A method for cross-border logistics information management based on a logistics management platform
By constructing a spatiotemporal alignment and virtual mapping topology for multi-source heterogeneous data in cross-border logistics, and combining Monte Carlo simulation and the improved SPEA2 algorithm, the problems of insufficient delay transmission analysis and weak reassignment decision-making mechanism in cross-border logistics are solved. Dynamic quantification and multi-objective optimization of default risk are realized, and the intelligent scheduling capability of cross-border logistics is improved.
Patent Information
- Application Number
- CN202611023094.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-10
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-07-10
AI Technical Summary
Existing logistics management platforms lack in-depth delay transmission analysis and forward-looking risk quantification capabilities in cross-border logistics. Their reassignment decision-making mechanisms have weak global optimization capabilities, making it difficult to achieve global optimal balance. Furthermore, multi-objective decisions are prone to getting stuck in local optima.
By constructing a spatiotemporal alignment and virtual mapping topology for multi-source heterogeneous data, combined with Monte Carlo stochastic simulation and the improved SPEA2 algorithm, and employing a topology truncation mechanism based on a dynamic reference grid, cross-border logistics route reconstruction is optimized, enabling dynamic quantification of default risk and adaptive route reconstruction for multi-objective decision-making.
It significantly improves the timeliness of default risk identification and the global convergence capability of multi-objective decision-making in cross-border logistics scenarios, enhances the balance between economic efficiency and timeliness of relocation schemes, and realizes comprehensive perception and intelligent scheduling optimization of cross-border logistics.
Smart Images

Figure CN122529417B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart logistics and supply chain management, and in particular to a method for cross-border logistics information management based on a logistics management platform. Background Technology
[0002] With the deepening of globalization, cross-border logistics has expanded rapidly, characterized by long operational chains, numerous participants, and significant susceptibility to uncertain external disturbances. To ensure supply chain resilience, existing logistics management platforms typically rely on multi-source data collection and processing technologies. These platforms aggregate heterogeneous data such as order status, physical trajectory, and load factor from various nodes, storing and visualizing it to assist dispatchers in monitoring the real-time location and node status of packages. When local congestion or delays occur in the logistics network, existing dispatch systems often use static thresholds or experience-based rules for early warning, with manual judgment required to trigger reassignment operations and select alternative routes. While this model meets basic logistics monitoring and dispatching needs to some extent, it has gradually revealed significant shortcomings when facing highly dynamic and tightly coupled cross-border logistics networks.
[0003] The primary drawback of existing technologies lies in their lack of in-depth analysis of delay propagation and proactive risk quantification capabilities. Traditional methods often remain at the level of passive monitoring of the current state of a single node, failing to construct a virtual mapping topology reflecting the global physical network. This makes it difficult to accurately capture the spatiotemporal dependence and propagation mechanisms of congestion between upstream and downstream nodes. Regarding estimated arrival times and default risk assessment, existing technologies often rely on historical average durations or simple linear predictions, neglecting the probabilistic distribution characteristics of dwell times at each node and transportation durations on each route. This deterministic prediction cannot objectively reflect the randomness and cumulative lag effects in the real logistics environment, leading to severely delayed default risk warnings. Systems often only passively trigger intervention after delays actually occur, missing the golden window for early scheduling and loss mitigation.
[0004] Another core flaw in existing technologies lies in the weak global optimization capability of reassignment decision-making mechanisms, which are prone to getting trapped in local optima. Cross-border route reconfiguration involves multiple conflicting optimization objectives, such as capacity reserves, customs clearance speed, and reassignment costs. Existing systems typically use weighted summation to transform multi-objective problems into single-objective problems. This subjective weighting approach severs the real competitive relationships between objectives and makes it difficult to present a complete Pareto front for decision-makers to choose from. When introducing evolutionary algorithms for multi-objective optimization, the truncation mechanism of traditional algorithms relies on global density estimation, which faces problems of computational redundancy and imbalance in diversity maintenance as the population iterates. When the number of archived individuals exceeds the limit, traditional clustering truncation often indiscriminately deletes individuals in dense regions, easily deleting grid boundary extreme points representing extreme trade-offs. This causes the algorithm to converge to local optima and the solution set distribution to degenerate, making it difficult for the generated route reconfiguration scheme to achieve a global optimal balance between timeliness and economy.
[0005] Therefore, how to provide a cross-border logistics information management method based on a logistics management platform is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0006] One objective of this invention is to propose a cross-border logistics information management method based on a logistics management platform. This invention fully integrates key steps such as spatiotemporal alignment of multi-source heterogeneous data, virtual mapping topology delay propagation analysis, Monte Carlo stochastic simulation, improved SPEA2 algorithm optimization, and control command issuance and execution. It constructs a logistics adaptive routing reconstruction process with dynamic quantification of default risk, adaptive grid partitioning of the target space, and Pareto optimal search for routing schemes. This achieves comprehensive perception of the entire link delay status, proactive risk warning, and intelligent scheduling optimization without sharing the original plaintext set. This invention improves the external archive truncation operation by introducing a topology truncation mechanism based on a dynamic reference grid. It prioritizes deleting individuals within highly crowded grids and forcibly retains grid boundary extreme points. It possesses advantages such as probabilistic risk quantification, high adaptability of grid partitioning, strong diversity of optimal solution sets, and complete preservation of boundary solutions. This significantly improves the timeliness of default risk identification, global convergence capability of multi-objective decision-making, and the balance between economic efficiency and timeliness of relocation schemes in complex cross-border logistics scenarios. Therefore, it effectively solves problems such as delayed delay propagation perception, distorted deterministic risk assessment, and easy loss of boundary solutions during truncation operations in existing methods.
[0007] A cross-border logistics information management method based on a logistics management platform according to an embodiment of the present invention includes the following steps: S1. Collect multi-source heterogeneous data from the entire cross-border logistics chain, extract features from the status data of various logistics node entities, and obtain local status feature data. S2. Perform unified alignment and fusion processing of local status feature data and multi-source heterogeneous data in terms of timestamps and spatial coordinates, and map them to the global cross-border logistics spatiotemporal coordinate system to obtain a standardized logistics status dataset. S3. Based on the standardized logistics status dataset, construct a virtual mapping topology, analyze the spatiotemporal dependence and physical transmission relationship between each logistics node, calculate and extract the capacity congestion and delay transmission status in the logistics link, and obtain full-link status evolution characteristic data. S4. Based on the full-link state evolution characteristic data, perform multiple random simulations and statistical calculations on the running trajectory of each node to obtain the probability distribution range of the dynamically predicted arrival time and the potential default risk assessment parameters. S5. Based on the probability distribution range of the dynamically predicted arrival time and the default risk assessment parameters, determine whether the package triggers the service level agreement default risk, construct a multi-objective optimization function, use the improved SPEA2 algorithm for comprehensive optimization calculation, introduce a topology truncation mechanism based on dynamic reference grid, adaptively divide multi-dimensional grid and calculate the local congestion coefficient of each grid, search the Pareto optimal solution set, and obtain the optimal cross-border logistics adaptive route reconstruction scheme. S6. Generate reassignment control instructions based on the optimal cross-border logistics adaptive routing reconstruction scheme, and execute logistics reassignment operations.
[0008] Optionally, S1 specifically includes: S11. Collect order status, physical trajectory coordinates, vehicle load rate and speed, and external disturbance factors for each node as multi-source heterogeneous data. S12. Build data cleaning rules on the local side of the logistics node, extract status data representing the operation of various logistics node entities from multi-source heterogeneous data, identify and remove missing values and abnormal jump values exceeding the preset abnormal threshold in the status data, and complete the preprocessing of the status data. S13. Without centrally sharing the original plaintext data of each logistics provider, read the preprocessed status data on the local side of the logistics node, and calculate the temporal change rate and state extreme value of the status data within the preset time window as feature dimensions. S14. Perform normalization mapping on the extracted feature dimensions, and then concatenate and combine them according to the entity identifier of the logistics node to output the local state feature data.
