Search

CN-121981635-A - Order matching method based on knowledge graph

CN121981635ACN 121981635 ACN121981635 ACN 121981635ACN-121981635-A

Abstract

The invention relates to the technical field of big data analysis, in particular to an order matching method based on a knowledge graph, which comprises the steps of obtaining order data to be processed, carrying out multi-mode feature extraction on the order data to obtain order feature data, mapping in the knowledge graph to generate cargo entity nodes, calling vehicle model nodes in the knowledge graph, carrying out multi-mode matching verification based on the cargo entity nodes and the vehicle model nodes, verifying geometric suitability based on a cargo voxel set and an effective loading voxel module, verifying operation operability based on an operation entry vector and a loading port vector set, carrying out transportation suitability analysis driven by historical image data, screening vehicles passing verification to generate a single vehicle scheme, simultaneously executing a load decomposition algorithm to disassemble the cargo entity nodes into subtasks, retrieving vehicle combinations, carrying out behavior consistency verification based on a historical track and a speed cooperation entropy, and outputting a candidate scheme set comprising the single vehicle scheme and the multi-vehicle cooperation scheme.

Inventors

  • ZHU WEI

Assignees

  • 无锡新宝锡物流有限公司

Dates

Publication Date
20260505
Application Date
20260129

Claims (8)

