An AI model-based logistics intelligent decision method and system

Abstract

The application relates to the technical field of intelligent logistics decision-making, and discloses an AI model-based intelligent logistics decision-making method and system.The method comprises the following steps: acquiring full-link logistics data, performing hierarchical desensitization fusion on the full-link logistics data to obtain target logistics data graphs; extracting logistics space-time features of the target logistics data graphs, performing feature learning on a preset decision expert model according to the logistics space-time features, and obtaining multiple decision submodels; performing multi-scenario simulation state analysis on the full-link logistics data by using the decision submodels, performing model optimization on the decision submodels according to the results of the multi-scenario simulation state analysis, and obtaining a target logistics decision model; performing logistics trajectory modeling on the full-link logistics data by using the target logistics decision model, and obtaining a target decision trajectory; and performing decision analysis on the full-link logistics data according to the target decision trajectory, and obtaining multiple logistics decision information.The application can realize intelligent logistics decision-making and improve the accuracy of logistics decision-making.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics decision-making technology, and in particular to an intelligent logistics decision-making method and system based on an AI model. Background Technology

[0002] In the current era of globalized e-commerce development, existing logistics decision-making methods fail to anonymize sensitive information such as customer privacy and trade secrets in the end-to-end logistics data, which can easily lead to data security and privacy leaks, thereby reducing the accuracy of subsequent logistics decisions.

[0003] Furthermore, traditional logistics decision-making models are mostly built based on fixed rules and preset parameters, making it difficult to adjust and optimize them according to changes in actual scenarios. This results in their inability to effectively adapt to the needs of multiple scenarios. For example, when faced with traffic congestion or severe weather, traditional logistics decision-making models cannot promptly replan logistics routes and delivery plans, leading to low reliability of logistics trajectory prediction and consequently reduced efficiency in logistics decision-making.

[0004] Therefore, how to achieve intelligent logistics decision-making and improve the accuracy of logistics decisions has become an urgent problem to be solved. Summary of the Invention

[0005] This invention provides an AI-based intelligent decision-making method for logistics, the main purpose of which is to solve the problems of low intelligence and low accuracy in logistics decision-making.

[0006] Firstly, to achieve the above objectives, the present invention provides a logistics intelligent decision-making method based on an AI model, comprising:

[0007] Acquire end-to-end logistics data, perform layered desensitization and fusion on the end-to-end logistics data, and obtain the target logistics data map;

[0008] Extract the spatiotemporal features of the target logistics data map, and perform feature learning on the preset decision expert model based on the spatiotemporal features of the logistics to obtain multiple decision sub-models;

[0009] The decision sub-models are used to perform multi-scenario simulation state analysis on the full-link logistics data, and the decision sub-models are optimized based on the results of the multi-scenario simulation state analysis to obtain the target logistics decision model.

[0010] The target logistics decision model is used to model the logistics trajectory of the entire logistics chain data to obtain the target decision trajectory.

[0011] Based on the target decision trajectory, the entire logistics data is analyzed to obtain multiple logistics decision information.

[0012] Secondly, the present invention also provides a logistics intelligent decision-making system based on an AI model, the system comprising:

[0013] The data anonymization and fusion module is used to acquire end-to-end logistics data, perform hierarchical anonymization and fusion on the end-to-end logistics data, and obtain a target logistics data map;

[0014] The model feature learning module is used to extract the spatiotemporal features of the target logistics data map, and to perform feature learning on the preset decision expert model based on the spatiotemporal features of the logistics to obtain multiple decision sub-models;

[0015] The scenario simulation analysis module is used to perform multi-scenario simulation state analysis on the full-link logistics data using multiple decision sub-models, and to optimize the decision sub-models based on the results of the multi-scenario simulation state analysis to obtain the target logistics decision model.

[0016] The logistics trajectory modeling module is used to model the logistics trajectory of the entire-link logistics data using the target logistics decision model to obtain the target decision trajectory.

[0017] The logistics decision analysis module is used to perform decision analysis on the entire logistics data based on the target decision trajectory to obtain multiple logistics decision information.

[0018] In this embodiment of the invention, the layered desensitization technology avoids the problems of overprotection or information leakage caused by the traditional "one-size-fits-all" desensitization through a hierarchical processing mechanism, significantly improving the compliance of logistics data; it constructs a structured target logistics data map, integrating the scattered layered desensitized logistics data into a logically coherent map structure, greatly improving data utilization efficiency; based on map structure analysis and multi-dimensional information decomposition technology, it can automatically extract complex features in both time and space dimensions, and transform discrete data into a structured feature set through spatiotemporal coupling modeling, solving the problems of fragmented spatiotemporal data and weak correlation in traditional logistics systems.

[0019] Secondly, by dynamically grouping spatiotemporal features according to core decision dimensions and training independent decision sub-models for each group, the complexity and computational cost of the models are significantly reduced, improving the decision accuracy of the sub-models in specific business scenarios and greatly enhancing the automation and efficiency of logistics decision-making. Through parallel processing of data simulations in multiple scenarios (such as urban delivery, cross-regional transportation, and extreme weather response), the efficiency of logistics decision analysis is significantly improved. Model optimization based on simulation results employs dynamic parameter adjustment and structural pruning techniques, significantly reducing computational resource consumption while ensuring decision accuracy and avoiding overfitting issues caused by traditional single-scenario training. The target logistics decision model processes end-to-end logistics data in real time and rapidly completes trajectory modeling through distributed parallel computing, effectively solving the problem that traditional static planning is ill-suited to dynamic scenarios. By driving the decision analysis of end-to-end logistics data through the target decision trajectory, tedious manual screening and calculation are eliminated, greatly shortening the decision analysis cycle and generating multiple highly accurate and reliable logistics decision information sets, significantly improving the intelligence and accuracy of logistics decision-making. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating an AI-based intelligent logistics decision-making method according to an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of a process for hierarchical desensitization and fusion of the entire logistics data according to an embodiment of the present invention;

[0023] Figure 3 This is a schematic diagram illustrating the process of using the target logistics decision model to model the logistics trajectory of the entire logistics chain data, as provided in an embodiment of the present invention.

[0024] Figure 4 A functional block diagram of an AI-based intelligent logistics decision-making system provided in an embodiment of the present invention;

[0025] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0026] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.

