Supply chain AI risk control risk agent effect evaluation and optimization method and system
By establishing a multi-dimensional quantitative indicator system for risk agency effectiveness and using AI model evaluation, and dynamically optimizing resource allocation, the problems of single-dimensional risk agency evaluation and rigid resource allocation in supply chain finance have been solved, thereby improving risk control efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CCCC(XIAMEN)INFORMATION CO LTD
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-12
AI Technical Summary
Existing supply chain finance platforms have a single dimension for risk agency assessment, rigid resource allocation, and a lack of effective mechanisms for eliminating inefficient agents, resulting in one-sided assessment results and wasted resources, which affects risk control efficiency.
Establish a multi-dimensional quantitative indicator system for risk agency performance, use AI models to evaluate agency performance in real time, dynamically allocate tasks through load balancing algorithms, and establish an elimination mechanism to optimize resource allocation.
It achieves optimal allocation of risk agency resources, improves the overall stability of supply chain risk control and the efficiency of risk disposal, and is applicable to various supply chain finance scenarios.
Smart Images
Figure CN122198646A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of supply chain finance risk control technology, and in particular to a method and system for evaluating and optimizing the effectiveness of AI-based risk control agents in the supply chain. Background Technology
[0002] In supply chain finance, risk agency mechanisms are crucial for mitigating risk and improving risk management efficiency. While existing supply chain finance platforms have adopted risk agency models, significant shortcomings exist in practical application: First, the evaluation of risk agency effectiveness is often one-dimensional, relying solely on compensation amount or speed without adequately considering key factors such as risk coverage, cost control, and compliance, leading to biased assessments. Second, rigid allocation of agency resources, employing a fixed distribution model, fails to dynamically adjust task allocation based on agency performance, resulting in wasted high-quality resources and inefficient resource utilization. Third, the lack of an effective mechanism for eliminating inefficient agencies means that some agencies with persistent issues such as untimely compensation and insufficient risk management capabilities cannot be promptly removed, impacting overall risk control agency efficiency.
[0003] While existing technologies have established multiple lines of risk control, they lack a systematic evaluation and optimization of the performance of risk brokerage institutions. Furthermore, related supply chain finance risk control technologies primarily focus on transaction risk identification, neglecting the dynamic management of brokerage resources. Therefore, there is an urgent need for a technological solution capable of comprehensively evaluating the effectiveness of risk brokerage and dynamically optimizing resource allocation to address the shortcomings of existing technologies. Summary of the Invention
[0004] The purpose of this invention is to propose a method for evaluating and optimizing the effectiveness of risk agency in AI-based supply chain risk control, aiming to solve the problems of single evaluation dimensions, rigid resource allocation, and insufficient agency efficiency in existing supply chain risk control.
[0005] This invention is implemented as follows: a method for evaluating and optimizing the effectiveness of AI-based risk control agency in supply chains, the method comprising the following steps: Establish a quantitative indicator system for the effectiveness of risk agency; Real-time collection of supply chain transaction data, risk agency execution data, risk control result data, and basic information of agency institutions; deduplication, outlier removal, and standardization processing of the collected data to form a standardized evaluation dataset; An AI evaluation model based on the fusion of gradient boosting tree and attention mechanism is constructed. The model is trained using a standardized evaluation dataset, and the model parameters are optimized through cross-validation to obtain the optimal evaluation model. The standardized data collected in real time is input into the optimal evaluation model, and the comprehensive evaluation score and individual indicator score of each agency are output. Based on the comprehensive evaluation score, an agent resource optimization model is constructed. A load balancing algorithm is used to allocate supply chain risk control tasks of different risk levels to agents of corresponding evaluation levels, and the task acceptance limit and permissions of the agents are adjusted. Set comprehensive evaluation score thresholds and individual indicator score thresholds, formulate an elimination mechanism, stop assigning new tasks to agencies that meet the elimination mechanism, and update the agency resource pool.
[0006] Another objective of this invention is to propose a supply chain AI risk control agency effect evaluation and optimization system.
