A Sheet Metal Parts Supply Chain Scheduling Method and System Based on Multi-Source Data

Abstract

This invention relates to the field of intelligent manufacturing and supply chain scheduling technology, specifically to a sheet metal parts supply chain scheduling method and system based on multi-source data. The method includes: acquiring underlying equipment wear status data, logistics transient fluctuation data, and raw material batch deviation data for each supply chain node in the sheet metal parts supply chain network, and fusing them to construct a node status dataset; calculating a node status deviation index by weighted summation of various status data, and comparing it with a preset threshold to trigger local task transfer; extracting the current manufacturing task of the stressed node, obtaining the available capacity of adjacent nodes, and generating candidate reconfiguration schemes containing alternative paths and material redirection information; calculating the cascade disturbance damping coefficient and scheduling plan fluctuation index, screening target reconfiguration schemes, and issuing pre-execution transfers and self-balancing, thereby achieving resilient flow before local risk spreads and reducing final delivery delays. This invention solves the problem that existing scheduling methods struggle to effectively fuse and identify node states.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and supply chain scheduling technology, specifically to a method and system for scheduling sheet metal parts supply chains based on multi-source data. Background Technology

[0002] Currently, the scheduling mechanism of the sheet metal parts supply chain is usually based on fixed plans, node capacity statistics, or single business data. When minor disturbances such as equipment wear, logistics fluctuations, or raw material batch deviations gradually accumulate at the nodes of the supply chain, the existing methods are difficult to effectively integrate and identify the node status and trigger local scheduling intervention in advance. When a node experiences a processing anomaly or delivery delay, existing scheduling methods are usually unable to accurately relocate the affected tasks. Each adjustment may be accompanied by scheduling changes and material redirection of a large number of unrelated nodes, reducing the response efficiency and overall stability of supply chain scheduling. Summary of the Invention

[0003] The purpose of this invention is to provide a sheet metal parts supply chain scheduling method and system based on multi-source data. This avoids scheduling changes and material redirection of numerous unrelated nodes when actual processing anomalies or delivery delays occur at nodes. Furthermore, it can effectively fuse and identify node states and trigger local scheduling interventions in advance, achieving precise pre-emptive transfer of affected tasks. This improves the response efficiency and overall stability of supply chain scheduling. Specifically, the technical solution of this invention is as follows: The sheet metal parts supply chain scheduling method based on multi-source data includes the following steps: Step S1: Obtain multi-source heterogeneous perturbation data of each supply chain node in the sheet metal parts supply chain network and perform fusion processing to construct a node status dataset; wherein, the multi-source heterogeneous perturbation data includes the wear status data of the underlying equipment, the transient fluctuation data of logistics, and the batch deviation data of raw materials; Step S2: Based on the node status dataset, determine the node status deviation index of each supply chain node; wherein, the node status deviation index is used to quantify the degree to which the supply chain node deviates from the ideal flow state. Step S3: If the node state deviation index is greater than the preset stress pressure threshold, a local task transfer instruction is generated; if the node state deviation index is not greater than the stress pressure threshold, the original scheduling plan is maintained. Step S4: In response to the local task transfer instruction, extract the current manufacturing tasks of supply chain nodes whose node state deviation index is greater than the stress pressure threshold, and obtain the available capacity of adjacent nodes in the sheet metal supply chain network as cross-node resources to generate candidate reconfiguration schemes; evaluate the cascade disturbance damping coefficient based on the ratio of the difference in the final delivery node delay time before and after the implementation of the candidate reconfiguration scheme to the expected delay time before implementation, and determine the scheduling plan fluctuation index based on the task change rate and time offset of the affected nodes after the implementation of the candidate reconfiguration scheme; Step S5: Based on the cascaded disturbance damping coefficient and the scheduling plan fluctuation index, select the target reconfiguration scheme from the candidate reconfiguration schemes and issue it for execution.

[0004] Preferably, the fusion processing of multi-source heterogeneous perturbation data in step S1 specifically includes: A fusion process of time series alignment and semantic feature extraction is performed on multi-source heterogeneous perturbation data to output a node state dataset.

[0005] Preferably, obtaining cross-node resources in step S4 specifically includes: Match the available capacity of adjacent nodes with compatible sheet metal process parameters as cross-node resources.

[0006] Preferably, step S5 specifically includes: If the cascade disturbance damping coefficient of the candidate reconfiguration scheme is greater than the preset damping coefficient threshold, and the scheduling plan fluctuation index is less than the preset fluctuation index threshold, then the corresponding candidate reconfiguration scheme will be marked as the target reconfiguration scheme. If the above conditions are not met, the matching range of cross-node resources is updated, and the step of generating candidate reconstruction schemes is re-executed; if the number of times the matching range is updated reaches the preset number of rounds and there are still no candidate reconstruction schemes that meet the conditions, the transfer of new tasks is frozen and rate limiting protection is implemented on the corresponding supply chain nodes. The time difference between data fusion and the issuance of the target reconstruction plan will be persistently stored as the cross-node reconstruction response latency.

[0007] The sheet metal parts supply chain scheduling system based on multi-source data includes the following modules: The data acquisition and fusion module is used to acquire and fuse multi-source heterogeneous perturbation data from various supply chain nodes in the sheet metal parts supply chain network to construct a node status dataset. The multi-source heterogeneous perturbation data includes wear status data of underlying equipment, transient fluctuation data of logistics, and batch deviation data of raw materials. The stress determination module is used to determine the node state deviation index of each supply chain node based on the node state dataset; the node state deviation index is used to quantify the degree to which the supply chain node deviates from the ideal flow state. The pre-reconstruction trigger module is used to generate a local task transfer instruction when the node state deviation index is greater than a preset stress pressure threshold. The reconfiguration scheme evaluation module is used to respond to local task transfer instructions, obtain the available capacity of adjacent nodes in the sheet metal supply chain network as cross-node resources, generate candidate reconfiguration schemes, and determine the cascade disturbance damping coefficient and scheduling plan fluctuation index corresponding to the candidate reconfiguration schemes. The self-balancing execution module is used to select a target reconfiguration scheme from candidate reconfiguration schemes and issue it for execution based on the cascaded disturbance damping coefficient and the scheduling plan fluctuation index.

[0008] Preferably, the data acquisition and fusion module is specifically used for: A fusion process of time series alignment and semantic feature extraction is performed on multi-source heterogeneous perturbation data to output a node state dataset.

[0009] Preferably, when acquiring cross-node resources, the reconstruction scheme evaluation module is specifically used for: Match the available capacity of adjacent nodes with compatible sheet metal process parameters as cross-node resources.

