High-speed light-load railway freight logistics method and system based on the Internet of Things and big data

Abstract

The present invention provides a high-speed light-load railway freight logistics method and system based on the Internet of Things and big data. The method includes the following steps: obtaining historical data and coordinates of agricultural logistics parks in the target area, and using them as logistics nodes to construct a regional logistics topology map. Calculating the correlation strength between nodes and constructing a hypergraph structure. Obtaining future logistics orders and available train information and constructing path constraints. Generate multiple logistics hyperpaths for each order in the hypergraph to ensure that the sum of the throughput of each node is 0. Calculate the node heat and construct load balancing constraints. Finally, based on the load balancing constraints and the shortest path goals, select the optimal combination from the hyperpaths to generate the optimal logistics strategy for the target area. The present invention comprehensively considers historical data, network structure and real-time order information to achieve efficient and balanced logistics scheduling optimization.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent logistics control, and specifically relates to a high-speed light-load railway freight logistics method and system based on the Internet of Things and big data. Background Art

[0002] In today's globalized economy and digital transformation, the logistics industry faces unprecedented challenges and opportunities. With the booming e-commerce industry, increasing cross-border trade, and rising consumer demand for fast delivery, logistics networks are becoming increasingly complex and expansive. Simultaneously, the rapid development of emerging technologies such as the Internet of Things, big data, and artificial intelligence has brought new possibilities to the logistics industry, but also raised higher demands.

[0003] Traditional logistics management systems were initially designed primarily to solve basic logistics scheduling and resource allocation problems. These systems are typically based on relatively simple algorithms and rules, such as shortest path algorithms or greedy algorithms. They perform well in small-scale, relatively stable logistics networks, effectively scheduling transportation routes, managing inventory, and scheduling human resources. However, these traditional systems exhibit numerous limitations when faced with the complex modern logistics environment. Specifically, traditional systems often employ static or semi-static models, making them difficult to adapt to the highly dynamic nature of modern logistics networks. Furthermore, these systems typically focus on local optimization and lack a global perspective. This may result in achieving the optimal allocation strategy for a specific route segment or node, but the overall network efficiency may be unsatisfactory. Summary of the Invention

[0004] The present invention provides a high-speed light-load railway freight logistics method and system based on the Internet of Things and big data, so as to solve the problem that the existing logistics management system is difficult to improve the logistics scheduling efficiency in a complex logistics system.

[0005] In a first aspect, the present invention provides a high-speed light-load railway freight logistics method based on the Internet of Things and big data, the method comprising the following steps:

[0006] Obtain historical logistics data and logistics park coordinates of all agricultural logistics parks in the target area;

[0007] The agricultural logistics park is used as a logistics node, and node attributes are assigned to the logistics node in combination with the historical logistics data and the logistics park coordinates. A regional logistics topology map of the target area is constructed based on the train track topology of the target area and in combination with all the logistics nodes;

[0008] Calculating the node association strength between the logistics nodes in the regional logistics topology graph based on the node attributes;

[0009] Constructing a hypergraph structure in the regional logistics topology graph by combining the node association strength and the logistics node edge, and converting the regional logistics topology graph into a regional logistics hypergraph;

[0010] Obtaining a list of logistics orders and a list of available trains within a preset time period in the future for the target area, and identifying logistics order information for each logistics order in the logistics order list, wherein the logistics order information includes a logistics starting point, a logistics destination, a logistics throughput, and a logistics freight type;

[0011] Constructing path constraints based on the logistics order information and the available train list, and generating multiple logistics super paths for each logistics order in the regional logistics hypergraph based on the path constraints, wherein the sum of the logistics throughputs of all the logistics nodes in each logistics super path is 0;

[0012] Calculating the node heat of each logistics node according to all the logistics super paths;

[0013] A node load balancing constraint condition is constructed based on the node heat. Based on the node load balancing constraint condition and with the shortest total path as the optimization goal, the optimal super path combination is selected from all the logistics super paths to generate the optimal logistics strategy for the target area.

[0014] Optionally, the agricultural logistics park is used as a logistics node, node attributes are assigned to the logistics node in combination with the historical logistics data and the logistics park coordinates, and a regional logistics topology map of the target area is constructed based on the train track topology of the target area and in combination with all the logistics nodes, including the following steps:

[0015] Using the agricultural logistics park as a logistics node;

[0016] For each of the logistics nodes, assign node coordinate attributes to the logistics node according to the logistics park coordinates;

[0017] Extracting the historical logistics freight types of the logistics node and the historical average throughput of each different historical logistics freight type from the historical logistics data using natural language technology;

[0018] Constructing a historical logistics matrix corresponding to the logistics node in combination with the historical logistics freight types and the historical average throughput;

[0019] Assigning the historical logistics matrix as the node logistics attribute of the logistics node;

[0020] If there are M freight train tracks in the train track topology of the target area between any two of the agricultural logistics parks, M logistics node edges are generated between the corresponding two logistics nodes to obtain a regional logistics topology map of the target area.

[0021] Optionally, the calculating the node association strength between the logistics nodes in the regional logistics topology graph based on the node attributes includes the following steps:

[0022] For any two logistics nodes, the transportation distance and average transportation time between the two logistics nodes are calculated according to the node coordinate attributes;

[0023] Based on the node logistics attributes, the logistics matrix similarity and throughput similarity between the two logistics nodes are calculated respectively;

[0024] The node association strength between the two logistics nodes is calculated by combining the transportation distance, the average transportation time, the logistics matrix similarity and the throughput similarity.

[0025] Optionally, the step of constructing a hypergraph structure in the regional logistics topology graph by combining the node association strength and the logistics node edge, and converting the regional logistics topology graph into a regional logistics hypergraph comprises the following steps:

[0026] In the regional logistics topology graph, for a logistics node subset composed of any multiple logistics nodes, if the node association strength between any two logistics nodes in the logistics node subset is greater than or equal to a preset association strength threshold, or the number of node edges of the logistics node edges between any two logistics nodes in the logistics node subset is greater than or equal to a preset edge number threshold, then a logistics hyperedge is generated for the logistics node subset;

[0027] For any of the logistics hyperedges, the hyperedge weight of the logistics hyperedge is calculated by combining the sum of the node association strengths of all the logistics nodes in the logistics node subset corresponding to the logistics hyperedge and the sum of the number of node edges;

[0028] Determining a hyperedge lifecycle of the logistics hyperedge based on a time decay function;

[0029] The hyperedge importance of the logistics hyperedge is calculated by combining the hyperedge weight of the logistics hyperedge, the hyperedge lifecycle, and the number of the logistics nodes in the logistics node subset corresponding to the logistics hyperedge;

[0030] Sorting all the logistics hyperedges in descending order of their importance, and performing hyperedge pruning on the sorted logistics hyperedges according to a preset last-place screening ratio;

[0031] Constructing a hyperedge tensor representation for all the retained logistics hyperedges to obtain a hypergraph structure corresponding to each logistics hyperedge;

[0032] All the hypergraph structures are mapped into the low-dimensional regional logistics topology map using hypergraph embedding technology to form a regional logistics hypergraph.

[0033] Optionally, the calculation formula for the node association strength is as follows:

[0034]

[0035] Where: D ij represents the node association strength between logistics node i and logistics node j, K ij represents the logistics matrix similarity between logistics node i and logistics node j, Q ij represents the throughput similarity between logistics node i and logistics node j, represents the weight of the mth freight train track between logistics node i and logistics node j, represents the transportation distance corresponding to the mth freight train track between logistics node i and logistics node j, It represents the average transportation time corresponding to the mth freight train track between logistics node i and logistics node j.

[0036] Optionally, constructing path constraints based on the logistics order information and the available train list, and generating multiple logistics hyperpaths for each logistics order in the regional logistics hypergraph based on the path constraints includes the following steps:

[0037] Path constraints are constructed based on the logistics order information and the available train list. The path constraints are as follows:

[0038]

[0039] Where: G is a binary decision variable indicating whether logistics order p is transported on freight train track q using a freight train of train type d. p represents the logistics throughput of logistics order p, C d N represents the upper limit of the loading capacity of a freight train of train type d. d represents the number of available trains of freight train type d, K represents the total number of train types of freight trains, A binary decision variable indicating whether logistics order p1 is transported on freight train track q using a freight train of train type d, A binary decision variable indicating whether logistics order p2 is transported on freight train track q using a freight train of train type d, Indicates the types of logistics freight included in logistics order p1. Indicates the types of logistics freight included in logistics order p2. Represents the goods compatibility matrix between logistics order p1 and logistics order p2;

[0040] Based on the path constraints and the hyperedge weights, a K shortest path algorithm is used in the regional logistics hypergraph to generate multiple logistics hyperpaths for each logistics order;

[0041] Counting the total number of super paths of all the logistics super paths; if the total number of super paths exceeds a preset path quantity threshold, respectively calculating the sum of the logistics throughputs of all the logistics nodes in each of the logistics super paths, and removing the logistics super paths whose sum of the logistics throughputs is not equal to 0;

[0042] If the total number of hyperpaths is less than or equal to the path number threshold, the hyperedge weight is adjusted according to a preset random perturbation range, and the K shortest path algorithm is used to generate multiple new logistics hyperpaths for each logistics order until the total number of hyperpaths exceeds the path number threshold;

[0043] Repeat the above-mentioned path removal step and new path generation step until the total number of super paths exceeds the path quantity threshold and the sum of the logistics throughput corresponding to any of the logistics super paths is equal to 0.

[0044] Optionally, the step of calculating the node heat of each logistics node according to all the logistics super paths includes the following steps:

[0045] For each of the logistics nodes, the node capacity of the logistics node is calculated according to the node logistics attributes of the logistics node;

[0046] According to all the logistics super paths, the node appearance frequency of the logistics node in the logistics super path and the node expected logistics volume of the logistics node in a future preset time period are counted;

[0047] The initial node heat of the logistics node is calculated by combining the node capacity, the node occurrence frequency and the expected logistics volume of the node, and the time decay function is introduced into the initial node heat to obtain the basic node heat of the logistics node;

[0048] Based on the basic node heat of all the logistics nodes and according to the connection relationship between the logistics nodes in all the logistics super paths, a first-order node heat transfer matrix between all the logistics nodes is constructed;

[0049] Using matrix multiplication and calculating the first-order node heat transfer matrix to obtain a high-order node heat transfer matrix between all the logistics nodes;

[0050] The basic node heats of all the logistics nodes are corrected by the high-order node heat transfer matrix to obtain the node heats of all the logistics nodes.

