Management method and system of vehicle networking control unit

Abstract

The invention provides a management method and system of a vehicle networking control unit. The method comprises the steps of obtaining various operation data of a vehicle-mounted T-BOX; the types of the operation data comprise hardware state data, software operation data and vehicle operation state data; extracting a plurality of operation feature representations from the plurality of operation data; inputting the plurality of operation feature representations into a pre-trained fault prediction set model to obtain a fault prediction result of the vehicle-mounted T-BOX; wherein the fault prediction set model comprises a fault prediction overall model and at least one fault prediction sub-model for a specific fault type of the vehicle-mounted T-BOX; inputting the various operation data into a pre-constructed health state evaluation model, and determining a health state evaluation result of the vehicle-mounted T-BOX; and integrating the fault prediction result and the health state evaluation result of the vehicle-mounted T-BOX, and executing a fault processing action. In this way, the accuracy of vehicle-mounted T-BOX fault prediction, the comprehensiveness of health assessment and the initiative of fault processing can be improved, and the fault occurrence rate is effectively reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicles, in particular to a management method and system of a vehicle networking control unit. BACKGROUND

[0002] With the rapid development of vehicle intelligence and networking, the vehicle T-BOX (Telematics BOX, vehicle networking control unit) as the core device for realizing remote communication, data interaction and intelligent control of vehicles, its importance is increasingly prominent. However, in the complex vehicle operating environment, T-BOX faces many potential risks such as hardware aging, software failure, load increase, communication anomaly, etc. Once a fault occurs, it may cause the vehicle remote function to fail, data transmission to be interrupted, and seriously affect the normal operation of the vehicle and user experience.

[0003] The existing vehicle T-BOX fault monitoring technology mainly realizes simple fault detection by collecting basic operating parameters (such as voltage, temperature, fault code). For example, by presetting threshold values to judge whether a fault occurs (such as temperature exceeding the standard, communication interruption), and generating an alarm information after the fault occurs.

[0004] However, the existing fault monitoring is mostly based on simple judgment of a single fault code or a small number of operating parameters, which cannot comprehensively and deeply analyze the health status related to T-BOX, resulting in low accuracy and reliability of fault monitoring. In addition, the fault monitoring and processing of T-BOX can only be passively reported and repaired after the fault occurs, lacking effective preventive maintenance means, and it is difficult to discover potential faults in advance and take corresponding measures, increasing the risk and cost caused by vehicle faults. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a management method and system of a vehicle networking control unit, which can solve the problems of T-BOX fault early warning lag, one-sided health assessment, and low efficiency of passive maintenance after fault occurrence in the prior art, and achieve the purposes of improving the accuracy of vehicle T-BOX fault prediction, the comprehensiveness of health assessment, and the initiative of fault handling, and effectively reducing the fault occurrence rate.

[0006] The embodiment of the present application provides a management method of a vehicle networking control unit, which comprises: Obtaining a plurality of operating data of a vehicle T-BOX; the types of the operating data include hardware state data, software operating data and vehicle operating state data; Extracting a plurality of operating feature representations from the plurality of operating data; input the plurality of operation feature representations into a pre-trained fault prediction ensemble model to obtain a fault prediction result of the vehicle-mounted T-BOX; wherein the fault prediction ensemble model comprises a fault prediction overall model and at least one fault prediction sub-model for a specific fault type of the vehicle-mounted T-BOX; input the plurality of operation data into a pre-constructed health state evaluation model to determine a health state evaluation result of the vehicle-mounted T-BOX; integrate the fault prediction result and the health state evaluation result of the vehicle-mounted T-BOX to perform a fault handling action.

[0007] The embodiment of the application further provides a management system of a vehicle networking control unit, and the management system comprises: a data acquisition module configured to acquire a plurality of operation data of a vehicle-mounted T-BOX; the types of the operation data comprise hardware state data, software operation data and vehicle operation state data; a feature extraction module configured to extract a plurality of operation feature representations from the plurality of operation data; a fault prediction module configured to input the plurality of operation feature representations into a pre-trained fault prediction ensemble model to obtain a fault prediction result of the vehicle-mounted T-BOX; wherein the fault prediction ensemble model comprises a fault prediction overall model and at least one fault prediction sub-model for a specific fault type of the vehicle-mounted T-BOX; a health evaluation module configured to input the plurality of operation data into a pre-constructed health state evaluation model to determine a health state evaluation result of the vehicle-mounted T-BOX; a fault handling module configured to integrate the fault prediction result and the health state evaluation result of the vehicle-mounted T-BOX to perform a fault handling action.

[0008] The embodiment of the application further provides an electronic device, comprising a processor, a memory and a bus, the memory stores machine readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, and the machine readable instructions are executed by the processor to perform the steps of the management method of the vehicle networking control unit as described above.

[0009] The embodiment of the application further provides a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program is executed by the processor to perform the steps of the management method of the vehicle networking control unit as described above.