[0007] The system includes: The indicator system construction module is used to establish a quantitative indicator system for the effectiveness of risk agency. The data acquisition and preprocessing module is used to collect supply chain transaction data, risk agency execution data, risk control result data, and basic information of agency institutions in real time; it performs deduplication, outlier removal, and standardization on the collected data to form a standardized evaluation dataset. The AI model training and evaluation module is used to build an AI evaluation model based on the fusion of gradient boosting tree and attention mechanism. The model is trained using a standardized evaluation dataset, and the model parameters are optimized through cross-validation to obtain the optimal evaluation model. The standardized data collected in real time is input into the optimal evaluation model, and the comprehensive evaluation score and individual indicator score of each agency are output. The agent resource dynamic optimization module is used to build an agent resource optimization model based on the comprehensive evaluation score, and use a load balancing algorithm to allocate supply chain risk control tasks of different risk levels to agents of corresponding evaluation levels, and adjust the task acceptance limit and permissions of agents. The agent node elimination module is used to set the comprehensive evaluation score threshold and the individual indicator score threshold, formulate the elimination mechanism, stop the allocation of new tasks to agents that meet the elimination mechanism, and update the agent resource library.
[0008] Beneficial effects of the present invention This invention proposes a method and system for evaluating and optimizing the effectiveness of AI-based risk agency in supply chain risk control, relating to the field of supply chain finance risk control technology. The method establishes a quantitative indicator system for risk agency effectiveness, encompassing multiple dimensions such as timely compensation and risk coverage. It utilizes an AI model to collect real-time supply chain transaction data, risk agency execution data, and risk control result data to dynamically evaluate the service performance of agency institutions. Based on the evaluation results, an agency resource optimization model is constructed. This model dynamically allocates agency tasks and adjusts agency permissions through a load balancing algorithm and establishes a mechanism to eliminate inefficient agency nodes, achieving optimal allocation of risk agency resources. This invention solves the problems of single-dimensional risk agency evaluation, rigid resource allocation, and insufficient agency efficiency in existing supply chain risk control systems, significantly improving the overall stability and risk handling efficiency of AI-based supply chain risk control, and is applicable to various supply chain finance scenarios. Attached Figure Description
[0009] Figure 1 This is a flowchart of a preferred embodiment of the present invention for evaluating and optimizing the effectiveness of AI-based risk control agency in the supply chain; Figure 2 This is a structural diagram of a supply chain AI risk control risk agency effect evaluation and optimization system according to a preferred embodiment of the present invention. Detailed Implementation
[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. For ease of explanation, only the parts related to the embodiments of this invention are shown. It should be understood that the specific embodiments described herein are merely for explaining this invention and are not intended to limit this invention.
[0011] This invention proposes a method and system for evaluating and optimizing the effectiveness of AI-based risk agency in supply chain risk control, relating to the field of supply chain finance risk control technology. The method establishes a quantitative indicator system for risk agency effectiveness, encompassing multiple dimensions such as timely compensation and risk coverage. It utilizes an AI model to collect real-time supply chain transaction data, risk agency execution data, and risk control result data to dynamically evaluate the service performance of agency institutions. Based on the evaluation results, an agency resource optimization model is constructed. This model dynamically allocates agency tasks and adjusts agency permissions through a load balancing algorithm and establishes a mechanism to eliminate inefficient agency nodes, achieving optimal allocation of risk agency resources. This invention addresses the problems of single-dimensional risk agency evaluation, rigid resource allocation, and insufficient agency efficiency in existing supply chain risk control systems, significantly improving the overall stability and risk handling efficiency of AI-based supply chain risk control, and is applicable to various supply chain finance scenarios.