[0027] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] This application provides an AI-based intelligent logistics decision-making method, which can be executed by software or hardware installed on terminal devices or server-side devices. The server-side includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be a standalone server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0029] Reference Figure 1 The diagram shown is a flowchart illustrating an AI-based intelligent logistics decision-making method according to an embodiment of the present invention. In this embodiment, the AI-based intelligent logistics decision-making method includes:

[0030] S1. Obtain full-chain logistics data, perform layered desensitization and fusion on the full-chain logistics data, and obtain the target logistics data map.

[0031] In this embodiment of the invention, the full-link logistics data refers to data generated at all stages covering the entire logistics business process (from order generation to final delivery), including but not limited to order data, transportation data, warehousing data, and delivery data; the layered desensitization and fusion is a technology for multi-level processing of full-link logistics data, including data desensitization (removal of sensitive information) and data fusion (integration of multi-source data) to generate a target logistics data map, thereby realizing logistics data visualization and improving the efficiency and transparency of subsequent intelligent decision-making.

[0032] The order data includes customer order information, product details, delivery requirements, etc.; the transportation data includes vehicle trajectory, transportation time, route planning, abnormal events (such as delays, accidents), etc.; the warehousing data includes inventory, inbound and outbound records, warehouse location, shelf management, etc.; and the delivery data includes delivery information, logistics receipt status, etc.

[0033] like Figure 2 As shown in this embodiment of the invention, the step of performing hierarchical desensitization and fusion of the end-to-end logistics data to obtain the target logistics data map includes:

[0034] S11. Standardize the format of the entire logistics data to obtain standardized logistics data;

[0035] S12. Identify the sensitivity levels of each data in the standardized logistics data;

[0036] S13. Divide the standardized logistics data into multiple hierarchical sensitive logistics data of different levels according to the data sensitivity level;

[0037] S14. Perform data layering and desensitization on multiple hierarchical sensitive logistics data to obtain layered desensitized logistics data;

[0038] S15. Identify multiple hierarchical logistics data entities in the hierarchical desensitized logistics data, and perform logistics association analysis on the multiple hierarchical logistics data entities to obtain logistics data association relationships;

[0039] S16. Using the multiple hierarchical logistics data entities as nodes and the logistics data relationships as edges, generate an initial logistics data map of the full-link logistics data based on the nodes and the edges;

[0040] S17. Perform map optimization and verification on the initial logistics data map to obtain the target logistics data map.

[0041] In this embodiment of the invention, the format standardization process includes data cleaning and format unification. Data cleaning includes removing duplicate data, correcting erroneous data, and filling in missing values. By comparing the unique identifiers of the entire logistics data (such as order numbers and tracking numbers), completely duplicate logistics data entries are deleted. Logically contradictory logistics data (such as "delivery time is earlier than order time") can also be identified. Through manual verification or automatic correction by the rule engine, non-critical missing fields (such as customer remarks) are filled with default values ​​(such as "none"). The format unification includes unifying the date and time format, that is, converting all time fields to the "year-month-day hour:minute" format (such as "2025-10-10 14:00"), and unifying fields such as weight and volume to international units (such as kilograms and cubic meters), thereby generating standardized logistics data with consistent structure and high readability.

[0042] In detail, it identifies fields in standardized logistics data that contain personal privacy (such as name, phone number, address), trade secrets (such as supplier quotations, inventory quantity), or compliance requirements (such as ID card number, bank card number) through keyword matching or regular expressions. Optionally, it can be combined with business scenarios to make judgments. For example, customer addresses are sensitive data in ordinary logistics data, but in cross-border logistics, the sensitivity level may be reduced due to customs clearance requirements.

[0043] The data sensitivity level can be divided into "high sensitivity" (such as ID card number), "medium sensitivity" (such as phone number), and "low sensitivity" (such as order product name). Specifically, the data sensitivity level is automatically marked by a rule engine (such as Drools) or a machine learning model (such as a classification model based on historical labeled data) to provide a basis for subsequent desensitization.

[0044] Furthermore, logistics data of “high sensitivity”, “medium sensitivity” and “low sensitivity” are grouped according to data sensitivity level and stored in different datasets or database tables. Optionally, high sensitivity logistics data is stored in an encrypted database, while medium and low sensitivity logistics data is stored in a regular database to generate hierarchical sensitive logistics data divided by sensitivity level, which facilitates targeted processing.

[0045] Specifically, highly sensitive logistics data is subjected to strong desensitization (such as completely replacing it with virtual values), moderately sensitive logistics data is subjected to data encryption (such as masking or generalization), and low-sensitivity logistics data is subjected to data masking (retaining the original value or only performing format conversion). This process generates layered desensitized logistics data to ensure that sensitive information is irreversible while maintaining the availability of the layered desensitized logistics data.

[0046] In detail, layered logistics data entities (such as orders, vehicles, warehouses, and customers) are extracted from the layered and anonymized logistics data. For example, order number formats are matched using regular expressions (such as "ORD-20251010-001"), or entity names are identified from text fields using NLP technology (such as "Guangzhou Distribution Center"). Relationships between layered logistics data entities are established based on business logic. For example, "Order A" is associated with "Transport Order B" (by matching the order number with the transport order number), and "Vehicle C" is associated with "Warehouse D" (by the spatiotemporal overlap of vehicle GPS tracks and warehouse locations, thus clarifying the relationships between layered logistics data entities (such as "order-transport order" and "vehicle-warehouse"), providing a structural foundation for the subsequent construction of logistics data maps.

[0047] Furthermore, multiple hierarchical logistics data entities are used as nodes, and logistics data relationships are used as edges. The nodes and edges are imported into a graph database (such as Neo4j). For example, node types include "order", "vehicle", and "warehouse", and edge types include "transportation", "storage", and "belongs to". Attributes are added to the nodes and edges. That is, the "order" node contains attributes such as "order time" and "product name", and the "transportation" edge contains attributes such as "start time" and "end time".

[0048] Specifically, a graphical map, namely the initial logistics data map, is automatically generated using visualization tools for graph databases (such as Neo4j Browser). The initial logistics data map graphically displays entities and their relationships in the entire logistics data chain.

[0049] In detail, the initial logistics data graph undergoes graph optimization and verification, including integrity verification, accuracy verification, and consistency verification. Integrity verification verifies whether the hierarchical logistics data entities and relationships cover the entire business chain (e.g., whether all orders are associated with shipping orders). Accuracy verification ensures that the hierarchical logistics data entities and relationships are error-free by sampling and comparing the entire logistics data chain with the graph data. Consistency verification checks whether the attributes of the same hierarchical logistics data entity in the initial logistics data graph are consistent (e.g., whether the delivery time of order "A" is the same across all associated edges). Based on the results of the graph optimization and verification, the hierarchical logistics data entities, relationships, or attributes are adjusted to generate an accurate, complete, and efficient target logistics data graph, which can be directly used for logistics analysis, decision support, and other scenarios.

