Supply chain production, transportation and sales cooperative scheduling method and system, and storage medium

Abstract

The invention relates to the technical field of supply chain scheduling, and discloses a supply chain production-transportation-sales cooperative scheduling method and system, and a storage medium. The system comprises a production data acquisition module, a transportation data acquisition module and a sales data acquisition module which are respectively used for acquiring plans, schedules, inventory, routes, states, costs, orders, prediction and market demand data of corresponding links; the production, transportation and sales association modeling module is used for establishing an association model based on the collected data and generating a collaborative scheduling prediction result; the dynamic feature extraction module combines historical and real-time data to extract dynamic features and identify a scheduling demand spatial-temporal change rule; the resource optimization scoring module calculates the resource adjustment priority and the allocation proportion according to the dynamic characteristics and the prediction result, and generates an optimization scheduling scheme; and the self-adaptive scheduling execution module adjusts production, transportation and sales scheduling in real time based on the optimization scheme to realize collaborative adaptive change. The system can break data islands and improve supply chain cooperation efficiency and strain capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supply chain scheduling, in particular to a supply chain production-distribution-sales collaborative scheduling method and system and a storage medium. BACKGROUND

[0002] In the modern business environment, the complexity of the supply chain system continues to increase as the market scale expands, and the coordination difficulty between the three core links of production, transportation and sales gradually increases. Under the traditional mode, the production link often only formulates production scheduling schemes according to internal plans, lacks real-time response to changes in market demand, and leads to overstocking of some products, while hot products frequently run out of stock. The transportation link often uses fixed routes and periodic arrangements, and cannot dynamically adjust to the production progress and the urgency of sales orders, which not only increases unnecessary transportation costs, but also may affect customer satisfaction due to delayed delivery. The sales link is difficult to quickly pass the information to the production and transportation links after obtaining the market demand data, resulting in a significant disconnection between demand prediction and actual production and distribution.

[0003] The data collection and processing of each link are relatively independent, and the production data, transportation data and sales data are stored in different systems, lacking effective integration and correlation analysis mechanism. When market demand, raw material supply, transportation conditions and other factors change suddenly, it is difficult for enterprises to quickly identify the change pattern of scheduling demand, and to timely adjust the resource allocation strategy, thereby affecting the operation efficiency of the entire supply chain. For example, if the transportation link cannot obtain the urgency of the sales orders and the inventory situation of the production end in time due to the sudden outbreak of an epidemic in a certain region, it may choose to take a detour on a less efficient route, or prioritize the delivery of non-urgent orders, ultimately leading to the late delivery of some orders. At the same time, the traditional scheduling method lacks comprehensive utilization of historical data and real-time data, and cannot accurately predict the collaborative scheduling effect, and the subjective factors in setting the priority of resource adjustment are strong, which cannot realize the optimal allocation of resources, further restricting the improvement of the overall competitiveness of the supply chain. SUMMARY

[0004] The present application aims to provide a supply chain production-distribution-sales collaborative scheduling method and system and a storage medium to solve the problems raised in the background.

[0005] To achieve the above-mentioned purpose, the present application provides a supply chain production-distribution-sales collaborative scheduling system, which comprises: a production data acquisition module for acquiring production plan data, production progress data and inventory data; a transportation data acquisition module for acquiring transportation route data, transportation status data and transportation cost data; a sales data acquisition module for acquiring sales order data, sales forecast data and market demand data; The production-transportation-sale correlation modeling module is configured to establish a correlation model between production, transportation and sale based on production plan data, production progress data and inventory data obtained by the production data acquisition module, transportation route data, transportation status data and transportation cost data obtained by the transportation data acquisition module, and sales order data, sales forecast data and market demand data obtained by the sales data acquisition module, and generate a collaborative scheduling prediction result. The dynamic feature extraction module is configured to extract dynamic features of production, transportation and sale based on historical data and real-time data, and identify a spatio-temporal variation law of scheduling demand. The resource optimization scoring module is configured to calculate a resource adjustment priority and a resource allocation ratio based on the dynamic features extracted by the dynamic feature extraction module and the collaborative scheduling prediction result generated by the production-transportation-sale correlation modeling module, and generate an optimized scheduling scheme. The adaptive scheduling execution module is configured to execute real-time scheduling adjustment of production, transportation and sale based on the optimized scheduling scheme generated by the resource optimization scoring module, and ensure that the collaborative scheduling is adaptive to changes.

[0006] Preferably, the production data acquisition module specifically comprises: The production plan data, the production progress data and the inventory data are collected, and data cleaning, outlier removal and production feature extraction are performed. The production area is divided into multiple units, and the production state of each unit at different times is counted to construct a production state heat map. The production progress data is analyzed by applying a production trajectory clustering algorithm to identify typical production modes and key production paths. Based on the production trajectory clustering results, a production flow prediction function is established to predict the production flow of each key production path at a specific time point in combination with time factors.