[0009] Optionally, S2 specifically includes: S21. Extract time information from local state feature data and multi-source heterogeneous data, find the absolute timestamp corresponding to the generation of each data record, set a unified standard time base, calculate the difference between the timestamp of each data record and the standard time base, remove data records with a difference greater than the preset alignment interval, and retain data records with a difference less than or equal to the preset alignment interval to complete the unified alignment of timestamps. S22. Read the physical trajectory coordinates from the multi-source heterogeneous data, substitute the longitude and latitude values of the physical trajectory coordinates into the preset coordinate transformation formula, calculate and convert them into standard longitude and standard latitude in the global cross-border logistics spatiotemporal coordinate system, and complete the unified alignment of spatial coordinates. S23. Bind the local state feature data to the unified aligned timestamp and spatial coordinates of the corresponding data record, identify the local state feature data with the same timestamp and spatial coordinates, and merge the identified local state feature data. S24. Combine the merged local state feature data with the corresponding standard longitude and standard latitude column by column, map them to the global cross-border logistics spatiotemporal coordinate system, and output the standardized logistics state dataset.
[0010] Optionally, S3 specifically includes: S31. Read the standardized logistics status dataset, extract the node identifier and standard longitude and standard latitude corresponding to each data record in the standardized logistics status dataset, and use each node identifier as a vertex of the virtual mapping topology. S32. Find two data records in the standardized logistics status dataset that have the same package identifier and consecutive timestamps. Connect the two vertices corresponding to these two data records with line segments, and use the line segments as edges of the virtual mapping topology to complete the virtual mapping topology. S33. Calculate the difference between the standard longitude and standard latitude between the two vertices corresponding to each edge in the virtual mapping topology, use the difference to calculate the physical space distance between the two vertices, extract the difference between the timestamps of the data records corresponding to the two vertices as the actual transmission duration, divide the physical space distance by the actual transmission duration to calculate the transmission speed, and use the ratio of the transmission speed to the preset reference speed as the physical transmission relationship. S34. Read the load factor of the transportation vehicle corresponding to each vertex in the standardized logistics status dataset, and mark the status data where the load factor of the transportation vehicle is greater than the preset load threshold as the capacity congestion status. S35. In the virtual mapping topology, find the vertex with a capacity congestion state as the congestion source node, read the vehicle load rate of the downstream vertex directly connected to the congestion source node through the edge, and when the vehicle load rate of the downstream vertex increases beyond the preset increase range within the preset observation period, mark the downstream vertex as a capacity congestion state, and complete the extraction of capacity congestion and delay transmission state. S36. Combine the capacity congestion status, physical transmission relationship, and corresponding unified aligned timestamps with standard longitude and standard latitude of each vertex in the virtual mapping topology in columns to output the full-link state evolution characteristic data.
[0011] Optionally, S4 specifically includes: S41. Extract the virtual mapping topology corresponding to the package identifier in the full-link state evolution feature data, and use the virtual mapping topology corresponding to the package identifier as the running trajectory of each node. S42. Read the physical transmission relationship and capacity congestion status corresponding to each vertex of the virtual mapping topology in the full-link state evolution feature data, find the records in the historical database that are the same as the physical transmission relationship and capacity congestion status, calculate the mean and standard deviation of the dwell time of the same records in the historical database at each vertex, and use the mean and standard deviation of the dwell time as the probability distribution parameters of the dwell time of each vertex. S43. Read the physical transmission relationship corresponding to each edge of the virtual mapping topology in the full-link state evolution feature data, find the record with the same physical transmission relationship in the historical database, calculate the mean and standard deviation of the transportation time of the same record on each edge in the historical database, and use the mean and standard deviation of the transportation time as the probability distribution parameter of the transportation time of each edge. S44. Based on the mean and standard deviation of the probability distribution parameters of dwell time for each vertex, generate a preset number of random dwell time sampling values using a normal distribution random number generation function. Based on the mean and standard deviation of the probability distribution parameters of transportation time for each edge, generate a preset number of random transportation time sampling values using a normal distribution random number generation function. In each sampling, the random dwell time sampling values of all vertices on the running trajectory of each node corresponding to the package identifier and the random transportation time sampling values of all edges are added together sequentially to calculate the total transportation time of a single simulation. S45. Sort the total transportation time obtained from all preset number of simulations in ascending order of value, select a preset specific percentile of the total transportation time value as the boundary, and construct the probability distribution interval of the dynamically predicted arrival time. S46. Read the promised arrival time threshold specified in the cross-border logistics service level agreement, count the number of times the total transportation time derived from all preset number of times exceeds the promised arrival time threshold, divide the number of times by the preset number of times to calculate the ratio, and use it as a parameter for assessing potential default risk.
[0012] Optionally, S5 specifically includes: S51. Subtract the preset default threshold from the potential default risk assessment parameters to obtain the difference. If the difference is greater than 0, the package is determined to have triggered the service level agreement default risk; otherwise, it is determined not to have triggered the service level agreement default risk. S52. If it is determined that the package triggers the risk of service level agreement default, then the transit hub, alternative customs clearance port and alternative route are used as alternative decision variables. Read the real-time capacity reserve, customs clearance speed and relocation cost corresponding to each alternative decision variable. Use the real-time capacity reserve as the first objective component, the customs clearance speed as the second objective component, and the relocation cost multiplied by negative 1 to obtain the negative relocation cost as the third objective component. Add the first objective component, the second objective component and the third objective component to obtain the summation result, and use it as the fitness value of the multi-objective optimization function to complete the construction of the multi-objective optimization function. S53. Use alternative decision variables to combine the initial population and substitute them into the multi-objective optimization function for calculation. Select individuals in the initial population one by one and mark them as dominated individuals and non-dominated individuals. Calculate the grid side length of real-time capacity margin, customs clearance speed, and reassignment cost. Perform multi-dimensional grid adaptive partitioning based on dynamic reference grid. S54. Count the number of individuals contained in each multidimensional grid, multiply the side length of the real-time capacity reserve grid by the side length of the clearance speed grid and multiply it by the side length of the relocation cost grid to obtain the multidimensional grid volume, divide the number of individuals contained in each multidimensional grid by the multidimensional grid volume to obtain the local congestion coefficient of each multidimensional grid, and mark the multidimensional grid with the local congestion coefficient greater than the preset congestion threshold as a high-congestion grid. S55. When the number of individuals in the external archive exceeds the preset archive capacity, find the individuals located inside the high-congestion grid, calculate the difference between the real-time capacity reserve and the real-time capacity reserve of other individuals inside the same high-congestion grid and square it, calculate the difference between the clearance speed and the clearance speed of other individuals inside the same high-congestion grid and square it, calculate the difference between the reassignment cost and the reassignment cost of other individuals inside the same high-congestion grid and square it. S56. After adding the three square values, take the square root to obtain the Euclidean distance between the corresponding individual and other individuals in the same high-crowding grid. Select the two individuals with the smallest Euclidean distance, and delete the individual with the smaller Euclidean distance from the external archive first. At the same time, find the individual with the largest real-time transport capacity, the largest clearance speed, and the smallest reassignment cost in each multi-dimensional grid and mark it as the grid boundary extreme point. Force the grid boundary extreme point to be retained in the external archive. Repeat the above deletion and retention operations until the number of individuals in the external archive equals the preset archive capacity. S57. Randomly select two individuals from the external archive as parent individual one and parent individual two. Perform single-point crossover operations on the transit hub, alternative customs clearance port, and backup route of parent individual one and parent individual two to generate offspring individual one and offspring individual two. Randomly replace the transit hub, alternative customs clearance port, and backup route of offspring individual one and offspring individual two with other alternative decision variables with a preset mutation probability to complete the mutation operation. Merge the generated offspring individuals with the individuals in the external archive. Repeat the non-dominated sorting and the topology truncation mechanism based on the dynamic reference grid to improve the external archive truncation operation. After multiple generations of crossover and mutation iterations, output the current non-dominated solution set in the external archive as the Pareto optimal solution set. S58. Read the real-time capacity reserve, customs clearance speed and relocation cost of each individual in the Pareto optimal solution set and normalize them respectively. Add the normalized capacity reserve to the normalized customs clearance speed and subtract the normalized relocation cost to obtain the comprehensive evaluation value of each individual. Select the candidate decision variable combination corresponding to the individual with the highest comprehensive evaluation value to obtain the optimal cross-border logistics adaptive route reconstruction scheme.
