An intelligent risk control method and system for insurance

By acquiring vehicle risk information in real time through IoT devices and multimodal big data models, precise intervention measures can be implemented, overcoming the limitations of static assessment models in traditional vehicle insurance, realizing dynamic risk management and premium adjustment, and reducing accident rates and costs.

CN123071085APending Publication Date: 2026-11-10SHENZHEN BEIDOU CHENYU TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611171711.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-04
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Traditional vehicle insurance relies on static risk assessment models, which are difficult to adapt to complex and ever-changing traffic and driving environments. The lack of model iteration mechanisms makes it impossible to identify dynamic risks in real time, resulting in high claims costs and adverse selection problems.

Method used

By acquiring multi-dimensional risk information in real time through IoT monitoring devices, processing this information using a target multimodal large model, outputting real-time risk values, and implementing corresponding intervention measures based on the risk level, including safety alerts, voice warnings, and manual escort, and dynamically adjusting insurance rates.

Benefits of technology

It enables real-time collection and precise intervention of vehicle risks across the entire domain, reduces accident rates, dynamically adjusts premiums, improves the accuracy of risk identification, forms a closed-loop intelligent risk control system, and incentivizes safe driving.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123071085A_ABST
    Figure CN123071085A_ABST
Patent Text Reader

Abstract

This invention relates to an intelligent risk control method, device, system, and computer-readable storage medium for insurance. The method includes: acquiring multi-dimensional risk information of an insured vehicle in real time via an IoT monitoring device; processing the input multi-dimensional risk information using a target multimodal large model to obtain a target real-time risk value for the insured vehicle; determining the corresponding risk level based on the target real-time risk value and implementing intervention measures according to the corresponding risk level; wherein different risk levels correspond to different intervention measures; collecting the real-time risk value of the insured vehicle within a preset time after the implementation of intervention measures, comparing it with the target implemented risk value to generate a comparison result, and adjusting the insurance premium rate of the insured vehicle based on the comparison result. This embodiment can achieve risk visualization, adapt to new vehicle scenarios, and reduce vehicle accident rates.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of risk control technology, and in particular to an intelligent insurance risk control method and an intelligent insurance risk control system. Background Technology

[0002] With the rapid growth of car ownership, the auto insurance market has also developed rapidly. Traditional auto insurance mainly relies on static risk assessment models, which are typically based on factors such as the owner's historical claims records, basic vehicle information (e.g., model and age), and the driver's age and gender. Insurance companies use these static variables to determine premium rates and process claims after an accident. However, with advancements in technology and improved data analysis capabilities, traditional static pricing and risk control models are gradually showing their limitations and are struggling to adapt to the complex and ever-changing traffic and driving environments of modern society.

[0003] While existing technologies have developed vehicle-mounted terminal monitoring solutions, they generally suffer from the following problems: lack of model iteration mechanisms, making it difficult to adapt to new risk scenarios. Summary of the Invention

[0004] Therefore, in order to overcome at least some of the defects and deficiencies of the prior art, this invention proposes an intelligent insurance risk control method to adapt to the risk scenarios of new vehicles.

[0005] On one hand, an intelligent risk control method for insurance proposed in this invention includes acquiring multi-dimensional risk information of an insured vehicle in real time through an Internet of Things (IoT) monitoring device; wherein the multi-dimensional risk data includes abnormal driving behavior information of the driver, environmental status information of the vehicle's surroundings, static attribute information of the vehicle, and external correlation information; processing the input multi-dimensional risk information using a target multimodal large model to obtain a target real-time risk value of the insured vehicle; obtaining a corresponding risk level based on the target real-time risk value, and implementing intervention measures according to the corresponding risk level; wherein different risk levels correspond to different intervention measures; collecting the real-time risk value of the insured vehicle within a preset time after the implementation of the intervention measures, comparing it with the target real-time risk value to generate a comparison result, and adjusting the insurance premium rate of the insured vehicle based on the comparison result.

[0006] In one embodiment of the present invention, the abnormal driving behavior information of the driver includes at least one of the following collected by the driver state monitoring device: physiological fatigue, distracted driving, excessive driving time, smoking, making or receiving phone calls, not wearing a seat belt, taking both hands off the steering wheel, and deviation from the driving trajectory; the vehicle surrounding environment status information includes at least one of the following collected by the advanced driver assistance system: forward collision warning, lane departure warning, and blind spot monitoring information; the vehicle static attribute information includes at least one of the following: the vehicle model, service life, maintenance records, and historical accident records of the insured vehicle; the external associated information includes at least one of the following: regional weather warning, traffic index, and enterprise safety level.

[0007] In one embodiment of the present invention, the step of processing the input multi-dimensional risk information using a target multimodal large model to obtain the target real-time risk value of the insured vehicle includes: inputting the multi-dimensional risk information into an initial multimodal large model to obtain an initial prediction result group; classifying and labeling the initial prediction result group and manually correcting it to obtain a corrected risk prediction result group; iteratively processing the initial multimodal large model based on the corrected risk prediction information and the corresponding multi-dimensional risk information to obtain an iterative multimodal large model; evaluating the iterative multimodal large model to obtain a target multimodal large model; and processing the input multi-dimensional risk information in real time based on the target multimodal large model to obtain the target real-time risk value of the insured vehicle.

[0008] In one embodiment of the present invention, the model input information includes the proportion of drivers driving at night, the frequency of continuous driving time exceeding the limit, the frequency of fatigue alarms; the road section accident rate, the weather risk coefficient, the vehicle type risk coefficient; and the dynamic interaction risk characteristics between the driver's fatigue state and the slippery road surface.

[0009] In one embodiment of the present invention, the step of determining the corresponding risk level based on the real-time risk value and executing intervention measures according to the corresponding risk level specifically includes: when the real-time risk value is at a low risk level, executing a first type of intervention measure of pushing safety alert information to the user terminal; when the real-time risk value is at a medium risk level, executing a second type of intervention measure of sending voice warning information to the in-vehicle equipment of the insured vehicle and simultaneously providing targeted safety training content to the user terminal; and when the real-time risk value is at a high risk level, executing a third type of intervention measure of activating a human agent to conduct real-time voice communication through the IoT monitoring device and continuously and dynamically tracking the insured vehicle until the risk is eliminated.

