Logistics full-link risk control method and equipment based on AI large model, and storage medium

Abstract

The invention discloses a logistics full-link risk control method and device based on an AI large model and a storage medium, and relates to the technical field of intelligent logistics, and the method comprises the steps: obtaining a logistics business risk control requirement inputted by a user through an interactive interface; analyzing logistics business risk control requirements, extracting logistics risk factors, and generating a structured risk control rule based on an industry knowledge base and business data in combination with a machine learning algorithm and a large model; identifying and solving rule conflicts of the structured risk control rule by using a collision detection algorithm; dynamically adjusting a rule threshold based on business volatility to obtain an optimized risk control rule; and pushing the risk control rule to a service supervision system based on the message queue or the API interface, and judging whether to trigger interception, alarm or transportation path correction operation based on a service node of the service system. The logistics risk control efficiency is improved, and the requirements of modern logistics for accuracy and real-time performance are met.

Description

Technical Field

[0001] This application relates to the field of intelligent logistics technology, and in particular, to a method, device, and storage medium for full-link risk control of logistics based on an AI large model. Background Art

[0002] The logistics industry involves multiple complex scenarios such as cargo transportation, warehouse management, and delivery timeliness. Its risk control directly affects the enterprise operation cost and service quality. With the rapid development of e-commerce, the logistics network shows a trend of scale expansion and process complexity, and the types of risks are increasingly diversified (such as cargo loss, timeliness delay, compliance risks, etc.).

[0003] Traditional logistics risk control mainly relies on manual experience to formulate fixed rules (such as overloading warning, route deviation alarm, etc.) and conducts passive monitoring through a supervision system. However, the generation of existing risk control rules depends on manual intervention: production and research personnel need to manually analyze user requirements and convert business language into rule logic recognizable by a computer, which has a high technical threshold and is time-consuming and laborious, resulting in low efficiency of logistics risk control. Therefore, improvement is urgently needed. Summary of the Invention

[0004] In order to improve the efficiency of logistics risk control and meet the requirements of modern logistics for accuracy and real-time performance, this application provides a method, device, and storage medium for full-link risk control of logistics based on an AI large model.

[0005] In a first aspect, the invention object of this application is achieved by adopting the following technical solutions: A method for full-link risk control of logistics based on an AI large model, including: Obtaining the logistics business risk control requirements input by a user through an interaction interface; Analyzing the logistics business risk control requirements, extracting logistics risk elements, and combining machine learning algorithms with a large model to generate structured risk control rules based on an industry knowledge base and business data; Using a collision detection algorithm to identify and resolve rule conflicts of the structured risk control rules; dynamically adjusting rule thresholds based on business volatility to obtain optimized risk control rules; Pushing risk control rules to a business supervision system based on a message queue or an API interface, and determining whether to trigger interception, alarm, or transportation path correction operations based on business nodes of the business system.

[0006] By adopting the above technical solutions, a logistics full-link risk control method based on an AI large model is provided. The AI large model can efficiently and intelligently analyze the risk control requirements of logistics operations input by users (which are unstructured risk control requirements at this time), and combine machine learning and large model capabilities to automatically generate structured rules that comply with industry specifications (such as "when the tonnage of hazardous chemicals > 1000 tons and passing through sensitive areas, an alarm is triggered"), replacing the inefficient mode of traditional manual rule writing, improving the rule generation efficiency; and eliminating rule conflicts through a rule collision detection algorithm (such as misjudgment caused by the superposition of multiple rules), and dynamically adjusting the threshold based on business fluctuations (such as a sharp increase in the volume of shipments during holidays), so that the accuracy of risk control rules is improved and the risk control rules have the ability of dynamic optimization. In actual applications, a message queue (MQ) can also be used to achieve the second-level distribution of rules for high-risk scenarios (also known as real-time risk control rules), and combined with API interfaces to support batch deployment for offline scenarios, covering all-link nodes such as order creation, cargo tracking, and route planning, improving the timeliness of logistics risk interception. Therefore, this application can not only improve the efficiency of logistics risk control, but also meet the requirements of modern logistics for accuracy and real-time performance.

[0007] In a preferred example of this application: The generation of structured risk control rules includes: Performing semantic slicing on the logistics business risk control requirements through a word segmentation model, and using a BERT model to extract risk features including goods categories, transportation attributes, and environmental parameters to obtain logistics risk elements; Conducting classification modeling based on historical violation data to generate risk scoring rules; Combining a predefined compliance template library and the logistics risk elements, and generating standardized risk control clauses through a rule matching algorithm to obtain structured risk control rules.

[0008] By adopting the above technical solutions, the accuracy of risk element identification is improved based on the semantic feature extraction of the BERT model, supporting the generation of refined rules for complex scenarios (such as the transportation of hazardous chemicals); the risk scoring rules generated by historical data modeling improve the detection rate of violation behaviors and reduce the false alarm rate; the template library and rule matching algorithm improve the generation efficiency of standardized clauses and reduce the maintenance cost of the rule library.

[0009] In a preferred example of this application: The steps of dynamically adjusting the rule threshold further include: Using a sliding window algorithm to perform time series analysis on business data for a specified time period, and calculating the probability distribution of the occurrence of risk times; Based on the probability distribution characteristics, generating extreme scenario stress test data through Monte Carlo simulation, and dynamically correcting the upper and lower limits of the rule threshold; Based on the corrected threshold range parameters, using a reinforcement learning model to perform parameter optimization to determine the optimal threshold combination that maximizes the ratio of the risk interception rate to the business passing rate.

[0010] By adopting the above technical solutions, the risk probability prediction error rate is reduced to within 3% by sliding window time series analysis, and the threshold correction response time is improved to the minute level; Monte Carlo simulation can cover 99% of extreme scenarios, so as to improve the rule robustness and avoid system crashes caused by sudden risks; the reinforcement learning optimization makes the ratio of the interception rate to the passing rate reach Pareto optimality (such as 3:1), reducing business losses.

[0011] In a preferred example of this application: the service nodes include order fulfillment, transportation route, and delivery acceptance. Pushing risk control rules to the service supervision system based on message queues or API interfaces further includes: Through the collaboration of real-time risk control intelligent agents and offline risk control intelligent agents by the workflow orchestration engine, the real-time risk control intelligent agent processes immediate risk events based on message queues, and the offline risk control intelligent agent periodically analyzes historical data through API interfaces and updates the risk control rule library; Adopting a dual-mode risk control decision tree, performing real-time interception and offline risk assessment in parallel at the service nodes, and determining the final risk control strategy through a weighted voting mechanism; Constructing a logistics risk simulation environment based on digital twin technology, simulating the impact of rule changes on the entire logistics link, obtaining risk assessment data, and generating a risk assessment report and synchronizing it to the service supervision system.

