Supply chain multi-node real-time cooperative scheduling and emergency response system and scheduling method

Abstract

The invention relates to the technical field of dispatching and emergency response, in particular to a supply chain multi-node real-time collaborative dispatching and emergency response system and method, and the system comprises a distributed data collection module which is used for obtaining the inventory data, logistics state and equipment operation parameters of each node in real time; the digital twin modeling engine is used for constructing a dynamic virtual mapping model of the supply chain network; the collaborative decision center generates a multi-objective optimization scheduling scheme based on a reinforcement learning algorithm; the emergency response trigger is used for automatically starting a graded emergency plan through abnormal mode recognition; according to the method, second-level response is realized through millisecond-level data synchronization and edge calculation, so that decision timeliness is improved, cross-node cooperation efficiency is improved by adopting multi-agent game and federated learning, and the punctuality rate of orders and the toughness index of the network are improved through multi-target Pareto optimization on the premise of controllable cost.

Description

Technical Field

[0001] The present invention relates to the technical field of scheduling and emergency response, and in particular to a supply chain multi-node real-time collaborative scheduling and emergency response system and a scheduling method. Background Art

[0002] Currently, supply chain multi-node collaborative scheduling and emergency response systems mainly rely on static optimization models and centralized decision-making architectures. Typical scheduling methods usually perform offline planning based on historical data and trigger emergency responses through preset rules or simple thresholds. For example, some systems use linear programming or heuristic algorithms to optimize transportation routes and inventory allocation, while anomaly detection relies on fixed threshold alarms or manual experience judgment. In addition, existing technologies mostly use single-objective optimization and lack the real-time processing capability of dynamic multi-objective trade-offs.

[0003] Traditional systems rely on periodic data updates and are unable to cope with the rapid changes in emergencies, resulting in delayed decision-making and an inability to meet the highly dynamic needs of modern supply chains. In addition, most systems adopt a centralized decision-making model and lack an autonomous negotiation mechanism between nodes, which leads to rigid resource allocation and makes it difficult to adapt to the flexible needs of distributed supply chain networks. At the same time, traditional scheduling models often only consider single objectives such as cost or timeliness, making it difficult to achieve a Pareto optimal balance of multiple objectives in a complex environment.

[0004] Therefore, in order to solve the above problems, the present invention proposes a supply chain multi-node real-time collaborative scheduling and emergency response system and scheduling method. Summary of the Invention

[0005] In order to overcome the problems of insufficient real-time performance, limited collaborative capabilities and single optimization objectives of traditional systems, the present invention proposes a supply chain multi-node real-time collaborative scheduling and emergency response system and scheduling method.

[0006] The technical solution of the present invention is: a supply chain multi-node real-time collaborative scheduling and emergency response system and scheduling method, including:

[0007] Distributed data acquisition module, used to obtain real-time inventory data, logistics status and equipment operating parameters of each node;

[0008] A digital twin modeling engine for building dynamic virtual mapping models of supply chain networks;

[0009] Collaborative decision-making center generates multi-objective optimization scheduling solutions based on reinforcement learning algorithms;

[0010] Emergency response triggers automatically initiate hierarchical emergency plans through abnormal pattern recognition.

[0011] Preferably, the digital twin modeling engine includes a topology relationship construction unit, a state synchronization unit and a prediction and deduction unit. The topology relationship construction unit dynamically maintains the multi-dimensional associations between nodes through a graph database, and defines three types of edge attributes: logistics dependence intensity, capital settlement cycle, and information sharing level, to support real-time topology reconstruction; the state synchronization unit adopts a publish-subscribe model to achieve millisecond-level broadcasting of node status through Apache Kafka, with a synchronization delay of ≤300ms; the prediction and deduction unit integrates the supply chain risk knowledge graph, combines meteorological and political information, and uses an LSTM-GAN hybrid model to predict the probability of interruption in the next 72 hours, and outputs a risk heat map.