[0010] In one embodiment of the present invention, the third type of intervention further includes: pushing the information of drivers marked as high-risk to the safety management terminal of their respective enterprises and generating an offline visit task list.

[0011] In one embodiment of the present invention, the step of collecting the real-time risk value of the insured vehicle after a preset time following the implementation of the intervention measures, comparing it with the target implemented risk value to generate a comparison result, and adjusting the insurance premium rate of the insured vehicle based on the comparison result includes: collecting the real-time risk value of the insured vehicle at preset intervals; comparing the real-time risk value with the target implemented risk value to obtain a comparison result; if the real-time risk value is not higher than the target implemented risk value, decreasing the insurance premium rate of the insured vehicle; if the real-time risk value is higher than the target implemented risk value, increasing the insurance premium rate of the insured vehicle; generating risk performance information based on the real-time risk value, and feeding it back into the target multimodal large model for optimization.

[0012] On the other hand, an insurance intelligent risk control device based on the above-mentioned insurance intelligent risk control method proposed in this embodiment of the invention includes: a multi-dimensional risk information acquisition module, used to acquire multi-dimensional risk information of the insured vehicle in real time through an Internet of Things monitoring device; wherein, the multi-dimensional risk data includes driver abnormal driving behavior information, vehicle surrounding environment status information, vehicle static attribute information, and external correlation information; a target real-time risk value acquisition module, used to process the input multi-dimensional risk information using a target multimodal large model to obtain the target real-time risk value of the insured vehicle; an intervention measure implementation module, used to obtain the corresponding risk level according to the target real-time risk value, and execute intervention measures according to the corresponding risk level; wherein, different risk levels correspond to different intervention measures; and a comparison result generation module, used to collect the real-time risk value of the insured vehicle within a preset time after the implementation of the intervention measures, compare it with the target real-time risk value to generate a comparison result, and adjust the insurance premium rate of the insured vehicle according to the comparison result.

[0013] Furthermore, the present invention provides an intelligent insurance risk control system, comprising a memory and a processor connected to the memory. The memory stores a computer program, and the processor executes the intelligent insurance risk control method described above when running the computer program.

[0014] In another aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein the stored computer program, when executed by a processor, can implement the insurance intelligent risk control method as described above.

[0015] As can be seen from the above, the technical features of this invention can have the following beneficial effects: This step uses IoT devices to achieve real-time collection of information on vehicles, drivers, road environment, and external factors, avoiding assessment bias caused by single data; it uses machine learning algorithms to fuse and calculate multi-source data, outputting quantifiable real-time risk values ​​to achieve risk visualization; the system automatically matches intervention measures from mild to severe according to the risk level to achieve precise intervention; after the intervention is completed, the system automatically collects data such as the risk reduction rate, accident rate, and alarm frequency, which are used to dynamically adjust premiums and incentivize safe driving; on the other hand, the effective data is fed back to the model to continuously improve the accuracy of risk identification, forming a closed-loop intelligent risk control throughout the entire process; the risk control model is continuously iterated based on real-time data feedback to adapt to new vehicle risks (such as new energy vehicles); compliance and social responsibility: helping customers improve their safety management capabilities and reduce the accident rate of insured vehicles. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating an intelligent risk control method for insurance provided in the first embodiment of the present invention.

[0018] Figure 2 for Figure 1 A flowchart illustrating the specific steps of step S12.

[0019] Figure 3 for Figure 1 A flowchart illustrating the specific steps of step S13.

[0020] Figure 4 for Figure 1 A flowchart illustrating the specific steps of step S14.

[0021] Figure 5 and Figure 6 This is a flowchart illustrating the risk intervention process in a specific embodiment of the present invention.

[0022] Figure 7 for Figure 1 The flowchart shown illustrates the specific formation process of the target multimodal large model.

[0023] Figure 8 This is a schematic diagram of the system framework of an insurance intelligent risk control system provided in the second embodiment of the present invention.

[0024] Figure 9This is a schematic diagram of the architecture of a vehicle monitoring cloud platform, which is a specific embodiment of an insurance intelligent risk control system.

[0025] Figure 10 A schematic diagram of a module for an intelligent insurance risk control device provided in the third embodiment of the present invention. Figure 11 This is a schematic diagram of a module of an insurance intelligent risk control system provided in the fourth embodiment of the present invention.

[0026] Figure 12 This is a schematic diagram of a storage medium provided in the fifth embodiment of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] In traditional insurance business, insurance companies mainly rely on the "post-accident claims" model: when an insured is involved in a traffic accident (such as a vehicle collision, being hit by illegally parked vehicles, etc.), they submit a claim application, and the insurance company completes the compensation through damage assessment and claims verification. However, this model has the following core problems: 1. Passivity: Insurance companies only intervene after a disaster occurs, and cannot prevent or reduce the probability of accidents in advance; 2. High claims costs: Frequent accidents involving certain high-risk groups (such as car owners with poor driving habits and small and medium-sized enterprises with weak safety management) result in a persistently high overall claims ratio; 3. Adverse selection: High-risk individuals are more inclined to purchase insurance, further increasing the payout pressure on insurance companies; 4. Lack of data linkage: Insurance risk assessment relies heavily on static information (such as age and historical accident records), making it difficult to reflect dynamic risk changes in real time (such as abnormal behaviors such as driver's current fatigue state, excessive driving time and speed, route deviation, abnormal changes in mileage, etc.).

[0029] With the development of technologies such as the Internet of Things, big data analytics, and artificial intelligence, proactive risk control services (such as in-vehicle terminal monitoring of driving behavior and monitoring equipment status) to identify and intervene in risks in advance have become a key direction for cost reduction and efficiency improvement in the insurance industry. However, existing technologies mostly focus on single scenarios (such as only for auto insurance or only providing early warning functions), lacking a systematic solution covering the entire chain of "risk identification - precise intervention - dynamic identification - effect feedback". Based on this, this application proposes an intelligent risk control method for insurance.