  1. 1. The order matching method based on the knowledge graph is characterized by comprising the following steps of: acquiring order data to be processed, wherein the order data comprises text description and image information; carrying out multi-mode feature extraction on the order data to obtain order feature data, and mapping the order feature data in a knowledge graph to generate cargo entity nodes, wherein the cargo entity nodes comprise cargo voxel sets based on image reconstruction and job entry vectors; Calling vehicle model nodes in the knowledge graph, wherein the vehicle model nodes comprise an effective loading voxel module for mapping a carriage interface, a loading and unloading port vector set and historical portrait data for representing vehicle operation characteristics; performing multidimensional matching verification based on cargo entity nodes and vehicle model nodes, including verifying geometric suitability based on cargo voxel sets and valid load voxel modules, verifying job operability based on job entry vectors and load port vector sets, and transportation suitability analysis based on historical portrait data drives; meanwhile, a load decomposition algorithm is executed to decompose cargo entity nodes into subtasks, vehicle combinations are searched, behavior consistency verification is carried out based on historical track and speed cooperative entropy, and a multi-vehicle cooperative scheme is generated; And outputting a candidate scheme set containing a single-car scheme and a multi-car cooperative scheme.
  2. 2. The knowledge-graph-based order matching method of claim 1, wherein: The specific steps of mapping and generating the cargo entity nodes in the knowledge graph include: carrying out semantic analysis on the text description to obtain cargo parameters, wherein the cargo parameters comprise cargo types and cargo weights; three-dimensional point cloud reconstruction is carried out on multi-view image information by utilizing a multi-view geometric reconstruction algorithm, and point cloud data are subjected to rasterization processing to generate a cargo voxel set describing cargo geometric occupation information; the method comprises the steps of identifying operation interaction characteristic points in an image through a target detection algorithm, wherein the operation interaction characteristic points comprise a hoisting ring, forklift jacks and carrying handles; And packaging the goods voxel set and the operation entry vector in the order feature data as attribute values, and generating the goods entity node by instantiation in a knowledge graph.
  3. 3. The knowledge-graph-based order matching method of claim 1, wherein: the specific steps of calling the vehicle model nodes in the knowledge graph include: searching active capacity vehicle identifications in a space-time range based on the starting point coordinate information of the order and a preset time window, and indexing corresponding vehicle model nodes in a vehicle body library of the knowledge graph according to the searched active capacity vehicle identifications; The data structure of the vehicle model node comprises: the method comprises the steps of effectively loading a voxel module, and removing available loading space grids generated after a wheel package invasion area, a reinforcing rib bulge area and a vehicle-mounted equipment occupation area after three-dimensional modeling is carried out on the carriage inner space; a loading and unloading port vector set, based on a space normal direction set defined by a physical opening structure of a vehicle compartment, comprising a vertical vector defining a top opening direction, a horizontal vector defining a side opening direction and a longitudinal vector defining a tail opening direction; the historical portrait data comprises basic configuration parameters of the vehicle and an associated historical transportation case set, wherein the historical transportation case set records process parameters for executing different types of order data in past tasks, and the process parameters at least comprise vehicle body vibration spectrum data and cargo integrity rate feedback data in the transportation process.
  4. 4. The knowledge-graph-based order matching method of claim 1, wherein: The step of verifying geometric suitability based on the cargo voxel set and the valid load voxel module comprises: The method comprises the steps of taking a cargo voxel set of a cargo entity node as source data, taking an effective loading voxel module of a vehicle model node as target data, conducting multi-degree-of-freedom pose search and placement simulation on the source data in a virtual three-dimensional space coordinate system, calculating a Boolean intersection operation result and a Boolean difference operation result of the source data and the target data according to each search pose, and judging that geometric adaptation is passed if the pose exists, so that the Boolean intersection operation result is equivalent to the source data in geometric volume and the collision detection result of a non-loading area outside the source data and the target data is zero.
  5. 5. The knowledge-graph-based order matching method as set forth in claim 4, wherein: The specific steps of verifying job operability based on the job entry vector and the loadport vector set include: The method comprises the steps of selecting a loading and unloading port vector which is matched with the operation entry vector angle, and constructing a virtual operation path, taking the operation entry vector of a cargo entity node as a guide, and carrying out continuous pose transformation on a cargo voxel set along the virtual operation path; in the continuous pose transformation process, generating a dynamic sweep voxel set describing the space occupation state of the cargo in the whole loading and unloading motion whole process by a space superposition algorithm; calculating collision volumes of the dynamic sweep voxel set and door frames, uprights and non-loading areas defined in the virtual vehicle model nodes; if the collision volume is zero, judging that the operation feasibility check passes, and recording the minimum Euclidean distance between the cargo surface voxel and the vehicle obstacle in the simulation process, and recording the minimum Euclidean distance as a minimum safety gap.
  6. 6. The knowledge-graph-based order matching method of claim 1, wherein: the specific steps of the transportation suitability analysis based on the historical data drive comprise: Calculating cosine similarity of current order feature data and vehicle historical order data corresponding to vehicle model nodes, and searching out a historical transportation case with similarity higher than a preset threshold value; Inputting the current order feature data, the knowledge graph node features of the candidate vehicles and the interactive process parameters of the retrieved historical transportation cases into a pre-trained transportation adaptability network model; Analyzing the influence of the cargo attribute on the running state of the vehicle by utilizing the cross attention mechanism layer in the transportation adaptability network model, and outputting a vehicle loss risk coefficient; and carrying out weighted normalization processing on the vehicle loss risk coefficient and the cargo damage predicted value to generate a transportation suitability index representing the matching degree of the cargo and the vehicle, and judging that the transportation suitability check is not passed if the index is lower than a preset threshold value.
  7. 7. The knowledge-graph-based order matching method as set forth in claim 5, wherein: The specific steps for generating the bicycle scheme comprise: For each candidate vehicle passing through the geometric suitability, the operation feasibility and the transportation suitability verification, calculating a safety margin score of the candidate vehicle, wherein the safety margin score is calculated by weighting calculation based on the residual filling rate of an effective loading space, the operation entering fluency indexes of an operation entering vector and a loading and unloading port vector and the transportation suitability index; Calculating the residual filling rate, namely counting the total number of non-empty voxels contained in the cargo voxel set, counting the total number of all available voxels contained in the valid loading voxel module, subtracting the total number of non-empty voxels from the total number of all available voxels, and taking the ratio of the obtained difference value to the total number of all available voxels as the residual filling rate of the valid loading space; The calculation of the operation entrance fluency index comprises the steps of constructing a normalized mapping function based on a Sigmoid function, and calculating the minimum safety gap to obtain the operation entrance fluency index.
  8. 8. The knowledge-graph-based order matching method of claim 1, wherein: the specific steps for generating the multi-vehicle cooperative scheme comprise: Triggering a splitting instruction, analyzing a cargo packing list contained in the order data, and extracting physical size and quantity parameters of each independent unit; guiding a reverse three-dimensional packing algorithm based on the cargo packing list, dividing a cargo voxel set of the cargo entity node into a plurality of subtask nodes, and re-instantiating physical properties of each subtask node; Searching vehicle combinations capable of bearing subtask nodes respectively, calling positioning track data of each vehicle in the combinations under the same-period and same-type historic paths, and calculating the historic track similarity among the vehicles by using a dynamic time warping algorithm; extracting historical speed time sequences of all vehicles to construct a speed probability distribution model, and calculating speed distribution difference values among vehicle combinations by using a relative entropy algorithm, namely speed cooperative entropy; and screening vehicle combinations with speed cooperative entropy lower than a preset threshold and history track similarity higher than the preset threshold, and generating a multi-vehicle cooperative scheme so as to ensure the space-time synchronism of a plurality of vehicles in the transportation process.

