Intelligent logistics management method and system based on knowledge graph

Abstract

The invention relates to the field of logistics management, and particularly discloses an intelligent logistics management method and system based on a knowledge graph, and the method comprises the steps: recognizing key situation elements of a logistics support task in real time through a situation perception model; dynamically extracting related knowledge sub-graphs from a pre-constructed knowledge graph based on the situation elements; carrying out multi-dimensional evaluation on the materials in the knowledge sub-graph space and generating an allocation scheme; virtual execution and conflict detection are carried out on the scheme through a digital twin environment, and a final optimization scheme is generated through dynamic correction; the system comprises a multi-source data acquisition and context awareness module, a knowledge graph management and dynamic query module, a multi-dimensional evaluation and scheme generation module, a digital twinning and scheme optimization module and a scheme execution and feedback optimization module. According to the method, the problems of data islanding, low decision-making efficiency and lack of scheme feasibility verification are solved, the whole-process intelligent management from data perception to scheme optimization is realized, and the accuracy and reliability of logistics guarantee are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of logistics management technology, and in particular to an intelligent logistics management method and system based on knowledge graphs. Background Technology

[0002] Existing logistics management systems suffer from technical deficiencies, making it difficult to meet increasingly complex support needs. Firstly, at the data and knowledge layers, current technologies face significant challenges. The prevalent use of relational databases combined with digital document storage results in robust "data silos" formed from heterogeneous data from multiple sources, including procurement, warehousing, and operations. For example, there is a lack of effective semantic connections between technical parameters from the design department, inventory status from the warehousing department, and fault records from the user department, preventing the formation of a unified knowledge system. While knowledge graph technology has applications in other fields, its application in the specific scenario of logistics management remains superficial. Existing knowledge graph construction methods suffer from significant data redundancy and struggle to support efficient range queries and reasoning on continuous material attributes (such as inventory levels and service life). More importantly, the maintenance of traditional knowledge bases heavily relies on manual input and annotation, which is not only inefficient but also fails to effectively capture and integrate "tacit knowledge" scattered across various documents and personnel experience, leading to lagging system knowledge updates and an inability to adapt to dynamic changes in the logistics environment.

[0003] Secondly, at the decision support layer, the existing systems suffer from a severe lack of intelligence. When faced with resource shortages and the need to find alternative solutions, most systems can only provide simple searches based on keyword matching, while complex matching decisions still rely on the personal experience of managers. This method is inefficient and makes it difficult to guarantee the scientific validity and consistency of decisions. Especially in emergency response scenarios, the system lacks the ability to dynamically perceive and understand the task context (such as urgency, time and space constraints, and environmental conditions), and cannot generate targeted support plans based on real-time changing task requirements. The decision-making process becomes an open-loop system, lacking effective verification and optimization mechanisms.

[0004] Finally, there are significant gaps in existing technologies at the solution verification layer. Traditional systems generate resource allocation plans that lack forward-looking verification methods, failing to predict and identify potential resource conflicts, path conflicts, or time conflicts before implementation. This often leads to temporary adjustments during actual execution, impacting efficiency and introducing additional operating costs and risks. The system lacks a "digital sandbox" capable of simulating the complexities of the real world, making decision quality largely dependent on the decision-maker's experience and luck.

[0005] Therefore, there is an urgent need in this field for a novel logistics management method that can break down data silos, enable intelligent decision-making, and provide reliable verification. This method needs to be able to automatically construct and dynamically update a domain knowledge graph, possess context awareness and understanding capabilities, and fully verify and optimize decision-making solutions in a virtual environment, thereby comprehensively improving the intelligence level and support efficiency of logistics management. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent logistics management method and system based on knowledge graphs.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: An intelligent logistics management method based on knowledge graphs includes: Real-time collection of multi-source operational data from the logistics system; identification of key contextual elements of the current logistics support task through a pre-trained contextual classification model; the key contextual elements include task type, time constraints, resource availability, and environmental conditions. Based on the identified key contextual elements, knowledge subgraphs related to the current context are dynamically extracted from the pre-constructed logistics material knowledge graph. The knowledge subgraphs contain material entities and their relationships that are strongly associated with the current contextual elements. Based on the knowledge subgraph, available resources are evaluated in multiple dimensions, the context suitability of each resource is calculated, and a resource allocation plan is generated based on the evaluation results. The generated resource allocation plan is virtually executed in a digital twin environment. By simulating the actual execution process, potential resource conflicts and spatiotemporal constraint conflicts are detected. Based on the detection results, the plan is dynamically modified, and the final optimized plan is generated and pushed to the execution terminal.

[0008] As a further technical solution of the present invention, the real-time acquisition of multi-source operational data of the logistics system, and the identification of key contextual elements of the current logistics support task through a pre-trained contextual classification model, specifically includes: Through a predefined data interface protocol, device status data is collected in real time from IoT sensor nodes, task order data is extracted periodically from the business system database, and environmental parameter data is continuously obtained from the environmental monitoring API interface, and the multi-source operation data is aggregated into a unified data stream. The unified data stream is cleaned and fused. A pre-trained context classification model is used to perform in-depth analysis on the fused data to extract temporal features that characterize the urgency of tasks, statistical features that reflect resource priorities, and topological features that describe spatiotemporal constraints. Based on the extracted temporal, statistical, and topological features, the key contextual elements of the current logistics support task are dynamically weighted and output through a multi-level attention mechanism in a pre-trained context classification model.