[0030] [First Embodiment] like Figure 1 As shown, this embodiment of the invention provides an intelligent insurance risk control method. Specifically, the intelligent insurance risk control method provided by this embodiment includes, for example, the following steps: S11. Obtain multi-dimensional risk information of the insured vehicle in real time through IoT monitoring equipment; wherein, the multi-dimensional risk data includes driver abnormal driving behavior information, vehicle surrounding environment status information, vehicle static attribute information and external correlation information; S12. Process the input multi-dimensional risk information using a target multimodal large model to obtain the target real-time risk value of the insured vehicle; S13. Obtain the corresponding risk level based on the real-time risk value of the target, and implement intervention measures according to the corresponding risk level; wherein, different risk levels correspond to different intervention measures; S14. Collect the real-time risk value of the insured vehicle within a preset time after the intervention measures are implemented, compare it with the target risk value to generate a comparison result, and adjust the insurance premium rate of the insured vehicle according to the comparison result.

[0031] This method applies to insured objects, mainly insured vehicles, such as commercial trucks, passenger vehicles, new energy trucks, private vehicles, etc., and is used for insurance risk control management of these vehicles.

[0032] In step S11, the IoT detection device can be, for example, at least one of an active safety intelligent video terminal, a Beidou positioning module, and a millimeter-wave radar module. The active safety intelligent video terminal can include, for example, a DSM (Driver Status Monitor) camera, an ADAS (Advanced Driver Assistance Systems) camera, a BSD (Blind Spot Detection) camera, etc. Of course, in other embodiments, sensors, vehicle-mounted terminals, etc., can also be used to achieve the function of collecting information. The multi-dimensional risk information can include, for example, behavioral data, environmental data, static attribute data, and external correlation data. Behavioral data can be, for example, information on abnormal driving behavior of the driver; environmental data can be, for example, information on the surrounding environment of the insured vehicle; and static attribute data can be, for example, information on the static attributes of the insured vehicle. Specifically, as... Figure 4As shown, abnormal driving behavior information may include, for example, driver fatigue, frequency of distracted driving, excessive driving time, cumulative driving time exceeding 8 hours in a single day, smoking, making or receiving phone calls, using a mobile phone, not wearing a seatbelt, taking both hands off the steering wheel, vehicle speed, or deviation from the vehicle's driving trajectory. The surrounding environmental status information of the insured vehicle may include, for example, the distance to obstacles around the vehicle (via ADAS cameras), blind spot monitoring (obtained via BSD cameras), millimeter-wave radar monitoring, and road slipperiness / iciness (obtained via ADAS visual monitoring). The static attribute information of the insured vehicle may include, for example, vehicle model / years of use / maintenance records, company industry type / safety management level, and the policyholder's historical accident records. External related data may include, for example, regional disaster warnings (such as rainstorms and typhoons), traffic congestion indices, and safety management regulations for the company's location.

[0033] In step S12, the target multimodal large model can be obtained by training the initial multimodal large model, such as... Figure 6 As shown, the multi-dimensional information obtained above is input into the initial multimodal large model. After multiple training and iterations, the initial multimodal large model is transformed into a target multimodal large model. Then, the multi-dimensional information is input into the target multimodal large model, causing it to output a real-time risk value of 0-100 points. The higher the score, the higher the probability of an accident. Of course, the risk value data can be adjusted according to the actual situation, for example, it can be adjusted to a real-time risk value of 0-50 points or 100-200 points, and can be adjusted in real time according to different requirements. Specific details are not limited here.

[0034] In step S13, different risk levels can be divided according to the real-time risk value. In this embodiment, the implementation risk value of 0-100 points is divided into three different levels: high, medium, and low. For example, a high risk level is 71-100 points, a medium risk level is 41-70 points, and a low risk level is 0-40 points. Of course, in other embodiments, different risk levels can be divided, such as levels one, two, three, and four. The higher the level, the higher the risk. The risk level division can be set according to the actual scenario requirements, and is not limited here. Then, intervention measures corresponding to the risk level are automatically matched and executed according to the divided risk level to achieve early warning, early intervention, and early risk reduction.

[0035] In step S14, the system continuously collects the vehicle's risk performance (such as a decrease in alarms, no accidents, and a reduction in risk value) after a preset time following intervention and inputs it into the target multimodal large model to obtain a real-time risk value. This real-time risk value is then compared with the target real-time risk value, and the premium is adjusted upwards or downwards based on the comparison results. For example, if the preset time is 3 months, and the real-time risk value remains lower than the target real-time risk value within 3 months, the premium can be lowered; if the real-time risk value remains higher than the target real-time risk value, the premium will be increased. Alternatively, the premium can be dynamically adjusted based on the actual risk performance of the insured or the insured vehicle, such as no accidents in the past 3 months and the extent of the risk value reduction after intervention. 1. Customers whose risk level continues to decrease will be given a premium discount (e.g., a maximum reduction of 30%). 2. For clients whose risk values ​​have not improved or have increased, the rates will be maintained or further increased; Then, the intervention effect data (such as a certain type of intervention reducing the accident rate by X%) is simultaneously fed back to the target multimodal large model so that the target multimodal large model can iteratively optimize the scoring rules. This application utilizes IoT devices to achieve real-time, comprehensive data collection of vehicles, drivers, road conditions, and external information, avoiding assessment biases caused by single data sources. Through machine learning algorithms, it fuses and calculates multi-source data to output quantifiable real-time risk values, enabling risk visualization. The system automatically matches intervention measures from mild to severe based on risk level, achieving precise intervention. After intervention, the system automatically collects data such as the degree of risk reduction, accident rate, and alarm frequency. This data is used to dynamically adjust premiums and incentivize safe driving. Furthermore, it feeds effective data back into the model to continuously improve risk identification accuracy, forming a closed-loop intelligent risk control system. Based on real-time data feedback, the risk control model is continuously iterated to adapt to new vehicle risks (such as new energy vehicles). Compliance and social responsibility: Helping clients improve their safety management capabilities and reduce the accident rate of insured vehicles.

