Electric heavy truck intelligent insurance pricing method and system based on multi-dimensional risk assessment

Abstract

The application relates to the field of new energy automobile big data processing technology, and is an electric heavy truck intelligent insurance pricing method and system based on multidimensional risk assessment, which comprises a data preprocessing module that receives original data from a data acquisition unit, pre-processes the data and extracts features, acquires two vectors from a data storage unit and the data preprocessing module based on a double-branch feature fusion deep learning model, splices the two vectors, and outputs a vehicle comprehensive risk coefficient through a full connection layer and a Sigmoid activation function, calculates a final premium based on the vehicle comprehensive risk coefficient, generates an insurance policy, generates a risk early warning and a behavior suggestion, and pushes the risk early warning and the behavior suggestion to a user terminal. The application incorporates electrochemical microscopic indexes of a power battery into a risk assessment system, uses a double-branch feature fusion deep learning architecture, considers influences of long-term aging trends and short-term dangerous working conditions of the vehicle, effectively captures instantaneous high-risk behaviors such as low-temperature fast charging, and greatly improves generalization ability and accuracy of risk prediction.

Description

Technical Field

[0001] This invention relates to a smart insurance pricing method and system for electric heavy-duty trucks based on multi-dimensional risk assessment, belonging to the cross-technical fields of new energy vehicle big data processing, vehicle networking, and insurance financial technology. Background Technology

[0002] The penetration rate of electric heavy-duty trucks in ports, mines, and long-haul logistics is rapidly increasing. However, commercial insurance products for electric heavy-duty trucks still mainly use the pricing model of traditional fuel-powered commercial vehicles, or only adopt the basic UBI (Usage-Based Insurance) model.

[0003] For example, patent CN119762246A proposes a risk control analysis method and system based on vehicle network accident big data. It constructs a diverse training dataset containing target and disturbance labels, and generates high-quality sample data sequences by combining accident feature logic chains, thereby optimizing the parameters of the risk control knowledge prediction network. Patent CN121258536A provides an insurance fraud detection system and method based on vehicle data. This system encompasses a five-layer architecture: data acquisition, preprocessing, anomaly detection, risk assessment, and early warning decision-making. The method sequentially executes data acquisition and preprocessing, multi-dimensional feature comprehensive evaluation matrix generation, anomaly detection, and fraud risk assessment. The process includes steps such as risk scoring and early warning; it integrates multi-source data such as vehicle, driving, environment, and claims in real time and uses Kafka message queues to push early warnings; patent CN121636805A discloses a method, system, device and storage medium for recommending insurance products. This method collects multi-source raw data such as the operation, battery, historical business and real-time monitoring of the target vehicle, extracts and filters features to generate a core feature set for battery health prediction, and accurately predicts the battery health value based on a stacked regression model. Then, it combines target customer data and risk control data, dynamically configures recommendation logic through a rule engine and outputs personalized insurance product solutions. However, the aforementioned existing technologies still focus on driving behavior, accident characteristics, and insurance fraud, lacking in-depth perception and risk quantification of the micro-state of the power battery, failing to incorporate charging behavior into the risk assessment system, making it difficult to identify safety hazards at the electrochemical level, and failing to consider the "vehicle-battery separation" transfer scenario, thus failing to dynamically track the impact of the battery pack's historical health status on the real-time risk of the entire vehicle. Specifically, existing UBI models primarily rely on driving behavior data such as GPS trajectories, rapid acceleration, and rapid deceleration. In electric heavy-duty trucks, the power battery, as a core component, accounts for as much as 40%-50%, and its health directly affects the overall vehicle safety risk. Traditional models cannot assess the internal micro-state of the battery (such as lithium plating, internal resistance growth, and thermal runaway tendency), making it impossible for insurance companies to accurately identify the overall vehicle risk level. Furthermore, existing models do not incorporate electrochemical risks during the charging process into the overall vehicle risk assessment. Frequent high-temperature fast charging, overcharging, and low-temperature high-current charging significantly accelerate battery aging and increase the probability of vehicle fires. Current technology lacks correlation analysis between key indicators such as charging temperature rise rate and voltage consistency and overall vehicle risk; in addition, in heavy-duty truck battery swapping models, the battery pack is separated from the vehicle chassis. Traditional insurance is tied to the vehicle identification number (VIN), but cannot track the historical health of the battery used in the vehicle. When a vehicle uses a high-risk aging battery, insurance companies cannot identify the vehicle's current comprehensive risk, leading to inaccurate underwriting basis. Finally, existing pricing is mostly based on annual static pricing. It is unable to guide users to improve their driving or charging habits in real time through rate leverage, and lacks a closed-loop management system of risk warning, behavioral intervention, and rate adjustment.

