Intelligent supply chain optimization control method

Abstract

The invention relates to the technical field of supply chain management, and discloses an intelligent supply chain optimization management and control method. The method comprises the following steps: acquiring full-link real-time operation data of a supply chain through a distributed data acquisition module; extracting topological feature data by using a graph neural network; inputting the multi-target collaborative optimization model, and generating supply chain strategy optimization parameters by adopting a mixed integer linear programming framework and a dynamic constraint relaxation mechanism; constructing an adaptive resource scheduling model, realizing node dynamic scheduling by using an improved particle swarm algorithm, and outputting optimal configuration data; and establishing a layered supply chain management and control model which comprises a strategic planning layer, a dynamic coordination layer and an execution control layer, and realizing full-link intelligent optimization management and control. The invention also relates to risk event processing, strategic planning layer resource network planning and the like. The system can comprehensively collect data, optimize resource configuration, improve response speed, reduce cost, effectively control risks and improve the overall competitiveness of a supply chain.

Description

Technical Field

[0001] The present invention relates to the technical field of supply chain management, and in particular to an intelligent supply chain optimization and management method. Background Art

[0002] In today's globalized economy, supply chain management is crucial to the operation and development of enterprises. With the increasingly fierce market competition and the diversification and personalization of customer needs, the traditional supply chain management model faces many challenges.

[0003] Traditional supply chain data collection methods have serious limitations. On the one hand, the scope of data collection is limited, making it difficult to obtain real-time operational data for the entire supply chain. Typically, only a few key links are focused on, resulting in an incomplete understanding of the overall status of the supply chain. For example, in inventory management, one may only be able to obtain static inventory quantities in the warehouse, but not be able to know the dynamic changes in inventory in real time, such as the time and frequency of goods entering and leaving the warehouse. On the other hand, data collection is not timely, often relying on manual data collection and entry on a regular basis. This causes data updates to lag and fail to reflect market changes in a timely manner. In a rapidly changing market environment, companies may be unable to obtain accurate market demand data in a timely manner, resulting in a disconnect between production plans and actual demand, which in turn leads to inventory backlogs or stockouts.

[0004] The topology of a supply chain is complex and dynamic, making it difficult for traditional methods to effectively extract its key features. Supply chains involve numerous nodes, such as suppliers, logistics centers, and demand terminals. The relationships between these nodes are complex and constantly changing with market conditions and business operations. Previous analytical methods based on simple graph models or empirical evidence were unable to deeply explore the dynamic connections between nodes and the underlying topological structure. This resulted in companies lacking a clear understanding of the overall supply chain structure when making decisions, making it difficult to optimize resource allocation and coordinate operations across all links.

[0005] Resource allocation and scheduling are critical components of supply chain management, but traditional optimization methods are ineffective. Traditional resource allocation strategies often focus solely on a single objective, such as cost or response time, while neglecting other important factors. For example, a singular pursuit of low cost can lead to extended service response times, impacting customer satisfaction; conversely, an excessive focus on response time can significantly increase costs. Furthermore, traditional optimization models often employ static constraints and are unable to adapt to dynamic changes in the supply chain, such as fluctuations in supplier production capacity and temporary changes in logistics routes. This makes it difficult for companies to quickly adjust resource allocation strategies in the face of operational uncertainties, reducing the flexibility and adaptability of the supply chain.

[0006] Traditional supply chain management methods are relatively weak in terms of risk mitigation. With the increasing number of risk factors such as global economic instability, natural disasters, and public health incidents, the risks facing supply chains are becoming increasingly complex and diverse. Traditional approaches often only implement response measures after a risk occurs, lacking effective risk prediction and early warning mechanisms. Enterprises are unable to identify potential risks in advance and struggle to accurately assess the scope and extent of their impact. This leaves the supply chain vulnerable to severe shocks when risks occur, resulting in significant economic losses. For example, in the raw material supply chain, if a supplier is unable to deliver on time due to a sudden natural disaster, traditional supply chain management models may be unable to adjust production plans and procurement strategies in a timely manner, leading to production interruptions and disrupting normal business operations.

[0007] Faced with these challenges, existing supply chain management technologies are no longer able to meet the needs of enterprises in a complex and volatile market environment. Therefore, there is an urgent need for innovative intelligent supply chain optimization and management methods that can comprehensively and real-timely collect data, deeply analyze the supply chain topology, achieve multi-objective collaborative optimization, efficiently schedule resources, and effectively address various risks to enhance the overall competitiveness and sustainable development capabilities of the supply chain. Summary of the Invention

[0008] The purpose of the present invention is to provide an intelligent supply chain optimization management and control method to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing and controlling an intelligent supply chain, the method comprising:

[0010] Real-time operational data of the entire supply chain is acquired through a distributed data acquisition module. The real-time operational data includes market demand time series data, inventory status matrix, logistics node coordinates, supplier capacity vector, and risk event sequence. Topological features of the real-time operational data are extracted based on a graph neural network to generate supply chain network topological feature data. The supply chain network topological feature data is input into a pre-built multi-objective collaborative optimization model. The multi-objective collaborative optimization model uses a mixed integer linear programming framework and a dynamic constraint relaxation mechanism to iteratively optimize the resource allocation strategy to generate supply chain strategy optimization parameters.

