Urban and rural digital fusion co-rich support system based on circulation network reconstruction

Abstract

The invention relates to the technical field of digital data processing, and particularly discloses an urban and rural digital fusion co-rich support system based on circulation network reconstruction, which comprises a digital twin mapping module, a network toughness evaluation module, a dynamic reconstruction decision module and a collaborative execution engine. Real-time mapping and state evaluation are performed by constructing a circulation network digital twinborn body, a network reconstruction scheme is generated based on multi-objective optimization, cross-system resource cooperative execution and closed-loop feedback are driven, self-adaptive and flexible intelligent reconstruction of an urban and rural circulation network along with environment change is realized, and the network reconstruction efficiency is improved. And the overall network toughness, the circulation efficiency and the balance of the co-richness target are improved.

Description

Technical Field

[0001] This invention belongs to the field of digital data processing technology, specifically relating to a support system for urban-rural digital integration and common prosperity based on the reconstruction of circulation networks. Background Technology

[0002] Within the strategic framework of regional economic development, the optimal allocation and efficient circulation of urban and rural factors of production are core issues. The traditional urban-rural circulation system relies on physical infrastructure and relatively rigid organizational networks, aiming to promote the two-way flow of factors such as goods, capital, and information between urban and rural areas in order to narrow the development gap and promote coordinated prosperity.

[0003] Urban-rural digital integration, as an emerging technological direction, aims to break down information barriers between urban and rural areas and improve circulation efficiency and service quality by utilizing digital technologies such as big data, the Internet of Things, and cloud computing. Its basic goal is to empower and upgrade the traditional circulation system by building digital information channels and service platforms, thereby more accurately matching supply and demand and reducing transaction costs.

[0004] Building a hierarchical, fixed-structure digital distribution network lacks dynamic adjustment and flexible response capabilities. When faced with seasonal peaks in agricultural product sales, cyclical fluctuations in consumer demand, or supply chain disruptions caused by sudden public events, the established network structure struggles to quickly reorganize resource pathways and collaborative relationships, leading to a sharp drop in distribution efficiency and slow response. Furthermore, traditional systems lack sufficient depth in integrating multi-source, heterogeneous distribution data, making it difficult to support real-time assessment and intelligent optimization decisions regarding network resilience, node efficiency, and distribution costs. Summary of the Invention

[0005] The purpose of this invention is to provide a support system for urban-rural digital integration and common prosperity based on the reconstruction of circulation networks, so as to solve the technical contradictions of the existing urban-rural digital circulation network structure being fixed, lacking dynamic flexibility, and unable to cope with complex and ever-changing environments, resulting in decreased circulation efficiency and slow response.

[0006] To achieve the above objectives, this invention provides a support system for urban-rural digital integration and shared prosperity based on the reconstruction of circulation networks. This system includes a digital twin mapping module, a network resilience assessment module, a dynamic reconstruction decision module, and a collaborative execution engine.

[0007] The digital twin mapping module is used to construct and continuously update a virtual mirror of the urban and rural circulation network. This module accesses and integrates multi-source heterogeneous data from IoT terminals, enterprise resource planning systems, traffic management systems, e-commerce platforms, and public service databases to map circulation nodes, connection paths, and factor flows in the physical world in real time within the virtual space. Circulation nodes include, but are not limited to, origin warehouses, wholesale markets, logistics centers, community retail outlets, and online platform stores. Connection paths encompass highways, railways, cold chain trunk lines, and last-mile delivery routes. Factor flows are specifically represented by real-time status data of commodity flow, capital flow, and information flow. The digital twin mapping module further establishes node attribute vectors, which include the node's geographical location, storage capacity, processing capacity, service type, and real-time load rate. Simultaneously, the module establishes connection path attribute vectors, which include path length, traffic capacity, real-time congestion index, transportation cost, and reliability score. All vector data is stored in time-series format, forming a full-dimensional, high-fidelity digital twin of the circulation network.

[0008] The network resilience assessment module is used to quantitatively assess and diagnose bottlenecks in the overall resilience of the network based on real-time network status data provided by the digital twin mapping module. This module first defines and calculates a set of core network resilience indicators. These indicators include global connectivity efficiency, critical node vulnerability index, and path redundancy. Global connectivity efficiency is measured by calculating the average efficiency of the shortest path between any two nodes in the network, where path efficiency is defined as the reciprocal of the path's actual throughput. The critical node vulnerability index is calculated as follows: The node and all its associated connections are simulated for removal, and the percentage decrease in global connectivity efficiency after removal is recalculated; this percentage is defined as the node's vulnerability index. Path redundancy is quantified by statistically analyzing the number of alternative paths connecting any two main nodes in the network and the ratio of their total throughput to the throughput of the main path.

