A supply chain-oriented intelligent order management method and system

Abstract

The application relates to the technical field of order management, and discloses an intelligent order management method for a supply chain, which comprises the following steps: obtaining corresponding multi-modal data through order demand flow, production equipment state, logistics sensor dynamic information and inventory topological atlas; analyzing the order and equipment correlation based on a space-time graph convolution network to generate a production capacity allocation scheme; calculating a logistics path planning scheme, predicting inventory shortage risk and generating a replenishment suggestion; if there is a high-priority order, inserting the production capacity plan and adjusting the equipment process chain; if there is a resource conflict, dynamically allocating resources; if the path risk value exceeds a threshold value, triggering a standby path switching; adjusting the weighted parameter through an adaptive federal algorithm to generate a global strategy and issue the global strategy to a client; the client dynamically adjusts the local configuration and uploads execution effect data in real time; if an exception is detected, triggering global strategy regeneration and updating the model through federal learning incrementally. The application can realize efficient management of supply chain orders.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of order management, in particular to a supply chain-oriented intelligent order management method and system. BACKGROUND

[0002] The current supply chain order management system generally adopts a traditional ERP architecture and relies on manual rule configuration and a static optimization model to perform order allocation and resource scheduling. However, such a system lacks a real-time self-adaptive closed-loop optimization mechanism in a dynamic environment, which leads to the inability to quickly respond to sudden demand fluctuations and resource conflicts.

[0003] Especially in a high-concurrency order scenario, a traditional centralized optimization algorithm has an exponential increase in computational complexity, and it is difficult to generate a feasible solution within a limited time, resulting in order response delay and resource waste.

[0004] As known from the above, how to realize efficient management of supply chain orders remains to be solved. SUMMARY

[0005] In order to realize efficient management of supply chain orders, the application provides a supply chain-oriented intelligent order management method and system.

[0006] In a first aspect, the application provides a supply chain-oriented intelligent order management method, which adopts the following technical solution:

[0007] A supply chain-oriented intelligent order management method comprises the following steps:

[0008] An order demand stream corresponding to an ERP system is extracted, the order demand stream comprising order quantity, priority and delivery time constraints; a production equipment state corresponding to an Internet of Things sensor is collected, the production equipment state comprising equipment vibration spectrum, temperature curve and energy consumption fluctuation; logistics dynamic information corresponding to a logistics sensor is obtained, the logistics dynamic information comprising transportation path coordinates, carrier speed and environmental temperature and humidity; a warehouse RFID system generates an inventory topology map, the inventory topology map comprising three-dimensional goods positions and inventory distribution generated by the warehouse RFID system; the order demand stream, the production equipment state, the logistics dynamic information and the inventory topology map are aligned through a space-time synchronization protocol to obtain corresponding multi-modal data;

[0009] The order and the equipment state are analyzed based on the order demand stream and the production equipment state through a pre-set space-time graph convolution network to obtain a corresponding capacity allocation scheme; a multi-objective optimal path is calculated based on the logistics dynamic information to obtain a corresponding path planning scheme, the multi-objective optimal path comprising cost, timeliness and carbon emission; a Bayesian network is constructed based on the inventory topology map to predict a stockout risk in the next 72 hours to generate a corresponding replenishment suggestion;

[0010] determining whether there is a high-priority order based on the capacity allocation scheme, when a high-priority order is received, inserting a new task in the capacity allocation scheme through a lossless order insertion algorithm, determining an affected equipment process chain based on the new task and making adjustments, determining whether there are multiple orders competing for the same resource based on the capacity allocation scheme, if so, calling a fuzzy logic arbitrator to dynamically allocate resources, generating conflict resolution instructions and executing them, and triggering a backup path scheme when the risk value corresponding to the path planning scheme is greater than the preset risk threshold value;

[0011] The ERP system, Internet of Things sensors, logistics sensors and warehouse RFID system in the supply chain are used as federated learning clients. The warehouse RFID system re-trains the ST-GCN model based on equipment state data, extracts the corresponding capacity prediction error, and the logistics sensor optimizes the path model according to the path planning error to predict the probability of carbon emission exceeding the standard. The warehouse RFID system updates the Bayesian network through the inventory topology map to evaluate the corresponding restocking response time. An adaptive federated optimization algorithm is used to aggregate the parameters of each client, dynamically weight the nodes related to high-priority orders, and generate a corresponding global strategy. The production equipment weight of high-priority orders is higher.

[0012] The global strategy is sent to each client, which dynamically adjusts the local device configuration after decryption, including sensor sampling rate and environmental detection accuracy. After each client executes, the corresponding execution effect data of each client is obtained, and the execution effect data is uploaded. If an anomaly is detected, the global strategy is regenerated, and the global model is updated incrementally through federated learning.

[0013] Optionally, during the analysis process of the spatio-temporal graph convolution network, the method further comprises: mapping the production equipment state into a spatio-temporal graph structure, wherein the nodes represent production equipment, and the edge weights are dynamically calculated based on the process dependency relationship between equipment and the real-time load difference; short-term fluctuation patterns and long-term trend features of the production equipment state are extracted through multi-level spatio-temporal convolution kernels, and dynamic capacity allocation weights are generated in combination with order priority; when a decrease in equipment health is detected, the allocation weight of the associated order is automatically reduced, and a backup equipment activation protocol is triggered.

[0014] Optionally, during the process of calling the fuzzy logic arbitrator to dynamically allocate resources, the three-dimensional evaluation index of resource conflict includes order priority, equipment health, and delivery time margin, and the method further comprises: constructing a fuzzy rule base, when multiple orders compete for the same equipment, orders with a priority greater than 0.8 and an equipment health greater than 0.7 are preferentially allocated; if the priority difference of the conflicting orders is less than 0.2, the carbon emission influence factor corresponding to the conflicting orders is compared, and the scheme with the lowest carbon emission increment is selected; when generating conflict resolution instructions, a backup resource window is reserved synchronously.

[0015] Optionally, in the risk value calculation process of the path planning scheme, the method further comprises: constructing a corresponding risk probability model based on real-time traffic flow, weather warning data and historical accident statistics in logistics dynamic information; obtaining the congestion index and weather disaster warning level corresponding to the path, marking the path as a high-risk path when the congestion index of the path > 0.6 or the weather disaster warning level ≥ 3; when obtaining the corresponding path planning scheme, synchronously generating a corresponding backup path scheme, the backup path scheme pre-generates multiple candidate paths, and dynamically switches the candidate paths according to the real-time risk value, and the switching response time < 15 seconds.

[0016] Optionally, in the process of abnormality detection and regeneration of the global strategy, the method further comprises: determining a corresponding abnormality criterion, the abnormality criterion comprising an order delivery delay rate > 10% or carbon emission exceeding > 15%, when detecting the abnormality, calling local data of each client, locating the abnormal root equipment or path based on the local data; when incrementally updating the global model, retaining the historical optimal version of the decision branch.

[0017] Optionally, in the process of dynamically adjusting the local equipment configuration, the method further comprises: when the federal strategy indicates a high-risk area, the GPS sampling rate of the logistics sensor is increased from 1 Hz to 10 Hz; when the vibration sensor of the production equipment detects a spectrum anomaly, a millisecond-level high-frequency acquisition mode is started; the scanning interval of the warehouse RFID system is adjusted according to the inventory turnover rate to be ≤ 2 minutes when the turnover rate > 5 times / day.

[0018] Optionally, the adaptive federal optimization algorithm, the method further comprises: dynamically adjusting the aggregation weight according to the data quality of the client, the node weight is increased by 30% when the data integrity > 90%; the client parameters associated with high-priority orders are encrypted and isolated, an adversarial verification mechanism is introduced, the abnormal deviation of the client uploaded parameters is detected, and the weight is automatically reduced when the deviation degree is greater than the deviation threshold.