[0013] Optionally, S53 specifically includes: S531. Arrange and combine different alternative decision variables to generate multiple initial solutions. Treat each initial solution as an individual and combine all individuals to form an initial population. Substitute the real-time transportation capacity margin, customs clearance speed and reassignment cost of each individual in the initial population into the multi-objective optimization function for calculation to obtain the coordinate point of each individual in the target space. S532. Select individuals from the initial population one by one as the current individual. Compare the coordinates of the current individual in the target space with the coordinates of all other individuals in the target space. If any other individual is better than the current individual in all three dimensions of real-time transport capacity, clearance speed and reassignment cost, then mark the current individual as a dominated individual. If no other individual is better than the current individual in all three dimensions, then mark the current individual as a non-dominated individual. Summarize all non-dominated individuals to form the current non-dominated solution set and store the current non-dominated solution set in the external archive. S533. Read the current non-dominated solution set from the external archive, extract the coordinates of all individuals in the target space, find the maximum and minimum values of all coordinates in the real-time capacity reserve dimension, subtract the minimum value from the maximum value in the real-time capacity reserve dimension to obtain the real-time capacity reserve range, find the maximum and minimum values of all coordinates in the clearance speed dimension, subtract the minimum value from the maximum value in the clearance speed dimension to obtain the clearance speed range, find the maximum and minimum values of all coordinates in the reassignment cost dimension, subtract the minimum value from the maximum value in the reassignment cost dimension to obtain the reassignment cost range; S534. Divide the real-time capacity surplus range by the preset number of grid divisions to obtain the grid side length of the real-time capacity surplus, divide the clearance speed range by the preset number of grid divisions to obtain the grid side length of the clearance speed, and divide the relocation cost range by the preset number of grid divisions to obtain the grid side length of the relocation cost. S535. Iterate through each individual in the current non-dominated solution set. For each individual, subtract the minimum value in the real-time capacity margin dimension from the corresponding real-time capacity margin and divide it by the side length of the real-time capacity margin grid. Round the quotient down to obtain the real-time capacity margin index number. Subtract the minimum value in the clearance speed dimension from the corresponding clearance speed and divide it by the side length of the clearance speed grid. Round the quotient down to obtain the clearance speed index number. Subtract the minimum value in the reassignment cost dimension from the corresponding reassignment cost and divide it by the side length of the reassignment cost grid. Round the quotient down to obtain the reassignment cost index number. S536. The real-time capacity reserve index number, customs clearance speed index number and relocation cost index number are concatenated in sequence as a string. The value corresponding to the string is used as the grid number of the corresponding individual. Individuals with the same grid number are divided into the same multidimensional grid to complete the multidimensional grid adaptive division based on the dynamic reference grid. S537. Count the number of individuals contained in each multidimensional grid, multiply the side length of the real-time capacity reserve grid by the side length of the clearance speed grid and multiply it by the side length of the relocation cost grid to obtain the multidimensional grid volume, divide the number of individuals contained in each multidimensional grid by the multidimensional grid volume to obtain the local congestion coefficient of each multidimensional grid, and mark the multidimensional grid with the local congestion coefficient greater than the preset congestion threshold as a high-congestion grid.
[0014] Optionally, S6 specifically includes: S61. Read the optimal cross-border logistics adaptive route reconstruction scheme, extract the transit hubs, alternative customs clearance ports and backup routes contained in the optimal cross-border logistics adaptive route reconstruction scheme as reassignment target nodes, bind the reassignment target nodes with the original logistics waybill number item by item, and generate reassignment control instructions by splicing them together according to the preset instruction format. S62. The reassignment control instruction is sent to the front-end scheduling interface of the cross-border logistics management platform and displayed in a pop-up window. At the same time, the reassignment control instruction is sent to the vehicle terminal of the transportation vehicle corresponding to the original logistics waybill number. The vehicle terminal receives and parses the reassignment control instruction, extracts the reassignment target node, updates the navigation destination of the transportation vehicle according to the reassignment target node, and completes the logistics reassignment operation.
[0015] The beneficial effects of this invention are: This invention addresses the challenges of inconsistent data structures, high privacy risks from plaintext sharing, and large spatiotemporal benchmark deviations in cross-border logistics by constructing data cleaning rules and feature extraction mechanisms at the local level of logistics nodes. It employs local feature extraction and normalized mapping for data preprocessing, combined with timestamp alignment and coordinate system polynomial transformation to generate a standardized logistics status dataset. Through the standardized dataset's structural unification and physical mapping, a virtual mapping topology reflecting the spatiotemporal dependencies of the global network is constructed. This virtual mapping topology is then input into a Monte Carlo random simulation process, where a normal distribution random number generation function is used to sample vertex dwell time and edge transportation time, calculating the probability distribution interval of dynamically predicted arrival times. In the route reconstruction phase, combining the default risk assessment parameters with a multi-objective optimization function, an improved SPEA2 algorithm is used to perform initial population non-dominated sorting, high-crowding grid marking, and external archive truncation operations. By introducing a topology truncation mechanism based on a dynamic reference grid, crowded individuals located inside high-crowding grids are preferentially deleted while grid boundary extreme points are forcibly retained. The Pareto optimal solution set is output, and the scheme with the highest comprehensive evaluation value is selected to obtain the optimal cross-border logistics adaptive route reconstruction scheme. Furthermore, by analyzing the optimal cross-border logistics adaptive route reconstruction scheme, a reassignment control instruction is generated and sent to the vehicle terminal of the transport vehicle to update the navigation destination, completing the logistics reassignment operation. Ultimately, a closed-loop intelligent management and control system is achieved, enabling comprehensive perception of the entire cross-border logistics chain state evolution, forward-looking early warning of default risks, and multi-objective scheduling decisions. This effectively improves the fusion accuracy of multi-source heterogeneous data, the quantification capability of delay transmission risks, and the diversity and timeliness of global convergence in multi-objective optimization. Attached Figure Description
[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a cross-border logistics information management method based on a logistics management platform proposed in this invention; Figure 2 This is a flowchart of the extraction and Monte Carlo simulation of capacity congestion and delay transmission state based on virtual mapping topology proposed in this invention. Figure 3 This is a flowchart of the adaptive route reconstruction based on the improved SPEA2 algorithm and dynamic reference grid topology truncation mechanism proposed in this invention. Detailed Implementation
[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0018] refer to Figures 1-3 A method for cross-border logistics information management based on a logistics management platform includes the following steps: S1. Collect multi-source heterogeneous data of the entire cross-border logistics chain, including order status, physical trajectory coordinates, load rate and speed of transportation vehicles, and external disturbance factors of each node. Without centrally sharing the original plaintext data of each logistics provider, extract features from the status data of various logistics node entities to obtain the local status feature data of each logistics node entity. S2. Perform unified alignment and fusion processing of local status feature data and multi-source heterogeneous data in terms of timestamps and spatial coordinates, and map them to the global cross-border logistics spatiotemporal coordinate system to obtain a standardized logistics status dataset. S3. Based on the standardized logistics status dataset, construct a virtual mapping topology that reflects the physical logistics network, analyze the spatiotemporal dependence and physical transmission relationship between each logistics node, calculate and extract the capacity congestion and delay transmission status in the logistics link, and obtain full-link status evolution characteristic data. S4. Based on the full-link state evolution characteristic data, perform multiple random simulations and statistical calculations on the running trajectory of each node to obtain the probability distribution range of the dynamically predicted arrival time and the potential default risk assessment parameters. S5. Based on the probability distribution range of the dynamically predicted arrival time and the default risk assessment parameters, determine whether the package triggers the service level agreement default risk. If there is a default risk, the transit hub, alternative customs clearance port and alternative route are used as alternative decision variables. Combine the real-time capacity reserve, customs clearance speed and re-delivery cost of each alternative node to construct a multi-objective optimization function. The improved SPEA2 algorithm is used for comprehensive optimization calculation. A topology truncation mechanism based on dynamic reference grid is introduced to improve the external archive truncation operation. Based on the distribution of the current non-dominated solution set in the target space, multi-dimensional grids are adaptively divided and the local congestion coefficient of each grid is calculated. Individuals located inside the high-congestion grid are deleted first, while individuals located at the grid boundary extreme points are forcibly retained. After multiple generations of crossover and mutation iteration, the Pareto optimal solution set is searched. The scheme with the highest comprehensive evaluation value is selected from it to obtain the optimal cross-border logistics adaptive route reconstruction scheme. S6. Generate reassignment control instructions based on the optimal cross-border logistics adaptive routing reconstruction scheme, and execute logistics reassignment operations.