[0011] An adaptive resource scheduling model is constructed based on the supply chain strategy optimization parameters. The adaptive resource scheduling model takes minimizing total cost and minimizing service response time as optimization goals and uses an improved particle swarm algorithm to dynamically schedule supply chain nodes. The improved particle swarm algorithm introduces an inertia weight adaptive adjustment mechanism and a global optimal de-disturbance strategy. Optimal supply chain node configuration data is output based on the adaptive resource scheduling model.

[0012] A hierarchical supply chain management and control model is established based on the optimal supply chain node configuration data. The hierarchical supply chain management and control model includes a strategic planning layer, a dynamic coordination layer and an execution control layer. The strategic planning layer performs global resource network planning based on the supply chain strategy optimization parameters, the dynamic coordination layer performs real-time supply and demand matching optimization based on the optimal supply chain node configuration data, and the execution control layer realizes dynamic adjustment of logistics paths based on a distributed model predictive control algorithm; supply chain management and control instructions are output through the hierarchical supply chain management and control model to realize intelligent optimization and control of the entire chain.

[0013] Preferably, extracting topological features from the real-time operation data based on a graph neural network includes:

[0014] Construct a supply chain network topology graph, where nodes represent suppliers, logistics centers, and demand terminals, and edges represent the strength of associations between nodes and logistics paths. Calculate the dynamic association weights between nodes based on an edge-attention mechanism, where the edge-attention weights are derived by weighted fusion of the cosine similarity of node feature vectors and historical interaction frequencies.

[0015] A multi-layer graph convolutional network is used to propagate features of the topological graph. Each layer of graph convolution operation updates the current node features by aggregating the feature vectors and edge weight information of adjacent nodes. A gated recurrent unit is introduced to dynamically model the time series operation data, and the historical time series data is spliced with the current topological features to generate supply chain network topological feature data.

[0016] Preferably, the multi-objective collaborative optimization model adopts a mixed integer linear programming framework including:

[0017] Define a set of decision variables, including binary location variables, integer capacity allocation variables, and continuous logistics path variables; construct a multi-objective optimization function, including a total cost function, a response time function, and a carbon emission function, where the total cost function is calculated as a linear combination of fixed costs, transportation costs, and inventory holding costs; the response time function is determined by maximizing the service time of critical path nodes; and the carbon emission function is calculated based on the product of transportation distance and load capacity.

[0018] A dynamic constraint relaxation mechanism is introduced to transform hard constraints into soft constraints with penalty terms, and the constraint violation amount is dynamically adjusted through the Lagrange multiplier method; the branch and bound algorithm is used to iteratively solve the relaxed optimization problem to generate the supply chain strategy optimization parameters.

[0019] Preferably, the improved particle swarm algorithm introduces an inertia weight adaptive adjustment mechanism including:

[0020] Dynamically adjust the inertia weight based on a population diversity index calculated as the standard deviation of the Euclidean distance between particles. When the diversity is below a threshold, the inertia weight is reduced to enhance local search, and when the diversity is above the threshold, the inertia weight is increased to enhance global exploration.

[0021] A global optimal solution perturbation strategy is adopted to apply Gaussian noise perturbation to the historical global optimal solution to generate a set of candidate solutions. The non-inferior solutions are screened through the Pareto dominance relationship to update the global optimal solution.

[0022] Preferably, the dynamic coordination layer performs real-time supply and demand matching optimization based on the optimal supply chain node configuration data, including:

[0023] A bilateral auction model for supply and demand matching is constructed, which takes supplier quotes and demander bids as input and determines the optimal matching pair based on the Vickrey-Clarke-Groves mechanism;

[0024] A federated learning framework is used to collaboratively optimize the privacy data of both supply and demand sides, and the global matching model parameters are updated through encrypted gradient aggregation.

[0025] Preferably, the execution control layer implements dynamic adjustment of logistics paths based on a distributed model predictive control algorithm, including:

[0026] Divide the logistics network into multiple sub-areas, and build a local prediction model for each sub-area. The prediction model includes transportation delay, route congestion probability, and vehicle load status;

[0027] The alternating direction multiplier method is used to collaboratively optimize the sub-region models, and distributed solution under global consistency constraints is achieved by exchanging boundary state information.