[0007] Preferably, the production-transportation-sale correlation modeling module specifically comprises: The production data, the transportation data and the sales data are obtained, and data processing and fusion are performed. Based on multi-source data, an activity feature vector of production-transportation-sale activities is constructed, including activity intensity, activity type and participating resource information at a specific time point. A dual-mode scheduling adaptation system including a normal mode and an event mode is constructed, and an event influence factor triggering mode switching is defined. Based on historical scheduling data and production-transportation-sale activity features, a scheduling-activity correlation model is trained to output a predicted collaborative scheduling distribution.

[0008] Preferably, the dynamic feature extraction module specifically comprises: Based on historical scheduling data, a time-varying supply chain network graph model is constructed to represent the dynamic characteristics of the supply chain topology changing over time. The time-varying supply chain network graph is converted into a low-dimensional feature representation by using a time series graph embedding algorithm, and the reconstruction error and time consistency are optimized. Based on the graph embedding result, the typical spatio-temporal pattern of the scheduling distribution is analyzed, and the periodic pattern and abnormal pattern are extracted. A node load time series prediction model is established to identify a set of key nodes in the supply chain and generate a predicted graph of the future supply chain topology load distribution.

[0009] Preferably, the resource optimization scoring module specifically includes: A supply chain topology elasticity scoring system is established to evaluate the adaptability of the supply chain topology to scheduling changes. Based on the supply chain network graph and the collaborative scheduling prediction result, a set of key nodes and a set of redundant paths are identified. For the predicted scheduling changes, the adjustment priority score of each node in the topology is calculated. Based on the node adjustment priority score, the optimal allocation ratio of the supply chain resources is calculated to solve the constraint optimization problem.

[0010] Preferably, the adaptive scheduling execution module specifically includes: A resource reallocation control system is constructed to convert the resource allocation scheme into specific resource allocation instructions. Based on the resource allocation instructions, dynamic adjustment of production resources, transportation resources, and sales resources is realized, and a smooth transition strategy is adopted to avoid disruptions caused by sudden changes. Based on the resource adjustment result, the scheduling strategy is optimized to find a set of scheduling paths that optimize the overall performance of the supply chain. After implementing the scheduling adjustment, a real-time monitoring and feedback mechanism is established to trigger the adaptive adjustment process based on performance deviations.

[0011] Preferably, the scheduling-activity association model adopts a multi-layer perceptron structure, including: The input layer receives the activity feature vector. The hidden layer contains two sub-layers and introduces a time attention mechanism to assign different weights to activity features at different time points. The output layer outputs the predicted scheduling distribution, and the number of neurons is equal to the number of supply chain region divisions.

[0012] Preferably, the supply chain topology elasticity score is composed of the following indicators: Node utilization balance, evaluating the balance of load distribution of each node in the supply chain; The number of redundant paths evaluates the availability of alternative paths when a node fails in the supply chain. Key node connectivity, evaluating the connection status of key nodes in the supply chain.

[0013] Preferably, the present application further comprises a supply chain production, transportation and sales collaborative scheduling method, which comprises all the modules and method processes of the supply chain production, transportation and sales collaborative scheduling system as described above.

[0014] Preferably, the present application further comprises a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement a supply chain production, transportation and sales collaborative scheduling method as described above.

[0015] Compared with the prior art, the present application has the beneficial effects that: By setting the production data acquisition module, the transportation data acquisition module and the sales data acquisition module, the key data of each link of production, transportation and sales can be comprehensively obtained, the situation of data island of each link in the traditional mode is broken, and the comprehensive integration and sharing of data are realized. The production link can more accurately formulate the production plan by obtaining the sales order data, the sales forecast data and the market demand data, so as to avoid the problems of inventory accumulation or shortage caused by blind production scheduling; the transportation link can flexibly adjust the transportation route and the transportation state based on the production progress data, the inventory data and the urgency of the sales order, so as to reduce unnecessary transportation cost and at the same time guarantee the timely delivery of orders; the sales link can obtain the production progress and the transportation state information in time, provide more accurate delivery time estimation to the customers, and improve the customer experience.

[0016] The production, transportation and sales correlation modeling module establishes a correlation model based on the multidimensional data obtained by each acquisition module, can generate a collaborative scheduling prediction result more in line with actual demand, and helps enterprises to predict possible scheduling problems in advance. For example, by correlating the relationship among production progress, transportation cost and sales demand, the collaborative imbalance problem that may be caused by production delay or high transportation cost can be identified in advance, and countermeasures can be taken in advance. The dynamic feature extraction module extracts the dynamic features of production, transportation and sales in combination with historical data and real-time data, can accurately identify the spatio-temporal variation law of scheduling demand, so that the enterprise can acutely capture the influence of market demand fluctuation, transportation road condition change, production equipment failure and other sudden situations on scheduling work, and provide clear direction guidance for subsequent resource adjustment.

