Method for dynamically dividing freight hub radiation range based on multi-constraint modularity optimization

Abstract

The application discloses a kind of based on multi-constraint modularity optimization freight hub radiation range dynamic division method, comprising: obtaining the freight interaction data of multiple space units in preset time period;Freight interaction data is divided into multiple sub-data sets corresponding to different time periods;For each sub-data set, construct a space-time weighted network, network node represents space unit, and the edge weight between nodes is determined by freight interaction intensity, inter-node industrial correlation degree and transport mode capacity conversion factor;With multi-constraint modularity function as objective function, community division is carried out on each space-time weighted network, and multi-constraint modularity function fuses standard modularity, space proximity reward item and administrative boundary penalty item;Based on the community division result of each period, the attribution degree of hub node to each community is determined, and the radiation range of hub node is identified and synthesized, so that the radiation range difference of hub in different time periods such as day and night can be identified.

Description

Technical Field

[0001] This invention relates to the fields of transportation geography and complex network analysis, specifically to a method for dynamically dividing the radiation range of freight hubs based on multi-constraint modularity optimization. Background Technology

[0002] Delineating the catchment area of ​​freight hubs is a core fundamental issue in regional logistics planning and transportation infrastructure layout. Traditional methods mainly rely on geographical distance decay models or administrative boundary delineation. However, these methods are all based on static location assumptions, and their accuracy in determining the coverage area is fundamentally limited when dealing with modern logistics systems.

[0003] Specifically, modern freight systems exhibit highly complex heterogeneity in both space and time. On the one hand, the effective reach of logistics networks does not simply follow the principle of geographical proximity, but is deeply influenced by the nonlinear effects of supply chain collaboration and multimodal transport, often resulting in a "radiation leap" phenomenon to distant industrial belts. On the other hand, freight demand exhibits drastically different patterns in the time dimension. For example, time-sensitive freight, exemplified by nighttime port collection and distribution, shows a substantial difference in spatial flow compared to daytime.

[0004] The inability to incorporate this dynamic, non-linear spatiotemporal-industry coupling relationship into a unified analytical framework leads to the following problems: To maintain flexibility in identifying freight communities that span administrative regions and natural distances, the result is often spatially dispersed or "enclave" clusters that are difficult to manage administratively. On the other hand, if we rely on static administrative boundaries or distance attenuation assumptions to ensure operability, we will inevitably lose the ability to capture industrial drivers and time-related differences, leading to a significant deviation between the classification results and the actual freight flow. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides a dynamic partitioning method for the radiation range of freight hubs based on multi-constraint modularity optimization, in order to solve the problems in existing technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A dynamic partitioning method for the radiation range of freight hubs based on multi-constraint modularity optimization includes: Acquire freight interaction data from multiple spatial units within a preset time period; Based on a preset time slicing strategy, the freight interaction data is divided into multiple subsets corresponding to different time periods; For each of the aforementioned subsets of data, a spatiotemporal weighted network is constructed, where network nodes represent the spatial units, and the edge weights between nodes are jointly determined by the freight interaction intensity and the transportation mode capacity conversion factor. An improved community detection algorithm is adopted, with a multi-constraint modularity function as the objective function, to divide each of the spatiotemporal weighted networks into communities. The multi-constraint modularity function integrates standard modularity, spatial proximity reward, administrative boundary penalty, and industry association reward. Based on the community division results for each time period, the degree of affiliation of the hub node to each community is determined, and the radiation range of the hub node is identified and synthesized accordingly.

[0007] In one embodiment, the time slicing strategy includes at least dividing a day into daytime and nighttime periods.

[0008] In one embodiment, it further includes: If the same spatial unit is determined to belong to the radiation range of different hub nodes in the community division results at different time periods, then the spatial unit is marked as a time-shared hinterland. The time-shared hinterland is a spatial unit that is significantly radiated by different hub nodes at different time periods.

[0009] In one embodiment, the industry relevance is calculated based on the semantic similarity between the text feature vectors of industry-related entities within each spatial unit.

[0010] In one embodiment, the multi-constraint modularity function is expressed as follows: ; in, For multi-constraint modularity function, For standard modularity, For spatial proximity rewards, Administrative boundary penalty item, For industry-related rewards, , , For weight hyperparameters.

[0011] In one embodiment, the value of the spatial proximity reward is negatively correlated with the geographical distance between nodes belonging to the same community; The administrative boundary penalty item generates a penalty value when any two nodes belonging to the same community belong to different administrative regions; The value of the industry-related reward item is positively correlated with the industry similarity between nodes belonging to the same community.

[0012] In one embodiment, it further includes: In response to changes in transportation infrastructure or adjustments in industrial layout, the node set, edge weights, or industrial relevance of the spatiotemporal weighted network are updated, and community division and radiation range identification are re-executed to dynamically update the radiation range of hub nodes.

[0013] A dynamic partitioning system for the radiation range of freight hubs based on multi-constraint modularity optimization includes: The data acquisition module is used to acquire freight interaction data from multiple spatial units within a preset time period; The network construction module is used to divide the data into multiple subsets according to a preset time slicing strategy, and to construct a spatiotemporally weighted network for each subset, wherein the edge weights are jointly determined by the freight interaction intensity and the transportation mode capacity conversion factor. The community discovery module is used to divide the spatiotemporal weighted networks into communities, with a multi-constraint modularity function that integrates standard modularity, spatial proximity reward, administrative boundary penalty and industry association reward as the objective function. The radiation range synthesis module is used to determine the degree of affiliation of the hub node to each community based on the community division results of each time period, and to identify and synthesize the radiation range of the hub node accordingly.