[0051] Optionally, constructing a node load balancing constraint condition based on the node heat, selecting an optimal super-path combination from all the logistics super-paths based on the node load balancing constraint condition and taking the shortest total path as the optimization goal to generate an optimal logistics strategy for the target area includes the following steps:

[0052] A node load balancing constraint condition is constructed based on the node heat, and the node load balancing constraint condition is:

[0053] max(B(v))-min(B(v))≤ε

[0054] Where: B(v) = H(v) / E(v), B(v) represents the node load balancing factor of any logistics node v, H(v) represents the node heat of the logistics node v, E(v) represents the node capacity of the logistics node v, and ε represents the maximum load imbalance;

[0055] Based on the node load balancing constraint condition and taking the shortest total path as the optimization goal, a multi-objective optimization function of the logistics strategy optimization process is constructed. The multi-objective optimization function is:

[0056]

[0057] Where: L(W x ) represents the total path length of all the logistics super paths in the x-th super path combination, represents the average value of the node load balancing factors of all the logistics nodes, α and β are weight coefficients, and α + β = 1;

[0058] Generating a plurality of initial super-path combinations from all the logistics super-paths by random combination;

[0059] Taking all the initial hyperpath combinations as the initial population, based on the multi-objective optimization function and using a hybrid metaheuristic algorithm combining a genetic algorithm and a simulated annealing algorithm, the optimal population individuals are selected from the initial population to obtain the optimal hyperpath combination;

[0060] An optimal logistics strategy for the target area is generated according to the optimal super path combination.

[0061] In the second aspect, the present invention also provides a high-speed light-load railway freight logistics system based on the Internet of Things and big data, including a memory, a processor, and a computer program stored on the memory and runnable on the processor. When the processor executes the computer program, it implements the high-speed light-load railway freight logistics method based on the Internet of Things and big data as described in the first aspect.

[0062] In a third aspect, the present invention also provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor, configures the processor to execute the high-speed light-load railway freight logistics method based on the Internet of Things and big data as described in the first aspect.

[0063] The beneficial effects of the present invention are:

[0064] The present invention provides an efficient, flexible and comprehensive optimization solution for complex regional logistics systems by comprehensively applying advanced technologies such as historical data analysis, network topology modeling and real-time order processing. Compared with the existing technology, the present invention has significant advantages and beneficial effects. First, the present invention more accurately describes the complex structure of the logistics network and the relationship between nodes by constructing a regional logistics hypergraph, thereby laying a solid foundation for subsequent path planning and resource allocation. Secondly, the present invention introduces innovative concepts such as node association strength and node heat, which enables the system to perceive the network status more intelligently and achieve load balancing during the scheduling process, effectively avoiding the local congestion problem common in traditional methods. Furthermore, the present invention adopts a strategy for generating multiple logistics super paths, which greatly increases the flexibility and adaptability of scheduling and can better cope with complex and changing logistics needs. In addition, the present invention simultaneously considers the two objectives of shortest path and load balancing during the optimization process, and achieves an improvement in the overall performance of the system through multi-objective optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of a high-speed, light-load railway freight logistics method based on the Internet of Things and big data in one embodiment of the present application.

[0066] Figure 2 This is a schematic diagram of a regional logistics topology map in one embodiment of the present application.

[0067] Figure 3 This is a schematic diagram of a regional logistics hypermap in one embodiment of the present application. DETAILED DESCRIPTION

[0068] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0069] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0070] In today's complex logistics environment, traditional logistics management systems face severe challenges. These systems primarily rely on simple algorithms and rules, such as shortest path algorithms or greedy algorithms, which perform well when handling small-scale, relatively stable logistics networks. However, as the scale and complexity of logistics networks continue to increase, the limitations of traditional systems are becoming increasingly prominent. Specifically, these systems use static or semi-static models, which make it difficult to adapt to the highly dynamic nature of modern logistics networks. Furthermore, traditional systems often focus on local optimization and lack a global perspective, resulting in the optimal allocation strategy for a specific route section or node, but unsatisfactory overall network efficiency. These issues seriously affect the key performance indicators of logistics systems, such as delivery efficiency, resource utilization, and cost control.

[0071] Consider a large agricultural region with multiple agricultural logistics parks connected by a high-speed, light-load rail network. Each logistics park has its own unique geographical location, logistics processing capabilities, and cargo type preferences. In this scenario, traditional logistics management systems struggle to effectively handle the following complexities: First, logistics demand is highly dynamic, with cargo flow patterns between logistics parks fluctuating frequently due to factors such as season, weather, and market demand. Second, different types of agricultural products require different transportation conditions, such as temperature control and timeliness, which complicates route planning. Furthermore, the number of trains, load capacities, and operating schedules within the high-speed, light-load rail network are all subject to dynamic change. In this scenario, traditional systems struggle to adjust logistics strategies in real time, resulting in uneven resource allocation. Some logistics parks may experience cargo backlogs while others experience idle resources.

[0072] If these technical problems cannot be effectively solved, serious consequences will occur. First, logistics efficiency will drop significantly, resulting in delays in the transportation of agricultural products, affecting product quality and market supply. Secondly, resource utilization will be low, resulting in increased operating costs and economic losses. Furthermore, due to the inability to accurately predict and respond to changes in logistics demand, the system may frequently be overloaded or idle, affecting the stability of the entire network. In addition, the lack of a global optimization perspective may lead to a contradiction between local optimization and overall benefits, such as some routes being overcrowded while other routes are idle. These problems not only affect the operational performance of a single logistics park, but also have a negative impact on the synergy and sustainability of the entire agricultural logistics network. Therefore, there is an urgent need for an advanced logistics management method that can comprehensively consider network dynamics, multi-dimensional constraints and global optimization. In the above-mentioned high-speed light-load railway freight logistics scenario, we are faced with complex problems such as dynamic changes in logistics demand, diversified transportation requirements and unbalanced resource allocation. In order to solve these technical problems, this application has conducted in-depth solution exploration.

[0073] First of all, considering that the cargo flow pattern between logistics parks changes frequently, a method that can reflect the status of the logistics network in real time is needed. Therefore, the idea of ​​obtaining the historical logistics data and logistics park coordinates of all agricultural logistics parks in the target area based on the Internet of Things technology is proposed. This method can provide an accurate and timely data basis for subsequent analysis. In order to fully understand the structure and characteristics of the logistics network, this application proposes to use agricultural logistics parks as logistics nodes, assign attributes to the nodes based on historical logistics data and logistics park coordinates, and construct a regional logistics topology map based on the train track topology of the target area. This method not only takes into account the physical network structure, but also incorporates historical logistics data, which can better reflect the actual logistics situation. However, a simple topology map may not be able to fully express the complex relationship between nodes. Therefore, this application further proposes a method for calculating the node association strength and constructs a hypergraph structure based on this. The hypergraph structure can better describe the complex relationship between multiple nodes and provide richer information for subsequent path planning.

[0074] In terms of path planning, traditional methods often only consider a single factor such as the shortest distance. This application proposes a more comprehensive method, which is to obtain a list of logistics orders and a list of available trains within a preset time period in the future, and construct path constraints based on this. This method not only takes into account logistics demand, but also takes into account available resources, and can better balance supply and demand. In order to further optimize the logistics strategy, this application introduces the concept of node heat and constructs node load balancing constraints based on this. This method can effectively avoid the situation where some nodes are overloaded while other nodes are idle, and improve the resource utilization of the entire network. Finally, this application proposes to use the shortest total path as the optimization goal, select the optimal super path combination from the multiple generated logistics super paths, and generate the optimal logistics strategy for the target area. This method takes into account both overall efficiency and load balancing, and can find better solutions in complex logistics environments.

[0075] Therefore, this application proposes a high-speed light-load railway freight logistics method based on the Internet of Things and big data. Figure 1 The figure is a flow chart of a high-speed light-load railway freight logistics method based on the Internet of Things and big data in one embodiment. Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps in the above process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps. Figure 1 As shown, the high-speed light-load railway freight logistics method based on the Internet of Things and big data disclosed in the present invention specifically includes the following steps:

[0076] S101. Obtain historical logistics data and logistics park coordinates of all agricultural logistics parks in the target area.

[0077] First, the geographic coordinates of each agricultural logistics park are precisely located using a geographic information system (GIS) and satellite positioning technology, using the WGS84 coordinate system, achieving meter-level accuracy. For example, the coordinates of a logistics park can be expressed as (39.9042°N, 116.4074°E). Simultaneously, historical logistics data from the past three to five years is collected, including daily inbound and outbound volumes, cargo types, and transportation methods. Data collection utilizes a multi-source fusion approach, integrating logistics park management systems, electronic documents, and IoT sensors to ensure comprehensiveness and accuracy. During data preprocessing, a sliding window median filter is used to remove outliers, with a window size set to seven days. Missing data is filled using multiple interpolation, taking into account time series characteristics and the influence of correlated variables. For example, a logistics park's average daily throughput in January 2023 might be 1,000 tons, primarily vegetables, fruit, and grain, 80% of which is transported by rail. Data normalization utilizes the Z-score method. This resulting data has zero mean and unit variance, facilitating subsequent analysis. Through this step, we can construct a data set D={(x1,y1,d1),...,(x n ,y n ,d n )}, where (x, y) is the coordinate and d is the historical data vector. This dataset provides a solid foundation for subsequent modeling. With over 98% data integrity, hourly temporal resolution, and spatial accuracy within 10 meters, it lays the foundation for accurately characterizing the dynamic characteristics of regional logistics networks.

[0078] S102. Take the agricultural logistics park as a logistics node, assign node attributes to the logistics node based on historical logistics data and logistics park coordinates, and construct a regional logistics topology map of the target area based on the train track topology of the target area and in combination with all logistics nodes.

[0079] Among them, the attribute vector A of each logistics node is defined as (a1, a2, ..., a k ), which includes the following key attributes: average daily throughput (tons / day), storage capacity (cubic meters), main cargo types (vector representation), handling efficiency (tons / hour), connected railway level (level 1-5), etc. For example, a node attribute might be A = (1000, 50000, [0.6, 0.3, 0.1], 100, 3), indicating an average daily throughput of 1000 tons, a storage capacity of 50,000 cubic meters, a cargo ratio of 60% vegetables, 30% fruit, and 10% grain, a handling efficiency of 100 tons / hour, and connected to a third-level railway.

[0080] Next, we construct an initial network structure based on the train track topology of the target area. This is represented by the adjacency matrix M from graph theory, where M[i][j] indicates whether nodes i and j are directly connected. Given the characteristics of the railway network, edge weights w[i][j] are introduced to represent the actual transportation distance or time between two nodes. For example, for a network with five nodes, its adjacency matrix might be:

[0081]

[0082] Among them, 1 means direct connection, and 0 means no connection.