[0009] As a further technical solution of the present invention, the step of dynamically extracting knowledge subgraphs related to the current context from a pre-constructed logistics material knowledge graph based on the identified key contextual elements specifically includes: The quantitative indicators of the key context elements are mapped to the initial query node set in the knowledge graph, and different initial traversal weights are assigned to each context element type. Execute context-driven multi-path graph traversal, starting from the initial query node set, and expand the search along the semantic relationship edges and spatiotemporal relationship edges between material entities. During the traversal, dynamically adjust the depth and breadth of path exploration according to the relationship type and the semantic relevance of the current context elements. During the traversal, the overall correlation strength between each accessed material entity and the current context elements is calculated in real time, and entities with correlation strength exceeding a preset threshold and their relationships are filtered out. The selected entities and relationships with high correlation strength are integrated to construct a concise knowledge subgraph oriented towards the current task context. This knowledge subgraph serves as a dedicated knowledge space for subsequent multi-dimensional evaluation.

[0010] As a further technical solution of the present invention, the step of mapping the quantitative indicators of key contextual elements to an initial query node set in a knowledge graph, and assigning different initial traversal weights to each contextual element type, specifically includes: Define quantitative indicators for key contextual elements, including quantitative values ​​for task type and urgency. Quantification of time constraint urgency Quantitative value of resource availability sufficiency and the quantitative value of the severity of environmental conditions ; The above quantitative indicators are mapped to initial query nodes in the knowledge graph. This constitutes the initial query node set. ; Different initial traversal weights are assigned to each context element type. These weights can be determined based on historical data and expert experience. ,satisfy .

[0011] As a further technical solution of the present invention, the execution context-driven multi-path graph traversal starts from the initial query node set and expands the search along the semantic relationship edges and spatiotemporal relationship edges between material entities. During the traversal process, the depth and breadth of path exploration are dynamically adjusted according to the relationship type and the semantic relevance of the current context elements, specifically including: In knowledge graph middle, For the collection of physical nodes of materials, This is the set of semantic and spatiotemporal relation edges between entities; starting from the initial query node set. Starting from the initial weights, initiate a multi-path graph traversal algorithm, such as an improved breadth-first search or depth-first search algorithm. During the traversal, for the currently visited node The depth and breadth of path exploration are dynamically adjusted based on the semantic and spatiotemporal relevance of the path to each initial query node; semantic relevance can be obtained through a pre-trained semantic similarity calculation model, such as... Represents a node With the initial query node The semantic similarity between nodes; the spatiotemporal relevance is calculated based on the spatiotemporal information of the resources represented by the nodes and the spatiotemporal constraints of the current task context, such as... and Representing nodes respectively Spatial and temporal correlation; The edge weights are dynamically adjusted based on the semantic relevance of the relation type and the elements in the current context. It can be represented as: ,in: For the edge The original weights, To adjust the coefficient, For the edge The semantic relevance of the relationship type to the elements in the current context. As a further technical solution of the present invention, the step of calculating the comprehensive association strength between each accessed material entity and the current contextual elements in real time during the traversal process, and filtering out entities with association strength exceeding a preset threshold and the relationships between them, specifically includes: For each accessed material entity node It calculates the overall correlation strength between itself and the elements of the current context in real time. The calculation formula is as follows: ,in: This represents the initial number of query nodes. For the initial query node The corresponding weights To balance the parameters of semantic relevance and spatiotemporal relevance, This is a balance parameter between spatial and temporal correlation. Comprehensive correlation strength With preset threshold If a comparison is made, If so, then retain the entity node. And its relationship edges with adjacent nodes.

[0012] As a further technical solution of the present invention, the selected highly correlated entities and relationships are integrated to construct a concise knowledge subgraph oriented towards the current task context. This knowledge subgraph serves as a dedicated knowledge space for subsequent multi-dimensional evaluation and specifically includes: The selected highly correlated entity nodes and their relationship edges are integrated to construct a concise knowledge subgraph tailored to the current task context. ,in It is a set of highly correlated entity nodes selected from the pool. It is a set of relational edges between entity nodes; Once the knowledge subgraph is constructed, it serves as a dedicated knowledge space for subsequent multi-dimensional evaluation, providing targeted knowledge support for the generation of resource allocation plans.

[0013] As a further technical solution of the present invention, the step of evaluating available resources in multiple dimensions based on the knowledge subgraph, calculating the context suitability of each resource, and generating a resource allocation plan based on the evaluation results specifically includes: Within the dedicated knowledge space formed by the knowledge subgraph, for each available material entity, its evaluation value in four dimensions—functional matching degree, spatiotemporal accessibility, cost-effectiveness, and risk coefficient—is calculated in parallel. The functional matching degree is obtained by calculating the semantic path similarity between the material attributes and the task requirements in the knowledge subgraph; the spatiotemporal accessibility is obtained by analyzing the topological relationship between the material location, transportation route, and task location in the knowledge subgraph. Based on the dynamic weight configuration of the current key situational elements, the evaluation values ​​of the four dimensions are weighted and fused to calculate the comprehensive situational suitability score of each material entity; among them, the weight of spatiotemporal accessibility is increased when the task is urgent, and the weight of cost-effectiveness is increased when resources are scarce. All available material entities are sorted according to the comprehensive situational suitability score. Based on the sorting results, total resource constraints, and multiple task requirements, one or more optimal material allocation schemes are generated through constraint satisfaction algorithm. The schemes specify the allocation quantity, scheduling path, and execution sequence of specific materials.