[0012] Preferably, the collaborative decision-making center includes a multi-objective optimizer, a conflict resolution algorithm and a dynamic route planner. The multi-objective optimizer converts transportation costs, inventory turnover rates and delivery punctuality rates into constrained mixed integer programming problems, and adopts a branch pricing algorithm to solve them; the conflict resolution algorithm introduces an improved Shapley value method, increases the node contribution time attenuation factor, and dynamically adjusts the benefit distribution weight; the dynamic route planner generates an optimal path set every 5 minutes based on the real-time updated road congestion index and capacity utilization rate through the Dijkstra-A* hybrid algorithm, and supports dynamic rerouting during transportation.

[0013] Preferably, the emergency response trigger includes a three-level early warning mechanism, a resource pooling scheduling module and a blockchain evidence storage unit. The three-level early warning mechanism sets inventory red lines, transportation delay thresholds and capacity gap rates as grading standards to trigger resource allocation plans at corresponding levels; the resource pooling scheduling module establishes a virtual emergency warehouse, automatically locks the available inventory of the three nearest nodes through smart contracts, and allocates transportation capacity using a Dutch auction algorithm; the blockchain evidence storage unit deploys a Hyperledger Fabric private chain to record the entire life cycle data of emergency events from triggering to release, including decision logs, resource allocation vouchers and post-audit traces.

[0014] Preferably, the system also includes an edge computing gateway, a federated learning framework and a digital certificate system. The edge computing gateway adopts the NVIDIA Jetson AGX Orin module to run a lightweight decision-making model locally on the node with a processing delay of <50ms. The federated learning framework is designed based on a gradient aggregation protocol with homomorphic encryption. Each node uploads encrypted model parameters, and the central server aggregates and updates the global model. The digital certificate system issues ERC-1155 standard tokens to record service transactions between nodes, and automatically settles through Solidity smart contracts, supporting minute-level settlement.

[0015] Preferably, the prediction and deduction unit includes a risk propagation analysis submodule, a resilience assessment submodule and a scenario construction library. The risk propagation analysis submodule constructs a directed weighted graph of the supply chain, uses the SIR infectious disease model to simulate the cascade effect of the interruption event, and outputs a list of key vulnerable nodes; the resilience assessment submodule defines the node resilience index RI = α·redundancy + β·response speed (α=0.6, β=0.4), and calculates the overall resilience of the network through Monte Carlo simulation; the scenario construction library stores a variety of typical scenarios including port closures and chip shortages, and each scenario is associated with a pre-trained response strategy neural network model.

[0016] Preferably, the real-time traffic situation awareness module of the dynamic route planner integrates GPS floating vehicle data, emergency accident information reported by roadside units, and NOAA weather radar data to construct a four-dimensional (x, y, t, risk) path cost function; the adaptive weight adjustment module dynamically sets the cost-timeliness weight ratio according to the urgency of the order, and the timeliness weight of S-level orders reaches 90%; the alternative path generation algorithm maintains K=5 optimal candidate paths, and when the main path is interrupted, it switches to the suboptimal path within 500ms based on the reinforcement learning model.

[0017] Preferably, the system also includes a human-computer collaborative interface, a knowledge graph engine and a simulation training system. The human-computer collaborative interface provides a Hololens 2AR visualization interface, superimposes the scheduling plan on the real warehousing scene, and supports gesture operation to adjust the cargo location allocation; the knowledge graph engine uses Neo4j to store emergency cases in the past five years, extracts event features through the BERT model, and builds a case similarity retrieval system; the simulation training system builds a Unity3D virtual environment, and trainees can practice the cross-node resource allocation process through VR headsets. The system scores in real time and generates improvement reports.

[0018] Preferably, the shared inventory electronic dashboard of the resource pooling scheduling module is connected to the WMS system of each node, and the available resources are displayed in three dimensions: material category, available quantity and geographical location; the capacity bidding platform is designed based on a smart contract of sealed second-price auction. After the shipper submits the transportation demand, the platform completes the carrier matching and price settlement within 3 minutes; the capability certification system crawls the historical performance data of third-party logistics companies and uses the XGBoost model to calculate the credit score. Companies with a score below 60 are not allowed to participate in emergency response.

