New energy automobile insurance risk assessment method and system

Abstract

The invention relates to the technical field of data analysis, and particularly discloses a new energy vehicle insurance risk assessment method and system, and the system comprises a data collection module which is used for collecting battery state data and vehicle driving information, and carrying out the analysis to obtain driving behavior data; the feature modeling module is used for respectively constructing a battery risk index, a driving risk index and an environment risk index; the real-time calculation module is used for storing hierarchical scoring rules including a battery health score calculation rule, a driving behavior score dynamic offset calculation rule and an environmental condition risk correction table in a rule warehouse; the server is also used for regularly scanning the rule warehouse and automatically compiling and loading a new rule when detecting that the rule file is changed; and the risk assessment module is used for dynamically calculating a battery health score, a driving behavior score offset and an environment correction coefficient according to the hierarchical scoring rule in the rule warehouse, and finally obtaining a risk score.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method and system for risk assessment of new energy vehicle insurance. Background Technology

[0002] In the current auto insurance sector, traditional pricing models rely on static factors (such as vehicle model and region) and historical claims data, which are completely inadequate for adapting to the unique risk structure of new energy vehicles (NEVs) (such as battery degradation, charging safety hazards, motor and electronic control failures, and the profound impact of driving behavior on energy consumption / battery). While existing usage-based insurance (UBI) schemes incorporate dynamic data, they generally suffer from three fundamental flaws: First, there is a serious lack of data utilization: the system mainly relies on general OBD / GPS / mobile phone sensor data, while deeply ignoring NEV's core data sources, resulting in a one-sided risk profile. Second, the risk models are homogenized: most are built based on data from fuel vehicles, lacking specific quantitative models for NEV risk factors, resulting in low accuracy and poor differentiation in risk assessment. Third, dynamic pricing is ineffective: rate adjustments are mostly based on travel or fixed-period aggregated data, which cannot respond to instantaneous high-risk events to achieve immediate, event-driven premium recalculation, and the technical architecture is difficult to support the real-time processing and low-latency model inference of massive high-frequency NEV data.

[0003] Therefore, there is an urgent need for a risk assessment method and system for new energy vehicle insurance that can achieve dynamic risk assessment. Summary of the Invention

[0004] One of the objectives of this invention is to provide a new energy vehicle insurance risk assessment system that enables dynamic risk assessment.

[0005] To solve the above-mentioned technical problems, this application provides the following technical solution: The new energy vehicle insurance risk assessment system includes: The data acquisition module is used to collect battery status data of the vehicle's power battery; it is also used to collect vehicle driving information and analyze it to obtain driving behavior data. The feature modeling module is used to construct battery risk index, driving risk index and environmental risk index respectively based on battery status data, driving behavior data and environmental condition data; The real-time calculation module is used to store the hierarchical scoring rules, including battery health score calculation rules, driving behavior score dynamic offset calculation rules, and environmental condition risk correction tables, in the rule repository; it is also used to periodically scan the rule repository, automatically compile and load new rules when rule file changes are detected; it is also used to call the data acquisition module to obtain battery status data for each vehicle and vehicle driving information, and call the risk assessment module. The risk assessment module is used to dynamically calculate the battery health score, driving behavior score offset, and environmental correction coefficient based on the hierarchical scoring rules in the rule repository, and finally obtain the risk score.

[0006] Furthermore, the battery status data includes SOH, the percentage of fast charging cycles, and the frequency of deep discharge. Vehicle driving information includes torque, longitudinal acceleration, and GPS positioning information. The data acquisition module is also used to analyze driving behavior data based on vehicle driving information, including rapid acceleration / deceleration and hill overload. Rapid acceleration / deceleration recognition, the calculation formula is as follows: ; in, The change in velocity within the time window. The length of the time window; The formula for ramp overload detection is: ; in, For real-time torque, d represents the elevation difference, and d represents the horizontal driving distance.

[0007] Furthermore, the feature modeling module is used to calculate the driving risk index using a dynamic scoring model, with the following formula: ; in, , These represent the number of occurrences of the i-th type of rapid acceleration event and rapid deceleration event, respectively. , For the i-th type of rapid acceleration and deceleration events, the weight coefficients are: For ramp overload duration, This is the overload penalty coefficient for the ramp.

[0008] Furthermore, the risk assessment module calculates the battery health score, and the calculation rules are as follows:

[0009] The risk assessment module is also used to calculate an intermediate score based on the battery health score, according to the following rules: ; in, Dynamic offset is defined for driving behavior; The risk assessment module is also used to obtain the final risk score through environmental coefficient multiplication correction, according to the following rules: ; in, This is a comprehensive correction factor for environmental conditions.