[0014] As a further technical solution of the present invention, the generated material allocation plan is virtually executed in a digital twin environment. Potential resource conflicts and spatiotemporal constraint conflicts are detected by simulating the actual execution process. Based on the detection results, the plan is dynamically corrected to generate a final optimized plan and push it to the execution terminal. Specifically, this includes: Based on the aforementioned material allocation plan and logistics material knowledge graph, a digital twin simulation environment is constructed that includes dynamic resource status, transportation network topology, and execution progress timeline. Using a discrete event simulation engine, the digital twin environment is driven to run according to the execution sequence set in the scheme. During the operation, one or more conflict types, such as resource allocation conflict, equipment usage time conflict and transportation route conflict, are monitored and recorded in real time. When a conflict is detected, a conflict resolution algorithm is triggered to search for feasible correction strategies based on the knowledge subgraph. The correction strategies include one or more combinations of resource replacement, task rescheduling, and path optimization. The revised solution is re-injected into the digital twin environment for iterative verification until all conflicts are resolved. The final optimized solution is then generated and pushed to each execution terminal via API. At the same time, the solution revision trajectory and optimization basis are displayed in a visual dashboard.

[0015] An intelligent logistics management system based on knowledge graphs, used to implement an intelligent logistics management method based on knowledge graphs, includes: The multi-source data acquisition and context awareness module is used to collect multi-source operational data of the logistics system in real time and identify key context elements of the current logistics support task through a pre-trained context classification model. The knowledge graph management and dynamic query module is used to dynamically extract knowledge subgraphs related to the current context from the pre-built logistics material knowledge graph based on the identified key context elements. The multi-dimensional assessment and solution generation module is used to conduct multi-dimensional assessments of available resources based on the knowledge subgraph, calculate the context suitability of each resource, and generate resource allocation plans based on the assessment results. The digital twin and solution optimization module is used to virtually execute the generated material allocation plan in a digital twin environment. By simulating the actual execution process, it detects potential resource conflicts and spatiotemporal constraint conflicts, and dynamically corrects the plan based on the detection results to generate the final optimized plan. The scheme execution and feedback optimization module is used to push the final optimized scheme to the execution terminal and collect scheme execution feedback data to optimize the system model.

[0016] The beneficial effects of this invention are as follows: 1. This method achieves precise decision-making and rapid response based on context awareness, significantly improving the agility and adaptability of logistical support. By constructing a dynamic context-aware model, this method identifies and quantifies key contextual elements of a task in real time, overcoming the shortcomings of rigid decision-making and disconnect from the environment in traditional systems. It transforms the fuzzy task environment into calculable contextual indicators, ensuring that the system can deeply understand implicit requirements such as task urgency and resource priority, thereby achieving a leap from "static resource matching" to "dynamic context adaptation".

[0017] 2. By dynamically focusing on knowledge subgraphs and using semantic reasoning, the algorithm overcomes the challenges of low efficiency and irrelevant redundancy in full graph queries, significantly optimizing computational resource utilization and decision accuracy. It abandons the crude approach of traversing the entire knowledge graph and instead employs a context-driven graph traversal algorithm to dynamically construct a highly concise knowledge subgraph strongly relevant to the current task as a dedicated decision space. This not only reduces retrieval time in a graph with millions of nodes from minutes to milliseconds, but more importantly, it filters out irrelevant information, ensuring that evaluation and decision-making are based entirely on the most relevant knowledge fragments.

[0018] 3. A digital twin-driven pre-execution and iterative optimization mechanism is introduced, constructing a complete decision-making verification closed loop, fundamentally improving the feasibility of the solution and the system risk prevention and control capabilities. This method adds a virtual execution and dynamic correction stage to the solution. By simulating the actual operation of the solution in a digital twin environment, approximately 80% of potential risks such as resource conflicts and spatiotemporal conflicts can be detected in advance, and iterative optimization can be automatically performed using conflict resolution algorithms. This realizes a new paradigm of "verification before execution," allowing solution defects that can only be discovered through trial and error in the real world under traditional models to be resolved in the virtual space in advance. Attached Figure Description

[0019] Figure 1 This is a flowchart of an intelligent logistics management method based on knowledge graphs proposed in this invention; Figure 2 This is a module diagram of an intelligent logistics management system based on knowledge graphs proposed in this invention; Figure 3 A module diagram of a pre-trained context classification model; Figure 4 Example diagram of a pre-constructed knowledge graph of logistics supplies; Figure 5 The topology diagram for step S4, digital twin virtual execution and dynamic optimization process; Figure 6 This is a schematic diagram comparing the effects of the present invention and the prior art in Example 1. Figure 1 ; Figure 7This is a schematic diagram comparing the effects of the present invention and the prior art in Example 1. Figure 2 . Detailed Implementation

[0020] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0021] Please see the appendix Figure 1 A knowledge graph-based intelligent logistics management method includes: S1. Collect multi-source operational data of the logistics system in real time, and identify key context elements of the current logistics support task through a pre-trained context classification model. The key context elements include task type, time constraints, resource availability and environmental conditions. like Figure 3 As shown, the pre-trained context classification model is a deep learning model trained on historical logistics task data. Its structure includes an input layer, a feature extraction layer, and an output layer.

[0022] The model consists of several layers: Input layer: receives feature vectors from multiple sources of operational data (device status, task orders, environmental parameters); Feature extraction layer: includes a CNN module (for extracting local features) and an LSTM module (for extracting temporal dependencies); Attention mechanism layer: dynamically weighted features using a multi-level attention mechanism; Output layer: outputs classifications of key contextual elements (task type, time constraints, resource availability, environmental conditions). This model can dynamically identify contextual elements such as task urgency and resource priority based on real-time data.