[0017] The resource optimization scoring module calculates resource adjustment priority and resource allocation proportion according to the dynamic characteristics and the collaborative scheduling prediction result, and generates an optimized scheduling scheme, thereby avoiding the disadvantage that resource allocation in the traditional scheduling depends on subjective judgment. The module can reasonably allocate production resources, transportation resources and sales resources according to actual conditions. For example, when the supply of raw materials is tight, raw materials are preferentially allocated to product production lines with strong market demand and large profit space; when transportation resources are limited, transportation vehicles are allocated according to the order urgency and the optimal transportation cost principle, so as to maximize the utilization efficiency of resources. The adaptive scheduling execution module can execute scheduling adjustment in real time based on the optimized scheduling scheme, so as to ensure that the collaborative scheduling can quickly adapt to changes in internal and external environments. When a sudden situation occurs, such as temporary interruption of transportation route or sudden increase of sales order, the system can quickly adjust production scheduling, transportation arrangement and sales strategy, maintain stable operation of each link of the supply chain, and avoid supply chain interruption or significant decrease in efficiency caused by sudden situations. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 a timing diagram of the supply chain production-transportation-sales collaborative scheduling system described in the present application; Figure 2 a flowchart of the working process of the production data acquisition module; Figure 3 a flowchart of the working process of the production-transportation-sales correlation modeling module; Figure 4 a flowchart of the working process of the dynamic characteristic extraction module; Figure 5 a flowchart of the working process of the adaptive scheduling execution module. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] Please refer to Figure 1 The present application provides a supply chain production-transportation-sales collaborative scheduling system, which comprises a production data acquisition module, a transportation data acquisition module, a sales data acquisition module, a production-transportation-sales correlation modeling module, a dynamic characteristic extraction module, a resource optimization scoring module and an adaptive scheduling execution module. The specific implementation is as follows: The production data acquisition module obtains production plan data, production progress data, and inventory data; the transportation data acquisition module obtains transportation route data, transportation status data, and transportation cost data; the sales data acquisition module obtains sales order data, sales forecast data, and market demand data. The production, transportation, and sales correlation modeling module establishes a correlation model between production, transportation, and sales based on the above multi-source data, and generates a collaborative scheduling prediction result. The dynamic feature extraction module extracts the dynamic features of production, transportation, and sales based on historical data and real-time data, and identifies the spatiotemporal variation law of scheduling demand. The resource optimization scoring module calculates the resource adjustment priority and resource allocation ratio based on the dynamic features and the collaborative scheduling prediction result, and generates an optimized scheduling scheme. The adaptive scheduling execution module executes real-time scheduling adjustment of production, transportation, and sales based on the optimized scheduling scheme, ensuring that the collaborative scheduling adapts to changes.

[0021] The system realizes the overall optimization of the supply chain through the collaborative operation of multiple modules. The production data acquisition module monitors the production status in real time, the transportation data acquisition module tracks the logistics dynamics, and the sales data acquisition module analyzes market demand. The production, transportation, and sales correlation modeling module integrates multi-source data and constructs a correlation model to predict scheduling demand. The dynamic feature extraction module extracts key patterns from time series data, the resource optimization scoring module evaluates supply chain elasticity and generates resource allocation strategies, and the adaptive scheduling execution module dynamically adjusts resource allocation and optimizes scheduling paths. The system realizes the whole-chain collaboration from data acquisition to scheduling execution through an automated process, improving the response speed and resource utilization efficiency of the supply chain.

[0022] Embodiment 1: see Figure 2 In the implementation process of the supply chain production, transportation, and sales collaborative scheduling system, the production data acquisition module undertakes the core functions of basic data acquisition and processing. The specific implementation of this module is based on a fictional but typical electronic product manufacturing enterprise. Its production system includes a final assembly workshop, a mainboard production workshop, a screen bonding workshop, a battery assembly workshop, and multiple raw material and finished product warehouses. These physical areas are divided into twenty-eight independent production units, each equipped with Internet of Things data acquisition equipment to record production plan data, production progress data, and inventory data in real time.

[0023] Production plan data is sourced from the enterprise's ERP system, containing product model, planned output, start time, deadline, bill of materials, and other information. Production progress data is collected through the on-site MES system of the workshop, recording the completion quantity, equipment status, time consumption, and yield rate of each work order in real time. Inventory data comes from the WMS system, dynamically updating the quantity, location, and status of raw materials, work-in-process, and finished products in each warehouse. After these raw data are transmitted to the production data acquisition module through the enterprise service bus, they are first cleaned. The cleaning process includes removing duplicate records caused by network transmission, correcting time alignment problems caused by inconsistent time stamp formats in different systems, and filling in data missing due to temporary equipment failure. Statistical process control methods are used to remove outliers, identifying and removing data points that deviate significantly from the normal range, such as abnormal spikes or zeros in single-hour output.

[0024] After data cleaning is complete, the module extracts production characteristics. From production plan data, it extracts planned completion rate, capacity utilization rate, and other characteristics; from production progress data, it extracts actual output rate, time efficiency, and equipment comprehensive efficiency; from inventory data, it extracts inventory turnover rate and inventory age structure. These characteristics are quantified as standardized indicators, forming a digital representation of production status.