[0010] Furthermore, it also includes a dynamic premium adjustment module, used to build a dynamic pricing model and calculate premiums according to the following rules: .

[0011] Furthermore, it also includes an early warning module, which is used to push early warning information based on the set early warning conditions; the early warning conditions include: battery health warning and charging behavior warning; It also includes a targeted service module, which is used to push differentiated value-added services based on the user's risk level. Differentiated value-added services for different risk levels include: providing battery swap coupons for high-risk users; and offering free roadside assistance services to low-risk users.

[0012] The second objective of this invention is to provide a method for assessing the insurance risks of new energy vehicles using the aforementioned system.

[0013] This solution incorporates core indicators such as State of Health (SOH), fast-charging frequency, and deep discharge frequency into battery state data, overcoming the limitations of traditional UBI solutions that cannot reflect the safety status of power batteries. In driving behavior analysis, it employs a rapid acceleration / deceleration recognition model and a slope overload detection formula, enabling the system to comprehensively capture instantaneous high-risk driving behaviors. In environmental risk construction, it introduces an environmental condition risk correction table, allowing the risk model to be dynamically calibrated based on extreme temperatures, severe weather, and high altitudes. Furthermore, this solution uses a real-time calculation module to uniformly store hierarchical scoring rules in rule files, automatically scanning, compiling, and loading them. This decouples scoring rules from business changes, allowing rule optimization without modifying the underlying code, thus ensuring the real-time nature and continuous maintainability of risk assessment. The risk assessment module dynamically calculates battery health scores, driving behavior offsets, and environmental correction coefficients based on the rule repository, forming a risk score that can be updated instantly with data changes. This achieves event-driven, continuously dynamic risk assessment capabilities that traditional insurance models cannot reach. Attached Figure Description

[0014] Figure 1 This is a logic block diagram of Implementation Example 3 of the New Energy Vehicle Insurance Risk Assessment System. Detailed Implementation

[0015] The following detailed description illustrates the specific implementation method: Example 1 The new energy vehicle insurance risk assessment system of this embodiment includes: a data acquisition module, a feature modeling module, a risk assessment module, and a real-time calculation module.

[0016] The data acquisition module is used to collect battery status data of the vehicle's power battery, including SOH (State of Health), fast charge cycle percentage, and deep discharge frequency. SOH is defined as the percentage of the current actual usable capacity of the battery to its rated capacity (SOH = CratedCcurrent × 100%). In this embodiment, the battery management system (BMS) calculates and stores this data in real time using the ampere-hour integration method combined with open-circuit voltage calibration.

[0017] FastChargeRatio is the ratio of the number of fast charging events to the total number of charging events within a statistical period (such as a calendar day or week). A fast charging event must meet two conditions: ① The charging power is greater than or equal to 80% of the battery's rated charging power (e.g., if the rated power is 100kW, the fast charging power must be ≥80kW); ② The initial State of Charge (SOC) is greater than 80% (to avoid misjudging necessary fast charging at low battery levels as high-risk behavior).

[0018] DepthOfDischarge (deep discharge frequency) is the number of discharge events where the State of Charge (SOC) is below 20% within a statistical period. The determination of a deep discharge event needs to be combined with the vehicle's operating status: when the vehicle is started (initial SOC value ≥ 20%), if the lowest SOC value is below 20% during driving and is not subsequently restored to above 20% through charging, it is counted as a deep discharge event.

[0019] The data acquisition module is also used to collect vehicle driving information, including torque, longitudinal acceleration, and GPS positioning information. Based on the vehicle driving information, driving behavior data is obtained, including rapid acceleration / deceleration and hill overload.

[0020] Rapid acceleration / deceleration recognition, the calculation formula is as follows: ; in, The change in velocity within the time window (unit: m / s). The time window length (in seconds).

[0021] The formula for ramp overload detection is: ; in, This represents the real-time torque of the motor (unit: N·m). d represents the elevation difference (in meters), and d represents the horizontal driving distance (in meters). The slope angle is given by the formula. This formula calculates the instantaneous slope overload risk value (in N·m·rad, which can be regarded as the torque-angle product) at a certain moment or for a certain road section.

[0022] This embodiment innovatively incorporates the load characteristics of the power system into the risk assessment system, breaking through the limitation of traditional UBI that relies solely on GPS speed.

