Cross-border freight dispatching method based on large model and knowledge graph and related device

Abstract

This application relates to a cross-border freight scheduling method and related apparatus based on a large model and knowledge graph. The method includes: receiving a freight scheduling request; inputting the freight scheduling request into a large model trained on cross-border freight samples, whereby the large model performs feature learning on the cross-border freight features in the freight scheduling request to obtain a path search request containing path constraints; inputting the path search request into a knowledge graph oriented towards the cross-border freight domain, whereby the knowledge graph performs multi-hop path search based on the path constraints to obtain a set of candidate intermodal transport solutions; and inputting the set of candidate intermodal transport solutions into the large model, whereby the large model performs reasoning and evaluation on the set of candidate intermodal transport solutions based on the path constraints to obtain the optimal intermodal transport solution. This method can improve the applicability of freight scheduling results and enhance the efficiency of intermodal transport collaboration.

Description

Technical Field

[0001] This application relates to the field of cross-border freight scheduling, and in particular to a cross-border freight scheduling method and related apparatus based on a large model and knowledge graph. Background Technology

[0002] In the field of cross-border freight scheduling, the evaluation of the cross-border freight scheduling process is crucial for achieving efficient coordination and stable execution of intermodal transport. Current methods rely on manual experience or rule-based configuration to evaluate the cross-border freight scheduling process. However, this approach struggles to adapt to dynamic scheduling needs, resulting in poor applicability of freight scheduling results and low efficiency in intermodal transport coordination. Summary of the Invention

[0003] Therefore, it is necessary to provide a cross-border freight scheduling method, apparatus, computer equipment, and computer-readable storage medium based on large models and knowledge graphs to address the above-mentioned technical problems. This is to solve the problems of poor applicability of freight scheduling results and low efficiency of intermodal transport collaboration in situations where the evaluation of the cross-border freight scheduling process is based on human experience or rule configuration.

[0004] Firstly, this application provides a cross-border freight scheduling method based on large models and knowledge graphs, including: Receive freight dispatch requests; The freight dispatch request is input into a large model trained on cross-border freight samples. The large model performs feature learning on the cross-border freight features in the freight dispatch request to obtain a path search request containing path constraints. The path search request is input into a knowledge graph for cross-border freight, and the knowledge graph performs a multi-hop path search based on the path constraints to obtain a set of candidate intermodal transport solutions. The set of candidate intermodal transport schemes is input into the large model, which then performs reasoning and evaluation on the set of candidate intermodal transport schemes based on the path constraints to obtain the optimal intermodal transport scheme.

[0005] Secondly, this application also provides a cross-border freight dispatching device based on a large model and knowledge graph, including: The input module is used to receive freight dispatch requests; The large model module is used to receive freight dispatch requests from the input module and perform feature learning on the cross-border freight features in the freight dispatch requests to obtain a path search request containing path constraints; wherein, the large model module includes a large model trained based on cross-border freight samples. The knowledge graph module is used to receive path search requests from the large model module and perform multi-hop path search based on the path constraints to obtain a set of candidate intermodal transport solutions; wherein, the knowledge graph module includes a knowledge graph oriented towards the cross-border freight field; The large model module is also used to: receive a set of candidate intermodal transport schemes from the knowledge graph module, and perform reasoning and evaluation on the set of candidate intermodal transport schemes based on the path constraints to obtain the optimal intermodal transport scheme.

[0006] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the above steps.

[0007] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the above steps.

[0008] The aforementioned cross-border freight scheduling method, apparatus, computer equipment, and computer-readable storage medium based on a large model and knowledge graph first receive freight scheduling requests, thus forming clear transportation demand information. Second, the large model performs feature learning on the cross-border freight characteristics in the freight scheduling request to obtain a path search request containing path constraints, thereby improving the correspondence between scheduling demands and path searches. Third, the knowledge graph performs multi-hop path searches based on the path constraints, thereby obtaining a set of candidate intermodal transport solutions that meet the path constraints, thus providing a candidate range that meets scheduling requirements. Finally, the large model performs reasoning and evaluation on the set of candidate intermodal transport solutions based on the path constraints, thereby further filtering to obtain the optimal intermodal transport solution. Based on this, the synergistic effect of the large model and knowledge graph in cross-border freight application scenarios improves the problem that methods based on manual experience or rule configuration are difficult to adapt to dynamic scheduling needs, thereby improving the applicability of freight scheduling results and enhancing the efficiency of intermodal transport collaboration. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a cross-border freight scheduling method based on a large model and knowledge graph in one embodiment. Figure 2This is a schematic diagram of a process for obtaining a path search request containing path constraints in one embodiment. Figure 3 This is a flowchart illustrating the process for obtaining the optimal intermodal transport plan in one embodiment; Figure 4 This is a structural block diagram of a cross-border freight dispatching device based on a large model and knowledge graph in one embodiment; Figure 5 This is an internal structure diagram of a computer device that implements a cross-border freight scheduling method based on a large model and knowledge graph in one embodiment. Figure 6 This is an internal structure diagram of a computer device that implements a cross-border freight scheduling method based on a large model and knowledge graph in one embodiment. Figure 7 This is an internal structure diagram of a computer-readable storage medium that implements a cross-border freight scheduling method based on a large model and knowledge graph in one embodiment. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0012] In one embodiment, such as Figure 1 As shown, a cross-border freight scheduling method based on a large model and knowledge graph is provided. This embodiment illustrates the method by applying it to a server. It can be understood that the method can also be applied to a terminal, or to a system including both a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps S100 to S400.

[0013] Step S100: Receive freight dispatch request.

[0014] For example, a freight dispatch request represents the overall dispatch demand information for a batch of goods to be transported during cross-border transportation; while the cross-border freight characteristics in the freight dispatch request represent the constraint information used to limit the selection of transportation routes, including transportation timeliness requirements, transportation cost requirements, transportation mode preferences, customs clearance node preferences, and transit node preferences. Among them, transportation timeliness requirements can mean completing cross-border transportation within a limited time, transportation cost requirements can mean controlling the overall transportation cost within a preset range, transportation mode preferences can mean prioritizing the use of a certain type of transportation, customs clearance node preferences can mean that customs clearance needs to be completed through a specific port node, and transit node preferences can mean that transportation needs to be completed through a specific transit node, warehousing node, or distribution node.

[0015] Step S200: Input the freight dispatch request into the large model trained based on cross-border freight samples. The large model performs feature learning on the cross-border freight features in the freight dispatch request to obtain a path search request containing path constraints.