[0014] A computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the method described above.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention decouples a single static network into multiple time-segmented sub-networks through a time-slicing mechanism. It can identify the differences in the radiation range of hubs at different times of day and night, and capture special logistics patterns driven by timeliness, such as "nighttime collection and distribution," thus overcoming the shortcomings of traditional static division.

[0016] By incorporating industry relevance into the edge weights, the division of the radiation range reflects the collaborative needs of the industrial chain rather than simply geographical proximity, which is more in line with the actual situation of the coupled development of "industry-transportation" in modern logistics. This allows us to discover and characterize logistics patterns such as "the strong cross-distance radiation of Yantian Port to the Dongguan electronic information industry belt".

[0017] Furthermore, by introducing an administrative boundary penalty term into the multi-constraint modularity function, a systematic inhibition is imposed on the merging of communities across administrative boundaries, enabling the algorithm to inherently weigh the feasibility of administrative management during the search for the optimal logistics cluster.

[0018] By using a transport capacity conversion factor, the frequency of different transport modes is converted into comparable and additive equivalent freight intensity. This allows for a quantitative assessment of the expansion effect of high-capacity transport modes such as rail and waterway on the hinterland of hubs, enabling network construction to more accurately reflect the actual effectiveness of the integrated transportation system.

[0019] By responding to changes in infrastructure or industrial layout and triggering rezoning, the method possesses dynamic responsiveness. It can provide quantitative pre-planning and post-assessment tools for scenarios such as the planning of new transportation corridors (e.g., the Shenzhen-Zhongshan Bridge) or industrial parks, supporting adaptive and dynamic adjustments to logistics planning. Attached Figure Description

[0020] Figure 1 A schematic diagram illustrating the workflow of a dynamic partitioning method for the radiation range of a freight hub based on multi-constraint optimization provided by this invention; Figure 1 The flowchart illustrates a dynamic partitioning method for the radiation range of a freight hub based on multi-constraint optimization, as provided in this invention.

[0021] Figure 2 The flowchart illustrates a dynamic partitioning method for the radiation range of a freight hub based on multi-constraint optimization, as provided in this invention.

[0022] Figure 3 This is a schematic diagram of the components of the improved modularity function in one embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram of the community detection iterative process based on an improved community detection algorithm in one embodiment of the present invention.

[0024] Figure 5 This is an example diagram comparing the daytime radiation range of key hubs in the Greater Bay Area in one embodiment of the present invention.

[0025] Figure 6 This is an example diagram comparing the nighttime radiation range of key hubs in the Greater Bay Area, as shown in one embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Delineating the catchment area of ​​freight hubs is a core fundamental issue in regional logistics planning and transportation infrastructure layout. Traditional methods mainly rely on geographical distance attenuation models or administrative boundary delineation, which have the following technical shortcomings: First, the static location assumption. Traditional methods assume that the radiation range of a hub decreases monotonically with geographical distance, failing to capture the "radiation leap" phenomenon caused by nonlinear factors such as industrial agglomeration and multimodal transport. For example, the radiation intensity of a port to a distant electronics and information industry belt may exceed its radiation intensity to a nearby non-industrial area.

[0028] Second, the limitations of single-modal transport. Existing studies only consider single modes of transport such as highways or ports, ignoring the expansion effect of multimodal transport on the hinterland, and cannot scientifically assess the role of low-cost, high-capacity transport modes such as railways and waterways in enhancing the radiation range of hubs.

[0029] Third, it ignores spatiotemporal heterogeneity. Existing methods do not distinguish the characteristics of freight transport at different times of day and night, and cannot identify special radiation patterns driven by time-sensitive demands such as "nighttime collection and distribution". Taking the Guangdong-Hong Kong-Macao Greater Bay Area as an example, the characteristic that the nighttime freight origin-destination ratio of Yantian Port exceeds 60% is difficult to reflect in static models, resulting in a serious deviation between the radiation range and the actual freight flow.

[0030] Fourth, fragmented administrative boundaries. Traditional divisions based on administrative regions fail to reveal the "hidden logistics communities" that cross administrative boundaries. For example, the equipment manufacturing logistics circle of Guangzhou-Foshan-Zhaoqing and the electronic information logistics circle of Shenzhen-Dongguan have spatial overlap, but existing methods cannot quantify, identify, and represent this cross-regional collaborative relationship.

[0031] In recent years, community detection algorithms in complex networks (such as the Louvain algorithm) have been attempted to be used for traffic network analysis, but existing methods still have shortcomings: the standard modularity function only considers the network topology and does not introduce real-world constraints such as spatial proximity, industry association, and time slicing; there is a lack of dynamic partitioning mechanisms for the spatiotemporal heterogeneity of freight origin-destination flows; and the conflict between cross-administrative boundary community partitioning and administrative operability has not been resolved.

[0032] The following section uses the Guangdong-Hong Kong-Macao Greater Bay Area as the research area, Yantian Port and Nansha Port as the core hubs, and district and county-level administrative units as the basic spatial nodes to provide a detailed description of the method of this invention.

[0033] like Figures 1 to 5 As shown, this invention provides a method for dynamically dividing the radiation range of freight hubs based on multi-constraint optimization, such as... Figure 1-2 As shown, the method includes the following steps: S100: Acquire freight interaction data of multiple spatial units within a preset time period; First, data was collected from the study area, including highway ETC transaction data, heavy truck GPS trajectory data, port container gate data, and railway freight train data. The raw data was then cleaned to remove abnormal records (such as those with excessively short (less than 5 minutes) or excessively long (more than 12 hours) travel times), and spatiotemporal alignment and map matching were performed to map vehicle trajectories to road networks and spatial units. Specifically, truck GPS trajectory data was primarily used for spatiotemporal data analysis of freight vehicles and analysis of hub coverage areas, while highway ETC data was mainly used for verification and validation to ensure the accuracy of spatiotemporal data within the transportation area.