[0028] The method also includes a method for handling risk event sequences:

[0029] Extract features of risk event sequences based on spatiotemporal convolutional networks and generate risk propagation pattern vectors;

[0030] Construct a Bayesian inference network, input historical risk data and real-time features into the network, calculate the probability of risk occurrence and the scope of impact, and generate risk avoidance strategy parameters.

[0031] Preferably, the strategic planning layer performs global resource network planning including:

[0032] A hypergraph model is used to describe multi-level supply chain relationships, and hyperedges represent cross-level collaborative relationships;

[0033] The hypergraph is divided into communities based on the spectral clustering algorithm, key resource hub nodes are identified, and the global resource allocation topology is generated.

[0034] Preferably, the present invention further includes an intelligent supply chain optimization management and control system for implementing an intelligent supply chain optimization management and control method, the method comprising:

[0035] Data acquisition module, used to obtain real-time data from the entire supply chain through distributed sensors and IoT devices;

[0036] The feature fusion module extracts topological features and performs privacy protection on data based on graph neural networks and federated learning frameworks;

[0037] The optimization decision module uses mixed integer linear programming and improved particle swarm algorithm to generate multi-objective optimization strategies;

[0038] Hierarchical control modules, including strategic planning unit, dynamic coordination unit and execution control unit, realize global planning, real-time matching and path adjustment respectively;

[0039] The risk management module performs risk prediction and avoidance based on spatiotemporal convolutional networks and Bayesian networks.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention obtains real-time operational data of the entire supply chain through a distributed data acquisition module, covering key information such as market demand, inventory status, logistics node coordinates, supplier production capacity and risk event sequence, to ensure that enterprises have a comprehensive and timely understanding of the supply chain status. By using a graph neural network to construct a supply chain network topology map, extracting topological features based on an edge attention mechanism and a multi-layer graph convolutional network, and combining gated recurrent units to model time series data, it is possible to deeply explore the dynamic associations and potential topological structures between supply chain nodes, providing an accurate basis for subsequent decision-making. Compared with traditional data collection and analysis methods, this greatly improves the integrity of the data and the accuracy of the analysis, enabling enterprises to perceive market changes more keenly and adjust their strategies in advance.

[0042] The multi-objective collaborative optimization model uses a mixed-integer linear programming framework, comprehensively considering multiple objectives such as total cost, response time, and carbon emissions to construct a comprehensive optimization function. At the same time, a dynamic constraint relaxation mechanism is introduced to convert hard constraints into soft constraints with penalty terms. Dynamic adjustments are made through the Lagrange multiplier method, and the branch-and-bound algorithm is used for iterative solution. This allows for more flexible response to complex constraints and dynamic changes in the supply chain, generating more reasonable supply chain strategy optimization parameters. The adaptive resource scheduling model aims to minimize total cost and service response time. It uses an improved particle swarm algorithm that introduces an inertia weight adaptive adjustment mechanism and a global optimal de-disturbance strategy to dynamically schedule supply chain nodes, effectively preventing the algorithm from falling into local optimality. This improves the efficiency and quality of resource scheduling, enables efficient allocation of supply chain resources, reduces operating costs, and enhances customer satisfaction.

[0043] The hierarchical supply chain management and control model is divided into a strategic planning layer, a dynamic coordination layer, and an execution control layer. The strategic planning layer uses a hypergraph model and spectral clustering algorithm to plan the global resource network based on supply chain strategy optimization parameters, identify key resource hubs, and optimize the resource allocation topology, providing strategic guidance for the long-term development of the supply chain. The dynamic coordination layer, based on optimal supply chain node configuration data, constructs a bilateral auction model and utilizes a federated learning framework to achieve real-time supply and demand matching optimization. While protecting the privacy data of both supply and demand parties, it improves matching efficiency and fairness, ensuring supply and demand balance in the supply chain. The execution control layer, based on a distributed model predictive control algorithm, divides the logistics network into sub-regions to construct local prediction models, and uses the alternating direction multiplier method for collaborative optimization to achieve dynamic adjustment of logistics routes, improving the timeliness and accuracy of logistics distribution and reducing logistics costs.

[0044] Based on feature extraction from a spatiotemporal convolutional network, a Bayesian inference network is constructed for risk event sequences. This accurately calculates the probability of risk occurrence and the scope of impact, generating risk mitigation strategy parameters. This enables companies to predict risks in advance and take timely measures, such as adjusting inventory levels, optimizing logistics routes, and identifying alternative suppliers. This effectively reduces the impact of risks on the supply chain, enhances supply chain stability and risk resilience, and ensures normal business operations.