[0023] The data acquisition module is also used to obtain a basic environmental risk index. In this embodiment, a baseline value is determined by statistically analyzing the average accident rate of similar vehicles under different environmental conditions in history. For example, the baseline risk value under normal temperature, sunny weather, and low altitude conditions is 1.0.

[0024] The feature modeling module is used to construct the Battery Risk Index (BRI), Driving Risk Index (DRI), and Environmental Risk Index (ERI) based on battery status data, driving behavior data, and environmental condition data, respectively. Specifically, As a core component of a vehicle's powertrain, the battery's health and usage patterns directly impact operational safety. The Battery Risk Index (BRI) quantifies the risks posed by the battery's inherent condition and usage behavior; the calculation formula is as follows: ; in, , , The weight coefficients for State of Health (SOH), Fast Charge Ratio, and Depth of Discharge frequency are obtained through training and optimization using historical claims data. The training method in this embodiment is as follows: using historical vehicle claims records (including labels such as repair costs and accident levels) as the dependent variable, and 1-SOH (reflecting the degree of battery capacity degradation), FastChargeRatio (fast charging frequency), and DepthOfDischarge (deep discharge frequency) within the corresponding time window as independent variables, a multiple linear regression model or a gradient boosting tree (GBDT) regression model is constructed. The weight coefficients are iteratively optimized by minimizing the mean squared error (MSE) objective function until the generalization error of the model on the validation set tends to stabilize.

[0025] Driving behavior is a direct human factor affecting vehicle safety. The Driving Risk Index (DRI) is used to quantify the risk of dangerous driving behaviors such as rapid acceleration, rapid deceleration, and overloading on inclines. This embodiment uses a dynamic scoring model for calculation, and the formula is as follows: ; in, , These represent the number of occurrences of the i-th type of rapid acceleration event and rapid deceleration event, respectively; the event classification is based on the absolute value of longitudinal acceleration: A rapid and intense acceleration is defined as: aacc ≥ 3 m / s²; The general rapid acceleration is: 2m / s2 ≤ aacc < 3m / s2; The severe rapid deceleration is: adec ≤ -3m / s2; The general rapid deceleration is: -3m / s2 < adec ≤ -2m / s2 (the negative sign indicates the deceleration direction).

[0026] , is the weight coefficient of the i-th type of rapid acceleration and rapid deceleration events, which is generated by learning the relevance of historical accident data through the XGBoost machine learning model. Specifically, it includes: Data collection: Extract historical driving behavior data (including the timestamps, acceleration values, and durations of rapid acceleration / deceleration events) and corresponding accident data (accident type, collision severity, repair cost); Feature engineering: Integrate driving behavior features (such as event frequency and intensity distribution) with vehicle state features (such as vehicle speed and load) to construct a multi-dimensional input feature set; Model training: Using the accident repair cost as the label, adopt the XGBoost regression model to learn the non-linear relationship between features and labels, and determine the contribution weights of each event to the accident loss through feature importance analysis; Hyperparameter optimization: Adjust parameters such as the learning rate (learning rate) and maximum depth (maxdepth) of the model through five-fold cross-validation, and finally output the verified weight coefficients.

[0027] is the ramp overload duration, which is defined as the cumulative time when the road slope θ ≥ 15° and the continuous duration t ≥ 30s during the vehicle driving process, and is used to determine which driving sections belong to the ramp overload working condition. The slope θ is calculated by combining the on-vehicle inertial measurement unit (IMU) with the GPS elevation data (θ = arctan(Δh / Δd), where Δh is the longitudinal elevation change and Δd is the longitudinal driving distance). When calculating the total ramp overload risk, only during the covered time period is accumulated, so determines the interval of and [[ID=3,1]] gives the risk intensity within this interval.

[0028] is the ramp overload penalty coefficient, which is determined by statistically calculating the ratio of the accident rate under different ramp overload durations to the accident rate under the reference duration (t < 30s). Through historical data fitting, when increases by 1 minute, the accident rate increases by about 80%, so the empirical value is taken as 1.8 (adjusting the comprehensive accident rate increment and weight sensitivity).

[0029] Environmental conditions (such as extreme temperatures, severe weather, and high altitudes) significantly affect vehicle performance and driving safety. The Environmental Risk Index (ERI) dynamically adjusts the baseline risk level using an environmental condition correction factor table. The calculation formula is as follows: ; in, To establish a baseline environmental risk index that does not consider environmental factors, a benchmark value is determined by statistically analyzing the average accident rate of the same type of vehicles under different environmental conditions in history (e.g., the benchmark risk value is 1.0 under normal temperature, sunny weather, and low altitude conditions).