[0019] In a second aspect, the present application provides an intelligent order management system for a supply chain, which adopts the following technical scheme:

[0020] An intelligent order management system for a supply chain, comprising:

[0021] The multi-modal data acquisition module extracts the corresponding order demand flow through the ERP system, the order demand flow including order quantity, priority and delivery time constraints, collects the corresponding production equipment state through the Internet of Things sensor, the production equipment state including equipment vibration spectrum, temperature curve and energy consumption fluctuation, acquires the corresponding logistics dynamic information through the logistics sensor, the logistics dynamic information including transportation path coordinates, carrier speed and environmental temperature and humidity, acquires the inventory topology map generated by the warehouse RFID system, the inventory topology map including the three-dimensional product location and inventory distribution generated by the warehouse RFID system, and aligns the order demand flow, the production equipment state, the logistics dynamic information and the inventory topology map through the space-time synchronization protocol for acquiring the corresponding multi-modal data.

[0022] The capacity allocation scheme acquisition module analyzes the correlation between orders and equipment states based on the order demand flow and the production equipment state through a pre-set space-time graph convolution network, for acquiring the corresponding capacity allocation scheme; calculates a multi-objective optimal path based on the logistics dynamic information, for acquiring the corresponding path planning scheme, the multi-objective optimal path including cost, timeliness and carbon emissions; constructs a Bayesian network based on the inventory topology map to predict the out-of-stock risk in the next 72 hours, and generates the corresponding replenishment suggestion.

[0023] The judgment module is used to judge whether there is a high-priority order based on the capacity allocation scheme, and when a high-priority order is received, a lossless order insertion algorithm is used to insert a new task in the capacity allocation scheme, and the affected equipment process chain is determined based on the new task and adjusted; the capacity allocation scheme is used to judge whether multiple orders are competing for the same resource, and if so, a fuzzy logic arbitrator is called to dynamically allocate resources, generate conflict resolution instructions and execute them; when the risk value corresponding to the path planning scheme is greater than a preset risk threshold, a backup path scheme is triggered.

[0024] The global strategy generation module uses the ERP system, the Internet of Things sensor, the logistics sensor and the warehouse RFID system in the supply chain as federated learning clients, the warehouse RFID system re-trains the ST-GCN model based on the equipment state data, extracts the corresponding capacity prediction error, the logistics sensor optimizes the path model according to the path planning error, predicts the probability of carbon emission exceeding the standard, the warehouse RFID system updates the Bayesian network through the inventory topology map, and evaluates the corresponding replenishment response time; uses an adaptive federated optimization algorithm to aggregate the parameters of each client, dynamically weights the nodes related to high-priority orders, and generates the corresponding global strategy, the production equipment weight of high-priority orders being higher.

[0025] An updating module is configured to distribute the global strategy to each client, and each client is configured to dynamically adjust a local device configuration after decryption, wherein the local device configuration comprises a sensor sampling rate and an environmental detection accuracy; after each client executes, execution effect data corresponding to each client is obtained, and the execution effect data is uploaded; if an exception is detected, the global strategy is triggered to be regenerated, and the global model is incrementally updated through federated learning.

[0026] In a third aspect, the application provides a supply chain-oriented intelligent order management method, which adopts the technical scheme as follows:

[0027] A supply chain-oriented intelligent order management method comprises a processor, and the processor runs a program of the supply chain-oriented intelligent order management method according to any one of the above.

[0028] In a fourth aspect, the application provides a storage medium, which adopts the technical scheme as follows:

[0029] A storage medium stores a program of the supply chain-oriented intelligent order management method according to any one of the above.

[0030] In summary, the application has at least one of the following beneficial technical effects:

[0031] Firstly, by integrating multi-modal data of ERP order flow, device status, logistics track and inventory distribution, and based on a spatio-temporal graph convolution network and a Bayesian prediction model, order demand is matched with production, logistics and warehousing capacity in real time; such cross-system data alignment and correlation analysis capability can accurately predict capacity gap, out-of-stock risk and path risk, so as to plan resource allocation and path optimization in advance, and significantly reduce delivery delay and inventory accumulation problems.

[0032] Secondly, for high-priority order insertion, resource contention and other scenarios, the scheme uses a lossless order insertion algorithm and a fuzzy logic arbitrator, combines multi-dimensional indicators such as order priority, device health and carbon emission impact, and realizes dynamic priority allocation of resources and rapid conflict resolution; for example, by reserving a spare resource window and switching a high-risk path in real time (<15 seconds response), the time efficiency of key orders is ensured while avoiding waste of production capacity or failure of delivery caused by sudden failure or logistics interruption.

[0033] Finally, the data value of each system client is aggregated through federated learning, which not only protects data privacy, but also dynamically adapts the global strategy to supply chain fluctuations through adaptive weight adjustment and abnormal detection mechanism; for example, the weight of high-priority orders associated with devices is increased, and the sampling rate is adjusted, to ensure priority supply of resources on critical paths; and abnormal root cause analysis and model incremental update capability continuously eliminate local decision bias and improve the response speed and resource utilization efficiency of the overall supply chain. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a flow chart of a supply chain-oriented intelligent order management method according to an exemplary embodiment.

[0035] Figure 2 is a structural block diagram of a supply chain-oriented intelligent order management system according to an exemplary embodiment. DETAILED DESCRIPTION

[0036] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings.

[0037] In the description of the present specification, the description of the terms "certain embodiments", "one embodiment", "some embodiments", "exemplary embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiments or examples are included in at least one embodiment or example of the present application. In the present specification, the exemplary description of the above terms does not necessarily mean the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0038] The embodiments of the present application disclose a supply chain-oriented intelligent order management method, referring to Figure 1 , comprising:

[0039] S100, extracting the corresponding order demand flow through the ERP system, collecting the corresponding production equipment state through the Internet of Things sensor, obtaining the corresponding logistics dynamic information through the logistics sensor, obtaining the inventory topology map generated by the warehouse RFID system, aligning the order demand flow, the production equipment state, the logistics dynamic information and the inventory topology map through the space-time synchronization protocol, and obtaining the corresponding multi-modal data.

[0040] Among them, for the order demand flow extracted by the ERP system, the order demand flow contains the order quantity, the priority (such as emergency order, regular order), the delivery time constraint (such as the deadline), which can be synchronized through the ERP system API interface or real-time database, and the order generation time, the customer specified delivery window, the SKU (inventory unit) involved in the order, etc. can be provided for subsequent production plan Dynamic demand benchmark, ensure that the capacity allocation matches the actual market demand.

[0041] For production equipment status acquired by IoT sensor data, the production equipment status includes device vibration spectrum, temperature curve, and energy consumption fluctuation. The device running vibration frequency can be monitored by an acceleration sensor, and abnormal spectrum may indicate bearing wear or mechanical failure. The temperature curve is recorded by a thermocouple or infrared sensor in real time. High temperature may indicate overload or cooling system failure. Energy consumption fluctuation is recorded by an electric meter or current sensor. Abnormal fluctuation may reflect a decrease in device efficiency or potential failure.

[0042] Sensor data is transmitted to the cloud or edge computing nodes through an industrial IoT gateway (such as MQTT protocol). The sampling frequency is set according to the device type (such as 100 Hz for vibration data and 1 Hz for temperature data). Real-time monitoring of device health status can be realized, providing a basis for production capacity prediction and fault warning, and avoiding delivery delays caused by device downtime.

[0043] For logistics dynamic information acquired by logistics sensors, the logistics dynamic information includes transportation path coordinates, vehicle speed, and environmental temperature and humidity. Transportation path coordinates can be determined by GPS or Beidou positioning system to record the real-time position of the transportation vehicle. Vehicle speed can be obtained by vehicle OBD device or CAN bus interface to assist in path optimization. Environmental temperature and humidity can be monitored by temperature and humidity sensors to monitor the storage environment of goods during transportation (such as cold chain transportation requiring constant temperature).

[0044] Vehicle data is aggregated through a logistics management platform (such as TMS system), and a dynamic traffic model is constructed by combining historical path data, so as to optimize path planning, avoid congestion or extreme weather, and reduce transportation risks (such as temperature control failure leading to cargo damage).

[0045] For inventory topology map generated by warehouse RFID system data, the inventory topology map includes three-dimensional product location and inventory distribution generated by the warehouse RFID system. The three-dimensional product location is scanned by the RFID reader on the warehouse shelf, and the X / Y / Z coordinates of the goods are generated by combining UWB (ultra-wideband) positioning technology. The inventory distribution is the real-time inventory quantity of each SKU, and the out-of-stock or excess area is marked. The RFID reader scans periodically (such as every 2 minutes), and the inventory data is updated by combining the warehouse management system (WMS). Real-time inventory location can be accurately located, supporting fast picking and replenishment decision-making, and reducing manual inventory errors.