[0010] The vehicle networking control unit management method and system provided by the embodiment of the present application collect multi-dimensional rich operation data, realize accurate prediction of the fault of the vehicle-mounted T-BOX through the integrated architecture of the fault prediction set model, realize accurate feedback of the health state of the vehicle-mounted T-BOX through the health state evaluation model, and execute fault handling actions by comprehensively considering the fault prediction result and the health state evaluation result of the vehicle-mounted T-BOX, so as to actively prevent faults and effectively reduce the actual fault occurrence rate.

[0011] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are described in detail as follows. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0013] Figure 1 One of the flowcharts of the vehicle networking control unit management method provided by the embodiment of the present application is shown; Figure 2 The second flowchart of the vehicle networking control unit management method provided by the embodiment of the present application is shown; Figure 3 The structural schematic diagram of the vehicle networking control unit management system provided by the embodiment of the present application is shown; Figure 4 The structural schematic diagram of the electronic device provided by the embodiment of the present application is shown. DETAILED DESCRIPTION

[0014] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, every other embodiment obtained by those skilled in the art without creative labor belongs to the scope of protection of the present application.

[0015] Through research, it is found that the existing fault monitoring technology for vehicle T-BOX (vehicle networking control unit, also known as telematics control unit) mainly realizes simple fault detection through collection of basic operating parameters (such as voltage, temperature, fault code). For example, whether a fault occurs (such as temperature exceeding the standard, communication interruption) is judged through preset threshold, and alarm information is generated after the fault occurs.

[0016] However, the existing vehicle equipment fault monitoring method mainly has the following deficiencies: 1. Fault detection is mostly based on single fault code or limited operating parameters, which cannot comprehensively and deeply analyze the health status related to T-BOX, resulting in low accuracy and reliability of fault prediction.

[0017] 2. The fault monitoring and processing of T-BOX can only be passively reported and repaired after the fault occurs, lacking effective preventive maintenance means, and it is difficult to discover potential faults in advance and take corresponding measures, increasing the risk and cost caused by vehicle faults.

[0018] 3. The existing technology does not fully utilize advanced technologies such as big data and artificial intelligence in data processing and analysis, and does not sufficiently mine massive vehicle operation data, which cannot realize accurate prediction of T-BOX faults and real-time evaluation of health status; 4. The module chip and positioning chip of vehicle T-BOX are prone to performance degradation, high CPU load, high memory occupation, inaccurate positioning and other problems after long time power-on operation, and the existing technology lacks targeted preventive maintenance mechanism and cannot perform performance optimization without affecting user use.

[0019] 5. The storage module such as eMMC (embedded multimedia card) of T-BOX has the problem of life termination, and the existing technology lacks special monitoring and life prediction mechanism for the health status of eMMC and other storage.

[0020] Based on this, the embodiments of the present application provide a vehicle networking control unit management method to solve the problems of T-BOX fault early warning lag, one-sided health evaluation, and low efficiency of passive maintenance after fault occurrence in the prior art, so as to improve the accuracy of T-BOX fault prediction, the comprehensiveness of health evaluation, and the initiative of fault handling, and effectively reduce the fault occurrence rate.

[0021] Please refer to Figure 1 , Figure 1 The flowchart of one of the vehicle networking control unit management methods provided by the embodiments of the present application. The management method provided by the embodiments of the present application can be applied to a vehicle networking control unit management system, which can be realized based on software and / or hardware. As shown in Figure 1 , the management method comprises: S101, acquire multiple running data of the vehicle-mounted T-BOX.

[0022] In this step, the multiple running data of the vehicle-mounted T-BOX can be acquired by a data acquisition module in the management system. The data acquisition module is a basic data input unit of the management system, adopts a distributed architecture and a standardized interface, and acquires multiple source data of the vehicle-mounted T-BOX and the vehicle associated nodes. The types of the running data include: hardware state data, software running data, and vehicle running state data. The data acquisition module specifically includes: (1) hardware state data acquisition submodule Through the on-chip system (SoC) built-in ADC and the special interface controller, the core hardware parameters of the T-BOX are collected, such as: the temperature of the main chip and peripheral chip (precision ±0.5℃, sampling frequency 1Hz), the voltage of the power supply bus (error ≤0.2%) and the working current (0-5A, linearity ≤0.1%, sampling frequency 10Hz), the continuous power-on time of the chip (timing accuracy ±1s / day), the drift error of the positioning chip (open field plane accuracy ≤2m, sampling frequency 1Hz), the number of bad blocks of eMMC user area (24-hour period statistics), the execution frequency of eMMC CMD8 instruction (single power-on period record), the power cycle number (0-65535 count range), and the LVD low voltage alarm trigger frequency (1-hour period statistics).