[0019] This invention significantly improves the accuracy and timeliness of adaptive route reconstruction in cross-border logistics. By extracting local features without centrally sharing the original plaintext data of various logistics providers and mapping them to a global spatiotemporal coordinate system, it achieves unified modeling of multi-source heterogeneous data, enhancing the perception capability of the entire link state while ensuring data privacy. Based on the construction of virtual mapping topology, it can accurately calculate and extract the state of capacity congestion and delay transmission, truly reflecting the spatiotemporal dependencies of the physical logistics network. When facing potential performance default risks, through multiple random simulations, it dynamically generates the probability distribution range of the expected arrival time and risk assessment parameters, achieving a leap from static prediction to dynamic probabilistic extrapolation. In the route reconstruction stage, an improved SPEA2 algorithm and a topology truncation mechanism based on a dynamic reference grid are introduced. By adaptively dividing the multi-dimensional grid and calculating local congestion, redundant individuals in highly congested grids are preferentially eliminated while boundary extreme points are forcibly retained, effectively avoiding the loss of Pareto optimal solutions and ensuring the diversity and global optimality of optimal reconstruction schemes. This invention demonstrates stronger robustness and practicality in the complex cross-border logistics environment of multi-objective game theory, significantly reduces default risk and reshipment costs, and significantly improves the intelligent scheduling level of logistics networks in response to sudden congestion and dynamic delays.
[0020] In this embodiment, S1 specifically includes: S11. Collect order status, physical trajectory coordinates, vehicle load rate and speed, and external disturbance factors at each node as multi-source heterogeneous data for the entire cross-border logistics chain. S12. Construct data cleaning rules on the local side of the logistics node, extract status data representing the operation of various logistics node entities from multi-source heterogeneous data, identify and remove missing values and abnormal jump values exceeding the preset abnormal threshold in the status data, and complete the preprocessing of the status data. The preset abnormal threshold is ±3 times the standard deviation of the historical normal fluctuation range of the status data. S13. Without centrally sharing the original plaintext data of each logistics provider, read the preprocessed status data on the local side of the logistics node, and calculate the temporal change rate and state extreme value of the status data within a preset time window as feature dimensions. The preset time window is the past 1 hour. S14. Perform normalization mapping on the extracted feature dimensions, and concatenate and combine them according to the entity identifier of the logistics node to output the local state feature data of each logistics node entity.
[0021] In this embodiment, S2 specifically includes: S21. Extract time information from local state feature data and multi-source heterogeneous data, find the absolute timestamp corresponding to the generation of each data record, set a unified standard time base, calculate the difference between the timestamp of each data record and the standard time base, remove data records with a difference greater than the preset alignment interval, and retain data records with a difference less than or equal to the preset alignment interval to complete the unified alignment of timestamps. The preset alignment interval is 5 minutes. S22. Read the physical trajectory coordinates in the multi-source heterogeneous data, substitute the longitude and latitude values of the physical trajectory coordinates into the preset coordinate transformation formula, calculate and convert them into the standard longitude and standard latitude in the global cross-border logistics spatiotemporal coordinate system, and complete the unified alignment of spatial coordinates. The preset coordinate transformation formula is a polynomial transformation formula from WGS84 coordinate system to GCJ02 coordinate system. S23. Bind the local state feature data to the unified aligned timestamp and spatial coordinates of the corresponding data record, identify the local state feature data with the same timestamp and spatial coordinates, and merge the identified local state feature data. S24. Combine the merged local state feature data with the corresponding standard longitude and standard latitude column by column, map them to the global cross-border logistics spatiotemporal coordinate system, and output the standardized logistics state dataset.
[0022] In this embodiment, S3 specifically includes: S31. Read the standardized logistics status dataset, extract the node identifier and standard longitude and standard latitude corresponding to each data record in the standardized logistics status dataset, and use each node identifier as a vertex reflecting the virtual mapping topology of the physical logistics network. S32. Find two data records in the standardized logistics status dataset that have the same package identifier and consecutive timestamps. Connect the two vertices corresponding to these two data records with line segments. Use the line segments as edges of the virtual mapping topology to construct a virtual mapping topology that reflects the physical logistics network. S33. Calculate the difference between the standard longitude and standard latitude between the two vertices corresponding to each edge in the virtual mapping topology, use the difference to calculate the physical spatial distance between the two vertices, extract the difference between the timestamps of the data records corresponding to the two vertices as the actual transmission duration, divide the physical spatial distance by the actual transmission duration to calculate the transmission speed, and use the ratio of the transmission speed to the preset reference speed as the physical transmission relationship, wherein the preset reference speed is 60 kilometers per hour. S34. Read the vehicle load rate corresponding to each vertex in the standardized logistics status dataset, and mark the status data where the vehicle load rate is greater than a preset load threshold as a capacity congestion status, where the preset load threshold is 90%. S35. In the virtual mapping topology, find the vertices with capacity congestion as the source nodes of congestion. Read the vehicle load rate of downstream vertices directly connected to the source nodes of congestion through edges. When the vehicle load rate of a downstream vertex increases by more than a preset increase within a preset observation period, mark the downstream vertex as having capacity congestion. This completes the extraction of capacity congestion and delay propagation status. The preset observation period is 2 hours, and the preset increase is 20%. S36. Combine the capacity congestion status, physical transmission relationship, and corresponding unified aligned timestamps with standard longitude and standard latitude of each vertex in the virtual mapping topology in columns to output the full-link state evolution characteristic data.
[0023] In this embodiment, S4 specifically includes: S41. Extract the virtual mapping topology corresponding to the package identifier in the full-link state evolution feature data, and use the virtual mapping topology corresponding to the package identifier as the running trajectory of each node. S42. Read the physical transmission relationship and capacity congestion status corresponding to each vertex of the virtual mapping topology in the full-link state evolution feature data, find the records in the historical database that are the same as the physical transmission relationship and capacity congestion status, calculate the mean and standard deviation of the dwell time of the same records in the historical database at each vertex, and use the mean and standard deviation of the dwell time as the probability distribution parameters of the dwell time of each vertex. S43. Read the physical transmission relationship corresponding to each edge of the virtual mapping topology in the full-link state evolution feature data, find the record with the same physical transmission relationship in the historical database, calculate the mean and standard deviation of the transportation time of the same record on each edge in the historical database, and use the mean and standard deviation of the transportation time as the probability distribution parameter of the transportation time of each edge. S44. Based on the mean and standard deviation of the probability distribution parameters of dwell time for each vertex, a preset number of random dwell time sampling values are generated using a normal distribution random number generation function. Based on the mean and standard deviation of the probability distribution parameters of transportation time for each edge, a preset number of random transportation time sampling values are generated using a normal distribution random number generation function. In each sampling, the random dwell time sampling values of all vertices on the running trajectory of each node corresponding to the package identifier are sequentially added to the random transportation time sampling values of all edges to calculate the total transportation time of a single simulation. The preset number of simulations is 1000. S45. Sort the total transportation time obtained from all preset number of simulations in ascending order of value, select the total transportation time value of a preset specific percentile as the boundary, and construct the probability distribution interval of the dynamically predicted arrival time. The preset specific percentile is the 5th percentile and the 95th percentile. S46. Read the promised arrival time threshold specified in the cross-border logistics service level agreement, count the number of times the total transportation time derived from all preset number of times exceeds the promised arrival time threshold, divide the number of times by the preset number of times to calculate the ratio, and use it as a parameter for assessing potential default risk.