[0012] By adopting the above technical solutions, the collaboration of dual-mode intelligent agents shortens the full-link risk response time to within 200ms, improving the interception success rate of high-risk events; digital twin simulation improves the efficiency of rule change verification and reduces the risk of misoperation; the weighted voting mechanism improves the consistency of risk control strategies for multiple nodes.

[0013] In a preferred example of this application: the method further includes: Obtaining logistics basic data and risk control design parameters, and judging the first predicted risk value of each node and the second predicted risk value of the full link during the goods transportation process according to the logistics basic data and the risk control design parameters; Generating a logistics risk control comparison table for dynamically adjusting risk control strategies based on the risk control design parameters, the first predicted risk value, and the second predicted risk value; Obtaining a logistics risk control reference range including historical risk disposal records, and constructing a multi-modal risk control decision model by combining the logistics risk control comparison table and the logistics risk control reference range; Obtaining real-time logistics operation status data, and inputting the operation status data into the multi-modal risk control decision model to generate dynamic risk control instructions.

[0014] By adopting the above technical solution, obtaining logistics basic data and risk control design parameters, and combining the first predicted risk values of each node and the second predicted risk value of the whole link in the process of goods transportation, a comparison table for adjusting the risk control strategy can be dynamically generated, thereby enhancing the predictability and response speed of the system to risks in different transportation links and ensuring the pertinence of risk control measures.

[0015] In a preferred example of this application: The judgment of the first predicted risk value of each node and the second predicted risk value of the whole link in the process of goods transportation includes: Extract logistics basic data from the business database, where the logistics basic data includes goods attribute parameters, transportation network topology, carrier capacity matrix, and environmental monitoring data; the risk control design parameters include node risk thresholds, link fault tolerance rates, and emergency response levels; Analyze the coupling relationship between goods attribute parameters and transportation network topology through a spatio-temporal prediction algorithm to judge the first predicted risk value of the loading and unloading nodes; Use a graph neural network to analyze the interaction between the carrier capacity matrix and environmental monitoring data to determine the second predicted risk value of the transfer nodes.

[0016] By adopting the above technical solution, detailed basic data is extracted from the business database, and spatio-temporal prediction algorithms and graph neural network analysis are used to accurately judge the risk values of loading and unloading nodes and transfer nodes. This process not only improves the accuracy of risk assessment but also provides a scientific basis for subsequent risk control.

[0017] In a preferred example of this application: The generation of the logistics risk control comparison table specifically includes: Generate a risk propagation path based on Monte Carlo simulation, and calculate the theoretical risk propagation probability in combination with the node risk threshold; Extract feature vectors from historical risk events and construct a risk feature library through contrastive learning; Establish a mapping relationship between risk levels and disposal measures according to the matching degree between the theoretical risk propagation probability and the risk feature library; Calculate the difference in risk disposal time limits under different transportation scenarios, and generate a logistics risk control comparison table including the emergency response level; Or, The construction of the multi-modal risk control decision model specifically includes: Build a hybrid architecture integrating a time series prediction network and a graph attention mechanism, and the input end receives real-time operation status data and the logistics risk control comparison table; Set up a closed-loop decision-making unit including a reinforcement learning module, a policy generation module, and an effect evaluation module; Optimize the policy in a logistics risk simulation environment to generate a risk control decision tree with adaptive capabilities.

[0018] By adopting the above technical solutions, the accuracy and timeliness of risk disposal measures are improved. At the same time, the application of the closed-loop decision-making unit enables the system to have the ability of self-adaptive optimization. The multi-modal risk control decision-making model realizes the efficient processing of real-time operation state data and accurate risk prediction in complex environments. The application of the reinforcement learning module enables the system to continuously optimize risk control strategies in a changing business environment.

[0019] In a second aspect, the invention object of the present application is achieved by the following technical solutions: A logistics full-link risk control system based on an AI large model is applied to the logistics full-link risk control method based on an AI large model as described above. The system includes: An interaction module for obtaining the risk control requirements of the logistics business input by the user through a graphical user interface; An analysis and generation module for analyzing the risk control requirements of the logistics business, extracting the logistics risk elements therein, and generating structured risk control rules according to the industry knowledge base and historical business data by combining machine learning algorithms and pre-trained large models; An optimization module for identifying and resolving rule conflicts in the structured risk control rules using a collision detection algorithm and dynamically adjusting the rule thresholds according to business volatility to obtain optimized risk control rules; A push and execution module for pushing the optimized risk control rules into the business supervision system through message queue or API interface technology and judging whether to trigger interception, warning or transportation path correction operations at each node of the business system.

[0020] In a third aspect, the invention object of the present application is achieved by the following technical solutions: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above logistics full-link risk control method based on an AI large model are implemented.

[0021] In a fourth aspect, the invention object of the present application is achieved by the following technical solutions: A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above logistics full-link risk control method based on an AI large model are implemented.

[0022] In summary, the present application includes at least one of the following beneficial technical effects: 1. Automate the generation of risk control rules, significantly improving the efficiency of rule construction and reducing the cost of manual intervention; based on rule collision detection and dynamic threshold optimization mechanisms, effectively enhancing the environmental adaptability and execution stability of risk control rules; combined with message queue and dual-mode agent collaboration technology, achieving a deep integration of real-time interception and offline evaluation of end-to-end risks, and greatly improving the risk response speed and disposal coverage rate of logistics operations; 2. The spatio-temporal prediction algorithm analyzes the cargo-network coupling relationship, reducing the risk of missed inspections at loading and unloading nodes; the graph neural network captures the carrier-environment interaction effects, reducing the misjudgment rate of risks at transfer nodes. Brief Description of the Drawings

[0023] Figure 1 is a flowchart of a logistics end-to-end risk control method based on an AI large model in an embodiment of the present application; Figure 2 is another flowchart of a logistics end-to-end risk control method based on an AI large model in an embodiment of the present application; Figure 3 is a schematic diagram of a device in an embodiment of the present application. Detailed Embodiments

[0024] The following further elaborates on the present application with reference to the accompanying drawings.