[0045] The accompanying intelligent supply chain optimization and management system integrates multiple functional modules, including data collection, feature fusion, optimized decision-making, hierarchical control, and risk management, enabling intelligent management of the entire process from data collection to decision execution. The modules collaborate and work together to adjust supply chain strategies and resource allocation in a timely manner based on real-time data and changing market conditions. This optimizes the overall supply chain, enhances the company's market competitiveness and sustainable development capabilities, and creates greater economic and social benefits for the company. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a working principle diagram of the intelligent supply chain optimization management and control method according to the present invention;

[0047] Figure 2 This is a workflow diagram for topological feature extraction based on graph neural networks;

[0048] Figure 3 To improve the workflow of particle swarm optimization algorithm;

[0049] Figure 4 A workflow diagram for risk event sequence processing. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0051] See also Figure 1-4 , the present invention provides an intelligent supply chain optimization management and control method, the overall implementation scheme of which is as follows:

[0052] The distributed data acquisition module captures real-time operational data across the entire supply chain. This data includes market demand time series data, inventory status matrices, logistics node coordinates, supplier capacity vectors, and risk event sequences. Leveraging various sensors, IoT devices, and data interfaces, the distributed data acquisition module collects data from every link in the supply chain, ensuring real-time and comprehensive data. For example, warehouse sensors can be used to capture inventory status matrices, while supplier capacity vectors can be obtained through integration with supplier systems.

[0053] Using a graph neural network, we extract topological features from this real-time operational data to generate supply chain network topological feature data. This step treats each link in the supply chain as a node and the connections between them as edges, constructing a graph-structured data set. We then leverage the powerful feature extraction capabilities of graph neural networks to mine topological relationships within the supply chain network.

[0054] The supply chain network topology data is fed into a pre-built multi-objective collaborative optimization model. This model uses a mixed-integer linear programming framework and a dynamic constraint relaxation mechanism to iteratively optimize resource allocation strategies and generate supply chain strategy optimization parameters. The multi-objective collaborative optimization model comprehensively considers multiple objectives, including cost, response time, and carbon emissions, and uses a specific planning framework and optimization mechanism to identify the optimal resource allocation strategy.

[0055] An adaptive resource scheduling model is constructed based on the supply chain strategy optimization parameters. This model optimizes total cost and service response time, using an improved particle swarm optimization algorithm to dynamically schedule supply chain nodes. This algorithm incorporates an adaptive inertia weight adjustment mechanism and a global optimal de-disturbance strategy. Based on the optimization parameters, the adaptive resource scheduling model utilizes an improved algorithm to rationally schedule supply chain nodes to achieve the optimization objectives.

[0056] Based on the adaptive resource scheduling model, the optimal supply chain node configuration data is output. By improving the calculation of the particle swarm algorithm, a set of optimal supply chain node configuration solutions is obtained, providing a basis for subsequent management and control.

[0057] A hierarchical supply chain management and control model is established based on the optimal supply chain node configuration data. The model comprises a strategic planning layer, a dynamic coordination layer, and an execution control layer. The strategic planning layer performs global resource network planning based on the supply chain strategy optimization parameters, the dynamic coordination layer optimizes supply and demand matching in real time based on the optimal supply chain node configuration data, and the execution control layer dynamically adjusts logistics routes using a distributed model predictive control algorithm. This hierarchical supply chain management and control model manages and controls the supply chain from different levels and perspectives, ensuring its efficient operation.

[0058] By outputting supply chain control instructions through the layered supply chain control model, intelligent optimization and control of the entire supply chain can be achieved. Based on the optimization results of each layer, specific control instructions are generated to accurately control each link of the supply chain, achieving the purpose of intelligent optimization and control.

[0059] The implementation of the present invention will be further described below with reference to Examples 1 to 5.

[0060] Example 1:

[0061] In step 2, when topological feature extraction is performed on the real-time operation data based on the graph neural network, the specific implementation is as follows:

[0062] Construct a supply chain network topology diagram, where nodes represent suppliers, logistics centers, and demand terminals, and edges represent the strength of connections between nodes and logistics paths. For example, in an electronics supply chain, the chip supplier, assembly plant (which can be considered a logistics center), and electronics retailer (demand terminal) are nodes, and the freight transportation routes between them are edges. To quantify the strength of edge connections, calculations can be performed based on historical transportation volume, transportation frequency, and other data.

[0063] The dynamic association weights between nodes are calculated based on the edge attention mechanism. The edge attention weights are obtained by weighted fusion of the cosine similarity of the node feature vectors and the historical interaction frequency. Assume that the feature vectors of node i and node j are and The cosine similarity between them is The historical interaction frequency is f ij . Edge attention weight w ij It can be calculated by the following formula:

[0064]

[0065] Here, α is an adjustable weight coefficient, which is set according to the actual situation and its value range is usually between [0, 1].