[0019] Preferably, the method for real-time collaborative scheduling of multiple nodes in the supply chain and the scheduling of the emergency response system includes the following steps:

[0020] S1 uses a distributed data acquisition module to obtain inventory data, logistics status, and equipment operating parameters of each supply chain node in real time. The digital twin modeling engine constructs a dynamic virtual mapping model of the supply chain network based on this data. Its topology relationship construction unit uses a graph database to dynamically maintain multi-dimensional relationships between nodes, defining three types of edge attributes: logistics dependency intensity, capital settlement cycle, and information sharing level, supporting real-time topology reconstruction. The state synchronization unit adopts a publish-subscribe model and implements millisecond-level node status broadcasting through Apache Kafka. The prediction and deduction unit integrates the supply chain risk knowledge graph and combines it with the LSTM-GAN hybrid model to predict the interruption probability in the next 72 hours, outputting a risk heat map to provide data support for subsequent decision-making.

[0021] S2, the emergency response trigger identifies abnormal fluctuations through a three-level early warning mechanism and automatically triggers multi-node joint simulation and deduction. The risk propagation analysis submodule of the prediction and deduction unit constructs a directed weighted graph of the supply chain, uses the SIR infectious disease model to simulate the cascading effects of disruption events, and outputs a list of key vulnerable nodes. The resilience assessment submodule calculates the network's overall resilience index (RI) (0.6·redundancy+0.4·response speed) through Monte Carlo simulation. The scenario construction library calls a pre-trained neural network model and generates response strategies based on real-time meteorological and political data. The dynamic route planner of the collaborative decision-making center simultaneously integrates GPS, roadside unit, and weather radar data to construct a four-dimensional path cost function, providing real-time environmental parameters for simulation and deduction.

[0022] S3, the collaborative decision-making hub's multi-objective optimizer, transforms transportation costs, inventory turnover, and on-time delivery into mixed integer programming problems, solving them using a branch-and-price algorithm. The conflict resolution algorithm incorporates an improved Shapley value method, adds a time-decrease factor to node contributions, and dynamically adjusts benefit distribution weights to balance the demands of multiple parties. The resource pooling scheduling module locks available inventory at the three nearest nodes through smart contracts and allocates capacity using a Dutch auction algorithm to ensure fairness and efficiency in resource scheduling. The dynamic route planner generates an optimal path set every five minutes based on the Dijkstra-A* hybrid algorithm. Combined with an adaptive weight adjustment module and an alternative path generation algorithm, it ultimately outputs a Pareto-optimal solution.

[0023] S4, the system executes lightweight decision-making models locally at the node through the edge computing gateway, and provides real-time feedback on the execution data of the plan. The federated learning framework is based on the homomorphic encrypted gradient aggregation protocol, which uploads the encrypted model parameters of each node to the central server, aggregates and updates the global model to optimize the scheduling strategy. At the same time, the blockchain evidence storage unit records the entire life cycle data of the emergency event to ensure the traceability of the process. The digital certificate system uses ERC-1155 tokens and Solidity smart contracts to achieve minute-level service transaction settlement. The human-computer collaborative interface superimposes the actual execution effect on the digital twin, and combines the case similarity retrieval system of the knowledge graph engine to continuously improve the practice scenarios of the simulation training system to form a closed-loop optimization.

[0024] Beneficial effects of the present invention:

[0025] 1. By acquiring node operation data through a distributed data acquisition module and combining it with the state synchronization unit and edge computing network of the digital twin modeling engine, an end-to-end low-latency data processing chain is constructed, enabling the system to respond to supply chain fluctuations in seconds. Compared with traditional periodic update systems, this improves decision-making time by more than 90%, effectively resolving scheduling deviations caused by data lags.