[0023] S11. Define Time Series At each point in time Collect device status data from IoT sensor nodes Extract task order data from the business system database Obtain environmental parameter data from the environmental monitoring API interface. ; The above three types of data will be aggregated into a unified data stream. , is represented as: The merge operation includes time alignment and format standardization to ensure that the data is synchronized in time and consistent in format.

[0024] S12. For unified data streams The data is cleaned to remove noise and outliers, resulting in cleaned data. , is represented as: ; The cleaned data is then fused, and feature vectors are extracted using feature extraction methods. , is represented as: .

[0025] S13. Temporal characteristics representing the urgency of a task Using time series analysis methods to analyze task order data extract; Statistical characteristics reflecting resource priority Statistical analysis methods were used to analyze equipment status data. extract; Topological features describing spatiotemporal constraints Using topology analysis methods to analyze environmental parameter data extract.

[0026] S14. Define a multi-level attention mechanism (MAAM), whose input is temporal features. Statistical features and topological features The output is a weighted vector of key context elements. , is represented as: ; In MAAM, the attention weights for each class of features are first calculated: Where: Attention is an attention function, such as the Softmax function: ; Then, the features are weighted and fused: ; Finally, through context classification models right Classify and identify key contextual elements {Task type, time constraint, resource availability, environmental conditions}, represented as: .

[0027] S2. Based on the identified key contextual elements, dynamically extract knowledge subgraphs related to the current context from the pre-constructed logistics material knowledge graph. The knowledge subgraphs contain material entities and their relationships that are strongly associated with the current contextual elements. like Figure 4 As shown, the pre-built logistics material knowledge graph is constructed by integrating multi-source data from the logistics domain, including material entities, attributes, and their relationships. The knowledge graph construction process includes data extraction, data fusion, and knowledge storage. Material entities include equipment, spare parts, tools, etc., attributes include location, status, quantity, etc., and relationships include semantic relationships (such as substitution relationships and compatibility relationships) and spatiotemporal relationships (such as adjacent locations and overlapping times). The knowledge graph is stored using a graph database, supporting efficient graph traversal and querying.

[0028] S21. Define quantitative indicators for key contextual elements, including quantitative values ​​for task type and urgency. Quantification of time constraint urgency Quantitative value of resource availability sufficiency and the quantitative value of the severity of environmental conditions ; The above quantitative indicators are mapped to initial query nodes in the knowledge graph. This constitutes the initial query node set. ; Different initial traversal weights are assigned to each context element type. These weights can be determined based on historical data and expert experience. ,satisfy .

[0029] S22. In knowledge graphs middle, For the collection of physical nodes of materials, This is the set of semantic and spatiotemporal relation edges between entities; starting from the initial query node set. Starting from the initial weights, initiate a multi-path graph traversal algorithm, such as an improved breadth-first search or depth-first search algorithm. During the traversal, for the currently visited node The depth and breadth of path exploration are dynamically adjusted based on the semantic and spatiotemporal relevance of the path to each initial query node; semantic relevance can be obtained through a pre-trained semantic similarity calculation model, such as... Represents a node With the initial query node The semantic similarity between nodes; the spatiotemporal relevance is calculated based on the spatiotemporal information of the resources represented by the nodes and the spatiotemporal constraints of the current task context, such as... and Representing nodes respectively Spatial and temporal correlation; The edge weights are dynamically adjusted based on the semantic relevance of the relation type and the elements in the current context. It can be represented as: ,in: For the edge The original weights, To adjust the coefficient, For the edge The semantic relevance of the relationship type to the elements in the current context.

[0030] S23. For each accessed material entity node It calculates the overall correlation strength between itself and the elements of the current context in real time. The calculation formula is as follows: ,in: This represents the initial number of query nodes. For the initial query node The corresponding weights To balance the parameters of semantic relevance and spatiotemporal relevance, This is a balance parameter between spatial and temporal correlation. Comprehensive correlation strength With preset threshold If a comparison is made, If so, then retain the entity node. And its relationship edges with adjacent nodes.

[0031] S24. Integrate the selected highly correlated entity nodes and the relationship edges between them to construct a concise knowledge subgraph tailored to the current task context. ,in It is a set of highly correlated entity nodes selected from the pool. It is a set of relational edges between entity nodes; Once the knowledge subgraph is constructed, it serves as a dedicated knowledge space for subsequent multi-dimensional evaluation, providing targeted knowledge support for the generation of resource allocation plans.

[0032] S3. Based on the knowledge subgraph, perform multi-dimensional evaluation of available resources, calculate the context suitability of each resource, and generate a resource allocation plan based on the evaluation results; S31. Functional matching degree is measured by calculating the length of the shortest semantic path between the material attribute and the task requirement in the knowledge subgraph and the reciprocal of the sum of its edge weights. The formula is: ,in: Functional compatibility; , From the task requirement node To the physical body of the supplies The most succinct semantic path, It is the edge The weight represents the strength of semantic relevance; It is a very small positive number; S32. Spatiotemporal accessibility modeling utilizes the shortest path length from resources to the task location and task time constraints for comprehensive calculation. The formula is: ,in: For spatiotemporal accessibility, It is a physical material. Arrive at the mission location The shortest path length, It is the time constraint of the task. and It is the scaling factor; S33. Cost-effectiveness is calculated as the ratio of the value of the materials to the cost of their allocation. ,in: For cost-effectiveness, For the value of materials, The cost of invocation; the risk coefficient is derived from a predefined risk assessment model: ,in: For risk coefficient, For risk assessment models; S34. Overall Contextual Fit Score: Based on the dynamic weights of key contextual elements, the evaluation values ​​of the four dimensions are weighted and integrated, using the following formula: ,in: To ensure overall contextual adaptability, , , , The weights are for four dimensions, and the weights satisfy... The weights can be dynamically adjusted based on the urgency of the task and the scarcity of resources; S35. Based on the comprehensive situational suitability score, the resources are sorted. Combining the total resource constraints and task requirements, the constraint satisfaction algorithm is used to generate the optimal resource allocation plan. The objective function is: The constraints are: ,in: It is supplies The allocation quantity, It is the amount of resources consumed. This is the total available resources.