[0046] For spatiotemporal synchronization protocol alignment multi-modal data, the specific execution steps are: all data sources use UTC timestamps to ensure consistent time reference; heterogeneous data (such as vibration spectrum as time series data, inventory as graph structure data) is unified into JSON or Parquet format; taking order generation time as an anchor point, production equipment state (such as the temperature of a certain equipment at the time of order generation), logistics path (transportation track after order issuance) and inventory data (availability of materials required by the order) are mapped to the same time window (such as ±5 minutes). Through metadata such as device ID, order ID and SKU code, multi-modal data is associated to specific production processes or logistics paths.

[0047] Multi-modal data can eliminate time delay and spatial deviation between data sources, providing a consistent data view for subsequent analysis, for example: determining whether the raw materials of a certain order have been warehoused (inventory data) and the production equipment is in an available state (equipment state data), and predicting the impact of logistics delay on delivery time (comparison of logistics path and order deadline).

[0048] S200, based on the order demand flow and the production equipment state, the association between the order and the equipment state is analyzed by a pre-set spatiotemporal graph convolution network to obtain a corresponding capacity allocation scheme; a multi-objective optimal path is calculated based on logistics dynamic information to obtain a corresponding path planning scheme, the multi-objective optimal path including cost, timeliness and carbon emission; a Bayesian network is constructed based on an inventory topology graph to predict the out-of-stock risk in the next 72 hours to generate a corresponding replenishment suggestion.

[0049] Among them, for order and device association analysis, it specifically includes: determining that the input data is order demand flow (quantity, priority, delivery time) and production equipment state (vibration, temperature, energy consumption), then using a spatiotemporal graph convolution network (ST-GCN) to map the production equipment state to a spatiotemporal graph structure, analyzing the association between device health, process dependency relationship and order priority, and outputting a dynamic capacity allocation scheme to ensure that high-priority orders are matched with healthy devices first and resource conflicts are avoided.

[0050] Through the spatiotemporal graph convolution network, the device health state (such as vibration anomaly, temperature overrun) is combined with the order priority to adjust the capacity allocation in real time, avoiding capacity waste caused by device failure or inefficiency; high-priority orders (such as urgent orders) are allocated to devices with high health degree and balanced load first to ensure on-time delivery of critical orders (for example, the on-time delivery rate of high-priority orders is improved to more than 95%); through process dependency relationship analysis, multiple orders competing for the same device resource are avoided, and order backlog or delay caused by resource conflicts is reduced.

[0051] For logistics path optimization, specifically including: determining the input data as logistics dynamic information (path coordinates, speed, temperature and humidity); then calculating the optimal path through multi-objective optimization algorithm (cost, timeliness, carbon emission), and synchronously generating backup path to cope with risks; thereby outputting path planning scheme, balancing cost and timeliness, and reducing the risk of carbon emission exceeding standard.

[0052] Under the premise of meeting delivery time, the cost-optimal path is selected (such as selecting a route or a type of carrier with higher fuel efficiency), the total logistics cost is reduced (for example, the transportation cost is reduced by 10%-15%), the driving distance is reduced through path optimization or a low-emission carrier (such as an electric truck) is selected, the carbon emission is reduced (for example, the carbon emission is reduced by more than 20%), the traffic congestion, weather disasters and other risks are monitored in real time, the backup path switching is triggered (the response time is less than 15 seconds), and the delivery delay caused by unexpected conditions is avoided (for example, the path switching success rate is 100% under extreme weather conditions).

[0053] For inventory prediction and replenishment recommendation, specifically including: determining the input data as inventory topology map (three-dimensional storage location, inventory quantity); then predicting the out-of-stock risk in the next 72 hours through the construction of Bayesian network, combining historical sales and supplier delivery cycle; thereby outputting replenishment recommendation, including emergency procurement trigger threshold and dynamic inventory threshold adjustment.

[0054] By predicting the SKU-level out-of-stock probability 72 hours in advance (such as triggering an early warning when the out-of-stock probability of a high-value SKU is greater than 30%), the order cancellation or customer loss caused by out-of-stock can be avoided; by dynamically adjusting the safety inventory threshold (such as increasing the inventory by 20% when the demand fluctuates greatly), the inventory accumulation can be reduced (for example, the inventory turnover rate is increased by 25%). According to the prediction result and the supplier delivery cycle, an emergency procurement order or an adjustment of the procurement plan is generated, which can ensure that the materials arrive in time (for example, the loss caused by out-of-stock is reduced by 40%).

[0055] In the embodiment of the application, in the analysis process of the spatio-temporal graph convolution network, the method further includes:

[0056] S210, mapping the production equipment state into a spatio-temporal graph structure.

[0057] Among them, the node represents the production equipment, each production equipment (such as an injection molding machine or an assembly line) is taken as a node in the graph, the node attribute includes real-time state (vibration spectrum, temperature, energy consumption) and health score (based on sensor data comprehensive evaluation),

[0058] The edge weight is dynamically calculated by the process dependency between devices and the real-time load difference. The connection between devices is defined according to the production process flow (such as injection molding machine→assembly line). The edge weight between devices is dynamically calculated by two parts: process correlation, the closer the process, the higher the weight (such as adjacent process weight=0.8, non-adjacent=0.3); load balancing, the greater the load difference between device A and device B, the lower the weight (such as load difference> 30%, weight reduced by 50%). And need to recalculate the edge weight every 5 minutes according to the latest sensor data to ensure that the graph structure reflects the real-time resource state.

[0059] S220, extract short-term fluctuation patterns and long-term trend features of production equipment state through multi-level spatio-temporal convolution kernel, and generate dynamic capacity allocation weight combined with order priority.

[0060] Among them, the specific execution steps include: short-term fluctuation pattern extraction, using 1D convolution kernel to analyze device sensor data (such as instantaneous abnormality of vibration spectrum, temperature sudden rise), identify device health risk; long-term trend feature analysis, using time convolution kernel to analyze the long-term change of device energy consumption curve (such as monthly average growth trend of energy consumption), predict device aging or performance decline.

[0061] The process of generating dynamic weight specifically includes: health weight, low device health (such as vibration anomaly or high temperature) reduces the weight; load balancing weight, the weight of the device with too high load is reduced; priority weight, the weight of high-priority orders is improved (such as emergency order weight+20%); the health, load, and priority are weighted to generate the final capacity allocation weight. Thus, a matching table of devices and orders is generated, which marks the expected completion time and resource conflict markers to output the capacity allocation scheme:

[0062] By combining device health, load, and order priority, the on-time rate of key orders can be maximized (such as high-priority order delivery on-time rate improved to 98%); and the allocation of aging devices (health<0.7) can be reduced to avoid delivery delays caused by failures.

[0063] S230, when detecting that the device health is declining, automatically reduce the allocation weight of its associated orders, and trigger the standby device activation protocol.

[0064] The health monitoring and triggering process specifically includes: real-time monitoring of device health (such as vibration abnormalities, temperature over-limit, energy consumption fluctuations), triggering a response when the health is below the threshold (such as <0.6), then automatically reducing the production capacity allocation weight of the device, for example: when the health is <0.6, the weight is reduced by 50%; when the health is <0.4, almost no new task is allocated. And activate the standby device: start the standby device (such as a standby injection molding machine), control the preheating time within 30 minutes, reassign the unfinished orders of the original device to the standby device with a health of >0.8, and mark them as "high-risk orders", triggering manual intervention processes.

[0065] For example, when an injection molding machine fails due to high temperature, the system switches to the standby device within 5 minutes to avoid order delays; through preventive adjustment, the downtime loss caused by equipment failure (such as 4 hours for average fault repair) is reduced.