[0009] The network resilience assessment module further runs a bottleneck prediction submodule based on stress testing. This submodule injects pre-set disturbance scenarios into the digital twin model, including simulating a surge in production in a specific production area, a temporary interruption of a main road, or a decline in the service capacity of a large logistics node, and observes and records the degradation of network performance indicators. Specifically, the performance indicators are the average order fulfillment time and the increase in unit commodity circulation cost. By analyzing the load changes of each node and path under different disturbances and their impact weights on global indicators, this module identifies potential bottlenecks and vulnerable links in the current network topology and generates a diagnostic report containing specific bottleneck locations, vulnerability levels, and impact ranges.

[0010] The dynamic reconfiguration decision module receives the diagnostic report output by the network resilience assessment module and generates the optimal network reconfiguration scheme based on preset shared-prosperity optimization objectives and real-time constraints. The core of this module is a multi-objective optimization solver. The shared-prosperity optimization objectives specifically include maximizing the overall network circulation efficiency, minimizing the service accessibility cost of nodes in remote rural areas, and balancing the resource load rate among nodes in different regions. Real-time constraints include the available capacity limit, node processing capacity limitations, the cost budget for the reconfiguration operation, and the maximum allowable service interruption time.

[0011] The decision-making process of the dynamic reconfiguration decision module is as follows: First, the current network state, diagnosed bottleneck information, optimization objectives, and constraints are encoded as a mixed-integer programming problem. Decision variables include whether to activate backup nodes, whether to add, delete, or adjust capacity allocation for specific paths, and whether to offload loads to highly vulnerable nodes. Further, the solver uses a branch-and-bound algorithm combined with heuristic rules to search the solution space for a Pareto optimal solution set that satisfies all constraints. For each candidate reconfiguration scheme, the solver performs rapid simulation by calling the digital twin mapping module to predict key performance indicators after implementation. Finally, the module selects the final reconfiguration scheme from the Pareto optimal solution set based on a comprehensive utility function. The comprehensive utility function is a weighted sum of the optimization objectives, and the weights can be dynamically configured according to policy guidance.

[0012] The collaborative execution engine decomposes the refactoring schemes generated by the dynamic refactoring decision module into executable instructions, driving relevant resources and systems in the physical world to respond collaboratively. This engine comprises an instruction distribution submodule, a resource scheduling submodule, and a feedback loop submodule. The instruction distribution submodule transforms the refactoring scheme into standardized application programming interface (API) calls or workflow tasks, distributing them to relevant logistics management systems, warehouse management systems, traffic signal control systems, and third-party service platforms. The resource scheduling submodule coordinates and locks in the physical resources required for scheme execution, including spare vehicles, temporary storage space, and additional manpower, and ensures the responsibility and settlement of resource usage through smart contracts.

[0013] The feedback closed-loop submodule monitors the status feedback of each stage during the execution of the plan in real time, including events such as instruction confirmation, resource availability, task start and completion. This submodule compares the execution status with the expected progress. If a significant deviation or new external disturbance is detected, it immediately sends an interrupt signal and updated environmental data to the dynamic reconfiguration decision module, triggering a new round of evaluation and decision-making cycle, thereby realizing closed-loop adaptive control from perception, evaluation, decision-making to execution.

[0014] In one embodiment of the present invention, the digital twin mapping module employs a spatiotemporal graph neural network model for data fusion and state prediction. This model takes historical and real-time node and path attribute vectors as input, captures the spatial structural dependencies of the network through graph convolutional layers, and captures temporal dynamic evolution through gated recurrent unit layers. The model outputs predicted node load and path throughput values ​​for multiple future time steps. These predicted values ​​serve as a forward-looking extension of the digital twin, providing predictive information about future states for network resilience assessment and dynamic reconstruction decisions.

[0015] In one embodiment of the present invention, the calculation of the vulnerability index of key nodes in the network resilience assessment module is modified by incorporating the importance weight of nodes under the goal of common prosperity. The importance weight is calculated comprehensively based on the per capita income level, agricultural product dependence, and historical circulation growth rate of the region served by the node. For nodes serving low-income, high-dependence regions, their weight is increased accordingly, giving them greater attention in vulnerability assessment and ensuring that network reconstruction decisions prioritize the circulation stability of weak links in the path to common prosperity.