[0033] S4. The generated material allocation plan is virtually executed in a digital twin environment. Potential resource conflicts and spatiotemporal constraint conflicts are detected by simulating the actual execution process. The plan is dynamically modified based on the detection results, and the final optimized plan is generated and pushed to the execution terminal.

[0034] like Figure 5 As shown, the digital twin virtual execution and dynamic optimization process includes environment construction, driving and monitoring, intelligent correction, iterative optimization, and reliable delivery.

[0035] S41. The digital twin simulation environment is represented as... ,in: This represents a set of dynamic resource states, where each resource state... It includes information such as the type, quantity, and location of the resources; Indicates the transportation network topology. It is a set of nodes, representing warehouses, distribution centers, etc. It is a set of edges, representing the transportation paths between nodes, each edge It has attributes such as distance and transit time; It indicates the execution progress sequence, including the task's start time, end time, etc.; The function for updating the dynamic status of resources is: ,in: It is a change in the state of resources. It is time. Update resource status according to actual business rules.

[0036] S42. The simulation engine drives the digital twin environment to run according to the execution timing set in the scheme. The execution timing is represented as a time series. At each point in time Triggering corresponding events, such as resource allocation and transportation; During the simulation, conflict types are monitored and recorded in real time, and conflicts are represented as a set. , where: each conflict It includes information such as the type of conflict, when it occurred, and the resources involved; for example, a resource allocation conflict can be represented as: ,in: It is a type of conflict. It refers to the time when the conflict occurred. It refers to the resources involved.

[0037] S43. When a conflict is detected, the conflict resolution algorithm is triggered, and feasible correction strategies are searched based on the knowledge subgraph. The correction strategies are represented as a set. Each strategy This includes strategy types (such as resource replacement, task rescheduling, and path optimization) and specific operational steps; The choice of correction strategy is based on a certain evaluation function, such as: ,in: The best correction strategy, It's weight. These are the evaluation functions for the strategy's impact on resource utilization, time scheduling, and cost.

[0038] S44. Re-inject the revised solution into the digital twin environment for iterative verification until all conflicts are resolved. The iterative verification process can be represented by a loop structure: while do Knowledge Subgraph Search Correction Strategy Choose the best correction strategy Application correction strategy update scheme Detecting conflicts in the new scheme end while The final optimized solution is pushed to each execution terminal via API, and the solution correction trajectory and optimization basis are displayed in a visual dashboard. The push process can be represented as follows: ,in: This is the final optimized solution; the visual display includes information such as the previous revisions of the solution, the conflict resolution process, and changes in the contextual adaptability of each material entity.

[0039] Please see the appendix Figure 2 An intelligent logistics management system based on knowledge graphs, used to implement an intelligent logistics management method based on knowledge graphs, includes: The multi-source data acquisition and context awareness module is used to acquire multi-source operational data from the logistics system in real time and identify key contextual elements of the current logistics support task through a pre-trained context classification model; specifically, it includes: The data interface unit is used to collect multi-source operational data from IoT sensor nodes, business system databases, and environmental monitoring API interfaces through predefined data interface protocols; The data fusion unit is used to clean and fuse the collected multi-source operational data to form a unified data stream. The context element extraction unit is used to analyze the unified data stream using a pre-trained context classification model and output quantitative indicators of key context elements based on a multi-level attention mechanism.

[0040] The knowledge graph management and dynamic query module is used to dynamically extract knowledge subgraphs related to the current context from a pre-built logistics materials knowledge graph based on identified key contextual elements; specifically, it includes: The graph storage unit is used to store and manage entity, attribute, and relationship data of the logistics material knowledge graph; The context mapping unit is used to map the quantitative indicators of key context elements to the initial query node set in the knowledge graph; The graph traversal computation unit is used to execute the context-driven multi-path graph traversal algorithm, starting from the initial set of query nodes and expanding the search along semantic relation edges and spatiotemporal relation edges. The knowledge subgraph construction unit is used to filter and integrate highly related entities and relationships based on the entity association strength calculated during the traversal process, and to construct a knowledge subgraph oriented towards the current task context.

[0041] The multi-dimensional assessment and solution generation module is used to perform multi-dimensional assessments of available resources based on the knowledge subgraph, calculate the context suitability of each resource, and generate resource allocation plans based on the assessment results; specifically including: The multi-dimensional evaluation unit is used to perform parallel calculations of the evaluation values ​​of available resources in terms of functional matching degree, spatiotemporal accessibility, cost-effectiveness and risk coefficient within a dedicated knowledge space composed of knowledge subgraphs. The adaptability calculation unit is used to generate a comprehensive situation adaptability score for each material by weighting and integrating multi-dimensional evaluation values ​​based on the dynamic weight configuration of key situational elements. The scheme generation unit is used to generate the optimal material allocation scheme that satisfies the total resource constraints and multiple task requirements based on the comprehensive situational adaptability score ranking results and through constraint satisfaction algorithm.