[0012] Figure 1 This is a flowchart of a preferred embodiment of the present invention for evaluating and optimizing the effectiveness of AI-based risk control agency in the supply chain; the method includes the following steps: S1. Establish a quantitative indicator system for the effectiveness of risk agency; The quantitative indicator system includes core indicators and auxiliary indicators; the core indicators include at least the timeliness of compensation payment, risk coverage, compensation accuracy rate, and agency cost control rate; the auxiliary indicators include at least the agency response speed, risk handling compliance rate, and cooperation stability. Key metrics: Timeliness of compensation: This reflects the efficiency of the agency in completing compensation after a risk event occurs. It is calculated as the ratio of the actual compensation completion time to the agreed compensation time limit, with a value range of [0, 100]. The larger the value, the stronger the coverage. The agreed compensation time limit is set according to the risk event level (e.g., high-risk events ≤ 72 hours, medium-risk events ≤ 120 hours, low-risk events ≤ 168 hours).
[0013] Risk Coverage: Reflects the agency's ability to cover supply chain risks. It is calculated as the ratio of the number of successfully handled risk events to the total number of risk events. The value ranges from [0, 100], with a higher value indicating stronger coverage.
[0014] Claims accuracy rate: Reflects the compliance and accuracy of the agency's claims behavior, excluding erroneous claims, overpayments, etc.; it is calculated as the ratio of the number of risk events for which compliant claims are made to the total number of risk events for which the agency has made claims, with a value range of [0,100]. The larger the value, the higher the accuracy of the claims. Agency cost control rate: This reflects the agency's cost control capability. The budgeted agency cost is pre-determined based on factors such as task difficulty and risk level. It is calculated as the ratio of actual agency cost to budgeted agency cost, with a value ranging from [0, +∞). A value ≤ 100 indicates that cost control has met the target.
[0015] Auxiliary indicators: Agency response speed: The agency's first response time after a risk event is triggered, ≤24 hours is considered acceptable.
[0016] Risk management compliance rate: The proportion of risk management procedures, documents, etc. of the agency that comply with regulatory requirements and cooperation agreements is ≥90% to be considered compliant.
[0017] Stability of cooperation: The duration of cooperation between the agency and the platform, and the period during which there is no record of breach of contract.
[0018] In this embodiment of the invention, the weights of each indicator are determined by the Analytic Hierarchy Process (AHP); the core indicators account for 80% and the auxiliary indicators account for 20%; among the core indicators, the weights of timely compensation, risk coverage, compensation accuracy, and agency cost control rate are 30%, 25%, 20%, and 5%, respectively; among the auxiliary indicators, the weights of agency response speed, risk handling compliance rate, and cooperation stability are 8%, 7%, and 5%, respectively.
[0019] In this embodiment of the invention, the specific process of determining the weights of each indicator using the Analytic Hierarchy Process (AHP) is as follows: S11, establish a three-level hierarchical structure; Target layer: Comprehensive evaluation of risk agency effectiveness (core objective: to adapt to the supply chain risk control requirements of "efficiency first, compliance as a safety net"); Criteria Level: Core Indicators and Auxiliary Indicators (two types of criteria); Solution layer: Specific indicators (core indicators and auxiliary indicators); S12, construct a two-layer judgment matrix; The first layer (target layer → criteria layer): Supply chain risk control experts and agency operation personnel can be invited to conduct pairwise comparisons and scores of "core indicators" and "auxiliary indicators" based on the business logic that "core indicators directly determine agency efficiency, and auxiliary indicators are only supplementary" (1-9 scale method). For example, the importance of core indicators to the evaluation target is 4 times that of auxiliary indicators, and the scoring matrix is "1,4; 1 / 4,1". The second layer (criteria layer → scheme layer): compares and scores the specific indicators within the core indicators and the auxiliary indicators pairwise (e.g., in the core indicators, "timeliness of compensation > risk coverage"). S13, Consistency check and weight calculation; Perform CR checks (CR < 0.1) on both judgment matrices to ensure logical consistency; The weights are calculated using the eigenvalue method: the first layer calculates the weights of the criteria layer (80% for core indicators and 20% for auxiliary indicators), and the second layer calculates the weights of each specific indicator under the corresponding criteria layer (e.g., 30% for timely compensation within the core indicator). The final total weight of each specific indicator = weight of the criteria layer × weight of the indicator under the criteria layer (e.g., total weight of timely compensation = 80% × 30% = 24%).