[0025] In one embodiment, as Figure 1 shown, the present application discloses a logistics end-to-end risk control method based on an AI large model, which specifically includes the following steps: S1: Obtain the risk control requirements of the logistics operation input by the user through the interaction interface.

[0026] In this embodiment, a visual interaction interface (such as a Web-based management platform or an API interface) receives the logistics risk control requirements input by the user. The interaction interface supports natural language input (such as "trigger an alarm when the goods are hazardous chemicals and the tonnage exceeds 1000 tons") or structured parameter configuration (such as selecting "transportation node" as "loading address" and setting the "sensitive area" label, etc.).

[0027] S2: Analyze the risk control requirements of the logistics operation, extract logistics risk factors, and combine machine learning algorithms and large models to generate structured risk control rules based on the industry knowledge base and business data.

[0028] In this embodiment, the logistics risk factors include risk subjects, risk thresholds, spatial constraints, and business scenarios. Among them, the risk subjects include the type of goods (hazardous chemicals / ordinary goods); the risk thresholds include tonnage (>1000 tons), transportation time (22:00-6:00 at night); the spatial constraints include the loading address (industrial park), the unloading address (residential area); the business scenarios include order creation, transportation transfer, and delivery acceptance.

[0029] Specifically, step S20 includes: Generating structured risk control rules includes: S21: Perform semantic slicing on the risk control requirements of logistics operations through a word segmentation model, and use the BERT model to extract risk features including goods categories, transportation attributes, and environmental parameters to obtain logistics risk elements.

[0030] In this embodiment, a BERT tokenizer (based on a Chinese pre-trained model) is used to tokenize the input logistics risk control requirements of the user. For example, the input text "When the goods are hazardous chemicals and the tonnage exceeds 1000 tons, an alarm is triggered" is tokenized as: "[CLS] When the goods are hazardous chemicals and the tonnage exceeds 1000 tons, an alarm is triggered [SEP]", and named entity recognition is performed on the tokenized result through the BERT model to label the risk subject, attribute, and threshold: Risk subject: goods (entity type: goods category) Attribute: tonnage (entity type: transportation attribute) Threshold: 1000 tons (numeric parameter) Logical relationship: and (conditional conjunction) Specifically, using the output of the last hidden layer of the BERT model, semantic feature vectors are generated for each entity. For example: The feature vector of hazardous chemicals represents its danger level in the logistics risk scenario.

[0031] Key element screening: Calculate the weights of each feature through the TF-IDF algorithm, and screen out the core risk elements, such as high-weight features: hazardous chemicals (high danger level), tonnage > 1000 tons (quantified risk threshold).

[0032] S22: Perform classification modeling based on historical violation data to generate risk scoring rules.

[0033] In this embodiment, historical violation records (such as hazardous chemical overloading incidents in the past 3 years) are extracted from the business database, including fields: goods type, tonnage, transportation time, route, whether an accident is triggered, etc. Extract time features, spatial features, and interaction features from the historical violation data. Time features such as converting transportation time into discrete features (such as "night transportation": 22:00 - 6:00); spatial features such as geocoding the loading address and unloading address to extract regional risk labels (such as "sensitive area", "high-speed section"); interaction features such as calculating the correlation between the goods danger level and transportation time (such as the risk coefficient of hazardous chemicals during night transportation increases).

[0034] Specifically, an XGBoost algorithm is used to construct a binary classification model (violation / normal), and the input features include goods type, tonnage, time, area, etc.; the decision-making logic of the model is analyzed through SHAP values to generate an interpretable risk scoring rule: for example, risk score = 0.8×hazard level + 0.2×overload coefficient.

[0035] S23: Combine the predefined compliance template library and logistics risk elements, and generate standardized risk control terms through a rule matching algorithm to obtain structured risk control rules.

[0036] In this embodiment, the compliance template library contains multiple predefined rule templates covering scenario risk scenarios. For example, threshold trigger type: IF goods = hazardous chemicals AND tonnage > threshold THEN alarm; spatio-temporal constraint type: IF time ∈ sensitive period AND area ∈ restricted area THEN route correction; combined logic type: IF (goods = hazardous chemicals OR tonnage > 800) AND weather = heavy rain THEN intercept. Each rule template reserves variable slots for thresholds or sensitive periods, supporting dynamic filling.

[0037] Specifically, the rule matching algorithm uses regular expression matching technology to bind the risk elements extracted in S21 to the template slots, and then checks the syntax legality of the rules through a formal verification tool (such as ANTLR) to ensure no logical contradictions. The Apriori algorithm is used to remove duplicate rules.

[0038] S3: Use a collision detection algorithm to identify and resolve rule conflicts in the structured risk control rules; dynamically adjust the rule thresholds based on business volatility to obtain optimized risk control rules.

[0039] In this embodiment, the collision detection algorithm is based on a rule conflict matrix, compares the new rules with the risk control rules in the existing rule library to determine whether there are collisions and overlaps, and uses a priority arbitration algorithm (such as based on the rule generation timestamp or business importance), and sorts the conflicting rules according to predefined priority rules (such as business importance, generation time, risk level), and retains the high-priority rules.

[0040] Priority dimensions are as follows: Business hard regulations (such as safety rules required by regulations have the highest priority); Rule generation time (new rules overwrite old rules, and manual confirmation of exceptions is required); Risk impact scope (rules affecting the whole have a higher priority than local rules).

[0041] Use a weighted voting mechanism to assign priority scores to each conflicting rule, and finally select the rule with the highest score for execution.

[0042] Specifically, dynamically adjusting the rule thresholds includes: S31: Use the sliding window algorithm to perform time series analysis on the business data for a specified time period, and calculate the probability distribution of the occurrence of risk times.

[0043] In this embodiment, the sliding window algorithm slides on the data stream through a time window of a fixed length (such as 7 days), and calculates statistical features segment by segment; the risk event probability distribution refers to the occurrence frequency distribution (such as normal distribution) of risk behaviors (such as overloading of hazardous chemicals) in the time dimension.

[0044] Specifically, the sliding window algorithm divides the specified time period into multiple consecutive windows according to hour / day granularity (such as a 24-hour window sliding 1 hour); calculates the incidence rate of risk events (such as the proportion of overloaded orders) within each window, and statistically analyzes the time series correlation of risk times (such as the overloading rate on weekends is higher than that on weekdays), and then uses the maximum likelihood estimation method to fit the probability distribution of risk events (such as a normal distribution with μ = 0.3 and σ = 0.1).