[0030] The environmental condition comprehensive correction factor is obtained by selecting the corresponding correction factor according to Table 1 based on the current environmental conditions of the vehicle and multiplying them together.

[0031] Table 1 Environmental Condition Risk Correction Table

[0032] The risk assessment module sequentially calculates battery health score, dynamically overlays driving behavior score, and corrects for environmental factors to obtain the final risk score. Specifically, As a core component of a vehicle's powertrain, the battery's State of Health (SOH) is a fundamental input for risk assessment. The risk assessment module calculates the battery health score by quantifying the SOH index, using the following calculation rules:

[0033] The threshold settings in this embodiment are based on the statistical results of historical accident data: when SOH ≥ 90%, the battery capacity decay is slight, and the risks of short circuits and thermal runaway are extremely low, corresponding to the low-risk range (0-30 points); when 80% ≤ SOH < 90%, the battery capacity decay is significant, and the increased internal resistance leads to an increased risk of high temperature, corresponding to the medium-risk range (31-70 points); when SOH < 80%, the battery enters a rapid decay period, and the probability of thermal runaway is 3-5 times higher than that of normal conditions, corresponding to the high-risk range (71-100 points).

[0034] Driving behavior is a direct human factor affecting vehicle safety, and its dynamic changes need to be incorporated into risk assessment in real time. The risk assessment module also adjusts the Driving Behavior Index (DRI) dynamically based on the battery health score to achieve a real-time response to short-term dangerous driving behaviors, resulting in an intermediate score. The specific rules are as follows: ; in, The dynamic offset is calculated for driving behavior, and its value is positively correlated with the DRI score. Based on historical data, when DRI ≥ 70 (high-risk driving), ΔDRI is increased by 40 points; when DRI ≤ 30 < 70 (medium-risk driving), ΔDRI is increased by 20 points; and when DRI < 30 (low-risk driving), ΔDRI is zero. This dynamic overlay mechanism can sensitively capture instantaneous changes in driving risk. For example, if a driver suddenly accelerates sharply (high DRI) while the battery is in good health (low-risk score), the intermediate score will increase significantly, triggering a warning.

[0035] Environmental conditions (such as extreme temperatures, severe weather, and high altitudes) can nonlinearly amplify or suppress the basic risk level. The risk assessment module is also used to obtain the final risk score through environmental coefficient multiplication correction, thereby achieving the final quantification of risk. The specific rules are as follows: ; in, The environmental condition comprehensive correction factor is determined by the current environmental conditions, as shown in Table 1. The determination of environmental conditions requires cross-validation using multi-source data: the correction factor is set based on historical accident rate statistics: under the same battery and driving conditions, the accident rate increases by 50% when the temperature is <-10℃ compared to normal temperatures. =1.5), the accident rate doubles during heavy rain / snowy road conditions ( =2.0), the accident rate increases by 30% when the altitude is >3000m. =1.3). Through multiplication correction, the model can accurately reflect the "leverage effect" of environmental factors on risk—for example, a vehicle with a battery health score of 70 (high risk) and a driving score of +40 (high DRI) will experience a significant risk increase in heavy rain / snowy road conditions. =2.0), the final risk score will increase from 110 (70+40) to 220, significantly triggering a high-level warning.

[0036] The real-time calculation module stores the hierarchical scoring rules, including battery health score calculation rules, driving behavior score dynamic offset calculation rules, and environmental condition risk correction tables, in the form of Drools rule files (.drl) in a rule repository, such as GitLab or a database. In this embodiment, the rule files are described using a domain-specific language (DSL).

[0037] The real-time calculation module also uses Drools' KieScanner component to periodically scan the rule repository (e.g., every 5 minutes). When a rule file change is detected, it automatically compiles and loads the new rule, updating the rule without restarting the service. For example, when the operations backend adjusts the rainstorm road condition correction factor from 2.0 to 2.5, only the corresponding rule file needs to be modified and committed to the repository; the real-time calculation module will automatically apply the new rule in the next scanning cycle. This achieves logical decoupling between business rules and underlying code, meaning that when adjusting the SOH threshold or risk score, no modification to the compiled code is required; only updating the rule file is needed for the changes to take effect.

