidate workstation j is late, and the lateness time is determined by... express.

5. The method for scheduling the feeding path of a rhythmic workshop AGV based on residence risk perception as described in claim 3, characterized in that, In S2, for candidate workstation j, its own dwell pressure Represented as: ； In the formula, A preset positive number used to prevent the denominator from being zero, and with They have the same time dimension; This is the late amplification factor; This is the preset time scale parameter.

6. The method for scheduling the feeding path of a cycle-based AGV in a workshop based on residence risk perception, as described in claim 5, is characterized in that... In S3, assuming priority is given to accessing candidate workstation j, candidate workstation j is removed from the current set of unaccessed workstations, resulting in a set of remaining unaccessed workstations. For any candidate workstation x in this set, the service completion deadline, workstation priority, and node transfer distance from candidate workstation j to candidate workstation x are minimized and minimized within this set to obtain the normalized service completion deadline. Normalized chemical level priority and normalized node transition distance ; Construct a prediction ranking index for candidate workstation x based on the normalization results. : ； In the formula, , , These are non-negative ranking weight coefficients; according to Sort the remaining unvisited workstations in ascending order, select the top h workstations as predicted workstations, and construct the subsequent predicted access sequence for candidate workstation j. : ； In the formula, This represents the u-th predicted workstation in the subsequent predicted access sequence.

7. The method for scheduling the feeding path of a cycle-based AGV in a workshop based on residence risk perception as described in claim 6, characterized in that, In S4, candidate workstation j is taken as the prediction starting point. And let the service completion time of the predicted starting point be For the u-th predicted workstation in the subsequent predicted access sequence Its service completion time Completion time margin Late time They are respectively: ； ； ； In the formula, This represents the (u-1)th predicted workstation in the subsequent predicted access sequence. Service completion time; Indicates the predicted workstation To the predicted work station Transfer time; Indicates the predicted workstation The earliest permitted service time; Indicates the predicted workstation The length of stay for service; Indicates the predicted workstation The deadline for service completion; Downstream propagation pressure corresponding to candidate station j Represented as: ； In the formula, Represents the propagation attenuation coefficient and ; When the subsequent predicted access sequence corresponding to candidate workstation j is empty, let .

8. The method for scheduling the feeding path of a cycle-based AGV in a workshop based on residence risk perception, as described in claim 7, is characterized in that... In step S5, after calculating the self-retention pressure and downstream propagation pressure of candidate workstation j, for any candidate workstation j in the current unvisited workstation set U, the node transfer distance from the current node i to candidate workstation j, the downstream propagation pressure of candidate workstation j, the self-retention pressure, and the workstation priority are performed using minimum-maximum normalization within the current unvisited workstation set U to obtain the normalized node transfer distance. Normalized downstream transmission pressure Normalization of its own residence pressure and normalization of chemical grade priority ; Construct the access cost of candidate workstation j based on the normalization results. : ； In the formula, , , , These represent distance weight, downstream propagation pressure weight, self-retention pressure weight, and priority weight, respectively. according to Sort the candidate workstations in the currently unvisited workstation set U in ascending order, and select the candidate workstation with the lowest access cost as the next workstation to be visited: ； In the formula, Indicates the next workstation selected for this round; Will Add to the current AGV feeding access sequence and update the current node. Update the current completion time to the access time. The service completion time is then removed from the set U of currently unvisited workstations. The updated set of unvisited workstations is obtained. If it is not empty, the updated scheduling state is returned to S2 to continue execution until all workstations to be fed have been visited, and a complete AGV feeding access sequence is obtained.

Images