[0016] For example, the large model represents a parameterized computational model formed through learning from a large number of cross-border freight samples. These cross-border freight samples represent scheduling data records generated during historical cross-border transportation processes, including the origin, destination, corresponding cross-border freight characteristics, and final route selection results for each historical batch of goods. Through multiple rounds of iterative training on these cross-border freight samples, the large model establishes a correlation between transportation demand and freight scheduling. Specifically, the origin, destination, and corresponding cross-border freight characteristics of each cross-border freight sample are used as input data, and the corresponding route selection results are used as the target output. By comparing multiple rounds of sample inputs and results, the internal parameters of the model are repeatedly adjusted, allowing the model to gradually approximate the mapping relationship between input features and output results, thereby forming a stable representation of the correlation between transportation demand and route selection in the parameter space.

[0017] Building upon this foundation, the large-scale model performs feature learning on the cross-border freight characteristics in freight dispatch requests. Specifically, through the established relationship between transportation demand and freight dispatch within the large-scale model, it analyzes various cross-border freight characteristics to identify which features are time constraints, which are cost constraints, which are node selection constraints, and which are transportation tool constraints. This transforms the dispatch requirements in the original request into conditional expressions with clear constraints. The conditional expressions corresponding to each feature are then organized and merged to form path constraint conditions. These path constraint conditions represent a set of conditions used to limit the scope of the transportation route search, such as limiting the overall transportation time to no more than a preset duration, limiting transportation costs to within a preset cost range, limiting the route to pass through or avoid specific nodes, and limiting the priority use of designated transportation tools.

[0018] Furthermore, the path constraints are combined with the transportation origin and destination corresponding to the freight scheduling request to generate a path search request. The path search request represents the structured input content for knowledge graph path retrieval, which not only gives the start and end range of the path search, but also gives the constraints that need to be met during the path search process.

[0019] Step S300: Input the route search request into the knowledge graph for cross-border freight, and the knowledge graph performs multi-hop route search based on route constraints to obtain a set of candidate intermodal transport solutions.

[0020] For example, using the origin and destination of the transport as the retrieval boundary in the path search request, a multi-hop path search is performed in the knowledge graph. A complete transport path consists of multiple sequentially connected path segments. Each path segment represents a connection between two adjacent logistics nodes in the knowledge graph, describing the transport process of goods moving from one logistics node to another during actual transport. Based on this, starting from the logistics node corresponding to the origin, the path is gradually expanded along the connections between logistics nodes. At each path segment selection, the information corresponding to that segment is compared with path constraints to filter out path segments that do not meet time limits, cost limits, node selection limits, or transport tool restrictions, thus retaining only the path extension directions that satisfy the constraints. Intermediate nodes that satisfy the constraints are then used as new expansion nodes, and the above expansion and filtering process continues until multiple continuous path segments connecting the origin and destination are formed, constituting a complete transport path connecting the origin and destination.

[0021] Finally, all transportation routes that meet the route constraints and connect the origin and destination are organized to obtain a set of candidate intermodal transport schemes. Each candidate intermodal transport scheme corresponds to a transportation route consisting of multiple route segments connected sequentially. It is used to represent a type of executable cross-border transportation arrangement, which includes the complete node sequence from the origin node to the destination node, the transportation connection order of each route segment between adjacent nodes, the configuration of transportation vehicles and nodes in each route segment, and the distribution of transportation time and transportation costs in each route segment.

[0022] Optionally, the knowledge graph for cross-border freight includes multiple logistics node pairs and the corresponding flow relationships for each logistics node pair. The flow relationship represents the flow attribute information between adjacent logistics nodes, including at least one of capacity status, rate information, transportation timeliness, and node processing efficiency. Each logistics node represents one of the following: a port node, a transit node, a warehousing node, or a distribution node. Specifically, capacity status represents the transportation capacity between adjacent logistics nodes on the corresponding route segment, such as the number of available transport batches, the weight of goods that can be carried, or the remaining capacity of the transport vehicle; rate information represents the cost standard corresponding to the flow of goods between adjacent logistics nodes, such as the transportation cost per unit weight, the transportation cost per unit volume, or a fixed transportation cost; transportation timeliness represents the time required for goods to complete the flow between adjacent logistics nodes, such as the time interval from departure to arrival; and node processing efficiency represents the ability of goods to complete processing operations at the node, such as the number of goods that can be processed per unit time or the time required to process a single batch.

[0023] Port nodes represent logistics nodes used to complete customs clearance processing, indicating the entry and exit links in cross-border transportation, such as customs-supervised ports; transit nodes represent logistics nodes used for route changes during transportation, indicating the connection position of goods between different transportation routes, such as corresponding transshipment nodes; warehousing nodes represent logistics nodes used for temporary storage of goods, indicating the storage location of goods during transportation, such as corresponding warehouse facilities; and distribution nodes represent logistics nodes used for delivering goods to the receiving location, indicating the final node in the transportation process, such as corresponding to the final delivery location.

[0024] Step S400: Input the set of candidate intermodal transport schemes into the large model. The large model will then perform reasoning and evaluation on the set of candidate intermodal transport schemes based on the path constraints to obtain the optimal intermodal transport scheme.

[0025] For example, for each candidate intermodal transport scheme, the large model aligns and integrates the cross-border transport arrangements and route constraints reflected in each route segment. This embeds route constraints at the constraint level, such as time constraints, cost constraints, node selection constraints, and means of transport, into the corresponding route representation, thereby forming an aligned and integrated feature that characterizes the synergistic effect of route constraints and route structure information. For example, it integrates the time connection situation and time constraints of each route segment in the transport route, integrates the cost distribution and cost constraints of each route segment, integrates the node sequence passed through each route segment and node selection constraints, and integrates the combination of transport modes and means of transport of each route segment.

[0026] Based on this, the alignment and integration features corresponding to each candidate intermodal transport scheme are paired, and the matching degree of the two candidate intermodal transport schemes at each constraint level is compared and analyzed in each pair. By comparing their differences in time connection, cost accumulation, node arrangement and consistency of transport mode, the superiority or inferiority relationship between each pair of candidate schemes is determined. Then, the above pairwise comparison results are summarized so that the performance of each candidate intermodal transport scheme in multiple comparisons can be comprehensively reflected, thereby forming the evaluation results for each candidate intermodal transport scheme.