[0016] In one embodiment of the present invention, the multi-objective optimization solver of the dynamic reconstruction decision module integrates a reinforcement learning-based policy network as an auxiliary decision-making unit. This policy network, through offline training, learns which reconstruction actions can consistently improve the overall utility function value under different network states and perturbation modes. During online decision-making, when traditional optimization algorithms take too long to solve, the policy network can directly output high-quality reconstruction action suggestions or provide excellent initial solutions for branch-and-bound algorithms, thereby ensuring real-time decision-making in complex scenarios.

[0017] As one embodiment of the present invention, the feedback closed-loop submodule of the collaborative execution engine establishes a blockchain-based instruction storage and execution traceability mechanism. All issued reconstruction instructions, resource scheduling contracts, and key milestone events of their execution status are encrypted and recorded on a distributed ledger. This ensures the auditability, immutability, and multi-party trustworthiness of the entire reconstruction process, providing a reliable execution environment for complex collaborations across administrative regions and business entities, and reducing the risk of collaborative friction and disputes.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention achieves high-fidelity virtualization of all elements and states of the urban and rural circulation physical network by constructing a real-time updated digital twin mapping module, providing a unified and accurate data foundation for subsequent analysis and decision-making. Furthermore, the network resilience assessment module not only calculates static indicators but also proactively predicts potential bottlenecks through simulated stress tests, transforming problem discovery from passive response to proactive early warning, significantly improving the system's ability to anticipate risks.

[0019] 2. The dynamic reconfiguration decision module proposed in this invention formalizes the network optimization problem into a multi-objective optimization model with multiple constraints. Its decision-making process comprehensively weighs multiple co-prosperity objectives such as efficiency, fairness, and cost. By solving for the Pareto optimal solution and making choices based on the comprehensive utility function, the system can automatically generate scientific and balanced reconfiguration schemes under complex constraints. This fundamentally overcomes the limitations of traditional fixed networks or adjustments based on only a single rule, and realizes the adaptive and flexible adjustment of network structure as the environment changes.

[0020] 3. The collaborative execution engine of this invention achieves precise integration and closed-loop control from digital decision-making to physical execution. It decomposes abstract reconstruction schemes into specific executable instructions across systems and coordinates and locks physical resources, ensuring the implementation of the scheme. In particular, the feedback closed-loop mechanism integrating blockchain technology constructs a trustworthy and traceable collaborative execution environment, effectively solving trust and compliance issues in cross-entity collaboration, ensuring the smooth implementation of the reconstruction process in complex organizational relationships, and thus truly realizing the intelligent, flexible, and shared-prosperity-oriented reconstruction capabilities of urban and rural digital circulation networks. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the dynamic reconfiguration decision module in this invention; Figure 3 This is a logical flowchart of the digital twin mapping and network resilience assessment in this invention; Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the collaborative execution engine in this invention. Detailed Implementation

[0022] This invention provides a support system for urban-rural digital integration and shared prosperity based on the reconstruction of circulation networks, the overall technical architecture of which is shown in the attached figure. Figure 1 To be continued Figure 4 As shown, the system consists of four core functional modules: a digital twin mapping module, a network resilience assessment module, a dynamic reconstruction decision module, and a collaborative execution engine. These modules are tightly coupled and linked in a closed loop through standardized data interfaces and event-driven mechanisms, collectively forming an adaptive intelligent system with perception, assessment, decision-making, and execution capabilities.

[0023] First, the digital twin mapping module, serving as the system's data foundation and virtual mirror construction unit, is responsible for high-fidelity, full-dimensional digital modeling and real-time synchronization of the urban-rural circulation network in the physical world. Please refer to the appendix. Figure 3This module continuously collects raw data streams from IoT terminals, enterprise resource planning (ERP) systems, traffic management systems, e-commerce platforms, and public service databases through a multi-source heterogeneous data access layer. IoT terminals are deployed at key distribution nodes such as production warehouses, wholesale markets, logistics centers, and community retail outlets to collect real-time sensor data on temperature, humidity, inventory levels, operational status, and equipment operating parameters. ERP systems provide structured business data such as order information, product categories, batch traceability, and supplier relationships. Traffic management systems output dynamic road network information such as road traffic status, traffic light timings, accident warnings, and traffic restriction policies. E-commerce platforms contribute consumer-side data such as user order behavior, regional best-selling trends, return rates, and delivery timeliness feedback. Public service databases cover macroeconomic and socioeconomic indicators such as administrative divisions, population distribution, per capita income, agricultural output, and infrastructure coverage.