[0004] Therefore, there is an urgent need for an intelligent insurance pricing system that can accurately quantify the comprehensive risks of electric heavy-duty trucks, adapt to battery swapping models, and have dynamic rate adjustment capabilities. This system can achieve precise matching of risk and premium by integrating physical models and data-driven models. Summary of the Invention

[0005] To address the problems in the existing technology, this invention provides a smart insurance pricing method and system for electric heavy-duty trucks based on multi-dimensional risk assessment.

[0006] The technical solution adopted by this invention to solve its technical problem is: a smart insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment, applied to a cloud computing unit, including: Step S1: The data preprocessing module receives raw data from the data acquisition unit and performs data cleaning, alignment, anomaly detection, and feature extraction. Step S2: The feature fusion risk assessment server obtains two vectors from the data storage unit and the data preprocessing module based on the dual-branch feature fusion deep learning model, concatenates the two vectors, and outputs the vehicle comprehensive risk coefficient through a fully connected layer and a Sigmoid activation function. Step S3: The dynamic pricing engine calculates the final premium based on the vehicle's comprehensive risk coefficient and generates the policy; Step S4: The user intervention module generates risk warnings and behavioral suggestions based on the vehicle risk coefficient and pushes them to the user terminal.

[0007] Further, step S1 includes: The data preprocessing module receives raw data from the data acquisition unit and performs millisecond-level time alignment between the vehicle data and the charging pile data based on the UTC timestamp. use The principle is to eliminate outliers caused by sensor drift, and to fill in missing data caused by short-term communication interruptions using linear interpolation. A four-dimensional feature system is extracted from the cleaned data, and the features are standardized and then input into a dual-branch feature fusion deep learning model. The four-dimensional feature system includes static attribute features, core component aging features, driving behavior features, and refueling safety features.

[0008] Furthermore, the two vectors obtained in step S2 include a global risk baseline vector obtained from the data storage unit and a real-time dynamic risk vector obtained from the data preprocessing module.

[0009] Furthermore, the dual-branch includes a first branch and a second branch. The first branch is an identity and history encoding branch. In the first branch, the dual-branch feature fusion deep learning model adopts a multilayer perceptron (MLP) structure to map static attributes and aging features into a global risk baseline vector. The second branch is the real-time operating condition extraction branch. In the second branch, the dual-branch feature fusion deep learning model uses a one-dimensional convolutional neural network (1D-CNN) to scan time series data to capture local electrochemical anomalies, and combines it with a gated recurrent unit (GRU) to simulate the thermal accumulation process and output a real-time dynamic risk vector. The vehicle comprehensive risk coefficient The value ranges from 0 to 1, representing the overall probability of an insured event occurring in the next insurance period.

[0010] Further, step S3 includes: Obtain the vehicle benchmark insurance premium from the actuarial database; The dynamic pricing engine follows the pricing formula. Perform calculations; Once the calculation is complete, the dynamic pricing engine generates an electronic policy and executes the premium deduction instruction; in, The base premium, This refers to the real-time comprehensive risk coefficient output end-to-end by the dual-branch feature fusion deep learning model. As the benchmark risk threshold, This is the premium adjustment sensitivity coefficient.

[0011] Furthermore, the aforementioned and All of these figures were derived by the insurance company through regression actuarial calculations based on historical claims data.

[0012] Furthermore, in step S4, when the risk coefficient continuously exceeds the set threshold, the user intervention module automatically generates risk warning information and targeted behavioral suggestions, and pushes the warning information to the user terminal. Set an observation period to track user behavior improvement. If the user improves their behavior during the observation period, the module will automatically trigger a rate reduction instruction; if the risk persists, a manual underwriting process will be triggered.

[0013] A smart insurance pricing system for electric heavy-duty trucks based on multi-dimensional risk assessment is provided to implement the aforementioned smart insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment. The system includes: The data acquisition unit is used to collect vehicle driving data, battery management system (BMS) data, and electrical parameters during the charging process. Data storage unit for storing vehicle, battery, and insurance premium related data; The cloud computing unit is used to output the vehicle's comprehensive risk coefficient and generate policy risk warnings and behavioral suggestions; The user terminal is used by vehicle owners or fleet managers to view risk reports and receive intervention suggestions.

[0014] Furthermore, the data acquisition unit includes an on-board terminal and a power replenishment terminal. The on-board terminal is installed in the vehicle's cab and connected to the OBD, and is responsible for collecting vehicle driving data and battery management system (BMS) data. The power replenishment terminal is deployed at charging or battery swapping sites and is responsible for collecting electrical parameters during the charging process. The data storage unit includes a vehicle risk profile database, a battery health database, and an insurance premium actuarial database. The vehicle risk profile database stores historical vehicle risk data using the VIN as the unique index. The battery health database is used to record the association between battery pack identification codes and health status; The premium actuarial database is used to store benchmark premiums, rate coefficients, and historical claims data.