[0066] Through order and device association analysis, logistics path optimization, and inventory prediction, the intelligent management of the supply chain is realized: first, based on the spatiotemporal graph convolution network, the health status of the device is dynamically matched with the order priority to ensure the priority delivery of key orders and improve the utilization rate of the device by 15%-20%; second, a multi-objective path optimization algorithm is used to balance cost, time efficiency, and carbon emissions, reducing transportation costs and environmental impact while ensuring path reliability; finally, the Bayesian network is used to predict the risk of stockout and generate replenishment strategies to reduce inventory accumulation and stockout losses, ultimately promoting the supply chain from passive response to active prediction, achieving the coordinated optimization of efficiency, cost control, and sustainability.

[0067] S300, based on the production capacity allocation scheme, determine whether there is a high-priority order to be received, when a high-priority order is received, insert a new task in the production capacity allocation scheme through a lossless order insertion algorithm, determine the affected device process chain based on the new task and make adjustments; based on the production capacity allocation scheme, determine whether multiple orders are competing for the same resource, if so, call the fuzzy logic arbitrator to dynamically allocate resources, generate conflict resolution instructions and execute; when the risk value corresponding to the path planning scheme is greater than the preset risk threshold, trigger the standby path scheme.

[0068] Among them, for high-priority order insertion processing, the specific execution process includes: the input data is the real-time received high-priority order and the current production capacity allocation scheme; then insert a new task into the production plan through a lossless order insertion algorithm, analyze its impact on the device process chain (such as delaying other orders or resource occupation); thus output the adjusted production capacity allocation scheme. Thus, high-priority orders can be executed with priority, while minimizing interference with the original plan.

[0069] For resource conflict arbitration, the specific execution process includes: the input data is that multiple orders in the capacity allocation scheme compete for the same resource (such as equipment, raw materials); then the fuzzy logic arbitrator is called, and the resource is dynamically allocated based on three-dimensional evaluation indexes (order priority, equipment health, delivery time margin); and the conflict resolution instruction is obtained, which includes the resource allocation result and the backup resource reservation strategy.

[0070] For path risk response, the specific execution process includes: the input data is the risk value (such as congestion index, weather warning level) of the path planning scheme, when the risk value exceeds the threshold value, the backup path scheme is triggered, and the candidate path is dynamically switched in real time; and the updated path planning scheme ensures the reliability of transportation.

[0071] In the embodiment of the application, in the process of calling the fuzzy logic arbitrator to dynamically allocate resources, the method further includes:

[0072] S310, a fuzzy rule base is constructed, when multiple orders compete for the same equipment, orders with a priority > 0.8 and an equipment health > 0.7 are preferentially allocated.

[0073] Among them, the rule base definition, based on business requirements, sets fuzzy rules, for example: rule 1, if the order priority > 0.8 and the equipment health > 0.7, then the order is preferentially allocated; rule 2, if the equipment health < 0.5, then the new task is refused to be allocated.

[0074] When it is detected that multiple orders compete for the same equipment (such as order A and order B need an injection molding machine at the same time), the fuzzy rule base is called for evaluation, and then priority judgment is performed, that is, the priority score (0-1 interval) of each order is calculated, if an order meets the conditions of rule 1, then the resource is directly allocated.

[0075] By defining the priority rules, order backlog caused by manual intervention delay can be avoided, and key orders (such as high-value customer orders) can be ensured to be preferentially executed; equipment health protection can also be realized, and tasks can be refused to be allocated to low health degree equipment, thereby reducing secondary risks caused by equipment failure.

[0076] S320, if the priority difference of the conflict orders < 0.2, the carbon emission influence factor corresponding to the conflict orders is compared, and the scheme with the lowest carbon emission increment is selected.

[0077] Among them, the specific execution steps include: when the priority difference of the conflict order is <0.2 (such as order A=0.85, order B=0.80), entering the carbon emission comparison link, and then performing carbon emission calculation, the production and transportation carbon emission increment of the two orders can be analyzed (such as order A needs to start high energy consumption equipment, and order B can reuse existing resources); if the carbon emission increment of order A is 10% higher than that of order B, order B is preferentially allocated, and the environmental impact is reduced.

[0078] When the priorities are similar, the solution with lower carbon emission is selected, which meets the ESG (environment, society, and governance) requirements, reduces the carbon tax or compliance risk, and can also reduce energy waste and long-term operating costs (such as electricity bills and carbon quota purchase costs).

[0079] S330, when generating conflict resolution instructions, a standby resource window is reserved synchronously.

[0080] Among them, the Du Yu reserved window defines that when generating conflict resolution instructions, 10%-15% of the device capacity or inventory resources are reserved as standby, for example: if a device is allocated 80% capacity to order A, 20% is reserved as standby. If a higher priority order is inserted later, the standby resource can be called to respond quickly, avoiding the triggering of conflicts again. It should be noted that if the reserved resource is not used within 48 hours, it is released back to the public resource pool.

[0081] The reserved resource provides a buffer for sudden orders or device failures, reduces delivery failures caused by insufficient resources, and also avoids idle resources caused by excessive reservation, and realizes efficient use of resources through dynamic monitoring.

[0082] In addition, in the embodiment of the application, in the risk value calculation process of the path planning scheme, the method further includes:

[0083] S340, a corresponding risk probability model is constructed based on real-time traffic flow, weather warning data and historical accident statistics in logistics dynamic information.

[0084] Among them, the specific execution steps include:

[0085] Firstly, data integration: real-time traffic data, access to navigation API (such as Gaode, Baidu) to obtain real-time congestion index; weather warning, integrate weather bureau API disaster warning levels such as heavy rain and fog; historical accident data, statistics of the number and type of accidents (such as car accidents and landslides) in a certain section in the past three months.

[0086] Then, risk probability calculation:

[0087] The risk value of the path is calculated by weighting the three types of data, with congestion index accounting for 40%, weather warning accounting for 30%, and historical accidents accounting for 30%. For example, the congestion index of a certain path is 0.7 (high), and the heavy rain warning is level 3 (medium), the risk value = 0.7 x 0.4 + 0.3 x 0.3 + 0.3 x 0.3 = 0.46 (high risk).

[0088] For threshold setting, risk value > 0.6 or weather warning ≥ 3 level is marked as high risk path. By converting abstract risk into calculable value, it provides basis for path switching; thus it can avoid relying on artificial experience judgment and reduce path planning deviation.

[0089] S350, obtain the congestion index and weather disaster warning level corresponding to the path, and mark the path as high risk path when the congestion index of the path is > 0.6 or the weather disaster warning level is ≥ 3 level.

[0090] Specific execution steps include:

[0091] Firstly, real-time risk assessment is carried out, and the path risk value is updated every 5 minutes. If the threshold value is exceeded (such as congestion index > 0.6), it is marked as high risk. Secondly, backup path is pre-generated, and 3-5 candidate paths (such as detouring highway, switching transportation mode) are calculated in advance according to historical data and real-time traffic. Then, path switching preparation is carried out, and the path switching condition (such as the risk value of the original path continuously > 0.8 for more than 10 minutes) is preset, and the switching response time (which needs to be < 15 seconds) is tested.

[0092] Switching the path before congestion or disaster can reduce the risk of delay (such as switching to inland route before typhoon); and path switching within 15 seconds can avoid long-time stasis of the vehicle, which can ensure the delivery time limit.

[0093] S360, when obtaining the corresponding path planning scheme, a corresponding backup path scheme is generated synchronously, the backup path scheme pre-generates multiple candidate paths, and dynamically switches the candidate paths according to the real-time risk value, and the switching response time is < 15 seconds.

[0094] Specific execution steps include:

[0095] Firstly, triggering switching, when the risk value of the original path exceeds the threshold value and the backup path is available, the switching instruction is issued to the logistics management system. Then, real-time monitoring and deviation correction, the deviation between the actual path of the vehicle and the planned path is compared every 30 seconds, and automatic deviation correction (such as GPS deviation > 500 meters, re-planning).

[0096] Finally, post-switching evaluation, record the actual time consumption, carbon emission and other data after path switching, for subsequent model optimization.

[0097] For example, when a certain path is marked as high-risk due to heavy rain, the system switches to a backup route within 5 minutes, ensuring timely delivery of goods; and continuously improves the accuracy of the path model through feedback data after switching.