[0010] Preferably, the self-balancing execution module is specifically used for: If the cascade disturbance damping coefficient of the candidate reconfiguration scheme is greater than the preset damping coefficient threshold, and the scheduling plan fluctuation index is less than the preset fluctuation index threshold, then the corresponding candidate reconfiguration scheme will be marked as the target reconfiguration scheme. If the above conditions are not met, the reconstruction scheme evaluation module will be triggered to update the matching range of cross-node resources and regenerate candidate reconstruction schemes. The time difference between data fusion and the issuance of the target reconstruction plan will be persistently stored as the cross-node reconstruction response latency.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention fuses multi-source heterogeneous perturbation data to construct a node status dataset, enabling each supply chain node to be described not only by a single capacity value, but also by a combination of status slices reflecting health, material arrival stability, and process adaptability. This improves the ability to characterize the actual operating status of nodes and solves the problem that existing scheduling methods are unable to effectively fuse and identify node status. 2. This invention incorporates both time availability and process compatibility into the resource selection logic, which can avoid transferring tasks to nodes that do not have the corresponding plate thickness capability, mold conditions, surface protection conditions, or reference connection conditions simply because the node is idle. This reduces problems such as uncontrolled bending angles, surface damage, assembly deviations, and rework caused by blind transfer, thereby unifying delay control and quality stability. 3. This invention uses the ratio of the difference in the final delivery node delay time before and after the implementation of the candidate reconfiguration scheme to the delay time before implementation as the cascade disturbance damping coefficient. It also generates a scheduling plan fluctuation index by combining the task change rate and time offset of nodes that are not directly affected. Based on the damping coefficient threshold and the fluctuation index threshold, the target reconfiguration scheme is screened. This invention can select a scheme from multiple candidate schemes that can effectively absorb local disturbances without causing frequent oscillations of a large number of stable nodes. This solves the problem that once the existing adjustment method is changed, it is accompanied by a large number of irrelevant node scheduling changes and a decrease in overall stability. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application and the prior art, the accompanying drawings used in the description of the embodiments and the prior art will be briefly introduced below: Figure 1 A flowchart illustrating the sheet metal parts supply chain scheduling method based on multi-source data provided in this application embodiment; Figure 2 A schematic diagram of a sheet metal parts supply chain scheduling system based on multi-source data provided in this application embodiment. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] The sheet metal parts supply chain scheduling method based on multi-source data includes the following steps: Step S1: Obtain multi-source heterogeneous perturbation data of each supply chain node in the sheet metal parts supply chain network and perform fusion processing to construct a node status dataset; wherein, the multi-source heterogeneous perturbation data includes the wear status data of the underlying equipment, the transient fluctuation data of logistics, and the batch deviation data of raw materials; Step S2: Based on the node status dataset, determine the node status deviation index of each supply chain node; wherein, the node status deviation index is used to quantify the degree to which the supply chain node deviates from the ideal flow state. Step S3: If the node state deviation index is greater than the preset stress pressure threshold, a local task transfer instruction is generated; if the node state deviation index is not greater than the stress pressure threshold, the original scheduling plan is maintained. Step S4: In response to the local task transfer instruction, extract the current manufacturing tasks of supply chain nodes whose node state deviation index is greater than the stress pressure threshold, and obtain the available capacity of adjacent nodes in the sheet metal supply chain network as cross-node resources to generate candidate reconfiguration schemes; evaluate the cascade disturbance damping coefficient based on the ratio of the difference in the final delivery node delay time before and after the implementation of the candidate reconfiguration scheme to the expected delay time before implementation, and determine the scheduling plan fluctuation index based on the task change rate and time offset of the affected nodes after the implementation of the candidate reconfiguration scheme; Step S5: Based on the cascaded disturbance damping coefficient and the scheduling plan fluctuation index, select the target reconfiguration scheme from the candidate reconfiguration schemes and issue it for execution; This embodiment provides a sheet metal parts supply chain scheduling mechanism based on multi-source data, such as... Figure 1 As shown; specifically, the explanation revolves around the same industrial main line scenario: an order for a new energy vehicle charging cabinet shell needs to be delivered within 48 hours. The supply chain network includes steel coil leveling node, laser cutting node, CNC bending node, welding node, spraying node and final delivery node in sequence. Each node is distributed in two parks and one cooperative factory area. These types of housing parts are characterized by dense holes, numerous bends, and high requirements for paint appearance. Therefore, if there is a slight disturbance in the upstream process, it is very easy to cause a chain reaction of material accumulation, material shortage, or rework in the downstream. The core of this implementation method is not to reorder the whole thing after the failure has completely occurred, but to identify the supply chain nodes in a state of disturbance accumulation in advance before the equipment stops, the logistics chain is broken, and the materials become unstable in large quantities, and replace the global reordering with local forward transfer. The specific processing procedure is as follows: In step S1, the system continuously acquires multi-source heterogeneous perturbation data from each node and aggregates it into a node status dataset; the multi-source heterogeneity here includes at least three types of information: the first is the wear status data on the equipment side, such as the power attenuation of the laser cutting head, the frequency of springback compensation of the bending machine mold, the current fluctuation of the welding machine, the circulation filter resistance of the spray booth, etc.; the second is the transient fluctuation data on the logistics side, such as the arrival deviation of the transfer vehicles between parks, the occupancy time of the loading and unloading platform, temporary traffic control, changes in the forklift waiting queue, etc. Thirdly, there is the batch deviation data of raw materials, such as plate thickness fluctuation, yield strength dispersion, surface coating consistency, and springback differences in the same batch of materials. In industry, the above information corresponds to three types of real constraints: whether the equipment still has continuous and stable processing capabilities, whether the task materials can be delivered within the planning window, and whether the materials to be processed still meet the predetermined process window. The system aggregates these heterogeneous data according to nodes, time slices, and task batches to form a node status dataset, so that each node is no longer described by a single capacity value, but by a set of status slices that reflect its health, material arrival stability, and process adaptability. In step S2, the system calculates the node state deviation index based on the node state dataset; the weighting here is not simply an abstract numerical superposition, but reflects the difference in the impact of different disturbance sources on the continuity of actual production. Before performing weighted summation, the system performs benchmark mapping on heterogeneous state items such as wear status data, transient fluctuation data of logistics, and batch deviation data of raw materials. That is, the maximum and minimum values ​​are normalized, and their original dimensions are uniformly mapped to the same dimensionless interval. This avoids the calculation distortion caused by the direct superposition of equipment temperature drift in degrees Celsius, transfer deviation in minutes, and yield strength in megapascals due to different physical dimensions. For example, for laser cutting nodes, the destructive effect of cutting head thermal drift and protective lens contamination on continuous processing is often greater than that of short-term logistics jitter. For the spraying node, the state of the pretreatment liquid tank and the drying tunnel cycle time have a more direct impact on the appearance consistency; therefore, the system pre-combines the process experience library, historical delay events and node roles to configure corresponding weights for each normalized state item, forming a comprehensive deviation representation for each node. The specific quantitative extrapolation rules are as follows: the system assigns corresponding weight values ​​to the baseline equipment wear status data, logistics transient fluctuation data, and raw material batch deviation data after benchmark mapping, for example, respectively. , and ,and Then, the benchmark data and their corresponding weight values ​​are multiplied and summed to finally calculate the node state deviation index. This can be understood as the node state deviation index reflecting the degree of deviation between the node and the ideal stable flow state: the greater the deviation, the closer the node is to the critical state of transitioning from perturbation to substantial instability. To facilitate understanding, a simplified quantitative model can be used to illustrate the data flow. Assume that at a certain scheduling moment, the bending node corresponds to three types of state segments, denoted as D1, D2, and D3, where D1 represents the equipment wear-related state, D2 represents the material arrival-related state, and D3 represents the raw material batch difference-related state. If, after preprocessing, D1 shows accelerated mold wear, D2 shows a delay in upstream material arrival deviating from a preset time threshold, and D3 shows that the springback of the current batch of sheet metal exceeds the preset tolerance, then the system can mark the node as processable but with insufficient stable margin. The node state deviation index generated at this time is not for displaying an arithmetic process, but to sum up the risks in the three directions according to the weight combination set above, and merge them into an industrial state quantity that can trigger a scheduling action. In step S3, the system compares the node state deviation index with the preset stress threshold. When the index is higher than the threshold, it indicates that although the node may not have completely stopped, it has shown obvious signs of congestion or instability. At this time, the pre-reconstruction mechanism is triggered and a local task transfer instruction is generated. When the index is not higher than the threshold, it means that the current disturbance is still within the absorption range of the supply chain itself, and the original scheduling plan can be maintained. Based on the strong sequential dependence of the sheet metal manufacturing chain, if the adjustment is made after the bending machine is actually stopped, the downstream welding and painting will often be empty, while the upstream cutting may continue to produce semi-finished products that cannot be digested in time, ultimately causing a larger range of plan fluctuations. In step S4, the system responds to the local task transfer instruction, extracts the current manufacturing task of the supply chain node whose node state deviation index is greater than the state deviation threshold, and then retrieves the available capacity of the adjacent nodes from the network as cross-node resources; the adjacent nodes here can be parallel nodes of the same process, or they can be close-neighbor collaborative nodes that can undertake some of the tasks connecting the preceding and following processes. The system generates candidate reconfiguration schemes around these cross-node resources. Each scheme includes at least an alternative path and material redirection information. An alternative path means that a batch of tasks no longer flows through the stressed node along the original route, but is instead processed by other nodes that can take over before returning to the subsequent processes. Material redirection information means that sheet metal, semi-finished products, turnover tools, and electronic batch cards that flow with the process are reassigned together to avoid the disconnect between physical material flow and data flow. During the candidate solution evaluation phase, the system focuses on two dimensions: the first is the cascaded disturbance damping coefficient, which is used to determine whether the solution can truly absorb local risks locally, rather than amplifying the delay to the final delivery node. When calculating the delay attenuation ratio, to avoid calculation overflow errors caused by perturbations triggering reconfiguration in the ideal operating condition where there was no expected delay in the original plan (i.e., the expected delay time before implementation was 0), the system adds a preset small positive real constant as a smoothing term to the denominator, specifically the delay time before implementation. For example, denoted as... ,and ; If, after the implementation of a certain solution, the expected delay reduction at the final delivery node reaches the preset improvement threshold, it indicates that the solution has the delay reduction capability to meet the preset evaluation conditions; the second is the scheduling plan fluctuation index, used to determine whether the solution, in order to alleviate the processing pressure of a single node, causes a large number of originally stable nodes to change their plans; this index is not a fuzzy judgment, but has structured calculation rules: The system uses the ratio of the number of nodes whose planned start times have changed after the implementation of candidate reconfiguration schemes to the total number of nodes in the network that were not directly affected by physical