[0050] In this embodiment of the invention, the step of performing data layering and desensitization on multiple layers of sensitive logistics data to obtain layered desensitized logistics data includes:

[0051] Data replacement and desensitization are performed on highly sensitive logistics data from multiple hierarchical sensitive logistics data to obtain highly desensitized logistics data;

[0052] Data encryption and desensitization are performed on the medium-sensitive logistics data among the multiple hierarchical sensitive logistics data to obtain medium-desensitized logistics data;

[0053] Data masking and desensitization are performed on the low-sensitivity logistics data from the multiple hierarchical sensitive logistics data to obtain low-sensitivity logistics data;

[0054] The highly anonymized logistics data, the moderately anonymized logistics data, and the lowly anonymized logistics data are aggregated to obtain stratified anonymized logistics data.

[0055] In this embodiment of the invention, the data replacement and desensitization refers to the use of complete virtualization to replace highly sensitive logistics data (such as ID card number, bank card number, and real name), thereby completely replacing the highly sensitive logistics data with virtual values ​​that have no real meaning, ensuring that the original information cannot be reversed and restored.

[0056] In detail, the data encryption and desensitization includes symmetric encryption and asymmetric encryption. The same key is used to symmetrically encrypt medium-sensitive logistics data (such as customer phone numbers and order amounts), while asymmetric encryption algorithms such as public key encryption and private key decryption are used to encrypt medium-sensitive logistics data with extremely high security requirements (such as internal contract numbers of enterprises), ensuring that only authorized users can restore the data, thereby improving data security and availability.

[0057] The data masking and desensitization refers to partially hiding low-sensitivity logistics data (such as address and age). For example, it ensures that the masked low-sensitivity logistics data still conforms to the business format, thus protecting data privacy while retaining the statistical value of the data.

[0058] Furthermore, the highly anonymized, moderately anonymized, and lowly anonymized logistics data are aggregated to ensure consistency in field naming, units, and date formats. Optionally, a real-time anonymization interface is provided for the aggregated hierarchical anonymized logistics data, dynamically returning logistics data at different anonymization levels based on the data sensitivity level to meet the data security and availability requirements in different scenarios and improve data anonymization efficiency.

[0059] In this embodiment of the invention, the layered desensitization technology achieves a balance between data security and business needs through a differentiated strategy. The hierarchical processing mechanism avoids the problems of overprotection or information leakage caused by the traditional "one-size-fits-all" desensitization, significantly improving the compliance of logistics data. The desensitization fusion technology constructs a structured target logistics data map through data association and consistency verification, integrating the scattered layered desensitized logistics data into a logically coherent map structure, greatly improving data utilization efficiency, reducing manual intervention, and providing a reliable data foundation for intelligent decision-making in the logistics industry.

[0060] S2. Extract the spatiotemporal features of the target logistics data map, and perform feature learning on the preset decision expert model based on the spatiotemporal features of the logistics to obtain multiple decision sub-models.

[0061] In this embodiment of the invention, the spatiotemporal characteristics of logistics include time-dimensional characteristics and spatial-dimensional characteristics. The time-dimensional characteristics include timestamps of logistics events, periodic patterns (such as daily delivery peaks), and time-series dependencies (such as the time from order generation to delivery). The spatial-dimensional characteristics include logistics spatial coordinates, regional divisions (such as warehouse coverage areas), and spatial interaction relationships (such as cross-regional transfers).

[0062] The decision expert model is an intelligent decision-making framework built on knowledge and experience rules in the logistics field. It is used to solve complex decision-making problems in logistics scenarios. The decision expert model integrates the experience of logistics industry experts, business rules (such as delivery priority and inventory threshold) and optimization algorithms (such as route planning and resource scheduling). Through feature learning mechanism and by combining real-time logistics data, it adjusts decision-making strategies to obtain multiple decision sub-models, thereby improving the interpretability of the decision model and reducing computational complexity.

[0063] In this embodiment of the invention, extracting the spatiotemporal features of the target logistics data map includes:

[0064] The target logistics data map is decomposed into time dimension information and spatial dimension information respectively to obtain the time attribute data and spatial attribute data corresponding to each node in the target logistics data map;

[0065] Calculate the temporal dimension features of the temporal attribute data and the spatial dimension features of the spatial attribute data;

[0066] The time dimension features are associated and matched with the spatial dimension features to obtain the spatiotemporal coupling features;

[0067] The spatiotemporal coupling features are reduced in dimensionality to obtain the spatiotemporal features of the target logistics data map.

[0068] In this embodiment of the invention, each node (such as order, warehouse, vehicle) in the target logistics data graph is traversed using a graph database query language (such as Cypher), the node type and associated edges (such as transportation relationship, time sequence) are identified, and time-related fields, such as order generation time, vehicle arrival time, and warehouse operation timestamp, are extracted from the meta-logistics data of the nodes or edges to form a time attribute dataset; geographical coordinates (such as latitude and longitude), regional codes (such as administrative divisions), and spatial movement trajectories (such as GPS sequences) are extracted to construct a spatial attribute dataset.

[0069] In detail, the time feature refers to identifying periodic patterns (such as daily delivery peaks, seasonal fluctuations) in the time attribute dataset and calculating the data interval of each time attribute in the time attribute dataset (such as the average time from order generation to delivery); the spatial feature refers to identifying high-frequency activity areas (such as popular delivery areas) or abnormal spatial points (such as remote warehouses) in the spatial attribute dataset based on spatial coordinates or regional coding.

[0070] Specifically, temporal features (such as the time of an event) and spatial features (such as the location of an event) are associated through node IDs or time windows. For example, "an order was shipped from warehouse A (coordinates X, Y) at 10:00 AM," generating spatiotemporal coupled features. Highly correlated spatiotemporal coupled features (such as "order volume" and "number of delivery vehicles" may be strongly correlated) are removed through correlation analysis (such as mutual information). Alternatively, unsupervised learning (such as autoencoders) or statistical methods (such as the idea of ​​principal component analysis) are used to retain the spatiotemporal coupled features that have the greatest impact on decision-making (such as "spatiotemporal heat value" which comprehensively reflects the activity of a region and time). This ensures that the dimensionality-reduced features can still be mapped to specific business scenarios (such as the combination of "peak hours + popular areas" replacing the original multidimensional data). Finally, high-dimensional spatiotemporal features of logistics with business interpretability are obtained, providing efficient input for subsequent decision-making models.