[0026] 2. A multi-agent game architecture is adopted to dynamically calculate node contributions through an improved Shapley value algorithm in the collaborative decision-making center. A federated learning framework is introduced to achieve encrypted parameter aggregation, allowing each node to autonomously negotiate resource allocation plans while protecting data privacy. This improves the efficiency of cross-node collaboration while reducing the risk of single point failures caused by centralized decision-making.

[0027] 3. By transforming transportation costs, inventory turnover rate, and delivery on-time rate into a constrained mixed integer programming problem, solving the Pareto front solution through a branch-and-price algorithm, and combining it with the adaptive weight module of the dynamic routing planner to achieve multi-objective dynamic trade-offs, this approach not only ensures cost control but also improves order on-time rate and network resilience. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Shown is a schematic diagram of the system framework of the present invention;

[0029] Figure 2 Shown is a schematic diagram of the system framework connection of the present invention;

[0030] Figure 3 What is shown is a schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0032] See also Figure 1 and Figure 2 The present invention provides an embodiment: a supply chain multi-node real-time collaborative scheduling and emergency response system, comprising:

[0033] Distributed data acquisition module, used to obtain real-time inventory data, logistics status and equipment operating parameters of each node;

[0034] A digital twin modeling engine for building dynamic virtual mapping models of supply chain networks;

[0035] Collaborative decision-making center generates multi-objective optimization scheduling solutions based on reinforcement learning algorithms;

[0036] Emergency response triggers automatically initiate hierarchical emergency plans through abnormal pattern recognition.

[0037] The digital twin modeling engine includes a topology relationship construction unit, a state synchronization unit and a prediction and deduction unit. The topology relationship construction unit dynamically maintains multi-dimensional associations between nodes through a graph database, defines three types of edge attributes: logistics dependency intensity, capital settlement cycle, and information sharing level, and supports real-time topology reconstruction; the state synchronization unit adopts a publish-subscribe model and implements millisecond-level broadcasting of node status through Apache Kafka, with a synchronization delay of ≤300ms; the prediction and deduction unit integrates the supply chain risk knowledge graph, combines external data sources such as meteorology and politics, uses an LSTM-GAN hybrid model to predict the probability of interruption in the next 72 hours, and outputs a risk heat map.

[0038] The collaborative decision-making center includes a multi-objective optimizer, a conflict resolution algorithm and a dynamic route planner. The multi-objective optimizer converts transportation costs, inventory turnover rate and delivery on-time rate into constrained mixed integer programming problems and adopts a branch and price algorithm to solve them. The conflict resolution algorithm introduces an improved Shapley value method, increases the node contribution time attenuation factor (λ=0.85), and dynamically adjusts the benefit distribution weight. The dynamic route planner generates an optimal path set every 5 minutes based on the real-time updated road congestion index (RCI) and capacity utilization rate through the Dijkstra-A* hybrid algorithm, supporting dynamic rerouting during transportation.

[0039] The three-level early warning mechanism of the emergency response trigger sets the inventory red line (20%), transportation delay threshold (48 hours), and capacity gap rate (15%) as grading standards to trigger the corresponding level of resource allocation plan; the resource pooling scheduling module establishes a virtual emergency warehouse, automatically locks the available inventory of the three nearest nodes through smart contracts, and allocates transportation capacity using the Dutch auction algorithm; the blockchain evidence storage unit deploys the Hyperledger Fabric private chain to record the entire life cycle data of the emergency event from triggering to release, including decision logs, resource allocation vouchers and post-audit traces.

[0040] The system also includes an edge computing gateway, a federated learning framework, and a digital credential system. The edge computing gateway uses the NVIDIA Jetson AGX Orin module to run lightweight decision-making models locally on the node with a processing delay of less than 50ms. The federated learning framework is designed based on a gradient aggregation protocol with homomorphic encryption. Each node uploads encrypted model parameters, and the central server aggregates and updates the global model. The digital credential system issues ERC-1155 standard tokens to record service transactions between nodes, such as warehouse sharing and capacity leasing, and automatically settles through Solidity smart contracts, supporting minute-level settlement.