[0038] The real-time computing module also calls the data acquisition module to obtain battery status data and vehicle driving information for each vehicle. It then calls the risk assessment module to dynamically calculate battery health scores, driving behavior score offsets, and environmental correction coefficients based on the hierarchical scoring rules in the rule repository. This ensures that the risk assessment logic is consistent with the latest business rules, ultimately yielding a risk score. Vehicle risk data is characterized by high concurrency (10-20 sensor data points per second per vehicle) and strong timeliness (risk events must be assessed within 500ms). By constructing a streaming computing pipeline, end-to-end low-latency processing of "data access - hierarchical computing - result output" can be achieved.

[0039] Specifically, the system consumes real-time vehicle sensor data streams via Flink Kafka Consumer. The data format is Avro serialized VehicleEvent objects (containing fields such as timestamp, SOC, vehicle speed, acceleration, and ambient temperature). A three-tiered processing operator chain is built based on Flink's DataStream API to sequentially perform battery health score calculation, driving behavior score overlay, and environmental coefficient correction. Battery health score operator: Receives raw data stream, extracts SOH field, calculates battery health score according to Drools rules, and outputs BatteryScoreEvent object; Driving behavior score operator: The frequency of rapid acceleration and deceleration events is counted through a sliding window (window size of 1 minute, sliding step size of 30 seconds), the XGBoost model is called to calculate the real-time DRI value, the offset ΔDRI is generated by combining the Drools rule, and the DrivingScoreEvent object is output. Environmental correction operator: It integrates GPS altitude, camera road condition images, and meteorological API data, and obtains correction coefficients by matching environmental conditions through Drools rules. Finally, the total risk score is calculated, and a FinalRiskScoreEvent object is output.

[0040] Latency optimization: Optimize external data queries (such as weather API calls) through Flink's asynchronous I / O mechanism, converting synchronous blocking operations into asynchronous non-blocking operations; at the same time, enable the checkpoint mechanism (every 5 minutes) to ensure data consistency during fault recovery, ultimately achieving an end-to-end processing latency of ≤500ms (P99 metric).

[0041] This embodiment also provides a method for assessing the insurance risks of new energy vehicles, using the system described above.

[0042] This embodiment achieves comprehensive collection of battery SOH, fast charging events, high-intensity discharge events, torque, longitudinal acceleration, and GPS information, significantly enhancing the coverage of risk factors compared to existing solutions relying on OBD or single GPS data. It rapidly applies acceleration / deceleration calculation formulas and slope overload detection formulas to real-time map vehicle power load and driver control behavior, thereby improving the accuracy of driving behavior recognition. The feature modeling module performs hierarchical modeling of battery risk index, driving risk index, and environmental risk index, enabling various risk factors to be quantified through independent models, solving the problem of severe coupling and difficulty in interpretation of risks across different dimensions in traditional models. The risk assessment module also constructs a multi-level scoring system based on battery health score, combined with dynamic deviations in driving behavior and environmental corrections, enabling rapid response to instantaneous factors such as short-term high-risk driving and extreme environments, significantly enhancing the dynamic adaptability of risk scoring. Combined with real-time updates from a rule repository, the scoring system in this embodiment can implement rule changes within milliseconds, ensuring that risk assessment is always consistent with the latest business logic.

[0043] Example 2 The difference between this embodiment and Embodiment 1 is that this embodiment also includes a dynamic premium adjustment module, used to construct a dynamic pricing model and calculate the premium. The specific rules are as follows: ； The benchmark premium is determined in advance by the insurance company based on the vehicle's basic attributes (such as vehicle model, age, and purchase price), insurance terms (such as compulsory traffic accident liability insurance and commercial insurance type), and industry average level, serving as the base value for risk adjustment (e.g., the benchmark premium for a certain vehicle model is 5,000 yuan / year).

[0044] The risk score is the final output of the real-time calculation module (range 0-220 points), which comprehensively reflects the overall level of vehicle battery health (BRI), driving behavior (DRI), and environmental risk (ERI). The higher the risk score, the greater the premium increase (e.g., a score of 100 corresponds to a 100% increase in premium).

[0045] The regional coefficient is dynamically adjusted based on the environmental risk of the vehicle's usual location and is determined through correlation analysis of historical accident data and meteorological / geographical data. For example: Low-risk areas (no record of heavy rain / snow, altitude ≤2000m): coefficient 0.9; Medium-risk areas (occasional heavy rain / snow, altitude 2000-3000m): coefficient 1.1; High-risk areas (frequent heavy rain / snow, altitude > 3000m): coefficient 1.3.