[0071] In this embodiment of the invention, the step of performing feature learning on a preset decision expert model based on the spatiotemporal characteristics of logistics to obtain multiple decision sub-models includes:

[0072] Obtain the core decision dimensions of the decision expert model, and construct a feature dimension mapping table based on the spatiotemporal characteristics of logistics and the core decision dimensions;

[0073] Based on the feature dimension mapping table, the spatiotemporal features of the logistics are grouped to obtain multi-dimensional feature data;

[0074] Based on the multi-dimensional feature data, the decision expert model is subjected to group feature learning to obtain multiple group decision models;

[0075] Model validation was performed on multiple of the aforementioned grouping decision models to obtain the model validation percentage error.

[0076] Determine whether the model validation percentage error is greater than a preset validation error threshold;

[0077] If the model validation percentage error is greater than the validation error threshold, the grouped decision model is fine-tuned to obtain multiple decision sub-models;

[0078] If the model validation percentage error is less than or equal to the validation error threshold, then the multiple grouped decision models are used as decision sub-models.

[0079] In this embodiment of the invention, the core decision-making dimensions include cost optimization, timeliness assurance, and resource utilization. The spatiotemporal characteristics of logistics (such as "peak-hour delivery heat value" and "cross-regional transportation time") are associated with the core decision-making dimensions, and the correspondence between the spatiotemporal characteristics of logistics and the core decision-making dimensions is recorded in tabular form to obtain a feature dimension mapping table.

[0080] In detail, based on the dimensional classification in the feature dimension mapping table, the spatiotemporal features of logistics are divided into multiple groups; for example, the cost group includes spatiotemporal features of logistics such as "transportation distance" and "warehouse space occupancy rate", and the timeliness group includes spatiotemporal features of logistics such as "delivery delay rate" and "transfer dwell time". The spatiotemporal features of the same group are encapsulated into a structured dataset (such as JSON format), i.e., multi-dimensional feature data, to ensure that each group of multi-dimensional feature data only contains features related to a specific decision dimension.

[0081] Furthermore, an independent decision expert model is initialized for each multi-dimensional feature data. The multi-dimensional feature data of the corresponding group is input into the decision expert model and trained through supervised learning (such as "optimal cost solution" in labeled historical data) or reinforcement learning (such as simulated scheduling strategy feedback) to obtain multiple group decision models. The reserved validation dataset (such as the untrained part in historical logistics data) is used to predict each group decision model, and the deviation between the prediction result and the actual value is calculated to obtain the model validation percentage error (such as "cost model error is 15%)". The model validation percentage error is compared with the preset validation error threshold (such as "10%)" to determine whether the group decision model meets the accuracy requirements.

[0082] Specifically, if the model validation percentage error is greater than the validation error threshold, the analysis is conducted to determine whether the large error in the group decision model is due to missing features (such as not considering weather factors). If so, relevant features can be added and the model retrained to adjust the model hyperparameters (such as learning rate and regularization coefficient) or to change the algorithm (such as switching from decision tree to neural network). If the model validation percentage error is within the validation error threshold (such as "8%)", the current group decision model is directly solidified into a decision sub-model, and its applicable scenario is recorded (such as "urban delivery timeliness sub-model"). Each decision sub-model independently solves the logistics decision problem of a specific decision dimension, and can work collaboratively through an integration mechanism (such as the joint optimization of scheduling schemes by cost and timeliness sub-models).

[0083] In this embodiment of the invention, based on graph structure analysis and multi-dimensional information decomposition technology, complex features in both time and space dimensions can be automatically extracted. By transforming discrete data into a structured feature set through spatiotemporal coupling modeling, the problem of fragmented spatiotemporal data and weak correlation in traditional logistics systems is solved. Secondly, the spatiotemporal features are dynamically grouped according to the core decision dimensions, and an independent decision sub-model is trained for each group of features. This significantly reduces model complexity and computational overhead, improves the decision accuracy of the decision sub-model in specific business scenarios, and greatly enhances the automation level and efficiency of logistics decision-making.

[0084] S3. Utilize multiple decision sub-models to perform multi-scenario simulation state analysis on the end-to-end logistics data, and optimize the decision sub-models based on the results of the multi-scenario simulation state analysis to obtain the target logistics decision model.

[0085] In this embodiment of the invention, the multi-scenario simulation state analysis refers to reproducing the operational state of the entire logistics data under different conditions (such as seasonal fluctuations and emergencies) in a virtual simulation environment, and evaluating the performance of the decision sub-model.

[0086] In this embodiment of the invention, the step of using multiple decision sub-models to perform multi-scenario simulation state analysis on the end-to-end logistics data includes:

[0087] The end-to-end logistics data is adapted to the logistics simulation scenarios corresponding to multiple decision sub-models to obtain the adapted logistics dataset corresponding to the logistics simulation scenarios.

[0088] Based on the decision sub-model, a single-scenario simulation analysis is performed on the adapted logistics dataset to obtain the single-scenario simulation decision results.

[0089] The logistics simulation scenarios are cross-combined to generate cross-simulation scenarios;

[0090] In the cross-simulation scenario, the decision sub-model is used to perform multi-scenario simulation analysis on the adapted logistics dataset to obtain multi-scenario simulation decision results;

[0091] The consistency of the single-scenario simulation decision results and the multi-scenario simulation decision results is verified to obtain valid simulation decision results.

[0092] In this embodiment of the invention, for each decision sub-model corresponding to a logistics simulation scenario (such as "urban peak-hour delivery" or "cross-regional cold chain transportation"), its core constraints (such as time windows and temperature thresholds) are extracted. Based on the core constraints, data segments that meet the core constraints are selected from the full-link logistics data (such as retaining only order records during peak hours). The original full-link logistics data is then converted into a scenario-adaptive format through feature engineering (such as converting geographical coordinates into gridded heat values) to obtain an adapted logistics dataset.

[0093] In detail, the adapted logistics dataset is input into the corresponding decision sub-model (e.g., the "timeliness sub-model" only receives time-related feature data) to simulate the logistics operation status in this scenario (e.g., predicting the probability of delivery delay). The decision sub-model outputs single-scenario decision suggestions based on the learned rules or patterns (e.g., "suggest increasing delivery resources during peak hours") and records key indicators (e.g., cost increment, timeliness achievement rate), which are the single-scenario simulation decision results.