[0098] Through the three core mechanisms of lossless order insertion algorithm, fuzzy logic arbitration and dynamic path switching, the supply chain realizes flexible execution and intelligent decision-making: First, high-priority orders can be seamlessly inserted into the production plan, ensuring that critical tasks are executed first and minimizing disruption to the original plan (emergency order delay rate ≤ 5%); second, based on the priority, equipment health and carbon emissions fuzzy rule base, quickly resolve resource conflicts and reserve backup resources, conflict resolution efficiency improved by 70%; finally, through real-time risk assessment and millisecond-level path switching, delivery failure rate caused by logistics interruption is reduced by more than 60%. Overall, S300 enables the supply chain to respond quickly to complex scenarios such as sudden orders, equipment failures or extreme weather, significantly improving delivery on-time rate and resource utilization efficiency.

[0099] S400, the ERP system, Internet of Things sensors, logistics sensors and warehouse RFID system in the supply chain are used as federated learning clients, the warehouse RFID system re-trains the ST-GCN model based on equipment state data, extracts the corresponding capacity prediction error, the logistics sensor optimizes the path model according to the path planning error, predicts the probability of carbon emission exceeding, the warehouse RFID system updates the Bayesian network through the inventory topology map, and evaluates the corresponding restocking response time; use adaptive federated optimization algorithm to aggregate parameters of each client, dynamically weight the weight of nodes related to high-priority orders, generate corresponding global strategy, the production equipment weight of high-priority orders is higher.

[0100] Among them, the specific execution process includes:

[0101] 1. Federated learning client initialization

[0102] Client registration: ERP, Internet of Things sensors, logistics sensors, and warehouse RFID systems register as federated learning nodes through a secure protocol

[0103] Data rights division: ERP, sharing order history data (without sensitive customer information); Internet of Things sensors, sharing desensitized equipment state data (such as statistical characteristics of vibration spectrum); logistics sensors, sharing path planning error and carbon emission data; warehouse RFID, sharing inventory distribution and restocking response time data; data encryption, all data is encrypted locally, only encrypted model parameters are uploaded instead of raw data.

[0104] 2. Warehouse RFID system: ST-GCN model training and capacity error analysis

[0105] Local model training: input, device status data (vibration, temperature, energy consumption); target, optimize ST-GCN model, predict device future 24-hour productivity; training method, use local historical data (such as device failure records and productivity fluctuations) to adjust model parameters.

[0106] Error analysis: calculate the error between the model's predicted productivity and the actual productivity (e.g., an injection molding machine predicts 100 pieces / day, but actually produces 95 pieces, so the error is 5%); mark devices with high errors (e.g., error > 15%) to trigger device health re-examination.

[0107] By optimizing the model with local data, we can reduce the waste caused by prediction bias (e.g., error reduction of 20%). It is important to note that high error devices may indicate potential faults, triggering maintenance in advance (e.g., if a device has an error > 10% for 3 consecutive days, schedule maintenance).

[0108] 3. Logistics sensor: path model optimization and carbon emission prediction

[0109] Local model training: input, path planning error data (e.g., difference between actual and planned time consumption); target, optimize path model, predict the carbon emission increment of new paths; training method, combine real-time traffic data and historical accident records to adjust path weights (e.g., reduce the weight of congested road segments by 30%).

[0110] Carbon emission prediction: analyze carbon emission factors for different paths (e.g., electric truck emission coefficient is 0.2 kg / km, diesel vehicle is 0.5 kg / km); predict paths that exceed the threshold (e.g., a path with carbon emissions greater than 15% of the government limit).

[0111] By optimizing the path, we can reduce the delay caused by route deviation (e.g., path time consumption error from 15% to 5%); we can also avoid penalties caused by excessive emissions, supporting ESG goals (e.g., annual carbon emissions reduced by 10%).

[0112] 4. Warehouse RFID system: Bayesian network update and replenishment response evaluation

[0113] Bayesian network update: input, inventory topology map (three-dimensional storage location distribution, inventory quantity change); target, update the dependency relationship between nodes (e.g., the correlation between the out-of-stock probability of a certain SKU and the supplier's delivery cycle); training method, combine new replenishment data and demand fluctuations to adjust node probability distribution.

[0114] Replenishment response evaluation: Calculate the time required for inventory to return to the safety threshold after the replenishment instruction is issued (such as after the emergency order is triggered, the replenishment response time < 2 hours). Thus, it can reduce the shortage or overstock caused by prediction bias (such as a 20% increase in inventory turnover rate); by quickly responding to demand fluctuations, it reduces the loss of shortage (such as a 40% reduction in the shortage rate of high-value SKUs).

[0115] 5. Adaptive federated optimization and global strategy generation

[0116] Parameter aggregation: Each client uploads encrypted model parameters (such as ST-GCN weights, path model coefficients), and uses an adaptive federated optimization algorithm (such as an improved version of FedAvg) to aggregate parameters: high-priority order weight increases, and the parameter weight of key production equipment increases by 30% (such as the weight of a certain customer order = 1.3); data quality weighting, nodes with data integrity > 90% weight increased by 20%.

[0117] Anomaly detection and isolation: Detect parameter abnormal nodes (such as a certain sensor data deviates from the global mean by > 30%), automatically reduce weight or isolate.

[0118] By updating the parameters related to high-priority orders first, it can ensure that critical path resources are supplied first; by fusing data from multiple clients, it can improve the adaptability of the model to supply chain fluctuations (such as quickly adapting to the characteristics of new factory equipment when a new factory is added).

[0119] 6. Strategy execution and feedback loop

[0120] Strategy delivery: Encrypt and deliver global strategies (such as capacity allocation weights, path model parameters) to each client.

[0121] Local execution: ERP updates order priority rules; logistics system switches to optimized path; warehouse RFID system adjusts replenishment threshold.

[0122] Effect feedback: Collect execution data (such as order delivery on-time rate, actual carbon emissions), upload to the federated learning center; if an anomaly is detected (such as a delivery delay rate > 10%), trigger incremental model update (only repair abnormal modules).

[0123] Through continuous iteration of the model, it can avoid local strategy bias (such as a factory equipment failure causing global capacity prediction bias); and after an anomaly is triggered, the model is repaired and the strategy is updated within 48 hours.

[0124] Integrate ERP, Internet of Things sensors, logistics sensors, and warehouse RFID systems as distributed clients through federated learning technology. Under the premise of protecting data privacy, achieve global optimization of the supply chain through local model training and global parameter aggregation: Collaborate with each client (such as warehouse RFID optimization capacity prediction model, logistics sensor optimization path and carbon emission prediction, RFID inventory replenishment model) to improve prediction accuracy, while dynamically weighting high-priority order-related resources (such as key equipment weight promotion) to ensure that urgent orders are executed first (on-time rate > 98%); and through real-time feedback and adaptive parameter updating, quickly respond to equipment failures, path interruptions, and other sudden scenarios (such as path switching response < 15 seconds), ultimately enabling the supply chain to evolve itself, significantly improving efficiency and reducing risks, becoming a core competitive advantage for enterprises.

[0125] S500, the global strategy is issued to each client, and the client decrypts and dynamically adjusts the local device configuration, including sensor sampling rate, environmental detection accuracy; after each client executes, the corresponding execution effect data of each client is obtained, and the execution effect data is uploaded, if an exception is detected, the global strategy is regenerated, and the global model is updated incrementally through federated learning.

[0126] Among them, the specific execution process includes:

[0127] 1. Global strategy encryption and distribution

[0128] First, strategy packaging: encrypt and package the global strategy generated by S400 (such as device configuration parameters, path weight, replenishment threshold) to ensure data security. Then, client distribution: distribute the encrypted strategy to each federated learning client (ERP, Internet of Things sensors, logistics sensors, warehouse RFID systems) through a secure channel (such as HTTPS). Finally, decryption and verification: the client decrypts the strategy using a private key and verifies the data integrity (such as through hash verification).

[0129] Based on the above three steps, the strategy can be prevented from being intercepted or tampered with, ensuring the confidentiality and reliability of the supply chain strategy; ensuring that each client only receives configuration parameters related to it (such as warehouse RFID systems only receiving inventory replenishment strategies).