[0036] Furthermore, the abnormal driving behavior information of the driver includes at least one of the following collected by the driver state monitoring equipment: physiological fatigue, distracted driving, excessive driving time, smoking, making or receiving phone calls, not wearing a seat belt, taking both hands off the steering wheel, and deviation from the driving trajectory; the vehicle's surrounding environment status information includes at least one of the following collected by the advanced driver assistance system: forward collision warning, lane departure warning, and blind spot monitoring information; the vehicle's static attribute information includes at least one of the following: the vehicle model, years of use, maintenance records, and historical accident records of the insured vehicle; the external related information includes at least one of the following: regional weather warnings, traffic index, and enterprise safety level.

[0037] Specifically, the driver abnormal driving behavior information package is collected by the DSM camera, steering wheel sensor, gyroscope, etc., and includes: Physiological fatigue: Characteristics of fatigue include closing eyes, yawning, looking down, and nodding. Distracted driving: prolonged visual deviation, using mobile phones, smoking, drinking water, etc.; Violations: Not wearing a seatbelt, taking both hands off the steering wheel, sudden acceleration, sudden deceleration; Driving abnormalities: trajectory deviation, route deviation, speeding, continuous driving time exceeding the limit.

[0038] The vehicle's surrounding environment information is collected by ADAS cameras, millimeter-wave radar, and blind spot monitoring systems, and includes: Forward collision warning, following too closely; Lane departure warning, driving over the lane lines; Pedestrians / vehicles entering blind spots; The road surface is slippery, icy, flooded, and the visibility is low in foggy weather.

[0039] The vehicle's static attribute information can be obtained, for example, from an insurance company's underwriting system: Vehicle model, tonnage / seat capacity, and usage type (commercial / non-commercial); Service life, cumulative mileage, maintenance and repair records; The number of accidents, the amount of claims, and the percentage of liability in the past 1-3 years.

[0040] The external information is obtained in real time through third-party interfaces (such as meteorological bureaus, traffic management departments, and related enterprises). Weather: Warnings for heavy rain, typhoon, dense fog, and snow / ice. Traffic: Road congestion index, accident-prone road sections; For enterprises: safety level, violation rate, and accident rate of transportation enterprises.

[0041] This embodiment achieves multi-faceted and refined capture of risk causes by integrating driver behavior and environmental state data, significantly improving the accuracy and comprehensiveness of risk identification and laying the foundation for precise modeling.

[0042] Furthermore, such as Figure 2 As shown, step S12 specifically includes: S121. Input the multi-dimensional risk information into the initial multimodal large model to obtain the initial prediction result group; S122. Classify and label the initial prediction result group and manually correct it to obtain the corrected risk prediction result group; S123. Based on the corrected risk prediction information and the corresponding multi-dimensional risk information, the initial multimodal large model is iteratively processed to obtain the iterative multimodal large model; S124. Evaluate the iterated multimodal large model to obtain the target multimodal large model; S125. Based on the target multimodal large model, the input multidimensional risk information is processed in real time to obtain the real-time risk value of the insured vehicle.

[0043] Specifically, such as Figure 7 As shown, in step S121, an initial multimodal large model can be constructed using existing technologies such as the Nvidla series (Mim inference) and / or Huawei Alles series (MindIE inference). This initial multimodal large model can be trained, for example, with historical accident samples as labels and real-time collected data as features. The real-time collected data can be obtained, for example, by filtering and cleaning from multi-dimensional risk information, to obtain the initial prediction result group of the initial multimodal large model before training.

[0044] In step S122, since the initial prediction result group consists of multiple data points, the data needs to be classified first. After classification, it is determined whether the data matches or does not match the actual result. If it matches, no correction is needed; if it does not match, correction is required. The correction step can be performed manually. For example, if a driver blinks multiple times while driving, and the IoT device collects data in real time, for instance, it might capture the driver closing their eyes. When this data is input into the initial multimodal model, it might be attributed to fatigue driving. However, this behavior does not actually constitute fatigue driving, leading to a misjudgment of the prediction result by the initial multimodal model. In this case, manual judgment of this part of the video can correct the prediction result of this behavior. The corrected result data is then input into the initial multimodal model, causing the initial multimodal model to map this behavior to another normal result, thereby achieving iteration of the initial multimodal model.

[0045] In step S123, the mapping relationship is changed after manual correction. The changed mapping relationship is then input into the initial multimodal large model, so that the mapping relationship of the initial multimodal large model is iterated. This process is equivalent to training the initial multimodal large model. If the final result is correct, no iteration is needed. If the prediction result does not match the fact, iterative correction of the mapping relationship is required.

[0046] In step S124, the iterated multimodal large model also needs to be evaluated. This can be done manually to assess the model's accuracy, or through a human-machine collaboration approach. For example, high-quality test data needs to be prepared first. Public benchmark datasets covering various fields (such as GLUE, HumanEval, etc.) can be used for general capability testing; alternatively, a custom dataset can be constructed, containing user questions (Prompt) and reference answers (Completion), to test the model's suitability for specific business scenarios. Then, appropriate evaluation methods and scoring criteria are selected based on business needs. For example, for scenarios with fixed answers, text similarity algorithms (such as ROUGE, exact matching) can be chosen; for open-ended questions or complex dialogues, a "referee model" (such as Qianwen-Max) can be configured, using a custom scorer Prompt to guide the referee model on how to score the model's answers (e.g., a 0-5 score system). Finally, an evaluation task is created on an evaluation platform (such as Alibaba Cloud Bailian, PAI, etc.), selecting the model being evaluated, the associated dataset, and the evaluation dimensions, and starting the automated evaluation. The system will automatically run inference and calculate scores for each metric. Then, manual annotation can be introduced (e.g., expert evaluation, crowdsourced scoring), or a mechanism of AI initial screening and manual review can be adopted. Furthermore, A / B testing and canary releases can be used to collect user feedback in a real online environment, forming a closed loop of evaluation-optimization-re-evaluation. Finally, after the evaluation is completed, the overall score and pass rate can be viewed through radar charts, data details, etc. As business needs and model iterations occur, the test set needs to be updated regularly and the weights of evaluation metrics dynamically adjusted to prevent data leakage and ensure continuous adaptation to the latest scenarios. The multimodal large model obtained after evaluation through the above methods is the required target multimodal large model.

[0047] In step S125, after obtaining the target multimodal large model, the previously acquired real-time driving behavior, environment, vehicle, and external information are input into the target multimodal large model. The target multimodal large model will then output a real-time risk value of 0 to 100, which is used to reflect the current accident probability or accident risk probability of the vehicle.