[0042] The digital twin and solution optimization module is used to virtually execute the generated resource allocation plan in a digital twin environment. It detects potential resource conflicts and spatiotemporal constraint conflicts by simulating the actual execution process, and dynamically corrects the plan based on the detection results to generate a final optimized plan. Specifically, it includes: The simulation environment construction unit is used to build a digital twin environment that includes the dynamic status of resources, transportation network topology, and execution progress timeline based on the material allocation plan and logistics material knowledge graph. The conflict detection unit is used to drive the operation of the digital twin environment through the discrete event simulation engine, and to monitor resource allocation conflicts, equipment usage time conflicts, and transportation route conflicts in real time. The dynamic correction unit is used to trigger the conflict resolution algorithm when a conflict is detected, and to search and implement resource replacement, task rescheduling or path optimization correction strategies based on knowledge subgraph search. The iterative verification unit is used to re-inject the modified scheme into the digital twin environment for iterative verification until a conflict-free final optimized scheme is generated.

[0043] The scheme execution and feedback optimization module is used to push the final optimized scheme to the execution terminal and collect scheme execution feedback data to optimize the system model; specifically, it includes: The solution push unit is used to push the final optimized solution to each execution terminal through a standardized API interface; The execution monitoring unit is used to display the progress and status of the solution execution in real time on a visual dashboard; The feedback collection unit is used to collect actual effect data and feedback information during the implementation of the plan. The model optimization unit is used to update the weight coefficients of contextual elements and optimize the parameters of the decision model based on the execution feedback data.

[0044] Example 1 1. Implementation Scenario Setting The effectiveness of this invention was verified in an emergency equipment maintenance and support mission at a large manufacturing enterprise. The specific scenario was: a critical piece of equipment on the production line experienced a sudden malfunction, requiring the urgent allocation of spare parts and maintenance resources, with the requirement to restore production within 4 hours.

[0045] 2. Specific Implementation Process Step 1: Context Awareness and Data Acquisition Real-time acquisition of equipment sensor data (fault codes, location information); Obtain emergency repair orders from the production scheduling system; Receive environmental data (traffic conditions, weather conditions); The pre-trained scenario classification model identifies the following: task urgency (0.9 / 1.0), time constraint (4 hours), available maintenance teams (3 groups), and spare parts inventory status.

[0046] Step 2: Dynamic Construction of Knowledge Subgraph From the full knowledge graph containing 150,000 entities and 450,000 relationships; Dynamically extract relevant knowledge subgraphs based on contextual elements; Generate a dedicated knowledge space containing relevant spare parts, technical personnel, tools and equipment (containing only about 800 core entities).

[0047] Step 3: Intelligent Assessment and Solution Generation A multi-dimensional evaluation was conducted on the 32 available spare parts in the knowledge subgraph; Calculate the scenario adaptability of each spare part (range 0.65-0.92). Three alternative allocation plans were generated, including: Optimal solution: Adaptability 0.92, time taken 3.5 hours; Alternative Option A: Fit 0.85, Time 2.8 hours; Alternative Option B: Fit 0.78, Time Taken 4.2 Hours; Step 4: Virtual Execution and Dynamic Optimization Simulate the execution of the scheme in a digital twin environment; A path conflict was detected with the optimal solution (main road construction). The route was automatically corrected to a detour, and the adjusted plan had a suitability of 0.89, taking 3.8 hours. After two rounds of iterative verification, all conflicts were eliminated.

[0048] Step 5: Implementation and Feedback The final solution will be pushed to mobile devices; Actual results: 3.9 hours completed, 23% cost savings; Collect execution data to update the weights of contextual elements.

[0049] 3. The effects of the present invention and the prior art are compared in Table 1 and below. Figure 6 , Figure 7 As shown Table 1: Comparison of the effects of the present invention and the prior art 4. Effect Analysis Through the verification and comparative analysis of the above embodiments, the solution of the present invention demonstrates significant advantages in the following aspects: Significant efficiency improvements: The complete decision-making cycle from data collection to solution generation has been shortened from several hours using traditional methods to less than 10 minutes, reducing emergency response time by more than 50%.

[0050] Excellent decision-making quality: Through scenario adaptation and multi-dimensional evaluation, the accuracy of the solution reaches over 92%, and the first-time execution success rate is increased to 90% to 95%, significantly reducing the number of adjustments during the execution process.

[0051] Effective risk control: Virtual execution in a digital twin environment can identify and resolve more than 85% of potential conflicts in advance, minimizing operational risks.

[0052] Significant economic benefits: Overall operating costs have been reduced by 25% to 35%, and resource utilization has increased to around 90%, achieving the goals of refined management and cost reduction and efficiency improvement.

[0053] This embodiment fully demonstrates the effectiveness and superiority of the present invention in a real-world environment, providing reliable technical support for modern logistics management.

[0054] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects: This method achieves context-aware, precise decision-making and rapid response, significantly improving the agility and adaptability of logistical support. By constructing a dynamic context-aware model, it identifies and quantifies key contextual elements of a task in real time, overcoming the rigidity and environmental disconnect inherent in traditional systems. Its innovation lies in transforming the ambiguous task environment into calculable contextual indicators, enabling all subsequent steps to unfold based on this dynamic context. This mechanism ensures the system deeply understands implicit requirements such as task urgency and resource priority, thus achieving a leap from "static resource matching" to "dynamic context adaptation." Practical applications show that this method can increase the speed of solution generation in complex scenarios by approximately 5 times and improve the alignment of decision results with actual task requirements by more than 40%.