[0083] In order to describe the network characteristics more accurately, we introduce the edge attribute vector E=(e1,e2,...,e m ), including transportation capacity (tons / day), average transportation speed (km / h), line condition score (0-1), etc. Combining node attributes and edge attributes, a complete regional logistics topology graph G = (V, E, A, W) is constructed, where V is the node set, E is the edge set, A is the node attribute set, and W is the edge attribute set. A weighted graph data structure is used for storage. Each node contains its attribute information and a list of adjacent nodes, and each edge contains its attribute information. To improve the expressiveness of the graph, a time-varying factor is introduced, and the static topology graph is expanded into a dynamic time series graph G(t) to capture the changes in network characteristics in different time periods (such as weekdays / weekends, peak season / off-season). A sliding time window (such as 7 days) is used to update the attributes of the graph to ensure that the model can reflect the dynamic characteristics of the network.

[0084] S103. Calculate the node association strength between each logistics node in the regional logistics topology diagram based on the node attributes.

[0085] Among them, a node association strength function R(i,j) can be defined to quantify the degree of association between node i and node j.

[0086] The function can take into account any one or more of the following factors:

[0087] 1. Logistics flow similarity: S_flow(i,j)=1-|F_i-F_j| / max(F_i,F_j), where F_i and F_j are the average daily logistics flows of nodes i and j respectively.

[0088] 2. Cargo type similarity: Using cosine similarity, S_cargo(i,j) = (C_i·C_j) / (||C_i||||C_j||), where C_i and C_j are vectors representing the proportion of cargo types.

[0089] 3. Geographic proximity: S_geo(i,j) = exp(-d_ij / d_0), where d_ij is the geographic distance between the two nodes and d_0 is the characteristic distance (e.g., 100km).

[0090] 4. Transport capacity compatibility: S_cap(i,j) = min(T_i,T_j) / max(T_i,T_j), where T_i and T_j are the processing capabilities of the node.

[0091] 5. Historical collaboration frequency: S_hist(i,j) = H_ij / max(H_i,H_j), where H_ij is the number of historical collaborations between the two nodes, and H_i and H_j are the total number of collaborations between each node.

[0092] In order to capture the dynamic characteristics of the network, a time decay factor is introduced: R_t(i,j) = R(i,j) exp(-λΔt), where Δt is the time interval since the last interaction and λ is the decay coefficient. Next, a node association strength matrix M_R is constructed, where M_R[i][j] = R_t(i,j). In order to improve computational efficiency, a sparse matrix storage method is used, and only elements with an association strength exceeding a threshold (such as 0.1) are retained. In order to more comprehensively describe the relationship between nodes, a second-order association is introduced: for nodes i and k, if there is an intermediate node j that makes R_t(i,j) and R_t(j,k) both high, then i and k are considered to have a potential strong association. This can be achieved through matrix multiplication M_R 2 Finally, community discovery algorithms (such as the Louvain method) are applied to identify closely connected subgroups in the logistics network based on the association strength matrix, which helps with subsequent path planning and resource allocation. Through this multi-dimensional and dynamic association strength calculation method, more than 90% of the important relationships in the logistics network can be accurately captured, and the degree of consistency between the identified key node pairs and actual operational data exceeds 95%. This provides a reliable foundation for subsequent hypergraph construction and path optimization, effectively improving the overall efficiency and resilience of the logistics network.

[0093] S104. Construct a hypergraph structure in the regional logistics topology graph by combining the node association strength and the logistics node edge, and convert the regional logistics topology graph into a regional logistics hypergraph.

[0094] Among them, hypergraph is an advanced data structure in graph theory, which allows an edge (called hyperedge) to connect multiple nodes. This feature is very suitable for describing complex logistics network relationships. First, define the hyperedge generation rule. Based on the node association strength matrix M_R calculated in the previous step, set the association strength threshold θ (such as 0.6). For any node subset S = {v1, v2, ..., v k}, a hyperedge e is generated if the following conditions are met:

[0095]

[0096] This ensures that there are strong connections between the nodes within the hyperedge and no nodes that should be included are omitted.

[0097] The weight w(e) of a hyperedge is defined as the geometric mean of the association strengths of all the node pairs it contains:

[0098] w(e)=(∏ ij M_R[i][j])^(1 / C(k,2)), where C(k,2) is the number of combinations of 2 out of k nodes.

[0099] To capture the dynamic nature of logistics networks, a time-varying hypergraph H(t) is introduced. A hyperedge lifecycle function, L(e,t), is defined to represent the validity of a hyperedge e at time t: L(e,t) = exp(-λ(t-t0)), where t0 is the creation time of the hyperedge and λ is the decay coefficient. In practical applications, a large number of hyperedges may be generated. To improve computational efficiency, a hyperedge pruning strategy is employed. A hyperedge importance metric, I(e), is defined as: I(e) = w(e)*|Ve|*L(e,t), where |Ve| is the number of nodes contained in the hyperedge. Only hyperedges ranked in the top K% by importance are retained.

[0100] Finally, hypergraph embedding techniques (such as Hyper-SAGNN) are applied to map the hypergraph structure into a low-dimensional vector space. Each node v and hyperedge e is mapped to a d-dimensional vector (typically d = 128). The embedding process takes into account node attributes, hyperedge structure, and temporal dynamics. This hypergraph-based representation method can more comprehensively capture the complex relationships in the logistics network, improving the network structure representation capability by 30-40% compared to traditional graph models.

[0101] S105. Obtain a list of logistics orders and a list of available trains in the target area within a preset time period in the future, and identify the logistics order information of each logistics order in the logistics order list.

[0102] A preset time period, typically 7-30 days in the future, is determined. This time window should be long enough for effective planning but not too long to avoid cumulative forecast errors. The available train list can be obtained from the target area's train management system. It includes the freight train types of all empty freight trains within the preset future time period, as well as the number of empty freight trains within each freight train type. More specific information is included in the list, including the number of carriages, loading limits, and freight type restrictions. For the logistics order list, a time series forecasting model can be used to estimate future order volumes. Here, the SARIMA (Seasonal Autoregressive Integrated Moving Average) model is combined with a machine learning approach: the SARIMA(p,d,q)(P,D,Q)m model, where p, d, and q represent the autoregressive order, differencing order, and moving average order, respectively; P, D, and Q represent the corresponding parameters of the seasonal component; and m represents the seasonal period. Model parameters are determined using maximum likelihood estimation. To improve forecast accuracy, the XGBoost algorithm is combined with external factors such as historical order data, holiday information, and weather forecasts as feature inputs. The model is evaluated using RMSE (root mean square error) and MAPE (mean absolute percentage error), and the model parameters are continuously updated through the rolling forecast method.

[0103] After order forecasting is completed, information is identified and extracted for each order. Order information includes:

[0104] 1. Logistics starting point (x_start, y_start): Use geocoding technology to convert the address into longitude and latitude coordinates.

[0105] 2. Logistics end point (x_end, y_end): also obtained using geocoding technology.

[0106] 3. Logistics throughput Q: measured in tons or cubic meters. For irregularly shaped goods, 3D scanning technology is used to accurately measure the volume.

[0107] 4. Logistics freight type C: adopts a standardized cargo classification and coding system, such as the HS code (Harmonized International Commodity Classification Code).

[0108] During the information extraction process, natural language processing (NLP) techniques are used to process unstructured order descriptions. The specific steps are as follows: Stop words and punctuation are removed, and word stemming is performed. Key information such as location, date, and quantity are identified. Relationships between entities, such as origin, destination, and cargo type, are determined. The extracted information is mapped to a standard format. Finally, all order information is consolidated into a standardized data structure, such as:

[0109] Order={ID,(x_start,y_start),(x_end,y_end),Q,C,t_expected,priority}

[0110] Where t_expected is the expected delivery time, and priority is the order priority (calculated based on factors such as customer level and timeliness of goods).

[0111] S106. Construct path constraints based on the logistics order information and the available train list, and generate multiple logistics hyperpaths for each logistics order in the regional logistics hypergraph based on the path constraints.

[0112] Among them, first, the loading capacity constraint is constructed. Taking into account the characteristics of different goods, the cargo compatibility matrix is ​​introduced to indicate that two different types of goods m and n can be loaded together. Next, a logistics hyperpath is generated in the regional logistics hypergraph H = (V, E). The hyperpath P is defined as a series of connected hyperedges e1, e2, ..., en, so that the starting point s∈e1, the end point t∈en, and for any adjacent ei and ei+1, there is at least one common node. To generate multiple candidate hyperpaths for each order, a modified version of the K shortest path algorithm (Yen'algorithm) can be used. When applying this algorithm on a hypergraph, the particularity of the hyperedges needs to be considered. In order to ensure that the generated hyperpath satisfies the condition that the sum of the logistics throughput of all logistics nodes is 0, a flow balance constraint can be introduced, as follows:

[0113] ∑e∈δ+(v)f(e)-∑e∈δ-(v)f(e)=b(v),

[0114] where δ+(v) and δ-(v) are the sets of hyperedges starting and ending at v, respectively, f(e) is the flow on hyperedge e, and b(v) is the net outflow of node v (Q for the starting point, -Q for the end point, and 0 for other nodes).

[0115] Finally, for each order, K candidate hyperpaths that satisfy all constraints are generated. To increase path diversity, random perturbations are introduced during the path generation process. This method generates multiple feasible logistics hyperpaths for each order, satisfying load constraints while ensuring flow balance at logistics nodes.

[0116] S107. Calculate the node heat of each logistics node based on all logistics super paths.

[0117] The node heat index H(v) reflects the importance and load of node v in the entire logistics network. The calculation of node heat can consider the following factors:

[0118] Path frequency: The frequency with which a node appears in all candidate hyperpaths.

[0119] Traffic weight: the expected logistics volume passing through the node.

[0120] Time distribution: load conditions in different time periods.

[0121] Node capacity: The upper limit of a node's processing power.

[0122] To capture the temporal dynamics of heat, a time decay factor can be introduced. Next, a node heat matrix is ​​constructed to represent the heat transfer between two different nodes. This can be calculated using a random walk model. To more comprehensively describe the heat relationships between nodes, higher-order heat transfer can be introduced, with the order of transfer typically being 2 or 3. Based on node heat, hotspots and potential bottlenecks in the logistics network can be identified.

[0123] S108. Construct node load balancing constraints based on node heat. Based on the node load balancing constraints and with the shortest total path as the optimization goal, select the optimal super-path combination from all logistics super-paths to generate the optimal logistics strategy for the target area.

[0124] First, we construct the node load balancing constraint. Define the load balancing factor B(v):

[0125] B(v)=H(v) / C(v)

[0126] Where H(v) is the popularity of node v and C(v) is the processing capacity of node v. Ideally, we want B(v) of all nodes to be as close as possible. Therefore, we define the load balancing constraint:

[0127] max(B(v))-min(B(v))≤ε,

[0128] Where ε is the maximum load imbalance allowed, usually set to 0.2-0.3.