[0046] The premium dynamic adjustment module is also used to adjust premiums according to preset adjustment modes. The preset adjustment modes include periodic adjustments and event-driven adjustments, taking into account both long-term risk trends and short-term abnormal risk responses.

[0047] The periodic adjustment recalculates premiums monthly, based on aggregating risk scores (such as monthly average BRI, DRI, and average environmental values ​​of the vehicle's usual location) using historical data from the past 30 days. For example, if vehicle A's average SOH in January is 85% (medium-risk BRI = 50 points), average DRI is 40 points (medium-risk driving score + 20 points), and the usual location is in a medium-risk area (coefficient 1.1), then the January premium = 5000 × (1 + 70 / 100) × 1.1 = 5000 × 1.7 × 1.1 = 9350 yuan; if in February the SOH drops to 75% (high-risk BRI = 80 points) and the DRI rises to 80 points (high-risk driving score + 40 points), then the February premium = 5000 × (1 + 120 / 100) × 1.1 = 5000 × 2.2 × 1.1 = 12100 yuan, significantly reflecting the impact of increased risk on premiums.

[0048] Event-driven adjustments trigger a real-time premium recalculation when a high-risk event occurs to the vehicle. Triggering conditions include: If the power of a single charge exceeds the threshold (e.g., fast charging power > 120kW, triggering an increase in the weight of FastChargeRatio in BRI); SOH weekly attenuation rate >2% (far exceeding the normal attenuation level of 0.5%-1%); Five consecutive sudden deceleration events (DRI sudden deceleration weighting increased). After an event is triggered, the risk score is immediately recalculated and the premium is updated. For example, if vehicle B's SOH weekly degradation rate reaches 3% due to battery thermal runaway, the premium will be increased by 20% immediately until the risk event is resolved (such as SOH recovery after battery repair).

[0049] It also includes an early warning module, used to set two levels of warning thresholds (yellow and red) based on pre-defined warning conditions, and push warning information via APP, SMS, or in-vehicle terminals to guide users to proactively reduce risks. Warning conditions include: Battery health alerts monitor the battery's State of Health (SOH) weekly degradation rate (calculated as: Weekly degradation rate = SOH last week / SOH this week - SOH last week × 100%). A yellow alert is triggered when the weekly degradation rate exceeds 1% (a notification is sent stating "Battery health is declining rapidly; it is recommended to have it inspected at an authorized service center soon"). If the weekly degradation rate exceeds 2%, it is upgraded to a red alert (an additional notification is sent stating "The battery is at risk of abnormal wear; it is recommended to contact customer service immediately to arrange a specialized inspection"). This alert is implemented by comparing the SOH data uploaded in real time by the BMS with a historical degradation trend model. The threshold is set based on historical accident data—vehicles with a weekly SOH degradation rate > 1% have a 4 times higher probability of thermal runaway in the next 3 months compared to normal vehicles.

[0050] The system provides charging behavior alerts by tracking the number of consecutive fast charges to 100% SOC. If this is triggered three times consecutively, charging habit optimization suggestions are pushed (e.g., "Frequent full charges accelerate battery degradation; it is recommended to charge to 80% SOC next time to extend battery life"). If triggered five times consecutively, it escalates to a high-risk alert (fast charging is restricted with a message stating, "To ensure battery safety, fast charging permission has been temporarily disabled and needs to be lifted at a service center"). This alert is implemented by statistically analyzing the SOC endpoint value and the fast charging marker (>80% SOC start) in the charging event log. The threshold is based on battery cycle life experiments—after five consecutive full charges, the battery capacity degradation rate is 30% higher than in normal charging mode.

[0051] It also includes a targeted service module, used to push differentiated value-added services based on the user's risk level (divided by total risk score: low risk <80 points, medium risk 80-150 points, high risk >150 points), thereby improving user stickiness and risk management effectiveness. Differentiated value-added services for different risk levels include: For high-risk users, battery swap coupons are offered (connected to the partner battery swap station system, users can redeem free battery swaps through the APP), guiding users to reduce fast charging and deep discharging behaviors (battery swapping can avoid the damage to the battery caused by fast charging). For example, if vehicle C has a total risk score of 200 points (high risk), the system will automatically issue 2 battery swap coupons (worth 300 yuan) and indicate on the APP homepage that "using battery swapping can reduce the risk of battery degradation".

[0052] For low-risk users, free roadside assistance is offered (in partnership with rescue organizations, covering basic services such as jump-starting, tire changing, and towing), enhancing users' perception of insurance services. For example, if vehicle D has a total risk score of 50 (low risk), the system will offer one free roadside assistance service each year during the user's birthday month, and will notify the user via SMS, "Thank you for maintaining good driving habits; one free roadside assistance service is offered as a gift."