[0130] 2. Client dynamically adjusts local device configuration

[0131] First, device configuration analysis: the client analyzes the parameters in the global strategy, for example:

[0132] Sensor sampling rate adjustment: IoT sensors increase the vibration sampling rate of critical equipment (e.g., high-priority order production equipment) from 1 Hz to 10 Hz; Environmental monitoring accuracy optimization: Logistics sensors adjust temperature and humidity monitoring accuracy based on path risk level (e.g., high-risk path accuracy is increased to ±0.5°C); Replenishment threshold update: Warehouse RFID system adjusts the safety stock threshold of a certain SKU from 100 to 120.

[0133] Then the configuration takes effect: dynamically adjust device parameters without manual intervention (e.g., remotely control sensors through API).

[0134] Through high-frequency sampling of critical equipment, faults can be detected earlier (e.g., 3 hours earlier warning of bearing abnormalities), and monitoring accuracy can be improved in high-risk scenarios, reducing decision bias caused by data errors.

[0135] 3. Execution effect data collection and upload

[0136] First, data collection: after the client executes the strategy, real-time collection of execution effect data, including: device side, adjusted capacity achievement rate, device failure rate; logistics side, actual path time, actual carbon emissions; warehouse side, replenishment response time, inventory turnover rate.

[0137] Then data encryption and upload: encrypt the data and upload it to the federated learning center to ensure privacy and security.

[0138] For example, verify whether path optimization reduces actual time consumption (e.g., original path time 3 hours → optimized 2.5 hours); if a certain device failure rate suddenly rises, it is marked as an abnormal trigger point.

[0139] 4. Abnormal detection and global strategy regeneration

[0140] First, set the abnormal threshold: preset the abnormal threshold of key indicators (e.g., delivery delay rate > 10%, carbon emissions exceed 15%).

[0141] Then real-time detection: the federated learning center compares the execution effect data with the expected target, and if it exceeds the threshold, it triggers an alarm; for example: the actual carbon emissions of a certain path are 20% higher than the predicted value, marked as abnormal.

[0142] Finally, strategy regeneration: root cause analysis, trace the association between abnormality and strategy (e.g., path model does not consider weather mutation); call S200-S400 modules, re-run capacity allocation, path planning and federated learning optimization, generate new strategy.

[0143] For example, when a typhoon causes path interruption, a backup path solution can be generated within 5 minutes; and through continuous iteration, decision-making failure caused by environmental changes can be reduced.

[0144] 5. Federated learning incremental update global model

[0145] First, incremental data integration: add the execution data of abnormal scenarios (such as path data under extreme weather) as incremental samples to the training set.

[0146] Then, local model fine-tuning: each client fine-tunes the local model based on the incremental data (such as optimizing the path model in the rainstorm scenario for logistics sensors).

[0147] Finally, parameter aggregation and update: use federated learning incremental algorithm (such as FedProx) to aggregate the fine-tuned parameters and update the global model, where only the abnormal related modules are updated (such as only optimizing the path model, keeping the stable parameters of the capacity model), that is, keeping the historical optimal version of the decision branch.

[0148] After accumulating extreme weather data, the path planning model's prediction accuracy in the rainstorm scenario is improved by 30%; and incremental update avoids full model retraining, saving computing resources (such as reducing training time by 50%).

[0149] It should be noted that S500 as a closed-loop execution and feedback optimization module of the supply chain, by encrypting the global strategy and issuing it to each client (such as ERP, sensor, RFID system), dynamically adjusting the local device configuration (such as sensor sampling rate, environmental monitoring accuracy), and collecting execution effect data in real time, forming a closed loop of "strategy execution - effect feedback - abnormal response - model evolution": when detecting abnormal situations such as delivery delay, carbon emission exceeding standard or equipment failure, the global strategy is regenerated within 5 minutes, and the model is updated incrementally through federated learning (such as optimizing prediction by combining extreme weather path data), ultimately realizing the dynamic response capability and adaptive evolution of the supply chain - both quickly solving abnormalities (such as path interruption recovery time < 15 minutes) and improving prediction accuracy (such as 40% improvement in extreme scenario path planning accuracy), while ensuring data privacy, significantly enhancing the efficiency and risk resistance of the supply chain.

[0150] In the embodiments of the present application, in the process of abnormality detection and global strategy regeneration, the method further comprises: determining the corresponding abnormality criterion, the abnormality criterion comprising order delivery delay rate > 10% or carbon emission exceeding standard > 15%, when detecting abnormality, calling local data of each client, and locating abnormal root equipment or path based on local data.

[0151] Specifically, the method comprises the following steps:

[0152] 1. Abnormal criterion setting and threshold definition: Threshold definition, order delivery delay rate > 10% or carbon emission exceeding > 15% triggers abnormality; Dynamic adjustment, temporarily adjust threshold according to scene (such as promotion period, extreme weather) (such as holiday delay rate tolerance increased to 15%). Thus, false positives or false negatives can be avoided, and response is only initiated in critical abnormalities.

[0153] 2. Real-time monitoring and abnormal triggering: Data comparison, compare real-time indicators with thresholds every 5 minutes, and mark abnormalities if they exceed the thresholds (such as a certain path delay rate = 12%); Classification marking, distinguish between local abnormalities (device failure) or global abnormalities (extreme weather). Thus, quick response can be achieved, triggering abnormal processing within 5 minutes and reducing the scope of influence.

[0154] 3. Client data retrieval and root cause localization: Data request, retrieve relevant client data (such as fault device log, congestion road section traffic); Root cause analysis, locate the root cause through time alignment and correlation analysis (such as "device X vibration anomaly causes order delay"). Thus, accurate positioning can be achieved to avoid over-adjustment of global strategies and only fix the problem root cause.

[0155] 4. Global strategy regeneration: Local abnormality, only adjust the affected module (such as reduce the weight of the faulty device); Global abnormality, re-run capacity allocation, path planning and federated learning optimization to generate new strategies; Root cause injection, add abnormal scenario data to the training set to optimize the model's prediction ability for similar problems. Thus, quick fixes can be achieved, generating new strategies within 5 minutes, such as switching paths within 15 minutes in extreme weather.

[0156] By accurately setting abnormal criteria (order delay rate > 10% or carbon emission exceeding > 15%), real-time monitoring and root cause localization, S500's abnormal handling process can trigger a response within 5 minutes, generate a repair strategy (such as dynamic path switching) within 15 minutes for local problems (such as device failure) or global risks (such as extreme weather), and optimize the model by injecting abnormal scenario data (such as a 40% increase in rain path planning accuracy), all while ensuring data privacy. This enables precise repair, rapid response, and continuous evolution, significantly reducing supply chain risks (delivery failure rate reduced by 60%) and improving resource utilization efficiency and risk resistance.

[0157] In the embodiments of the present application, in the process of dynamically adjusting the local device configuration, the method further comprises: when the federal strategy indicates a high-risk area, the GPS sampling rate of the logistics sensor is increased from 1 Hz to 10 Hz; the vibration sensor of the production device starts a millisecond-level high-frequency acquisition mode when it detects a frequency spectrum anomaly; the scanning interval of the warehouse RFID system is adjusted according to the inventory turnover rate to be ≤ 2 minutes when the turnover rate > 5 times / day.

[0158] The specific execution process includes the following steps:

[0159] 1. High-risk area triggers logistics sensor GPS high-frequency sampling: trigger condition, federal strategy marks a certain area as high-risk (such as congestion index > 0.6 or weather warning ≥ 3 levels); configuration adjustment, the GPS sampling rate of the logistics sensor is increased from 1 Hz (1 time / second) to 10 Hz (10 times / second), and real-time high-frequency collection of vehicle position data is performed; execution range, only for logistics equipment in high-risk areas (such as trucks in transit).

[0160] 2. Millisecond-level high-frequency collection triggered by abnormal spectrum of vibration sensor: trigger condition, the vibration sensor of the production equipment detects abnormal spectrum (such as a sudden increase in a specific frequency component, indicating bearing wear or looseness); configuration adjustment, the vibration sensor starts millisecond-level high-frequency collection mode (such as increasing the sampling rate from 100 Hz to 10 kHz), capturing subtle vibration changes; execution range, only for abnormal equipment (such as injection molding machines, CNC machine tools).