[0129] Next, we define the optimization objective function. Taking into account the two objectives of minimizing the total path and balancing the load, we construct a multi-objective optimization function. To solve this NP-hard problem, we employ a hybrid meta-heuristic algorithm combining an improved genetic algorithm (GA) with simulated annealing (SA). The resulting optimal logistics strategy includes the optimal hyperpath for each order, the expected load at each node, the overall network load balance, the total transport distance, and the estimated completion time. This hybrid meta-heuristic optimization method can quickly find high-quality solutions in complex logistics networks.

[0130] In one embodiment, an agricultural logistics park is used as a logistics node, and node attributes are assigned to the logistics node based on historical logistics data and logistics park coordinates. A regional logistics topology map of the target area is constructed based on the train track topology of the target area and combined with all logistics nodes, including the following steps:

[0131] Use agricultural logistics parks as logistics nodes;

[0132] For each logistics node, assign node coordinate attributes to the logistics node according to the logistics park coordinates;

[0133] Utilize natural language technology to extract historical logistics freight types at logistics nodes and the historical average throughput of each different historical logistics freight type from historical logistics data;

[0134] Combine historical logistics freight types and historical average throughput to build a historical logistics matrix corresponding to the logistics node;

[0135] Assign the historical logistics matrix as the node logistics attributes of the logistics node;

[0136] If there are M freight train tracks between any two agricultural logistics parks in the train track topology of the target area, then M logistics node edges are generated between the corresponding two logistics nodes to obtain the regional logistics topology map of the target area.

[0137] In this implementation, agricultural logistics parks are first used as logistics nodes. For each logistics node, node coordinate attributes are then assigned based on the logistics park's coordinates. This process involves multiple technical areas, including geographic information system (GIS) technology, coordinate system conversion, and precision control. Specifically, the WGS84 (World Geographic System 1984) global geographic coordinate system is commonly used. It is currently the most widely used geographic coordinate system and is compatible with most GPS devices and map services. The precise coordinates of each agricultural logistics park can be obtained through various methods, such as field measurements using a high-precision GPS receiver, which can achieve meter-level or even centimeter-level accuracy. After obtaining the coordinates, coordinate system conversion and precision calibration are required. For example, a local coordinate system (such as the Gauss-Krüger projection) can be converted to the WGS84 coordinate system. The conversion formula is as follows: (φ, λ) = f(x, y), where (φ, λ) are the latitude and longitude in WGS84, (x, y) are the coordinates in the local coordinate system, and f is a complex conversion function that typically requires specialized GIS software (such as ArcGIS).

[0138] To improve coordinate accuracy, differential GPS (DGPS) or real-time kinematic (RTK) measurement can be used, which can improve accuracy to the centimeter level. At the same time, the effects of earth curvature and projection distortion need to be considered, especially when spanning large areas. Coordinate data is stored using double-precision floating-point numbers (double) to ensure sufficient accuracy. To handle dynamic changes in coordinate data (such as changes in the center point due to logistics park expansion), a regular update mechanism is implemented.

[0139] Next, natural language processing techniques are used to extract historical logistics data from logistics nodes to identify the historical freight types and the historical average throughput of each freight type. This step first requires the collection and organization of a large amount of historical logistics data, including but not limited to freight manifests, transportation records, and warehouse logs. Data preprocessing involves operations such as text cleaning, formatting standardization, and deduplication. Subsequently, natural language processing (NLP) techniques, particularly named entity recognition (NER) and relation extraction (RE) algorithms, are applied to identify and extract freight types and related information. For example, a model based on a BiLSTM-CRF (bidirectional long short-term memory network combined with conditional random fields) is used to identify freight type, quantity, and time information from text. Model training requires a large amount of annotated data, and semi-supervised learning methods such as distant supervision or active learning can be used to reduce manual annotation costs. The identified freight types are then classified and integrated using a hierarchical clustering algorithm to handle synonyms and hyponymy (e.g., "apple" and "fruit").

[0140] When extracting throughput information, it is necessary to consider the conversion and standardization of different units of measurement. Time series analysis methods such as moving average or exponential smoothing are used to calculate the historical average throughput of each type of cargo. To deal with noise and outliers in the data, robust statistics techniques such as the median absolute deviation (MAD) method can be applied to identify and process abnormal data points. Ultimately, a structured data table containing the freight type and the corresponding average throughput is generated for each logistics node. This method can not only extract key information from unstructured text, but also process large-scale data sets to provide accurate historical logistics pattern analysis.

[0141] Next, a historical logistics matrix corresponding to each logistics node needs to be constructed by combining historical freight types and historical average throughput. This process first requires defining a standardized freight type classification system, such as the International Standard Commodity Classification Code (SITC) or the Harmonized System of Customs Description and Coding (HS Code). For each logistics node, an n×m matrix M is created, where n is the number of standardized freight types and m is the time dimension (e.g., divided by month). The matrix element M[i,j] represents the average throughput of the i-th freight type in the j-th time unit. To handle sparse data, matrix factorization techniques such as singular value decomposition (SVD) or non-negative matrix factorization (NMF) can be used to reduce dimensionality and fill missing values. Considering the seasonal and cyclical characteristics of logistics data, time series decomposition methods such as STL (Seasonal and Trend decomposition using Loess) can be applied to separate trend, seasonal, and residual components. Ultimately, each logistics node will have a corresponding standardized historical logistics matrix. This representation method not only compresses data storage space but also preserves rich temporal and categorical information, providing a structured data foundation for subsequent pattern recognition and predictive analysis. Finally, assigning the historical logistics matrix to the node logistics attributes of the logistics node is to convert static data into dynamic network features, which can be achieved by using a graph database (such as Neo4j) or a specialized network analysis library (such as NetworkX).

[0142] If any two agricultural logistics parks share M freight train tracks in the target region's train track topology, then M logistics node edges are generated between the corresponding two logistics nodes, resulting in a regional logistics topology map for the target region. Specifically, detailed train track topology information for the target region must first be obtained, including the number of tracks, type (e.g., high-speed, conventional), and capacity. This information can be obtained from the railway department or gathered through a combination of remote sensing image analysis and field research. For each pair of logistics nodes, the direct rail connection between them is examined. A Depth-First Search (DFS) or Breadth-First Search (BFS) algorithm is used to explore all possible paths between the two nodes, identifying M different freight train tracks. Each identified track generates a corresponding edge in the logistics topology map. These edges not only represent physical connections but also require attributes to describe their characteristics, such as distance, average travel time, capacity, and reliability index. Edge weights can be a weighted combination of these attributes, with the weight coefficients determined using multi-objective optimization methods. Given the complexity of the railway network, some edges may be directed (one-way transport) or bidirectional. Therefore, directed graphs or mixed graphs are needed to represent this complexity.

[0143] In one embodiment, calculating the node association strength between logistics nodes in a regional logistics topology graph based on node attributes includes the following steps:

[0144] For any two logistics nodes, the transportation distance and average transportation time between the two logistics nodes are calculated based on the node coordinate attributes;

[0145] Based on the node logistics attributes, the logistics matrix similarity and throughput similarity between two logistics nodes are calculated respectively;

[0146] The node association strength between two logistics nodes is calculated by combining the transportation distance, average transportation time, logistics matrix similarity and throughput similarity.

[0147] In this implementation, calculating the transport distance and average transport time between any two logistics nodes based on the node coordinate attributes is a fundamental step in building a logistics network model. First, using the latitude and longitude coordinates (φ1, λ1) and (φ2, λ2) of each node, the Haversine formula is used to calculate the great-circle distance. This method accounts for the curvature of the Earth and is more accurate than straight-line distance. For shorter distances, the Vincenty formula can be used for higher accuracy. Considering that actual transport routes typically do not follow perfectly great-circle paths, a path coefficient k (typically between 1.1 and 1.5) is introduced, where the actual transport distance D = k*d. Next, calculating the average transport time requires considering multiple factors: base transport time, loading and unloading time, and transit time. Base transport time T_base = D / v, where v is the average transport speed (determined by the mode of transport, such as rail or road). Loading and unloading times T_load and T_unload can be estimated based on the type of cargo and the efficiency of the logistics park facilities. If transit occurs, the transit time T_transfer is also added. Therefore, the total average delivery time, T, is T_base + T_load + T_unload + T_transfer. Finally, the calculated delivery distances and average delivery times are stored in an n×n matrix, where n is the number of logistics nodes. The matrix element (i, j) represents the delivery distance and delivery time from node i to node j.

[0148] Next, the logistics matrix similarity and throughput similarity between the two logistics nodes are calculated based on the node logistics attributes. First, for the calculation of logistics matrix similarity, considering the high-dimensional nature of the logistics matrix, the cosine similarity method is used. Assume that the logistics matrices of nodes A and B are MA and MB, respectively. These matrices are flattened into vectors VA and VB. The cosine similarity calculation formula is: cos_sim(A,B) = (VA·VB) / (||VA||||VB||), where · represents the dot product and ||V|| represents the Euclidean norm of the vector. This method can effectively capture the similarity of logistics patterns and is not affected by the absolute numerical size. To handle the possible missing values ​​in the logistics matrix, matrix completion techniques such as singular value decomposition (SVD) or collaborative filtering algorithms can be used. For high-dimensional sparse matrices, principal component analysis (PCA) can be used to reduce the dimensionality before calculating the similarity to improve computational efficiency.

[0149] When calculating throughput similarity, we use the normalized Euclidean distance to account for the scale differences between different logistics nodes. Let the throughput vectors of nodes A and B be TA and TB, respectively (each element represents the throughput of a different cargo type). First, perform min-max normalization on the vectors: T'i = (Ti - min(T)) / (max(T) - min(T)). Then, calculate the Euclidean distance of the normalized vectors: dist(A, B) = sqrt(Σ(T'Ai - T'Bi)) 2 Throughput similarity is defined as: sim_throughput(A,B) = 1 / (1+dist(A,B)). This method not only takes into account the absolute difference in throughput, but also reduces the impact of scale effects through normalization. In order to comprehensively consider the importance of different types of goods, the cargo importance weight vector W can be introduced. The modified throughput similarity calculation formula is: sim_throughput_w(A,B) = 1 / (1+sqrt(ΣWi(T'wAi-T'wBi) 2 )).