[0027] In other words, the alignment fusion feature is used to characterize the differences in the candidate intermodal transport schemes under the corresponding path constraints. That is, although each candidate intermodal transport scheme corresponds to the same path constraints, their degree of fit is different. Therefore, by comparing the alignment fusion features in pairs, the degree of fit of different candidate intermodal transport schemes under path constraints can be distinguished.

[0028] Finally, the priority order of each candidate intermodal transport scheme is determined based on the evaluation results of all pairwise comparisons, and the candidate intermodal transport scheme set is sorted accordingly. Then, the candidate intermodal transport scheme in the optimal position in the sorting results is output as the optimal intermodal transport scheme, so as to realize the scheme selection process based on the synergistic effect of path constraints and path structure information.

[0029] In the aforementioned cross-border freight scheduling method based on a large model and knowledge graph, in step S100, a freight scheduling request is received, thereby forming clear transportation demand information; in step S200, the large model performs feature learning on the cross-border freight characteristics in the freight scheduling request to obtain a path search request containing path constraints, thereby improving the correspondence between scheduling demand and path search; in step S300, the knowledge graph performs multi-hop path search based on path constraints, thereby obtaining a set of candidate intermodal transport solutions that meet the path constraints, thus providing a candidate range that meets scheduling requirements; in step S400, the large model performs reasoning and evaluation on the set of candidate intermodal transport solutions based on path constraints, thereby further filtering to obtain the optimal intermodal transport solution; based on this, in the entire technical solution, through the synergistic effect of the large model and knowledge graph in cross-border freight application scenarios, the problem that methods based on manual experience or rule configuration are difficult to adapt to dynamic scheduling requirements is improved, thereby improving the applicability of freight scheduling results and improving the efficiency of intermodal transport collaboration.

[0030] In one exemplary embodiment, such as Figure 2 As shown, step S200 involves “learning features from the cross-border freight characteristics in the freight scheduling request using a large model to obtain a path search request containing path constraints”, which includes steps S201 to S203.

[0031] Step S201: Encode the cross-border freight features in the freight dispatch request to obtain the feature vectors corresponding to each feature dimension.

[0032] For example, various cross-border freight characteristics are divided according to preset feature dimensions, so that various constraint information such as transportation timeliness requirements, transportation cost requirements, transportation mode preferences, customs clearance node preferences, and flow node preferences are respectively assigned to the corresponding feature dimensions. Then, the values ​​in each feature dimension are encoded, which is equivalent to converting the values ​​in each feature dimension into corresponding coded components according to preset encoding rules, and arranging multiple coded components under the same feature dimension in a preset order to form a corresponding feature vector. Here, a feature vector is represented as an array structure composed of multiple coded components in sequence, used to characterize the overall description result of the constraint information under that feature dimension. For example, transportation timeliness requirements can correspond to [timeliness urgency encoding, deadline limit encoding, delay allowable range encoding, time adjustment range encoding], where timeliness urgency encoding is used to indicate the urgency of the overall transportation task, deadline limit encoding is used to indicate whether the transportation must be completed before the specified time, delay allowable range encoding is used to indicate the maximum time range of allowed delay, and time adjustment range encoding is used to indicate the time interval that can be adjusted under different route selections.

[0033] For example, transportation cost requirements can correspond to [cost ceiling code, cost fluctuation range code, cost control level code, cost adjustment range code]. The cost ceiling code indicates the maximum amount that transportation costs must not exceed, the cost fluctuation range code indicates the range of cost changes allowed under different route choices, the cost control level code indicates the strictness of cost constraints, and the cost adjustment range code indicates the acceptable range of cost differences under the premise of meeting the cost ceiling.

[0034] For example, transportation mode preference can correspond to [preferred mode code, restricted mode code, combination restricted code, and substitution allowed code]. The preferred mode code is used to indicate the preferred transportation mode or set of transportation modes, the restricted mode code is used to indicate the transportation mode that is not allowed, the combination restricted code is used to indicate whether different transportation modes are allowed to be combined, and the substitution allowed code is used to indicate whether different transportation modes are allowed to be substituted for each other.

[0035] For example, customs clearance node preferences can correspond to [priority port code, prohibited port code, and mandatory port code]. The priority port code represents the set of port nodes that are preferred to be passed through, the prohibited port code represents the set of port nodes that are not allowed to be passed through, and the mandatory port code represents the port nodes that must be passed through in the path.

[0036] For example, the preference for circulation nodes can correspond to [priority node code, restricted node code, transit number code, node adjustment code]. The priority node code is used to indicate the range of nodes that are preferred as transit nodes, warehousing nodes, or delivery nodes. The restricted node code is used to indicate the range of nodes that are not allowed to be transit nodes, warehousing nodes, or delivery nodes. The transit number code is used to indicate the maximum range of allowed transit numbers. The node adjustment code is used to indicate the range of transit nodes, warehousing nodes, or delivery nodes that can be replaced or adjusted in different path selection processes.

[0037] Step S202: Based on the feature vectors corresponding to each feature dimension, perform feature learning on the constraint semantics in the cross-border freight process to obtain path constraint parameters corresponding to the path search structure of the knowledge graph.

[0038] For example, after obtaining the feature vectors corresponding to each feature dimension, the feature vectors are input into a large model, and the values ​​of the encoded components of the feature vectors under different feature dimensions are analyzed to learn the constraint semantics in the cross-border freight process. The constraint semantics represent the transportation restriction relationships formed by the interaction between different feature dimensions. Specifically, by analyzing the combination relationships of the encoded components under different feature dimensions such as transportation timeliness requirements, transportation cost requirements, transportation mode preferences, customs clearance node preferences, and flow node preferences, the constraints of each feature dimension are associated under a unified expression. For example, when the urgency code corresponding to the transportation timeliness requirement has a high value and the deadline restriction code is a strict restriction, while the priority code in the transportation mode preference is limited to a single transportation mode and the alternative allowance code has a low value, it can be determined that the route needs to be completed through a fixed transportation mode in a short time, thus forming a joint restriction on transportation time and transportation mode.

[0039] For example, when the upper limit code for transportation cost requirements is low and the cost control level code is high, while the number of transfers code for flow-type node preferences is low, it can be determined that the route needs to be completed with fewer transfers and the overall cost is limited, thus forming a joint restriction on the route length and transfer process. For another example, when the mandatory port code for customs clearance node preferences restricts specific ports and the prohibited port code excludes some nodes, while the priority node code for flow-type node preferences restricts specific transfer nodes, it can be determined that the route needs to meet the combined restriction of port nodes and flow nodes in node selection.