[0048] Furthermore, such as Figure 3 As shown, step S13 specifically includes: S131. When the real-time risk value is at a low risk level, execute the first type of intervention measure of pushing security alert information to the user terminal. S132. When the real-time risk value is at a medium risk level, execute the second type of intervention by sending a voice warning message to the vehicle-mounted equipment of the insured vehicle and simultaneously providing targeted safety training content to the user terminal. S133. When the real-time risk value is at a high-risk level, a third type of intervention is initiated, in which a human agent conducts real-time voice communication through the IoT monitoring device to provide escort and continuously and dynamically tracks the insured vehicle until the risk is eliminated.

[0049] Specifically, in this embodiment, steps S131-S133 are not sequential; that is, intervention measures are executed according to the risk level of the real-time risk value.

[0050] Specifically, this embodiment divides real-time risk values ​​into three risk levels for intervention, achieving precise, closed-loop, and efficient intervention, as illustrated in the following example: 1. Low risk (0-40 points): First-level intervention (mild suggestion): Triggering conditions: Good driving behavior, no high-frequency alarms, and normal road conditions; Intervention method: Push positive incentive prompts to the user terminal (or APP / mini-program), such as "Your recent driving behavior is good, keep it up"; Objective: To provide positive reinforcement and maintain good driving habits.

[0051] 2. Medium risk (41-70 points): Second type of intervention (early warning + training) Triggering conditions: Alarms such as fatigue, distraction, speeding, and following too closely are triggered. Intervention methods: The in-vehicle device provides a real-time voice warning: "You are driving while fatigued, please take a rest." Simultaneously push targeted safety training videos / courseware (driver power bank system); Objective: To correct the situation promptly and prevent the risk from escalating.

[0052] 3) High risk (71-100 points): Third type of intervention (human escort + strong control) Triggering conditions: continuous fatigue, multiple violations, high-risk sections of highways / mountainous areas, and combined risks; Intervention methods: The live operator immediately initiates in-vehicle voice intercom / telephone support; Continuously monitor the driver's status until the risk is eliminated; Requires parking and rest, and mandates on-the-ground rest control; Objective: To forcibly block high-risk activities and prevent accidents from occurring.

[0053] By employing a tiered strategy of alerts, warnings, and mandatory intervention, precise matching of risk response measures and optimal resource allocation are achieved. For low-risk clients, encouragement is the primary approach, while intervention intensity gradually increases for medium- and high-risk clients, ensuring that high-risk situations are addressed promptly and effectively, avoiding customer resentment or insufficient intervention caused by a "one-size-fits-all" approach.

[0054] Furthermore, the third type of intervention also includes: pushing the information of drivers marked as high-risk to their respective companies' safety management terminals and generating a list of offline visit tasks.

[0055] Furthermore, such as Figure 4 As shown, step S14 specifically includes: S141. Collect the real-time risk value of the insured vehicle at preset intervals; S142. Compare the real-time risk value with the target implementation risk value to obtain a comparison result; S143. If the real-time risk value is not higher than the target implementation risk value, reduce the insurance premium rate of the insured vehicle. S144. If the real-time risk value is higher than the target implementation risk value, increase the insurance premium rate for the insured vehicle. S145. Generate risk performance information based on the real-time risk value and feed it back to the target multimodal large model for optimization.

[0056] Specifically, in step S141, the preset time can be set according to the actual situation, such as after a one-month interval, that is, monitoring the insured vehicle every month, every quarter, or every six months, 1 month, or 3 months. For example, if monitoring is done every month, the real-time risk value of the insured vehicle is obtained at a one-month interval and compared with the target real-time risk value obtained when implementing intervention measures.

[0057] In step S142, the comparison results can be divided into two categories, for example: the real-time risk value is not higher than the target implementation risk value and the real-time risk value is higher than the target implementation risk value.

[0058] Steps S143 and S144 are executed simultaneously. Either step S143 or S144 is executed based on the comparison results. For example, after intervention measures, customers whose real-time risk values ​​continue to decrease are given a premium discount (e.g., a maximum reduction of 30%); after intervention measures, customers whose risk values ​​do not improve or increase are maintained or have their rates further increased.

[0059] In step S145, after the premium is adjusted, the real-time risk value of the vehicle to be reimbursed is generated simultaneously to produce risk performance information (such as a certain type of intervention measure that reduces the accident rate of this type of risk performance information by X%), which is then fed back to the target multimodal large model, thereby enabling iterative optimization of the scoring rules.

[0060] Specifically, for scenarios involving commercial vehicles and corporate fleets, high-risk interventions are enhanced with closed-loop supervision at the corporate end: the system automatically pushes the list of high-risk drivers and vehicles to the corporate safety management platform / APP, and then reminds the corporate safety administrator to conduct safety education, interviews, and suspension of work for rectification; it also automatically generates an offline visit task list, including: driver's name, license plate number, reason for risk; alarm type, time of occurrence, risk level; suggested rectification measures, and follow-up visit time limit; after the company completes the rectification, the results are uploaded, and the system archives them in a closed loop, forming a four-party risk control closed loop of platform—vehicle—driver—company.

[0061] This embodiment links individual driver risk with corporate safety management responsibility, forming a three-dimensional intervention network of online early warning and offline outreach. It breaks through the limitations of relying solely on in-vehicle equipment for intervention and significantly improves the success rate of correcting persistent high-risk behaviors.

[0062] like Figures 5-6 As shown, the insurance intelligent risk control method in this embodiment is as follows: 1. Preliminary preparation: Communicate with customers who need insurance to reach a consensus. Communicate with customers in advance before they purchase insurance to ensure their understanding and cooperation, so as to do a good job in the preliminary coordination and preparation work before risk control access.

[0063] 2. Establish a risk control group: Insurance companies, risk control providers, and enterprises jointly establish a risk control group; The insurance company monitors the early warning information and protection measures pushed in the group to keep track of the company's situation; Risk control providers send risk alerts and information on safeguard measures to the group chat. Relevant companies cooperated in the integration process by providing drivers' relevant information.