[0150] Finally, the node association strength between two logistics nodes is calculated by combining transportation distance, average transportation time, logistics matrix similarity, and throughput similarity. This process requires integrating indicators from different dimensions into a single association strength indicator. This comprehensive association strength calculation method not only considers physical distance and time factors, but also incorporates business similarity measures, providing a comprehensive relationship evaluation foundation for subsequent network analysis and optimization. In this embodiment, the calculation formula for node association strength is as follows:

[0151]

[0152] Where: D ijrepresents the node association strength between logistics node i and logistics node j, K ij represents the logistics matrix similarity between logistics node i and logistics node j, Q ij represents the throughput similarity between logistics node i and logistics node j, represents the weight of the mth freight train track between logistics node i and logistics node j, represents the transportation distance corresponding to the mth freight train track between logistics node i and logistics node j, It represents the average transportation time corresponding to the mth freight train track between logistics node i and logistics node j.

[0153] In one embodiment, combining node association strength and logistics node edges to construct a hypergraph structure in a regional logistics topology graph, and converting the regional logistics topology graph into a regional logistics hypergraph includes the following steps:

[0154] In the regional logistics topology graph, for a logistics node subset composed of any multiple logistics nodes, if the node association strength between any two logistics nodes in the logistics node subset is greater than or equal to the preset association strength threshold, or the number of node edges between any two logistics nodes in the logistics node subset is greater than or equal to the preset edge number threshold, then a logistics superedge is generated for the logistics node subset;

[0155] For any logistics hyperedge, the hyperedge weight of the logistics hyperedge is calculated by combining the sum of the node association strengths of all logistics nodes in the logistics node subset corresponding to the logistics hyperedge and the sum of the number of node edges;

[0156] Determine the hyperedge life cycle of logistics hyperedge based on time decay function;

[0157] The hyperedge importance of the logistics hyperedge is calculated by combining the hyperedge weight, hyperedge life cycle and the number of logistics nodes in the logistics node subset corresponding to the logistics hyperedge;

[0158] Sort all logistics hyperedges in descending order of their importance, and perform hyperedge pruning on the sorted logistics hyperedges according to the preset last-place screening ratio;

[0159] Construct hyperedge tensor representations for all retained logistics hyperedges and obtain the hypergraph structure corresponding to each logistics hyperedge;

[0160] Hypergraph embedding technology is used to map all hypergraph structures into a low-dimensional regional logistics topology graph to form a regional logistics hypergraph.

[0161] In this embodiment, in the regional logistics topology map, for each generated subset S, two conditions need to be verified: the node association strength condition and the node edge quantity condition. The node association strength condition requires that the association strength R(i,j) between any two nodes i and j in the subset satisfies R(i,j)≥θ, where θ is a preset association strength threshold. This threshold can be determined through data analysis or expert experience, and a typical value may be between 0.7 and 0.9. The node edge quantity condition requires that the number of logistics node edges E(i,j) between any two nodes i and j in the subset satisfies E(i,j)≥η, where η is a preset edge quantity threshold. The selection of this threshold should take into account the connectivity and density of the network, and the possible value range is 1 to 5. This method can not only capture complex multi-node relationships, but also flexibly adjust the hyperedge generation strategy according to different threshold settings.

[0162] Reference Figure 2 , Figure 2 is a regional logistics topology diagram in one implementation, including logistics nodes A, B, C, D, E, F, and G. A solid line between nodes represents a logistics node edge, and the D value between nodes represents the node association strength. Assuming that the association strength threshold is 0.8 and the edge number threshold is 3, the node association strength between any two nodes among logistics nodes A, B, and C exceeds the association strength threshold, meeting the generation condition of logistics hyperedge. Therefore, a logistics hyperedge e1 between logistics nodes A, B, and C can be generated, as shown in Figure 3 Similarly, the number of node edges between logistics nodes BD and BE exceeds the edge number threshold, and the node association strength between nodes DE exceeds the association strength threshold, so a logistics hyperedge e2 between logistics nodes B, D and E can be generated, as shown in Figure 3 shown.

[0163] For any logistics hyperedge, the hyperedge weight of the logistics hyperedge can be calculated by combining the sum of the node association strengths of all logistics nodes in the logistics node subset corresponding to the logistics hyperedge and the sum of the number of node edges. This process first requires defining a calculation model for the hyperedge weight, which should be able to comprehensively reflect the association strength and connection density between nodes. To capture nonlinear relationships, an exponential function can be introduced. Considering the comparability of hyperedges of different sizes, a scale adjustment factor can also be introduced. The final hyperedge weight formula is as follows:

[0164] W=((ΣR(i,j)) / C(n,2))^α*((ΣE(i,j)) / C(n,2))^β*log(n)

[0165] Where R(i,j) is the strength of the association between nodes i and j, E(i,j) is the number of edges between nodes i and j, n is the number of nodes included in the hyperedge, C(n,2) is the number of combinations of 2 out of n nodes, used for normalization, α and β are weight coefficients, where α + β = 1. The log(n) term balances the weights of large and small hyperedges. This hyperedge weight calculation method not only comprehensively considers node association strength and connection density, but also more accurately reflects the actual importance of the hyperedge than simple node or edge counts.

[0166] Next, the hyperedge lifecycle of a logistics hyperedge is determined based on a time decay function. A commonly used time decay function is the exponential decay function: L(t) = exp(-λ(t-t0)), where t is the current time, t0 is the creation time of the hyperedge, and λ is the decay coefficient. The choice of the decay coefficient λ is crucial, as it determines how quickly the importance of the hyperedge decreases over time. λ can be determined by analyzing historical data using maximum likelihood estimation or least squares methods. To improve computational efficiency, a discretization method can be used to convert the continuous time function into discrete time steps, with the lifecycle value updated once per time step. Furthermore, an event-driven update mechanism can be implemented to trigger recalculation of the lifecycle when significant changes occur in the network structure. This method for determining the hyperedge lifecycle based on a time decay function not only reflects the dynamic characteristics of the logistics network but also provides a quantitative basis for the maintenance and updating of hyperedges.

[0167] Calculating the hyperedge importance of a logistics hyperedge by combining its hyperedge weight, lifecycle, and the number of logistics nodes in the corresponding logistics node subset is a key step in assessing the importance of a hyperedge within the entire network. This process requires the design of a computational model that comprehensively considers multiple factors. The calculation formula is as follows: I = W^α*L^β*N^γ, where W is the hyperedge weight, L is the current lifecycle value, N is the number of nodes included in the hyperedge, and α, β, and γ are weight exponents. This formula reflects the mutual influence of various factors through a multiplication form. The selection of weight exponents can be determined through various methods, such as expert evaluation, sensitivity analysis, or optimization using machine learning methods (such as random forests).

[0168] To balance the influence of different dimensions, each factor can be normalized: I = (W / W_max)^α*(L / L_max)^β*(N / N_max)^γ, where W_max, L_max, and N_max are the maximum values ​​of the corresponding factors across all hyperedges. Considering the potential for nonlinear relationships between different factors, a more complex functional form can be introduced: I = f(W)*g(L)*h(N), where f, g, and h can be logarithmic functions, sigmoid functions, or other functions to capture sensitivity to changes across different ranges. To reflect the temporal dynamics of hyperedge importance, a time decay factor can be introduced: I(t) = I*exp(-λ(t_now - t_update)), where t_now is the current time, t_update is the last update time, and λ is the decay coefficient. This hyperedge importance calculation method not only comprehensively considers multiple characteristics of a hyperedge but also reflects its dynamic importance in the network.

[0169] Next, all logistics hyperedges are sorted in descending order of hyperedge importance. After the sorting is completed, the last-place screening ratio p (usually between 0.1 and 0.3) needs to be determined. The number of screened hyperedges is in Indicates rounding down. The pruning process starts from the end of the sorted list and removes the last k hyperedges. However, simple last-place screening may cause sudden changes in network connectivity. In order to maintain the overall structure of the network, more complex pruning strategies can be adopted: 1) Connectivity maintenance: Before removing a hyperedge, check whether it will cause the network to split. If so, keep the hyperedge; 2) Progressive pruning: Instead of removing all low-importance hyperedges at once, prune them in small increments multiple times and evaluate the network performance each time; 3) Conditional pruning: Set pruning conditions, such as pruning only when the importance of the hyperedge is lower than the threshold θ and the node degree is greater than the threshold d. In order to evaluate the impact of pruning, network indicators before and after pruning can be calculated, such as average path length, clustering coefficient, centrality, etc. If the indicator change exceeds the preset threshold, some pruning operations can be rolled back.

[0170] Next, we construct a hyperedge tensor representation for all the retained logistics hyperedges and obtain the hypergraph structure corresponding to each logistics hyperedge. This process first requires designing a tensor structure suitable for hypergraph representation. Considering that logistics hyperedges may connect different numbers of nodes, a third-order tensor can be used. To represent a hypergraph, N is the total number of nodes and K is the maximum hyperedge cardinality (i.e., the maximum number of nodes a single hyperedge can contain). The tensor element T[i,j,k]=w indicates that the connection strength between nodes i and j in the k-th order relationship is w. For hyperedges with a cardinality less than K, the remaining dimensions can be padded with zeros. To improve storage and computational efficiency, a sparse tensor representation can be used, storing only non-zero elements. The construction process is as follows:

[0171] 1) Initialize an empty sparse tensor T;

[0172] 2) For each retained hyperedge e, obtain its node set V(e) and weight w(e);

[0173] 3) For each node pair (i, j) in V(e), update T[i, j, |V(e)|-1]+=w(e), where |V(e)| is the cardinality of the hyperedge e. Considering the directionality of the logistics relationship, the tensor can be expanded to the fourth order The last dimension represents the direction of the relationship (0 for undirected, 1 for directed). This tensor-based hypergraph representation method not only effectively captures complex multi-node relationships but also flexibly adapts to logistics networks of varying scale and complexity by adjusting the tensor structure and storage strategy.

[0174] Finally, hypergraph embedding technology is used to map all hypergraph structures into a low-dimensional regional logistics topology, forming a regional logistics hypergraph. This process requires selecting an appropriate hypergraph embedding algorithm. One possible approach is to extend the Node2Vec algorithm to the hypergraph scenario, called HyperNode2Vec. The core idea of ​​this algorithm is to generate node sequences through random walks and then use the Skip-gram model to learn low-dimensional representations of the nodes. The specific steps are as follows:

[0175] 1) Define the hyperedge random walk strategy: p(e|v)∝w(e)*|e|^α, where v is the current node, e is the hyperedge containing v, w(e) is the hyperedge weight, |e| is the hyperedge cardinality, and α is the influence factor of the hyperedge size;

[0176] 2) Generate a large number of random walk sequences;

[0177] 3) Apply the Skip-gram model to each sequence: maximize logP(N(v)|f(v)), where N(v) is the network neighborhood of v and f(v) is the feature representation of v;

[0178] 4) Use negative sampling and stochastic gradient descent to optimize model parameters. In order to maintain the global structure of the hypergraph, additional regularization terms can be introduced, such as the hyperedge preservation loss: L_hyper = Σ||Σf(v) / |e|-g(e)||2, where g(e) is the representation of the hyperedge e. Considering the dynamic characteristics of the logistics network, dynamic embedding techniques can be adopted, such as using recurrent neural networks (RNN) or temporal convolutional networks (TCN) to capture the temporal evolution of node representations. In order to process large-scale networks, distributed computing frameworks such as Apache Spark can be used to parallelize the embedding calculation process. In the selection of the dimension of the embedding space, adaptive methods can be used, such as determining the optimal dimension by analyzing the curve of the embedding quality (such as reconstruction error) as the dimension changes.