[0040] Based on this, according to the transportation constraints formed by the interactions between the different feature dimensions, the constraint semantics are mapped to the path search structure in the knowledge graph. The path search structure represents a conventional path-finding method in the knowledge graph, starting from the transportation origin and gradually expanding to the transportation destination. It determines the extension order, connection method, and node selection range of the path through the connection relationships between nodes. Specifically, by mapping the constraint semantics corresponding to the combined constraints of each feature dimension to the corresponding positions in the path search structure, the constraints can act on the node selection, path connection, and path extension process of the path, thereby generating path constraint parameters. These path constraint parameters represent the constraint methods for path extension in parameter form, such as combined constraint parameters used to limit the total path duration range, path cost range, port node selection, and transfer nodes and transfer times.

[0041] For example, when the urgency code corresponding to the transportation timeliness requirement is high and the deadline limit code is strict, while the priority code in the transportation mode preference is limited to a single transportation mode and the alternative allowable code is low, it can be determined that the route needs to be completed in a short time using a fixed transportation mode. This combination of constraints is then mapped to the path extension position in the path search structure, so that the path can only select path segments that meet the time limit and have a fixed transportation mode, thereby obtaining the path constraint parameters related to transportation time and transportation mode.

[0042] For example, when the upper limit of cost in transportation cost requirements is low and the cost control level is high, and the number of transfers in flow-type node preferences is low, this combination of constraints is mapped to the path connection position in the path search structure. This makes the path control the number of transfers and limit the overall cost during the connection process, thereby obtaining the path constraint parameters related to transportation cost and number of transfers.

[0043] For example, when the required port code in the preference of customs clearance nodes restricts specific ports and prohibits the exclusion of some nodes by port codes, while the priority node code in the preference of transit nodes restricts specific transit nodes, this combination of restrictions is mapped to the node selection position in the path search structure. This ensures that the path must simultaneously meet the selection conditions of both port nodes and transit nodes during the node selection process, thereby obtaining the path constraint parameters related to the selection of customs clearance nodes and transit nodes.

[0044] Step S203: Based on the path constraint parameters, perform constraint injection on the path search structure in the knowledge graph to obtain the path constraint conditions and construct a path search request containing the path constraint conditions.

[0045] For example, constraining the path search structure based on path constraint parameters is equivalent to embedding the path constraint parameters corresponding to different combinations of constraints into different positions within the path search structure, giving each combination of constraints a clear role in the path search process. For instance, path constraint parameters related to transportation time and mode of transport are injected into the path extension position, ensuring that the path can only select path segments that meet the time limit and mode of transport requirements when expanding gradually. Path constraint parameters related to transportation cost and number of transfers are injected into the path connection position, ensuring that the path must simultaneously meet the cost range and number of transfers restrictions when connecting different path segments. Path constraint parameters related to the selection of customs clearance nodes and circulation nodes are injected into the node selection position, ensuring that the path can only select nodes that meet the port passage conditions and circulation node restrictions during the node selection process.

[0046] Through the above methods, the path search structure is constrained and controlled at three levels: path extension, path connection, and node selection, thus forming path constraint conditions. Furthermore, these path constraint conditions are integrated with the transportation origin and destination to construct a path search request, which serves as the input for subsequent path searches. Therefore, step S202 processes the feature vectors to map the constraint semantics to the path search structure, obtaining path constraint parameters to determine the position and expression of each constraint within the path search structure. Step S203 then applies these path constraint parameters to the path search structure, ensuring the constraints actually take effect during path extension, path connection, and node selection, thereby restricting the path search process. Thus, step S202 focuses on the structural correspondence of constraints, while step S203 focuses on the actual effect of the constraints.

[0047] In this embodiment, in step S201, feature vectors are formed by feature encoding of cross-border freight characteristics, thereby achieving a unified expression of various constraints; in step S202, feature learning is performed on the constraint semantics based on each feature vector, and corresponding processing is performed with the path search structure to generate path constraint parameters, thereby determining the position and expression of constraints in the path search structure; in step S203, constraints are injected into the path search structure based on the path constraint parameters, thereby forming a constraint-controlled path search expression; based on this, in the entire technical solution, continuous processing from constraint expression to structure correspondence to path action is achieved, making the path search process consistent with scheduling constraints.

[0048] In an exemplary embodiment, step S202, "based on the feature vectors corresponding to each feature dimension, performs feature learning on the constraint semantics in the cross-border freight process to obtain path constraint parameters corresponding to the path search structure of the knowledge graph," includes steps S2021 to S2023.

[0049] Step S2021: Determine the weight coefficients corresponding to each feature dimension based on the differences in cross-border freight characteristics represented by the feature vectors corresponding to each feature dimension.

[0050] For example, by analyzing the values ​​of each coded component in the feature vectors corresponding to feature dimensions such as transportation timeliness requirements, transportation cost requirements, transportation mode preferences, customs clearance node preferences, and flow node preferences, the differences in cross-border freight characteristics reflected by each feature dimension can be determined. For instance, when the urgency code for transportation timeliness requirements has a high value, the deadline restriction code is strictly limited, and the delay allowance code has a low value in the feature vector, it can be determined that the time restriction reflected by this feature dimension is strong. When the cost ceiling code for transportation cost requirements has a low value and the cost control code has a high value in the feature vector, it can be determined that the cost restriction reflected by this feature dimension is strong. When both the mandatory port code and the prohibited port code have clear values ​​in the feature vector corresponding to customs clearance node preferences, it can be determined that this feature dimension has a clear restrictive effect on the selection of port nodes.

[0051] Then, based on the constraint strength, constraint boundary, and degree of restriction clarity reflected by the feature vectors corresponding to each individual feature dimension, the degree of influence of each feature dimension in the overall constraint expression is quantified, thereby determining the weight coefficients corresponding to each feature dimension. The constraint strength indicates the degree to which the feature dimension restricts path selection, which can be judged by the values ​​of the relevant coded components in the corresponding feature vector. For example, the tighter the time limit and the lower the cost limit, the higher the constraint strength. The constraint boundary indicates the degree to which the feature dimension limits the range of path options, which is reflected in the width of the allowed range. For example, the smaller the time adjustment interval and the smaller the cost fluctuation range, the more convergent the constraint boundary. The degree of restriction clarity indicates whether the feature dimension provides clear selection or exclusion conditions, which is reflected in whether there are clear constraints such as mandatory nodes, prohibited nodes, or restrictions on transportation methods.