[0024] In this embodiment, S5 specifically includes: S51. Subtract the preset default threshold from the potential default risk assessment parameters to obtain the difference. If the difference is greater than 0, it is determined that the package has triggered the service level agreement default risk; otherwise, it is determined that the service level agreement default risk has not been triggered. The preset default threshold is 0.3. S52. If it is determined that the package triggers the risk of service level agreement default, then the transit hub, alternative customs clearance port and alternative route are used as alternative decision variables. Read the real-time capacity reserve, customs clearance speed and relocation cost corresponding to each alternative decision variable. Use the real-time capacity reserve as the first objective component, the customs clearance speed as the second objective component, and the relocation cost multiplied by negative 1 to obtain the negative relocation cost as the third objective component. Add the first objective component, the second objective component and the third objective component to obtain the summation result, and use it as the fitness value of the multi-objective optimization function to complete the construction of the multi-objective optimization function. S53. Use alternative decision variables to combine the initial population and substitute them into the multi-objective optimization function for calculation. Select individuals in the initial population one by one and mark them as dominated individuals and non-dominated individuals. Calculate the grid side length of real-time capacity margin, customs clearance speed, and reassignment cost. Perform multi-dimensional grid adaptive partitioning based on dynamic reference grid. S54. Count the number of individuals contained in each multidimensional grid, multiply the side length of the real-time capacity reserve grid by the side length of the clearance speed grid and multiply by the side length of the relocation cost grid to obtain the multidimensional grid volume, divide the number of individuals contained in each multidimensional grid by the multidimensional grid volume to obtain the local congestion coefficient of each multidimensional grid, and mark the multidimensional grid with the local congestion coefficient greater than the preset congestion threshold as a high congestion grid, wherein the preset congestion threshold is 5; S55. When the number of individuals in the external archive exceeds the preset archive capacity, find the individuals located inside the high-congestion grid, calculate the difference between the real-time capacity reserve and the real-time capacity reserve of other individuals inside the same high-congestion grid and square it, calculate the difference between the clearance speed and the clearance speed of other individuals inside the same high-congestion grid and square it, calculate the difference between the reassignment cost and the reassignment cost of other individuals inside the same high-congestion grid and square it. The preset archive capacity is 100. S56. After adding the three square values, take the square root to obtain the Euclidean distance between the corresponding individual and other individuals in the same high-crowding grid. Select the two individuals with the smallest Euclidean distance, and delete the individual with the smaller Euclidean distance from the external archive first. At the same time, find the individual with the largest real-time transport capacity, the largest clearance speed, and the smallest reassignment cost in each multi-dimensional grid and mark it as the grid boundary extreme point. Force the grid boundary extreme point to be retained in the external archive. Repeat the above deletion and retention operations until the number of individuals in the external archive equals the preset archive capacity. S57. Randomly select two individuals from the external archive as parent individual one and parent individual two. Perform a single-point crossover operation on the transit hub, alternative customs clearance port, and backup route of parent individual one and parent individual two to generate offspring individual one and offspring individual two. Randomly replace the transit hub, alternative customs clearance port, and backup route of offspring individual one and offspring individual two with other alternative decision variables with a preset mutation probability to complete the mutation operation. Merge the generated offspring individuals with the individuals in the external archive. Repeat the non-dominated sorting and the improved external archive truncation operation based on the topology truncation mechanism of dynamic reference grid. After multiple generations of crossover and mutation iteration, output the current non-dominated solution set in the external archive as the Pareto optimal solution set. The preset mutation probability is 0.1. S58. Read the real-time capacity reserve, customs clearance speed and relocation cost of each individual in the Pareto optimal solution set and normalize them respectively. Add the normalized capacity reserve to the normalized customs clearance speed and subtract the normalized relocation cost to obtain the comprehensive evaluation value of each individual. Select the candidate decision variable combination corresponding to the individual with the highest comprehensive evaluation value to obtain the optimal cross-border logistics adaptive route reconstruction scheme.
[0025] In this embodiment, S53 specifically includes: S531. Arrange and combine different alternative decision variables to generate multiple initial solutions. Treat each initial solution as an individual and combine all individuals to form an initial population. Substitute the real-time transportation capacity margin, customs clearance speed and reassignment cost of each individual in the initial population into the multi-objective optimization function for calculation to obtain the coordinate point of each individual in the target space. S532. Select individuals from the initial population one by one as the current individual. Compare the coordinates of the current individual in the target space with the coordinates of all other individuals in the target space. If any other individual is better than the current individual in all three dimensions of real-time transport capacity, clearance speed and reassignment cost, then mark the current individual as a dominated individual. If no other individual is better than the current individual in all three dimensions, then mark the current individual as a non-dominated individual. Summarize all non-dominated individuals to form the current non-dominated solution set and store the current non-dominated solution set in the external archive. S533. Read the current non-dominated solution set from the external archive, extract the coordinates of all individuals in the target space, find the maximum and minimum values of all coordinates in the real-time capacity reserve dimension, subtract the minimum value from the maximum value in the real-time capacity reserve dimension to obtain the real-time capacity reserve range, find the maximum and minimum values of all coordinates in the clearance speed dimension, subtract the minimum value from the maximum value in the clearance speed dimension to obtain the clearance speed range, find the maximum and minimum values of all coordinates in the reassignment cost dimension, subtract the minimum value from the maximum value in the reassignment cost dimension to obtain the reassignment cost range; S534. Divide the real-time capacity surplus range by the preset grid division number to obtain the real-time capacity surplus grid side length, divide the clearance speed range by the preset grid division number to obtain the clearance speed grid side length, and divide the relocation cost range by the preset grid division number to obtain the relocation cost grid side length. The preset grid division number is 10. S535. Iterate through each individual in the current non-dominated solution set. For each individual, subtract the minimum value in the real-time capacity margin dimension from the corresponding real-time capacity margin and divide it by the side length of the real-time capacity margin grid. Round the quotient down to obtain the real-time capacity margin index number. Subtract the minimum value in the clearance speed dimension from the corresponding clearance speed and divide it by the side length of the clearance speed grid. Round the quotient down to obtain the clearance speed index number. Subtract the minimum value in the reassignment cost dimension from the corresponding reassignment cost and divide it by the side length of the reassignment cost grid. Round the quotient down to obtain the reassignment cost index number. S536. The real-time capacity reserve index number, customs clearance speed index number and relocation cost index number are concatenated in sequence as a string. The value corresponding to the string is used as the grid number of the corresponding individual. Individuals with the same grid number are divided into the same multidimensional grid to complete the multidimensional grid adaptive division based on the dynamic reference grid. S537. Count the number of individuals contained in each multidimensional grid, multiply the side length of the real-time capacity reserve grid by the side length of the clearance speed grid and multiply it by the side length of the relocation cost grid to obtain the multidimensional grid volume, divide the number of individuals contained in each multidimensional grid by the multidimensional grid volume to obtain the local congestion coefficient of each multidimensional grid, and mark the multidimensional grid with the local congestion coefficient greater than the preset congestion threshold as a high-congestion grid.
[0026] This invention achieves dynamic optimization and precise Pareto front maintenance for multi-objective route reconstruction in cross-border logistics by introducing an improved SPEA2 algorithm combined with a dynamic reference grid topology truncation mechanism. Transit hubs, customs clearance ports, and backup routes are used as decision variables. A multi-objective fitness function is constructed using capacity reserves, customs clearance speed, and negative reassignment costs. The initial population is non-dominated and the target space range is calculated. Individuals are adaptively partitioned into a multi-dimensional grid using a floor index. During external archiving truncation, a local congestion coefficient is calculated based on the grid volume. Redundant individuals with the smallest Euclidean distance within highly congested grids are preferentially removed, while boundary extreme points are forcibly retained. This invention maintains the diversity and boundary integrity of the solution set distribution under complex scenarios involving intense multi-objective game dynamics of capacity fluctuations, customs clearance delays, and cost constraints. The routing schemes selected through normalized comprehensive scoring significantly suppress premature convergence of inferior solutions and the loss of high-quality boundary solutions.
[0027] The improved SPEA2 algorithm of this invention is similar to the original SPEA2 algorithm in that both retain the core architecture of the multi-objective evolutionary algorithm, namely, retaining non-dominated individuals by constructing an external archive, generating offspring and merging them by crossover mutation, performing non-dominated sorting based on dominance relationship, and using crowding information to maintain the diversity and boundary integrity of population distribution, ensuring the uniformity of the Pareto front.
[0028] The difference lies in that this invention breaks away from the limitations of the original SPEA2 algorithm, which uniformly calculates crowding for all individuals using a fixed distance and randomly truncates redundant individuals. Instead, it introduces a topological truncation mechanism based on a dynamic reference grid. Building upon the original SPEA2 algorithm's direct calculation of Euclidean distance in the target space to assess individual similarity, this invention incorporates an adaptive partitioning step using a dynamic reference grid. The system first extracts the range of the non-dominated solution set in each dimension, divides it by a preset partition number to generate the multidimensional grid side length, and then calculates the index number by rounding down, precisely partitioning individuals into the corresponding multidimensional grids. When the external archive exceeds capacity, this invention does not blindly compare distances globally, but instead calculates the local crowding coefficient by statistically analyzing the ratio of the number of individuals within the grid to the grid volume, accurately locating highly crowded grids. In the truncation operation, Euclidean distances are preferentially calculated for individuals within highly crowded grids, and redundant points are deleted, while grid boundary extreme points are forcibly retained.