[0066] A multi-layer graph convolutional network is used to propagate features of the topological graph. Each layer of graph convolution operation updates the current node feature by aggregating the feature vectors and edge weight information of adjacent nodes. Taking the l-th layer of graph convolution as an example, the feature update formula of node i is

[0067]

[0068] in is the feature vector of node i in layer l, N(i) is the set of neighbor nodes of node i, d i and d j are the degrees of nodes i and j, respectively, W l is the weight matrix of the lth layer, and σ is the activation function, such as the ReLU function.

[0069] Gated recurrent units (GRUs) are introduced to dynamically model time-series operational data, concatenating historical time-series data with current topological features to generate supply chain network topological feature data. Gated recurrent units (GRUs) effectively handle long-term dependencies in time-series data. GRUs are used to model market demand time-series data and generate time-series features. These time-series features are then concatenated with topological features derived from graph convolutional networks across the feature dimension to generate richer supply chain network topological feature data, providing more comprehensive input information for subsequent multi-objective collaborative optimization models.

[0070] Example 2:

[0071] The purpose of this embodiment is to explain in detail how the multi-objective collaborative optimization model uses a mixed integer linear programming framework and how to combine it with a dynamic constraint relaxation mechanism to optimize and solve the resource allocation strategy, thereby obtaining more reasonable supply chain strategy optimization parameters. When the multi-objective collaborative optimization model in step 3 uses a mixed integer linear programming framework, the specific implementation is as follows:

[0072] Define a set of decision variables, including binary location variables, integer capacity allocation variables, and continuous logistics path variables. Take a simple supply chain scenario as an example. Assume there are m potential logistics center locations, and the binary location variable x i (i=1,2,…,m) indicates whether to choose the i-th location to establish a logistics center, x i =1 means selection, x i =0 means no selection; there are n suppliers and k demand terminals, and the integer capacity allocation variable y ij represents the quantity of products that supplier i allocates to demand terminal j; continuous logistics path variable z ij Represents the logistics transportation volume from supplier i to demand terminal j.

[0073] Construct a multi-objective optimization function, which includes the total cost function, response time function and carbon emission function. The total cost function C is calculated by the linear combination of fixed cost, transportation cost and inventory holding cost. Assume that the fixed cost of establishing logistics center i is F i , the unit transportation cost from supplier i to demand terminal j is t ij , the inventory holding cost per unit product in the logistics center is h, and the inventory quantity is I, then the total cost function The response time function T is determined by the maximum service time of the critical path nodes. Assume that the service time of node l on the critical path is s l , then T=max{s l The carbon emission function E is calculated based on the product of transportation distance and load, assuming that the transportation distance from supplier i to demand terminal j is d ij , the transport load is w ij ,but Introducing a dynamic constraint relaxation mechanism, the hard constraints are converted into soft constraints with penalty terms, and the constraint violation amount is dynamically adjusted through the Lagrange multiplier method. For example, for capacity constraints (S i is the capacity of supplier i), when this constraint is violated, a penalty term is introduced where λ i is the Lagrange multiplier. By continuously adjusting the Lagrange multiplier, the constraint violation is kept within an acceptable range while optimizing the objective function.

[0074] A branch-and-bound algorithm is used to iteratively solve the relaxed optimization problem and generate supply chain strategy optimization parameters. This algorithm branches the decision variables, gradually narrowing the range of feasible solutions. It also utilizes the concepts of upper and lower bounds to prune branches that are unlikely to contain the optimal solution, thereby improving solution efficiency. During each iteration, the solution to the current subproblem is calculated and the upper and lower bounds are updated until an optimal solution that satisfies the requirements is found. This results in supply chain strategy optimization parameters, such as optimal location decisions, capacity allocation plans, and logistics routing plans.

[0075] Example 3:

[0076] The purpose of this embodiment is to elaborate on how the inertia weight adaptive adjustment mechanism and global optimal de-disturbance strategy of the improved particle swarm algorithm play a role in the adaptive resource scheduling model, so as to achieve more efficient dynamic scheduling of supply chain nodes and achieve the optimization goals of minimizing total cost and shortening service response time. The introduction of the inertia weight adaptive adjustment mechanism of the improved particle swarm algorithm in step 4 includes:

[0077] The inertia weight is dynamically adjusted based on the population diversity index, which is calculated by the standard deviation of the Euclidean distance between particles. Assume that there are N particles in the particle swarm, and the position vector of each particle is The average position vector of the particle swarm is The standard deviation σ of the Euclidean distance between particles can be expressed by the formula Calculation. When the diversity index σ is lower than the set threshold σ th When , it means that the particle swarm tends to converge. At this time, reduce the inertia weight w, such as using the formula where w max and w min are the maximum and minimum values of the inertia weight, respectively, to enhance the local search capability and enable the algorithm to find the optimal solution more finely in the current local area. When the diversity index σ is higher than the threshold σ th When the inertia weight w is increased, the formula is used σ max It is the maximum value of the diversity index to enhance the global exploration ability, allowing the algorithm to search the solution space more widely and avoid falling into the local optimum.