[0052] Based on the above three aspects, the degree of influence of each feature dimension in the overall constraint expression is comprehensively evaluated. When a feature dimension has high constraint strength, narrow constraint boundary and clear restriction conditions, its corresponding weight coefficient is set to a high value; when only some indicators are strong, its weight coefficient is set to a medium value; when the constraint strength is weak, the boundary is wide and the restriction is unclear, its weight coefficient is set to a low value.

[0053] Step S2022: Based on the weight coefficients corresponding to each feature dimension, perform feature weighting and fusion on the feature vectors corresponding to each feature dimension in each feature dimension combination to obtain the feature fusion result.

[0054] For example, for any combination of feature dimensions, the feature vectors corresponding to the feature dimensions participating in the combination are processed according to their weight coefficients. This ensures that feature dimensions with strong constraints, convergent boundaries, and clear limitations occupy a larger proportion in the fusion process, while feature dimensions with weaker constraints occupy a smaller proportion, thus forming a feature fusion result corresponding to that combination of feature dimensions. Under each combination of feature dimensions, the feature vectors of each feature dimension are integrated in a unified expression structure to obtain the feature fusion results corresponding to different combinations of feature dimensions. Here, the feature fusion result represents the overall constraint expression after the combined effect of multiple types of constraint information in the corresponding combination of feature dimensions.

[0055] For example, in the combination of feature dimensions for transportation timeliness requirements and transportation mode preferences, the feature vectors are weighted and fused according to their corresponding weight coefficients. When the weight coefficient corresponding to transportation timeliness requirements is high, the limitation on the time range dominates in the feature fusion result, while the limitation on the range of transportation modes is superimposed, thus reflecting the combined constraint that the path must first meet the time constraint and take into account the transportation mode constraint. As another example, in the combination of feature dimensions for transportation cost requirements and preferences for transit nodes, the feature fusion is weighted and fused according to their corresponding weight coefficients. When the weight coefficients of transportation cost requirements and preferences for transit nodes are similar, the limitations on the cost range and the range of transit times in the feature fusion result work together, thus reflecting the combined constraint that the path must simultaneously meet the cost constraint and the transit constraint.

[0056] Step S2023: Based on the feature fusion results, perform parameter mapping between the constraint semantics in the cross-border freight process and the path search structure in the knowledge graph, so that the constraint semantics correspond to the path search structure, and obtain the path constraint parameters.

[0057] For example, in the combined constraints reflected by the fusion results of each feature, the constraint semantics corresponding to the combined constraints are mapped to the corresponding positions in the path search structure, so that the constraints can act on the node selection, path connection, and path extension processes of the path, thereby generating path constraint parameters. Optionally, the path extension position, path connection position, and node selection position in the path search structure do not form a one-to-one correspondence with the feature dimension combination, but are determined according to the objects on which the combined constraints formed by each feature dimension combination are applied; specifically, when the combined constraints formed by a certain feature dimension combination act on the path segment selection or path duration control in the path extension process, the feature dimension combination corresponds to the path extension position; when its combined constraints act on the connection relationship or transfer process between path segments, it corresponds to the path connection position; when its combined constraints act on the node filtering or node selection range in the path, it corresponds to the node selection position; therefore, the same position can correspond to multiple feature dimension combinations, as long as the combination restricts the same type of object, it is classified into the corresponding position, thereby realizing the unified mapping of different combined constraints in the path search structure.

[0058] In this embodiment, in step S2021, the difference analysis of the feature vectors corresponding to each feature dimension is performed and the weight coefficients are determined to distinguish the degree of influence of different feature dimensions; in step S2022, based on the weight coefficients corresponding to each feature dimension, the feature vectors corresponding to each feature dimension are subjected to feature weighted fusion in each feature dimension combination to obtain the feature fusion result reflecting the combination constraints; in step S2023, the constraint semantics are mapped to the path search structure according to the feature fusion result to generate path constraint parameters corresponding to the path search structure; based on this, the entire technical solution realizes continuous processing from feature difference identification, combination constraint formation to path constraint parameter generation, so that the constraint information in the cross-border freight application scenario can be more targeted to the path search process.

[0059] In one exemplary embodiment, such as Figure 3 As shown, step S400, "the large model performs reasoning and evaluation on the candidate intermodal transport scheme set based on path constraints to obtain the optimal intermodal transport scheme", includes steps S401 to S402.

[0060] Step S401: Extract features from each candidate intermodal transport scheme in the candidate intermodal transport scheme set, and align and fuse the extracted features with the path constraints to obtain the aligned and fused features corresponding to each candidate intermodal transport scheme.

[0061] For example, for each candidate intermodal transport scheme, the large model aligns and integrates the cross-border transport arrangements and route constraints reflected in each route segment, embedding constraints such as time constraints, cost constraints, node selection constraints, and means of transport constraints into the corresponding route expression, thereby forming an aligned and integrated feature that characterizes the synergistic effect of route constraints and route structure information.

[0062] In essence, when processing each candidate intermodal transport scheme, it first expands the scheme according to the order of its route segments, so that each route segment corresponds to its origin and destination nodes, origin and destination times, and corresponding costs, forming a continuous sequence of route segments. Based on this, since all candidate intermodal transport schemes satisfy the route constraints, the matching determination is no longer performed during processing; instead, a detailed analysis of the degree of matching is conducted. Specifically, the route constraints are applied item by item to the sequence of route segments. For time constraints, the origin and destination times of each route segment are compared segment by segment, and adjacent segments are compared... The time connectivity between route segments is analyzed to characterize the matching degree of the routes within a time range. For cost constraints, the costs of each route segment are accumulated and compared with the cost range to characterize the matching degree of route costs within the defined interval. For node selection constraints, each node traversed by a route segment is matched one by one to characterize the matching degree of route nodes with the allowed set of nodes. For transportation mode constraints, the transportation modes corresponding to each route segment are compared segment by segment to characterize the matching degree of transportation mode combinations with the constraints. After completing the above processing, the matching degree of each route segment under different constraints is unified and integrated to form a corresponding alignment and fusion feature, which is used to characterize the difference in the matching degree of the candidate intermodal transport scheme under the premise of satisfying the route constraints.