[0020] S2 collects supply chain transaction data, risk agency execution data, risk control result data, and basic information of agency institutions in real time; it also performs deduplication, outlier removal, and standardization on the collected data to form a standardized evaluation dataset. In this embodiment of the invention, data can be collected by connecting to supply chain finance platforms, agency business systems, risk control systems, and regulatory systems via API interfaces; Supply chain transaction data: including procurement contract data, logistics and distribution data, fund settlement data, invoice data, etc.; Risk-based agency execution data includes claims application data, claims completion time, claims amount, and agency cost details. Risk control results data include: risk event level, risk occurrence time, risk handling results, and whether compensation is compliant; Basic information about the agency: including its qualification level, duration of cooperation, and history of defaults.
[0021] Data preprocessing includes: Data cleaning: Outlier detection algorithms (such as the IQR rule) are used to remove data that exceeds a reasonable range, such as data whose compensation time exceeds three times the agreed time limit; Data standardization: Normalize the indicator data of different dimensions and convert them to the range of [0,100] to facilitate comprehensive evaluation; Data completion: Missing data are completed using interpolation or AI model-based prediction methods (such as linear regression, random forest, LSTM, GRU, etc.) to ensure that the dataset completeness is ≥98%.
[0022] S3. Construct an AI evaluation model based on the fusion of gradient boosting tree and attention mechanism. Train the model using a standardized evaluation dataset, optimize the model parameters through cross-validation, and obtain the optimal evaluation model. Input the standardized data collected in real time into the optimal evaluation model, and output the comprehensive evaluation score and individual indicator score of each agency. In this embodiment of the invention, the training and evaluation process of the AI evaluation model includes: Model Construction: The basic evaluation model is constructed using the XGBoost algorithm (Gradient Boosting Tree algorithm). An attention mechanism is introduced to assign dynamic weights to core indicators, and the influence weights of each indicator are adjusted in real time according to the type of risk event and the supply chain scenario. For example, for the supply chain risk control agency evaluation scenario, the loss function of XGBoost is customized to introduce a risk agency compensation cost weight term (loss function = basic mean square error + α × compensation cost deviation, where α is a scenario-based adjustment coefficient with a value of 0.8-1.2).
[0023] Model Training: The standardized evaluation dataset is divided into a training set and a test set according to a preset ratio (e.g., 7:3). The training set is used for model parameter learning, and the test set is used for model performance validation. The key hyperparameters of the model are globally optimized using a preset hyperparameter optimization algorithm, so that the model's evaluation accuracy on the test set is greater than or equal to a preset accuracy threshold (e.g., 95%), and the mean squared error is less than or equal to a preset error threshold (e.g., 0.05).
[0024] The hyperparameter optimization algorithm includes, but is not limited to, hyperparameters such as grid search optimization learning rate (0.01-0.1), tree depth (3-10), and number of iterations (100-1000). Real-time evaluation: Input the pre-processed real-time data into the trained AI evaluation model. The model outputs the comprehensive evaluation score (out of 100) and individual indicator scores for each agency. Set the evaluation frequency (e.g., one evaluation cycle every 30 days) and the real-time evaluation result update frequency (e.g., evaluation result update frequency ≥ 1 time / 24 hours).
[0025] For example, in this embodiment of the invention, when training the AI evaluation model, four core indicators are incorporated into the quantitative evaluation system, and their weighting in the overall score is clearly defined: Agency cost control rate (core indicator): accounts for 5% of the overall score. Together with the timeliness of claims payment, risk coverage, and accuracy of claims payment, it constitutes the core indicator score (total weight 80%). The model uses this indicator to quantify the cost control capability of the agency. A cost control rate ≤100% gets full marks, and >100% deducts points proportionally (e.g., 120% gets 3 points, 150% gets 0 points). Agent response speed (auxiliary indicator): accounts for 8% of the overall score. Response time ≤ 24 hours gets full marks, 24-48 hours gets 50% of the score, and > 48 hours gets 0 points. Risk handling compliance rate (auxiliary indicator): accounts for 7% of the overall score. A compliance rate of ≥90% gets full marks, 80%-90% gets 50% of the score, and <80% gets 0 marks. Stability of cooperation (auxiliary indicator): accounting for 5% of the overall score. A full score is awarded for a cooperation duration of ≥2 years with no breach of contract, 50% of the score is awarded for 1-2 years with no breach of contract, and 0 points are awarded for <1 year or with a breach of contract record. The four core indicators, together with other auxiliary indicators, constitute a complete scoring system. The model outputs the comprehensive evaluation score and individual indicator scores of the agency through this system, providing data support for subsequent resource optimization and elimination mechanisms.