damage as the task change rate, and the average absolute time difference between the original planned time and the new scheduled start time of the affected nodes as the time offset. The task change rate and time offset are normalized to eliminate dimensional differences, and then the scheduling plan fluctuation index is calculated by summing the two values ​​using empirical weighting coefficients. For example, they are denoted as... and ,and ; Frequent plan changes should be avoided in industrial settings because welding fixtures, painting hangers, logistics vehicles, and personnel shifts are all organized around a set rhythm. If a local anomaly causes repeated changes to the whole system, it will lead to chaos in on-site execution. Therefore, the system should not only consider whether the plan can effectively reduce delays, but also the degree of negative impact of the plan on the stability of the overall scheduling. In one specific embodiment, if the current bending node B is in a high deviation state, the original path of the batch T1 being processed is cutting A → bending B → welding C → spraying D → delivery E; the system search finds that there is spare capacity in parallel bending node B2 and bending node B3 in the cooperative factory area, so schemes P1 and P2 are formed; scheme P1 is A → B2 → C → D → E, and the materials only need to be transferred within the park. Option P2 is A→B3→C→D→E, but requires cross-park transfer; if option P1 reduces the final delivery delay more significantly and only affects a small number of pending tasks, then its cascading disturbance damping effect is better and the planning disturbance is smaller; if option P2 can take over the capacity, but introduces a longer logistics chain and more task rearrangements, then its volatility index is higher. In step S5, the system selects a target reconstruction scheme based on the cascaded disturbance damping coefficient and the scheduling plan fluctuation index, and sends the scheme to the relevant nodes to perform the pre-transfer and self-balancing of the current manufacturing task; Self-balancing refers to the process of making localized adjustments between several elements, such as tasks, materials, tooling, transportation, and time windows, without completely restarting the entire plant's scheduling, so that the network can return to a relatively stable operating trajectory. After the target plan is issued, the original stressed nodes can reduce their load, adjacent nodes can fill in the gaps, downstream nodes can maintain continuous and stable material input, and the final delivery node can avoid sudden and significant delays. Furthermore, if the node status deviation index of a certain supply chain node is greater than the status deviation threshold, but its current task is subject to highly specialized tooling constraints and there are no nodes that can take over in the short term, the system will not forcibly transfer the task. Instead, it will mark the node as a node to be locally protected and take load reduction measures, such as freezing the upstream to continue feeding it, slowing down the previous cutting cycle, prioritizing the arrival of key spare parts and materials, and generating a minimum cycle strategy for downstream nodes. If there are missing parts in the multi-source data, such as the logistics platform not transmitting the vehicle location for a short time, the system can temporarily increase the judgment weight of the equipment side and material side status, and add a data credibility insufficient mark to the node. The task transfer will only be executed when the credibility meets the minimum requirement, so as to avoid triggering incorrect scheduling based on distorted input. In the above-mentioned new energy vehicle charging cabinet housing order scenario, the system detected an abnormal increase in the frequency of mold compensation actions at the main bending node during the night shift. At the same time, the springback deviation of this batch of cold-rolled steel sheets was higher than the threshold of the benchmark batch, and the arrival time of work-in-process from the cutting workshop was delayed and deviated from the tolerance range. A single change is not enough to determine the fault, but the combination of the three indicates that the bending node is in a state of disturbance accumulation. Therefore, before the equipment stops, the system transfers the task of two pallet door panel side panels to the backup bending node in the same park, and simultaneously redirects the corresponding electronic process card and turnover rack; in this way, the downstream welding station can continue to receive materials in the original shift, and the spraying and hanging sequence does not need to be completely overturned. The purpose of this step is to transform discrete and inconsistent perturbation signals into executable state quantities for supply chain scheduling, and to complete the smooth transfer of tasks before local risks spread into actual line stoppages, thereby achieving the resilient flow and local self-balancing of the sheet metal supply chain under continuous dynamic disturbances. Furthermore, step S1 involves fusing the multi-source heterogeneous perturbation data, specifically including: The time series alignment and semantic feature extraction of multi-source heterogeneous perturbation data are fused to output a node state dataset. This embodiment provides a fusion mechanism for multi-source heterogeneous perturbation data; specifically, in the aforementioned continuous scenario, simply putting equipment, logistics and material data in the same database cannot support effective scheduling, because the update cycle, expression method and engineering meaning of these data are completely different; Equipment signals may be reported on a second-by-second basis, logistics status may be refreshed on a minute-by-minute basis, and material batch information is often only updated during incoming inspection or material requisition. Without time series alignment and semantic feature extraction, the system may easily mistake the status at different time levels for the real status at the same moment, thus causing incorrect judgments. Specifically, time series alignment addresses the question of whether different data sources at the same node are showing the same production stage; the system uses manufacturing task batches and process windows as common references, rather than simply using the acquisition clock as the only coordinate. For example, during the 08:00-08:30 window of processing batch T1 at the bending node, the equipment side can collect continuous vibration and compensation action records, the logistics side can collect the receipt time of the semi-finished products of this batch, and the material side can collect the springback level mark of the steel plates of this batch; the system uniformly merges these data from different cycles into the status slot of bending node-batch T1-current window, thereby avoiding mismatching the status of the previous batch of materials with the status of the current batch of equipment; Semantic feature extraction addresses the problem of how to transform raw signals into a scheduling-understandable state. Equipment vibration curves, temperature rise waveforms, logistics trajectory points, and inspection text records are not directly equivalent to scheduling meanings. The system needs to extract them into semantic tags with engineering orientation. For example, the continuously increasing temperature rise of the cutting head and the power compensation action can be extracted as a decrease in processing energy efficiency. Loading is completed but the park gate is congested, which is a sign of increased uncertainty in short-term transfer; yield dispersion of the same batch of plates is a sign of decreased forming consistency. These semantic labels are more interpretable in the subsequent calculation of the node state deviation index and are also easier to connect with the process rule base. As an example, a simplified scenario model is used to illustrate the alignment process: Let the equipment source data sequences be Q1, Q2, and Q3, representing the bending compensation records at 08:05, 08:10, and 08:15, respectively; let the logistics source data sequences be L1 and L2, representing the material arrival status at 08:00 and 08:20, respectively; and let the material source data M1 represent the springback grouping results of this batch before it enters the production line. The system does not simply concatenate Q1, Q2, Q3, L1, L2, and M1 in chronological order, but instead maps them to the same state slot N corresponding to T1 and the bending window. Within this state slot, three semantic segments are extracted: high equipment load, tight material connection, and material springback deviating from the baseline, which are used as a node snapshot in the node state dataset. As a fault tolerance mechanism, if a data source is missing or delayed, such as the logistics system failing to send back vehicle arrival information in a timely manner, the system can first use the most recent reliable logistics status and the current task cycle to construct a temporary placeholder feature and mark the feature as pending confirmation; if subsequent supplementary data arrives, the corresponding status slot will be filled back. If the missing data is not filled before the scheduling decision, the system will reduce the weight of this feature when the node state deviates from the exponential in subsequent calculations to prevent individual missing data from amplifying the misjudgment of the node state deviating from the exponential. If there is a drift in the timestamps of different data sources, such as the device gateway time being two minutes faster than the scheduling master station, the system will first perform clock correction and then perform window merging to avoid forming pseudo-delay. In the same charging cabinet housing order chain, during the bending window of the night shift from 08:00 to 08:30, the equipment side continuously reported an increase in mold compensation, the logistics side showed that the arrival of the previous cutting semi-finished products was later than planned, and the material side showed that the current roll material came from a higher strength batch; after time alignment, these three types of information were confirmed as a common status within the same batch and the same process window; After semantic extraction, the system does not retain the original noise curve as a direct criterion, but instead forms a node description with reduced processing stability margin; subsequent modules no longer need to interpret the original code and can directly enter the stage of calculating the node state deviation index based on the node state dataset. The purpose of this step is to transform the original information with different beats, expressions, and physical meanings into a consistent state description for scheduling scenarios, thereby making the node state dataset comparable, interpretable, and directly callable. Furthermore, step S4, obtaining cross-node resources, specifically includes: Match the available capacity of adjacent nodes with compatible sheet metal process parameters as cross-node resources; This embodiment provides a cross-node resource matching mechanism; specifically, in the aforementioned scheme, if only the availability of idle equipment is used as the basis for task transfer, although capacity can be shifted in form, it often leads to new quality problems in the sheet metal manufacturing site. The reason is that even if adjacent nodes belong to the same process, their mold specifications, tonnage range, bending compensation model, fixture reference, machinable plate thickness and surface protection conditions may be different. If the compatibility of these process parameters is ignored and the task is transferred directly, it is easy to cause consequences such as uncontrolled bending angle, out-of-tolerance indentation, increased weld assembly gap or increased surface defects before spraying. Therefore, this embodiment further introduces process parameter compatibility conditions when matching production capacity. Specifically, when the system searches for available capacity of adjacent nodes, it not only checks whether the node is idle within the time window, but also performs a compatibility comparison of its process parameters; the compatible parameters may include at least: equipment capacity parameters, such as maximum plate thickness, maximum bending length, available die opening, and pressure range; Process accuracy parameters, such as angle control capability and whether the springback compensation model is compatible with the current material grade; surface protection parameters, such as whether it supports anti-pressure damage treatment for coated plates or appearance surfaces; connection parameters, such as whether it has a reference system consistent with subsequent welding fixtures; only when adjacent nodes simultaneously meet the two conditions of time availability and process compatibility will they be included in the cross-node resource pool. This supplement is necessary because the previous solution has obvious defects under extreme conditions: if task transfer is only based on delay compression, the system may assign the high-appearance-requirement door panel task to the bending node, which has available space but no soft pressure protection tooling, and eventually expose pressure marks after painting, which leads to rework and creates new cascading disturbances; therefore, this embodiment uses process compatibility as a prerequisite for resource matching to avoid turning delay into quality loss from the source. For ease of understanding, a simplified scenario model can be used. Assume that the main bending node B is processing batch T1, and the candidate adjacent nodes are B2 and B3. B2 has the ability to bend 2.0 mm sheet metal, uses the same series of molds as the main node, and supports surface coating protection. Although B3 is also available, it is only suitable for 1.5 mm ordinary internal parts and has no current fixture reference. In this case, the system can mark B2 as compatible and transferable in the resource table, and mark B3 as available in time but incompatible in process. When generating candidate refactoring schemes, the system will only focus on B2, or only use B3 for some low-requirement sub-parts, without migrating the entire task package to B3. As a fault-tolerance mechanism, if all adjacent nodes have some incompatibility, the system can execute a hierarchical transfer strategy, that is, only transfer the subtasks with lower process requirements and less sensitivity to mold and surface protection, while leaving the highly sensitive tasks on the original node to wait for recovery. If a compatible node exists but the cross-park logistics time exceeds the preset delivery grace period, causing it to fail to complete within the delivery window despite process adaptation, the system will