[0053] This embodiment adds a dynamic premium adjustment module and an early warning module to the existing embodiment, enabling risk assessment results to be linked with insurance business processes and improving the automation capabilities of business operations. The dynamic pricing model combines real-time risk scores, regional risk coefficients, and benchmark premiums, and uses a linear premium adjustment formula to achieve monthly recalculation or event-driven instant recalculation. This overcomes the shortcomings of traditional annual pricing, which cannot reflect the dynamic risk characteristics of new energy vehicles, and allows premiums to accurately reflect risk changes such as battery degradation and deterioration in driving behavior.

[0054] Example 3 like Figure 1 As shown, the open sharing of vehicle operation data (such as battery SOC curves and driving behavior trajectories) is of great value for risk assessment and insurance pricing, but it also faces serious privacy risks—the raw data may contain sensitive information such as user driving habits and charging preferences, which, if maliciously used, could lead to user identity leakage, property loss, or privacy violations. Traditional data anonymization methods (such as simple anonymization) are easily cracked by correlation attacks and cannot meet strict privacy protection requirements.

[0055] To avoid directly uploading raw sensitive data to the cloud, this embodiment also includes an in-vehicle edge terminal (such as a T-BOX terminal, based on an ARM Cortex-A7 chip with a computing power of ≥4 TOPS) and a cloud server.

[0056] The vehicle-mounted edge device includes the data acquisition module and feature modeling module described in Embodiment 1, and also includes a desensitization module; The cloud server includes the risk assessment module and real-time computing module described in Embodiment 1, and also includes a data aggregation module; Data preprocessing and privacy protection are completed at the vehicle's edge, with only anonymized risk indices uploaded to the cloud, fundamentally blocking the source of personal information leakage. Specifically: The vehicle edge computing unit is used to acquire real-time vehicle operation data (such as SOC time series curves, charging power, and rapid acceleration events) from vehicle sensors. The desensitization module at the vehicle edge computing unit uses lightweight feature extraction algorithms to extract statistical features strongly correlated with risk assessment from the raw data, such as SOC mean, SOC variance, FastChargeRatio, and DeepDischarge frequency, while discarding sensitive information that can be associated with individuals, such as timestamps and geographical locations. For example, for the SOC curve (time series data SOC(t), t∈[t0,t1]), statistical feature extraction might include: ; By combining FastChargeRatio (number of fast charges / total number of charges) and DepthOfDischarge (number of times SOC < 20%), an intermediate value of the Battery Risk Index (BRI) is generated. Finally, only the aggregated risk values ​​such as BRI, DRI (driving risk index), and ERI (environmental risk index) are uploaded to the cloud server.

[0057] By extracting features from the vehicle's edge, the raw data (such as the timestamp distribution of the SOC curve and the specific time period of charging behavior) is thoroughly filtered out. The cloud server only receives risk indices without individual identifiers. Even if the data is illegally obtained, it is impossible to infer the driving habits or charging preferences of a specific user from a single risk index, thus achieving individual-level privacy protection.

[0058] To support the analysis needs of group data in insurance pricing, risk research, and other business operations, in this embodiment, the cloud server is also used to statistically aggregate the anonymized risk index through a data aggregation module, and to achieve ε-differential privacy by adding Laplace noise, ensuring that the statistical characteristics of the group data are not affected by the leakage of individual information. Specifically: The risk index is grouped and statistically analyzed by region (e.g., "City A"), vehicle model (e.g., "2023 Model 3"), and time period (e.g., "June 2024"), and a group average is calculated (e.g., "Average BRI of 2023 Model 3 in City A = 65"). The aggregation formula is: ； in, Let n be the risk index of the i-th user in the group, and n be the group size.

[0059] To prevent attackers from inferring individual data from group statistics (such as inferring a user's BRI value from the average BRI), the remote server also adds Laplace noise to the aggregated statistics to achieve ε-difference privacy. The probability density function of the Laplace distribution is: ; in, b is the position parameter (set to 0), and b is the scale parameter (b=ϵ1).

[0060] ε is the core parameter of differential privacy, controlling the trade-off between the strength of privacy protection and data availability. In this embodiment, ε is set to 0.1 (typical industry value range 0.1-1) based on business needs. A smaller ε results in stronger privacy protection but also a larger statistical error. Experiments show that when ε=0.1, the contribution of a single user to the group average is diluted by noise to less than 5% of the original value (standard deviation b=10). Attackers cannot infer individual information from statistical values, while the relative error of the group average is controlled within 8%, meeting the data accuracy requirements of businesses such as insurance pricing.