[0179] In one embodiment, path constraints are constructed based on logistics order information and a list of available trains. Generating multiple logistics hyperpaths for each logistics order in a regional logistics hypergraph based on the path constraints includes the following steps:

[0180] Construct path constraints based on logistics order information and available train lists;

[0181] Based on the path constraints and the hyperedge weights, the K shortest path algorithm is used in the regional logistics hypergraph to generate multiple logistics hyperpaths for each logistics order;

[0182] Count the total number of super paths for all logistics super paths. If the total number of super paths exceeds the preset path number threshold, calculate the sum of the logistics throughput of all logistics nodes in each logistics super path and remove the logistics super paths with a sum of logistics throughput not equal to 0.

[0183] If the total number of hyperpaths is less than or equal to the path number threshold, the hyperedge weights are adjusted according to the preset random perturbation range, and the K shortest path algorithm is used to generate multiple new logistics hyperpaths for each logistics order until the total number of hyperpaths exceeds the path number threshold;

[0184] Repeat the above path removal step and new path generation step until the total number of super paths exceeds the path number threshold and the sum of the logistics throughput corresponding to any logistics super path is equal to 0.

[0185] In this implementation, first, the logistics order information needs to be structured and processed to extract key fields such as the starting point, destination, cargo type, weight, volume, expected arrival time, etc. For each field, data cleaning and standardization are required, such as unified address format, unit conversion, etc. Next, the list of available trains is processed, including information such as train number, route, departure time, arrival time, and load limit. This information needs to be connected to the real-time train scheduling system to ensure the timeliness of the data. Next, the following path constraints can be constructed based on the above information:

[0186]

[0187] Where: G is a binary decision variable indicating whether logistics order p is transported on freight train track q using a freight train of train type d. p represents the logistics throughput of logistics order p, C d N represents the upper limit of the loading capacity of a freight train of train type d. d represents the number of available trains of freight train type d, K represents the total number of train types of freight trains, A binary decision variable indicating whether logistics order p1 is transported on freight train track q using a freight train of train type d, A binary decision variable indicating whether logistics order p2 is transported on freight train track q using a freight train of train type d, Indicates the types of logistics freight included in logistics order p1. Indicates the types of logistics freight included in logistics order p2. Represents the goods compatibility matrix between logistics order p1 and logistics order p2.

[0188] Next, based on the path constraints and hyperedge weights, a K-shortest path algorithm is used within the regional logistics hypergraph to generate multiple logistics hyperpaths for each logistics order. First, a suitable K-shortest path algorithm, such as the Yen algorithm or the Eppstein algorithm, must be selected. Taking the Yen algorithm as an example, its basic principle is to find the shortest path; systematically remove edges from the shortest path and recalculate it to find the next shortest path; and repeat this process until K paths are found. Implementing this algorithm within a hypergraph structure requires some modifications: defining the "remove" operation for a hyperedge as setting its weight to infinity rather than physically deleting it. Considering the dynamic nature of logistics networks, an incremental K-shortest path algorithm can be implemented: when the network structure undergoes minor changes, local adjustments based on existing paths are made instead of recalculating all paths. To ensure that the generated paths satisfy the previously defined constraints, constraint checking is required during the path search process. Lagrangian relaxation can be used to convert some hard constraints into penalty terms and include them in the objective function. This ensures constraint satisfaction while avoiding excessive contraction of the search space. To increase path diversity, random perturbations can be introduced: in each iteration, the weights of some hyperedges are randomly adjusted with a small probability. This helps explore more possible paths. This hypergraph path generation method based on K-shortest paths not only provides diverse path options but also effectively balances multiple optimization objectives.

[0189] Next, the total number of superpaths for all logistics superpaths is counted. If the total number of superpaths exceeds a preset path number threshold, the sum of the logistics throughput of all logistics nodes in each logistics superpath is calculated, and logistics superpaths with a total logistics throughput not equal to 0 are removed. This process involves path counting and threshold comparison. The path number threshold N_threshold is set to control computational complexity and ensure solution diversity. Typically, N_threshold can be set to K times the number of orders (K is the number of paths generated for each order). Path total count can be implemented using a simple counter. When the total number of paths P_total > N_threshold, the path screening mechanism is triggered. The core of the screening process is to calculate the sum of the logistics throughput of each superpath. For each superpath R, its total throughput is calculated as: T(R) = Σt(n), where n∈R and t(n) is the logistics throughput of node n. After the calculation is complete, paths with T(R) = 0 are retained, and paths with T(R) ≠ 0 are removed. The theoretical basis of this screening method is that nodes with a throughput of 0 may represent newly built or underutilized logistics nodes, and selecting these paths helps balance the network load.

[0190] If the total number of hyper-paths is less than or equal to the path quantity threshold, adjust the hyper-edge weights according to a preset random perturbation range, and use the K shortest path algorithm to generate multiple new logistics hyper-paths for each logistics order until the total number of hyper-paths exceeds the path quantity threshold. First, the way and range of random perturbation need to be defined. A possible method is to use Gaussian noise: w'(e) = w(e) * (1 + N(0, σ2)), where w(e) is the original weight of hyper-edge e, and N(0, σ2) is a Gaussian distribution with a mean of 0 and a variance of σ2. The choice of σ is crucial and can be set as a small proportion of the weight average, such as 5%. To avoid the weight becoming negative or too large, upper and lower limits can be set: w_min ≤ w'(e) ≤ w_max. The perturbation process can adopt a batch processing method, randomly selecting a certain proportion (such as 10%) of hyper-edges for adjustment each time, which can introduce sufficient randomness while maintaining the overall structure of the network. After the weight adjustment, the K shortest path algorithm needs to be run again. Considering the computational efficiency, an incremental K shortest path algorithm can be adopted: using the previously calculated results, only update the affected paths. Specifically, for each original path, calculate its length under the new weight. If the length change exceeds the threshold δ, recalculate the path. The generation process of new paths is similar to the previous steps.

[0191] Repeat the above path removal steps and new path generation steps until the total number of hyper-paths exceeds the path quantity threshold and the total logistics throughput corresponding to any logistics hyper-path is equal to 0. To avoid falling into a local optimal solution, the idea of simulated annealing can be introduced: accept a slightly worse solution with a certain probability. Specifically, after each iteration, calculate the quality Q(current) of the current solution. If Q(current) < Q(best), accept the current solution with a probability P = exp((Q(current) - Q(best)) / T), where T is the temperature parameter, which gradually decreases with iteration: T_i = T_0 * γ^i, and γ is the cooling rate (such as 0.98). This iterative optimization method can effectively search for high-quality path combinations in a complex solution space through dynamic adjustment and multi-strategy combination. Practice shows that this method can improve the solution quality compared with simple greedy algorithms or static programming methods, and can adapt to the dynamic changes of the logistics network at the same time.

[0192] In one implementation, calculating the node heat of each logistics node based on all logistics hyper-paths includes the following steps:

[0193] For each logistics node, calculate the node capacity of the logistics node according to the node logistics attributes of the logistics node;

[0194] According to all logistics hyper-paths, count the node occurrence frequency of the logistics node in the logistics hyper-path and the expected node logistics volume of the logistics node within a preset future time period;

[0195] The initial node heat of the logistics node is calculated by combining the node capacity, node occurrence frequency and node expected logistics volume. The time decay function is introduced into the initial node heat to obtain the basic node heat of the logistics node.

[0196] Based on the basic node heat of all logistics nodes and the connection relationship between logistics nodes in all logistics super paths, a first-order node heat transfer matrix between all logistics nodes is constructed;

[0197] The high-order node heat transfer matrix between all logistics nodes is calculated using matrix multiplication and based on the first-order node heat transfer matrix;

[0198] The basic node heat of all logistics nodes is corrected through the high-order node heat transfer matrix to obtain the node heat of all logistics nodes.

[0199] In this implementation, calculating the node capacity of each logistics node based on its node logistics attributes is a key step in quantifying the processing capacity of an agricultural logistics park. First, valid information must be extracted from the historical logistics matrix. This matrix contains different freight types and their historical average throughput, and can be represented as M = {(c1,v1),(c2,v2),...,(cn,vn)}, where ci represents the freight type and vi represents the corresponding historical average throughput. Calculating node capacity requires considering several factors: 1) Total throughput: C_total = Σvi, representing the sum of the throughput of all freight types; 2) Freight type diversity: D = -Σ(pi*logpi), where pi = vi / C_total, a diversity metric based on information entropy; 3) Maximum single-class throughput: C_max = max(vi), reflecting the node's ability to handle a single freight type; 4) Logistics park area: A, which can be obtained from the node coordinate attributes or specified separately; and 5) Equipment sophistication index: E, reflecting the technological level of the logistics park. Taking these factors into account, we can construct a node capacity calculation formula: C = α*C_total + β*D + γ*C_max + δ*A + ε*E, where α, β, γ, δ, and ε are weight coefficients that need to be adjusted based on actual conditions. This multi-factor node capacity calculation method not only fully reflects the processing capabilities of the agricultural logistics park but also adapts to dynamic changes and emergencies in the logistics network.

[0200] Calculating the frequency of occurrence of logistics nodes within all logistics hyperpaths, as well as the expected logistics volume of these nodes within a predetermined time period, is a key step in assessing the importance of agricultural logistics parks and predicting future loads. Node occurrence frequency can be calculated using a weighted frequency method: F(n) = Σ(wi*I(n,Pi)) / Σwi, where I(n,Pi) is an indicator function that is 1 if node n appears in path Pi and 0 otherwise, and wi is the weight of path Pi. Predicting the expected logistics volume at a node involves time series analysis and forecasting techniques. A basic forecast can be performed using the ARIMA model: V(t) = c + φ1V(t-1) + ... + φpV(tp) + θ1ε(t-1) + ... + θqε(tq), where φi and θi are model parameters and ε(t) is white noise. Taking into account the seasonal characteristics of agricultural logistics, the seasonal ARIMA model can be introduced: (1-L^s)(1-L)^dV(t)=(1+θ1L+...+θqL^q)(1+Θ1L^s+...+ΘQL^sQ)ε(t), where L is the lag operator and s is the seasonal cycle.