[0026] S4. Based on the comprehensive evaluation score, an agent resource optimization model is constructed. A load balancing algorithm is used to allocate supply chain risk control tasks of different risk levels to agents of corresponding evaluation levels, and the task acceptance limit and permissions of the agents are adjusted. In this embodiment of the invention, the agency is divided into three levels based on the comprehensive evaluation score: High-quality agents: score ≥ 85 points, possessing high risk handling and cost control capabilities; Qualified Agent: 60 points ≤ score < 85 points, possesses basic risk management capabilities, but needs continuous improvement; Agents needing improvement: Score < 60 points, insufficient risk management capabilities, business scope needs to be limited.
[0027] Task allocation optimization: A load balancing algorithm is used to dynamically allocate supply chain risk control tasks according to risk level (high, medium, low) and task complexity based on the agency's level, current task saturation, and historical performance. Premium Agents: Accept high-risk, highly complex tasks; unlimited task limits can be set. Qualified Agent: Accepts low to medium risk and moderately complex tasks; can set the task acceptance limit to not exceed the first percentage (e.g., 80%) of its highest historical completed task. Agents to be improved: Only accept low-risk, simple and complex tasks; the task acceptance limit can be set to not exceed the second largest percentage of its highest historical completion amount (e.g., 50%).
[0028] Permission Adjustment: The system permissions of agencies will be dynamically adjusted based on the evaluation results. High-performing agencies will be granted priority in order acceptance and fast approval. Agencies that need improvement will have approval nodes added and their independent decision-making authority restricted.
[0029] In this embodiment of the invention, the load balancing algorithm may be either the Weighted Least Connections (WLC) algorithm or the Risk-Adaptive Weighted Round Robin (RA-WRR) algorithm; For example, in one embodiment of the present invention, the agent resource optimization model uses the weighted least connections (WLC) algorithm for task allocation: First, basic weights are assigned to agents of different levels. The weight of a high-quality agent (score ≥ 85 points) is set to 5, the weight of a qualified agent (score ≤ 60 points < 85 points) is set to 3, and the weight of an agent that needs improvement (score < 60 points) is set to 1. At the same time, the number of risk control tasks (i.e., "connections") that each agent has not yet completed is counted in real time. The algorithm calculation formula is: task allocation priority = agent weight / number of currently uncompleted tasks. Agents with higher priority values are given priority to accept new tasks. For example: High-quality agent A (weight 5) currently has 10 unfinished tasks and a priority of 0.5; high-quality agent B (weight 5) currently has 5 unfinished tasks and a priority of 1. High-risk and urgent risk control tasks will be assigned to agent B first. If qualified agent C (weight 3) currently has 1 unfinished task and a priority of 3, but its weight is only 3, it is still lower than the priority of high-quality agent B (priority of 1) (the core logic of the algorithm is "weight priority, rebalancing the number of tasks"), ensuring that high-risk tasks are always undertaken by high-quality agents, while avoiding task overload of a single high-quality agent.