mark it as a backup resource rather than the preferred resource. If the process parameters of a certain batch are not fully described, such as the appearance attributes not being explicitly maintained in the process card, the system will prioritize matching according to the stricter standards to avoid hidden quality risks caused by lenient matching. In the charging cabinet housing order, the main bending node is processing the cabinet door exterior side panel. These parts require stable straightness of the folded edge, no indentations on the surface, and a tight correspondence with the welding positioning pin holes in the future. The system search found that although the spare bending node in the same park has a lower production capacity, it is equipped with the same type of upper and lower dies and an exterior protective film device. Although the bending node in another cooperating factory has sufficient cycle time, its die opening is more suitable for internal reinforcing ribs. Therefore, the system only transferred the cabinet door exterior side panel to the former, instead of transferring all parts to the latter. The purpose of this step is to incorporate the availability of the capability to undertake the task and whether the quality requirements are met after the task is undertaken into the resource matching logic, so as to achieve the executability and quality stability of cross-node transfer. Furthermore, step S5 specifically includes: If the cascade disturbance damping coefficient of the candidate reconfiguration scheme is greater than the preset damping coefficient threshold, and the scheduling plan fluctuation index is less than the preset fluctuation index threshold, then the corresponding candidate reconfiguration scheme will be marked as the target reconfiguration scheme. If the above conditions are not met, the matching range of cross-node resources is updated, and the step of generating candidate reconstruction schemes is re-executed; if the number of times the matching range is updated reaches the preset number of rounds and there are still no candidate reconstruction schemes that meet the conditions, the transfer of new tasks is frozen and rate limiting protection is implemented on the corresponding supply chain nodes. The time difference between data fusion and the issuance of the target reconstruction plan will be persistently stored as the cross-node reconstruction response latency. This embodiment provides a target reconstruction scheme screening and response latency recording mechanism; specifically, based on the above, simply generating multiple candidate schemes is not enough, because different schemes may perform differently in terms of delivery recovery capability and the degree of disturbance to the original plan; if only the goal is to minimize the final delay, the system may frequently affect a large number of irrelevant nodes. If only the stability of the plan is pursued, the system may not be able to respond adequately to real risks. Therefore, this embodiment further introduces a dual threshold screening mechanism of damping coefficient threshold and volatility index threshold, and performs resource range update and regeneration when the conditions are not met, while recording the response latency of the entire pre-reconstruction. Specifically, the damping coefficient threshold is used to determine whether a candidate solution has sufficient delay absorption capacity; specifically, if the delay reduction of the final delivery node after the solution is implemented does not reach the preset threshold, the solution is removed from the candidate pool; the volatility index threshold is used to determine whether the disruption of the original scheduling order by the solution is still within a controllable range; that is, the system only accepts solutions that can effectively absorb local disturbances without pulling the global scheduling into frequent oscillations; both conditions must be met simultaneously for the candidate solution to be marked as the target solution. If a candidate solution fails the dual threshold screening, the system does not abandon it directly, but updates the matching range of cross-node resources and regenerates the candidate solution. The update may include expanding the geographical range, expanding the time window, allowing some intermediate buffer inventory to intervene, or adding more parallel nodes under the premise of process compatibility. The reason for adopting this mechanism is that the upper-level solution may face the problem of too narrow a resource field of view in complex disturbance scenarios: for example, when only looking for alternative nodes within the local park, although the plan can be guaranteed to be stable, the delay reduction is limited; while after appropriately expanding to the collaborative park, a more suitable take-off point may be found; by expanding the matching range to regenerate the reconstruction solution when the standard is not met, the early locking of low-quality solutions can be avoided. When recording the latency of cross-node reconstruction response, the system takes the moment when the multi-source heterogeneous perturbation data is fused as the starting point and the moment when the target solution is officially issued to the execution node as the ending point. This time difference reflects the overall agility of the system from seeing the risk to taking action. After storing this indicator for a long time, it can be used to optimize threshold settings, resource allocation and system communication links. For example, if it is found that the latency of night shift cross-park scheduling exceeds the preset standard latency threshold for a long time, it indicates that the problem may not be in the algorithm itself, but may be in the confirmation link of the cooperating node or the logistics linkage mechanism. In one specific embodiment, suppose solutions P1, P2, and P3 are generated for the same stressed node; P1 can reduce the delay, but requires changes to a large number of original welding sequences; P2 has a smaller impact on the original plan, but the reduction in delivery delay is insufficient; P3 can significantly bring back the final delivery node delay, while only affecting a small number of unstarted tasks; then the system selects P3 as the target solution; if none of P1, P2, and P3 simultaneously meet both thresholds, the system can expand the matching scope from parallel nodes in the same park to collaborative nodes in the same city + buffer inventory linkage, and regenerate a new set of solutions; Furthermore, if no solution meets the dual threshold conditions after a preset number of matching rounds of updates, the system can enter a conservative execution mode: freeze the transfer of new tasks, implement rate limiting protection only for stressed nodes, and output a manual review prompt to the management end to avoid repeated system reconstruction under low feasibility conditions; if a key timestamp is found to be missing when recording the response latency, such as the execution node not yet sending back order confirmation, the system first records a semi-closed-loop latency, and then updates it to a full closed-loop latency after the response is sent back to complete it, ensuring that historical data can be continuously accumulated without interruption due to single point of failure. In the scenario of charging cabinet casing orders, the system initially found only two bending alternatives within the main park. One of them could absorb the pressure of the main node, but it would force the overall production scheduling of the painting fixtures to be postponed. The other hardly changed the painting process, but it could only take on manufacturing tasks that accounted for a smaller proportion than the preset effective ratio, and had limited help in the final delivery. Based on this, the system determined that neither of the two options was ideal, expanded the matching scope to the collaborative factory area, and regenerated the solution by combining the transfer train schedule. Finally, a solution was found that could both handle the key appearance parts and not change most of the welding cycle time. The system also recorded the time difference from the completion of the multi-source data fusion during the night shift to the issuance of the target solution, as the response latency of this cross-node reconstruction. The purpose of this step is to establish a dual screening threshold that effectively absorbs disturbances and suppresses planned oscillations, and to form a basis for sustainable optimization through incubation period tracking, thereby achieving more robust upfront restructuring decisions. The sheet metal parts supply chain scheduling system based on multi-source data includes the following modules: The data acquisition and fusion module is used to acquire and fuse multi-source heterogeneous perturbation data from various supply chain nodes in the sheet metal parts supply chain network to construct a node status dataset. The multi-source heterogeneous perturbation data includes wear status data of underlying equipment, transient fluctuation data of logistics, and batch deviation data of raw materials. The stress determination module is used to determine the node state deviation index of each supply chain node based on the node state dataset; the node state deviation index is used to quantify the degree to which the supply chain node deviates from the ideal flow state. The pre-reconstruction trigger module is used to generate a local task transfer instruction when the node state deviation index is greater than a preset stress pressure threshold. The reconfiguration scheme evaluation module is used to respond to local task transfer instructions, obtain the available capacity of adjacent nodes in the sheet metal supply chain network as cross-node resources, generate candidate reconfiguration schemes, and determine the cascade disturbance damping coefficient and scheduling plan fluctuation index corresponding to the candidate reconfiguration schemes. The self-balancing execution module is used to select a target reconfiguration scheme from candidate reconfiguration schemes and issue it for execution based on the cascaded disturbance damping coefficient and the scheduling plan fluctuation index. This embodiment provides a sheet metal parts supply chain scheduling system based on multi-source data, such as... Figure 2 As shown; specifically, the system can be deployed on the scheduling master station server of a manufacturing enterprise, or in a cloud-edge collaborative architecture, where the edge side is responsible for collecting the status of workshop equipment and logistics, and the central side is responsible for cross-node solution evaluation and distribution; the system works around the aforementioned new energy vehicle charging cabinet shell order scenario, and the modules form a continuous closed loop, rather than isolated functional stacking; Specifically, the data acquisition and fusion module is responsible for establishing data connections with equipment gateways, manufacturing execution systems, warehouse management systems, enterprise resource planning systems, quality inspection systems, and logistics positioning terminals; its acquisition objects include equipment operating signals, task execution status, inventory and in-transit information, batch inspection records, etc., and merging them into node status datasets; After receiving the dataset, the stress determination module performs differentiated state assessments on different nodes to generate corresponding node state deviation indices. Based on this, the pre-reconstruction trigger module determines which nodes' node state deviation indices have entered the risk pre-intervention range and generates local task transfer instructions. After receiving the instruction, the reconfiguration scheme evaluation module searches for adjacent resources in the network, generates candidate schemes, and evaluates their impact on delivery delays and plan stability; the self-balancing execution module then selects target schemes from them and simultaneously distributes them to relevant workshop terminals, logistics scheduling terminals, and execution dashboards. The above modules can be projected as software functional units or further mapped as a hardware and software collaborative structure. For example, the data acquisition and fusion module can consist of an industrial gateway, database services, and semantic processing services; the node state deviation index and reconfiguration scheme evaluation module can be deployed in the scheduling engine server; the self-balancing execution module can be linked with the manufacturing execution system dispatch interface, automated guided vehicle scheduling interface, electronic tag system, and Kanban system. In this way, the system can not only perceive the underlying perturbations but also deliver decisions directly to the execution site. For ease of explanation, a simplified scenario-based module flow diagram can be provided; let node N1 be the main bending node and N2 be the backup bending node; the data acquisition and fusion module obtains equipment status package A from N1, material arrival status package B from the logistics system, and material batch package C from the quality inspection system, and outputs node snapshot S for N1; the node status deviation index is calculated by weighted summation of the status data in S, and the node status deviation index R1 is obtained. After the pre-reconstruction trigger module determines that the state deviation index R1 of the node is greater than the state deviation threshold, it generates a transfer instruction M1; the reconstruction scheme evaluation module constructs a scheme set {P1, P2} accordingly; the self-balancing execution module selects P1 and sends it to N2 and related logistics nodes; the whole process reflects the module's serial collaborative relationship. As a fault tolerance mechanism, if the data acquisition and fusion module detects an interruption of a critical data source, such as the field gateway going offline, the system can switch to degrade mode, call the data snapshot within the most recent stable window as a temporary basis, and at the same time increase the level of manual confirmation. If a node's state deviates from the exponent and cannot form a reliable result, such as a node having severe state conflicts, the pre-reconstruction triggering module will not directly trigger the transfer, but will instead place the node in the observation state; if the reconstruction scheme evaluation module finds that the resource pool is empty, the self-balancing execution module can only issue load reduction and buffering commands, without executing cross-node transfer; When processing charging cabinet shell orders during the night shift, the data acquisition and fusion module first detects the simultaneous presence of wear signals on the main bending node equipment, in-transit logistics jitter, and material batch differences; the node