[0024] After entering the digital twin mapping module, the aforementioned multi-source data first undergoes data cleaning and format standardization to eliminate noise issues such as timestamp discrepancies, inconsistent units, and missing fields. Subsequently, the module maps the cleaned data to a unified semantic space based on a predefined ontology model. Building upon this, the module constructs two types of core attribute vectors: node attribute vectors and connection path attribute vectors. The node attribute vectors represent the state of each circulation node in the form of quintuples, specifically including geographical location, storage capacity, processing capacity, service type (such as classification tags for direct delivery from production sites, regional distribution, community self-pickup, and online fulfillment), and real-time load rate (defined as the ratio of the current processing workload to the maximum processing capacity, ranging from 0 to 1). The connection path attribute vector also describes the characteristics of each connection path in a quintuple, including path length, capacity (unit: vehicles / hour or tons / day), real-time congestion index (calculated based on historical average speed and current measured speed, the higher the value, the more congested), transportation cost (unit: yuan / ton·km), and reliability score (calculated based on historical punctuality rate, accident rate, weather impact, etc., with a value range of 0 to 100).

[0025] All attribute vectors are organized into time series by timestamp and stored in a distributed time-series database. This database uses a columnar storage structure, supporting millisecond-level writes and efficient aggregation queries. The digital twin mapping module further utilizes a spatiotemporal graph neural network model to fuse historical and real-time vectors. This model contains two core components: a graph convolutional layer and a gated recurrent unit layer. The graph convolutional layer operates on the network topology at the current moment, capturing spatial dependencies between nodes through neighborhood aggregation operations; the gated recurrent unit layer expands along the time dimension, learning the dynamic evolution of node and path states. Model training employs supervised learning, using node load and path throughput capabilities for the next 15, 30, and 60 minutes as prediction targets. After training, the model is deployed on edge computing nodes, capable of outputting predicted values ​​for multiple future time steps online. These predicted values ​​are appended to the end of the digital twin's time series, forming a forward-looking extended state, providing future state predictions for subsequent modules.

[0026] Secondly, the network resilience assessment module, based on real-time and predictive network data output by the digital twin mapping module, quantitatively assesses the overall resilience level of the circulation network and diagnoses bottlenecks. Please refer to the appendix for further details. Figure 3 This module first calculates three core network resilience indicators: global connectivity efficiency, critical node vulnerability index, and path redundancy.

[0027] The calculation process for global connectivity efficiency is as follows: Assume there are a total of [number] networks in the network. If there are multiple circulation nodes, then all node pairs need to be calculated. , The shortest path between () and (). The weight of a path is defined as the reciprocal of its capacity; that is, the lower the capacity, the higher the path weight. The shortest path is solved using Dijkstra's algorithm. Let... For nodes To the node The path efficiency is defined as the shortest path length (i.e., the sum of weights). Global connectivity efficiency This is the arithmetic mean of the path efficiency for all non-diagonal node pairs, calculated using the following formula:

[0028] The higher this indicator, the stronger the overall network connectivity and the smoother the flow of information and goods.

[0029] The calculation of the critical node vulnerability index incorporates a common-prosperity-oriented importance weight correction mechanism. For any node k, it is first simulated to remove it from the network (while simultaneously removing all its associated connections), and then the global connectivity efficiency after removal is recalculated. Vulnerability Index Preliminary definition However, to reflect the goal of common prosperity, the index needs to be multiplied by an importance weight. . The vulnerability index is synthesized by weighting three sub-factors: the inverse of the region's per capita income level (normalized, with higher weights for lower-income regions), dependence on agricultural product exports (defined as the proportion of agricultural product circulation volume processed at this node to the total circulation volume), and the historical average monthly growth rate of circulation volume over the past 12 months. The three sub-factors are assigned weights of 0.4, 0.4, and 0.2 respectively, and after linear weighting, are scaled to a range of 1 to 3. The final vulnerability index is... This design ensures that nodes serving economically disadvantaged areas with high agricultural dependence receive higher priority in the assessment.

[0030] Path redundancy is calculated for any pair of major nodes (such as a provincial capital and a county center, or a large production area and a core consumption area). Let the capacity of the main path be... The set of alternative paths is Each path The passage capacity is Then path redundancy Defined as If no alternative path exists, then =0. This metric reflects the network's fault tolerance capability when the main path fails.