[0034] The acquired truck GPS trajectory data was sampled at 30-second intervals, averaging approximately 63 million records per day, while highway ETC transaction data averaged approximately 7 million records per day. Toll station locations were matched to district / county-level administrative units to construct a "toll station-district / county" mapping table. Abnormal records with travel times less than 5 minutes or greater than 12 hours were removed, ensuring that at least 95% of the cleaned data was valid. The district / county-level administrative regions of the 11 cities in the Greater Bay Area were used as basic nodes, with 26 key hub nodes, including Yantian Port, Shekou Port, Nansha Port, Huangpu Old Port, Dongguan Port, Shunde New Port, Xiaolan Port, and Xiangzhou Port, individually labeled.

[0035] The contents of the truck GPS trajectory data are shown in Table 1 below. The information mainly includes the freight vehicle ID, timestamp, vehicle GPS coordinates, and GPS speed data.

[0036] Table 1. Structure and Example of Truck GPS Track Data: The daily ETC transaction data of highways averages about 7 million records. The data table structure is shown in Table 2, which mainly includes vehicle ID, entrance ID, exit ID, entrance station name, exit station name, vehicle type, vehicle weight, entrance time, and exit time.

[0037] Table 2. Structure and Example of Highway Toll Collection Data Table: Subsequently, the study area was divided into several basic spatial units, which could be district or county-level administrative regions or regular grid units, and key freight hub nodes (such as major ports, airports, railway freight stations, and highway logistics parks) were marked at the same time.

[0038] For example, in data processing, toll station locations are matched to district / county-level administrative units to construct a "toll station-district / county" mapping table; the effective data ratio after cleaning is no less than 95%. The district / county-level administrative regions of the 11 cities in the Greater Bay Area are used as basic spatial unit nodes (110 in total), and 26 key hub nodes, including Yantian Port, Shekou Port, Nansha Port, Huangpu Old Port, Dongguan Port, Shunde New Port, Xiaolan Port, and Xiangzhou Port, are separately labeled to provide a standardized and structured data foundation for subsequent network construction.

[0039] In addition, to support the subsequent calculation of industrial relevance, this step also collects POI data of industrial parks within each spatial unit and extracts their industrial type label information.

[0040] S200. Based on the preset time slicing strategy, the freight interaction data is divided into multiple subsets corresponding to different time periods. After acquiring the data, the freight interaction dataset acquired in step S100 within a complete time period is divided into multiple sub-datasets corresponding to different time periods, based on a preset time slicing strategy.

[0041] This avoids ignoring temporal heterogeneity and failing to distinguish freight patterns at different times. After the division, the freight connections that were originally processed uniformly throughout the day are decoupled into the time domain channel, allowing the algorithm to capture temporal characteristics such as "a port only has strong radiation to a certain area at night".

[0042] In this embodiment, the time slicing parameter T=2 is set to divide a day into a daytime period (6:00 to 18:00) and a nighttime period (18:00 to 6:00 the next day), forming daytime and nighttime subsets respectively. This time slicing is based on the fact that ports exhibit significant nighttime cargo handling characteristics. Taking Yantian Port as an example, its nighttime cargo origin-destination ratio exceeds 60%, and there are substantial differences in cargo flow direction and coverage between daytime and nighttime.

[0043] As needed, it can be further subdivided into weekday periods and weekend / holiday periods to capture the different radiation logics of weekday productive freight and holiday consumer freight. This allows for targeted identification and comparison of hub radiation patterns at different times of day and night, generating differentiated radiation range division results.

[0044] When constructing the network, the set of nodes V contains two types of logical entities: ordinary spatial unit nodes and hub nodes. Both are equal in status within the network and participate uniformly in community partitioning. When a hub is physically located within a spatial unit (e.g., Yantian Port is located in Yantian District), the hub node and the spatial unit node are considered two independent nodes, connected by edges established from actual internal freight flows (such as short-haul transportation between the port area and its surrounding district / county warehouses). This allows for a precise characterization of the different interactions between the hub and its adjacent and more distant areas.

[0045] S300. For each subset of data, construct a spatiotemporal weighted network. The edge weights are jointly determined by the freight interaction strength, the industrial correlation between nodes, and the transportation capacity conversion factor. In this embodiment, a weighted undirected network, i.e., a spatiotemporal weighted network, is constructed for each subset of data (e.g., daytime subset and nighttime subset). The nodes in the network include all spatial unit nodes and hub nodes, and the edges between nodes represent freight transport connections. To achieve dynamic partitioning, a time-slice parameter is first set, dividing the day into multiple time periods (including at least daytime and nighttime periods), thereby constructing sub-networks for each time period.

[0046] When calculating edge weights, the frequency of different modes of transportation in multi-source data is comprehensively considered, and a transportation mode conversion factor is introduced to reflect the large-capacity advantage of rail and water transport compared to road transport—for example, the capacity of a railway train can be converted into the equivalent number of trips of multiple road trucks. Furthermore, industry relevance is introduced, calculating the industry similarity between nodes based on industrial park POI data or input-output tables, enabling network weights to reflect the need for industrial collaboration. Specifically, based on the industrial park POI data of each district and county, an industry type TF-IDF vector is constructed, and cosine similarity is calculated to derive the industry relevance. .

[0047] Among them, edge weight comprehensively reflects the strength of freight connections. Edge weight is not determined by the single factor of freight frequency, but by the following three essential elements, reflecting the organic integration of information on "time and space-industry-capacity".