status deviation index is calculated based on this data and is greater than the status deviation threshold; the pre-reconstruction trigger module then initiates a local task transfer; The reconstruction scheme evaluation module compares the connection paths between the main park and the cooperative plant; the self-balancing execution module finally sends the reconstruction results to the backup bending node, the park logistics dispatching station and the welding line Kanban simultaneously; on-site personnel can continue production according to the updated task package without waiting for the plant-wide rescheduling; The purpose of this system is to integrate multi-source perception, risk assessment, local reconstruction, and execution linkage into a unified scheduling closed loop, thereby achieving supply chain-level proactive intervention and resilient operation. Furthermore, the data acquisition and fusion module is specifically used for: The time series alignment and semantic feature extraction of multi-source heterogeneous perturbation data are fused to output a node state dataset. This embodiment provides a refined implementation mechanism for the system-side data acquisition and fusion module. Specifically, in the previous system solution, if the module only undertakes the function of acquisition and forwarding, subsequent modules still need to face the problems of original data timing conflicts and expression differences, resulting in a slower overall link response, and different modules may have inconsistent understandings of the same data. To this end, this embodiment directly integrates time series alignment and semantic feature extraction into the data acquisition and fusion module, so that the data entering the scheduling engine already has a unified time sequence and clear industrial semantics. Specifically, the module may include a time normalization subunit, a window mapping subunit, and a semantic extraction subunit; the time normalization subunit is used to correct clock deviations from different data sources; the window mapping subunit is used to merge data with different sampling frequencies into the same task process window; The semantic extraction subunit is used to convert raw numerical values, trajectories and text into schedulable states such as accelerated wear, instability in transit, and batch forming sensitivity; the module outputs not an unfiltered and unprocessed raw data stream, but a state dataset organized by nodes, which can be directly called for subsequent deviation evaluation. The necessity lies in the fact that if these processing actions are distributed to various subsequent modules, it will cause duplicate calculations and the same device waveform will be interpreted with different meanings in different modules, reducing scheduling consistency. Therefore, it is more suitable to build a unified state base by placing the fusion action in the acquisition module. A simplified scenario model can be provided; assume the field gateway uploads device data packets G1 and G2, the logistics platform uploads status packet W1, and the quality inspection system uploads batch record Q1; after time normalization, these data are mapped to the node B-batch T1-08:00 window; after semantic extraction, a tag group is formed {increased equipment load, deteriorated material delivery continuity, sensitive material rebound}; finally, a status snapshot S1 is output to subsequent modules, instead of distributing G1, G2, W1, and Q1 as is; As a fault-tolerance mechanism, if the semantic extraction subunit has not yet established mapping rules for certain new device codes, the original fields can be retained and unexplained status markers can be added. At the same time, the rule base can be supplemented through the manual maintenance interface. If the time drift of a data source exceeds the allowable range is found during the time normalization process, it will not be merged into the current window for the time being, but will be put into the queue to be corrected to prevent incorrect alignment from polluting the node status dataset. In the night shift scheduling of charging cabinet shell orders, after the data acquisition and fusion module receives the bending machine compensation record, the automated guided vehicle transfer return, and the steel plate batch inspection results, it first unifies their time base, then maps them to the same process window, extracts the semantic state of the node where the processing stability allowance decreases, and outputs it; subsequent modules no longer need to interpret the original code and can directly enter the deviation judgment. The purpose of this mechanism is to pre-organize the raw heterogeneous data into a unified, understandable, and callable node state representation, thereby achieving stable input to the system-level scheduling link; Furthermore, when acquiring cross-node resources, the refactoring scheme evaluation module is specifically used for: Match the available capacity of adjacent nodes with compatible sheet metal process parameters as cross-node resources; This embodiment provides a refined resource screening mechanism for a system-side reconfiguration scheme evaluation module. Specifically, in the previous system scheme, if the reconfiguration scheme evaluation module only screens resources based on available capacity and geographical proximity, although it can quickly generate candidate schemes, it is easy to misjudge idle resources as available resources in the sheet metal scenario. Therefore, this embodiment embeds process parameter compatibility verification into the resource search process of this module. Specifically, the reconfiguration scheme evaluation module may further include a resource retrieval subunit, a process compatibility verification subunit, and a transferability marking subunit; the resource retrieval subunit initially identifies adjacent nodes that can be carried over in time based on local task transfer instructions; The process compatibility verification subunit compares the equipment capabilities, mold combinations, plate thickness applicable range, surface protection conditions, reference fixtures, and subsequent connection requirements of these nodes; the transferability marking subunit classifies adjacent nodes into three categories based on the comparison results: fully compatible, acceptable, partially compatible, acceptable subtasks, and unacceptable; subsequently, candidate refactoring schemes are generated only based on the first two categories of resources; The reason for this design is that if the system only discovers quality risks during the evaluation phase, a lot of scheduling time is often wasted, and it may even lead the site to mistakenly believe that the solution is feasible. Moving the process compatibility judgment to the resource search phase can reduce the number of invalid solutions and improve the efficiency and accuracy of refactoring. A simplified scenario model can be used as an example. Suppose the resource retrieval subunit finds adjacent nodes X1, X2, and X3. After process compatibility verification, X1 meets the requirements for plate thickness, mold, and appearance protection, and is marked as Class A. X2 only meets the plate thickness requirements but lacks appearance protection, and is marked as Class B, and can only handle internal reinforcement parts. X3 has production capacity but insufficient bending length, and is marked as Class C. Therefore, in the solution generation stage, appearance parts can be mapped to X1, internal parts can be mapped to X2, and X3 can be eliminated. As a fault-tolerance mechanism, if the process parameter database is incomplete, for example, if a cooperative plant temporarily changes a mold but has not yet updated the system, the module can request the node side to quickly confirm before marking it as acceptable; if the process compatibility verification result conflicts with the historical execution result, for example, if the system determines it is acceptable but the historical rework rate is high, then the verification conditions will be tightened by referring to the historical bad records first; if all resources are judged to be unacceptable, the module will mark this reconstruction as buffer-only and not transferable, and send it back to the execution module for protective cycle control; In the order processing of charging cabinet housing, the reconstruction scheme evaluation module found two spare bending nodes and one outsourced bending node. After process compatibility verification, it was found that only one spare node has both the anti-indentation capability of cabinet door appearance parts and the consistency of the reference fixture, while the other is only suitable for reinforcing rib parts. Although the outsourced node is available, its mold length is insufficient. Based on this, the module restricts the transfer of door panel appearance parts to the former, without expanding the transfer range without constraints. The purpose of this mechanism is to integrate process manufacturability constraints into the resource assessment process in advance, so as to ensure that the quality of candidate reconfiguration solutions is controllable and their implementation is feasible. Furthermore, the self-balancing execution module is specifically used for: If the cascade disturbance damping coefficient of the candidate reconfiguration scheme is greater than the preset damping coefficient threshold, and the scheduling plan fluctuation index is less than the preset fluctuation index threshold, then the corresponding candidate reconfiguration scheme will be marked as the target reconfiguration scheme. If the above conditions are not met, the reconstruction scheme evaluation module will be triggered to update the matching range of cross-node resources and regenerate candidate reconstruction schemes. The time difference between data fusion and the issuance of the target reconstruction plan will be persistently stored as the cross-node reconstruction response latency. This embodiment provides a refined closed-loop mechanism for a system-side self-balancing execution module. Specifically, in the previous system solution, if the self-balancing execution module only assumes the simple execution role of receiving and issuing the plan, it is difficult to guarantee that the final issued plan will simultaneously take into account delivery recovery and plan stability. Therefore, this embodiment centrally sets up dual threshold judgment, resource range callback, and response latency retention in this module, so that it is not only responsible for execution, but also for final closed-loop control. Specifically, the self-balancing execution module may include a threshold management subunit, a dual-condition judgment subunit, a callback reconstruction subunit, a scheme issuance subunit, and a latency recording subunit; the threshold management subunit maintains the damping coefficient threshold and volatility index threshold for different product families and different delivery levels; the dual-condition judgment subunit reviews each candidate scheme; When a candidate solution meets the requirements of sufficient delay absorption capacity and sufficiently small planned disturbance, the solution distribution subunit marks it and distributes it to the target node; if not, the callback refactoring subunit returns a range expansion request to the refactoring solution evaluation module; the latency recording subunit records the entire time from risk identification to solution execution. This design further strengthens the previous layer system solution. In complex supply chain environments, the quality of candidate solutions cannot be guaranteed by generating them once. The system needs to have the ability to self-callback and re-evaluate. Otherwise, it is easy to quickly execute suboptimal solutions, causing subsequent global rescheduling involving more supply chain nodes. Placing the callback logic in the self-balancing execution module is conducive to forming a closed loop of evaluation-screening-reconstruction if not satisfied-execution after satisfaction. In one specific embodiment, the candidate solution set is set as {F1, F2}. After review by the dual-condition judgment subunit, it is found that F1 significantly improves the final delivery, but it will change the production schedule of multiple stable nodes, so it is not approved; F2 has little impact on the original schedule and can still restore the critical delivery window, so it is approved and marked as the target solution; if both F1 and F2 are not approved, the reconstructing subunit is called back to issue a resource expansion request, for example, to expand the search boundary from within the park to the collaborative factory area, and wait for a new solution set {F3, F4} again; after the target solution is determined and issued, the latency recording subunit records the total time of this round of the process and writes it into the historical database. Furthermore, if no usable solution is found after a preset number of callback reconstructions, the self-balancing execution module enters conservative mode, issuing only rate limiting, buffering, and manual confirmation instructions, and does not continue to automatically expand the matching range to prevent non-convergent iterative calculations; if the threshold management subunit finds that there is no mature threshold template for the current product family, it can call a similar historical template as an initial reference, and manually review and revise it after this execution; if a certain time point is missing during the incubation period recording process, the module should at least save the known key node time to ensure that the bottleneck of the link can still be analyzed later. In the night shift anomaly handling of charging cabinet shell orders, the self-balancing execution module first reviewed the two transfer plans formed in the main park. It was found that although one of them could shorten the delivery delay, it would cause welding and painting to be re-queued. The other only required adjusting a small number of unstarted door panel tasks and could maintain the stability of the painting fixture sequence. Therefore, the latter was selected as the target plan and issued. If neither of these two solutions is satisfactory, the module will automatically request the refactoring solution evaluation module to expand the search boundary to the collaborative plant area. After the final solution is determined, the system will archive the entire process from the identification of the abnormal fusion of the bending node to the completion of the target solution, providing a basis for subsequent night shift resource allocation optimization. The purpose of this mechanism is to base the final execution on quantifiable dual-condition screening and a reversible reconstruction loop, and to continuously accumulate system response capability data through the incubation period, thereby achieving more robust local forward transfer and supply chain self-balancing.