[0031] After completing the static index calculations, the network resilience assessment module initiates the bottleneck prediction submodule. This submodule pre-defines several typical disturbance scenarios, including a 200% surge in citrus production in a major producing area and a 72-hour closure of a section of the Beijing-Hong Kong-Macau Expressway due to construction. For each scenario, the module injects corresponding parameter disturbances into the digital twin (such as increasing the output load of a specific node, setting the capacity of a path to zero, or reducing the processing capacity of a node to 30%) and runs a lightweight simulation engine. During the simulation, the module continuously monitors two key performance degradation indicators: average order fulfillment time (the average time from order placement to receipt, in hours) and unit commodity circulation cost (total circulation cost divided by total circulation tons, in yuan / ton). By analyzing the load change rate of each node under disturbances, the magnitude of changes in path utilization, and their sensitivity to global indicators, the module identifies high-risk bottlenecks. For example, if a rural distribution center's load rate soars to 120% in the citrus surge scenario, and its vulnerability index exceeds the threshold of 85, it is marked as a highly vulnerable bottleneck node. Finally, the module generates a structured diagnostic report, which includes the bottleneck location, vulnerability level (high / medium / low), impact range (measured by the number of affected nodes and the area), and recommended intervention directions (such as activating backup nodes, increasing capacity, and adjusting routing strategies).

[0032] Third, the dynamic reconfiguration decision module receives the aforementioned diagnostic report and, based on the current network state, optimization objectives, and constraints, generates the optimal network reconfiguration scheme. Please refer to the appendix. Figure 2 The core of this module is a multi-objective mixed-integer programming solver. The optimization objectives include three aspects: maximizing the overall network efficiency (i.e., maximizing...). The objectives are to minimize the service accessibility cost of remote rural nodes and balance the resource load rate among nodes in different regions. Service accessibility cost is defined as the weighted distance from any rural node to the nearest logistics node at level three or higher, with weights taking into account terrain difficulty, road grade, and public transportation coverage. Load balancing is measured by the reciprocal of the standard deviation of the average load rate of nodes in each region; the smaller the standard deviation, the higher the load balancing.

[0033] The constraints include four categories: the upper limit of available capacity (determined by the total number of currently registered vehicles and driver scheduling), hard limits on node processing capacity (no overload operation is allowed), budget for reconfiguration operation costs (such as the rental cost of activating backup warehouses, and the total cost of fuel and labor for temporarily dispatched vehicles must not exceed the preset limit), and the maximum allowable service interruption time (service suspension caused by any reconfiguration operation must not exceed 2 hours).

[0034] The decision variables are a mixture of discrete and continuous types, specifically including: binary variables. This indicates whether to enable the first option. One backup node (0 for no, 1 for yes); continuous variable , indicating that it is assigned to the first The additional capacity percentage for each connecting route (values ​​ranging from 0 to 1); and integer variables. , representing the amount of tasks (in units per day) that are diverted from the highly vulnerable node k to other nodes.

[0035] The solver first encodes the problem into a standard mixed-integer programming form, with the objective function being a weighted sum of three optimization objectives, and the weight coefficients... , , It can be dynamically allocated according to the provincial common prosperity policy (such as...) =0.5, =0.3, =0.2). Subsequently, the solver uses a branch and bound algorithm to search for feasible solutions that satisfy all constraints in the solution space. To accelerate convergence, the solver integrates a reinforcement learning-based policy network as an auxiliary decision-making unit. This policy network takes the network state vector (including node load, path congestion, vulnerability index, etc.) as input and outputs the probability distribution of reconstruction actions. The network learns a long-term utility maximizing policy on millions of historical perturbation-response samples through offline training. During online decision-making, if the branch and bound algorithm does not find a feasible solution within 5 seconds, the policy network directly outputs the highest probability action; if a partial solution has been found, it is used as the initial solution to guide the search. For each candidate solution, the solver calls the digital twin mapping module to perform a fast forward simulation (simulating only the next 2 hours, with a step size of 5 minutes) to predict the outcome after the solution is implemented. Service accessibility cost and load standard deviation. Finally, the module is based on the comprehensive utility function:

[0036] For the predicted cost of service accessibility, The scheme with the highest utility value is selected as the final output, and the variable marked with an apostrophe is the simulation prediction value.

[0037] Fourth, the collaborative execution engine is responsible for translating abstract refactoring schemes into collaborative actions in the physical world. Please refer to the appendix. Figure 4 The engine includes an instruction dispatch submodule, a resource scheduling submodule, and a feedback loop submodule.

[0038] The instruction dispatch submodule parses the decision variables in the reconstruction scheme and generates a standardized instruction set. For example, if If =1, then an instruction to activate the backup warehouse #n is generated, encapsulated as a RESTful API call, with the target address being the interface endpoint of the corresponding warehouse management system; if =0.3, then an instruction to increase the capacity of cold chain trunk line #e by 30% is generated, which is converted into the green wave band adjustment parameters of the traffic signal control system; if If the value is 5000, then 5000 orders will be generated from the node. Traffic is routed to the node. and The workflow tasks are pushed to the logistics company's task scheduling platform. All instructions include a unique transaction identifier, execution time limit, and confirmation requirements.