[0015] Furthermore, the cloud computing unit includes a feature fusion risk assessment server, a dynamic pricing engine, a user intervention module, and a data preprocessing module; The feature fusion risk assessment server is equipped with a dual-branch feature fusion deep learning model and outputs a comprehensive vehicle risk coefficient. The dynamic pricing engine calculates premiums and generates policies based on risk coefficients. The user intervention module generates risk warnings and behavioral suggestions and pushes them to the user terminal. The data preprocessing module performs data cleaning, alignment, anomaly detection, and feature extraction in the cloud.

[0016] The beneficial effects of this invention are: Compared with the prior art, the present invention has the following significant advantages: 1. This invention incorporates electrochemical microscopic indicators of power batteries, such as lithium plating risk integral, internal resistance consistency, charging temperature rise rate, and thermal runaway tendency, into the risk assessment system. Utilizing a dual-branch feature fusion deep learning architecture, this invention can consider both the long-term aging trend of the vehicle and the impact of short-term hazardous operating conditions, effectively capturing instantaneous high-risk behaviors such as low-temperature fast charging. This significantly improves the generalization ability and accuracy of risk prediction, overcomes the shortcomings of traditional models in identifying electrochemical hazards, and achieves precise quantification of the overall vehicle risk.

[0017] 2. For heavy-duty truck battery swapping, this invention innovatively establishes a dynamic tracking mechanism with the Vehicle Identification Number (VIN) as the insured entity and the Battery Pack Identification Number (BID) as a dynamic risk factor. The system tracks and calculates the historical health and usage records of the swapped batteries in real time, dynamically linking them to the vehicle's risk profile. This allows insurance companies to accurately grasp the current comprehensive risk of the battery-swapping vehicle, ensuring the accuracy of underwriting and pricing bases and completely solving the problem of invalid underwriting basis in the "vehicle-battery separation" scenario.

[0018] 3. This invention abandons the traditional annual static pricing model and establishes a dynamic management closed loop of "risk warning - behavioral intervention - premium adjustment". By pushing targeted behavior improvement suggestions to users and coordinating with dynamic premium adjustment leverage, it proactively guides users to optimize their driving and charging habits, constructing a proactive risk control closed loop, significantly reducing the loss ratio and extending battery life.

[0019] 4. The dynamic pricing engine adjusts rates periodically based on a comprehensive risk coefficient, resulting in higher premiums for high-risk vehicles and lower premiums for low-risk vehicles, significantly improving premium fairness. Furthermore, fleet managers and drivers can monitor vehicle risk status and premium composition in real time through user terminals, eliminating the need for manual risk data collection. This not only improves billing transparency but also greatly enhances fleet risk management efficiency, achieving a precise match between risk and price. Attached Figure Description

[0020] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the overall system architecture of the intelligent insurance pricing method and system for electric heavy-duty trucks based on multi-dimensional risk assessment of the present invention. Figure 2 This is a diagram of a dual-branch risk assessment model in the intelligent insurance pricing method and system for electric heavy-duty trucks based on multi-dimensional risk assessment, as described in this invention. Detailed Implementation

[0021] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0022] Please see Figures 1-2 This embodiment provides a smart insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment. This method is applied to a cloud computing unit, which is the core processing unit of the system, responsible for data cleaning, model inference, premium calculation, and user intervention strategy generation. The method includes the following steps: Step S1: The data preprocessing module receives raw data from the data acquisition unit and performs data cleaning, alignment, anomaly detection, and feature extraction. Specifically, the data preprocessing module receives raw data from the data acquisition unit and performs millisecond-level time alignment between the vehicle data and the charging pile data based on the UTC timestamp; use The principle is to eliminate outliers caused by sensor drift, and to fill in missing data caused by short-term communication interruptions using linear interpolation. A four-dimensional feature system is extracted from the cleaned data, and the features are standardized and then input into a dual-branch feature fusion deep learning model. The four-dimensional feature system includes static attribute features, core component aging features, driving behavior features, and refueling safety features.

[0023] Step S2: The feature fusion risk assessment server obtains two vectors from the data storage unit and the data preprocessing module based on the dual-branch feature fusion deep learning model, concatenates the two vectors, and outputs the vehicle comprehensive risk coefficient through a fully connected layer and a Sigmoid activation function. The two vectors obtained in step S2 include the global risk baseline vector obtained from the data storage unit and the real-time dynamic risk vector obtained from the data preprocessing module.

[0024] The structure of the dual-branch risk assessment model is shown below. Figure 2 The dual-branch system includes a first branch and a second branch. The first branch is the identity and history encoding branch. In the first branch, the dual-branch feature fusion deep learning model adopts a multilayer perceptron (MLP) structure to map static attributes and aging features into a global risk baseline vector. The second branch is the real-time operating condition extraction branch. In the second branch, the dual-branch feature fusion deep learning model uses a one-dimensional convolutional neural network (1D-CNN) to scan time series data to capture local electrochemical anomalies, and combines it with a gated recurrent unit (GRU) to simulate the thermal accumulation process and output a real-time dynamic risk vector. Vehicle comprehensive risk coefficient The value ranges from 0 to 1, representing the overall probability of an insured event occurring in the next insurance period.