[0045] S32: Based on the probability distribution characteristics, generate extreme scenario stress test data through Monte Carlo simulation, and dynamically correct the upper and lower limits of the rule thresholds.

[0046] In this embodiment, Monte Carlo simulation refers to generating a large number of extreme scenario data through random sampling to evaluate the performance of the system under rare events; extreme scenario stress test refers to simulating extreme risk events that exceed the conventional thresholds (such as a sudden 500% increase in the transportation volume of hazardous chemicals).

[0047] Specifically, based on the statistical results of S31, extract the distribution parameters of risk events (such as mean μ, variance σ), and define the boundary conditions of extreme scenarios (such as tonnage > 1500 tons, transportation time < safety time limit), use the Monte Carlo method to generate random numbers that conform to the probability distribution (such as generating 10,000 simulated orders), conduct random sampling, and superimpose extreme conditions (such as a sharp increase in transportation volume during holidays, extreme weather) on the simulated data; generate violation samples (such as "hazardous chemical tonnage = 2000 tons and passing through sensitive areas"). Through simulation, calculate the risk interception rate and false alarm rate under different thresholds.

[0048] S33: Based on the corrected threshold range parameters, use the reinforcement learning model to perform parameter optimization to determine the optimal threshold combination that maximizes the ratio of the risk interception rate to the business passing rate.

[0049] In this embodiment, reinforcement learning (RL) refers to a machine learning method that optimizes decision-making strategies through a trial-and-error mechanism. The core elements include state, action, and reward. The ratio of the risk interception rate to the business passing rate is an indicator used to measure the effectiveness of risk control rules.

[0050] Specifically, first define the state space, action space parameters, and reward function. The state space defines the dynamic factors affecting the threshold (such as the current volume of shipments, weather level, regional risk level); the action space is the range of adjustable threshold parameters (such as the transport weight T ∈ [800, 1200] tons), and the reward function is like Reward = α1 × interception rate + β1 × (1 - false alarm rate) + γ1 × business continuity; where α1, β1, and γ1 are weight coefficients used to balance the interception effect and business losses.

[0051] Use a Deep Q-Network (DQN) to process the high-dimensional state space, construct a neural network structure. The input layer is the current state vector, and the Q values of the output layer correspond to the expected benefits of different threshold actions. Parameter optimization means that the large model agent performs threshold adjustment actions in the simulation environment, calculates the cumulative benefits according to the reward function, and finally outputs the threshold combination that maximizes Reward (such as T = 1050 tons, false alarm rate ≤ 5%).

[0052] S4: Push the risk control rules to the business supervision system based on the message queue or API interface, and determine whether to trigger interception, alarm, or transportation route correction operations based on the business nodes of the business system.

[0053] In this embodiment, the business nodes include order fulfillment, transportation route, and delivery acceptance.

[0054] Specifically, step S4 includes: S41: Through the collaboration of the real-time risk control agent and the offline risk control agent by the workflow orchestration engine, the real-time risk control agent processes immediate risk events based on the message queue, and the offline risk control agent periodically analyzes historical data through the API interface and updates the risk control rule library.

[0055] In this embodiment, the workflow orchestration engine is a scheduling system used to coordinate multiple heterogeneous agents (such as real-time risk control and offline risk control) to work together according to preset logic; the real-time risk control agent is an AI module (such as Flink + TensorFlow) that processes immediate risk events based on the message queue (MQ) in real time; the offline risk control agent is a batch processing module (such as Spark + Hadoop) that periodically analyzes historical data through the API interface and updates the rule library.

[0056] Specifically, the real-time risk control link deploys real-time risk control rules (such as "hazardous chemical tonnage > 1000 tons") to the Kafka message queue and sets up a priority queue; the rule execution content is like the Flink real-time computing engine listening to the message queue, parsing order data (goods, tonnage, address), and matching the rule conditions; if a rule is triggered (such as overloading), an interception interface is called through the HTTP API (such as notifying the security team).

[0057] The offline risk control link schedules tasks through Airflow, calls the business system API to pull historical order data, cleans the data and stores it in HDFS to form an offline analysis data lake. When the rules are updated, the new rules are pushed to the rule engine database through the RESTful API.

[0058] S42: A dual-mode risk control decision tree is used to perform real-time interception and offline risk assessment in parallel at business nodes, and the final risk control strategy is determined through a weighted voting mechanism.

[0059] In this embodiment, the dual-mode risk control decision tree is a hybrid decision model that combines real-time interception (online learning) and offline evaluation (batch learning); the weighted voting mechanism allocates voting weights based on the confidence of real-time risk control (high timeliness) and offline risk control (high accuracy).

[0060] Specifically, at the order fulfillment node, the real-time risk control agent intercepts overloaded orders (response time < 200ms); the offline risk control agent obtains the historical overload event frequency through the API and calculates the abnormal index of the area to which the order belongs. At the transportation path node, the real-time risk control detects the vehicle deviating from the scheduled route (GPS positioning drift > 50 meters); the offline risk control evaluates the safety risk level of the road section based on the historical trajectory data.

[0061] S43: Build a logistics risk simulation environment based on digital twin technology, simulate the impact of rule changes on the entire logistics chain, obtain risk assessment data, and generate a risk assessment report that is synchronized to the business supervision system.

[0062] In this embodiment, the risk assessment report is an analysis document used to quantify the impact of rule changes on business indicators (interception rate, false alarm rate, timeliness). The risk assessment report includes a comparison of risk heat maps before and after the rule change; key indicators such as interception rate, false alarm rate, business loss rate, and recommended rule optimization directions (such as adjusting the speed limit threshold).

[0063] In one embodiment, if Figure 2 As shown in the figure, the full-link risk control method for logistics based on AI big model also includes: S10: Obtain logistics basic data and risk control design parameters, and determine the first predicted risk value of each node in the cargo transportation process and the second predicted risk value of the entire link according to the logistics basic data and risk control design parameters.

[0064] In this embodiment, the logistics basic data includes cargo attributes, transportation network topology (road grade, node location), carrier capacity matrix (transportation timeliness, safety score), and environmental monitoring data (weather, road conditions); the risk control design parameters include node risk thresholds (such as "speed limit 50 km / h in sensitive areas"), link fault tolerance rates (such as "allowing 1 path deviation"), and emergency response levels (such as "immediate interception required for red alerts"). The first predicted risk value refers to the risk probability of a single transportation node (such as loading, transfer, unloading) (such as "risk value of storing hazardous chemicals in high-temperature areas is 0.8"); the second predicted risk value refers to the comprehensive risk value of the entire link (from shipment to receipt) (such as "overall link risk value is 0.6, triggering a yellow alert").