[0094] Furthermore, the logistics simulation scenario is broken down into independent dimensions (such as time, space, and resource type), and optional values ​​for each dimension are defined (such as the time dimension including "weekdays," "weekends," and "holidays"). Cross-simulation scenarios are generated by permutation and combination (such as "weekends + city center + cold chain transportation"), while ensuring that the combination meets business logic (such as excluding high-temperature periods for cold chain transportation). In the cross-simulation scenario, multiple decision sub-models are called in parallel (such as simultaneously simulating the decision-making of "cost sub-model" and "timeliness sub-model" under resource competition scenarios), and the output conflicts of each decision sub-model are recorded (such as the cost model suggesting reducing vehicles, and the timeliness model suggesting increasing vehicles). An arbitration mechanism (such as weighted voting or priority rules) is introduced to resolve output conflicts, thereby generating a comprehensive decision that takes into account multiple objectives (such as "increasing some vehicles within the controllable cost range to ensure timeliness"), i.e., multi-scenario simulation decision results.

[0095] The process involves comparing common dimensions (such as resource allocation) in single-scenario and multi-scenario simulation decision results, checking for logical contradictions (such as suggesting increasing vehicles in a single scenario while suggesting decreasing them in a multi-scenario scenario), retaining validated decision results (such as both single-scenario and multi-scenario scenarios supporting "increase 20% of vehicles"), and marking or correcting abnormal results (such as extreme suggestions due to insufficient data in multi-scenario scenarios) to generate a valid simulation decision set.

[0096] In this embodiment of the invention, the step of optimizing the decision sub-model based on the results of multi-scenario simulation state analysis to obtain the target logistics decision model includes:

[0097] Based on the results of multi-scenario simulation state analysis, the parameters of the decision sub-model to be optimized are determined, and a parameter optimization strategy is generated based on the parameters of the model to be optimized.

[0098] The decision sub-model is initially optimized according to the parameter optimization strategy to obtain the initial optimized decision model;

[0099] Based on the initial optimization decision model, candidate pruning objects and pruning granularity control parameters are generated;

[0100] Obtain the edge weights of the candidate pruning objects, and generate a weight importance score for each candidate pruning object based on the edge weights;

[0101] The target pruning object is selected from the candidate pruning objects based on the weighted importance score;

[0102] The target pruning object corresponding to the initial optimization decision model is deleted according to the pruning granularity control parameters to obtain the target logistics decision model.

[0103] In this embodiment of the invention, performance indicators (such as prediction accuracy and decision stability) of the decision sub-model are extracted from the results of multi-scenario simulation state analysis. Scenarios that do not meet expectations (such as large timeliness prediction deviation under extreme weather conditions) are marked. Feature importance analysis is used to locate the model parameters to be optimized that cause the model performance to decline (such as weights in neural networks and splitting thresholds in decision trees). Based on the type of model parameters to be optimized (such as continuous value parameters and discrete rule parameters) and the optimization objective (such as improving accuracy and reducing computation time), an appropriate optimization method (such as grid search and Bayesian optimization) is selected to generate a parameter optimization strategy.

[0104] Specifically, according to the parameter range and step size in the parameter optimization strategy, the parameters to be optimized in the decision sub-model are gradually modified (such as increasing the number of hidden layer nodes in the neural network and adjusting the maximum depth of the decision tree). The adjusted model is tested on the validation dataset, and the changes in key indicators (such as accuracy improvement and overfitting risk) are recorded. If the model performance does not improve significantly or degrades in continuous iterations, the adjustment is terminated, the current optimal parameter combination is retained, and the initial optimized decision model is generated.

[0105] Furthermore, the initial optimization decision model is converted into an interpretable structure (such as branch nodes in a decision tree or connection edges in a neural network). Potential pruning objects (such as nodes in a decision tree with fewer than a threshold number of covered samples) are marked according to preset rules (such as branches used infrequently or connection edges with weights close to zero). Based on the model complexity and business requirements, the coarseness of pruning (such as pruning by layer or pruning by a single node) is set, and pruning granularity control parameters are generated.

[0106] In detail, edge weights are converted into weight importance scores through normalization (e.g., mapping weight values ​​to intervals). Weight importance thresholds are set according to the model compression objective (e.g., reducing the number of parameters) (e.g., retaining the top 70% of edges by importance). Candidate pruning objects with importance below the threshold are filtered out. At the same time, it is verified whether the remaining objects meet business constraints (e.g., the model must still be able to cover all logistics scenarios after pruning). Target pruning objects are deleted according to the pruning granularity control parameters (e.g., removing the lowest weight connection edge in the neural network, or merging similar branches in the decision tree). The pruned model is then retrained in a lightweight manner (e.g., adjusting the remaining parameters in a few iterations) to compensate for the information loss caused by pruning, confirm that it can meet business requirements, and thus improve the accuracy of logistics decisions.

[0107] In this embodiment of the invention, data simulation under different logistics scenarios (such as urban delivery, cross-regional transportation, and extreme weather response) is processed in parallel through multi-scenario simulation states, which significantly improves the analysis efficiency of logistics decision-making. The model optimization based on simulation results adopts dynamic parameter adjustment and structural pruning techniques, which greatly reduces the consumption of computing resources while ensuring the accuracy of decision-making, and ensures that the optimized model is robust under different business scenarios, avoiding the overfitting problem caused by traditional single-scenario training.

[0108] S4. Use the target logistics decision model to model the logistics trajectory of the entire logistics chain data to obtain the target decision trajectory.

[0109] In this embodiment of the invention, the logistics trajectory modeling refers to the process of performing spatiotemporal sequence analysis on the entire logistics data based on the target logistics decision model. By extracting key features such as timestamps and spatial coordinates from the entire logistics data, the movement path of logistics objects (such as vehicles and goods) is constructed. The path is then dynamically corrected by combining business rules (such as route compliance and load limits) and real-time influencing factors (such as congestion and weather) to form an executable target decision trajectory, providing accurate decision-making basis for transportation scheduling, resource allocation and other links.