[0061] The distributed architecture of the vehicle edge and cloud server in this embodiment achieves privacy protection for raw vehicle operating data without sacrificing the accuracy of risk assessment. The vehicle edge uses lightweight algorithms such as SOC statistical feature extraction, fast charging frequency calculation, and deep discharge frequency extraction to convert raw time-series data into de-identified risk-related features and generate aggregated indicators such as BRI, DRI, and ERI locally, thus avoiding the uploading of raw data to the cloud and completely blocking the path of privacy leakage. The cloud server uses a data aggregation module to achieve group statistics across vehicles, regions, and vehicle models. Simultaneously, it combines a Laplace noise addition mechanism to achieve ε-differential privacy, ensuring that the aggregated statistical results cannot be used to infer the specific behavior of individual users, thereby meeting regulatory requirements and the needs of insurance companies for large-scale statistical analysis. Without increasing user risk, the availability and stability of statistical indicators are maintained, ensuring the effective operation of business such as insurance pricing and risk area classification.

[0062] The above are merely embodiments of the present invention. The invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are able to access all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A new energy vehicle insurance risk assessment system, characterized in that, include: The data acquisition module is used to collect battery status data of the vehicle's power battery; It is also used to collect vehicle driving information and analyze driving behavior data; The feature modeling module is used to construct battery risk index, driving risk index and environmental risk index respectively based on battery status data, driving behavior data and environmental condition data; The real-time calculation module is used to store the hierarchical scoring rules, including battery health score calculation rules, driving behavior score dynamic offset calculation rules, and environmental condition risk correction tables, in the rule repository; it is also used to periodically scan the rule repository, automatically compile and load new rules when rule file changes are detected; it is also used to call the data acquisition module to obtain battery status data for each vehicle and vehicle driving information, and call the risk assessment module. The risk assessment module is used to dynamically calculate the battery health score, driving behavior score offset, and environmental correction coefficient based on the hierarchical scoring rules in the rule repository, and finally obtain the risk score.

2. The new energy vehicle insurance risk assessment system according to claim 1, characterized in that: The battery status data includes SOH, percentage of fast charging cycles, and frequency of deep discharge. Vehicle driving information includes torque, longitudinal acceleration, and GPS positioning information. The data acquisition module is also used to analyze driving behavior data based on vehicle driving information, including rapid acceleration / deceleration and hill overload. Rapid acceleration / deceleration recognition, the calculation formula is as follows: ； in, The change in velocity within the time window. The length of the time window; The formula for ramp overload detection is: ； in, For real-time torque, d represents the elevation difference, and d represents the horizontal driving distance.

3. The new energy vehicle insurance risk assessment system according to claim 2, characterized in that: The feature modeling module is used to calculate the driving risk index using a dynamic scoring model, with the following formula: ； in, , These represent the number of occurrences of the i-th type of rapid acceleration event and rapid deceleration event, respectively. , For the i-th type of rapid acceleration and deceleration events, the weight coefficients are: For ramp overload duration, This is the overload penalty coefficient for the ramp.

4. The new energy vehicle insurance risk assessment system according to claim 3, characterized in that: The risk assessment module calculates the battery health score according to the following rules: ； The risk assessment module is also used to calculate an intermediate score based on the battery health score, according to the following rules: ； in, Dynamic offset is defined for driving behavior; The risk assessment module is also used to obtain the final risk score through environmental coefficient multiplication correction, according to the following rules: ； in, This is a comprehensive correction factor for environmental conditions.

5. The new energy vehicle insurance risk assessment system according to claim 4, characterized in that: It also includes a dynamic premium adjustment module, used to build a dynamic pricing model and calculate premiums according to the following rules: 。 6. The new energy vehicle insurance risk assessment system according to claim 5, characterized in that: It also includes an early warning module, which is used to push early warning information based on the set early warning conditions; Warning conditions include: battery health warning and charging behavior warning; It also includes a targeted service module, which is used to push differentiated value-added services based on the user's risk level. Differentiated value-added services for different risk levels include: providing battery swap coupons for high-risk users; and offering free roadside assistance services to low-risk users.

7. A risk assessment method for new energy vehicle insurance, characterized by: Use the system according to any one of claims 1-6.

Images