[0025] Step S3: The dynamic pricing engine calculates the final premium based on the vehicle's comprehensive risk coefficient and generates the policy; Specifically, the benchmark vehicle insurance premium is obtained from the actuarial database; The dynamic pricing engine follows the pricing formula. Perform calculations; After the calculation is completed, the dynamic pricing engine generates an electronic policy and executes the premium deduction instruction, ensuring that the output of the AI ​​model directly drives the dynamic adjustment of the rate and supports monthly or weekly pricing loops. in, The base premium, This refers to the real-time comprehensive risk coefficient output end-to-end by the dual-branch feature fusion deep learning model. The baseline risk threshold represents the standard risk level. This is the premium adjustment sensitivity coefficient.

[0026] and All of these figures were derived by the insurance company through regression actuarial calculations based on historical claims data.

[0027] Step S4: The user intervention module generates risk warnings and behavioral suggestions based on the vehicle risk coefficient and pushes them to the user terminal unit.

[0028] Specifically, in step S4, when the risk coefficient continuously exceeds the set threshold, the user intervention module automatically generates risk warning information and targeted behavioral suggestions, such as avoiding low-temperature fast charging, and pushes the warning information to the user terminal unit. The user intervention module pushes the warning information to the user terminal unit through the API interface. Set an observation period to track user behavior improvement. If the user improves their behavior during the observation period, the module will automatically trigger a rate reduction instruction; if the risk persists, a manual underwriting process will be triggered.

[0029] This embodiment provides a smart insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment. The method is applied to a data acquisition unit, which is responsible for acquiring and initially digitizing the raw physical signals. The data acquisition unit mainly includes two types of equipment: an on-board terminal and a power replenishment terminal. The vehicle terminal reads vehicle driving status data and battery management system (BMS) data in real time and uploads the collected data to the cloud computing unit in real time. Specifically, the vehicle-mounted terminal is installed in the vehicle's driver's cab and connects to the vehicle's OBD interface via a physical interface. This terminal has a built-in communication module capable of reading real-time vehicle driving status data and battery management system (BMS) data. The collected data includes vehicle speed, three-dimensional acceleration, latitude and longitude coordinates, motor speed, total battery pack voltage, individual cell voltage, battery temperature, SOC (State of Charge), and SOH (State of Health). The vehicle-mounted terminal uploads the collected data to the cloud in real-time via a 4G or 5G mobile communication network. The data sampling frequency is set to 0.1Hz to 1Hz to ensure the capture of instantaneous driving behaviors such as rapid acceleration and deceleration, as well as fluctuations in battery voltage.

[0030] The charging terminal communicates with the vehicle's BMS via a charging communication protocol (such as GB / T27930), collects electrical parameters during the charging process, and uploads the collected data to the cloud computing unit in real time. Specifically, the energy replenishment terminal is deployed at charging stations or battery swapping stations, and includes a charging pile controller or a battery swapping station management system. During vehicle charging or battery swapping operations, the energy replenishment terminal communicates with the vehicle's BMS via a charging communication protocol (such as GB / T27930) to collect electrical parameters during the charging process. The collected data includes charging current, charging voltage, charging power, battery temperature rise curve, individual cell voltage consistency data, and insulation resistance. The energy replenishment terminal uploads the data to the cloud via wired or wireless networks, supporting both batch uploads after charging is complete and real-time uploads during charging. The sampling frequency is set to 0.1Hz to 1Hz to quantify the safety risks of energy replenishment activities. This embodiment provides a smart insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment. This method is applied to a data storage unit, which is responsible for the persistent storage, index management, and secure backup of massive amounts of heterogeneous data. It mainly includes three core databases: The vehicle risk profile database uses the vehicle identification number (VIN) as the unique primary key index. The database stores basic static information of vehicles, historical risk coefficient records, insurance claim records, current transfer records, and driving behavior statistical characteristics. Specifically, the database adopts a distributed relational database architecture and supports high-concurrency read and write operations. It is used to establish a risk profile for the entire life cycle of a vehicle, ensuring that all risk assessments and pricing operations are associated with the vehicle as the main entity.

[0031] The battery health database uses the battery pack identification number (BID) as a unique primary key index, and also establishes a field associated with the VIN; Specifically, the database stores data such as the battery's chemical system type, rated capacity, manufacturing date, SOH decay trajectory, equivalent full cycle count, fast charging ratio, and extreme temperature exposure duration. The database also records the battery pack's circulation history, enabling the tracking of changes in its health status when batteries are swapped between different vehicles, and supporting dynamic risk tracking in battery swapping scenarios.