[0063] Optionally, the matching degree is obtained by quantifying the difference between the actual values ​​in the candidate intermodal transport scheme and the constraint range in the path constraints. For example, by calculating the deviation ratio between the path time and the time limit, the deviation ratio between the path cost and the cost range, the compliance ratio of whether the node completely belongs to the allowed set, and the compliance ratio of whether the transport mode meets the limiting requirements. The above results are uniformly converted into numerical form so that the matching degree can reflect the degree of closeness of the candidate intermodal transport scheme to each path constraint in terms of numerical magnitude.

[0064] Step S402: Perform reasoning and evaluation on the alignment and fusion features corresponding to each candidate intermodal transport scheme to obtain the evaluation results corresponding to each candidate intermodal transport scheme, and sort them according to the evaluation results corresponding to each candidate intermodal transport scheme to obtain the optimal intermodal transport scheme.

[0065] For example, the alignment and fusion features corresponding to each candidate intermodal transport scheme are combined in pairs, and the matching degree of the two candidate intermodal transport schemes at each constraint level is compared and analyzed in each pair. That is, by comparing their differences in time connection, cost accumulation, node arrangement and consistency of transport mode, the superiority and inferiority relationship between each pair of candidate schemes is determined, thereby obtaining the optimal intermodal transport scheme.

[0066] In essence, the alignment and fusion features are composed of matching components corresponding to multiple different constraint levels. Therefore, when comparing any two candidate intermodal transport schemes, the matching components in the alignment and fusion features corresponding to both are first extracted, and the matching components at the same constraint level are compared accordingly. Specifically, for the time matching component, the magnitude of the deviation between the two within the time range is compared to determine the quality of their time matching; for the cost matching component, the magnitude of the cost deviation between the two is compared to determine the quality of their cost control; for the node matching component, the quality of their node arrangement is determined by comparing the satisfaction ratio or deviation between the two in node selection; and for the transport mode matching component, the quality of their transport mode combination is determined by comparing the difference in the degree of compliance between the two.

[0067] After comparing each matching component, all candidate intermodal transport schemes are compared in pairs, and the number of times each candidate intermodal transport scheme performs well or poorly in all combinations is counted to form the corresponding evaluation results. Then, the candidate intermodal transport schemes are sorted according to the evaluation results, and the candidate intermodal transport scheme in the best position in the sorting results is selected as the optimal intermodal transport scheme output.

[0068] For example, when comparing any two candidate intermodal transport schemes pairwise, if one candidate intermodal transport scheme has a better match degree than the other candidate intermodal transport scheme in at least one of the time matching component, cost matching component, node matching component, or mode of transport matching component, and is not worse than the other in the remaining matching components, then the candidate intermodal transport scheme is recorded as being better than the other once; if one candidate intermodal transport scheme has a better match degree than the other candidate intermodal transport scheme in most matching components, then the candidate intermodal transport scheme is also recorded as being better than the other once; if the two candidate intermodal transport schemes have different advantages and disadvantages in each matching component, then their superiority or inferiority relationship is determined according to the degree of difference in each matching component, and the one with the smaller difference is recorded as being better than the other once; if the two candidate intermodal transport schemes have the same degree of match in each matching component, then the number of superiority or inferiority performances is not counted; after completing the pairwise comparisons between all candidate intermodal transport schemes, the number of superiority or inferiority performances of each candidate intermodal transport scheme in all comparison processes is accumulated to serve as its corresponding evaluation result.

[0069] In this embodiment, in step S401, feature extraction of candidate intermodal transport schemes is performed and aligned and fused with path constraints to achieve a unified representation of the matching degree of each candidate intermodal transport scheme under path constraints. In step S402, reasoning and evaluation are performed based on the aligned and fused features corresponding to each candidate intermodal transport scheme to determine the superiority or inferiority relationship between different candidate intermodal transport schemes and form corresponding evaluation results. Based on this, the entire technical solution realizes continuous processing from the representation of candidate intermodal transport schemes to the determination of superiority or inferiority relationships, enabling different candidate intermodal transport schemes to be effectively distinguished and ranked under the same path constraints.

[0070] In an exemplary embodiment, step S402, which involves “inferring and evaluating the alignment and fusion features corresponding to each candidate intermodal transport scheme to obtain the evaluation results corresponding to each candidate intermodal transport scheme”, includes step S4021.

[0071] Step S4021: Combine the alignment and fusion features corresponding to each candidate intermodal transport scheme in pairs, and perform difference inference calculation on the combined alignment and fusion features to obtain the relative advantages and disadvantages between each candidate intermodal transport scheme, and determine the evaluation result corresponding to each candidate intermodal transport scheme based on the relative advantages and disadvantages.

[0072] For example, when comparing any two candidate intermodal transport schemes, the time matching component, cost matching component, node matching component, and transport mode matching component are first extracted from the alignment and fusion features of the two schemes respectively, and the matching components at the same constraint level are compared. On this basis, the correspondence between the two candidate intermodal transport schemes and different elements is compared at the combination level of intermodal transport routes and intermodal transport nodes, intermodal transport routes and intermodal transport modes, and intermodal transport nodes and intermodal transport modes, so that the correspondence between route structure and node arrangement, route structure and transport mode, and node arrangement and transport mode are reflected.

[0073] Specifically, in the process of calculating differences at the combination level, for the combination of intermodal routes and intermodal nodes, the node sequences corresponding to the route segments in the two candidate intermodal schemes are aligned and compared segment by segment to determine the degree of matching between the node arrangement and the route extension order, thereby determining the differences between the route structure and the node configuration; for the combination of intermodal routes and intermodal modes, the order of the route segments and the distribution of the corresponding modes of transport are compared segment by segment to determine the degree of matching between the modes of transport in the route structure, thereby determining the differences between the route structure and the modes of transport; for the combination of intermodal nodes and intermodal modes, the correlation between the node locations and the corresponding modes of transport is compared to determine the degree of matching between the mode of transport selection at the nodes, thereby determining the differences between the node arrangement and the modes of transport.

[0074] For example, based on the differences between the above-mentioned constraint levels and combination levels, the overall superiority or inferiority relationship between the two candidate intermodal transport schemes is determined, and the determination result is used as the comparison result under the combination. After completing the pairwise combination and difference reasoning calculation between all candidate intermodal transport schemes, the comparison results of each candidate intermodal transport scheme in all combinations are summarized, and the corresponding evaluation result is determined based on its superiority or inferiority performance in each combination, thereby realizing the reasoning evaluation process based on alignment and fusion features.