[0065] Specifically, step S10 includes: S101: Extract the logistics basic data from the business database. The logistics basic data includes cargo attribute parameters, transportation network topology, carrier capacity matrix, and environmental monitoring data; the risk control design parameters include node risk thresholds, link fault tolerance rates, and emergency response levels.

[0066] In this embodiment, the cargo attribute parameters include cargo types (such as hazardous chemicals, fresh produce), volume, weight, hazard levels (such as UN numbers), and special transportation requirements (such as constant temperature, shockproof); the transportation network topology is a logistics network represented in a graph structure, with nodes being cities / ports / warehouses and edges being transportation routes (including road grades, traffic restrictions, distances); the carrier capacity matrix is a two-dimensional table with carrier IDs horizontally and capacity indicators (including timeliness scores, accident rates, safety certification levels) vertically; the environmental monitoring data refers to meteorological data (temperature, rainfall, wind), traffic data (congestion index, accident records), and geographical information (terrain).

[0067] Specifically, map the cargo types to risk levels in advance (such as hazardous chemicals = 5, electronic products = 1) and calculate the length of the transportation route.

[0068] S102: Analyze the coupling relationship between the cargo attribute parameters and the transportation network topology through a spatio-temporal prediction algorithm to determine the first predicted risk value of the loading and unloading nodes.

[0069] In this embodiment, the spatio-temporal prediction algorithm combines the LSTM long short-term memory network and the GCN graph convolutional network, and is a model that simultaneously considers the dynamic changes in time series and the spatial topology relationship. The loading and unloading nodes refer to the operation points where goods are loaded or unloaded (such as warehouses, ports); the coupling relationship refers to the mutual influence between the cargo attributes and the transportation network topology.

[0070] Specifically, the analytical operation of the spatio-temporal prediction algorithm includes the processing of time series by LSTM. Based on the historical transportation time series of goods, it outputs a time dynamic feature vector for obtaining transportation (such as seasonal fluctuations and periodic extensions). The analytical operation of the spatio-temporal prediction algorithm also uses a GCN graph convolutional network to process the graph structure. Based on the adjacency matrix of the transportation network and node features (road grade, risk grade), it outputs a spatial association feature vector of the nodes, and then uses a fully connected network (FCN) to predict the risk value of the loading and unloading nodes. For example, the risk value of the loading and unloading nodes of a hazardous chemical order on a mountain road is predicted to be 0.85 (high risk).

[0071] S103: Analyze the interaction effect between the carrier capacity matrix and the environmental monitoring data using a graph neural network, and determine the second predicted risk value of the transfer node.

[0072] In this embodiment, the interaction effect refers to the synergistic effect between the carrier capacity and the environmental monitoring data (such as the delay rate of carrier A increasing by 30% in rainy weather).

[0073] Specifically, first construct a carrier-environment interaction graph, where the nodes are carrier IDs and environmental monitoring point IDs; the edges are the historical performance of the carrier in a specific environment (such as the delay rate of carrier A on a certain section of the road). Then obtain and aggregate the node features. The carrier node features include time efficiency frequency division, accident rate, and safety certification, and the environmental node features include temperature, rainfall, and road surface conditions.

[0074] Calculate the interaction weight between the carrier node and the environmental node based on the attention mechanism: , where is the attention weight; is the query matrix, representing the "demand" of the current node for attention to other nodes; is the transposed matrix of the key matrix, and the key matrix represents the "feature identification" of other nodes; is used to calculate the similarity score matrix of each node to all other nodes; is the scaling factor; the softmax function normalizes the similarity matrix into a probability distribution (the sum of all elements is 1).

[0075] Then use a graph attention network (GNN) layer to process the carrier features and environmental features, obtain the output value of the graph attention network, and determine the risk level of the transfer node (including high, medium, and low risk levels) based on the output value of the graph attention network and then output it.

[0076] S20: Generate a logistics risk control comparison table for dynamically adjusting the risk control strategy based on the risk control design parameters, the first predicted risk value, and the second predicted risk value.

[0077] In this embodiment, step S20 includes: S201: Generate a risk propagation path based on Monte Carlo simulation, and calculate the theoretical risk propagation probability in combination with the node risk threshold.

[0078] In this embodiment, the theoretical risk propagation path refers to the path where a risk event spreads from one node to other nodes in the logistics network (such as "hazardous chemical leakage → pollute the transport vehicle → pollute the unloading warehouse"). The node risk threshold is the preset upper limit of node risk tolerance (such as "warehouse storage hazardous chemical risk threshold = 0.7").

[0079] Specifically, first define the risk event types (such as leakage, delay, overload) and their propagation rules (such as "the propagation probability of leakage risk between adjacent nodes is 0.3"). For example, the propagation path of a hazardous chemical leakage event on the transport path may include "loading node → transport transfer → unloading node".

[0080] Configure the simulation parameters of Monte Carlo simulation: first input the node risk thresholds of each node for different risks (such as sensitive area risk threshold = 0.8) and the link fault tolerance rate (such as allowing 1 path deviation); then set the Monte Carlo simulation parameters (such as the number of sampling times = 10,000 times, the number of risk propagation steps = 3 steps or 4 steps); simulate the generation path of risk propagation: randomly generate the risk starting node (such as "loading warehouse"), and select the next-hop node according to the probability (such as "transport vehicle" → "unloading dock").

[0081] Calculate the total risk value of each path: , where i is the node identifier; then calculate and statistically analyze the probability of reaching or exceeding the node risk threshold in all simulated paths to obtain the theoretical risk propagation probability: .

[0082] S202: Extract the feature vectors from historical risk events, and construct a risk feature library through contrastive learning.

[0083] In this embodiment, the feature vector is a numerical vector representing the characteristics of a risk event; contrastive learning refers to learning the similarity characteristics of risk events by comparing positive and negative sample pairs; the risk feature library is a database for storing the characteristics of historical risk events and their corresponding disposal effects.

[0084] Specifically, extract risk event records from the historical database, including: goods type, transportation time, weather conditions, disposal measures, results (success / failure); then use the TF-IDF algorithm to extract text features (such as the keyword weights of "hazardous chemicals" and "sensitive areas"), generate semantic feature vectors through Word2Vec (such as "heavy rain" → [0.3, 0.7, 0.2]), and conduct contrastive learning training of positive and negative samples. Positive samples are events of the same risk level (such as two "hazardous chemical leakage" events); negative samples are events of different risk levels (such as "hazardous chemical leakage" and "ordinary goods delay"); use a Siamese network to train a feature embedding model to make the vector distances of similar events closer, and then associate and store the trained feature vectors with the disposal results to form a queryable risk feature library.