[0048] (a) Freight Interaction Intensity For ETC data, the passage records of vehicles between toll stations are directly counted, and the OD flow is summarized according to the spatial unit to which the origin and destination toll stations belong. For GPS data, the origin and destination districts of the trajectory are determined through map matching, and the frequency of freight traffic between districts is counted. This embodiment combines the statistical results of the above two data sources as the traffic frequency of the k-th mode of transportation between node i and node j within time period t. .

[0049] (ii) Transport capacity conversion factor of transport mode This invention introduces a transport capacity conversion factor for different modes of transport. This is used to convert the single-pass frequency of different modes of transportation into the equivalent pass frequency under the benchmark mode of transportation, so as to reflect the capacity differences of different modes of transportation and avoid the situation of only considering the pass frequency while ignoring the capacity differences of different modes of transportation.

[0050] This embodiment uses road transport as the benchmark and sets the capacity conversion factors for each mode of transport as follows: =1.0, =1.5, =2.0. Its physical meaning is: the carrying capacity of a railway freight train can be converted into 1.5 times the equivalent number of trips of a single road truck, and the carrying capacity of a waterway liner can be converted into 2.0 times the equivalent number of trips of a single road truck. This conversion factor is based on the standard unit carrying capacity of different modes of transport. For example, the rated load capacity of a standard heavy-duty freight truck is 40 tons, and the load capacity of a standard freight train is approximately 60 tons; therefore, the railway conversion factor is 1.5. The load capacity of a standard inland waterway container ship is approximately 80 tons; therefore, the waterway conversion factor is 2.0. By introducing a capacity conversion factor, the comparative advantages of different modes of transportation are scientifically quantified, enabling the constructed network to more accurately reflect the shaping effect of comprehensive transportation costs on the hinterland. For example, the expansion effect of the opening of a railway freight line on the inland hinterland will be more realistically reflected in this network model with an equivalent weight of 1.5 times, solving the problem of traditional methods underestimating the actual hinterland of multimodal transport hubs.

[0051] (III) Industrial linkages between nodes By calculating the industrial linkage between any two spatial units This is integrated into edge weights to avoid relying solely on physical distance and traffic while ignoring the need for industry chain collaboration.

[0052] In this embodiment, based on the POI data of industrial parks within each spatial unit, industry type labels (such as electronic information, equipment manufacturing, chemical industry, biomedicine, etc.) are extracted to construct the TF-IDF text feature vector of the industry type for each spatial unit. The cosine similarity between any two vectors is then calculated, which serves as the industry correlation between nodes. The cosine similarity value ranges from [0,1]. A larger value indicates that the industrial structures of the two spatial units are more similar or that there is an upstream and downstream collaborative relationship. By calculating the industrial similarity between nodes based on industrial park POI data or input-output tables through industrial correlation, the network weights can reflect the needs of industrial collaboration.

[0053] As an optional implementation method, the degree of industrial linkage can also be calculated based on the consumption coefficient or allocation coefficient between industries in the input-output table.

[0054] (iv) Comprehensive calculation of edge weights Based on the above three factors, the formula for calculating edge weight is as follows: (1) in, Indicates time period internal nodes With nodes The edge weights (overall freight intensity) between them. Use time period numbers (e.g., daytime t=1, nighttime t=2). , For any two nodes (spatial unit nodes or hub nodes) in the network. The mode of transport (road, rail, water). For the first This mode of transportation during this period Frequency of passage within the area For the first The conversion factor for each mode of transport (e.g., 1 for road transport, 1.5 for rail transport, and 2.0 for water transport).

[0055] In this case, the time slice can be set to T=2, divided into daytime (6:00-18:00) and nighttime (18:00-6:00 the next day). The frequency of freight traffic between districts and counties within each time period is calculated: for ETC data, vehicle trajectories are directly statistically analyzed; for GPS data, the origin and destination districts and counties are determined through map matching. The transport mode conversion factor is set: 1 for highways, 1.5 for railways, and 2.0 for waterways. The edge weights are calculated according to formula (1).

[0056] The resulting spatiotemporal weighted network prioritizes edge weights that reflect the absolute intensity and comprehensive transport capacity of freight interactions across different time periods. Industrial relevance, on the other hand, is introduced as an independent structural constraint during the subsequent community discovery objective function optimization phase, guiding the partitioning results to align with the spatial organization logic of the industrial chain. This step-by-step integration design enables the network itself to clearly capture the spatiotemporal heterogeneity of freight connections.

[0057] The network connection strength between two spatial units is proportional to their physical freight frequency, the capacity level of the transportation modes involved, and the degree of matching of their industrial structures. This constructs a highly integrated "information-rich" weighted network that combines spatiotemporal dynamics, economic and technological factors, and industrial collaboration information, laying a data foundation for subsequent accurate and interpretable community discovery.

[0058] S400. An improved community detection algorithm is adopted, using a multi-constraint modularity function as the objective function, to perform community partitioning on each spatiotemporal weighted network. For each of the aforementioned spatiotemporally weighted networks (daytime network and nighttime network), an improved community detection algorithm is used to independently perform community partitioning. The unique feature of this algorithm is that its objective function is not the standard modularity, but a multi-constraint modularity function that integrates the standard modularity, spatial proximity reward term, and administrative boundary penalty term.

[0059] It should be noted that standard modularity-driven community discovery only considers the edge density of the network topology and lacks consideration of real spatial constraints. This may result in spatially discrete "enclave" communities or cross-domain "super communities" that cannot be implemented, lacking geographical interpretability and administrative operability.