[0064] 3. Customer visits and data investigation: After the integration work is completed, for example, within two weeks, visit customers to understand the current management status and provide feedback on the data detection and investigation. Data detection and investigation is used to check whether there are any abnormalities in the equipment data. If there are abnormalities, the company needs to propose rectification to the equipment supplier to ensure that the equipment is free of abnormalities.

[0065] 4. Risk control and protection work: (1) Data analysis: The risk control provider provides risk information and driver data, and then the person in charge supervises the implementation of rectification. Specifically: a. If the device is obstructed, the risk control provider will issue a risk alert and push it to the group. The person in charge will then contact the driver by phone to make corrections. If the driver makes corrections, the camera will return to normal. If the driver does not make corrections, the alert will be pushed to the group and the team leader or person in charge will be notified to follow up and make corrections. b. In case of fatigued driving, the risk control provider will issue a risk alert and push it to the group. If it is a false alarm, the device will issue a voice reminder. If the driver is genuinely fatigued, the device will issue a voice reminder and simultaneously contact the driver by phone for assistance. After the driver answers the phone, the provider will continue to monitor the driver's status. If the driver is still fatigued, the provider will call the driver again to remind them to stop and rest. If the driver does not answer the phone, the alert will be pushed to the group and the team leader or person in charge will be notified for follow-up. c. Equipment online troubleshooting: If the equipment has not been online for 48 hours, it may be due to the vehicle not being dispatched or the equipment malfunctioning. If the equipment is malfunctioning, the equipment supplier shall carry out repairs. d. Other alarm handling: For driving such as making or receiving phone calls, using mobile phones, speeding, driving beyond permitted time, taking both hands off the steering wheel, not wearing a seat belt, etc., the risk control provider will issue a risk alarm and push it to the group. The alarm data will be pushed to the group daily / weekly / monthly in the form of a checklist or ledger for the team leader or person in charge to manage the drivers.

[0066] All of the above alarms will be responded to 24 hours a day.

[0067] (2) When an accident occurs, the person in charge shall cooperate in retrieving the video and completing the analysis within 48 hours after the accident. The person in charge shall be responsible for coordination and preparation for the subsequent cause of the accident and the determination of responsibility.

[0068] 5. Safety Training: Drivers will be trained in the first month after the integration is completed. Enterprises classified as high-risk are advised to receive monthly training, that is, once a month.

[0069] Specific Implementation Example 1: Active Risk Control for Commercial Trucks Data collection: Active safety terminal DSM camera (monitoring driver's facial expression recognition fatigue state, smoking, making or receiving phone calls, and other abnormal driving behaviors), ADAS camera (monitoring driving behavior habits), and Beidou positioning (recording driving trajectory and speed) are installed on insured commercial trucks; at the same time, the weather API is connected to obtain real-time road weather, etc. Risk modeling: The model detected that a truck driver was continuously triggering physiological fatigue alarms. Combined with the fact that the current road section is a curve with a high incidence of accidents, the real-time risk value rose to 85 points (high risk). Intervention measures: The system immediately issues a warning via in-vehicle voice prompt: "Fatigue driving detected. Please rest at the nearest service area." A human customer service representative immediately monitors the driver's condition in real time via intercom until the condition improves and follows up on the driver's rest time. At the same time, drivers identified as high-risk are targeted with safety training videos, and offline visits are made to the company to communicate and resolve the risks. Feedback on the results: Within two weeks after the intervention, the driver showed no signs of physiological fatigue while driving, and the system reassessed his risk score, which dropped to 30 points (low risk). No accidents occurred. [Second Embodiment] like Figure 8 As shown, the second embodiment of the present invention provides an insurance intelligent risk control system 200. As... Figure 6 As shown, the insurance intelligent risk control system 200 includes, for example, a data acquisition layer 201, a data processing layer 202, a risk control engine layer 203, and an interaction layer 204.

[0070] Specifically, the data acquisition layer 201 is used to acquire multi-dimensional risk information of the insured vehicle through IoT terminals, user terminals, and third-party interfaces; the data processing layer 202 is used to clean and filter false alarms of the multi-dimensional risk information, and includes a feature extraction module 2021; the risk control engine layer 203 includes a risk modeling module 2031, an intervention decision module 2032, and a premium calculation module 2033; the interaction layer 204 communicatively connects the data processing layer 202 and the risk control engine layer 203, and includes an insurance institution management backend, a user terminal, and an enterprise security management port; wherein, the feature extraction module 2021 is used to process the multi-dimensional risk information. The system generates feature vectors from multi-dimensional risk data; the risk modeling module 2031 processes the multi-dimensional risk information based on machine learning algorithms to obtain the real-time risk value of the insured vehicle; the intervention decision module 2032 determines the corresponding risk level based on the real-time risk value and executes intervention measures according to the corresponding risk level; different risk levels correspond to different intervention measures; the premium adjustment module 2034 collects the risk performance information of the insured vehicle after the intervention measures and adjusts the insurance premium rate of the insured vehicle; the interaction layer 204 executes the intervention measures and displays the risk information.

[0071] In this embodiment, the insurance intelligent risk control system adopts a four-layer architecture to automate the entire process of data access, processing, decision-making, and output: 1. Data Acquisition Layer IoT terminals: DSM / ADAS / BeiDou / Radar / Vehicle-mounted equipment; User terminals: Driver APP, vehicle owner mini-program; Third-party interfaces: meteorological, traffic, traffic police, and enterprise data.

[0072] 2. Data Processing Layer Data cleaning: Remove invalid alarms, duplicate data, and false alarms; Feature extraction: Converting raw data into feature vectors that the model can recognize.

[0073] 3. Risk control engine layer Risk modeling module: calculates risk values ​​in real time; Intervention decision-making module: Matches levels and outputs intervention strategies; Premium adjustment module: Dynamically calculates premium rates based on risk performance.

[0074] 4. Interaction Layer Insurance institution back-end system: View risk distribution, intervention effectiveness, and loss ratio; User terminal: Receive alerts, view ratings, and receive training; Enterprise safety management platform: View high-risk lists and supervise rectification.