[0085] S203: Establish a mapping relationship between risk levels and disposal measures according to the matching degree between the theoretical risk propagation probability and the risk feature library.

[0086] In this embodiment, the matching degree refers to the similarity score (such as cosine similarity) between the theoretical risk propagation probability and the historical feature library; the risk level is the severity of the risk divided according to the matching degree (such as high, medium, low); the disposal measure is the coping strategy for different risk levels (such as interception, detour, release).

[0087] Specifically, the similarity calculation between the propagation probability vector of the current risk event and the historical feature library is as follows: , where is the cosine similarity, and the risk level is divided in combination with a preset similarity threshold: High risk (similarity ≥ 0.8): Intercept immediately; Medium risk (0.3 ≤ similarity < 0.8): Strengthen monitoring; Low risk (similarity < 0.3): Normal passage.

[0088] Specifically, the formula definition of the cosine similarity is: , where, is the current risk event vector; is the historical feature library vector; the numerator is the vector dot product, representing the feature synergy effect; the denominator is the product of the vector norms, normalizing the direction consistency; the data dimensions of the input vectors include the risk diffusion speed, the spatio-temporal influence range, and the resource pressure index.

[0089] Exemplarily, the example scenario is that an abnormal image is found during the security inspection of cross-border logistics goods, and it is necessary to evaluate the similarity with historical high-risk events and extract the current event vector (i.e., diffusion speed: 80% of the benchmark value; influence range: 60% of the link; resource pressure: 30% redundancy), and the historical high-risk event vector The calculation process is as follows: 0.8 × 0.75 + 0.6 × 0.55 + 0.3 × 0.25 = 0.6 + 0.33 + 0.075 = 0.995.

[0090] Magnitude product: , . Cosine similarity = ≈1.07. Since in the calculation of cosine similarity, the theoretical value range should be [-1, 1], the calculation result is "clamped" to 1.0 at this time. It is determined that: matching degree = 1.0, and the risk level is high risk, then the "interception" action is triggered and the emergency team is activated.

[0091] Based on historical successful cases, recommend the optimal disposal measures for each risk level and establish associations.

[0092] S204: Calculate the difference in risk disposal time limits under different transportation scenarios and generate a logistics risk control comparison table including the emergency response level.

[0093] Specifically, the difference in risk disposal time limits refers to the difference between the time required for risk disposal in different scenarios and the standard time (such as "the night disposal time limit is 20 minutes slower than the day"); the emergency response level is the response priority divided according to the time limit difference (set different emergency response priorities: for example, a red response requires immediate handling, and a green response can be postponed).

[0094] Specifically, divide the scenarios according to the transportation type (such as hazardous chemicals, fresh food) and region (such as city, mountain area). For example, Scenario A: Hazardous chemicals are transported at night through the mountain area; Scenario B: Fresh food products are transported through the city on a rainy day; among them, the difference in risk disposal time limits is the difference between the actual disposal time and the standard disposal time.

[0095] Divide the response levels according to the time limit difference and the risk level: For example, in the high-risk scenario of hazardous chemicals transportation, if the time limit difference ≥ 10 minutes, the response level is red. In the medium-risk scenario of fresh food transportation, if the time limit difference ≤ 5 minutes, the response level is yellow.

[0096] Integrate the risk level, disposal measures and response level to generate a comparison table: ① Risk level: high risk, propagation probability threshold: ≥0.8, disposal measure: intercept immediately; emergency response level: red; ② Risk level: medium risk, propagation probability threshold: 0.3 - 0.8, disposal measure: strengthen monitoring; emergency response level: yellow; ③ Risk level: low risk, propagation probability threshold: <0.3, disposal measure: normal passage; emergency response level: green.

[0097] S30: Obtain the logistics risk control reference range containing historical risk disposal records, and construct a multi-modal risk control decision-making model by combining the logistics risk control comparison table and the logistics risk control reference range.

[0098] In this embodiment, the historical risk disposal records include detailed disposal data of past risk events (such as disposal time, measures, results, costs); the logistics risk control reference range is the boundary values of reasonable disposal time limit, resource consumption, risk tolerance, etc. (such as "the disposal time limit for high-risk events ≤ 30 minutes") based on historical data statistics.

[0099] Specifically, extract historical risk event data from business systems (such as WMS, TMS), including: risk types (such as overloading, route deviation, cargo damage), disposal records (such as interception time, detour route, emergency resource call), and result labels (success / failure, cost, time limit). Calculate the average disposal time limit and standard deviation of various risk types (such as "the average time limit for high-risk events is 28 minutes, ±5 minutes"), statistically calculate the upper and lower limits of the corresponding disposal costs (such as "the average cost is 400, and the maximum does not exceed 800"), and define the acceptable minimum success rate (such as "the success rate of high-risk events ≥ 95%") to obtain the logistics risk control reference range.

[0100] Specifically, the steps of constructing the multi-modal risk control decision-making model specifically include: S301: Build a hybrid architecture that integrates a time series prediction network and a graph attention mechanism, and the input end receives real-time operation status data and the logistics risk control comparison table.

[0101] In this embodiment, the time series prediction network is a neural network that processes time series data (such as transportation time limit, risk event occurrence frequency), and the LSTM network is used in this application; the graph attention mechanism (GAT) is an attention model used to capture the dynamic relationships between nodes (such as warehouses, transportation nodes) in the logistics network; the hybrid architecture is a deep learning framework that simultaneously integrates time series modeling and graph structure modeling. The real-time operation status data includes GPS positioning data (vehicle location, speed), sensor data (cargo temperature, humidity), and business data (order status, transportation path).

[0102] Specifically, the input layer of the time series prediction network receives historical time series data (such as the failure rate of a certain node in the past 7 days), and the encoding layer uses LSTM or Transformer to encode time-dependent relationships; the output layer predicts the future risk probability (such as "the failure probability of node A within the next 2 hours is 0.6").