[0015] It should be noted that 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A sheet metal parts supply chain scheduling method based on multi-source data, characterized in that, Includes the following steps: Step S1: Obtain multi-source heterogeneous perturbation data of each supply chain node in the sheet metal parts supply chain network and perform fusion processing to construct a node status dataset; wherein, the multi-source heterogeneous perturbation data includes the wear status data of the underlying equipment, the transient fluctuation data of logistics, and the batch deviation data of raw materials; Step S2: Based on the node status dataset, determine the node status deviation index of each supply chain node; wherein, the node status deviation index is used to quantify the degree to which the supply chain node deviates from the ideal flow state; Step S3: If the node state deviation index is greater than the preset stress pressure threshold, a local task transfer instruction is generated; if the node state deviation index is not greater than the stress pressure threshold, the original scheduling plan is maintained. Step S4: In response to the local task transfer instruction, extract the current manufacturing tasks of the supply chain nodes whose node state deviation index is greater than the stress pressure threshold, and obtain the available capacity of adjacent nodes in the sheet metal supply chain network as cross-node resources to generate candidate reconfiguration schemes; evaluate the cascade disturbance damping coefficient based on the ratio of the difference in the final delivery node delay time before and after implementing the candidate reconfiguration scheme to the expected delay time before implementation, and determine the scheduling plan fluctuation index based on the task change rate and time offset of the affected nodes after the implementation of the candidate reconfiguration scheme; Step S5: Based on the cascaded disturbance damping coefficient and the scheduling plan fluctuation index, select the target reconstruction scheme from the candidate reconstruction schemes and issue it for execution.