[0030] For example, in another embodiment of the present invention, the agent resource optimization model uses a risk-adaptive weighted round-robin (RA-WRR) algorithm for task allocation: First, a basic weight is assigned to each agent (8 for high-quality agents, 5 for qualified agents, and 2 for agents needing improvement). Then, the agent's specific risk control capabilities are combined (e.g., Agent A is skilled in handling fund misappropriation risks in the construction supply chain, with an adaptation weight of +2; Agent B is skilled in handling fraudulent transaction risks in the manufacturing supply chain, with an adaptation weight of +2). The comprehensive weight is calculated as: basic weight of the agent's level + risk type adaptation weight. The algorithm allocates tasks in round-robin according to the comprehensive weight ratio. For example, if there are high-quality agents A (overall weight 10), high-quality agents B (overall weight 8), and qualified agents C (overall weight 5), then in every 23 rounds of high-risk tasks, A will undertake 10, B 8, and C 5. Only A / B will undertake high-risk tasks specifically for the construction / manufacturing industry, while C will only undertake general medium-risk tasks. At the same time, after each round of inquiry, the algorithm automatically updates the real-time evaluation score and suitability weight of the agency (if A makes two consecutive mistakes in handling the risk of misappropriation of funds, the suitability weight will drop to 0), dynamically adjusting the round of inquiry ratio to ensure the task proportion of high-quality agents and achieve precise matching between risk types and agency capabilities.
[0031] S5. Set comprehensive evaluation score thresholds and individual indicator score thresholds, formulate an elimination mechanism, stop assigning new tasks to agencies that meet the elimination mechanism, and update the agency resource pool.
[0032] In one embodiment of the present invention, a comprehensive evaluation score threshold is set (e.g., 60 points). Set thresholds for individual core indicators: risk coverage threshold (e.g., 60 points), payout accuracy threshold (e.g., 70%). The elimination mechanism is as follows: Upon meeting any of the following elimination conditions, the elimination process is triggered: For agencies that trigger the elimination conditions, the system automatically suspends their new task assignment permissions, removes them from the agency resource pool, and updates the resource configuration scheme. In a specific implementation, an elimination warning notification can be sent to the platform operator; the operator completes manual review within a preset time limit (e.g., within 7 working days), and upon successful review, officially removes the agency from the agency resource pool and updates the resource configuration scheme.
[0033] Elimination criteria include: If an agency's overall evaluation score is lower than the overall evaluation score threshold for N consecutive evaluation cycles (N is 1-3 cycles, each cycle is 15-30 days, depending on the efficiency of business operations); In a single assessment, the payout accuracy rate is lower than the payout accuracy rate threshold and the risk coverage is lower than the risk coverage threshold; Any entity found to have committed a major compliance risk event (such as malicious erroneous claims, falsification of risk control data, etc.) will be immediately eliminated. In another embodiment of the present invention, the single core indicator may further include the timeliness of compensation payment; correspondingly, the elimination mechanism may further include the following elimination conditions: If the timeliness of claims payment is below the timeliness threshold for two consecutive assessment cycles (≤80 points), or if the timeliness of claims payment is below the serious non-compliance threshold for a single core indicator in a single assessment (this threshold is 70%-80% of the basic threshold for timeliness of claims payment, corresponding to serious inefficiency, such as ≤60 points); Optionally, in some embodiments of the present invention, the elimination mechanism may further include the following elimination conditions: If the agency cost control rate exceeds the first threshold (e.g., 150%) for two consecutive evaluation periods, or if the agency cost control rate exceeds the second threshold (e.g., 200%) in a single evaluation, it indicates severe cost out-of-control. The agent's response speed is greater than the first threshold for agent response speed (e.g., 72 hours) for two consecutive evaluation cycles, or the agent's response speed is greater than the second threshold for agent response speed (e.g., 120 hours, indicating severe response lag) in a single evaluation. The compliance rate of risk handling is less than the first threshold (e.g., 80%) for two consecutive assessment cycles, or the compliance rate is less than the second threshold (e.g., 70%, serious compliance deficiency) in a single assessment. If a new major breach of contract occurs in the stability of the cooperation (such as maliciously delaying tasks or providing false compliance materials), or if the task is refused to be assigned without justifiable reason for one consecutive assessment cycle; Corresponding to the supply chain AI risk control risk agency effect evaluation and optimization method described in the above embodiment, Figure 2 This diagram illustrates a structural block diagram of a supply chain AI risk control risk agency effectiveness evaluation and optimization system provided in an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The system includes:
[0034] The indicator system construction module is used to establish a quantitative indicator system for the effectiveness of risk agency. The data acquisition and preprocessing module is used to collect supply chain transaction data, risk agency execution data, risk control result data, and basic information of agency institutions in real time; it performs deduplication, outlier removal, and standardization on the collected data to form a standardized evaluation dataset. The AI model training and evaluation module is used to build an AI evaluation model based on the fusion of gradient boosting tree and attention mechanism. The model is trained using a standardized evaluation dataset, and the model parameters are optimized through cross-validation to obtain the optimal evaluation model. The standardized data collected in real time is input into the optimal evaluation model, and the comprehensive evaluation score and individual indicator score of each agency are output. The agent resource dynamic optimization module is used to build an agent resource optimization model based on the comprehensive evaluation score, and use a load balancing algorithm to allocate supply chain risk control tasks of different risk levels to agents of corresponding evaluation levels, and adjust the task acceptance limit and permissions of agents. The agent node elimination module is used to set the comprehensive evaluation score threshold and the individual indicator score threshold, formulate the elimination mechanism, stop the allocation of new tasks to agents that meet the elimination mechanism, and update the agent resource library.