[0041] The prediction and deduction unit includes a risk propagation analysis submodule, a resilience assessment submodule and a scenario construction library. The risk propagation analysis submodule constructs a directed weighted graph of the supply chain, uses the SIR infectious disease model to simulate the cascading effects of interruption events, and outputs a list of key vulnerable nodes; the resilience assessment submodule defines the node resilience index RI = α·redundancy + β·response speed (α=0.6, β=0.4), and calculates the overall resilience of the network through Monte Carlo simulation; the scenario construction library stores 12 typical scenarios including port closures and chip shortages, and each scenario is associated with a pre-trained response strategy neural network model.

[0042] The dynamic route planner's real-time traffic situation awareness module integrates GPS floating vehicle data (updated at 1Hz), emergency information reported by roadside units (RSUs), and NOAA weather radar data to construct a four-dimensional (x, y, t, risk) path cost function. The adaptive weight adjustment module dynamically sets the cost-time weight ratio based on the urgency of the order (divided into three levels: S / A / B), with the time weight for S-level orders reaching 90%. The alternative path generation algorithm maintains K = 5 optimal candidate paths. When the main path is interrupted, it switches to the suboptimal path within 500ms based on a reinforcement learning model (PPO algorithm).

[0043] The system also includes a human-computer collaborative interface, a knowledge graph engine and a simulation training system. The human-computer collaborative interface provides a Hololens 2AR visualization interface, superimposes the scheduling plan on the real warehousing scene, and supports gesture operation to adjust the cargo allocation. The knowledge graph engine uses Neo4j to store emergency cases in the past five years and extracts event features through the BERT model. The model features include typhoon level and impact radius, and builds a case similarity retrieval system. The simulation training system builds a Unity3D virtual environment, which includes 20 preset disaster scenarios. Trainees can use VR headsets to practice cross-node resource allocation processes. The system scores in real time and generates improvement reports.

[0044] The shared inventory electronic dashboard of the resource pooling scheduling module is connected to the WMS system of each node, and displays the available resources in three dimensions: material category (raw materials / semi-finished products / finished products), available quantity, and geographical location; the transportation capacity bidding platform is designed based on the smart contract of the sealed second-price auction (Vickrey auction). After the shipper submits the transportation demand, the platform completes the carrier matching and price settlement within 3 minutes; the capability certification system crawls the historical performance data (punctuality rate, cargo damage rate) of third-party logistics companies and uses the XGBoost model to calculate the credit score (ranging from 0 to 100 points). Companies with a score below 60 are not allowed to participate in emergency response.

[0045] See also Figure 3 , further, the specific workflow of the present invention is described in detail:

[0046] S1 uses a distributed data acquisition module to obtain inventory data, logistics status, and equipment operating parameters of each supply chain node in real time. The digital twin modeling engine constructs a dynamic virtual mapping model of the supply chain network based on this data. Its topology relationship construction unit uses a graph database to dynamically maintain multi-dimensional relationships between nodes, defining three types of edge attributes: logistics dependency intensity, capital settlement cycle, and information sharing level, supporting real-time topology reconstruction. The state synchronization unit adopts a publish-subscribe model and implements millisecond-level node status broadcasting through Apache Kafka. The prediction and deduction unit integrates the supply chain risk knowledge graph and combines it with the LSTM-GAN hybrid model to predict the interruption probability in the next 72 hours, outputting a risk heat map to provide data support for subsequent decision-making.