[0029] Based on the above improvements, the beneficial effects of this invention are that, through dynamic grid adaptive partitioning and local congestion assessment, the improved SPEA2 algorithm can accurately truncate high-density areas, avoiding the high complexity and risk of accidental deletion brought about by the global calculation of the original SPEA2; the forced retention of boundary extreme points effectively overcomes the defect of traditional truncation operations that are prone to losing Pareto front endpoint solutions, significantly improving the diversity of solution set distribution and global convergence; this design enhances the optimization accuracy of the algorithm in complex scenarios of intense multi-objective game of capacity, customs clearance and cost, providing a better decision scheme for route reconstruction.
[0030] In this embodiment, S6 specifically includes: S61. Read the optimal cross-border logistics adaptive route reconstruction scheme, extract the transit hub, alternative customs clearance port and backup route included in the optimal cross-border logistics adaptive route reconstruction scheme as the re-delivery target node, bind the re-delivery target node with the original logistics waybill number item by item, and generate a re-delivery control instruction containing the re-delivery target node and the original logistics waybill number according to the preset instruction format. The preset instruction format is JSON format. S62. The reassignment control instruction is sent to the front-end scheduling interface of the cross-border logistics management platform and displayed in a pop-up window. At the same time, the reassignment control instruction is sent to the vehicle terminal of the transportation vehicle corresponding to the original logistics waybill number. The vehicle terminal receives and parses the reassignment control instruction, extracts the reassignment target node, updates the navigation destination of the transportation vehicle according to the reassignment target node, and completes the logistics reassignment operation.
[0031] Example 1: To verify the feasibility of this invention in the field of multi-objective scheduling and dynamic risk management in cross-border logistics, it was deployed in the intelligent scheduling and management platform of a large-scale cross-border integrated logistics enterprise in the Guangdong-Hong Kong-Macao Greater Bay Area. This enterprise covers 12 main cross-border transportation routes from Shenzhen and Guangzhou to Hong Kong, with a total of 35 core transit hubs, 18 alternative customs clearance ports, and over 2,500 various cross-border transportation vehicles, handling an average of over 42,000 cross-border parcels and containers daily. Cross-border logistics scenarios involve cross-border customs, multimodal transport, and complex customs policies, making their transportation networks extremely vulnerable and often facing sudden capacity bottlenecks and delivery time default risks. Typical emergencies include: temporary port congestion leading to customs clearance stagnation, mainline vehicle breakdowns causing chain delays, and severe weather causing drastic fluctuations in transportation time. The platform receives over 2.5TB of multi-source heterogeneous data daily, including GPS tracks, vehicle CAN bus data, port passage monitoring, customs clearance checkpoint patency, and meteorological data. Traditional logistics scheduling systems mainly rely on human experience and static shortest path rules, resulting in slow response, poor optimization capabilities, and an inability to dynamically adapt to the spatiotemporal changes of cross-border networks. Especially during peak holiday periods or in scenarios with multiple shifts, uneven distribution of transport capacity, large-scale delays, and high default claims are common problems.
[0032] In practical deployment, the method of this invention cleans and extracts features from the aforementioned multi-source heterogeneous data at the node local side. Addressing the issues of inconsistent timestamps and coordinate systems, it performs time synchronization and coordinate system polynomial transformation to generate a standardized logistics status dataset. Based on this, the system constructs a virtual mapping topology reflecting the spatiotemporal dependence of the physical logistics network. When the load factor of a hub node exceeds 90%, the system marks it as a capacity congestion state and propagates the delay status downstream along the topology edges, thereby accurately extracting the full-link state evolution characteristics. For the extracted congestion and delay features, the system does not use deterministic time calculations but instead inputs them into a Monte Carlo random simulation process. It fits the normal distribution parameters of the dwell time and edge transportation time of each node using historical data, performs one thousand random samplings and accumulations, and obtains the probability distribution range of the dynamically predicted arrival time and potential default risk assessment parameters. When the default risk assessment parameter exceeds a preset threshold of 0.3, the system triggers an adaptive route reconstruction mechanism. At this point, transit hubs, alternative customs clearance ports, and backup routes are listed as alternative decision variables. The system constructs a multi-objective optimization function to maximize real-time capacity reserve, maximize customs clearance speed, and minimize reassignment cost. During the solution process, the system introduces an improved SPEA2 algorithm for Pareto optimization. The initial population is stored in an external archive after non-dominated sorting. The system calculates the range of the three dimensions—real-time capacity reserve, customs clearance speed, and reassignment cost—and divides it by a preset number of partitions for multi-dimensional adaptive grid partitioning. Each individual is assigned to a corresponding grid, and the system counts the number of individuals within each grid and calculates the local congestion coefficient. When the number of external archives exceeds the preset capacity of 100, a topological truncation mechanism based on a dynamic reference grid is introduced. This mechanism prioritizes deleting individuals with the smallest Euclidean distance within highly congested grids while forcibly retaining grid boundary extreme points. This mechanism effectively avoids the problem of losing boundary solutions during truncation operations in traditional methods, ensuring the diversity and boundary integrity of the optimal solution set. After multiple generations of crossover and mutation iterations, a Pareto optimal solution set is output. Finally, the scheme with the highest comprehensive evaluation value is selected to generate control commands, which are then sent to the vehicle terminal for reassignment. Table 1 below shows the comparison data between the method of this invention and the traditional rule-based scheduling system in cross-border logistics multi-objective route reconstruction tasks during a three-month trial period: Table 1. Performance Comparison of the Invention and Traditional Methods in Adaptive Routing Reconstruction of Cross-border Logistics
[0033] Based on the comparative data shown in Table 1, it can be seen that the adaptive route reconstruction method based on the improved SPEA2 algorithm and dynamic reference grid topology truncation mechanism proposed in this invention has significant performance advantages over the traditional rule scheduling method in terms of dynamic risk management of cross-border logistics. In particular, it has achieved comprehensive improvement in key indicators such as optimization success rate, computation time, retention of solution set diversity, and improvement of overall fulfillment rate.
[0034] In terms of optimization success rate and number of successful relocations, this invention maintains a high level of over 92% in four typical sudden risk scenarios, far exceeding the average of approximately 62% for traditional systems. For example, in the "sudden congestion at the port" scenario, traditional systems rely on manually pre-set backup route pools for blind target rotation, failing to consider the real-time carrying capacity of downstream nodes. This results in numerous relocation schemes experiencing congestion again during actual execution, achieving only a 57.38% optimization success rate. In contrast, this invention, by integrating the extraction of capacity congestion status from virtual mapping topology with Monte Carlo simulation, can accurately predict the timeliness of relocation paths. Combined with the improved SPEA2 algorithm for optimization in the global target space, it achieves a 92.38% optimization success rate, effectively improving the ability to avoid sudden congestion.
[0035] In terms of computational efficiency, this invention significantly reduces the computational complexity of non-dominated sorting and congestion calculations in high-dimensional target spaces through adaptive grid partitioning and topology truncation mechanisms, thereby significantly shortening the response time for route reconstruction. In four risk scenarios, the average computation time is reduced from over 40 seconds using traditional methods to approximately 12 seconds, a nearly three-fold speedup. This ensures that the system can generate the optimal reassignment plan in a very short time when facing concurrent risks such as holiday warehouse overload. For example, the traditional computation for "holiday warehouse overload warning" takes an average of 40.88 seconds, while this invention only takes 13.12 seconds, greatly improving the real-time handling capability of logistics networks in dealing with large-scale node congestion.