[0055] By dynamically focusing on knowledge subgraphs and using semantic reasoning, the problems of low efficiency and irrelevant redundancy in full graph queries have been overcome, greatly optimizing the utilization of computing resources and the accuracy of decision-making.

[0056] The innovation of this method lies in abandoning the extensive approach of traversing the entire knowledge graph. Instead, it employs a context-driven graph traversal algorithm to dynamically construct a highly concise knowledge subgraph strongly relevant to the current task as a dedicated decision space. This not only reduces the retrieval time in a graph with millions of nodes from minutes to milliseconds, but more importantly, it filters out irrelevant information, ensuring that evaluation and decision-making are based entirely on the most relevant knowledge fragments. This method fully leverages the semantic association characteristics of the knowledge graph, performing deep reasoning through calculations such as semantic path similarity, thereby increasing the accuracy of decision-making from approximately 65% ​​of traditional methods to over 90%.

[0057] This approach introduces a digital twin-driven pre-execution and iterative optimization mechanism, constructing a complete decision-making verification closed loop that fundamentally improves the feasibility of solutions and the system's risk control capabilities. The most forward-looking innovation lies in the addition of virtual execution and dynamic correction stages. By simulating the actual operation of the solution in a digital twin environment, approximately 80% of potential risks such as resource conflicts and spatiotemporal conflicts can be detected in advance, and iterative optimization can be automatically performed using conflict resolution algorithms. This realizes a new paradigm of "verification before execution," addressing solution defects that can only be discovered through trial and error in the real world under traditional models in the virtual space. This mechanism increases the first-time execution success rate of solutions from approximately 60% to over 95%, significantly reducing operational risks and cost losses caused by solution infeasibility, forming a complete closed loop from intelligent generation to reliable delivery.

[0058] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

[0059] This invention is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this specification. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. An intelligent logistics management method based on knowledge graphs, characterized in that, include: Real-time collection of multi-source operational data from the logistics system; identification of key contextual elements of the current logistics support task through a pre-trained contextual classification model. Based on the identified key contextual elements, knowledge subgraphs related to the current context are dynamically extracted from the pre-constructed logistics material knowledge graph; Based on the knowledge subgraph, available resources are evaluated in multiple dimensions, the context suitability of each resource is calculated, and a resource allocation plan is generated based on the evaluation results. The generated resource allocation plan is virtually executed in a digital twin environment. By simulating the actual execution process, potential resource conflicts and spatiotemporal constraint conflicts are detected. Based on the detection results, the plan is dynamically modified, and the final optimized plan is generated and pushed to the execution terminal.

2. The intelligent logistics management method based on knowledge graphs according to claim 1, characterized in that, The real-time acquisition of multi-source operational data from the logistics system, through a pre-trained context classification model, identifies key contextual elements of the current logistics support task, specifically including: Through a predefined data interface protocol, device status data is collected in real time from IoT sensor nodes, task order data is extracted periodically from the business system database, and environmental parameter data is continuously obtained from the environmental monitoring API interface, and the multi-source operation data is aggregated into a unified data stream. The unified data stream is cleaned and fused. A pre-trained context classification model is used to perform in-depth analysis on the fused data to extract temporal features that characterize the urgency of tasks, statistical features that reflect resource priorities, and topological features that describe spatiotemporal constraints. Based on the extracted temporal, statistical, and topological features, the key contextual elements of the current logistics support task are dynamically weighted and output through a multi-level attention mechanism in a pre-trained context classification model.

3. The intelligent logistics management method based on knowledge graphs according to claim 1, characterized in that, The step of dynamically extracting knowledge subgraphs related to the current context from a pre-constructed logistics material knowledge graph based on identified key contextual elements specifically includes: The quantitative indicators of the key context elements are mapped to the initial query node set in the knowledge graph, and different initial traversal weights are assigned to each context element type. Execute context-driven multi-path graph traversal, starting from the initial query node set, and expand the search along the semantic relationship edges and spatiotemporal relationship edges between material entities. During the traversal, dynamically adjust the depth and breadth of path exploration according to the relationship type and the semantic relevance of the current context elements. During the traversal, the overall correlation strength between each accessed material entity and the current context elements is calculated in real time, and entities with correlation strength exceeding a preset threshold and their relationships are filtered out. The selected entities and relationships with high correlation strength are integrated to construct a concise knowledge subgraph oriented towards the current task context. This knowledge subgraph serves as a dedicated knowledge space for subsequent multi-dimensional evaluation.

4. The intelligent logistics management method based on knowledge graphs according to claim 3, characterized in that, The process of mapping the quantitative indicators of key contextual elements to an initial set of query nodes in a knowledge graph, and assigning different initial traversal weights to each contextual element type, specifically includes: Define quantitative indicators for key contextual elements, including quantitative values ​​for task type and urgency. Quantification of time constraint urgency Quantitative value of resource availability sufficiency and the quantitative value of the severity of environmental conditions ; The above quantitative indicators are mapped to initial query nodes in the knowledge graph. This constitutes the initial query node set. ; Different initial traversal weights are assigned to each context element type. These weights can be determined based on historical data and expert experience. ,satisfy .