Description

Order matching method based on knowledge graph Technical Field The invention relates to the technical field of big data analysis, in particular to an order matching method based on a knowledge graph. Background With the rapid development of the modern logistics industry, the network freight platform and the intelligent logistics system are widely applied to a car-cargo matching scene. Existing order matching techniques rely primarily on structured text labels and simplified geometric parameters for filtering. Specifically, the weight, total volume and type labels of goods in an order are generally extracted and compared with rated load, outline size and vehicle type in a vehicle database, and a three-dimensional boxing algorithm based on cuboid bounding boxes is introduced in part of the method to estimate the loading rate. However, the above prior art has a significant technical problem in practical application, namely, the matching model is severely disjointed from physical reality, resulting in frequent occurrence of pseudo-matching. Existing bounding box algorithms ignore the topology details of the physical entities. For example, there are often areas within the vehicle cargo compartment where the wheel housing is raised, ribs or refrigeration units are not loaded, and the cargo may have grooves or irregular protrusions. Second, there is a lack of feasibility verification of the dynamic job process. The prior art only focuses on the static suitability of the cargo in the final loading position, but ignores the kinematic constraints during loading and unloading. Finally, the suitability of the hidden features is ignored, the existing matching logic only usually considers the physical space and the load, does not consider the coupling relation between the running characteristics of the vehicle and the cargo attributes, and also fails to effectively solve the problem of time-space asynchronism caused by the difference of the running habits of the vehicles in the multi-vehicle cooperative task. Therefore, how to construct a refined order matching method capable of integrating high-precision physical space constraint, dynamic operation process verification and hidden transportation characteristics is a technical problem to be solved in the technical field of current logistics. For this purpose, an order matching method based on a knowledge graph is provided. Disclosure of Invention The invention aims to provide an order matching method based on a knowledge graph, which screens matching schemes passing verification through analysis of geometric suitability, operation operability and transportation suitability. In order to achieve the above object, the present invention provides an order matching method based on a knowledge graph, including: acquiring order data to be processed, wherein the order data comprises text description and image information; carrying out multi-mode feature extraction on the order data to obtain order feature data, and mapping the order feature data in a knowledge graph to generate cargo entity nodes, wherein the cargo entity nodes comprise cargo voxel sets based on image reconstruction and job entry vectors; Calling vehicle model nodes in the knowledge graph, wherein the vehicle model nodes comprise an effective loading voxel module for mapping a carriage interface, a loading and unloading port vector set and historical portrait data for representing vehicle operation characteristics; performing multidimensional matching verification based on cargo entity nodes and vehicle model nodes, including verifying geometric suitability based on cargo voxel sets and valid load voxel modules, verifying job operability based on job entry vectors and load port vector sets, and transportation suitability analysis based on historical portrait data drives; meanwhile, a load decomposition algorithm is executed to decompose cargo entity nodes into subtasks, vehicle combinations are searched, behavior consistency verification is carried out based on historical track and speed cooperative entropy, and a multi-vehicle cooperative scheme is generated; And outputting a candidate scheme set containing a single-car scheme and a multi-car cooperative scheme. The specific steps of mapping and generating the cargo entity nodes in the knowledge graph include: carrying out semantic analysis on the text description to obtain cargo parameters, wherein the cargo parameters comprise cargo types and cargo weights; three-dimensional point cloud reconstruction is carried out on multi-view image information by utilizing a multi-view geometric reconstruction algorithm, and point cloud data are subjected to rasterization processing to generate a cargo voxel set describing cargo geometric occupation information; the method comprises the steps of identifying operation interaction characteristic points in an image through a target detection algorithm, wherein the operation interaction characteristic points comprise a hoisting