[0025] Based on these processed data, the module divides the entire production area into twenty-eight units, each corresponding to an independent production line or warehouse. The production status of each unit at different time points is statistically analyzed, with dimensions including unit time output, equipment operating status, and material consumption speed. The statistical results are used to construct a production status heat map, with the horizontal axis representing time series and the vertical axis representing production unit number. The color depth ranges from light to dark, representing the intensity of production activities from low to high. Through this visualization method, active and bottleneck areas in different time periods can be identified intuitively. For example, from 9 am to 11 am, multiple units in the general assembly workshop will show concentrated distribution of dark color, indicating the production peak period; while during the night, the screen fitting workshop may continue to show moderate activity, indicating that the workshop implements continuous production mode.

[0026] The production trajectory clustering algorithm is applied to analyze the production progress data. The algorithm treats the manufacturing process of each product as a trajectory, which is composed of a series of consecutive production event nodes, such as material delivery, component assembly, quality testing, finished product storage, etc. The algorithm uses a density-based spatial clustering method to identify high-density trajectory groups with similar paths. By analyzing the data from the past three months, the algorithm identifies six typical production patterns. One of the patterns is labeled as the "standard motherboard assembly pattern", which is characterized by the path: taking components from the warehouse, going through the four core processes of soldering, reflow soldering, testing, and assembly, and finally flowing into the final assembly workshop. Another pattern is the "emergency order processing pattern", which significantly shortens the path and skips some testing steps, directly entering the final assembly. The key production paths are determined by analyzing the trajectory passing frequency and time consumption. The soldering process in the motherboard workshop and the final testing process in the final assembly workshop are identified as critical paths, as more than 80% of the trajectories must pass through these two nodes, and their processing speed directly affects the overall production cycle.

[0027] Based on the production trajectory clustering results, a production flow prediction function is established. The function considers multiple factors, including historical same-period flow, current work-in-process quantity, equipment health status, planned production scheduling, etc. The function uses time series analysis combined with machine learning methods for modeling. For each key production path, such as the soldering process in the motherboard workshop, the prediction function receives the actual processing volume in the past 24 hours, the planned production volume in the next 8 hours, the equipment warning status, etc. as input, and outputs the estimated number of workpieces passing through at future time points. The prediction process takes into account the differences between weekdays and weekends, seasonal demand fluctuations, and known planned maintenance events. The prediction results are represented in time series form, for example, the prediction shows that the flow of the soldering process will peak at 10 am the next day, with 350 motherboards passing through per hour. These prediction data provide key inputs for subsequent production-distribution-sales coordination scheduling, enabling the system to foresee production bottlenecks and make resource adjustments in advance. The entire implementation process embodies the complete chain from raw data collection to valuable information extraction, laying a data foundation for supply chain coordination scheduling through multi-level data processing and analysis.

[0028] Example 2: see Figure 3The implementation of the production-distribution-sales correlation modeling module is based on the actual operation scenario of a large consumer goods manufacturing enterprise that produces home appliance products and has multiple production bases, distribution centers, and retail networks. The module first obtains information from three independent data sources: production data from the manufacturing execution system of each factory, including real-time capacity utilization, work-in-process inventory, production line status, etc.; transportation data from the logistics management platform, including vehicle location, transportation time efficiency, transportation cost, warehouse turnover rate, etc.; and sales data from the customer relationship management system and e-commerce platform, covering order details, forecasted sales, promotional activities, market trends, etc. After these heterogeneous data are transmitted to the modeling module through the enterprise data bus, systematic data processing and fusion are performed.

[0029] The data processing stage performs standardized conversion on multi-source information. The time dimension is converted to UTC timestamp and adjusted to the local time zone; the measurement units are converted to standard international units; and the data format is reconstructed in accordance with the unified JSON specification. Data fusion is achieved through entity analysis technology, based on product SKU codes, order numbers, and transportation order numbers, etc. business identifiers, data records from different systems are associated to the same business entity. For example, a retail order is associated to a specific factory production batch, a designated warehouse inventory unit, and a corresponding transportation vehicle dispatch order.

[0030] Based on the fused multi-source data, the module constructs production-distribution-sales activity feature vectors. Each feature vector corresponds to a fifteen-minute time window and contains three types of core information: activity intensity is represented by quantitative indicators such as production man-hours, transportation ton-kilometers, and sales order amount; activity type is identified by classification coding, distinguishing between regular production, emergency replenishment, promotional transportation, normal sales, and other different business scenarios; and participating resource information details the resource identifiers involved, such as production line numbers, transportation vehicle IDs, and salesperson codes. These vectors form a time series data stream that accurately depicts the dynamic running state of the supply chain system.