[0110] like Figure 3 As shown in this embodiment of the invention, the step of using the target logistics decision model to model the logistics trajectory of the entire-link logistics data to obtain the target decision trajectory includes:

[0111] S41. Extract the trajectory association dataset from the full-link logistics data, and obtain the timestamp and spatial coordinates of the trajectory association dataset;

[0112] S42. Generate the initial logistics trajectory of the full-link logistics data based on the timestamp and the spatial coordinates;

[0113] S43. Use the target logistics decision model to perform multi-dimensional trajectory analysis on the initial logistics trajectory to obtain multi-dimensional trajectory influence parameters;

[0114] S44. Correct the trajectory anomaly of the initial logistics trajectory according to the multi-dimensional trajectory influence parameters to obtain the corrected logistics trajectory;

[0115] S45. Based on the modified logistics trajectory, perform logistics feasibility verification on the full-link logistics data, and optimize the modified logistics trajectory according to the results of the logistics feasibility verification to obtain the target decision trajectory.

[0116] In this embodiment of the invention, data directly related to the logistics trajectory (such as order transportation records, vehicle GPS positioning, and warehouse entry and exit times) are filtered from the full-chain logistics data, and irrelevant information (such as customer contact information) is removed. Through data parsing technology, the timestamps (such as order delivery time and vehicle arrival time) and spatial coordinates (such as latitude and longitude or gridded location codes) of each trajectory-related data in the trajectory-related dataset are extracted to form a trajectory-related dataset containing spatiotemporal dimensions, providing a foundation for subsequent trajectory construction.

[0117] In detail, spatial coordinates are sorted according to timestamps, and discrete location points are connected in chronological order to form an initial logistics trajectory. For missing or abnormal spatiotemporal data (such as coordinate breaks caused by GPS signal loss), interpolation algorithms (such as linear interpolation and prediction interpolation based on historical patterns) are used to complete the trajectory to ensure the continuity of the initial logistics trajectory.

[0118] Specifically, the initial logistics trajectory is decomposed into multiple dimensions using a target logistics decision model to analyze key factors affecting the logistics trajectory (such as time delays caused by traffic congestion, reduced loading and unloading efficiency caused by warehouse congestion, and route changes caused by abnormal weather). The target logistics decision model quantifies the degree of influence of each factor on the trajectory by comparing historical data with real-time logistics information (such as congestion increasing the time of a certain route by 30%), and generates trajectory impact parameters that include multiple dimensions such as time, cost, and risk.

[0119] Furthermore, based on multi-dimensional trajectory impact parameters, abnormal segments in the initial logistics trajectory (such as segments where actual travel time far exceeds prediction) are marked and corrected. For example, if delays are caused by traffic congestion, the model can suggest switching to an alternative route; if loading and unloading delays are caused by warehousing issues, subsequent transportation schedules can be adjusted to generate a corrected logistics trajectory that better reflects actual operating conditions. The corrected logistics trajectory is then substituted into the full-chain logistics data to verify whether it meets all business constraints (such as vehicle load limits, timeliness contract requirements, and route compliance).

[0120] If the verification fails (e.g., a certain route exceeds the vehicle's range), the trajectory is adjusted and corrected (e.g., by adding a transfer charging point). Through multiple iterations and optimizations, a target decision trajectory that meets both actual operating conditions and optimized decision objectives (e.g., lowest cost, highest timeliness) is finally generated as the final basis for logistics execution.

[0121] In this embodiment of the invention, the entire logistics data is processed in real time according to the target logistics decision model, and trajectory modeling is completed quickly through distributed parallel computing. This significantly shortens the time consumption of traditional serial processing, meets the stringent real-time requirements of the logistics industry, effectively solves the problem that traditional static planning is difficult to adapt to dynamic scenarios, and improves the efficiency and accuracy of subsequent logistics decisions.

[0122] S5. Based on the target decision trajectory, perform decision analysis on the full-link logistics data to obtain multiple logistics decision information.

[0123] In this embodiment of the invention, the decision analysis refers to the process of making decisions and scheduling the entire logistics data based on the target decision trajectory; the logistics decision information includes transportation scheduling instructions (such as vehicle dispatch time and route selection), warehousing operation suggestions (such as inventory allocation and loading and unloading sequence), and risk warnings (such as delay prediction and abnormal intervention), which can provide accurate and dynamic decision guidance for each link of logistics.

[0124] In this embodiment of the invention, the step of performing decision analysis on the end-to-end logistics data based on the target decision trajectory to obtain multiple logistics decision information includes:

[0125] The target decision trajectory is visually reconstructed to generate a visual logistics trajectory map;

[0126] The visualized logistics trajectory map is matched and associated with the full-link logistics data to obtain an associated logistics dataset.

[0127] Decision parameter clustering analysis was performed on the associated logistics dataset to obtain core logistics parameters;

[0128] A multi-objective logistics decision evaluation matrix is ​​constructed based on the core logistics parameters, and the entropy weight decision influence of the multi-objective logistics decision evaluation matrix is ​​calculated.

[0129] Based on the entropy weight decision influence, each related logistics data in the related logistics dataset is sorted by decision priority to generate a target related logistics dataset.

[0130] The feasibility of the target-related logistics dataset is verified by a pre-set decision tree verification model, resulting in multiple logistics decision information.

[0131] In this embodiment of the invention, the spatiotemporal information (such as timestamps and spatial coordinates) in the target decision trajectory is converted into graphical elements, and a dynamic and visual logistics trajectory map is generated through a map engine or a custom drawing tool. The visual logistics trajectory map connects key nodes (such as starting point, transit station, and destination) with lines, and distinguishes different states (such as normal transportation and delay warning) with color, thickness, or animation effects. At the same time, labels of logistics object attributes (such as vehicle type and cargo type) are superimposed, so that the visual logistics trajectory map can intuitively reflect the status scheduling of the entire logistics data and facilitate quick human understanding.

[0132] This involves accurately matching nodes (such as time points and geographical locations) in the visualized logistics trajectory map with records in the full-chain logistics data. By filtering through time windows (such as matching data within a certain time range before and after the trajectory node) and verifying spatial range (such as matching data within a certain distance around the trajectory node), it is ensured that each trajectory node can be associated with the corresponding original full-chain logistics data (such as order status and vehicle sensor data), forming a related logistics dataset that includes the relationship between the trajectory and the original data.

[0133] Specifically, key parameters affecting decision-making (such as transportation time, cost, and risk level) are extracted from the associated logistics dataset. The dimensionality of the parameters is reduced by feature reduction techniques (such as principal component analysis). Then, clustering algorithms (such as K-means or hierarchical clustering) are used to group similar parameters and identify the core parameter groups that have the greatest impact on logistics decisions (such as high-frequency delay sections and high-cost transportation periods), providing a focus for subsequent decisions. A multi-objective evaluation matrix is ​​constructed based on the core logistics parameters, treating each core logistics parameter as an evaluation dimension (such as timeliness, cost, and safety).