[0078] Using the global optimal solution perturbation strategy, Gaussian noise perturbation is applied to the historical global optimal solution to generate a candidate solution set, and the non-inferior solution is screened through the Pareto dominance relationship to update the global optimal solution. Assume that the historical global optimal solution is Apply Gaussian noise to it Get candidate solutions After generating multiple candidate solutions, they are screened based on the Pareto dominance relationship. If solution A is not inferior to solution B in all objectives and is superior to solution B in at least one objective, then solution A is said to dominate solution B. The set of all non-inferior solutions is used as the new global optimal solution candidate set. From these, appropriate solutions are selected to update the global optimal solution, thereby guiding the particle swarm to search for more optimal areas, improving the algorithm's convergence speed and solution quality, and achieving more reasonable dynamic scheduling of supply chain nodes.

[0079] Example 4:

[0080] In step 6, when the dynamic coordination layer performs real-time supply and demand matching optimization based on the optimal supply chain node configuration data, the specific implementation is as follows:

[0081] A bilateral auction model for supply and demand matching is constructed, taking supplier quotes and demander bids as inputs and determining the optimal matching pair based on the Vickrey-Clarke-Groves mechanism. For example, assume multiple suppliers and multiple demanders participate in the matching process. Suppliers quote product prices based on their costs and expected profits, while demanders bid prices based on their needs and budgets. The core idea of the Vickrey-Clarke-Groves mechanism is that for each participant, their payoff or benefit depends on their marginal impact on the overall outcome. In supply and demand matching, this mechanism ensures that each participant truthfully declares their quotes and bids, thereby achieving efficient resource allocation. Specifically, for each supplier, the payoff is the difference between the social welfare of all suppliers and demanders after matching, excluding itself, and the social welfare after matching itself. For each demander, the price paid is the difference between the social welfare of all suppliers and demanders after matching, excluding itself, and the social welfare after matching itself (a positive value indicates a payoff; a negative value indicates a gain). In this way, the optimal supply and demand matching pair is determined, balancing the interests of both parties.

[0082] A federated learning framework is used to collaboratively optimize the private data of both supply and demand parties, updating the global matching model parameters through encrypted gradient aggregation. In real-world supply chains, suppliers and demanders each possess their own private data, such as cost structures and demand preferences. The federated learning framework allows all parties to train models without sharing their original data. In practice, each party trains the matching model locally using its own data and calculates the model's gradients. These gradients are then encrypted using encryption technology and uploaded to a central server. The central server aggregates the encrypted gradients, decrypts them, and distributes them to all parties. Each party updates its local model parameters based on the aggregated gradients, continuously iterating and optimizing to ensure that the global matching model better adapts to actual supply and demand conditions while protecting the privacy of all parties.

[0083] Example 5:

[0084] When the control layer implements dynamic adjustment of logistics paths based on the distributed model predictive control algorithm in step 6, the specific implementation is as follows:

[0085] The logistics network is divided into multiple sub-regions. Based on factors such as logistics flow, geographical areas, and traffic conditions, the entire logistics network can be reasonably divided into multiple relatively independent but interrelated sub-regions. Taking a large multinational logistics network as an example, it can be divided according to different countries or regions, and each sub-region contains several logistics nodes. A local prediction model is constructed for each sub-region, which incorporates key factors such as transportation delay, path congestion probability, and vehicle load status. Transportation delay is determined by analyzing historical transportation data, road construction plans, weather impacts, etc.; path congestion probability is calculated with the help of real-time traffic data and road condition prediction models; vehicle load status is obtained in real time by on-board sensors. Assume that the local prediction model of a sub-region is: transportation time T = α × T d +β×P c +γ×L+δ, where T d Is shipping delayed, P c is the path congestion probability, L is the vehicle load status, α, β, γ are weight coefficients, and δ is a constant term. These parameters can be obtained through historical data regression analysis.

[0086] The alternating direction multiplier method is used to collaboratively optimize the sub-region models. Within each sub-region, the locally optimal logistics path is determined based on the local prediction model and the sub-region's internal logistics constraints, such as warehouse capacity and vehicle quantity limits. Sub-regions exchange boundary status information, such as cargo flow at boundary nodes and vehicle arrival times, to achieve distributed solution under global consistency constraints. In each iteration, sub-regions update their local models and optimization results based on boundary information transmitted from adjacent sub-regions, gradually achieving global optimization for the entire logistics network, enabling dynamic adjustment of logistics paths and improving logistics efficiency.