[0047] S2, the emergency response trigger identifies abnormal fluctuations through a three-level early warning mechanism and automatically triggers multi-node joint simulation and deduction. The risk propagation analysis submodule of the prediction and deduction unit constructs a directed weighted graph of the supply chain, uses the SIR infectious disease model to simulate the cascading effects of disruption events, and outputs a list of key vulnerable nodes. The resilience assessment submodule calculates the network's overall resilience index (RI) (0.6·redundancy+0.4·response speed) through Monte Carlo simulation. The scenario construction library calls a pre-trained neural network model and generates response strategies based on real-time meteorological and political data. The dynamic route planner of the collaborative decision-making center simultaneously integrates GPS, roadside unit, and weather radar data to construct a four-dimensional path cost function, providing real-time environmental parameters for simulation and deduction.

[0048] S3, the collaborative decision-making hub's multi-objective optimizer, transforms transportation costs, inventory turnover, and on-time delivery into mixed integer programming problems, solving them using a branch-and-price algorithm. The conflict resolution algorithm incorporates an improved Shapley value method, adds a time-decrease factor to node contributions, and dynamically adjusts benefit distribution weights to balance the demands of multiple parties. The resource pooling scheduling module locks available inventory at the three nearest nodes through smart contracts and allocates capacity using a Dutch auction algorithm to ensure fairness and efficiency in resource scheduling. The dynamic route planner generates an optimal path set every five minutes based on the Dijkstra-A* hybrid algorithm. Combined with an adaptive weight adjustment module and an alternative path generation algorithm, it ultimately outputs a Pareto-optimal solution.

[0049] S4, the system executes lightweight decision-making models locally at the node through the edge computing gateway, and provides real-time feedback on the execution data of the plan. The federated learning framework is based on the homomorphic encrypted gradient aggregation protocol, which uploads the encrypted model parameters of each node to the central server, aggregates and updates the global model to optimize the scheduling strategy. At the same time, the blockchain evidence storage unit records the entire life cycle data of the emergency event to ensure the traceability of the process. The digital certificate system uses ERC-1155 tokens and Solidity smart contracts to achieve minute-level service transaction settlement. The human-computer collaborative interface superimposes the actual execution effect on the digital twin, and combines the case similarity retrieval system of the knowledge graph engine to continuously improve the practice scenarios of the simulation training system to form a closed-loop optimization.

[0050] Through the above steps, millisecond-level data synchronization and edge computing achieve second-level response, thereby improving decision-making timeliness. Multi-agent game and federated learning are used to improve cross-node collaboration efficiency. Multi-objective Pareto optimization is used to improve the on-time rate of orders and the network resilience index under the premise of controllable costs, so as to solve the problems of traditional systems with insufficient real-time performance, limited collaborative capabilities and single optimization goals.

Claims

1. Supply chain multi-node real-time collaborative scheduling and emergency response system, characterized by: Includes: Distributed data acquisition module, used to obtain real-time inventory data, logistics status and equipment operating parameters of each node; A digital twin modeling engine for building dynamic virtual mapping models of supply chain networks; Collaborative decision-making center generates multi-objective optimization scheduling solutions based on reinforcement learning algorithms; Emergency response triggers automatically initiate hierarchical emergency plans through abnormal pattern recognition.

2. The supply chain multi-node real-time collaborative scheduling and emergency response system according to claim 1 is characterized by: The digital twin modeling engine includes a topology relationship construction unit, a state synchronization unit, and a prediction and deduction unit. The topology relationship construction unit dynamically maintains multi-dimensional associations between nodes through a graph database, and defines three types of edge attributes: logistics dependency intensity, capital settlement cycle, and information sharing level, supporting real-time topology reconstruction; the state synchronization unit adopts a publish-subscribe model and implements millisecond-level broadcasting of node status through Apache Kafka, with a synchronization delay of ≤300ms; the prediction and deduction unit integrates the supply chain risk knowledge graph, combines meteorological and political information, uses an LSTM-GAN hybrid model to predict the probability of interruption in the next 72 hours, and outputs a risk heat map.