[0039] The resource scheduling submodule is responsible for coordinating the implementation of physical resource guarantee instructions. When a backup warehouse needs to be activated, the module queries the resource pool for available temporary storage space (including idle factory buildings, community activity center conversion sites, etc.), automatically locks the right to use it through a smart contract, and the contract terms include the usage duration, payment method (billed by the hour), and division of safety responsibilities. For capacity expansion needs, the module connects to a shared transportation platform, publishes bidding tasks, selects carriers with high credit ratings and matching vehicle types, and signs electronic waybills via blockchain. In terms of manpower scheduling, the module links with a labor platform to push temporary employment needs to nearby registered riders or warehouse workers' terminals based on task location and skill requirements.

[0040] The feedback loop submodule establishes an end-to-end execution monitoring chain. Each issued instruction, after execution by the target system, must return a structured status code (e.g., received, resources in place, task started, completed, execution failed). The submodule compares the actual progress with the expected timeline in real time. If any critical node delay exceeds a threshold (e.g., 15 minutes), or if a new external disturbance is received (e.g., a red rainstorm warning issued by the meteorological department), an interruption mechanism is immediately triggered. The interruption signal, along with an updated environmental snapshot (containing the latest node status, path status, and disturbance details), is sent back to the dynamic reconstruction decision module, initiating a new round of evaluation-decision-making. All key events in the entire execution process—including instruction issuance, contract signing, resource locking, and task completion—are hashed and encrypted before being written to the consortium blockchain ledger. This ledger is jointly maintained by local governments, logistics companies, e-commerce platforms, and other parties to ensure that all operations are traceable and tamper-proof, providing legally credible credentials for cross-entity collaboration.

[0041] In summary, this embodiment constructs a high-fidelity virtual network through a digital twin mapping module, achieves proactive bottleneck early warning through a network resilience assessment module, generates multi-objective optimization schemes through a dynamic reconfiguration decision module, and ensures accurate cross-system implementation through a collaborative execution engine. These four components form a closed-loop adaptive control flow. The system not only improves circulation efficiency but also ensures that the benefits of network reconfiguration are tilted towards rural and disadvantaged areas through a shared-prosperity weighting mechanism and service accessibility optimization, truly providing technical support for urban-rural digital integration.

[0042] Based on the aforementioned embodiments, this embodiment further refines the specific implementation architecture and training mechanism of the spatiotemporal graph neural network in the digital twin mapping module, and enhances the network resilience assessment module's ability to cope with extreme complex disturbances.

[0043] The spatiotemporal graph neural network used in the digital twin mapping module consists of three stacked layers. The first layer is the spatiotemporal embedding layer, which maps the original node attribute vectors and path attribute vectors to a 128-dimensional latent space. Node embedding uses a fully connected neural network, while path embedding combines the embeddings of the start and end nodes with the path's own attributes before mapping. The second layer is the core spatiotemporal convolutional layer, which contains two sub-modules: a spatial attention graph convolutional module and a temporal gated recurrent module. In each iteration, the spatial attention graph convolutional module assigns dynamic weights to the neighbors of each node. The weight calculation is based on node type similarity, path reliability score, and real-time load difference, and is normalized using the softmax function. This mechanism allows the model to focus on the neighboring nodes that have the greatest impact on the current state. The temporal gated recurrent module uses two stacked gated recurrent units with a hidden state dimension of 256 to capture long-term dependencies of node states. The third layer is the prediction decoding layer, which consists of two fully connected network layers and outputs the future... =The node load prediction and path capacity prediction values ​​are calculated at 12 time steps (one step every 5 minutes). The model loss function uses weighted mean square error, giving higher weight to high load periods (such as 10:00-14:00 daily).

[0044] The model training employs a two-stage strategy. The first stage involves offline pre-training on three years of historical data, with the data divided into training, validation, and test sets in an 8:1:1 ratio. The Adam optimizer is used, with an initial learning rate of 0.001 and a batch size of 64, training for 200 epochs until the validation loss converges. The second stage involves online fine-tuning. Every day at midnight, the system incrementally updates the model parameters using the previous day's actual data, reducing the learning rate to 0.0001 and training for only 10 epochs to avoid overfitting. The model is deployed on a Kubernetes cluster, supporting horizontal scaling to handle high-concurrency prediction requests.