[0087] In step 7, the risk event sequence processing method uses a spatiotemporal convolutional network to extract features from the risk event sequence. Risk events are organized into a spatiotemporal data structure based on their chronological order and location, and then fed into the spatiotemporal convolutional network. The convolutional layers in the network use convolution kernels to slide across the spatiotemporal dimensions, extracting features of risk events at different times and locations, such as the changing trends in risk frequency across different regions and the fluctuations in risk intensity over time. This generates a risk propagation pattern vector, revealing the spatiotemporal evolution of risk.

[0088] A Bayesian inference network is constructed, taking historical risk data and real-time features extracted by a spatiotemporal convolutional network as input. Based on Bayes' theorem, the network calculates conditional probabilities between nodes to determine the probability of risk occurrence and the scope of its impact. For example, for supplier disruption risk, the probability of a supplier disruption and the range of downstream companies potentially affected are calculated based on historical supplier disruption data, current supplier production status, and real-time supply market information. Based on these results, risk mitigation strategy parameters are generated, such as adjusting safety stock levels and identifying alternative suppliers, to ensure supply chain stability.

[0089] During global resource network planning at the strategic planning level in step 8, a hypergraph model is used to describe multi-tiered supply chain relationships. In a complex supply chain, there are multiple tiers, including raw material suppliers, component manufacturers, finished product assemblers, and retailers. Nodes in the hypergraph represent enterprises at different levels, while hyperedges represent cross-tier collaborative relationships. For example, a raw material supplier directly establishes a long-term cooperation agreement with a finished product assembler. This complex relationship can be clearly represented in the hypergraph model.

[0090] Hypergraphs are partitioned into communities based on a spectral clustering algorithm. This algorithm uses the eigenvectors of the hypergraph's Laplacian matrix to divide nodes into distinct communities. In the supply chain, these communities correspond to different functional modules or interest groups. By analyzing the connection strength and resource flow of nodes within each community, key resource hubs are identified. These nodes play a crucial role in resource allocation and information transmission. Based on the community partitioning and key node identification results, a global resource allocation topology is generated, providing a basis for long-term strategic planning of the supply chain, optimizing resource allocation, and enhancing the overall competitiveness of the supply chain.

[0091] The present invention also includes an intelligent supply chain optimization management and control system, the system comprising:

[0092] Data acquisition module, used to obtain real-time data from the entire supply chain through distributed sensors and IoT devices;

[0093] The feature fusion module extracts topological features and performs privacy protection on data based on graph neural networks and federated learning frameworks;

[0094] The optimization decision module uses mixed integer linear programming and improved particle swarm algorithm to generate multi-objective optimization strategies;

[0095] Hierarchical control modules, including strategic planning unit, dynamic coordination unit and execution control unit, realize global planning, real-time matching and path adjustment respectively;

[0096] The risk management module performs risk prediction and avoidance based on spatiotemporal convolutional networks and Bayesian networks.

[0097] The implementation of the system is described in the above embodiment and will not be described in detail in the specification.

[0098] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0099] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent supply chain optimization management and control method, characterized in that: include: The distributed data acquisition module acquires real-time operational data across the entire supply chain, including market demand time series data, inventory status matrix, logistics node coordinates, supplier capacity vectors, and risk event sequences. A graph neural network is used to extract topological features from the real-time operational data to generate supply chain network topology feature data. Inputting the supply chain network topology feature data into a pre-built multi-objective collaborative optimization model, wherein the multi-objective collaborative optimization model uses a mixed integer linear programming framework combined with a dynamic constraint relaxation mechanism to iteratively optimize the resource allocation strategy and generate supply chain strategy optimization parameters; An adaptive resource scheduling model is constructed based on the supply chain strategy optimization parameters. The adaptive resource scheduling model takes minimizing total cost and minimizing service response time as optimization goals, and adopts an improved particle swarm algorithm to dynamically schedule supply chain nodes. The improved particle swarm algorithm introduces an inertia weight adaptive adjustment mechanism and a global optimal de-disturbance strategy. Outputting optimal supply chain node configuration data based on the adaptive resource scheduling model; A hierarchical supply chain management and control model is established based on the optimal supply chain node configuration data. The hierarchical supply chain management and control model includes a strategic planning layer, a dynamic coordination layer and an execution control layer. The strategic planning layer performs global resource network planning based on the supply chain strategy optimization parameters, the dynamic coordination layer performs real-time supply and demand matching optimization based on the optimal supply chain node configuration data, and the execution control layer realizes dynamic adjustment of logistics paths based on a distributed model predictive control algorithm; supply chain management and control instructions are output through the hierarchical supply chain management and control model to realize intelligent optimization and control of the entire chain.