[0060] In this embodiment, three modification terms are added to the standard modularity function to improve the scientific rigor and operability of community division. The first term is a spatial proximity reward, which encourages geographically close nodes to be grouped into the same community, preventing the algorithm from producing extremely dispersed or "enclave"-like community structures and maintaining spatial continuity in the division results. The second term is an administrative boundary penalty, which penalizes the merging of communities across administrative boundaries (such as across cities or provinces), balancing the division results between naturally formed logistics clusters and administrative feasibility. The third term is an industry association reward, which encourages nodes with high industry similarity to be grouped into the same community, reflecting the driving role of supply chain collaboration in logistics organization. By adjusting the weight parameters of each modification term, the modularity function can simultaneously consider spatial continuity, administrative operability, and industry matching, achieving multi-objective optimization.

[0061] (I) Construction of Multi-Constraint Modularity Function The multi-constraint modularity function The expression form is: (2) in, This is the improved multi-constraint modularity value (the higher the value, the better the quality of community partitioning). Standard modularity (considering only network topology). For spatial proximity rewards, Administrative boundary penalty item, For industry-related rewards, This is the spatial proximity weight hyperparameter, with values ​​ranging from [0.1, 0.5], such as 0.3. This is a hyperparameter for the administrative boundary penalty weight, with a value range of [0, 0.5], such as 0.2. The hyperparameter for the industry-related reward weight has a value range of [0.3, 0.7], such as 0.5. The hyperparameter is determined by grid search or Bayesian optimization, with the goal of maximizing the combined score of modularity and interpretability.

[0062] The definitions and functions of each component are as follows: Standard modularity The salience of network community structure relative to random networks was measured, driving the algorithm to discover clusters of nodes with dense freight connections. The calculation method is as follows: (3) in, It is the sum of the weights of all edges in the network (total freight intensity). For each element in the network adjacency matrix, wij(t) is a value if there is an edge between nodes i and j, and 0 otherwise. Let be the degree of node i (the sum of the weights of all edges connected to node i). The degree of node j, The Kronecker function is defined as follows: δ=1 when node i and node j belong to the same community, and δ=0 when node i and node j belong to different communities. Let i be the community number to which node i belongs. is the community number to which node j belongs.

[0063] Spatial proximity reward Its value is a weighted sum of the reciprocals of the geographical distances between nodes belonging to the same community, and is negatively correlated with the geographical distances between nodes. This embodiment uses the straight-line distance between the locations of district and county governments. Take the reciprocal 1 / As a reward weight: , For nodes and Geographical distance; in, The geographical distance between node i and node j (in this embodiment, the straight-line distance between the locations of district and county governments) is used. The reward value is negatively correlated with the geographical distance: the closer the distance, the higher the reward for being classified into the same community.

[0064] During the optimization process, assigning geographically proximate nodes to the same community yields a positive reward, thereby suppressing the formation of "enclave" communities (i.e., communities that are spatially discontinuous and consist of isolated patches), and ensuring the spatial continuity of the partitioning results. This is an algorithmic implementation of the first law of geography.

[0065] Administrative boundary penalty items When any two nodes belonging to the same community are located in different administrative regions, this item applies a penalty value to the modularity, thereby inhibiting community merging across administrative regions. In this embodiment, a boundary indicator factor is used. If nodes i and j belong to the same city, =0; if they belong to different cities =1; For cross-border scenarios such as Shenzhen and Hong Kong, further settings can be made. =2 to reinforce the punishment: , in, The boundary indicator factor is 1 for cross-city nodes and 0 for same-city nodes. If nodes i and j belong to the same city, bij = 0 (no penalty). If nodes i and j belong to different cities, bij = 1 (basic penalty). If nodes i and j belong to different countries / regions (such as Shenzhen and Hong Kong), bij = 2 (enhanced penalty). A penalty value is only generated when cross-boundary nodes are assigned to the same community (δ = 1). Industry-related incentive items Its value is related to the industry similarity between nodes belonging to the same community. A positive correlation is established, and the industry correlation calculated in step S300 is directly used as the industry similarity. , in, The industry similarity between node i and node j is denoted as . The industry association degree is obtained by constructing a TF-IDF vector based on the POI data of each spatial unit industrial park and calculating the cosine similarity. The value range is [0,1]. The reward value is positively correlated with the industry similarity: the higher the industry matching degree, the higher the reward obtained by being classified into the same community.

[0066] The rewards will include nodes with upstream and downstream collaboration or similar industry clusters into the same logistics community, so that the scope of coverage is accurately anchored to the inherent logic of the spatial organization of the industrial chain, rather than just based on geographical proximity or traffic volume.

[0067] α, β, and δ are weighted hyperparameters that control the influence of spatial proximity, administrative boundary penalties, and industrial linkages on the overall optimization objective, respectively. This embodiment uses a grid search within a three-dimensional parameter space of α∈[0.1,0.5], β∈[0,0.5], and δ∈[0.3,0.7]. The optimal objective is determined by the highest combined score of modularity and interpretability of the partitioning results, resulting in optimal values ​​of α=0.3, β=0.2, and δ=0.5. Optional implementation methods also include automatic hyperparameter tuning methods such as Bayesian optimization.

[0068] Using the multi-constraint modularity function as the objective function, the improved Louvain algorithm is independently executed for each time-segment sub-network in an iterative optimization process, as follows: Figure 4 As shown, it specifically includes the following stages: Phase 1: Initialization.

[0069] Each node in the network is initialized as an independent community. Taking the nighttime network as an example, 136 nodes are initialized (110 district / county nodes + 26 hub nodes), and each node is an independent community.

[0070] Phase Two: Local Movement Phase.