[0032] The premium actuarial database is stored using a composite index of vehicle model and VIN. The database stores benchmark premium data, rate coefficients, historical claims data, and statistical distributions of risk factors. Specifically, the rate coefficient includes the weights of health factors, charging factors, and driving factors. This database provides actuarial basis and historical reference data for the dynamic pricing engine, and supports insurance companies in adjusting weighting coefficients based on actual claims.

[0033] The data storage unit employs a hot and cold data separation mechanism. High-frequency access data from the past six months is stored in a high-speed cache database to ensure read speed, while historical data older than six months is archived in object storage to reduce costs. Data backup utilizes a multi-copy redundancy mechanism to ensure data security and immutability.

[0034] This embodiment provides a smart insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment, applied to a user terminal unit, including: User terminals communicate with cloud computing units via the Internet; Receive various data from cloud computing unit communications and display or pop up the received data; Provide a feedback portal for users to submit information.

[0035] Specifically: The user terminal unit is responsible for the interaction between the system and the user, and mainly includes the mobile app and the web management backend.

[0036] The user terminal communicates with the cloud computing unit via the internet. In the risk report viewing function, the terminal displays the current vehicle's risk coefficient, risk level, and risk composition analysis, helping users understand the basis for premium calculations. In the premium billing viewing function, the terminal displays premium details, rate adjustment explanations, and historical premium comparisons, improving billing transparency. In the intervention suggestion receiving function, the terminal displays risk warnings and improvement suggestions via pop-ups or push notifications, and provides a feedback entry point for users to confirm receipt or submit an appeal. The terminal interface allows fleet administrators to view the risk status of multiple vehicles in batches, facilitating fleet risk management for logistics companies.

[0037] This embodiment provides an intelligent insurance pricing system for electric heavy-duty trucks based on multi-dimensional risk assessment, used to implement the aforementioned intelligent insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment. The system is applied to a cloud computing unit and includes: The data acquisition unit is used to collect vehicle driving data, battery management system (BMS) data, and electrical parameters during the charging process. The data acquisition unit includes an on-board terminal and a charging terminal. The on-board terminal is installed in the vehicle's cab and connected to the OBD, and is responsible for collecting vehicle driving data and battery management system (BMS) data. The power replenishment terminal is deployed at charging or battery swapping sites and is responsible for collecting electrical parameters during the charging process; Data storage unit for storing vehicle, battery, and insurance premium related data; The data storage unit includes a vehicle risk profile database, a battery health database, and an insurance premium actuarial database; The vehicle risk profile database stores historical vehicle risk data using the VIN as the unique index. The battery health database is used to record the correlation between battery pack identification codes and health status; The premium actuarial database is used to store benchmark premiums, rate coefficients, and historical claims data.

[0038] The cloud computing unit is used to output the vehicle's comprehensive risk coefficient and generate policy risk warnings and behavioral suggestions; The cloud computing unit includes a feature fusion risk assessment server, a dynamic pricing engine, a user intervention module, and a data preprocessing module; The feature fusion risk assessment server deploys a dual-branch feature fusion deep learning model and outputs the comprehensive risk coefficient of the vehicle. A dynamic pricing engine calculates premiums and generates policies based on risk factors. The user intervention module generates risk warnings and behavioral suggestions and pushes them to the user terminal unit. The data preprocessing module performs data cleaning, alignment, anomaly detection, and feature extraction in the cloud.

[0039] This embodiment provides an intelligent insurance pricing system for electric heavy-duty trucks based on multi-dimensional risk assessment, applied to a data acquisition unit. The system includes: The on-board unit (OBD data acquisition) is installed in the vehicle's cab, connected to the OBD, and is responsible for collecting vehicle driving data and battery management system (BMS) data.

[0040] Energy replenishment terminals (charging piles / battery swapping stations) are deployed at charging / battery swapping sites and are responsible for collecting electrical parameters during the charging process.

[0041] This embodiment provides an intelligent insurance pricing system for electric heavy-duty trucks based on multi-dimensional risk assessment, applied to a data storage unit. The system includes: The vehicle risk profile database stores historical vehicle risk data using the VIN as the unique index.

[0042] The battery health database records the correlation between the battery pack identification number (BID) and the state of health (SOH). The premium actuarial database stores benchmark premiums, premium rates, and historical claims data.

[0043] This embodiment provides an intelligent insurance pricing system for electric heavy-duty trucks based on multi-dimensional risk assessment, applied to a user terminal. The system includes: User terminals (App / Web) allow vehicle owners or fleet managers to view risk reports and receive intervention recommendations.

[0044] This embodiment provides an application of an intelligent insurance pricing system based on a fleet of FAW Jiefang JH6 electric battery-swapping heavy trucks; This embodiment uses a fleet of 30 FAW Jiefang JH6 electric battery-swapping heavy trucks belonging to a logistics and transportation company as an example to describe in detail the specific implementation process of the technical solution of the present invention. This fleet mainly operates in short-haul transportation scenarios from mining areas to ports, uses battery swapping for energy replenishment, and is insured by a property insurance company.