2. The method according to claim 1, characterized in that Extracting topological features of the real-time operation data based on the graph neural network includes: Construct a supply chain network topology graph, where nodes represent suppliers, logistics centers, and demand terminals, and edges represent the strength of associations between nodes and logistics paths. Calculate the dynamic association weights between nodes based on an edge-attention mechanism, where the edge-attention weights are derived by weighted fusion of the cosine similarity of node feature vectors and historical interaction frequencies. A multi-layer graph convolutional network is used to propagate features of the topological graph. Each layer of graph convolution operation updates the current node features by aggregating the feature vectors and edge weight information of adjacent nodes. A gated recurrent unit is introduced to dynamically model the time series operation data, and the historical time series data is spliced with the current topological features to generate supply chain network topological feature data.

3. The method according to claim 1, characterized in that The multi-objective collaborative optimization model adopts a mixed integer linear programming framework including: Define a set of decision variables, including binary location variables, integer capacity allocation variables, and continuous logistics path variables; construct a multi-objective optimization function, including a total cost function, a response time function, and a carbon emission function, where the total cost function is calculated as a linear combination of fixed costs, transportation costs, and inventory holding costs; the response time function is determined by maximizing the service time of critical path nodes; and the carbon emission function is calculated based on the product of transportation distance and load capacity. A dynamic constraint relaxation mechanism is introduced to transform hard constraints into soft constraints with penalty terms, and the constraint violation amount is dynamically adjusted through the Lagrange multiplier method; the branch and bound algorithm is used to iteratively solve the relaxed optimization problem to generate the supply chain strategy optimization parameters.

4. The method according to claim 1, wherein The improved particle swarm algorithm introduces an inertia weight adaptive adjustment mechanism including: Dynamically adjust the inertia weight based on a population diversity index calculated as the standard deviation of the Euclidean distance between particles. When the diversity is below a threshold, the inertia weight is reduced to enhance local search, and when the diversity is above the threshold, the inertia weight is increased to enhance global exploration. A global optimal solution perturbation strategy is adopted to apply Gaussian noise perturbation to the historical global optimal solution to generate a set of candidate solutions. The non-inferior solutions are screened through the Pareto dominance relationship to update the global optimal solution.

5. The method according to claim 1, wherein The dynamic coordination layer performs real-time supply and demand matching optimization based on the optimal supply chain node configuration data, including: A bilateral auction model for supply and demand matching is constructed, which takes supplier quotes and demander bids as input and determines the optimal matching pair based on the Vickrey-Clarke-Groves mechanism; A federated learning framework is used to collaboratively optimize the privacy data of both supply and demand sides, and the global matching model parameters are updated through encrypted gradient aggregation.

6. The method according to claim 1, characterized in that The execution control layer implements dynamic adjustment of logistics routes based on the distributed model predictive control algorithm, including: Divide the logistics network into multiple sub-areas, and build a local prediction model for each sub-area. The prediction model includes transportation delay, route congestion probability, and vehicle load status; The alternating direction multiplier method is used to collaboratively optimize the sub-region models, and distributed solution under global consistency constraints is achieved by exchanging boundary state information.

7. The method according to claim 1, characterized in that It also includes methods for handling risk event sequences: Extract features of risk event sequences based on spatiotemporal convolutional networks and generate risk propagation pattern vectors; Construct a Bayesian inference network, input historical risk data and real-time features into the network, calculate the probability of risk occurrence and the scope of impact, and generate risk avoidance strategy parameters.

8. The method according to claim 1, characterized in that The strategic planning layer conducts global resource network planning, including: A hypergraph model is used to describe multi-level supply chain relationships, and hyperedges represent cross-level collaborative relationships; The hypergraph is divided into communities based on the spectral clustering algorithm, key resource hub nodes are identified, and the global resource allocation topology is generated.

9. An intelligent supply chain optimization management and control system, used to implement the method according to any one of claims 1 to 8, characterized in that: include: Data acquisition module, used to obtain real-time data from the entire supply chain through distributed sensors and IoT devices; The feature fusion module extracts topological features and performs privacy protection on data based on graph neural networks and federated learning frameworks; The optimization decision module uses mixed integer linear programming and improved particle swarm algorithm to generate multi-objective optimization strategies; Hierarchical control modules, including strategic planning unit, dynamic coordination unit and execution control unit, realize global planning, real-time matching and path adjustment respectively; The risk management module performs risk prediction and avoidance based on spatiotemporal convolutional networks and Bayesian networks.

Images