3. The supply chain multi-node real-time collaborative scheduling and emergency response system according to claim 1 is characterized by: The collaborative decision-making hub includes a multi-objective optimizer, a conflict resolution algorithm, and a dynamic routing planner. The multi-objective optimizer transforms transportation costs, inventory turnover, and on-time delivery into a constrained mixed integer programming problem and solves it using a branch-and-price algorithm. The conflict resolution algorithm introduces an improved Shapley value method, adds a time-dependent decay factor to node contributions, and dynamically adjusts the weight of benefit distribution. The dynamic route planner generates an optimal path set every 5 minutes based on the real-time updated road congestion index and capacity utilization rate using the Dijkstra-A* hybrid algorithm, supporting dynamic rerouting during transportation.

4. The supply chain multi-node real-time collaborative scheduling and emergency response system according to claim 1 is characterized by: The emergency response trigger includes a three-level early warning mechanism, a resource pooling scheduling module, and a blockchain evidence storage unit. The three-level early warning mechanism sets inventory red lines, transportation delay thresholds, and capacity gap rates as grading standards to trigger resource allocation plans at corresponding levels. The resource pooling scheduling module establishes a virtual emergency warehouse, automatically locks the available inventory of the three nearest nodes through smart contracts, and allocates transportation capacity using a Dutch auction algorithm. The blockchain evidence storage unit deploys a Hyperledger Fabric private chain to record the entire life cycle data of emergency events from triggering to resolution, including decision logs, resource allocation vouchers, and post-audit traces.

5. The supply chain multi-node real-time collaborative scheduling and emergency response system according to claim 1 is characterized by: The system also includes an edge computing gateway, a federated learning framework, and a digital certificate system. The edge computing gateway uses the NVIDIA Jetson AGX Orin module to run lightweight decision-making models locally on the node with a processing delay of less than 50ms. The federated learning framework is designed based on a gradient aggregation protocol with homomorphic encryption. Each node uploads encrypted model parameters, and the central server aggregates and updates the global model. The digital certificate system issues ERC-1155 standard tokens to record service transactions between nodes and automatically settles through Solidity smart contracts, supporting minute-level settlement.

6. The supply chain multi-node real-time collaborative scheduling and emergency response system according to claim 2 is characterized by: The prediction and deduction unit includes a risk propagation analysis submodule, a resilience assessment submodule and a scenario construction library. The risk propagation analysis submodule constructs a directed weighted graph of the supply chain, uses the SIR infectious disease model to simulate the cascading effects of interruption events, and outputs a list of key vulnerable nodes; the resilience assessment submodule defines the node resilience index RI = α·redundancy + β·response speed (α=0.6, β=0.4), and calculates the overall resilience of the network through Monte Carlo simulation; the scenario construction library stores a variety of typical scenarios including port closures and chip shortages, and each scenario is associated with a pre-trained response strategy neural network model.

7. The supply chain multi-node real-time collaborative scheduling and emergency response system according to claim 3 is characterized by: The real-time traffic situation awareness module of the dynamic route planner integrates GPS floating vehicle data, emergency accident information reported by roadside units, and NOAA weather radar data to construct a four-dimensional (x, y, t, risk) path cost function; The adaptive weight adjustment module dynamically sets the cost-time weight ratio based on the urgency of the order, with the time weight for S-level orders reaching 90%. The alternative path generation algorithm maintains K = 5 optimal candidate paths. When the main path is interrupted, it switches to the suboptimal path within 500ms based on the reinforcement learning model.

8. The supply chain multi-node real-time collaborative scheduling and emergency response system according to claim 1 is characterized by: The system also includes a human-machine collaborative interface, a knowledge graph engine and a simulation training system. The human-machine collaborative interface provides a Hololens 2AR visualization interface, superimposes the scheduling plan on the real warehouse scene, and supports gesture operation to adjust the cargo allocation; the knowledge graph engine uses Neo4j to store emergency cases in the past five years, extracts event features through the BERT model, and builds a case similarity retrieval system; the simulation training system builds a Unity3D virtual environment, and trainees can use VR headsets to practice the cross-node resource allocation process. The system scores in real time and generates improvement reports.