[0075] Furthermore, the IoT terminal includes at least one of the following: an active safety intelligent video terminal, a driver status monitoring system camera, an advanced driver assistance system camera, a BeiDou positioning module, and a millimeter-wave radar module. Specifically: The system supports access to mainstream proactive security hardware: 1. Active safety intelligent video terminal: integrated DSM+ADAS all-in-one machine; 2. DSM camera: Monitors driver fatigue, distraction, smoking, and phone use; 3. ADAS camera: monitors for forward collisions, lane departures, and road conditions; 4. Beidou positioning module: real-time location, speed, trajectory, and mileage; 5. Millimeter-wave radar module: blind spot monitoring and near-field obstacle warning.

[0076] Furthermore, such as Figure 8 As shown, the data processing layer 202 further includes a data cleaning module 2022 for filtering the multi-dimensional risk information before the feature extraction module 2021 processes it, in order to remove invalid alarm data and false alarm data. Specifically, the data cleaning module 2022 performs the following cleaning rules: 1) Remove invalid alarms caused by obstruction, low light, or no target; 2) Filter false alarms that are single-instantaneous and have no continuous risk; 3) Merge duplicate alarms for the same event; 4) Retain valid, continuous, and high-risk alarm data for model calculation to improve the accuracy of risk control.

[0077] Furthermore, the aforementioned intelligent insurance risk control system is applicable to insurance risk control management for commercial freight vehicles, passenger vehicles, new energy freight vehicles, and private vehicles. Specifically... This system is highly versatile and can cover all categories of auto insurance products: 1. Commercial freight vehicles: long-haul logistics trucks, dump trucks, and hazardous materials trucks; 2. Passenger vehicles: buses, minibuses, ride-hailing vehicles, taxis; 3. New energy freight vehicles: electric logistics vehicles and refrigerated trucks; 4. Private vehicles: family sedans, SUVs, and MPVs. The system can configure differentiated risk models and intervention strategies based on different vehicle types.

[0078] like Figure 9 As shown, the insurance intelligent risk control system 200 provided in this embodiment can perform specific monitoring and risk control through the vehicle monitoring cloud platform system.

[0079] Specifically, a vehicle monitoring cloud platform system may include, for example, an application layer, a business layer, and a device layer.

[0080] The application layer includes the operator's app, the PC management platform, and the driver's app.

[0081] The business layer includes an AI video intelligent analysis engine, vehicle management, alarm management, and monitoring statistics. The AI ​​video intelligent analysis engine includes functions such as video content parsing, violation detection, driving behavior recognition, and intelligent false alarm filtering. It receives uploaded alarm videos, uses AI risk identification to filter false alarms, and finally issues warnings for information indicating high-risk behavior. Vehicle management involves static vehicle attribute information, such as vehicle files, vehicle binding, and vehicle location. Alarm management includes real-time alarm functions, AI intelligent filtering, alarm push notifications, alarm handling, and alarm statistics. Monitoring statistics include real-time monitoring, vehicle listing, online statistics, mileage statistics, report analysis, and data export.

[0082] The device layer includes, for example, positioning devices, sensors, and communication protocols.

[0083] Specifically, the positioning device may include, for example, a GPS locator, a Beidou locator, a 4G device, a WiFi positioning device, an LBS positioning device, and a 5G device.

[0084] The sensors may include, for example, an oil level sensor, a temperature sensor, a camera, an accelerometer, a barometric pressure sensor, etc.

[0085] The communication protocol may include, for example, the JT / T 808 protocol, the JT / T 809 protocol, the MQTT protocol, the TCP / IC protocol, the HTTP protocol, and the WebSocket protocol.

[0086] It should be noted that the insurance intelligent risk control method implemented by the insurance intelligent risk control system 200 provided in this embodiment is as described in the first embodiment above, and therefore will not be described in detail here. Optionally, the various modules and other operations or functions in the second embodiment are respectively for implementing the methods of the first embodiment of the present invention, and the beneficial effects are the same as those of the first embodiment. For the sake of brevity, they will not be described in detail here.

[0087] [Third Embodiment] like Figure 10 As shown, the third embodiment of the present invention provides an insurance intelligent risk control device 300. The insurance intelligent risk control device 300 includes, for example, a dimensional risk information acquisition module 301, a target real-time risk value acquisition module 302, an intervention measure implementation module 303, and a comparison result generation module 304.

[0088] Specifically, the multi-dimensional risk information acquisition module 301 is used to acquire multi-dimensional risk information of the insured vehicle in real time through IoT monitoring equipment; wherein, the multi-dimensional risk data includes driver abnormal driving behavior information, vehicle surrounding environment status information, vehicle static attribute information and external correlation information.

[0089] The target real-time risk value acquisition module 302 is used to process the input multi-dimensional risk information using a target multimodal large model to obtain the target real-time risk of the insured vehicle.

[0090] The intervention module 303 is used to obtain the corresponding risk level based on the real-time risk value of the target, and to execute intervention measures according to the corresponding risk level; wherein, different risk levels correspond to different intervention measures. The comparison result generation module 304 is used to collect the real-time risk value of the insured vehicle within a preset time after the intervention measures are implemented, compare it with the target risk value to generate a comparison result, and adjust the insurance premium rate of the insured vehicle according to the comparison result. For the specific working process and technical effects of the insurance intelligent risk control device 300 in this embodiment, please refer to the description of the first embodiment above.

[0091] [Fourth Embodiment] like Figure 11 As shown, a third embodiment of the present invention provides an insurance intelligent risk control system 400. The insurance intelligent risk control system 400 includes, for example, a memory 410 and a processor 420 connected to the memory 410. The memory 410 may be, for example, a non-volatile memory, on which a computer program 411 is stored. The processor 420 may be, for example, an embedded processor. When the processor 420 runs the computer program 411, it executes an insurance intelligent risk control method according to the first embodiment described above.

[0092] For the specific working process and technical effects of the insurance intelligent risk control system 400 in this embodiment, please refer to the description of the first embodiment above.