[0036] Regarding Pareto solution diversity and boundary solution retention rate, this invention maintains a diversity index above 0.89 and a boundary solution retention rate exceeding 92% across all scenarios, while traditional methods typically have a diversity index around 0.6 and a boundary solution retention rate below 60%. The traditional SPEA2 algorithm, when handling high-dimensional multi-objective problems, often loses Pareto front endpoint solutions due to the random deletion of highly crowded individuals during truncation operations. This leads to severe convergence of scheduling schemes, failing to meet the diverse needs of cost-sensitive and time-sensitive customers in cross-border logistics. In contrast, this invention introduces a topology truncation mechanism based on a dynamic reference grid. This mechanism prioritizes eliminating redundant individuals within highly crowded grids and forcibly retains boundary extreme points, resulting in a more uniform and comprehensive distribution of the output optimal solution set. This ensures that the subsequent comprehensive scoring mechanism can select the optimal routing scheme that balances capacity, customs clearance, and cost.
[0037] The invention also demonstrates significant advantages in improving capacity utilization and overall fulfillment rate. The average capacity utilization rate is increased by over 24%, and the overall fulfillment rate by approximately 18%. Compared to the approximately 10% and 6% increases of traditional methods, this significantly reduces resource waste caused by uneven capacity allocation and high compensation costs due to delays, substantially lowering enterprise operating costs and enhancing the overall resilience of the logistics network. Overall, in key cross-border logistics scenarios such as "port congestion," "vehicle breakdowns," "weather delays," and "holiday warehouse overload," this invention achieves efficient, accurate, and globally optimal dynamic route reconstruction by integrating spatiotemporal topology modeling and an improved multi-objective evolutionary algorithm, filling the technical shortcomings of traditional methods in dynamic risk control of complex spatiotemporal networks.
[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for cross-border logistics information management based on a logistics management platform, characterized in that, Includes the following steps: S1. Collect multi-source heterogeneous data from the entire cross-border logistics chain, extract features from the status data of various logistics node entities, and obtain local status feature data. S2. Perform unified alignment and fusion processing of local status feature data and multi-source heterogeneous data in terms of timestamps and spatial coordinates, and map them to the global cross-border logistics spatiotemporal coordinate system to obtain a standardized logistics status dataset. S3. Based on the standardized logistics status dataset, construct a virtual mapping topology, analyze the spatiotemporal dependence and physical transmission relationship between each logistics node, calculate and extract the capacity congestion and delay transmission status in the logistics link, and obtain full-link status evolution characteristic data. S4. Based on the full-link state evolution characteristic data, perform multiple random simulations and statistical calculations on the running trajectory of each node to obtain the probability distribution range of the dynamically predicted arrival time and the potential default risk assessment parameters. S5. Based on the probability distribution range of the dynamically predicted arrival time and the default risk assessment parameters, determine whether the package triggers the service level agreement default risk, construct a multi-objective optimization function, use the improved SPEA2 algorithm for comprehensive optimization calculation, introduce a topology truncation mechanism based on dynamic reference grid, adaptively divide multi-dimensional grid and calculate the local congestion coefficient of each grid, search the Pareto optimal solution set, and obtain the optimal cross-border logistics adaptive route reconstruction scheme. S6. Generate re-dispatch control instructions based on the optimal cross-border logistics adaptive routing reconstruction scheme, and execute logistics re-dispatch operations; S5 specifically includes: S51. Subtract the preset default threshold from the potential default risk assessment parameters to obtain the difference. If the difference is greater than 0, the package is determined to have triggered the service level agreement default risk; otherwise, it is determined not to have triggered the service level agreement default risk. S52. If it is determined that the package triggers the risk of service level agreement default, then the transit hub, alternative customs clearance port and alternative route are used as alternative decision variables. Read the real-time capacity reserve, customs clearance speed and relocation cost corresponding to each alternative decision variable. Use the real-time capacity reserve as the first objective component, the customs clearance speed as the second objective component, and the relocation cost multiplied by negative 1 to obtain the negative relocation cost as the third objective component. Add the first objective component, the second objective component and the third objective component to obtain the summation result, and use it as the fitness value of the multi-objective optimization function to complete the construction of the multi-objective optimization function. S53. Use alternative decision variables to combine the initial population and substitute them into the multi-objective optimization function for calculation. Select individuals in the initial population one by one and mark them as dominated individuals and non-dominated individuals. Calculate the grid side length of real-time capacity margin, customs clearance speed, and reassignment cost. Perform multi-dimensional grid adaptive partitioning based on dynamic reference grid. S54. Count the number of individuals contained in each multidimensional grid, multiply the side length of the real-time capacity reserve grid by the side length of the clearance speed grid and multiply it by the side length of the relocation cost grid to obtain the multidimensional grid volume, divide the number of individuals contained in each multidimensional grid by the multidimensional grid volume to obtain the local congestion coefficient of each multidimensional grid, and mark the multidimensional grid with the local congestion coefficient greater than the preset congestion threshold as a high-congestion grid. S55. When the number of individuals in the external archive exceeds the preset archive capacity, find the individuals located inside the high-congestion grid, calculate the difference between the real-time capacity reserve and the real-time capacity reserve of other individuals inside the same high-congestion grid and square it, calculate the difference between the clearance speed and the clearance speed of other individuals inside the same high-congestion grid and square it, calculate the difference between the reassignment cost and the reassignment cost of other individuals inside the same high-congestion grid and square it. S56. After adding the three square values, take the square root to obtain the Euclidean distance between the corresponding individual and other individuals in the same high-crowding grid. Select the two individuals with the smallest Euclidean distance, and delete the individual with the smaller Euclidean distance from the external archive first. At the same time, find the individual with the largest real-time transport capacity, the largest clearance speed, and the smallest reassignment cost in each multi-dimensional grid and mark it as the grid boundary extreme point. Force the grid boundary extreme point to be retained in the external archive. Repeat the above deletion and retention operations until the number of individuals in the external archive equals the preset archive capacity. S57. Randomly select two individuals from the external archive as parent individual one and parent individual two. Perform single-point crossover operations on the transit hub, alternative customs clearance port, and backup route of parent individual one and parent individual two to generate offspring individual one and offspring individual two. Randomly replace the transit hub, alternative customs clearance port, and backup route of offspring individual one and offspring individual two with other alternative decision variables with a preset mutation probability to complete the mutation operation. Merge the generated offspring individuals with the individuals in the external archive. Repeat the non-dominated sorting and the topology truncation mechanism based on the dynamic reference grid to improve the external archive truncation operation. After multiple generations of crossover and mutation iterations, output the current non-dominated solution set in the external archive as the Pareto optimal solution set. S58. Read the real-time capacity reserve, customs clearance speed and relocation cost of each individual in the Pareto optimal solution set and normalize them respectively. Add the normalized capacity reserve to the normalized customs clearance speed and subtract the normalized relocation cost to obtain the comprehensive evaluation value of each individual. Select the candidate decision variable combination corresponding to the individual with the highest comprehensive evaluation value to obtain the optimal cross-border logistics adaptive route reconstruction scheme.
2. The cross-border logistics information management method based on a logistics management platform according to claim 1, characterized in that, S1 specifically includes: S11. Collect order status, physical trajectory coordinates, vehicle load rate and speed, and external disturbance factors for each node as multi-source heterogeneous data. S12. Build data cleaning rules on the local side of the logistics node, extract status data representing the operation of various logistics node entities from multi-source heterogeneous data, identify and remove missing values and abnormal jump values exceeding the preset abnormal threshold in the status data, and complete the preprocessing of the status data. S13. Without centrally sharing the original plaintext data of each logistics provider, read the preprocessed status data on the local side of the logistics node, and calculate the temporal change rate and state extreme value of the status data within the preset time window as feature dimensions. S14. Perform normalization mapping on the extracted feature dimensions, and then concatenate and combine them according to the entity identifier of the logistics node to output the local state feature data.
3. The cross-border logistics information management method based on a logistics management platform according to claim 1, characterized in that, S2 specifically includes: S21. Extract time information from local state feature data and multi-source heterogeneous data, find the absolute timestamp corresponding to the generation of each data record, set a unified standard time base, calculate the difference between the timestamp of each data record and the standard time base, remove data records with a difference greater than the preset alignment interval, and retain data records with a difference less than or equal to the preset alignment interval to complete the unified alignment of timestamps. S22. Read the physical trajectory coordinates from the multi-source heterogeneous data, substitute the longitude and latitude values of the physical trajectory coordinates into the preset coordinate transformation formula, calculate and convert them into standard longitude and standard latitude in the global cross-border logistics spatiotemporal coordinate system, and complete the unified alignment of spatial coordinates. S23. Bind the local state feature data to the unified aligned timestamp and spatial coordinates of the corresponding data record, identify the local state feature data with the same timestamp and spatial coordinates, and merge the identified local state feature data. S24. Combine the merged local state feature data with the corresponding standard longitude and standard latitude column by column, map them to the global cross-border logistics spatiotemporal coordinate system, and output the standardized logistics state dataset.