[0161] 3. Dynamic adjustment of RFID scanning interval driven by inventory turnover rate: trigger condition, the warehouse RFID system monitors that the inventory turnover rate of a certain SKU is > 5 times / day (high-frequency warehouse entry and exit); configuration adjustment, the RFID scanning interval is shortened from the default 5 minutes to ≤ 2 minutes, real-time tracking of inventory changes; execution range, only for high-turnover SKUs (such as best-selling goods or out-of-stock materials).

[0162] By dynamically adjusting the local device configuration, the supply chain realizes precise and intelligent resource adaptation at key links: in high-risk areas, the GPS sampling rate of the logistics sensor is increased to 10 Hz, the positioning accuracy is improved by 10 times, supporting millisecond-level path correction (such as avoiding sudden congestion), reducing the risk of delay; in equipment health monitoring, the vibration sensor warns of faults (such as bearing cracks) 3-5 hours in advance through 10 kHz high-frequency collection, reducing unplanned equipment downtime by 40%; in inventory management, the RFID scanning interval of high-frequency turnover SKUs is shortened to 2 minutes, the out-of-stock response speed is accelerated, the out-of-stock rate of high-value SKUs is reduced by 30%, and the over-scanning of low-turnover inventory is avoided. These adjustments through scenario-based resource configuration improve data accuracy and decision-making speed while reducing the waste of computing power, forming a closed loop of dynamic response and resource optimization, significantly enhancing the risk resistance and operational efficiency of the supply chain.

[0163] In the embodiments of the present application, the adaptive federal optimization algorithm also includes: dynamically adjusting the aggregation weight according to the client data quality, the node weight is increased by 30% when the data integrity is > 90%; the client parameters associated with high-priority orders are encrypted and isolated, an adversarial verification mechanism is introduced, the abnormal deviation of the client uploaded parameters is detected, and the weight is automatically reduced when the deviation is greater than the deviation threshold.

[0164] The specific execution process includes the following steps:

[0165] 1. Dynamic weight adjustment of data quality: First, data evaluation, quantifying the quality of client data (such as data completeness, timeliness, consistency); then weight adjustment rules, if the client data completeness is > 90%, its weight in global parameter aggregation is increased by 30% (such as original weight 1.0→1.3); corresponding execution range: all client participating in federated learning (such as ERP, Internet of Things sensors, logistics sensors).

[0166] 2. Encryption isolation and adversarial verification of high-priority order parameters: First, encryption isolation, independently encrypting and storing the client parameters associated with high-priority orders (such as key equipment weight, path model parameters), and isolating them from other ordinary order parameters. Then adversarial verification, which includes: deviation detection, comparing the client uploaded parameters with the global model mean to calculate the deviation (such as the client value of parameter X deviates from the global mean by > 15%); threshold triggering, if the deviation is greater than the threshold (such as 15%), it is determined as abnormal, and the weight of the client is automatically reduced (such as weight from 1.3→0.7). Finally, the corresponding execution range, high-priority order related clients (such as key production equipment, core logistics path).

[0167] Through the adaptive federated optimization algorithm, the supply chain realizes dynamic weight adjustment driven by data quality and secure isolation of high-priority parameters: the weight of the client with data completeness > 90% is increased by 30%, ensuring that high-quality data dominates the global model; the parameters associated with high-priority orders are encrypted and isolated, and the deviation of the parameters is detected in real time through adversarial verification (such as automatically reducing the weight when the deviation is > 15%), preventing malicious attacks or abnormal data interference; ultimately improving the credibility, security and anti-interference ability of the federated learning model (such as reducing the influence of abnormal nodes by 50%), ensuring the priority supply of key order resources while accelerating the convergence efficiency of the model, providing reliable decision support for global optimization of the supply chain.

[0168] The embodiments of the application disclose a supply chain-oriented intelligent order management system, referring to Figure 2 , comprising:

[0169] The multi-modal data acquisition module 001 extracts the corresponding order demand flow through the ERP system, the order demand flow including the order quantity, priority and delivery time constraint, collects the corresponding production equipment state through the Internet of Things sensor, the production equipment state including the equipment vibration spectrum, temperature curve and energy consumption fluctuation, acquires the corresponding logistics dynamic information through the logistics sensor, the logistics dynamic information including the transportation path coordinates, carrier speed and environment temperature and humidity, acquires the warehouse RFID system generated inventory topology map, the inventory topology map including the three-dimensional goods location and inventory distribution generated by the warehouse RFID system, aligns the order demand flow, production equipment state, logistics dynamic information and inventory topology map through the space-time synchronization protocol, and is used for acquiring the corresponding multi-modal data.

[0170] The capacity allocation scheme acquisition module 002 analyzes the correlation between the order and the equipment state based on the order demand flow and the production equipment state through the pre-set space-time graph convolution network, and is used for acquiring the corresponding capacity allocation scheme; calculates the multi-objective optimal path based on the logistics dynamic information, acquires the corresponding path planning scheme, and the multi-objective optimal path includes cost, time efficiency and carbon emission; constructs a Bayesian network based on the inventory topology map to predict the out-of-stock risk in the next 72 hours, and generates the corresponding replenishment suggestion.

[0171] The judgment module 003 is used for judging whether there is a high-priority order based on the capacity allocation scheme, when receiving a high-priority order, inserting a new task in the capacity allocation scheme through a lossless order insertion algorithm, determining the affected equipment process chain based on the new task and adjusting; based on the capacity allocation scheme, it is used for judging whether there are multiple orders competing for the same resource, if so, a fuzzy logic arbitrator is called to dynamically allocate resources, conflict resolution instructions are generated and executed; when the risk value corresponding to the path planning scheme is greater than the preset risk threshold, the standby path scheme is triggered.

[0172] The global strategy generation module 004 takes the ERP system, the Internet of Things sensor, the logistics sensor and the warehouse RFID system in the supply chain as the federated learning client, the warehouse RFID system re-trains the ST-GCN model based on the equipment state data, extracts the corresponding capacity prediction error, the logistics sensor optimizes the path model according to the path planning error, predicts the probability of carbon emission exceeding the standard, the warehouse RFID system updates the Bayesian network through the inventory topology map, and evaluates the corresponding replenishment response time; using an adaptive federated optimization algorithm to aggregate the parameters of each client, dynamically weighting the weight of the node related to the high-priority order, for generating the corresponding global strategy, the production equipment weight of the high-priority order is higher.

[0173] The updating module 005 distributes the global strategy to each client, the client dynamically adjusts the local device configuration after decryption, the local device configuration includes sensor sampling rate, environment detection accuracy; after each client executes, the execution effect data corresponding to each client is obtained, the execution effect data is uploaded, if an exception is detected, the global strategy is triggered to be regenerated, and the global model is incrementally updated through federated learning.

[0174] The embodiment of the application further discloses a supply chain-oriented intelligent order management system, comprising a processor, and a program of the supply chain-oriented intelligent order management method in any one of the above embodiments is run in the processor.

[0175] The embodiment of the application further discloses a storage medium, and the storage medium stores the program of the supply chain-oriented intelligent order management method in any one of the above embodiments.

[0176] Although the embodiments of the application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the application.