[0045] Step S01, Application Scenarios and System Deployment: In terms of vehicle and equipment configuration, the fleet consists of 30 FAW Jiefang JH6 electric heavy-duty trucks, each equipped with an on-board terminal (OBD data acquisition), which is physically connected to the vehicle's OBD interface. The fleet is equipped with one battery swapping station and one charging station, and the battery swapping station management system has data upload capabilities. A cloud computing system is deployed, including a data preprocessing module, a feature fusion risk assessment server, a dynamic pricing engine, and a user intervention module, and a vehicle risk profile database, a battery health database, and an insurance premium actuarial database are established.

[0046] In terms of user terminal configuration, the fleet manager and 30 drivers have all registered and installed a WeChat mini program that includes charging and battery swapping functions and insurance services, in order to receive risk reports, premium bills and intervention suggestions.

[0047] In terms of basic information entry, during the system initialization phase, vehicle information for 30 vehicles was entered into the vehicle risk profile database, with the VIN serving as the unique index for the insured entity. The battery pack identification numbers (BIDs) of the 35 battery packs in the fleet were entered into the battery health database, and an initial VIN-BID association was established. The base premium was set at RMB 20,000 per vehicle per year based on the historical loss ratio of the JH6 model.

[0048] Step S02, Data Acquisition and Storage Implementation: Taking vehicle number 001 in the fleet (VIN code ending in 12345) as an example, the onboard terminal collects vehicle driving data at a frequency of 1Hz, including vehicle speed, acceleration, latitude and longitude, motor speed, etc., and simultaneously collects battery management system (BMS) data, including total battery pack voltage, individual cell voltage, battery temperature, SOC, and SOH. When vehicle 001 enters the battery swapping station for charging, the station terminal reads the current, voltage, temperature rise curve, and insulation resistance data during the charging process through the charging communication protocol, with a sampling frequency of 0.1Hz. All data is uploaded to the cloud computing unit in real time via the 4G network.

[0049] The cloud-based data storage unit receives uploaded data. The vehicle risk profile database, indexed by the last four digits of the VIN (Vehicle Identification Number) 12345, stores the vehicle's historical risk coefficients, accident records, and battery swapping records. The battery health database, indexed by the currently bound battery pack BID, stores the battery pack's chemical system (lithium iron phosphate), rated capacity (600kWh), SOH (State of Health) degradation trajectory, and cycle count. The premium actuarial database stores the base premium of 20,000 yuan for this vehicle model and the initial premium rate coefficient. Data is stored using a hot-cold separation system, with data from the past 12 months stored in a high-speed cache to ensure fast query response times.

[0050] Step S03, Risk Assessment and Calculation Implementation: The cloud-based data preprocessing module receives the raw data from vehicle 001. First, it aligns the vehicle data with the battery swapping station data at the millisecond level based on the UTC timestamp. Then, it uses... Abnormal temperature data caused by sensor drift (such as instantaneous temperature jumps exceeding 50°C) are excluded in principle. For missing data caused by communication interruptions, linear interpolation is used to fill in the gaps. Finally, a four-dimensional feature system is extracted, including static attributes (vehicle model, battery capacity), component aging (cumulative throughput, fast charging ratio), driving behavior (number of rapid accelerations, proportion of heavy-load hill climbs), and charging safety (charging temperature rise rate, number of low-temperature charging cycles).

[0051] The feature fusion risk assessment server loads a pre-trained dual-branch feature fusion deep learning model. The Identity and History Encoding branch (MLP) receives static attributes and component aging features, outputting a global risk baseline vector. The Real-Time Operating Condition Extraction branch (1D-CNN+GRU) receives nearly 7 days of driving and charging time-series data, capturing electrochemical anomalies and thermal accumulation processes, and outputting a real-time dynamic risk vector. The two vectors are concatenated and passed through a fully connected layer and a Sigmoid activation function to output the comprehensive risk coefficient for the next insurance period for vehicle 001. .

[0052] During winter operation, the model detected that vehicle 001 underwent high-current fast charging 10 times in the past 30 days when the ambient temperature was below -5°C, and the initial SOC of charging was generally above 80%. The model calculated that the rate of lithium plating damage was significantly increased, leading to a higher overall risk coefficient for the vehicle. The value rose from an initial 0.3 to 0.75, indicating a high-risk level.

[0053] Step S04, Dynamic Pricing and Intervention Implementation: The dynamic pricing engine obtains the comprehensive risk coefficient of vehicle 001. According to the pricing formula Calculate and set the final premium. The maximum and minimum value boundaries, to prevent a vehicle's Extremely low (close to 0) and At higher values, the result within the parentheses may become negative, leading to absurd results such as negative or zero premiums.