[0093] [Fifth Embodiment] like Figure 12 As shown, the fourth embodiment of the present invention provides a storage medium, such as a computer-readable storage medium 500. The computer-readable storage medium 500 is, for example, a non-volatile memory, such as a magnetic medium (e.g., hard disk, floppy disk, and magnetic tape), an optical medium (e.g., CD-ROM and DVD), a magneto-optical medium (e.g., optical disc), and a hardware device specifically configured for storing and executing computer-executable instructions (e.g., read-only memory (ROM), random access memory (RAM), flash memory, etc.). Computer-executable instructions 510 are stored on the computer-readable storage medium 500. The computer-readable storage medium 500 can be executed by one or more processors or processing devices to implement an insurance intelligent risk control method as described in the first embodiment above.

[0094] Furthermore, it is understood that the foregoing embodiments are merely illustrative examples of the present invention. Provided that the technical features do not conflict, the structure is not contradictory, and the purpose of the invention is not violated, the technical solutions of the various embodiments can be arbitrarily combined and used.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent risk control method for insurance, characterized in that, include: The system uses IoT monitoring devices to acquire multi-dimensional risk information about the insured vehicle in real time. The multi-dimensional risk data includes information on abnormal driving behavior of the driver, information on the status of the vehicle's surrounding environment, information on the vehicle's static attributes, and external related information. The multi-dimensional risk information input is processed using a target multimodal large model to obtain the target real-time risk value of the insured vehicle; Based on the real-time risk value of the target, the corresponding risk level is obtained, and intervention measures are implemented according to the corresponding risk level; wherein, different risk levels correspond to different intervention measures; Collect the real-time risk value of the insured vehicle after a preset time following the implementation of the intervention measures, compare it with the target risk value to generate a comparison result, and adjust the insurance premium rate of the insured vehicle based on the comparison result.

2. The method according to claim 1, characterized in that, The abnormal driving behavior information of the driver includes at least one of the following collected by the driver status monitoring device: physiological fatigue, distracted driving, driving beyond the permitted time, smoking, making or receiving phone calls, not wearing a seat belt, taking both hands off the steering wheel, and deviation from the driving trajectory; the vehicle surrounding environment status information includes at least one of the following collected by the advanced driver assistance system: forward collision warning, lane departure warning, and blind spot monitoring information. The vehicle static attribute information includes at least one of the following: vehicle model, service life, maintenance records, and historical accident records; the external related information includes at least one of the following: regional weather warnings, traffic index, and enterprise safety level.

3. The method according to claim 1, characterized in that, The process of using a target multimodal large model to process the input multidimensional risk information to obtain the target real-time risk value of the insured vehicle includes: The multi-dimensional risk information is input into the initial multimodal large model to obtain the initial prediction result set; The initial prediction result group is classified, labeled, and manually corrected to obtain the corrected risk prediction result group. The initial multimodal large model is iteratively processed based on the corrected risk prediction information and the corresponding multidimensional risk information to obtain the iterative multimodal large model. The iteratively evaluated multimodal large model yields the target multimodal large model. Based on the target multimodal large model, the multidimensional risk information input is processed in real time to obtain the real-time risk value of the insured vehicle.

4. The method according to claim 3, characterized in that, The model input information includes the proportion of drivers driving at night, the frequency of continuous driving time exceeding the limit, the frequency of fatigue alarms; the road section accident rate, the weather risk coefficient, the vehicle type risk coefficient; and the dynamic interaction risk characteristics between the driver's fatigue state and the slippery road surface.

5. The method according to claim 1, characterized in that, The process of determining the corresponding risk level based on the real-time risk value and implementing intervention measures according to the corresponding risk level specifically includes: When the real-time risk value is at a low risk level, the first type of intervention measure is implemented, which pushes a security alert message to the user terminal. When the real-time risk value is at the medium risk level, a second type of intervention is executed, which sends a voice warning message to the vehicle-mounted equipment of the insured vehicle and simultaneously provides targeted safety training content to the user terminal. When the real-time risk value is at a high-risk level, a third type of intervention is initiated, in which a human agent conducts real-time voice communication through the IoT monitoring device to provide support and continuously and dynamically tracks the insured vehicle until the risk is eliminated.

6. The method according to claim 5, characterized in that, The third type of intervention also includes: pushing the information of drivers marked as high-risk to their respective companies' safety management terminals and generating a list of offline visit tasks.

7. The method according to any one of claims 1-6, characterized in that, The process of collecting the real-time risk value of the insured vehicle at a preset time after the intervention measures are implemented, comparing it with the target implementation risk value to generate a comparison result, and adjusting the insurance premium rate of the insured vehicle based on the comparison result includes: Collect the real-time risk value of the insured vehicle at preset intervals; The real-time risk value is compared with the target implementation risk value to obtain the comparison result; If the real-time risk value is not higher than the target implementation risk value, the insurance premium rate for the insured vehicle shall be reduced. If the real-time risk value is higher than the target implementation risk value, the insurance premium rate for the insured vehicle shall be increased. Risk performance information is generated based on the real-time risk value and fed back into the target multimodal large model for optimization.

8. An insurance intelligent risk control device applied to the method of any one of claims 1-7, characterized in that, include: The multi-dimensional risk information acquisition module is used to acquire multi-dimensional risk information of the insured vehicle in real time through IoT monitoring devices; wherein, the multi-dimensional risk data includes abnormal driving behavior information of the driver, environmental status information of the vehicle's surroundings, static attribute information of the vehicle, and external related information. The target real-time risk value acquisition module is used to process the input multi-dimensional risk information using a target multimodal large model to obtain the target real-time risk value of the insured vehicle. The intervention module is used to obtain the corresponding risk level based on the real-time risk value of the target, and to execute intervention measures according to the corresponding risk level; wherein, different risk levels correspond to different intervention measures. The comparison result generation module is used to collect the real-time risk value of the insured vehicle within a preset time after the intervention measures are implemented, compare it with the target risk value to generate a comparison result, and adjust the insurance premium rate of the insured vehicle according to the comparison result.

9. An intelligent insurance risk control system, characterized in that, include: A memory and a processor connected to the memory, the memory storing a computer program, the processor executing the insurance intelligent risk control method as described in any one of claims 1-7 when running the computer program.

10. A computer-readable storage medium, characterized in that, include: The computer-readable storage medium stores a computer program, which, when executed by a processor, can implement the insurance intelligent risk control method as described in any one of claims 1-7.