2. The sheet metal parts supply chain scheduling method based on multi-source data according to claim 1, characterized in that, The step S1, which involves fusing the multi-source heterogeneous perturbation data, specifically includes: The multi-source heterogeneous perturbation data is fused by time series alignment and semantic feature extraction to output the node state dataset.

3. The sheet metal parts supply chain scheduling method based on multi-source data according to claim 1, characterized in that, The step S4 of obtaining cross-node resources specifically includes: The available capacity of adjacent nodes with compatible sheet metal process parameters is used as the cross-node resource.

4. The sheet metal parts supply chain scheduling method based on multi-source data according to claim 1, characterized in that, Step S5 specifically includes: If the cascaded disturbance damping coefficient of the candidate reconstruction scheme is greater than a preset damping coefficient threshold, and the scheduling plan fluctuation index is less than a preset fluctuation index threshold, then the corresponding candidate reconstruction scheme is marked as the target reconstruction scheme. If the above conditions are not met, the matching range of the cross-node resources is updated, and the step of generating candidate reconstruction schemes is re-executed; if the number of times the matching range is updated reaches a preset number of rounds and there are still no candidate reconstruction schemes that meet the conditions, the transfer of new tasks is frozen and rate limiting protection is implemented on the corresponding supply chain nodes. The time difference between data fusion and the issuance of the target reconstruction scheme will be persistently stored as the cross-node reconstruction response latency.