[0134] In detail, the weights of each dimension in the multi-objective logistics decision evaluation matrix are calculated using the entropy weight method. First, the information entropy of each core logistics parameter under different decision scenarios is calculated (reflecting the dispersion and importance of the parameter). Then, the weights are assigned according to the information entropy, so that the core logistics parameters with a large amount of information can obtain higher decision influence, ensuring that the multi-objective logistics decision evaluation matrix can objectively reflect the comprehensive role of each core logistics parameter in decision-making.

[0135] Furthermore, each piece of related logistics data in the related logistics dataset is weighted and scored according to the entropy weight decision influence degree. The weights of dimensions such as timeliness and cost are combined with the actual data values ​​to calculate the comprehensive decision priority. The data is then sorted from high to low priority to generate the target related logistics dataset, ensuring that high-priority data (such as orders that are about to be delayed or high-cost transportation tasks) can be processed first.

[0136] Specifically, the target-related logistics dataset is input into a pre-defined decision tree validation model. Through layer-by-layer rule matching, the decision feasibility of each target-related logistics data is verified, and logistics decision information containing specific operational suggestions (such as switching routes or adjusting loading and unloading order) is generated. At the same time, the reasons for validation failures are recorded to provide feedback for the optimization of the above-mentioned target logistics decision model.

[0137] In this embodiment of the invention, the decision analysis of the entire logistics data is driven by the target decision trajectory, eliminating the need for tedious manual screening and calculation, greatly shortening the decision analysis cycle, and generating multiple highly accurate and reliable logistics decision information, thus significantly improving the intelligence and accuracy of logistics decision-making.

[0138] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0139] like Figure 4 The diagram shown is a functional block diagram of an AI-based intelligent logistics decision-making system provided in an embodiment of the present invention.

[0140] This disclosure provides an AI-based intelligent logistics decision-making system, which corresponds one-to-one with the AI-based intelligent logistics decision-making method described in the previous embodiment. For example... Figure 4 As shown, the AI-based intelligent logistics decision-making system 100 includes a data anonymization and fusion module 101, a model feature learning module 102, a scenario simulation analysis module 103, a logistics trajectory modeling module 104, and a logistics decision analysis module 105. Detailed descriptions of each functional module are as follows:

[0141] The data anonymization and fusion module 101 is used to acquire full-chain logistics data, perform hierarchical anonymization and fusion on the full-chain logistics data, and obtain a target logistics data map.

[0142] The model feature learning module 102 is used to extract the spatiotemporal features of the target logistics data map, and to perform feature learning on the preset decision expert model based on the spatiotemporal features of the logistics to obtain multiple decision sub-models;

[0143] The scenario simulation analysis module 103 is used to perform multi-scenario simulation state analysis on the full-link logistics data using multiple decision sub-models, and to optimize the decision sub-models based on the results of the multi-scenario simulation state analysis to obtain the target logistics decision model.

[0144] The logistics trajectory modeling module 104 is used to model the logistics trajectory of the entire-link logistics data using the target logistics decision model to obtain the target decision trajectory.

[0145] The logistics decision analysis module 105 is used to perform decision analysis on the full-link logistics data based on the target decision trajectory to obtain multiple logistics decision information.

[0146] In this invention, the specific limitations of the AI ​​model-based intelligent logistics decision-making system can be found in the above-described limitations of the AI ​​model-based intelligent logistics decision-making method, and will not be repeated here. Each module in the aforementioned AI model-based intelligent logistics decision-making system 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 operations corresponding to each module.