[0054] Among them, the benchmark premium The threshold is 20,000 yuan, assuming the actuarial benchmark risk threshold is set at 20,000 yuan. The adjustment coefficient is 0.3. The value is 0.178. This is because the low-temperature fast charging behavior was captured by the model, resulting in the vehicle's... The premium coefficient increased significantly from 0.3 to 0.75. Substituting this into the formula, the premium coefficient for this vehicle increased by approximately 8% the following month (i.e., 1 + 0.178 × (0.75 - 0.3) = 1.08), meaning the premium was adjusted to 21,600 yuan. The system automatically generated an electronic policy and a premium adjustment explanation.

[0055] The user intervention module detected a risk coefficient exceeding the threshold of 0.7 and automatically generated a risk warning. The warning was then pushed to the WeChat mini-program terminals of the vehicle's driver (001) and the fleet administrator via the internet. The warning specifically stated: "Your vehicle has been frequently fast-charging at high SOC in low-temperature environments, posing a high risk of battery lithium plating and thermal runaway. It is recommended to switch to a battery swapping station's heated slow-charging mode or preheating before charging." Simultaneously, the system sets a 30-day observation period, informing the user that if they improve their behavior, the following month's insurance premium may return to the base rate or even be reduced.

[0056] After receiving the warning, the fleet manager adjusted the scheduling of vehicle 001, stipulating that the battery heating function of the battery swapping station must be used for slow charging during low-temperature periods. The following month, the cloud data preprocessing module detected that the number of low-temperature fast charging cycles for vehicle 001 had dropped to 0, and the charging temperature rise rate had returned to normal. The feature fusion risk assessment server recalculated... The premium rate dropped to 0.35. Based on this, the dynamic pricing engine restored the premium coefficient to the benchmark for the following month and added a 5% safe driving bonus, resulting in an actual premium of 19,000 yuan.

[0057] Step S05, Implementation of Risk Tracking in Battery Swapping Scenarios: In the third month of operation, vehicle 001 replaced its original battery pack (BID_A) with a new battery pack (BID_B) at the battery swapping station. The battery swapping station reported the vehicle and battery replacement information to the cloud in real time.

[0058] Upon receiving the BID change signal, the cloud system immediately queries the historical records of BID_B in the battery health database. Assuming BID_B is a new battery in good health, with a SOH of 98% and no high-risk charging records, the system incorporates this battery's health characteristics into the vehicle's 001 dual-branch feature fusion deep learning model, directly impacting... If BID_B is a high-risk aging battery, the system will temporarily increase the current risk factor of vehicle 001 and notify the owner. In this embodiment, because the battery has been replaced with a healthy one, the overall risk factor of vehicle 001 is further stabilized, ensuring that the insurance coverage accurately reflects the current state of the vehicle.

[0059] Record keeping. The system writes this battery swap record into the vehicle risk profile database, linking it to the last four digits of the VIN code, 12345. Simultaneously, the usage record for BID_A is archived, while the usage record for BID_B begins to accumulate. Regardless of how the battery circulates, the insurance policy remains linked to the VIN code. Battery risk is dynamically incorporated into the overall vehicle risk assessment. Under the battery swapping model, the ownership and maintenance responsibility for the battery belong to the battery swapping station or battery bank. Any increase in insurance premiums due to battery risk is the responsibility of the battery bank, not the fleet.

[0060] Through the implementation of this embodiment, the fleet of 30 JH6 heavy trucks achieved the following results: After 6 months of operation, no battery thermal runaway incidents occurred in the fleet; the collision accident rate decreased by approximately 15% due to driving behavior intervention; the overall loss ratio is expected to decrease by 20%; by using rate leverage to guide drivers to optimize charging habits, the average battery capacity degradation rate of the fleet decreased by 10%, and the battery life is expected to be extended by 1 year; premiums for high-risk vehicles increased, while premiums for low-risk vehicles decreased, achieving a precise match between risk and price; and drivers' acceptance of premium adjustments significantly improved. The fleet administrator can monitor the risk status of all vehicles through the App, eliminating the need for manual statistics and improving management efficiency by more than 50%.

[0061] This embodiment verifies the feasibility and effectiveness of the technical solution of the present invention in the scenario of electric battery swapping heavy trucks, and realizes a closed loop of intelligent insurance pricing with the vehicle as the main body, integrating battery micro-risks, and dynamically adjusting premiums.

[0062] Although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A smart insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment, characterized by: Applications in cloud computing units include: S1. The data preprocessing module receives raw data from the data acquisition unit and performs data cleaning, alignment, anomaly detection, and feature extraction. S2. The feature fusion risk assessment server obtains two vectors from the data storage unit and the data preprocessing module based on the dual-branch feature fusion deep learning model. The two vectors are concatenated and the comprehensive vehicle risk coefficient is output through a fully connected layer and a Sigmoid activation function. S3, the dynamic pricing engine calculates the final premium based on the vehicle's comprehensive risk coefficient and generates an insurance policy; S4. The user intervention module generates risk warnings and behavioral suggestions based on the vehicle risk coefficient and pushes them to the user terminal.