Claims

1. A smart order management method for supply chain, characterized in that, include: The order demand flow is extracted from the ERP system, including order quantity, priority, and delivery time constraints. The status of the corresponding production equipment is collected through IoT sensors, including equipment vibration spectrum, temperature curve, and energy consumption fluctuation. The corresponding logistics dynamic information is obtained through logistics sensors, including transportation path coordinates, vehicle speed, and ambient temperature and humidity. The inventory topology map generated by the warehouse RFID system is obtained, including the three-dimensional location of goods and inventory distribution generated by the warehouse RFID system. The order demand flow, the production equipment status, the logistics dynamic information, and the inventory topology map are aligned through a spatiotemporal synchronization protocol to obtain the corresponding multimodal data. Based on the order demand flow and the production equipment status, the correlation between orders and equipment status is analyzed through a pre-set spatiotemporal graph convolutional network to obtain the corresponding capacity allocation scheme; Based on the logistics dynamic information, calculate the multi-objective optimal path and obtain the corresponding path planning scheme. The multi-objective optimal path includes cost, timeliness, and carbon emissions. Based on the inventory topology map, construct a Bayesian network to predict the stockout risk in the next 72 hours and generate corresponding replenishment suggestions. Based on the capacity allocation scheme, determine whether a high-priority order is received. When a high-priority order is received, insert a new task into the capacity allocation scheme using a lossless order insertion algorithm. Based on the new task, determine the affected equipment process chain and make adjustments. Based on the capacity allocation scheme, determine whether multiple orders are competing for the same resource. If so, call the fuzzy logic arbitrator to dynamically allocate resources, generate conflict resolution instructions, and execute them. When the risk value corresponding to the route planning scheme is greater than the preset risk threshold, the backup route scheme is triggered. The ERP system, IoT sensors, logistics sensors, and warehouse RFID system in the supply chain are used as federated learning clients. The warehouse RFID system retrains the ST-GCN model based on equipment status data and extracts the corresponding capacity prediction error. The logistics sensors optimize the path model based on the path planning error and predict the probability of carbon emission exceeding the standard. The warehouse RFID system updates the Bayesian network through the inventory topology map and evaluates the corresponding replenishment response time. An adaptive federated optimization algorithm is used to aggregate parameters from each client, dynamically weight the nodes related to high-priority orders, and generate corresponding global strategies. Production equipment for high-priority orders has a higher weight. The global policy is distributed to each client. After decryption, the client dynamically adjusts its local device configuration, which includes sensor sampling rate and environmental detection accuracy. After execution by each client, the execution effect data corresponding to each client is obtained and uploaded. If an anomaly is detected, the global policy is regenerated and the global model is incrementally updated through federated learning.

2. The intelligent order management method for supply chain as described in claim 1, characterized in that, In the analysis of spatiotemporal graph convolutional networks, the methods also include: The state of production equipment is mapped to a spatiotemporal graph structure, where nodes represent production equipment and edge weights are dynamically calculated based on the process dependencies between equipment and the real-time load difference. The short-term fluctuation patterns and long-term trend characteristics of production equipment status are extracted by multi-level spatiotemporal convolution kernels, and dynamic capacity allocation weights are generated by combining order priority. When a device's health is detected to be declining, the allocation weight of its associated orders is automatically reduced, and the standby device activation protocol is triggered.

3. The intelligent order management method for supply chain as described in claim 1, characterized in that, In the process of dynamically allocating resources by invoking a fuzzy logic arbitrator, the three-dimensional evaluation indicators for resource conflicts include order priority, equipment health, and delivery time margin. The methods also include: Construct a fuzzy rule base so that when multiple orders compete for the same device, priority is given to allocating orders with a priority > 0.8 and a device health score > 0.

7. If the priority difference between conflicting orders is less than 0.2, the carbon emission impact factors corresponding to the conflicting orders are compared, and the option with the lowest carbon emission increment is selected. When generating conflict resolution instructions, reserve a spare resource window simultaneously.

4. The intelligent order management method for supply chain as described in claim 1, characterized in that, The methods used in calculating the risk value of a path planning scheme also include: A corresponding risk probability model is constructed based on real-time traffic flow, weather warning data, and historical accident statistics from logistics dynamic information. Obtain the congestion index and weather disaster warning level corresponding to the route. When the congestion index of the route is greater than 0.6 or the weather disaster warning level is greater than or equal to level 3, the route is marked as a high-risk route. When obtaining the corresponding path planning scheme, the corresponding backup path scheme is generated simultaneously. The backup path scheme pre-generates multiple candidate paths and dynamically switches candidate paths according to the real-time risk value, with a switching response time of less than 15 seconds.

5. The intelligent order management method for supply chain as described in claim 1, characterized in that, The method also includes the following steps in the process of anomaly detection and global policy regeneration: The corresponding anomaly criteria are determined, including order delivery delay rate > 10% or carbon emission exceedance > 15%. When an anomaly is detected, local data from each client is retrieved, and the root cause device or path of the anomaly is located based on the local data. When incrementally updating the global model, the decision branches of the historical best version are retained.

6. The intelligent order management method for supply chain as described in claim 1, characterized in that, The methods also include: When federal policy designates high-risk areas, the GPS sampling rate of logistics sensors is increased from 1Hz to 10Hz; vibration sensors on production equipment activate millisecond-level high-frequency acquisition mode when they detect abnormal spectrum; and the scanning interval of the warehouse RFID system is adjusted according to the inventory turnover rate so that the scanning interval is ≤2 minutes when the turnover rate is >5 times / day.

7. The intelligent order management method for supply chain as described in claim 1, characterized in that, Adaptive federated optimization algorithms, and other methods include: Aggregate weights are dynamically adjusted based on client data quality, with nodes having a data integrity of >90% receiving a 30% weight increase; client parameters associated with high-priority orders are encrypted and isolated, and an adversarial verification mechanism is introduced to detect abnormal deviations in parameters uploaded by the client, automatically reducing weight if the deviation exceeds the deviation threshold.

8. A smart order management system for the supply chain, characterized in that, include: The multimodal data acquisition module extracts the corresponding order demand flow through the ERP system. The order demand flow includes order quantity, priority, and delivery time constraints. It collects the corresponding production equipment status through IoT sensors. The production equipment status includes equipment vibration spectrum, temperature curve, and energy consumption fluctuation. It acquires the corresponding logistics dynamic information through logistics sensors. The logistics dynamic information includes transportation path coordinates, vehicle speed, and ambient temperature and humidity. It acquires the inventory topology map generated by the warehouse RFID system. The inventory topology map includes the three-dimensional goods location and inventory distribution generated by the warehouse RFID system. The order demand flow, the production equipment status, the logistics dynamic information, and the inventory topology map are aligned through a spatiotemporal synchronization protocol to acquire the corresponding multimodal data. The capacity allocation scheme acquisition module analyzes the correlation between orders and equipment status using a pre-set spatiotemporal graph convolutional network based on the order demand flow and the production equipment status, in order to obtain the corresponding capacity allocation scheme. Based on the logistics dynamic information, calculate the multi-objective optimal path and obtain the corresponding path planning scheme. The multi-objective optimal path includes cost, timeliness, and carbon emissions. Based on the inventory topology map, a Bayesian network is constructed to predict the stockout risk in the next 72 hours and generate corresponding replenishment suggestions. The judgment module is used to determine whether a high-priority order is received based on the capacity allocation scheme. When a high-priority order is received, a new task is inserted into the capacity allocation scheme through a lossless order insertion algorithm. Based on the new task, the affected equipment process chain is determined and adjusted. The module is also used to determine whether multiple orders are competing for the same resource based on the capacity allocation scheme. If so, a fuzzy logic arbitrator is called to dynamically allocate resources, generate conflict resolution instructions, and execute them. When the risk value corresponding to the route planning scheme is greater than the preset risk threshold, the backup route scheme is triggered. The global strategy generation module uses the ERP system, IoT sensors, logistics sensors, and warehouse RFID system in the supply chain as federated learning clients. The warehouse RFID system retrains the ST-GCN model based on equipment status data and extracts the corresponding capacity prediction error. The logistics sensors optimize the path model based on the path planning error and predict the probability of carbon emission exceeding the standard. The warehouse RFID system updates the Bayesian network through the inventory topology map and evaluates the corresponding replenishment response time. An adaptive federated optimization algorithm is used to aggregate parameters from each client and dynamically weight the nodes related to high-priority orders to generate corresponding global strategies. The production equipment of high-priority orders has a higher weight. The update module distributes the global policy to each client. After decryption, the client dynamically adjusts the local device configuration, which includes sensor sampling rate and environmental detection accuracy. After each client executes the code, it obtains the execution effect data corresponding to each client and uploads the execution effect data. If an anomaly is detected, the global policy is triggered to be regenerated, which is used to incrementally update the global model through federated learning.

9. A smart order management system for the supply chain, characterized in that, Includes a processor in which a program for a supply chain-oriented intelligent order management method as described in any one of claims 1-7 is running.

10. A storage medium, characterized in that, The program stores a supply chain-oriented intelligent order management method as described in any one of claims 1-7.

Images