[0031] A dual-mode scheduling pre-adaptation system is constructed, including normal mode and event mode. The normal mode corresponds to the daily operation state of the supply chain, and the event mode is used to handle unexpected situations. The system defines thirteen core indicators as event impact factors, including order satisfaction rate fluctuation coefficient, transportation delay index, and capacity abnormal deviation value. These factors are updated through real-time data stream calculation, and when any factor exceeds the preset threshold for two consecutive periods, the system automatically triggers mode switching. For example, when a sudden increase in transportation delay index is detected and continues to exceed the threshold, indicating that the logistics network is abnormal, the system immediately switches to event mode and starts emergency scheduling plans.

[0032] Based on the accumulated historical scheduling data and the characteristics of production and distribution activities, a scheduling-activity association model is trained. This model uses a deep learning architecture and is trained using historical data from the past eighteen months. The input features in the training samples include a feature vector composed of activity intensity, type, and resource information, and the output label is the actual resource scheduling decision distribution. The model learns the complex mapping relationship between different activity patterns and optimal scheduling strategies through a large number of samples. The trained model can output a predicted scheduling distribution for the future time period based on the real-time activity feature vector, which represents the required resource allocation proportion for each supply chain node in the form of probability. For example, the model may predict that the distribution center in the East China region needs to increase the inventory allocation by twenty percent, while the factory in South China needs to reduce the production output by fifteen percent. This prediction provides quantitative decision-making basis for the coordinated scheduling of the entire supply chain system, enabling the system to optimize resource allocation based on data-driven methods.

[0033] Example 3: Referring to Figure 4 The implementation of the dynamic feature extraction module and the resource optimization scoring module is based on a nationwide retail enterprise's supply chain network, which includes multiple regional distribution centers, city warehouses, and front-end stores. The dynamic feature extraction module first constructs a time-varying supply chain network graph model based on historical scheduling data. The model abstracts supply chain entities as nodes, including suppliers, factories, regional warehouses, distribution centers, and retail stores, totaling one hundred and twenty nodes. The logistics relationships between nodes are abstracted as edges, including transportation routes, inventory allocation paths, and information flow channels, totaling more than two hundred edges. The dynamic characteristics of the network graph are represented by time series snapshots, each corresponding to a four-hour time window, recording the state attributes of each node such as inventory level, processing capacity, order backlog, and the flow attributes of each edge such as transportation volume, cost, and time efficiency. Node attributes include capacity limits, current load rates, processing speeds, etc.; edge attributes include transportation capacity, actual flow, cost coefficients, etc. Through consecutive time snapshots, the model can fully capture the dynamic evolution process of the supply chain network over time.

[0034] The time-varying supply chain network graph is converted into a low-dimensional feature representation using a time series graph embedding algorithm. This algorithm uses a recurrent neural network-based architecture and is implemented through an encoder-decoder framework. The encoder compresses the graph structure information at each time step into a fixed-dimensional vector representation, and the decoder attempts to reconstruct important features of the original graph from the embedded vector. The optimization objective includes two aspects: the reconstruction error requires the embedded vector to maximize the preservation of the topological characteristics and node attribute information of the original graph; the time series consistency constraint requires the embedded vector to change smoothly between adjacent time steps, reflecting the continuity of network evolution. The embedding process maps each node at each time step to a 128-dimensional vector, which captures both the node's static attributes and its dynamic behavior characteristics in the network.

[0035] Based on the graph embedding results, typical spatio-temporal patterns of the dispatch distribution are analyzed. Clustering methods in unsupervised learning are adopted to analyze the embedding vectors, and identify the repeatedly appearing pattern categories. Periodic patterns are characterized by a 24-hour cycle, such as the surge in the morning delivery center's outbound volume and the evening aggregation of store sales data. Abnormal patterns include vector features corresponding to sudden demand surges, transportation disruptions, or inventory anomalies. The identification of these patterns helps understand the regularity and abnormal situations of the supply chain network behavior.

[0036] A set of key nodes in the supply chain is identified, and a node load time series prediction model is established. The identification of key nodes is based on the comprehensive evaluation of multiple network centrality indicators, including degree centrality, closeness centrality, and betweenness centrality. Nodes with high degree centrality represent connections to many other nodes; nodes with high closeness centrality indicate shorter average distances to other nodes; and nodes with high betweenness centrality indicate that a large number of shortest paths pass through them. By weighted calculation of these indicators, key nodes such as national central warehouses, regional hub warehouses, and major city distribution centers are identified. For these key nodes, a load time series prediction model is established, which considers multiple variables such as historical load data, seasonal factors, promotional activity effects, and external environmental factors. The model outputs the load prediction values of each key node within the next 72 hours, generating a supply chain topology load distribution prediction graph. This graph visualizes the expected load levels of each node in the network at each time point in the future in the form of a heat map.

[0037] The resource optimization scoring module establishes a supply chain topology resilience scoring system. The resilience score is composed of three core indicators: node utilization balance evaluates the uniformity of load distribution in the network, calculated as the inverse function of node load variance; redundant path number measures the network's ability to maintain connectivity when some nodes or edges fail; and key node connectivity assesses the accessibility of important nodes to other nodes. The resilience score calculation formula is:

[0038] where: represents the resilience score, denotes the standard deviation of node load, represents the mean of node load, is the number of redundant paths, is the total number of paths, is the actual number of connections of key nodes, is the theoretical maximum number of connections of key nodes, , , are the weight coefficients of each indicator, configured according to the characteristics of the supply chain.