[0045] In the network resilience assessment module, this embodiment introduces a composite disturbance stress testing mechanism. Traditional single disturbance tests are insufficient to reflect the complexity of the real world; therefore, the module constructs a library of composite disturbance scenarios, such as the triple superposition of main road disruption, extreme high temperatures, and e-commerce promotions. Each composite disturbance is composed of multiple basic disturbances combined according to logical relationships. The module uses the Monte Carlo method to randomly sample disturbance combinations, injecting 3 to 5 simultaneously occurring disturbance events in each simulation. The simulation engine is upgraded to an event-driven architecture, supporting causal propagation between disturbances. For example, a main road disruption causes vehicles to detour, which in turn triggers secondary road congestion, ultimately increasing last-mile delivery costs. The module records the nonlinear degradation curve of network performance under composite disturbances and calculates the resilience collapse threshold—the minimum combination of disturbance strengths required for performance indicators to drop to 50% of normal values. This threshold serves as the ultimate measure of network resilience, guiding long-term infrastructure investment planning.

[0046] Furthermore, the reinforcement learning policy network for the dynamic reconstruction decision module employs deep learning. Network architecture. The state space contains 100-dimensional network feature vectors, and the action space is discretized into 20 typical reconstruction operations (such as activating class A backup warehouses, increasing the capacity of the BC route by 20%, etc.). The reward function is designed as the increment of the comprehensive utility function, i.e. , This represents the overall utility function value of the network at time t (i.e., before the current reconstruction action is performed). The network stores transition samples through an experience replay mechanism. , Indicates the state at a given moment. Indicates the moment of action. Indicates a reward at any time. This indicates the current state at any given moment, and a batch update is performed every 1000 steps. To improve exploration efficiency, the following approach is adopted: Strategy, Initial =0.9, which decays exponentially to 0.01 with the number of training steps. During online inference, the policy network outputs an action suggestion every 500 milliseconds to ensure a response to sudden disturbances within 1 second.

[0047] The collaborative execution engine's blockchain notarization mechanism utilizes the Hyperledger Fabric consortium blockchain framework. In terms of channel design, an independent channel is established for each city to ensure data isolation. Smart contracts are written in Go and include three main functions: instruction verification, resource locking, and state updating. All transactions are written to the ledger after consensus is reached through a sorting service, and a Merkle tree structure ensures data integrity. Auditors (such as market regulators) can verify execution compliance in real time through read-only nodes, without relying on a centralized log system.

[0048] Through the above enhancements, this embodiment significantly improves the system's accuracy in predicting future states, the depth of its assessment of extreme and complex risks, and the robustness of its decision-making execution, providing stronger technical support for the stable operation of urban and rural circulation networks in highly uncertain environments.

Claims

1. A support system for urban-rural digital integration and shared prosperity based on the reconstruction of circulation networks, characterized in that: include: The digital twin mapping module is used to build and continuously update a virtual mirror of the urban and rural circulation network; The network resilience assessment module is used to quantitatively assess and diagnose bottlenecks in the overall resilience of the circulation network based on the real-time network status data provided by the digital twin mapping module. The dynamic reconfiguration decision module is used to receive the diagnostic report output by the network resilience assessment module and generate the optimal network reconfiguration scheme based on the preset co-enrichment optimization objective and real-time constraints. The collaborative execution engine is used to decompose the reconstruction scheme generated by the dynamic reconstruction decision module into executable instructions and drive relevant resources and systems in the physical world to respond collaboratively.

2. The urban-rural digital integration and shared prosperity support system based on the reconstruction of circulation networks according to claim 1, characterized in that, The digital twin mapping module accesses and integrates multi-source heterogeneous data from IoT terminals, enterprise resource planning systems, traffic management systems, e-commerce platforms, and public service databases to map circulation nodes, connection paths, and element flows in the physical world in real time in virtual space. The module establishes node attribute vectors, which include the node's geographical location, storage capacity, processing capacity, service type, and real-time load rate. It also establishes connection path attribute vectors, which include path length, traffic capacity, real-time congestion index, transportation cost, and reliability score. All vector data is stored in time-series format, forming a full-dimensional digital twin of the circulation network.

3. The urban-rural digital integration and common prosperity support system based on circulation network reconstruction according to claim 2, characterized in that, The network resilience assessment module defines and calculates a set of core network resilience indicators, including global connectivity efficiency, critical node vulnerability index, and path redundancy. The network resilience assessment module runs a bottleneck prediction submodule based on stress testing. The bottleneck prediction submodule injects preset disturbance scenarios into the digital twin model, observes and records the decay of network performance indicators, and identifies potential bottlenecks and vulnerable links in the current network topology by analyzing the load changes of each node and path under different disturbances and their impact weights on global indicators. It then generates a diagnostic report containing the specific bottleneck location, vulnerability level, and impact range.