[0071] Iterate through all nodes. For each node, attempt to move it from its current community to the community of its neighboring nodes, and calculate the change in modularity before and after the move based on the multi-constraint modularity function. Only perform a move if the modularity increases after the move. Repeat the scan of all nodes until no node movement increases the modularity. In the first round of local movements in the nighttime network of this embodiment, a node's modularity increases significantly when it moves from an independent community to the community where the port is located, and the move is accepted; repeat the scan until all node movements do not increase the modularity, obtaining the first round of community partitioning results.

[0072] Phase 3: Network Aggregation Phase.

[0073] The communities formed in Phase 2 are shrunk into "super nodes". The edge weights within the communities are aggregated into the self-loop weights of the super nodes, and the edge weights between communities are aggregated into the edge weights between the super nodes, thus constructing a new aggregation network.

[0074] Phase 4: Iterative convergence.

[0075] Repeat the local movement phase of Phase 2 and the network aggregation phase of Phase 3 on the aggregated network until the change in modularity is less than a preset threshold (e.g., 0.001) or the preset maximum number of iterations is reached. After the algorithm converges, output the stable community partitioning results of the subnetwork for that time period.

[0076] Ultimately, each time period sub-network will output a stable set of community division results. Each community represents a "hub-hinterland" functional cluster, reflecting the spatial agglomeration pattern (spatial unit) of the actual logistics organization within that time period.

[0077] By organically integrating the three correction terms of "spatial proximity reward," "administrative boundary penalty," and "industry association reward" with the standard modularity into a single, multi-objective-guided optimization function, this invention cleverly solves the deep-seated technical contradiction of "how to make the results of purely data-driven community discovery directly interpretable and operable in the real world."

[0078] S500. Based on the community division results of each time period, determine the degree of affiliation of the hub node to each community, and identify and synthesize the radiation range of the hub node accordingly. After the community division results stabilize at each time period, the dynamic identification and synthesis of the radiation range are performed.

[0079] (a) Calculation of the hub's affiliation with each community For each hub node h, calculate the degree of affiliation of the hub to each community c at each time period t. The degree of affiliation is defined as the proportion of the sum of edge weights between hub node h and all nodes in community c to the total sum of edge weights between hub node h and all nodes in the network. The calculation formula is as follows: (4) in, Let h be the degree of affiliation of hub h to community c in time period t. These are the key nodes to be analyzed (such as Yantian Port and Nansha Port). Represents any community. The set of all nodes in the network. Let be the edge weight between hub h and any node j within community c in the network during time period t. ( When a preset threshold (e.g., 0.3) is used, the community is determined. As a hub During the period The radiation range.

[0080] (II) Determination of Radiation Range When belonging When the value is greater than or equal to the preset threshold λ, community c is determined to be the radiation range of hub h in that time period t.

[0081] In this embodiment, λ is 0.3.

[0082] The affiliation threshold λ can be set to 0.3 as needed. That is, when the freight connection intensity between a hub and a community accounts for more than 30% of its total connection intensity, the community is considered to be significantly influenced by the hub, and the direction has significant business dominance.

[0083] In sensitivity testing, when λ increases from 0.2 to 0.4, the number of identified communities within the radiation range decreases slightly, but the core hinterland composition remains stable. The core indirect hinterland of the hub does not undergo drastic changes due to small shifts in the threshold. In practical applications, λ can be adjusted by planners based on the hub density and competitive relationships within the study area.

[0084] Using the Guangdong-Hong Kong-Macao Greater Bay Area as the research area, the community affiliation degree of 26 key hub nodes was calculated in both daytime and nighttime networks. For example, Yantian Port's community affiliation degree in the nighttime network, which includes Longgang District, Pingshan District, southeastern towns of Dongguan, and some streets in Huiyang District of Huizhou, reached 0.65, exceeding the threshold of 0.3. This community was identified as a core component of Yantian Port's nighttime radiation range. In contrast, the affiliation degree distribution in the daytime network was relatively balanced, and the radiation range exhibited a wider but sparser characteristic.

[0085] (III) Synthesis of Comprehensive Radiation Range The combined radiation range of the hub node is obtained by taking the union of its radiation ranges at different times (day and night). The combined radiation range reflects the overall hinterland coverage of the hub under different weather conditions. Figure 5 and Figure 6 The illustrations show a comparison of the radiation range of key hub nodes in the Greater Bay Area during the day and night, which can intuitively show the spatial differences in the day and night radiation patterns.

[0086] (iv) Identification and marking of time-sharing hinterland Furthermore, this step implements an important labeling operation: if the same spatial unit belongs to the radiation range of different hub nodes in the community division results at different time periods (such as daytime and nighttime), then the spatial unit is marked as "time period shared hinterland", and its time period switching mode is output, that is, recording which hub radiation the spatial unit belongs to during the daytime and which hub radiation it belongs to at nighttime.

[0087] For example, in this embodiment, the Dongguan Songshan Lake area is mainly influenced by Dongguan Port and surrounding highway hubs during the day (reflecting its high-frequency short-distance delivery demand for electronic components during the day), while at night it is mainly influenced by Yantian Port (reflecting its demand for finished products to be concentrated at the port for export at night). This area is marked as the "time-sharing hinterland" of Yantian Port and Dongguan Port, and the time-switching mode is recorded as "belonging to Dongguan Port during the day and belonging to Yantian Port at night".

[0088] This accurately reveals and characterizes the dynamic flexibility of the hinterland and the competitive and cooperative relationship between hubs in the time dimension, providing a new and quantitative basis for decision-making on multimodal transport timetable coordination, flexible warehousing layout, and time-based differentiated logistics services, and solving the problem that existing technologies cannot characterize the situation of flexible hinterland across time periods.

[0089] One implementation also includes dynamically updating the radiation range in response to external changes. When significant changes occur within the study area, the following update process is triggered.