5. The intelligent logistics management method based on knowledge graphs according to claim 4, characterized in that, The execution context-driven multi-path graph traversal starts from the initial query node set and expands the search along the semantic and spatiotemporal relationship edges between material entities. During the traversal, the depth and breadth of path exploration are dynamically adjusted based on the relationship type and the semantic relevance of the current context elements. Specifically, this includes: In knowledge graph middle, For the collection of physical nodes of materials, This is the set of semantic and spatiotemporal relation edges between entities; starting from the initial query node set. Starting from the initial weights, initiate a multi-path graph traversal algorithm, such as an improved breadth-first search or depth-first search algorithm. During the traversal, for the currently visited node The depth and breadth of path exploration are dynamically adjusted based on the semantic and spatiotemporal relevance of the path to each initial query node; the semantic relevance can be obtained through a pre-trained semantic similarity calculation model. Represents a node With the initial query node The semantic similarity between nodes is used; the spatiotemporal relevance is calculated based on the spatiotemporal information of the resources represented by the nodes and the spatiotemporal constraints of the current task context. and Representing nodes respectively Spatial and temporal correlation; The edge weights are dynamically adjusted based on the semantic relevance of the relation type and the elements in the current context. It can be represented as: ,in: For the edge The original weights, To adjust the coefficient, For the edge The semantic relevance of the relationship type to the elements in the current context.

6. The intelligent logistics management method based on knowledge graphs according to claim 5, characterized in that, The process of calculating the comprehensive association strength between each accessed material entity and the current contextual elements in real time during the traversal, and filtering out entities with association strength exceeding a preset threshold and the relationships between them, specifically includes: For each accessed material entity node It calculates the overall correlation strength between itself and the elements of the current context in real time. The calculation formula is as follows: ,in: This represents the initial number of query nodes. For the initial query node The corresponding weights To balance the parameters of semantic relevance and spatiotemporal relevance, This is a balance parameter between spatial and temporal correlation. Comprehensive correlation strength With preset threshold If a comparison is made, If so, then retain the entity node. And its relationship edges with adjacent nodes.

7. The intelligent logistics management method based on knowledge graphs according to claim 6, characterized in that, The process involves integrating the selected highly correlated entities and relationships to construct a concise knowledge subgraph tailored to the current task context. This knowledge subgraph serves as a dedicated knowledge space for subsequent multi-dimensional evaluation and specifically includes: The selected highly correlated entity nodes and their relationship edges are integrated to construct a concise knowledge subgraph tailored to the current task context. ,in It is a set of highly correlated entity nodes selected from the pool. It is a set of relational edges between entity nodes; Once the knowledge subgraph is constructed, it serves as a dedicated knowledge space for subsequent multi-dimensional evaluation, providing targeted knowledge support for the generation of resource allocation plans.

8. The intelligent logistics management method based on knowledge graphs according to claim 1, characterized in that, Based on the knowledge subgraph, the available resources are evaluated from multiple dimensions, the context suitability of each resource is calculated, and a resource allocation plan is generated based on the evaluation results. Specifically, this includes: Within the dedicated knowledge space formed by the knowledge subgraph, for each available material entity, its evaluation values ​​in four dimensions—functional matching degree, spatiotemporal accessibility, cost-effectiveness, and risk coefficient—are calculated in parallel. Based on the dynamic weight configuration of the current key context elements, the evaluation values ​​of the four dimensions are weighted and fused to calculate the comprehensive context fit score of each material entity. All available material entities are sorted according to the comprehensive situational suitability score. Based on the sorting results, total resource constraints, and multiple task requirements, one or more optimal material allocation schemes are generated through the constraint satisfaction algorithm.

9. The intelligent logistics management method based on knowledge graphs according to claim 1, characterized in that, The process of virtually executing the generated resource allocation plan in a digital twin environment, detecting potential resource conflicts and spatiotemporal constraint conflicts by simulating the actual execution process, dynamically correcting the plan based on the detection results, generating a final optimized plan, and pushing it to the execution terminal, specifically includes: Based on the aforementioned material allocation plan and logistics material knowledge graph, a digital twin simulation environment is constructed that includes dynamic resource status, transportation network topology, and execution progress timeline. Using a discrete event simulation engine, the digital twin environment is driven to run according to the execution sequence set in the scheme. During the operation, one or more conflict types, such as resource allocation conflict, equipment usage time conflict and transportation route conflict, are monitored and recorded in real time. When a conflict is detected, the conflict resolution algorithm is triggered to search for feasible correction strategies based on the knowledge subgraph. The revised solution is re-injected into the digital twin environment for iterative verification until all conflicts are resolved. The final optimized solution is then generated and pushed to each execution terminal via API. At the same time, the solution revision trajectory and optimization basis are displayed in a visual dashboard.

10. An intelligent logistics management system based on knowledge graphs, characterized in that, An intelligent logistics management method based on knowledge graphs, as described in any one of claims 1-9, comprises: The multi-source data acquisition and context awareness module is used to collect multi-source operational data of the logistics system in real time and identify key context elements of the current logistics support task through a pre-trained context classification model. The knowledge graph management and dynamic query module is used to dynamically extract knowledge subgraphs related to the current context from the pre-built logistics material knowledge graph based on the identified key context elements. The multi-dimensional assessment and solution generation module is used to conduct multi-dimensional assessments of available resources based on the knowledge subgraph, calculate the context suitability of each resource, and generate resource allocation plans based on the assessment results. The digital twin and solution optimization module is used to virtually execute the generated material allocation plan in a digital twin environment. By simulating the actual execution process, it detects potential resource conflicts and spatiotemporal constraint conflicts, and dynamically corrects the plan based on the detection results to generate the final optimized plan. The scheme execution and feedback optimization module is used to push the final optimized scheme to the execution terminal and collect scheme execution feedback data to optimize the system model.

Images