[0201] The initial node heat of a logistics node is calculated by combining node capacity, node frequency, and expected node volume. Introducing a time decay function into the initial node heat to determine the basic node heat of a logistics node is a key step in quantifying the dynamic importance of an agricultural logistics park. The initial node heat can be calculated using a weighted summation method: H_init = w1*C+w2*F+w3*V, where C is the node capacity, F is the node frequency, V is the expected volume, and w1, w2, and w3 are the corresponding weights. Weights can be determined using the Analytic Hierarchy Process (AHP): a judgment matrix is ​​constructed, eigenvectors are calculated, and the weights of each factor are obtained. Next, based on the basic node heat of all logistics nodes and the connectivity between logistics nodes in all logistics hyperpaths, a first-order node heat transfer matrix is ​​constructed between all logistics nodes. This first-order heat transfer matrix P can be represented as an N×N matrix, where N is the total number of agricultural logistics parks and Pij represents the probability of heat transfer from node i to node j. The calculation of matrix elements can be based on the direct connection relationship between nodes and the flow of goods: Pij = wij / Σkwik, where wij is the connection weight between nodes i and j, which can be defined as: wij = Σe∈Ef(e)*I(i,j∈e), E is the set of all logistics super paths, f(e) is the flow or importance of path e, and I(i,j∈e) is an indicator function, which is 1 when i and j are both in path e, and 0 otherwise.

[0202] Next, matrix multiplication is used to calculate the high-order node heat transfer matrix between all logistics nodes based on the first-order node heat transfer matrix, thereby capturing the long-distance heat propagation effect in the agricultural logistics network. The high-order transfer matrix Pk can be obtained by raising the first-order matrix P to the kth power: Pk = P^k. This calculation process reflects the distribution of heat after k steps of propagation. To improve computational efficiency, especially for large-scale agricultural logistics networks, optimized matrix power algorithms can be used, such as the binary decomposition method: P16 = ((P^2)^2)^2^2. This can reduce the computational complexity from O(n^3k) to O(n^3log(k)). Considering that only a limited number of propagation orders may be required in practical applications, a truncated order K can be defined. The weighted sum from P1 to PK is calculated: P_total = Σk = 1 ~ K(αk*P^k), where αk is the kth-order weight, which can be set to αk = β^(k-1). β is the attenuation coefficient, reflecting the impact of distance on heat propagation. Important nodes are given higher weights during the calculation. To account for possible community structures within the network, a block matrix technique can be employed, first calculating high-order transfers within a community, then between communities. Furthermore, to capture potential cyclic effects, a self-loop coefficient can be introduced into the high-order calculations: after each exponentiation operation, the diagonal elements are appropriately incremented. This high-order heat transfer analysis based on matrix multiplication can not only reveal long-distance interactions within agricultural logistics networks but also uncover potential heat propagation paths and patterns.

[0203] Finally, the basic node heat of all logistics nodes is corrected by the high-order node heat transfer matrix to obtain the node heat of all logistics nodes. The correction process can be expressed as: H_final = f(H_base, P_total), where H_base is the basic heat vector, P_total is the high-order transfer matrix, and f is the correction function. A simple correction method is linear combination: H_final = (1-α)*H_base+α*(P_total*H_base), where α is the balance parameter that controls the influence of global information. In order to capture nonlinear effects, a more complex correction function can be used, such as using a neural network model: H_final = NN(H_base, P_total*H_base), where NN is a trained neural network that can be trained using historical data. Considering that different agricultural logistics parks may have different sensitivities to global information, a node-specific correction coefficient can be introduced: H_final[i] = (1-αi)*H_base[i] + αi*(P_total[i,:]*H_base), where αi is the correction coefficient for node i and can be determined based on node characteristics (such as scale and connectivity). This global information-based heat correction method not only balances local and global characteristics but also captures the complex interactions within agricultural logistics networks, providing a more comprehensive decision-making basis for resource scheduling and risk prevention. By adjusting the correction function and parameters, this method can flexibly adapt to agricultural logistics networks of different types and dynamic characteristics.

[0204] In one embodiment, a node load balancing constraint condition is constructed based on node popularity. Based on the node load balancing constraint condition and with the shortest total path as the optimization goal, the optimal super-path combination is selected from all logistics super-paths to generate the optimal logistics strategy for the target area, including the following steps:

[0205] Construct node load balancing constraints based on node heat;

[0206] Based on the node load balancing constraint and taking the shortest total path as the optimization goal, a multi-objective optimization function for the logistics strategy optimization process is constructed;

[0207] Generate multiple initial hyperpath combinations from all logistics hyperpaths by random combination;

[0208] All initial hyperpath combinations are used as the initial population, and a hybrid metaheuristic algorithm based on a multi-objective optimization function and a combination of genetic algorithm and simulated annealing algorithm is used to select the best population individuals from the initial population to obtain the optimal hyperpath combination.

[0209] Generate the optimal logistics strategy for the target area based on the optimal hyperpath combination.

[0210] In this embodiment, constructing node load balancing constraints based on node heat is a key step in achieving logistics network optimization. The goal of the node load balancing constraint is to ensure that the loads of each node in the network are relatively balanced, avoiding the situation where some nodes are overloaded while other nodes are idle. The construction process first needs to define the node load balancing factor, which can be expressed as: B(v) = H(v) / E(v), where B(v) represents the node load balancing factor of any logistics node v, H(v) represents the node heat of logistics node v, and E(v) represents the node capacity of logistics node v. The load balancing constraint can be expressed as: max(B(v))-min(B(v))≤ε, where ε represents the maximum load imbalance.

[0211] Next, based on the node load balancing constraint and taking the shortest total path as the optimization goal, a multi-objective optimization function for the logistics strategy optimization process is constructed. The multi-objective optimization function can be expressed as:

[0212]

[0213] Where L(W x ) represents the total path length of all logistics super paths in the xth super path combination, represents the average value of the node load balancing factor for all logistics nodes, α and β are weight coefficients, and α + β = 1. This multi-objective optimization function not only balances efficiency and balance but also flexibly adapts to various complex constraints and optimization requirements.

[0214] Next, a variety of initial hyperpath combinations are generated from all logistics hyperpaths through random combination. This process first requires defining the structure of the combination, which can be expressed as a vector C = (p1, p2, ..., pn), where pi represents the hyperpath selected for the i-th order. The random combination generation process can be carried out in the following way: for each order i, a path pi is randomly selected from its available hyperpath set Pi. In order to improve the quality and diversity of the combination, preference random selection can be adopted: P(pi) ∝ exp(-β*L(pi)), where L(pi) is the length of path pi, and β is a temperature parameter that controls the degree of randomness. In order to ensure that the generated combination meets the basic constraints, a rejection sampling strategy can be adopted: repeatedly generate combinations until the constraints are met.

[0215] Then, using all initial hyperpath combinations as the initial population, a hybrid metaheuristic algorithm combining a genetic algorithm and a simulated annealing algorithm is used to select the optimal population individuals from the initial population based on a multi-objective optimization function to obtain the optimal hyperpath combination. This hybrid algorithm combines the global search capabilities of the genetic algorithm (GA) with the local search capabilities of the simulated annealing algorithm (SA). The main steps of the algorithm include:

[0216] 1) Initialization: Use the initial combination of the previous steps as the initial population.

[0217] 2) Fitness evaluation: Use a multi-objective optimization function to evaluate the fitness of each individual.

[0218] 3) Selection: The parent individuals are selected using the tournament selection method.

[0219] 4) Crossover: Generate offspring using uniform crossover or sequential crossover.

[0220] 5) Mutation: Randomly mutate the offspring with a certain probability.

[0221] 6) SA local search: SA is applied to some individuals for local optimization. The cooling function of SA can be defined as: T(t) = T0*α^t, where T0 is the initial temperature and α is the cooling rate.

[0222] 7) Elite retention: retain a certain proportion of the best individuals and directly enter the next generation.

[0223] 8) Population update: replace some individuals according to fitness.

[0224] 9) Termination judgment: reaching the maximum number of iterations or fitness convergence. To handle multi-objective problems, the NSGA-II algorithm can be used for non-dominated sorting and crowding calculation. To improve the algorithm's adaptability, adaptive parameter adjustment can be used: the crossover rate pc and mutation rate pm are dynamically adjusted as the population diversity changes. Considering the dynamic nature of the problem, a dynamic population strategy can be designed: the population size and structure can be dynamically adjusted according to environmental changes. To avoid premature convergence, a restart mechanism can be introduced: when the population diversity drops below a certain threshold, the population is partially reinitialized. In the SA phase, an adaptive neighborhood search strategy can be used: the neighborhood structure can be dynamically adjusted based on search history. To balance exploration and exploitation, an ε-greedy strategy can be used: the current optimal solution is selected with a probability of 1-ε and random exploration is performed with a probability of ε. Considering the potentially different importance of different decision variables, a weighted coding strategy can be used: different weights can be assigned to different decision variables. Furthermore, to handle constraints, an adaptive penalty function method can be used: the penalty factor can be dynamically adjusted based on the feasibility of the population. This hybrid metaheuristic algorithm not only effectively balances global search and local optimization but is also adaptable to complex multi-objective optimization problems.

[0225] Finally, the optimal logistics strategy for the target area is generated based on the optimal hyperpath combination. This process first requires analyzing the optimal hyperpath combination and extracting specific route information for each order. For each route, a detailed list of all nodes traversed, the transportation method used, and the estimated time of arrival (ETA) is required. Based on this information, a detailed transportation plan is then constructed, including a timetable: a detailed shipping, transit, and arrival schedule is developed for each order. Resource allocation: Transportation vehicles, storage space, and other resources are allocated based on the route. Personnel scheduling: Operators are assigned to each node to ensure sufficient manpower support. Cost budgeting: Detailed transportation and storage costs are calculated based on the selected route. Risk assessment: Potential risk points, such as weather impacts and traffic congestion, are identified and corresponding contingency plans are developed. To improve the operational effectiveness of the strategy, a hierarchical strategy structure can be designed: a top-level overall network strategy, a middle-level regional strategy, and a bottom-level node-specific operational guidelines. To account for potential deviations in actual implementation, a flexible strategy is designed: key decision points and alternative plans are defined to allow for minor adjustments during execution.

[0226] The present invention also discloses a high-speed, light-load railway freight logistics system based on the Internet of Things and big data, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements the high-speed, light-load railway freight logistics method based on the Internet of Things and big data as described in any one of the above embodiments.

[0227] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0228] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD) or flash memory card (FC) equipped on the computer device. In addition, the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output. This application does not impose any restrictions on this.

[0229] The present invention also discloses a computer-readable storage medium, which stores instructions. When the instructions are executed by a processor, the processor is configured to execute the high-speed light-load railway freight logistics method based on the Internet of Things and big data described in any one of the above embodiments.