[0075] Specifically, in determining the overall superiority / inferiority of the intermodal transport schemes, the differences at the constraint level and the combination level are first evaluated separately. At the constraint level, the results of comparing time matching, cost matching, node matching, and mode of transport matching components are analyzed to determine the superiority / inferiority of the two candidate intermodal transport schemes under each individual constraint condition. At the combination level, the results of comparing the combinations of intermodal transport routes and nodes, intermodal transport routes and modes, and intermodal transport nodes and modes are analyzed to determine the superiority / inferiority of the two candidate intermodal transport schemes under different element synergies. Based on this, the evaluation results at the constraint level and the evaluation results at the combination level are integrated. That is, when one candidate intermodal transport scheme is superior to the other at both the constraint level and the combination level, it is determined to be superior overall. When it has different advantages at the constraint level and the combination level respectively, the overall superiority / inferiority is determined based on the differences between the two types of evaluation results. When it has no significant advantage in either type of evaluation result, the superiority / inferiority is determined to be consistent, thus obtaining the overall superiority / inferiority relationship between the two candidate intermodal transport schemes.

[0076] For example, when comparing any two candidate intermodal transport schemes pairwise, if candidate A is superior to candidate B in 3 out of 4 matching components at the constraint level and not inferior in the remaining 1, and is also superior to B in all 3 comparisons at the combination level, then A has 3 advantages at the constraint level and 3 advantages at the combination level, for a total of 6 advantages, while B has no advantages. Therefore, A is judged to be superior to B overall. If candidate A has 3 advantages over B and 1 inferior to B at the constraint level, and 1 advantage over B and 2 inferior to B at the combination level, then A has 4 advantages, and B has 3 advantages. Therefore, A is judged to be superior to B overall. If candidate A and B each have 2 advantages over each other at the constraint level and 1 advantage over each other at the combination level, then the number of advantages they have is the same, and their relative superiority / inferiority relationship is considered to be consistent.

[0077] In essence, both the constraint-level and combination-level analyses are based on the premise that the candidate intermodal transport schemes have met the route constraints. The constraint-level analysis focuses on comparing the satisfaction of individual conditions such as time, cost, nodes, and modes of transport, while the combination-level analysis focuses on comparing the rationality of the route structure and node arrangement, the route structure and modes of transport, and the synergistic relationships between nodes and modes of transport. Therefore, they focus on different aspects. In actual evaluation, a candidate intermodal transport scheme may perform better at the individual constraint level but weaker at the combination level. This does not constitute a conflict but rather reflects differences under different evaluation perspectives. By integrating the results of the two types of comparisons, a more comprehensive judgment can be obtained by balancing the satisfaction of individual constraints with the synergistic relationships of multiple factors.

[0078] In this embodiment, in step S4021, the alignment and fusion features corresponding to the candidate intermodal transport schemes are combined in pairs, and the difference inference calculation is performed on the combined alignment and fusion features to identify the degree of matching deviation of different candidate intermodal transport schemes at each combination level. Based on this, in the whole technical solution, the processing flow from combination level comparison to determination of superiority and inferiority is realized, so that the differences between candidate intermodal transport schemes can be distinguished under a unified standard and used for ranking determination.

[0079] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0080] Based on the same inventive concept, this application also provides a cross-border freight scheduling device based on a large model and knowledge graph for implementing the cross-border freight scheduling method based on a large model and knowledge graph as described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the cross-border freight scheduling device based on a large model and knowledge graph provided below can be found in the limitations of the cross-border freight scheduling method based on a large model and knowledge graph described above, and will not be repeated here.

[0081] In one exemplary embodiment, such as Figure 4As shown, a cross-border freight dispatching device based on a large model and knowledge graph is provided, including: an input module 100, a large model module 200, and a knowledge graph module 300, wherein: Input module 100 is used to receive freight dispatch requests; The large model module 200 is used to receive freight dispatch requests from the input module and perform feature learning on the cross-border freight features in the freight dispatch requests to obtain a path search request containing path constraints; wherein, the large model module includes a large model trained based on cross-border freight samples. The knowledge graph module 300 is used to receive path search requests from the large model module and perform multi-hop path search based on path constraints to obtain a set of candidate intermodal transport solutions; the knowledge graph module includes a knowledge graph for the cross-border freight field. The large model module 200 is also used to: receive a set of candidate intermodal transport schemes from the knowledge graph module, and perform reasoning and evaluation on the set of candidate intermodal transport schemes based on path constraints to obtain the optimal intermodal transport scheme.

[0082] In an exemplary embodiment, the large model module 200 is further configured to: encode the cross-border freight features in the freight scheduling request to obtain feature vectors corresponding to each feature dimension; perform feature learning on the constraint semantics in the cross-border freight process based on the feature vectors corresponding to each feature dimension to obtain path constraint parameters corresponding to the path search structure of the knowledge graph; and inject constraints into the path search structure in the knowledge graph based on the path constraint parameters to obtain path constraint conditions and construct a path search request containing path constraint conditions.

[0083] In an exemplary embodiment, the large model module 200 is further configured to: determine the weight coefficients corresponding to each feature dimension based on the differences in cross-border freight characteristics represented by the feature vectors corresponding to each feature dimension; perform feature weighted fusion on the feature vectors corresponding to each feature dimension in each feature dimension combination based on the weight coefficients corresponding to each feature dimension to obtain the feature fusion result; and perform parameter mapping between the constraint semantics in the cross-border freight process and the path search structure in the knowledge graph based on the feature fusion result, so that the constraint semantics correspond to the path search structure to obtain the path constraint parameters.

[0084] In an exemplary embodiment, the knowledge graph for the cross-border freight field includes multiple logistics node pairs and the flow relationship corresponding to each logistics node pair; wherein, the flow relationship represents the flow attribute information between adjacent logistics nodes, which includes at least one of capacity status, rate information, transportation timeliness and node processing efficiency; each logistics node represents one of port node, transit node, warehousing node or distribution node.

[0085] In an exemplary embodiment, the large model module 200 is further configured to: extract features from each candidate intermodal transport scheme in the candidate intermodal transport scheme set, and align and fuse the extracted features with the path constraints to obtain the aligned and fused features corresponding to each candidate intermodal transport scheme; perform reasoning and evaluation on the aligned and fused features corresponding to each candidate intermodal transport scheme to obtain the evaluation results corresponding to each candidate intermodal transport scheme, and sort and determine the optimal intermodal transport scheme based on the evaluation results corresponding to each candidate intermodal transport scheme.