[0103] The graph nodes of the graph attention mechanism are the key nodes (warehouses, transfer stations, sensitive areas) in the logistics network, and the edges are the association relationships between the nodes (transportation routes, risk propagation routes), and can calculate the weights of the relationships between nodes through the multi-head attention mechanism; for example, the association weight between node B and node C is 0.8, and the weight between node B and node D is 0.3. Concatenate the output of the time series prediction (such as the risk probability) with the node importance score of the graph attention mechanism; then comprehensively combine the results of the two modalities through weighted summation.

[0104] S302: Set up a closed-loop decision-making unit including a reinforcement learning module, a policy generation module, and an effect evaluation module.

[0105] In this embodiment, the reinforcement learning module is a machine learning module that optimizes the decision-making strategy by interacting with the simulation environment; the policy generation module generates specific risk control instructions (such as interception, detour) based on the model output; the effect evaluation module is used to quantitatively evaluate the policy effect.

[0106] Specifically, the parameters of the state space of the reinforcement learning module include: the current risk level (high / medium / low), the remaining time limit (such as "2 hours remaining until the delivery time"), and the resource occupancy (such as "the number of available security personnel"); the parameters of the action space include interception, detour along a safe route, notify the emergency team, and release; the reward function R = α × risk avoidance - β × time delay - γ × cost, where α, β, and γ are weight coefficients.

[0107] The evaluation metrics of the effect evaluation module are the risk avoidance rate (the proportion of successfully preventing risk events), the false alarm rate (the proportion of normal events intercepted), and the business loss rate (the time delay cost caused by risk control), and the evaluation results are fed back to the reinforcement learning module to optimize the strategy.

[0108] S303: Optimize the strategy in the logistics risk simulation environment to generate an adaptive risk control decision tree.

[0109] In this embodiment, the logistics risk simulation environment is a virtual environment (such as a digital twin platform) for simulating the real logistics network; the adaptive risk control decision tree is a decision tree model that dynamically adjusts the branch conditions according to the simulation results.

[0110] Specifically, the logistics risk simulation environment can dynamically configure parameters such as risk thresholds, resource capacities, and weather conditions. The policy optimization training includes: the agent executes actions (such as "detour") in the simulation environment, observes the development of risk events; optimizes the policy according to the preset reward function (such as "detouring causes a 10% increase in time delay but a 20% increase in the risk avoidance rate"); then generates a tree-like decision structure based on the policy output of reinforcement learning.

[0111] S40: Obtain the logistics operation status data in real time, and input the operation status data into the multi-modal risk control decision-making model to generate dynamic risk control instructions.

[0112] In this embodiment, the logistics operation status data is obtained in real time based on IOT devices, API interfaces, and message queues. Features strongly related to risks (such as "driving speed on mountain roads at night") are extracted from the logistics operation status data. The features include temporal features, spatial features, and business features.

[0113] Exemplarily, the temporal features include calculating the speed volatility (speed volatility = (current speed - 1-hour average) / 1-hour standard deviation). The spatial features include determining whether the vehicle enters sensitive areas (such as schools and residential areas) and calculating the estimated arrival time at the next node. The business features include extracting the order priority (such as "fresh food cold chain needs to be delivered with priority") and statistically calculating the current link congestion index (based on historical travel time).

[0114] The historical trends and real-time values in the statistical time dimension are counted, and the current location in the spatial dimension is associated with the risk attributes of upstream and downstream nodes. The GPS trajectory data is combined with weather API data (such as rainstorm warnings) to obtain multi-modal real-time monitoring data. The multi-modal risk control decision-making model uses decision trees for policy arbitration based on the multi-modal real-time monitoring data. Policy arbitration includes rule priority and model-assisted decision-making. For example, in terms of rule priority: if there are hard rules in the comparison table (such as "overloading must be intercepted"), the interception action is directly triggered. Model-assisted decision-making means that in scenarios not covered by the rules, recommended actions are generated based on the model scores. Then, in combination with predefined executable operation templates (such as "send warning text messages" and "trigger electronic fence interception"), corresponding instruction templates are selected according to the action type to issue dynamic risk control instructions.

[0115] It should be understood that the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0116] In one embodiment, a logistics full-link risk control system based on an AI large model is provided. This logistics full-link risk control system based on an AI large model corresponds to the logistics full-link risk control method based on an AI large model in the above embodiment.

[0117] The logistics full-link risk control system based on an AI large model includes an interaction module, a parsing and generating module, an optimization module, and a pushing and executing module. The detailed descriptions of each functional module are as follows: The interaction module is used to obtain the risk control requirements of logistics operations input by users through a graphical user interface; A parsing and generating module, which is used to parse the risk control requirements of logistics operations, extract the logistics risk elements therein, and generate structured risk control rules based on the industry knowledge base and historical business data by combining machine learning algorithms and pre-trained large models; An optimization module, which is used to identify and resolve rule conflicts in the structured risk control rules using a collision detection algorithm, and dynamically adjust the rule thresholds according to business volatility to obtain optimized risk control rules; A pushing and executing module, which is used to push the optimized risk control rules into the business supervision system through message queue or API interface technology, and determine whether to trigger interception, warning, or transportation route correction operations at each node of the business system.

[0118] For the specific limitations of the AI large model-based logistics full-link risk control system, reference can be made to the limitations of the AI large model-based logistics full-link risk control method in the above text, which will not be elaborated here; each module in the above AI large model-based logistics full-link risk control system can be implemented in whole or in part by software, hardware, and their combinations; the above modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0119] In one embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 3 shown. The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store structured risk control rules, collision detection algorithms, etc. The network interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements an AI large model-based logistics full-link risk control method.

[0120] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented: S1: Obtain the risk control requirements of logistics operations input by the user through an interactive interface; S2: Parse the risk control requirements of logistics operations, extract logistics risk elements, combine machine learning algorithms and large models, and generate structured risk control rules based on the industry knowledge base and business data; S3: Identify and resolve rule conflicts in the structured risk control rules using a collision detection algorithm; dynamically adjust the rule thresholds based on business volatility to obtain optimized risk control rules; S4: Push the risk control rules to the business supervision system based on a message queue or API interface, and determine whether to trigger interception, warning, or transportation route correction operations based on the business nodes of the business system.

[0121] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: S1: Obtain the risk control requirements of the logistics business input by the user through an interactive interface; S2: Analyze the risk control requirements of the logistics business, extract logistics risk factors, and combine machine learning algorithms and large models to generate structured risk control rules based on an industry knowledge base and business data; S3: Identify and resolve rule conflicts in the structured risk control rules using a collision detection algorithm; dynamically adjust the rule thresholds based on business volatility to obtain optimized risk control rules; S4: Push the risk control rules to the business supervision system based on a message queue or API interface, and determine whether to trigger interception, warning, or transportation route correction operations based on the business nodes of the business system.