[0147] In the embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0148] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0149] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0150] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented. Any references to memory, storage, databases, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0151] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0152] In the embodiments provided in this disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0153] It should be noted that, in this disclosure, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element limited by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0154] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An AI model-based logistics intelligent decision method, characterized in that, The method comprises: acquiring full-link logistics data, performing hierarchical desensitization fusion on the full-link logistics data to obtain a target logistics data graph; extracting logistics space-time features of the target logistics data graph, and performing feature learning on a preset decision expert model according to the logistics space-time features to obtain a plurality of decision sub-models; wherein the feature learning on the decision expert model according to the logistics space-time features to obtain a plurality of decision sub-models comprises: acquiring core decision dimensions of the decision expert model, constructing a feature dimension mapping table according to the logistics space-time features and the core decision dimensions; performing feature grouping on the logistics space-time features according to the feature dimension mapping table to obtain multi-dimensional feature data; performing grouped feature learning on the decision expert model according to the multi-dimensional feature data to obtain a plurality of grouped decision models; performing model verification on a plurality of the grouped decision models to obtain a model verification percentage error; determining whether the model verification percentage error is greater than a preset verification error threshold; if the model verification percentage error is greater than the verification error threshold, fine-tuning the grouped decision model to obtain a plurality of decision sub-models; and if the model verification percentage error is less than or equal to the verification error threshold, regarding a plurality of the grouped decision models as decision sub-models; performing multi-scenario simulation state analysis on the full-link logistics data by using a plurality of the decision sub-models, and performing model optimization on the decision sub-models according to a result of the multi-scenario simulation state analysis to obtain a target logistics decision model; performing logistics trajectory modeling on the full-link logistics data by using the target logistics decision model to obtain a target decision trajectory; performing decision analysis on the full-link logistics data according to the target decision trajectory to obtain a plurality of logistics decision information. 2.The AI model-based logistics intelligent decision method of claim 1, wherein, The hierarchical desensitization fusion on the full-link logistics data to obtain a target logistics data graph comprises: performing format standardization processing on the full-link logistics data to obtain standardized logistics data; identifying data sensitivity levels in the standardized logistics data; dividing the standardized logistics data into a plurality of different levels of hierarchical sensitive logistics data according to the data sensitivity levels; performing data hierarchical desensitization on a plurality of the hierarchical sensitive logistics data to obtain hierarchical desensitization logistics data; identifying a plurality of hierarchical logistics data entities in the hierarchical desensitization logistics data, and performing logistics correlation analysis on a plurality of the hierarchical logistics data entities to obtain logistics data correlation relationships; regarding a plurality of the hierarchical logistics data entities as nodes and the logistics data correlation relationships as edges, and generating an initial logistics data graph of the full-link logistics data according to the nodes and the edges; performing graph optimization verification on the initial logistics data graph to obtain a target logistics data graph. 3.The AI model-based logistics intelligent decision method of claim 2, wherein, The data hierarchical desensitization on a plurality of the hierarchical sensitive logistics data to obtain hierarchical desensitization logistics data comprises: performing data replacement desensitization on high-sensitive logistics data in a plurality of the hierarchical sensitive logistics data to obtain high desensitization logistics data; The medium sensitive logistics data in the multiple hierarchical sensitive logistics data is subjected to data encryption and desensitization to obtain medium desensitized logistics data; The low sensitive logistics data in the multiple hierarchical sensitive logistics data is subjected to data shielding desensitization to obtain low desensitized logistics data; The high desensitized logistics data, the medium desensitized logistics data and the low desensitized logistics data are subjected to data aggregation to obtain layered desensitized logistics data. 4.The AI model-based logistics intelligent decision method of claim 1, wherein, The logistics space-time features of the target logistics data graph are extracted, including: The target logistics data graph is subjected to time dimension information disintegration and space dimension information disintegration respectively to obtain time attribute data and space attribute data corresponding to each node in the target logistics data graph; The time dimension features of the time attribute data and the space dimension features of the space attribute data are calculated; The time dimension features are associated and matched with the space dimension features to obtain space-time coupling features; The space-time coupling features are subjected to feature dimension reduction to obtain logistics space-time features of the target logistics data graph. 5.The AI model-based logistics intelligent decision method of claim 1, wherein, The multiple decision sub-models are utilized to perform multi-scenario simulation state analysis on the full-link logistics data, including: The multiple decision sub-models are utilized to perform scenario-adaptive processing on the full-link logistics data to obtain an adaptive logistics data set corresponding to the logistics simulation scenario; The adaptive logistics data set is subjected to single-scenario simulation analysis according to the decision sub-model to obtain a single-scenario simulation decision result; The logistics simulation scenarios are cross-combined to generate cross-simulation scenarios; The adaptive logistics data set is subjected to multi-scenario simulation analysis in the cross-simulation scenarios by utilizing the decision sub-model to obtain a multi-scenario simulation decision result; The single-scenario simulation decision result and the multi-scenario simulation decision result are subjected to consistency verification to obtain an effective simulation decision result. 6.The AI model-based logistics intelligent decision method of claim 1, wherein, The decision sub-model is subjected to model optimization according to the results of multi-scenario simulation state analysis to obtain a target logistics decision model, including: The decision sub-model is subjected to model optimization according to the results of multi-scenario simulation state analysis to obtain a target logistics decision model, including: The decision sub-model is subjected to initial optimization according to the parameter optimization strategy to obtain an initial optimization decision model; The candidate pruning objects and pruning granularity control parameters are generated according to the initial optimization decision model; The edge weights of the candidate pruning objects are obtained, and a weight importance score of each candidate pruning object is generated according to the edge weights; The target pruning objects in the candidate pruning objects are screened out according to the weight importance score; The target pruning objects corresponding to the initial optimization decision model are deleted according to the pruning granularity control parameters to obtain a target logistics decision model. 7.The AI model-based logistics intelligent decision method of claim 1, wherein, The target logistics decision model is utilized to perform logistics trajectory modeling on the full-link logistics data to obtain a target decision trajectory, including: The trajectory association data set in the full-link logistics data is extracted, and the timestamps and spatial coordinates of the trajectory association data set are obtained; The initial logistics trajectory of the full-link logistics data is generated according to the timestamps and the spatial coordinates; The target flow decision model is used for multi-dimensional trajectory analysis on the initial flow trajectory, to obtain multi-dimensional trajectory influence parameters; The initial flow trajectory is corrected in terms of trajectory anomaly according to the multi-dimensional trajectory influence parameters, to obtain a corrected flow trajectory; The full-link flow data is verified in terms of flow feasibility based on the corrected flow trajectory, and the corrected flow trajectory is optimized in terms of trajectory according to the result of flow feasibility verification, to obtain a target decision trajectory. 8.The AI model-based logistics intelligent decision method of claim 1, wherein, The target decision trajectory is used for decision analysis on the full-link flow data, to obtain a plurality of flow decision information, including: The target decision trajectory is reconstructed in a visualized manner, to generate a visualized flow trajectory graph; The visualized flow trajectory graph is matched and associated with the full-link flow data, to obtain an associated flow data set; The associated flow data set is analyzed in terms of decision parameter clustering, to obtain core flow parameters; A multi-target flow decision evaluation matrix is constructed based on the core flow parameters, and an entropy weight decision influence degree of the multi-target flow decision evaluation matrix is calculated; Each associated flow data in the associated flow data set is sorted in terms of decision priority according to the entropy weight decision influence degree, to generate a target associated flow data set; The target associated flow data set is verified in terms of feasibility decision according to a preset decision tree verification model, to obtain a plurality of flow decision information.

9. An AI model-based logistics intelligent decision system, characterized in that, The system comprises: A data desensitization fusion module is configured to obtain full-link flow data, and perform hierarchical desensitization fusion on the full-link flow data, to obtain a target flow data graph; A model feature learning module is configured to extract flow space-time features of the target flow data graph, and learn features of a preset decision expert model based on the flow space-time features, to obtain a plurality of decision sub-models; wherein the learning features of the decision expert model based on the flow space-time features, to obtain a plurality of decision sub-models, comprises: obtaining core decision dimensions of the decision expert model, constructing a feature dimension mapping table based on the flow space-time features and the core decision dimensions; grouping the flow space-time features based on the feature dimension mapping table, to obtain multi-dimensional feature data; learning features of the decision expert model based on the multi-dimensional feature data, to obtain a plurality of grouped decision models; verifying the plurality of grouped decision models, to obtain a model verification percentage error; determining whether the model verification percentage error is greater than a preset verification error threshold; if the model verification percentage error is greater than the verification error threshold, fine-tuning the grouped decision models, to obtain a plurality of decision sub-models; if the model verification percentage error is less than or equal to the verification error threshold, the plurality of grouped decision models are used as decision sub-models; A scene simulation analysis module is configured to use the plurality of decision sub-models to analyze the full-link flow data in a plurality of scene simulation states, and optimize the decision sub-models based on the result of the analysis in the plurality of scene simulation states, to obtain a target flow decision model. The logistics trajectory modeling module is configured to perform logistics trajectory modeling on the full-link logistics data by using the target logistics decision model, to obtain a target decision trajectory. The logistics decision analysis module is configured to perform decision analysis on the full-link logistics data according to the target decision trajectory, to obtain a plurality of logistics decision information.

Images