[0230] Among them, the computer program can be stored in a machine-readable medium, the computer program includes computer program code, the computer program code can be in the form of source code, object code, executable file or certain middleware, etc. The machine-readable medium includes any entity or device that can carry computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the machine-readable medium includes but is not limited to the above-mentioned components.

[0231] Among them, the power transmission line comprehensive fault detection method in the above embodiment is stored in the computer-readable storage medium through the computer-readable storage medium, and is loaded and executed on the processor to facilitate the storage and application of the above method.

[0232] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as above, which are not provided in detail for the sake of simplicity.

[0233] The one or more embodiments of this application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application should be included in the scope of protection of this application.

Claims

1. A high-speed light-load railway freight logistics method based on the Internet of Things and big data, characterized in that: The steps include: Obtain historical logistics data and logistics park coordinates of all agricultural logistics parks in the target area; The agricultural logistics park is used as a logistics node, and node attributes are assigned to the logistics node in combination with the historical logistics data and the logistics park coordinates. A regional logistics topology map of the target area is constructed based on the train track topology of the target area and in combination with all the logistics nodes; Calculating the node association strength between the logistics nodes in the regional logistics topology graph based on the node attributes; Constructing a hypergraph structure in the regional logistics topology graph by combining the node association strength and the logistics node edge, and converting the regional logistics topology graph into a regional logistics hypergraph; Obtaining a list of logistics orders and a list of available trains within a preset time period in the future for the target area, and identifying logistics order information for each logistics order in the logistics order list, wherein the logistics order information includes a logistics starting point, a logistics destination, a logistics throughput, and a logistics freight type; Constructing path constraints based on the logistics order information and the available train list, and generating multiple logistics super paths for each logistics order in the regional logistics hypergraph based on the path constraints, wherein the sum of the logistics throughputs of all the logistics nodes in each logistics super path is 0; Calculating the node heat of each logistics node according to all the logistics super paths; A node load balancing constraint condition is constructed based on the node heat. Based on the node load balancing constraint condition and with the shortest total path as the optimization goal, the optimal super path combination is selected from all the logistics super paths to generate the optimal logistics strategy for the target area.

2. The high-speed light-load railway freight logistics method based on the Internet of Things and big data according to claim 1 is characterized in that: The agricultural logistics park is used as a logistics node, node attributes are assigned to the logistics node in combination with the historical logistics data and the logistics park coordinates, and a regional logistics topology map of the target area is constructed based on the train track topology of the target area and in combination with all the logistics nodes, including the following steps: Using the agricultural logistics park as a logistics node; For each of the logistics nodes, assign node coordinate attributes to the logistics node according to the logistics park coordinates; Extracting the historical logistics freight types of the logistics node and the historical average throughput of each different historical logistics freight type from the historical logistics data using natural language technology; Constructing a historical logistics matrix corresponding to the logistics node in combination with the historical logistics freight types and the historical average throughput; Assigning the historical logistics matrix as the node logistics attribute of the logistics node; If there are M freight train tracks in the train track topology of the target area between any two of the agricultural logistics parks, M logistics node edges are generated between the corresponding two logistics nodes to obtain a regional logistics topology map of the target area.

3. The high-speed light-load railway freight logistics method based on the Internet of Things and big data according to claim 2 is characterized in that: The step of calculating the node association strength between the logistics nodes in the regional logistics topology graph based on the node attributes comprises the following steps: For any two logistics nodes, the transportation distance and average transportation time between the two logistics nodes are calculated according to the node coordinate attributes; Based on the node logistics attributes, the logistics matrix similarity and throughput similarity between the two logistics nodes are calculated respectively; The node association strength between the two logistics nodes is calculated by combining the transportation distance, the average transportation time, the logistics matrix similarity and the throughput similarity.

4. The high-speed light-load railway freight logistics method based on the Internet of Things and big data according to claim 3 is characterized in that: The step of constructing a hypergraph structure in the regional logistics topology graph by combining the node association strength and the logistics node edge, and converting the regional logistics topology graph into a regional logistics hypergraph comprises the following steps: In the regional logistics topology graph, for a logistics node subset composed of any multiple logistics nodes, if the node association strength between any two logistics nodes in the logistics node subset is greater than or equal to a preset association strength threshold, or the number of node edges of the logistics node edges between any two logistics nodes in the logistics node subset is greater than or equal to a preset edge number threshold, then a logistics hyperedge is generated for the logistics node subset; For any of the logistics hyperedges, the hyperedge weight of the logistics hyperedge is calculated by combining the sum of the node association strengths of all the logistics nodes in the logistics node subset corresponding to the logistics hyperedge and the sum of the number of node edges; Determining a hyperedge lifecycle of the logistics hyperedge based on a time decay function; The hyperedge importance of the logistics hyperedge is calculated by combining the hyperedge weight of the logistics hyperedge, the hyperedge lifecycle, and the number of the logistics nodes in the logistics node subset corresponding to the logistics hyperedge; Sorting all the logistics hyperedges in descending order of their importance, and performing hyperedge pruning on the sorted logistics hyperedges according to a preset last-place screening ratio; Constructing a hyperedge tensor representation for all the retained logistics hyperedges to obtain a hypergraph structure corresponding to each logistics hyperedge; All the hypergraph structures are mapped into the low-dimensional regional logistics topology map using hypergraph embedding technology to form a regional logistics hypergraph.

5. The high-speed light-load railway freight logistics method based on the Internet of Things and big data according to claim 3 is characterized in that: The calculation formula of the node association strength is as follows: Where: D ij represents the node association strength between logistics node i and logistics node j, K ij represents the logistics matrix similarity between logistics node i and logistics node j, Q ij represents the throughput similarity between logistics node i and logistics node j, represents the weight of the mth freight train track between logistics node i and logistics node j, represents the transportation distance corresponding to the mth freight train track between logistics node i and logistics node j, It represents the average transportation time corresponding to the mth freight train track between logistics node i and logistics node j.

6. The high-speed light-load railway freight logistics method based on the Internet of Things and big data according to claim 4 is characterized in that: The step of constructing path constraints based on the logistics order information and the available train list, and generating multiple logistics hyperpaths for each logistics order in the regional logistics hypergraph based on the path constraints comprises the following steps: Path constraints are constructed based on the logistics order information and the available train list. The path constraints are as follows: Where: G is a binary decision variable indicating whether logistics order p is transported on freight train track q using a freight train of train type d. p represents the logistics throughput of logistics order p, C d N represents the upper limit of the loading capacity of a freight train of train type d. d represents the number of available trains of freight train type d, K represents the total number of train types of freight trains, A binary decision variable indicating whether logistics order p1 is transported on freight train track q using a freight train of train type d, A binary decision variable indicating whether logistics order p2 is transported on freight train track q using a freight train of train type d, Indicates the types of logistics freight included in logistics order p1. Indicates the types of logistics freight included in logistics order p2. Represents the goods compatibility matrix between logistics order p1 and logistics order p2; Based on the path constraints and the hyperedge weights, a K shortest path algorithm is used in the regional logistics hypergraph to generate multiple logistics hyperpaths for each logistics order; Counting the total number of super paths of all the logistics super paths; if the total number of super paths exceeds a preset path quantity threshold, respectively calculating the sum of the logistics throughputs of all the logistics nodes in each of the logistics super paths, and removing the logistics super paths whose sum of the logistics throughputs is not equal to 0; If the total number of hyperpaths is less than or equal to the path number threshold, the hyperedge weight is adjusted according to a preset random perturbation range, and the K shortest path algorithm is used to generate multiple new logistics hyperpaths for each logistics order until the total number of hyperpaths exceeds the path number threshold; Repeat the above-mentioned path removal step and new path generation step until the total number of super paths exceeds the path quantity threshold and the sum of the logistics throughput corresponding to any of the logistics super paths is equal to 0.

7. The high-speed light-load railway freight logistics method based on the Internet of Things and big data according to claim 6 is characterized in that: The step of calculating the node heat of each logistics node according to all the logistics super paths includes the following steps: For each of the logistics nodes, the node capacity of the logistics node is calculated according to the node logistics attributes of the logistics node; According to all the logistics super paths, the node appearance frequency of the logistics node in the logistics super path and the node expected logistics volume of the logistics node in a future preset time period are counted; The initial node heat of the logistics node is calculated by combining the node capacity, the node occurrence frequency and the expected logistics volume of the node, and the time decay function is introduced into the initial node heat to obtain the basic node heat of the logistics node; Based on the basic node heat of all the logistics nodes and according to the connection relationship between the logistics nodes in all the logistics super paths, a first-order node heat transfer matrix between all the logistics nodes is constructed; Using matrix multiplication and calculating the first-order node heat transfer matrix to obtain a high-order node heat transfer matrix between all the logistics nodes; The basic node heats of all the logistics nodes are corrected by the high-order node heat transfer matrix to obtain the node heats of all the logistics nodes.

8. The high-speed light-load railway freight logistics method based on the Internet of Things and big data according to claim 7 is characterized in that: The step of constructing a node load balancing constraint condition based on the node heat, selecting an optimal super path combination from all the logistics super paths based on the node load balancing constraint condition and taking the shortest total path as the optimization goal to generate an optimal logistics strategy for the target area includes the following steps: A node load balancing constraint condition is constructed based on the node heat, and the node load balancing constraint condition is: Where: B(v) = H(v) / E(v), B(v) represents the node load balancing factor of any logistics node v, H(v) represents the node heat of the logistics node v, E(v) represents the node capacity of the logistics node v, and ε represents the maximum load imbalance; Based on the node load balancing constraint condition and taking the shortest total path as the optimization goal, a multi-objective optimization function of the logistics strategy optimization process is constructed. The multi-objective optimization function is: Where: L(W x ) represents the total path length of all the logistics super paths in the x-th super path combination, represents the average value of the node load balancing factors of all the logistics nodes, α and β are weight coefficients, and α + β = 1; Generating a plurality of initial super-path combinations from all the logistics super-paths by random combination; Taking all the initial hyperpath combinations as the initial population, based on the multi-objective optimization function and using a hybrid metaheuristic algorithm combining a genetic algorithm and a simulated annealing algorithm, the optimal population individuals are selected from the initial population to obtain the optimal hyperpath combination; An optimal logistics strategy for the target area is generated according to the optimal super path combination.

9. A high-speed light-load railway freight logistics system based on the Internet of Things and big data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the high-speed light-load railway freight logistics method based on the Internet of Things and big data as described in any one of claims 1 to 8.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by a processor, the processor is configured to execute the high-speed light-load railway freight logistics method based on the Internet of Things and big data according to any one of claims 1 to 8.

Images