[0035] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by program instructions and related hardware. The program can be stored in a computer-readable storage medium, such as ROM, RAM, disk, optical disk, etc.
[0036] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for evaluating and optimizing the effectiveness of AI-based risk control agency in supply chains, characterized in that, The process includes the following steps: Establish a quantitative indicator system for the effectiveness of risk agency; Real-time collection of supply chain transaction data, risk agency execution data, risk control result data, and basic information of agency institutions; deduplication, outlier removal, and standardization processing of the collected data to form a standardized evaluation dataset; An AI evaluation model based on the fusion of gradient boosting tree and attention mechanism is constructed. The model is trained using a standardized evaluation dataset, and the model parameters are optimized through cross-validation to obtain the optimal evaluation model. The standardized data collected in real time is input into the optimal evaluation model, and the comprehensive evaluation score and individual indicator score of each agency are output. Based on the comprehensive evaluation score, an agent resource optimization model is constructed. A load balancing algorithm is used to allocate supply chain risk control tasks of different risk levels to agents of corresponding evaluation levels, and the task acceptance limit and permissions of the agents are adjusted. Set comprehensive evaluation score thresholds and individual indicator score thresholds, formulate an elimination mechanism, stop assigning new tasks to agencies that meet the elimination mechanism, and update the agency resource pool.
2. The supply chain AI risk control risk agency effect evaluation and optimization method as described in claim 1, characterized in that, The quantitative indicator system includes core indicators and auxiliary indicators; the core indicators include at least the timeliness of compensation, risk coverage, compensation accuracy, and agency cost control rate; the auxiliary indicators include at least the agency response speed, risk handling compliance rate, and cooperation stability.
3. The supply chain AI risk control risk agency effect evaluation and optimization method as described in claim 2, characterized in that, Key metrics include: Timeliness of compensation: This reflects the efficiency of the agency in completing compensation after a risk event occurs. It is calculated as the ratio of the actual compensation completion time to the agreed compensation time limit, with a value range of [0,100]. Risk coverage: Reflects the agency's ability to cover supply chain risks. It is calculated as the ratio of the number of successfully handled risk events to the total number of risk events, with a value range of [0, 100]. Claims accuracy rate: Reflects the compliance and accuracy of the agency's claims behavior, excluding erroneous claims, overpayments, etc.; it is calculated as the ratio of the number of risk events for which compliant claims are made to the total number of risk events for which the agency has made claims, with a value range of [0,100]. Agency cost control rate: Reflects the cost control capability of the agency. The budgeted agency cost is pre-determined based on factors such as task difficulty and risk level. It is calculated as the ratio of actual agency cost to budgeted agency cost, with a value range of [0,+∞). Supporting indicators include: Agency response speed: The time it takes for the agency to respond to a risk event after it is triggered; Risk management compliance rate: The percentage of risk management procedures, documentation, etc., of the agency that comply with regulatory requirements and cooperation agreements; Stability of cooperation: The duration of cooperation between the agency and the platform, and the period during which there is no record of breach of contract.