[0039] Based on the supply chain network graph and the collaborative scheduling prediction results, a set of critical nodes and a set of redundant paths are identified. The critical nodes are determined by network centrality analysis and business importance. The redundant paths are identified by path similarity analysis and functional equivalence evaluation. For the predicted scheduling changes, an adjustment priority score is calculated for each node in the topology. The score calculation is based on the node's elasticity contribution, current load rate, predicted load change amplitude, and impact on the overall network performance. A multi-criteria decision analysis method is used to consider these factors and generate a priority score between 0 and 100 for each node.

[0040] Based on the node adjustment priority score, the optimal allocation proportion of supply chain resources is calculated. This translates into a constrained optimization problem, with the objective function being to maximize the overall network elasticity score, the decision variables being the resource allocation amounts for each node, and the constraint conditions including resource total amount limits, time window constraints, and capacity upper limits. The solution uses a linear programming method to output the resource allocation proportions for each node, guiding actual scheduling decisions. The entire implementation process provides a systematic solution for supply chain network elasticity management and resource allocation through quantitative analysis and optimization calculation.

[0041] Example 4: Referring to Figure 5 The implementation of the adaptive scheduling execution module is based on a global supply chain network of a multinational manufacturing enterprise that produces industrial equipment, with manufacturing bases in North America, Europe, and Asia, multiple regional distribution centers, and hundreds of suppliers. The module first builds a resource reallocation control system that receives optimized scheduling schemes from the resource optimization scoring module. The schemes are transmitted in XML format, containing structured information such as resource type, allocation amount, target node, time window, etc. After parsing these schemes, the system generates specific resource allocation instruction sets. The instruction sets cover three categories: production resource adjustment instructions, transportation resource scheduling instructions, and sales resource allocation instructions. Production resource adjustment instructions include device start-stop commands, shift adjustment notifications, and personnel allocation plans; transportation resource scheduling instructions include vehicle route changes, transportation priority adjustments, and carrier allocation updates; sales resource allocation instructions involve inventory redistribution, promotion strategy adjustments, and order priority modifications. These instructions are distributed to various execution systems through an enterprise service bus, including manufacturing execution systems, transportation management systems, and customer relationship management systems.

[0042] Based on the resource allocation instructions, the module implements dynamic adjustment of three types of resources. In terms of production resource adjustment, when the instructions require an increase in the output of a certain type of equipment, the system will coordinate related factories to adjust production schedules, allocate additional raw material inventory, and adjust production line pace. During the execution of transportation resource adjustment, the system will recalculate the optimal transportation path based on the new routing instructions, allocate available transportation tools, and adjust the loading plan. When adjusting sales resources, the system will modify the inventory levels of each distribution center and update the order fulfillment strategy based on the inventory reallocation instructions. All adjustments use a smooth transition strategy, such as gradually increasing production capacity by five percent per hour until reaching the target value for production capacity increases, using a segmented switching method for transportation route changes by testing the feasibility of the new route before fully implementing it, and using a batch transfer method for inventory allocation to avoid disrupting normal operations. This gradual adjustment method minimizes the risk of business disruption.

[0043] Based on the results of resource adjustment, the module optimizes the scheduling strategy to find the optimal combination of scheduling paths for the overall performance of the supply chain. The optimization process considers multiple objectives: minimizing total cost, maximizing order fulfillment rate, and shortening average response time. The system uses a multi-objective optimization algorithm to analyze all possible scheduling path combinations and evaluate the comprehensive performance indicators of each combination. The optimization results are a set of Pareto optimal solutions, providing multiple feasible scheduling schemes for decision-makers to choose from. The final determined scheduling path combination is updated to all related systems to ensure that the entire supply chain executes a unified scheduling strategy.

[0044] After implementing scheduling adjustments, the module establishes a real-time monitoring and feedback mechanism. The monitoring system continuously tracks key performance indicators, including order completion time, inventory turnover rate, transportation punctuality rate, and capacity utilization rate. These actual values are compared with expected values to calculate performance deviations. When the deviation exceeds the preset threshold, the system automatically triggers an adaptive adjustment process. This process re-runs the resource optimization scoring and scheduling execution process, forming a closed-loop control. The feedback mechanism ensures that the system can continuously adapt to external environmental changes and internal state fluctuations.

[0045] The scheduling-activity association model uses a multi-layer perceptron structure, with the input layer receiving a 256-dimensional production, transportation, and sales activity feature vector. The hidden layer includes two sub-layers, the first with 512 neurons and the second with 256 neurons, both using ReLU activation functions. The model introduces a time attention mechanism that can automatically learn the importance weights of activity features at different time points, allowing the model to focus more on feature information at key time points. The output layer has the same number of neurons as the number of supply chain region divisions, which is 128 in this case, and uses a Softmax activation function to output the predicted scheduling distribution. The model's training uses the adaptive moment estimation algorithm, with a classification cross-entropy loss function, and updates network parameters through the backpropagation algorithm.