4. The cross-border logistics information management method based on a logistics management platform according to claim 1, characterized in that, S3 specifically includes: S31. Read the standardized logistics status dataset, extract the node identifier and standard longitude and standard latitude corresponding to each data record in the standardized logistics status dataset, and use each node identifier as a vertex of the virtual mapping topology. S32. Find two data records in the standardized logistics status dataset that have the same package identifier and consecutive timestamps. Connect the two vertices corresponding to these two data records with line segments, and use the line segments as edges of the virtual mapping topology to complete the virtual mapping topology. S33. Calculate the difference between the standard longitude and standard latitude between the two vertices corresponding to each edge in the virtual mapping topology, use the difference to calculate the physical space distance between the two vertices, extract the difference between the timestamps of the data records corresponding to the two vertices as the actual transmission duration, divide the physical space distance by the actual transmission duration to calculate the transmission speed, and use the ratio of the transmission speed to the preset reference speed as the physical transmission relationship. S34. Read the load factor of the transportation vehicle corresponding to each vertex in the standardized logistics status dataset, and mark the status data where the load factor of the transportation vehicle is greater than the preset load threshold as the capacity congestion status. S35. In the virtual mapping topology, find the vertex with a capacity congestion state as the congestion source node, read the vehicle load rate of the downstream vertex directly connected to the congestion source node through the edge, and when the vehicle load rate of the downstream vertex increases beyond the preset increase range within the preset observation period, mark the downstream vertex as a capacity congestion state, and complete the extraction of capacity congestion and delay transmission state. S36. Combine the capacity congestion status, physical transmission relationship, and corresponding unified aligned timestamps with standard longitude and standard latitude of each vertex in the virtual mapping topology in columns to output the full-link state evolution characteristic data.
5. The cross-border logistics information management method based on a logistics management platform according to claim 1, characterized in that, S4 specifically includes: S41. Extract the virtual mapping topology corresponding to the package identifier in the full-link state evolution feature data, and use the virtual mapping topology corresponding to the package identifier as the running trajectory of each node. S42. Read the physical transmission relationship and capacity congestion status corresponding to each vertex of the virtual mapping topology in the full-link state evolution feature data, find the records in the historical database that are the same as the physical transmission relationship and capacity congestion status, calculate the mean and standard deviation of the dwell time of the same records in the historical database at each vertex, and use the mean and standard deviation of the dwell time as the probability distribution parameters of the dwell time of each vertex. S43. Read the physical transmission relationship corresponding to each edge of the virtual mapping topology in the full-link state evolution feature data, find the record with the same physical transmission relationship in the historical database, calculate the mean and standard deviation of the transportation time of the same record on each edge in the historical database, and use the mean and standard deviation of the transportation time as the probability distribution parameter of the transportation time of each edge. S44. Based on the mean and standard deviation of the probability distribution parameters of dwell time for each vertex, generate a preset number of random dwell time sampling values using a normal distribution random number generation function. Based on the mean and standard deviation of the probability distribution parameters of transportation time for each edge, generate a preset number of random transportation time sampling values using a normal distribution random number generation function. In each sampling, the random dwell time sampling values of all vertices on the running trajectory of each node corresponding to the package identifier and the random transportation time sampling values of all edges are added together sequentially to calculate the total transportation time of a single simulation. S45. Sort the total transportation time obtained from all preset number of simulations in ascending order of value, select a preset specific percentile of the total transportation time value as the boundary, and construct the probability distribution interval of the dynamically predicted arrival time. S46. Read the promised arrival time threshold specified in the cross-border logistics service level agreement, count the number of times the total transportation time derived from all preset number of times exceeds the promised arrival time threshold, divide the number of times by the preset number of times to calculate the ratio, and use it as a parameter for assessing potential default risk.
6. The cross-border logistics information management method based on a logistics management platform according to claim 1, characterized in that, S53 specifically includes: S531. Arrange and combine different alternative decision variables to generate multiple initial solutions. Treat each initial solution as an individual and combine all individuals to form an initial population. Substitute the real-time transportation capacity margin, customs clearance speed and reassignment cost of each individual in the initial population into the multi-objective optimization function for calculation to obtain the coordinate point of each individual in the target space. S532. Select individuals from the initial population one by one as the current individual. Compare the coordinates of the current individual in the target space with the coordinates of all other individuals in the target space. If any other individual is better than the current individual in all three dimensions of real-time transport capacity, clearance speed and reassignment cost, then mark the current individual as a dominated individual. If no other individual is better than the current individual in all three dimensions, then mark the current individual as a non-dominated individual. Summarize all non-dominated individuals to form the current non-dominated solution set and store the current non-dominated solution set in the external archive. S533. Read the current non-dominated solution set from the external archive, extract the coordinates of all individuals in the target space, find the maximum and minimum values of all coordinates in the real-time capacity reserve dimension, subtract the minimum value from the maximum value in the real-time capacity reserve dimension to obtain the real-time capacity reserve range, find the maximum and minimum values of all coordinates in the clearance speed dimension, subtract the minimum value from the maximum value in the clearance speed dimension to obtain the clearance speed range, find the maximum and minimum values of all coordinates in the reassignment cost dimension, subtract the minimum value from the maximum value in the reassignment cost dimension to obtain the reassignment cost range; S534. Divide the real-time capacity surplus range by the preset number of grid divisions to obtain the grid side length of the real-time capacity surplus, divide the clearance speed range by the preset number of grid divisions to obtain the grid side length of the clearance speed, and divide the relocation cost range by the preset number of grid divisions to obtain the grid side length of the relocation cost. S535. Iterate through each individual in the current non-dominated solution set. For each individual, subtract the minimum value in the real-time capacity margin dimension from the corresponding real-time capacity margin and divide it by the side length of the real-time capacity margin grid. Round the quotient down to obtain the real-time capacity margin index number. Subtract the minimum value in the clearance speed dimension from the corresponding clearance speed and divide it by the side length of the clearance speed grid. Round the quotient down to obtain the clearance speed index number. Subtract the minimum value in the reassignment cost dimension from the corresponding reassignment cost and divide it by the side length of the reassignment cost grid. Round the quotient down to obtain the reassignment cost index number. S536. The real-time capacity reserve index number, customs clearance speed index number and relocation cost index number are concatenated in sequence as a string. The value corresponding to the string is used as the grid number of the corresponding individual. Individuals with the same grid number are divided into the same multidimensional grid to complete the multidimensional grid adaptive division based on the dynamic reference grid. S537. Count the number of individuals contained in each multidimensional grid, multiply the side length of the real-time capacity reserve grid by the side length of the clearance speed grid and multiply it by the side length of the relocation cost grid to obtain the multidimensional grid volume, divide the number of individuals contained in each multidimensional grid by the multidimensional grid volume to obtain the local congestion coefficient of each multidimensional grid, and mark the multidimensional grid with the local congestion coefficient greater than the preset congestion threshold as a high-congestion grid.
7. The cross-border logistics information management method based on a logistics management platform according to claim 1, characterized in that, S6 specifically includes: S61. Read the optimal cross-border logistics adaptive route reconstruction scheme, extract the transit hubs, alternative customs clearance ports and backup routes contained in the optimal cross-border logistics adaptive route reconstruction scheme as reassignment target nodes, bind the reassignment target nodes with the original logistics waybill number item by item, and generate reassignment control instructions by splicing them together according to the preset instruction format. S62. The reassignment control instruction is sent to the front-end scheduling interface of the cross-border logistics management platform and displayed in a pop-up window. At the same time, the reassignment control instruction is sent to the vehicle terminal of the transportation vehicle corresponding to the original logistics waybill number. The vehicle terminal receives and parses the reassignment control instruction, extracts the reassignment target node, updates the navigation destination of the transportation vehicle according to the reassignment target node, and completes the logistics reassignment operation.
Citation Information
Patent Citations
Cross-border e-commerce logistics dynamic matching optimization method and system based on big data driving
CN120409833A
Intelligent cross-border logistics scheduling optimization system
CN121189959A