[0122] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can 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), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0123] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is used as an example for illustration. In actual applications, the above functions can be assigned to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0124] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the features; and these modifications or replacements 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 application, and should all be included in the protection scope of the present application.

Claims

1. Logistics full-link risk control method based on AI large model, characterized in that, It includes: Obtain the risk control requirements of the logistics business input by the user through the interactive interface; Analyze the risk control requirements of the logistics business, extract logistics risk elements, and combine machine learning algorithms and large models to generate structured risk control rules based on the industry knowledge base and business data; Use the collision detection algorithm to identify and resolve the rule conflicts of the structured risk control rules; dynamically adjust the rule thresholds based on business volatility to obtain optimized risk control rules; Push the risk control rules to the business supervision system based on the message queue or API interface, and determine whether to trigger interception, warning, or transportation route correction operations based on the business nodes of the business system.

2. The AI large model-based full-link logistics risk control method according to claim 1, wherein, The generation of the structured risk control rules includes: Perform semantic slicing on the risk control requirements of the logistics business through a word segmentation model, and use the BERT model to extract risk characteristics including goods categories, transportation attributes, and environmental parameters to obtain logistics risk elements; Conduct classification modeling based on historical violation data to generate risk scoring rules; Combine the predefined compliance template library and the logistics risk elements, and generate standardized risk control terms through a rule matching algorithm to obtain structured risk control rules.

3. The AI large model-based full-link risk control method for logistics according to claim 1, wherein, The steps of dynamically adjusting the rule thresholds further include: Adopt a sliding window algorithm to perform time series analysis on the business data for a specified time period, and calculate the probability distribution of the occurrence of risk times; Based on the probability distribution characteristics, generate extreme scenario stress test data through Monte Carlo simulation, and dynamically correct the upper and lower limits of the rule thresholds; Based on the corrected threshold range parameters, use a reinforcement learning model to perform parameter optimization to determine the optimal threshold combination that maximizes the ratio of the risk interception rate to the business passing rate.

4. The AI large model-based full-link logistics risk control method according to claim 1, wherein, The business nodes include order fulfillment, transportation route, and delivery acceptance. Pushing the risk control rules to the business supervision system based on the message queue or API interface further includes: Through the collaboration of the real-time risk control intelligent agent and the offline risk control intelligent agent by the workflow orchestration engine, the real-time risk control intelligent agent processes instant risk events based on the message queue, and the offline risk control intelligent agent periodically analyzes historical data through the API interface and updates the risk control rule library; Adopt a dual-mode risk control decision tree to perform real-time interception and offline risk assessment in parallel at the business nodes, and determine the final risk control strategy through a weighted voting mechanism; Build a logistics risk simulation environment based on digital twin technology, simulate the impact of rule changes on the entire logistics link, obtain risk assessment data, and generate a risk assessment report and synchronize it to the business supervision system.

5. The AI large model-based full-link risk control method for logistics according to claim 1 or 2, characterized in that, The method further includes: Obtain logistics basic data and risk control design parameters, and based on the logistics basic data and the risk control design parameters, judge the first predicted risk value of each node and the second predicted risk value of the entire link during the goods transportation process; Based on the risk control design parameters, the first predicted risk value, and the second predicted risk value, generate a logistics risk control comparison table for dynamically adjusting the risk control strategy; Obtain the logistics risk control reference range including historical risk disposal records, and combine the logistics risk control comparison table and the logistics risk control reference range to build a multi-modal risk control decision model; Real-time obtain logistics operation status data, and input the operation status data into the multi-modal risk control decision model to generate dynamic risk control instructions.

6. The AI large model-based full logistics link risk control method according to claim 5, wherein, Judging the first predicted risk value of each node and the second predicted risk value of the entire link during the goods transportation process, including: Extracting logistics basic data from the business database, where the logistics basic data includes goods attribute parameters, transportation network topology, carrier capacity matrix, and environmental monitoring data; the risk control design parameters include node risk thresholds, link fault tolerance rates, and emergency response levels; Analyzing the coupling relationship between goods attribute parameters and transportation network topology through a spatio-temporal prediction algorithm to judge the first predicted risk value of the loading and unloading nodes; Using a graph neural network to analyze the interaction between the carrier capacity matrix and environmental monitoring data to determine the second predicted risk value of the transfer nodes.

7. The AI large model-based full logistics link risk control method according to claim 5, wherein, Generating a logistics risk control comparison table specifically includes: Generating a risk propagation path based on Monte Carlo simulation and calculating the theoretical risk propagation probability in combination with node risk thresholds; Extracting feature vectors from historical risk events and constructing a risk feature library through contrastive learning; Establishing a mapping relationship between risk levels and disposal measures according to the matching degree between the theoretical risk propagation probability and the risk feature library; Calculating the difference in risk disposal time limits under different transportation scenarios and generating a logistics risk control comparison table including emergency response levels; Or, Constructing the multi-modal risk control decision model specifically includes: Building a hybrid architecture integrating a time series prediction network and a graph attention mechanism, and receiving real-time operation status data and the logistics risk control comparison table at the input end; Setting up a closed-loop decision-making unit including a reinforcement learning module, a policy generation module, and an effect evaluation module; Performing policy optimization in a logistics risk simulation environment to generate a risk control decision tree with adaptive capabilities.

8. An AI large model-based full-link risk control system for logistics, characterized in that, Applied to the AI large model-based logistics full-link risk control method according to any one of claims 1-7, the system includes: An interaction module for obtaining the logistics business risk control requirements input by the user through a graphical user interface; An analysis and generation module for analyzing the logistics business risk control requirements, extracting the logistics risk elements therein, and generating structured risk control rules according to the industry knowledge base and historical business data in combination with machine learning algorithms and pre-trained large models; An optimization module for using a collision detection algorithm to identify and resolve rule conflicts in the structured risk control rules, and dynamically adjusting the rule thresholds according to business volatility to obtain optimized risk control rules; A push and execution module for pushing the optimized risk control rules to the business supervision system through message queue or API interface technology, and judging whether to trigger interception, warning, or transportation path correction operations at each node of the business system.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it realizes the steps of the AI large model-based logistics full-link risk control method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it realizes the steps of the AI large model-based logistics full-link risk control method according to any one of claims 1 to 7.

Images