[0086] In an exemplary embodiment, the large model module 200 is further configured to: combine the alignment and fusion features corresponding to each candidate intermodal transport scheme in pairs, perform difference inference calculation on the combined alignment and fusion features to obtain the relative superiority-inferiority relationship between each candidate intermodal transport scheme, and determine the evaluation result corresponding to each candidate intermodal transport scheme based on the relative superiority-inferiority relationship.

[0087] In an exemplary embodiment, the process of combining the alignment and fusion features corresponding to each candidate intermodal transport scheme in pairs involves the following combination levels: combination of intermodal transport path and intermodal transport node, combination of intermodal transport path and intermodal transport mode, and combination of intermodal transport node and intermodal transport mode.

[0088] The modules in the aforementioned cross-border freight dispatching device based on large models and knowledge graphs can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0089] In one exemplary embodiment, a computer device is provided, the computer device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps in the above-described method embodiments.

[0090] This computer device can be a server, and its internal structure diagram can be as follows: Figure 5As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores data related to cross-border freight scheduling. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network.

[0091] The computer device can be a terminal, and its internal structure diagram can be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the steps in the aforementioned cross-border freight scheduling method based on large models and knowledge graphs. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen; the input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs or touchpads set on the casing of the computer device, or external keyboards, touchpads or mice, etc.

[0092] Those skilled in the art will understand that Figure 5 or Figure 6The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0093] In one exemplary embodiment, such as Figure 7 The diagram shows the internal structure of a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the above-described method embodiments.

[0094] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0095] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0096] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A cross-border freight dispatching method based on a large model and a knowledge graph, characterized in that, The method includes: Receive freight dispatch requests; The freight dispatch request is input into a large model trained on cross-border freight samples. The large model performs feature learning on the cross-border freight features in the freight dispatch request to obtain a path search request containing path constraints. The path search request is input into a knowledge graph for cross-border freight, and the knowledge graph performs a multi-hop path search based on the path constraints to obtain a set of candidate intermodal transport solutions. The set of candidate intermodal transport schemes is input into the large model, which then performs reasoning and evaluation on the set of candidate intermodal transport schemes based on the path constraints to obtain the optimal intermodal transport scheme.

2. The method according to claim 1, characterized in that, The step of performing feature learning on the cross-border freight characteristics in the freight dispatch request using the large model to obtain a path search request containing path constraints includes: The cross-border freight characteristics in the freight dispatch request are feature-encoded to obtain feature vectors corresponding to each feature dimension; Based on the feature vectors corresponding to each feature dimension, feature learning is performed on the constraint semantics in the cross-border freight process to obtain path constraint parameters corresponding to the path search structure of the knowledge graph. Based on the path constraint parameters, constraint injection is performed on the path search structure in the knowledge graph to obtain path constraint conditions and construct a path search request containing the path constraint conditions.

3. The method according to claim 2, characterized in that, The step involves learning the constraint semantics in the cross-border freight process based on the feature vectors corresponding to each feature dimension, to obtain path constraint parameters corresponding to the path search structure of the knowledge graph, including: Based on the differences in cross-border freight characteristics represented by the feature vectors corresponding to each feature dimension, determine the weight coefficients corresponding to each feature dimension. Based on the weight coefficients corresponding to each feature dimension, the feature vectors corresponding to each feature dimension are weighted and fused in each feature dimension combination to obtain the feature fusion result. Based on the feature fusion results, parameter mapping is performed between the constraint semantics in the cross-border freight process and the path search structure in the knowledge graph, so that the constraint semantics correspond to the path search structure, and path constraint parameters are obtained.

4. The method according to any one of claims 1 to 3, characterized in that, The knowledge graph for cross-border freight includes multiple logistics node pairs and the flow relationships corresponding to each logistics node pair. The flow relationship refers to the flow attribute information between adjacent logistics nodes, which includes at least one of the following: capacity status, rate information, transportation timeliness and node processing efficiency; each logistics node represents one of the following: port node, transit node, warehousing node or distribution node.

5. The method according to claim 1, characterized in that, The process of using the large model to reason and evaluate the set of candidate intermodal transport solutions based on the path constraints to obtain the optimal intermodal transport solution includes: Feature extraction is performed on each candidate intermodal transport scheme in the candidate intermodal transport scheme set, and the extracted features are aligned and fused with the path constraints to obtain the aligned and fused features corresponding to each candidate intermodal transport scheme. The alignment and fusion features corresponding to each candidate intermodal transport scheme are inferred and evaluated to obtain the evaluation results of each candidate intermodal transport scheme. The optimal intermodal transport scheme is then determined by ranking the evaluation results of each candidate intermodal transport scheme.

6. The method according to claim 5, characterized in that, The process of reasoning and evaluating the alignment and fusion features corresponding to each candidate intermodal transport scheme to obtain the evaluation results for each candidate intermodal transport scheme includes: The alignment and fusion features corresponding to each candidate intermodal transport scheme are combined in pairs, and the difference inference calculation is performed on the combined alignment and fusion features to obtain the relative advantages and disadvantages between each candidate intermodal transport scheme. The evaluation result corresponding to each candidate intermodal transport scheme is determined based on the relative advantages and disadvantages.

7. The method according to claim 6, characterized in that, The process of combining the alignment and fusion features of each candidate intermodal transport scheme in pairs involves the following combination levels: combination of intermodal transport routes and intermodal transport nodes, combination of intermodal transport routes and intermodal transport modes, and combination of intermodal transport nodes and intermodal transport modes.

8. A cross-border freight dispatching device based on a large model and knowledge graph, characterized in that, The device includes: The input module is used to receive freight dispatch requests; The large model module is used to receive freight dispatch requests from the input module and perform feature learning on the cross-border freight features in the freight dispatch requests to obtain a path search request containing path constraints; wherein, the large model module includes a large model trained based on cross-border freight samples. The knowledge graph module is used to receive path search requests from the large model module and perform multi-hop path search based on the path constraints to obtain a set of candidate intermodal transport solutions; wherein, the knowledge graph module includes a knowledge graph oriented towards the cross-border freight field; The large model module is also used to: receive a set of candidate intermodal transport schemes from the knowledge graph module, and perform reasoning and evaluation on the set of candidate intermodal transport schemes based on the path constraints to obtain the optimal intermodal transport scheme.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

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

Images