[0090] (a) Triggering conditions Triggering conditions include, but are not limited to: events involving changes to transportation infrastructure (such as newly opened bridges, tunnels, railway lines, or port berths), or events involving adjustments to industrial layout (such as the construction of new large-scale industrial parks, industrial relocation, or the commissioning or relocation of major manufacturing bases).

[0091] (ii) Update operation When the above-mentioned triggering event is detected, perform the following operations: First, update the node set (e.g., add a new transportation hub node), edge weight (e.g., change the freight interaction intensity between nodes due to the opening of a new channel), or industry correlation (e.g., change the industry feature vector of a node due to the construction of a new industrial park) in the spatiotemporal weighted network accordingly; then, re-execute steps S200 to S500 and output the updated hub radiation range division results.

[0092] (III) Quantitative Assessment of Changes Furthermore, the rate of change of the radiation range can be calculated to quantify the impact of the external events on the hinterland of each hub.

[0093] The formula for calculating the rate of change of scope size is: , used to measure the physical spatial expansion or contraction of the radiation range, where, The comprehensive radiation range (set of spatial units) of the hub after the event. The comprehensive radiation range (set of spatial units) of the hub before the event occurred. Let be the cardinality of the difference between the two sets, which is the number of newly added spatial units.

[0094] Changes in industry matching degree: This is used to measure the change in the tightness of industrial linkages within the updated radiation range. A positive value indicates an improved match between the radiation range and the spatial organization of the industrial chain. The new radiation range after an external event (such as the opening of the Shenzhen-Zhongshan Bridge) occurs. The total of industry-related rewards within the region, The original radiation range before the event occurred. The sum of industry-related rewards within the region.

[0095] Shared hinterland change rate: , where S is the set of spatial units labeled as “time-sharing hinterland”. This refers to a set of spatial units that are marked as "time-sharing hinterland" before an external event occurs (e.g., before a new transportation corridor is opened). This refers to the set of spatial units that are marked as "time-sharing hinterland" after an external event occurs (i.e., after the network is re-divided according to the updated network). Represents a set Cardinality refers to the total number of spatial units identified as "time-sharing hinterlands" before an event occurs. Represents a set and The potential of the symmetric difference, specifically, is the sum of the number of newly added shared hinterlands and the number of shared hinterlands that have disappeared. This value reflects the absolute scale of changes in the shared hinterland pattern; this shared hinterland change rate index is used to measure changes in the competitive and cooperative relationship between hubs over time, specifically, quantifying the degree of impact of external events (such as the opening of a new passage) on the competitive and cooperative relationship between hubs over time. For example, if the opening of a new bridge makes it unnecessary for certain areas to belong to different hubs during the day and night (shared hinterland disappears), or if new time-sharing dependencies emerge, This will capture and measure the rate of structural change in this spatiotemporally elastic hinterland.

[0096] The above multidimensional indicators can be used to comprehensively assess the redistribution effect of events such as the opening of the Shenzhen-Zhongshan Bridge on the radiation range of freight hubs in terms of physical space, industrial space and temporal flexibility, providing scientific and precise quantitative support for planning decisions.

[0097] Parameter sensitivity analysis and robustness verification: To verify the robustness of this technical solution and the rationality of the hyperparameter selection, a systematic parameter sensitivity test was conducted.

[0098] When the spatial proximity weight α is changed from 0.1 to 0.5: when α=0.1, the geographical continuity of the community is weak, and there are a few enclaves; when α increases to 0.3, the geographical continuity of the community is significantly enhanced, and the community discovery effect driven by industry linkages is not excessively weakened; when α=0.5, the modularity decreases by about 0.05, and the over-reliance on geographical proximity leads to the incorrect segmentation of some cross-distance industry-linked communities. α=0.3 is the preferred value.

[0099] When the administrative boundary penalty weight β is changed from 0 to 0.5: When β=0, there are many cross-city communities, such as spatial units in Guangzhou and Foshan being merged into the same community; when β increases to 0.2, the number and size of cross-city communities are systematically reduced to a reasonable level, and the same community in Guangzhou and Foshan is split into two, with a decrease in modularity of 0.08, but a significant gain in administrative operability; when β=0.5, there are almost no cross-city communities, excessively severing the actual cross-regional logistics connections. β=0.2 is the preferred value.

[0100] When the industry reward weight δ is changed from 0.3 to 0.7: when δ=0.3, the guiding effect of industry linkage on community division is limited; when δ=0.5, the community binding relationship between Yantian Port and Dongguan Electronic Information Industry Belt is significantly strengthened, and the modularity increases by about 0.03; when δ=0.7, isolated industry communities that do not conform to reality are generated, and interpretability decreases. δ=0.5 is the preferred value.

[0101] The above-mentioned preferred hyperparameter combination (α=0.3, β=0.2, δ=0.5) can achieve the comprehensive optimality of modularity, geographical continuity, administrative operability and industry matching, proving the stability of the method in the parameter space.

[0102] A dynamic partitioning system for the radiation range of freight hubs based on multi-constraint modularity optimization includes: The data acquisition module is used to acquire freight interaction data from multiple spatial units within a preset time period; The network construction module is used to divide the data into multiple subsets according to a preset time slicing strategy, and to construct a spatiotemporally weighted network for each subset, wherein the edge weights are jointly determined by the freight interaction intensity and the transportation mode capacity conversion factor. The community discovery module is used to divide the spatiotemporal weighted networks into communities, with a multi-constraint modularity function that integrates standard modularity, spatial proximity reward, administrative boundary penalty and industry association reward as the objective function. The radiation range synthesis module is used to determine the degree of affiliation of the hub node to each community based on the community division results of each time period, and to identify and synthesize the radiation range of the hub node accordingly.