4. The urban-rural digital integration and common prosperity support system based on the reconstruction of circulation networks according to claim 3, characterized in that, The core of the dynamic reconfiguration decision module is a multi-objective optimization solver. The shared-rich optimization objectives include maximizing the overall network circulation efficiency, minimizing the service accessibility cost of remote rural nodes, and balancing the resource load rate among nodes in different regions. The real-time constraints include the upper limit of available capacity, node processing capacity limits, cost budget for reconfiguration operations, and the maximum allowable service interruption time. The decision-making process of the dynamic reconfiguration decision module is as follows: the current network state, diagnosed bottleneck information, optimization objectives, and constraints are encoded into a mixed-integer programming problem. The solver uses a branch-and-bound algorithm combined with heuristic rules to search for a Pareto optimal solution set that satisfies all constraints in the solution space. For each candidate reconfiguration scheme, the solver performs rapid simulation by calling the digital twin mapping module to predict the key performance indicators after the scheme is implemented. Finally, the dynamic reconfiguration decision module selects a final reconfiguration scheme from the Pareto optimal solution set based on the comprehensive utility function.

5. A support system for urban-rural digital integration and shared prosperity based on the reconstruction of a circulation network, as described in claim 4, is characterized in that... The collaborative execution engine comprises an instruction distribution submodule, a resource scheduling submodule, and a feedback loop submodule. The instruction distribution submodule transforms the refactoring scheme into standardized application programming interface (API) call instructions or workflow tasks, which are then distributed to relevant logistics management systems, warehouse management systems, traffic signal control systems, and third-party service platforms. The resource scheduling submodule is responsible for coordinating and securing the physical resources required for scheme execution, including spare vehicles, temporary storage space, and additional manpower, and ensures the rights, responsibilities, and settlement of resource usage through smart contracts. The feedback loop submodule monitors the status feedback of each stage during scheme execution in real time and compares the execution status with the expected progress. If a significant deviation or new external disturbance is detected, an interruption signal and updated environmental data are immediately sent to the dynamic refactoring decision module, triggering a new round of evaluation and decision-making.

6. The urban-rural digital integration and common prosperity support system based on circulation network reconstruction according to claim 5, characterized in that, The digital twin mapping module employs a spatiotemporal graph neural network model for data fusion and state prediction. The spatiotemporal graph neural network model takes historical and real-time node attribute vectors and connection path attribute vectors as inputs, captures the spatial structure dependence of the network through graph convolutional layers, and captures the dynamic evolution over time through gated recurrent unit layers. The spatiotemporal graph neural network model outputs node load prediction values ​​and path traversability prediction values ​​for multiple future time steps, which serve as a forward-looking extension of the digital twin.

7. A support system for urban-rural digital integration and shared prosperity based on the reconstruction of a circulation network, as described in claim 6, is characterized in that... The training of the spatiotemporal graph neural network model adopts a two-stage strategy: the first stage is offline pre-training on historical data; the second stage is online fine-tuning, in which the system incrementally updates the model parameters daily using the actual data from the previous day.

8. A system for supporting urban-rural digital integration and shared prosperity based on the reconstruction of a circulation network, as described in claim 7, is characterized in that... The calculation of the vulnerability index of key nodes in the network resilience assessment module is modified by incorporating the importance weight of nodes under the goal of common prosperity; the importance weight is calculated based on the per capita income level, agricultural product dependence and historical circulation growth rate of the area served by the node.

9. A support system for urban-rural digital integration and shared prosperity based on the reconstruction of a circulation network, as described in claim 8, is characterized in that... The calculation process of the critical node vulnerability index is as follows: simulate the removal of the node and all its associated connections, recalculate the percentage decrease in global network connectivity efficiency after removal, and define the vulnerability index of the node as such; the final vulnerability index is the vulnerability index multiplied by the importance weight.

10. A support system for urban-rural digital integration and shared prosperity based on the reconstruction of a circulation network, as described in claim 9, is characterized in that... The multi-objective optimization solver of the dynamic reconstruction decision module integrates a policy network based on reinforcement learning as an auxiliary decision-making unit. The policy network is trained offline to learn which reconstruction actions can stably improve the comprehensive utility function value in the long term under different network states and perturbation modes. During online decision-making, when the branch and bound algorithm takes too long to solve, the policy network directly outputs high-quality reconstruction action suggestions or provides a high-quality initial solution for the branch and bound algorithm.

Images