9. The supply chain multi-node real-time collaborative scheduling and emergency response system according to claim 4 is characterized by: The shared inventory electronic dashboard of the resource pool scheduling module is connected to the WMS system of each node, displaying available resources in three dimensions: material category, available quantity and geographical location; The capacity bidding platform is designed based on a smart contract for sealed second-price auctions. After the shipper submits the transportation demand, the platform completes carrier matching and price settlement within 3 minutes; the capability certification system crawls the historical performance data of third-party logistics companies and uses the XGBoost model to calculate the credit score. Companies with a score below 60 are not allowed to participate in emergency response.

10. A scheduling method for a supply chain multi-node real-time collaborative scheduling and emergency response system, using the supply chain multi-node real-time collaborative scheduling and emergency response system according to claims 1-9, characterized in that: The following steps are included: S1 uses a distributed data acquisition module to obtain inventory data, logistics status, and equipment operating parameters of each supply chain node in real time. The digital twin modeling engine constructs a dynamic virtual mapping model of the supply chain network based on this data. Its topology relationship construction unit uses a graph database to dynamically maintain multi-dimensional relationships between nodes, defining three types of edge attributes: logistics dependency intensity, capital settlement cycle, and information sharing level, supporting real-time topology reconstruction. The state synchronization unit adopts a publish-subscribe model and implements millisecond-level node status broadcasting through Apache Kafka. The prediction and deduction unit integrates the supply chain risk knowledge graph and combines it with the LSTM-GAN hybrid model to predict the interruption probability in the next 72 hours, outputting a risk heat map to provide data support for subsequent decision-making. S2, the emergency response trigger identifies abnormal fluctuations through a three-level early warning mechanism and automatically triggers multi-node joint simulation and deduction. The risk propagation analysis submodule of the prediction and deduction unit constructs a directed weighted graph of the supply chain, uses the SIR infectious disease model to simulate the cascading effects of disruption events, and outputs a list of key vulnerable nodes. The resilience assessment submodule calculates the network's overall resilience index (RI) (0.6·redundancy+0.4·response speed) through Monte Carlo simulation. The scenario construction library calls a pre-trained neural network model and generates response strategies based on real-time meteorological and political data. The dynamic route planner of the collaborative decision-making center simultaneously integrates GPS, roadside unit, and weather radar data to construct a four-dimensional path cost function, providing real-time environmental parameters for simulation and deduction. S3, the collaborative decision-making hub's multi-objective optimizer, transforms transportation costs, inventory turnover, and on-time delivery into mixed integer programming problems, solving them using a branch-and-price algorithm. The conflict resolution algorithm incorporates an improved Shapley value method, adds a time-decrease factor to node contributions, and dynamically adjusts benefit distribution weights to balance the demands of multiple parties. The resource pooling scheduling module locks available inventory at the three nearest nodes through smart contracts and allocates capacity using a Dutch auction algorithm to ensure fairness and efficiency in resource scheduling. The dynamic route planner generates an optimal path set every five minutes based on the Dijkstra-A* hybrid algorithm. Combined with an adaptive weight adjustment module and an alternative path generation algorithm, it ultimately outputs a Pareto-optimal solution. S4, the system executes lightweight decision-making models locally at the node through the edge computing gateway, and provides real-time feedback on the execution data of the plan. The federated learning framework is based on the homomorphic encrypted gradient aggregation protocol, which uploads the encrypted model parameters of each node to the central server, aggregates and updates the global model to optimize the scheduling strategy. At the same time, the blockchain evidence storage unit records the entire life cycle data of the emergency event to ensure the traceability of the process. The digital certificate system uses ERC-1155 tokens and Solidity smart contracts to achieve minute-level service transaction settlement. The human-computer collaborative interface superimposes the actual execution effect on the digital twin, and combines the case similarity retrieval system of the knowledge graph engine to continuously improve the practice scenarios of the simulation training system to form a closed-loop optimization.

Images