4. The supply chain AI risk control risk agency effect evaluation and optimization method as described in claim 2 or 3, characterized in that, The weights of each indicator are determined using the analytic hierarchy process (AHP).
5. The supply chain AI risk control risk agency effect evaluation and optimization method as described in claim 4, characterized in that, The training and evaluation process of the AI evaluation model includes: using the XGBoost algorithm to build a basic evaluation model, introducing an attention mechanism to assign dynamic weights to core indicators, and adjusting the influence weights of each indicator in real time according to the type of risk event and the supply chain scenario. Model training: The standardized evaluation dataset is divided into training set and test set according to a preset ratio. The training set is used for model parameter learning, and the test set is used for model performance verification. The key hyperparameters of the model are globally optimized by a grid search preset hyperparameter optimization algorithm so that the evaluation accuracy of the model on the test set is greater than or equal to the preset accuracy threshold and the mean squared error is less than or equal to the preset error threshold. Real-time evaluation: The pre-processed real-time data is input into the trained AI evaluation model, and the model outputs the comprehensive evaluation score and individual indicator score of each agency.
6. The supply chain AI risk control risk agency effect evaluation and optimization method as described in claim 5, characterized in that, The hyperparameter optimization algorithm includes, but is not limited to, grid search optimization algorithms for learning rate, tree depth, and number of iterations.
7. The method for evaluating and optimizing the effectiveness of AI-based risk control agency in supply chains as described in claim 1, characterized in that, The load balancing algorithm can be either the weighted least connections algorithm or the risk-adaptive weighted round-robin algorithm.
8. The method for evaluating and optimizing the effectiveness of AI-based risk control agency in supply chains as described in claim 1, characterized in that, The agent resource optimization model classifies agents into three levels—high-quality agents, qualified agents, and agents needing improvement—based on their comprehensive evaluation scores. High-quality agents undertake high-risk, high-complexity tasks, while qualified agents undertake low-to-medium risk, medium-complexity tasks. The system needs improvement in allowing agents to undertake low-risk, simple-to-complex tasks; and in dynamically adjusting the system permissions of agents based on evaluation results.
9. The method for evaluating and optimizing the effectiveness of AI-based risk control agency in supply chains as described in claim 1, characterized in that, The elimination mechanism is as follows: if any elimination condition is met, the elimination process is triggered: for an agency that triggers the elimination condition, the system automatically suspends its new task assignment permission, removes it from the agency resource pool, and updates the resource configuration scheme. Elimination criteria include: An agency's overall evaluation score is lower than the overall evaluation score threshold for N consecutive evaluation periods; N is 1-3 periods. In a single assessment, the payout accuracy rate is lower than the payout accuracy rate threshold and the risk coverage is lower than the risk coverage threshold; A major compliance risk event has occurred.
10. A supply chain AI risk control risk agency effect evaluation and optimization system, characterized in that, The system includes: The indicator system construction module is used to establish a quantitative indicator system for the effectiveness of risk agency. The data acquisition and preprocessing module is used to collect supply chain transaction data, risk agency execution data, risk control result data, and basic information of agency institutions in real time; it performs deduplication, outlier removal, and standardization on the collected data to form a standardized evaluation dataset. The AI model training and evaluation module is used to build an AI evaluation model based on the fusion of gradient boosting tree and attention mechanism. The model is trained using a standardized evaluation dataset, and the model parameters are optimized through cross-validation to obtain the optimal evaluation model. The standardized data collected in real time is input into the optimal evaluation model, and the comprehensive evaluation score and individual indicator score of each agency are output. The agent resource dynamic optimization module is used to build an agent resource optimization model based on the comprehensive evaluation score, and use a load balancing algorithm to allocate supply chain risk control tasks of different risk levels to agents of corresponding evaluation levels, and adjust the task acceptance limit and permissions of agents. The agent node elimination module is used to set the comprehensive evaluation score threshold and the individual indicator score threshold, formulate the elimination mechanism, stop the allocation of new tasks to agents that meet the elimination mechanism, and update the agent resource library.