2. The intelligent insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment according to claim 1, characterized in that: S1 includes: The data preprocessing module receives raw data from the data acquisition unit and performs millisecond-level time alignment between the vehicle data and the charging pile data based on the UTC timestamp. use The principle is to eliminate outliers caused by sensor drift, and to fill in missing data caused by short-term communication interruptions using linear interpolation. A four-dimensional feature system is extracted from the cleaned data, and the features are standardized and then input into a dual-branch feature fusion deep learning model. The four-dimensional feature system includes static attribute features, core component aging features, driving behavior features, and refueling safety features.

3. The intelligent insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment according to claim 1, characterized in that: The two vectors obtained in S2 include the global risk baseline vector obtained from the data storage unit and the real-time dynamic risk vector obtained from the data preprocessing module.

4. The intelligent insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment according to claim 1, characterized in that: The dual-branch system includes a first branch and a second branch. The first branch is the identity and history encoding branch. In the first branch, the dual-branch feature fusion deep learning model adopts a multilayer perceptron (MLP) structure to map static attributes and aging features into a global risk baseline vector. The second branch is the real-time operating condition extraction branch. In the second branch, the dual-branch feature fusion deep learning model uses a one-dimensional convolutional neural network (1D-CNN) to scan time series data to capture local electrochemical anomalies, and combines it with a gated recurrent unit (GRU) to simulate the thermal accumulation process and output a real-time dynamic risk vector. The vehicle comprehensive risk coefficient The value ranges from 0 to 1, representing the overall probability of an insured event occurring in the next insurance period.

5. The intelligent insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment according to claim 1, characterized in that: S3 includes: Obtain the vehicle benchmark insurance premium from the actuarial database; The dynamic pricing engine follows the pricing formula. Perform calculations; Once the calculation is complete, the dynamic pricing engine generates an electronic policy and executes the premium deduction instruction; in, The base premium, This refers to the real-time comprehensive risk coefficient output end-to-end by the dual-branch feature fusion deep learning model. As the benchmark risk threshold, This is the premium adjustment sensitivity coefficient.

6. The intelligent insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment according to claim 1, characterized in that: The and All of these figures were derived by the insurance company through regression actuarial calculations based on historical claims data.

7. The intelligent insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment according to claim 1, characterized in that: In S4, when the risk coefficient continuously exceeds the set threshold, the user intervention module automatically generates risk warning information and targeted behavior suggestions, and pushes the warning information to the user terminal. Set an observation period to track user behavior improvement. If the user improves their behavior during the observation period, the module will automatically trigger a rate reduction instruction; if the risk persists, a manual underwriting process will be triggered.

8. An intelligent insurance pricing system for electric heavy-duty trucks based on multi-dimensional risk assessment, characterized in that: The system is used to implement the intelligent insurance pricing method for electric heavy-duty trucks based on multi-dimensional risk assessment as described in any one of claims 1-7, the system comprising: The data acquisition unit is used to collect vehicle driving data, battery management system (BMS) data, and electrical parameters during the charging process. Data storage unit for storing vehicle, battery, and insurance premium related data; The cloud computing unit is used to output the vehicle's comprehensive risk coefficient and generate policy risk warnings and behavioral suggestions; The user terminal is used by vehicle owners or fleet managers to view risk reports and receive intervention suggestions.

9. The intelligent insurance pricing system for electric heavy-duty trucks based on multi-dimensional risk assessment according to claim 1, characterized in that: The data acquisition unit includes an on-board terminal and a power replenishment terminal. The on-board terminal is installed in the vehicle's cab and connected to the OBD, and is responsible for collecting vehicle driving data and battery management system (BMS) data. The power replenishment terminal is deployed at charging or battery swapping sites and is responsible for collecting electrical parameters during the charging process. The data storage unit includes a vehicle risk profile database, a battery health database, and an insurance premium actuarial database. The vehicle risk profile database stores historical vehicle risk data using the VIN as the unique index. The battery health database is used to record the association between battery pack identification codes and health status; The premium actuarial database is used to store benchmark premiums, rate coefficients, and historical claims data.

10. The intelligent insurance pricing system for electric heavy-duty trucks based on multi-dimensional risk assessment according to claim 1, characterized in that: The cloud computing unit includes a feature fusion risk assessment server, a dynamic pricing engine, a user intervention module, and a data preprocessing module; The feature fusion risk assessment server is equipped with a dual-branch feature fusion deep learning model and outputs a comprehensive vehicle risk coefficient. The dynamic pricing engine calculates premiums and generates policies based on risk coefficients. The user intervention module generates risk warnings and behavioral suggestions and pushes them to the user terminal. The data preprocessing module performs data cleaning, alignment, anomaly detection, and feature extraction in the cloud.

Images