5. A sheet metal parts supply chain scheduling system based on multi-source data, characterized in that, Includes the following modules: The data acquisition and fusion module is used to acquire and fuse multi-source heterogeneous micro-perturbation data from each supply chain node in the sheet metal parts supply chain network to construct a node status dataset; wherein, the multi-source heterogeneous micro-perturbation data includes wear status data of underlying equipment, transient fluctuation data of logistics, and batch deviation data of raw materials. The stress determination module is used to determine the node state deviation index of each of the supply chain nodes based on the node state dataset; wherein the node state deviation index is used to quantify the degree to which the supply chain node deviates from the ideal flow state; The pre-reconstruction trigger module is used to generate a local task transfer instruction when the node state deviation index is greater than a preset stress pressure threshold. The reconfiguration scheme evaluation module is used to respond to the local task transfer instruction, obtain the available capacity of adjacent nodes in the sheet metal supply chain network as cross-node resources, generate candidate reconfiguration schemes, and determine the cascade disturbance damping coefficient and scheduling plan fluctuation index corresponding to the candidate reconfiguration schemes. The self-balancing execution module is used to select a target reconstruction scheme from the candidate reconstruction schemes and issue it for execution based on the cascaded disturbance damping coefficient and the scheduling plan fluctuation index.

6. The sheet metal parts supply chain scheduling system based on multi-source data according to claim 5, characterized in that, The data acquisition and fusion module is specifically used for: The multi-source heterogeneous perturbation data is fused by time series alignment and semantic feature extraction to output the node state dataset.

7. The sheet metal parts supply chain scheduling system based on multi-source data according to claim 5, characterized in that, When acquiring cross-node resources, the reconstruction scheme evaluation module is specifically used for: The available capacity of adjacent nodes with compatible sheet metal process parameters is used as the cross-node resource.

8. The sheet metal parts supply chain scheduling system based on multi-source data according to claim 5, characterized in that, The self-balancing execution module is specifically used for: If the cascaded disturbance damping coefficient of the candidate reconstruction scheme is greater than a preset damping coefficient threshold, and the scheduling plan fluctuation index is less than a preset fluctuation index threshold, then the corresponding candidate reconstruction scheme is marked as the target reconstruction scheme. If the above conditions are not met, the reconstruction scheme evaluation module is triggered to update the matching range of the cross-node resources and regenerate the candidate reconstruction scheme. The time difference between data fusion and the issuance of the target reconstruction scheme will be persistently stored as the cross-node reconstruction response latency.

Images