[0103] In practice, the above methods can be deployed on cloud servers or local computing platforms, for example, through the collaborative implementation of the following functional modules: Data acquisition module: Connects to highway toll collection system, GPS data center, port management system and railway dispatching system, collects freight interaction data in real time or in batches, and performs data cleaning and preprocessing; Network construction module: It is configured with time slice parameters, transportation mode conversion factor table and industry correlation calculation engine, and constructs spatiotemporal weighted network for each time period according to preset strategy; Community Detection Module: Loads the improved Louvain algorithm and multi-constraint modularity function, with built-in configuration interfaces for hyperparameters α, β, and δ, and performs community detection in parallel across different time periods of the network; Radiation range synthesis module: Filters radiation ranges based on the degree of affiliation threshold λ, performs day and night analysis and aggregates the results, and identifies and marks the shared hinterland during the marked time periods; Dynamic prediction module: Monitors changes in infrastructure and industrial layout, triggers network updates and re-divisions, and outputs a quantitative report on the rate of change.

[0104] Each module exchanges data through a network interface or bus, and works together to dynamically divide and predict the coverage area of ​​the freight hub.

[0105] A computer-readable storage medium having a computer program stored thereon. When executed by a processor, the program implements all the method steps described above. The storage medium can be any non-volatile carrier capable of storing program code, such as a server hard drive, solid-state drive, optical disc, or USB flash drive.

[0106] The foregoing description of specific exemplary embodiments of the present invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it is obvious that many changes and variations can be made based on the above teachings. Although embodiments of the invention have been shown and described, these specific embodiments are merely explanations of the invention and are not intended to limit it. The specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. The purpose of selecting and describing exemplary embodiments is to explain the specific principles of the invention and its practical application, so that those skilled in the art, after reading this specification, can make modifications, substitutions, variations, and various choices and changes to the embodiments as needed without departing from the principles and spirit of the invention, provided that such modifications, substitutions, variations, and choices and changes are within the scope of the claims of the invention and are protected by patent law.

Claims

1. A method for dynamically dividing the radiation range of freight hubs based on multi-constraint modularity optimization, characterized in that, include: Acquire freight interaction data from multiple spatial units within a preset time period; Based on a preset time slicing strategy, the freight interaction data is divided into multiple subsets corresponding to different time periods; For each of the aforementioned subsets, a spatiotemporal weighted network is constructed, wherein network nodes represent the spatial units, and the edge weights between nodes are jointly determined by the freight interaction intensity and the transportation mode capacity conversion factor. An improved community detection algorithm is adopted, with a multi-constraint modularity function as the objective function, to divide each of the spatiotemporal weighted networks into communities. The multi-constraint modularity function integrates standard modularity, spatial proximity reward, administrative boundary penalty, and industry association reward. Based on the community division results for each time period, the affiliation degree of the hub node to each community is determined, and the radiation range of the hub node is identified and synthesized.

2. The method for dynamically dividing the radiation range of a freight hub based on multi-constraint modularity optimization according to claim 1, characterized in that, The time slicing strategy includes at least dividing a day into daytime and nighttime periods.

3. The method for dynamically dividing the radiation range of a freight hub based on multi-constraint modularity optimization according to claim 1, characterized in that, Also includes: If the same spatial unit is determined to belong to the radiation range of different hub nodes in the community division results at different time periods, then the spatial unit is marked as the time-shared hinterland.

4. The method for dynamically dividing the radiation range of a freight hub based on multi-constraint modularity optimization according to claim 1, characterized in that, The industrial relevance is calculated based on the semantic similarity between the text feature vectors of industry-related entities within each spatial unit.

5. The method for dynamically dividing the radiation range of a freight hub based on multi-constraint modularity optimization according to claim 1, characterized in that, The multi-constraint modularity function is: ； in, For multi-constraint modularity function, For standard modularity, For spatial proximity rewards, Administrative boundary penalty item, For industry-related rewards, , , For weight hyperparameters.

6. The method for dynamically dividing the radiation range of a freight hub based on multi-constraint modularity optimization according to claim 5, characterized in that, The value of the spatial proximity reward is negatively correlated with the geographical distance between nodes belonging to the same community; The administrative boundary penalty item generates a penalty value when any two nodes belonging to the same community belong to different administrative regions; The value of the industry-related reward item is positively correlated with the industry similarity between nodes belonging to the same community.

7. The method for dynamically dividing the radiation range of a freight hub based on multi-constraint modularity optimization according to claim 1, characterized in that, Also includes: In response to changes in transportation infrastructure or adjustments in industrial layout, the node set, edge weights, or industrial relevance of the spatiotemporal weighted network are updated, and community division and radiation range identification are re-executed to dynamically update the radiation range of hub nodes.

8. A dynamic partitioning system for the radiation range of freight hubs based on multi-constraint modularity optimization, characterized in that, include: The data acquisition module is used to acquire freight interaction data from multiple spatial units within a preset time period; The network construction module is used to divide the data into multiple subsets according to a preset time slicing strategy, and to construct a spatiotemporally weighted network for each subset, wherein the edge weights are jointly determined by the freight interaction intensity and the transportation mode capacity conversion factor. The community discovery module is used to divide the spatiotemporal weighted networks into communities, with a multi-constraint modularity function that integrates standard modularity, spatial proximity reward, administrative boundary penalty and industry association reward as the objective function. The radiation range synthesis module is used to determine the degree of affiliation of the hub node to each community based on the community division results of each time period, and to identify and synthesize the radiation range of the hub node accordingly.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method of any one of claims 1 to 7.

Images