[0046] Table 1: Resource Allocation Instruction Execution Status Tracking Table,

[0047] Referring to Table 1, the status tracking table is updated in real-time, providing managers with a complete view of instruction execution. The execution status of each instruction is visually displayed through color coding: green for completed, yellow for in execution, red for delayed or abnormal, and gray for not yet started. The execution progress field displays the current completion in percentage, calculated based on actual data feedback from each execution system. This fine-grained status tracking ensures the visualization and controllability of the resource reallocation process, making the supply chain scheduling adjustment process more transparent and efficient. The entire implementation process embodies a complete closed loop from instruction generation to execution monitoring, achieving adaptive operation of supply chain collaborative scheduling through systematic resource adjustment and continuous performance optimization.

[0048] Example 5: Implementation of Supply Chain Topology Elasticity Score Based on the operational network of an international logistics enterprise, which operates a supply chain system covering six continents, including more than two hundred key facilities such as major hub ports, regional distribution centers, cross-border transfer stations, and urban distribution nodes. The elasticity scoring system regularly evaluates the network structure composed of these facilities, using three core indicators for quantitative analysis. The node utilization balance index measures the load distribution of different nodes in the network, and evaluates the balance degree by calculating the standard deviation of the current load rate of each node. In the specific implementation, the system collects real-time load data of each node every fifteen minutes, including warehouse storage rate, cargo handling volume, order backlog volume, and other parameters. After normalization, the load rate variance of all nodes is calculated. The smaller the variance value, the more balanced the load between nodes, and the more reasonable the network resource utilization; the larger the variance value, the more unbalanced the resource allocation problem of some nodes being overloaded while others are idle. This indicator helps identify key nodes that need load adjustment.

[0049] The Redundancy 5 Path Number indicator assesses the network's resilience when some nodes fail. In the implementation process, the system first constructs a complete logistics path atlas, recording all feasible transportation routes between nodes. Through path similarity analysis algorithms, it identifies functionally replaceable parallel routes. For example, there may be three independent paths for transporting goods from Shanghai Port to Chicago Distribution Center: direct flight route, Arctic route via Anchorage transfer, and Middle East route via Dubai transfer. The system counts the number of such alternative paths with the same starting and ending points, and the more the number, the higher the network redundancy. This indicator considers the independence between paths to avoid counting routes sharing key nodes as effective redundancy. The system updates the path database every week to reflect the impact of newly opened or discontinued routes on network redundancy in a timely manner.

[0050] Key node connectivity monitors the connection status of important facilities and other nodes. The identification of key nodes combines network topology analysis and business impact assessment, selecting hub ports with a handling volume of more than 15%, distribution centers serving more than three countries, and specialized facilities handling special goods. The system continuously tracks the actual number of connections of these key nodes, i.e., the number of available inbound and outbound transportation channels, and compares it with the theoretical maximum number of connections. The higher the connection ratio, the deeper the integration of the key node with the network, and the smaller the impact of a single node failure on the overall network. This indicator is calculated by monitoring the routing data of transportation orders in real time. When the number of connection orders for a certain key node consistently decreases, the system will issue a connectivity warning.

[0051] The comprehensive calculation of the resilience score uses a weighted aggregation method. The three indicators are assigned different weight coefficients, and the weight values are dynamically adjusted according to the strategic focus of the enterprise. In the regular operation period, the node balance degree weight is higher; during the peak season or special events, the number of redundant paths weight is increased; for time-sensitive products, the key node connectivity weight is increased. The calculation process is automatically executed every hour, and the output score value is between zero and one hundred. The score result is divided into multiple levels, corresponding to different management response mechanisms. When the score is below the warning threshold, the system automatically triggers the network optimization process, prompting the need to increase redundant paths or adjust node load. The score result is visualized in the form of a heat map, intuitively presenting the resilience level of each region in the network. This quantitative evaluation mechanism provides an objective basis for the continuous optimization of the supply chain network, helping enterprise managers to identify weak links and implement targeted improvements. The entire implementation process embodies the systematic monitoring and evaluation capability of the health status of the supply chain network structure.

[0052] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0053] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A supply chain production, transportation, and sales collaborative scheduling system, characterized in that, include: The production data acquisition module is used to acquire production plan data, production progress data, and inventory data; The transportation data acquisition module is used to acquire transportation route data, transportation status data, and transportation cost data. The sales data acquisition module is used to obtain sales order data, sales forecast data, and market demand data; The production, transportation, and sales correlation modeling module establishes a correlation model between production, transportation, and sales based on the production planning data, production progress data, and inventory data obtained by the production data acquisition module, the transportation route data, transportation status data, and transportation cost data obtained by the transportation data acquisition module, and the sales order data, sales forecast data, and market demand data obtained by the sales data acquisition module, and generates collaborative scheduling prediction results. The dynamic feature extraction module extracts dynamic features of production, transportation, and sales based on historical and real-time data, and identifies the spatiotemporal variation patterns of scheduling needs. The resource optimization scoring module calculates the resource adjustment priority and resource allocation ratio based on the dynamic features extracted by the dynamic feature extraction module and the collaborative scheduling prediction results generated by the production, transportation and sales association modeling module, and generates an optimized scheduling scheme. The adaptive scheduling execution module, based on the optimized scheduling scheme generated by the resource optimization scoring module, performs real-time adjustments to the scheduling of production, transportation, and sales to ensure that collaborative scheduling adapts to changes.

2. The supply chain production, transportation, and sales collaborative scheduling system according to claim 1, characterized in that, The production data acquisition module specifically includes: Collect production plan data, production progress data, and inventory data, and perform data cleaning, outlier removal, and production feature extraction. The production area is divided into multiple units, and the production status of each unit at different times is statistically analyzed to construct a production status heatmap. Production trajectory clustering algorithms are applied to analyze production progress data to identify typical production patterns and critical production paths; Based on the clustering results of production trajectories, a production flow prediction function is established, which combines time factors to predict the production flow of each key production path at a specific time point.

3. The supply chain production, transportation, and sales collaborative scheduling system according to claim 2, characterized in that, The production, transportation, and sales correlation modeling module specifically includes: Acquire production data, transportation data, and sales data, and then process and integrate the data. Based on multi-source data, a feature vector of production, transportation and sales activities is constructed, which includes the activity intensity, activity type and participating resource information at a specific time point; Construct a dual-mode scheduling pre-adaptive system that includes normal mode and event mode, and define event impact factors to trigger mode switching; Based on historical scheduling data and the characteristics of production, transportation and sales activities, a scheduling-activity correlation model is trained to output the predicted collaborative scheduling distribution.

4. A supply chain production, transportation, and sales collaborative scheduling system according to claim 3, characterized in that, The dynamic feature extraction module specifically includes: Based on historical scheduling data, a time-varying supply chain network graph model is constructed to represent the dynamic characteristics of supply chain topology changes over time. The temporal graph embedding algorithm is applied to convert the time-varying supply chain network graph into a low-dimensional feature representation, thereby optimizing reconstruction error and temporal consistency. Based on the graph embedding results, typical spatiotemporal patterns of scheduling distribution are analyzed, and periodic and abnormal patterns are extracted. Identify the set of key nodes in the supply chain, establish a time-series prediction model for node load, and generate a prediction map of the future supply chain topology load distribution.

5. A supply chain production, transportation, and sales collaborative scheduling system according to claim 4, characterized in that, The resource optimization scoring module specifically includes: Establish a supply chain topology resilience scoring system to assess the adaptability of the supply chain topology to scheduling changes; Based on the supply chain network diagram and collaborative scheduling prediction results, identify the set of key nodes and the set of redundant paths; For the predicted scheduling changes, calculate the adjustment priority score for each node in the topology; Based on node adjustment priority scoring, the optimal allocation ratio of supply chain resources is calculated to solve the constraint optimization problem.

6. A supply chain production, transportation, and sales collaborative scheduling system according to claim 5, characterized in that, The adaptive scheduling execution module specifically includes: Construct a resource reallocation control system to convert resource allocation schemes into specific resource allocation instructions; Based on resource allocation instructions, dynamic adjustments to production, transportation, and sales resources are achieved, and a smooth transition strategy is adopted to avoid interruptions caused by sudden changes. Based on the resource adjustment results, optimize the scheduling strategy and find a set of scheduling paths to achieve the best overall performance of the supply chain. After implementing scheduling adjustments, a real-time monitoring and feedback mechanism is established to trigger adaptive adjustment processes based on performance deviations.

7. A supply chain production, transportation, and sales collaborative scheduling system according to claim 6, characterized in that, The scheduling-activity association model adopts a multilayer perceptron structure, including: The input layer receives feature vectors of production, transportation, and sales activities; The hidden layer contains two sub-layers and introduces a temporal attention mechanism to assign different weights to activity features at different time points; The output layer outputs a predicted scheduling distribution, and the number of neurons is equal to the number of supply chain region divisions.

8. A supply chain production, transportation, and sales collaborative scheduling system according to claim 7, characterized in that, The supply chain topology resilience score is composed of the following indicators: Node utilization balance assesses the balance of load distribution across nodes in the supply chain. The number of redundant paths is used to assess the availability of alternative paths in the event of node failure in the supply chain. Key node connectivity: Assess the connectivity status of key nodes in the supply chain.

9. A supply chain production, transportation, and sales collaborative scheduling method, characterized in that, It includes all modules and method processes of the supply chain production, transportation and sales collaborative scheduling system as described in any one of claims 1 to 9.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the supply chain production, transportation and sales collaborative scheduling method as described in claim 9.

Images