[0023] (2) software running data acquisition submodule Through the kernel interface and the application layer hook function in the system, the software running state is collected, such as: process PID, CPU occupancy rate (100ms sampling period), memory usage rate and state identifier; system memory usage rate, process exclusive memory and swap exchange frequency (1Hz sampling); software main / secondary version number and revision number; time, project ID and result of active diagnosis request (ISO 14229 protocol format); receiving time, coding, state and result of remote control instruction (log delay ≤10ms). The data is acquired by the kernel driver and the user state process in cooperation, and the ring buffer cache is adopted. (3) communication data acquisition submodule Through the network controller module and the protocol stack hook, the communication performance parameters are collected, supporting 4G / 5G / C-V2X multi-standard communication scenarios, such as: cellular communication RSRP and SINR values (1Hz sampling), uplink and downlink data throughput (1s period statistics, accuracy ±1%), and continuous 100 data packet loss rate (10s sampling period). (4) vehicle running state data acquisition submodule The vehicle state parameters are obtained through the MCU vehicle bus interface, such as vehicle speed (CAN bus transmission, accuracy ±0.5km / h, 10Hz update), engine speed (0-8000rpm, CAN bus acquisition), instantaneous fuel consumption (volume method calculation, cumulative error≤3%), vehicle locking state (BCM CAN message, 0-unlocked / 1-locked), sleep wake-up signal (ACC gear level detection, 0V sleep / 12V wake-up), battery voltage (10Hz sampling, calculate voltage standard deviation) and ignition switch state (ECU CAN message, OFF / ON / START).

[0024] The data can be collected through the ISO 11898 standard CAN FD and vehicle Ethernet (SOME / IP protocol), and the data frame priority scheduling is supported. The communication interface design of data collection adopts a multi-level reliable communication architecture. The CAN bus interface: high-speed CAN (500kbps) and CAN FD (2Mbps) mixed networking, end-to-end delay≤10ms, supporting fault tolerance; the vehicle Ethernet interface adopts 1000BASE-T1 standard, transmission rate≥1000Mbps, with priority scheduling capability; the interface circuit has≥2500Vrms isolation and overvoltage and overcurrent protection function. LIN bus can also be used to replace the CAN connection of low-rate sensors; 5G C-V2X direct communication (PC5 interface) is used to supplement the Ethernet function; PCIe is used to replace the vehicle Ethernet in high-bandwidth scenarios. Thermocouple temperature sensor (±1℃ precision) is used to replace the on-chip digital sensor; the bad block state is indirectly obtained through the SMART attribute of eMMC.

[0025] S102、From the plurality of operating data, a plurality of operating feature representations are extracted.

[0026] In this step, the collected multi-source heterogeneous raw data is systematically processed, the data quality is improved through standardization operation, and high-quality input features are provided for fault prediction and health assessment through feature engineering. For example, a pipeline processing architecture can be used to sequentially perform data cleaning, denoising, normalization, feature extraction and feature selection operations.

[0027] In one possible implementation, step S102 can include: S1021, preprocessing the plurality of operating data. Wherein the preprocessing method includes data cleaning and normalization.

[0028] Data cleaning is to process the abnormal values, missing values and invalid data in the original data, which improves the data quality, and specifically includes: Outlier detection and correction: 3σ principle combined with box plot analysis or Isolation Forest algorithm can be used to identify outliers. For continuous parameters such as temperature and voltage in hardware state data, when the data deviates from the mean by 3 times the standard deviation or exceeds the interval [Q1-1.5IQR, Q3+1.5IQR], it is marked as abnormal and corrected by sliding average method of adjacent 5 valid data points; for discrete parameters such as eMMC bad block number, when the jump amplitude exceeds 200% of the historical mean, trigger data validity check, if invalid, use the median of the previous 3 valid values to fill in.

[0029] Missing value processing: For local data loss caused by communication interruption, when the missing duration is ≤5s and the number of consecutive missing data points is ≤10, linear interpolation or K-nearest neighbor interpolation is used to complete the missing data; when the missing duration is >5s, mark the data segment integrity identifier as "low" and keep the timestamp boundary information before and after the missing data.

[0030] Noise filtering: For high-frequency noise in vehicle operating state data, Butterworth low-pass filter (cutoff frequency 10Hz, order 4) or wavelet transform is used for smoothing processing; for positioning chip drift error data containing Gaussian noise, Kalman filter algorithm (process noise variance Q=0.01, measurement noise variance R=0.1) or particle filter is used for noise reduction, and the signal-to-noise ratio of the filtered data is improved by ≥15dB.

[0031] Further, in order to eliminate the dimension difference of different data types, it is also necessary to map various data to a unified numerical range. The normalized data in the embodiments of the present application all retain the bidirectional mapping relationship between the original physical value and the normalized value, supporting the reverse analysis of the subsequent results. The methods that can be used include: Linear normalization: For parameters with clear physical range such as hardware temperature (-40℃~85℃) and voltage (9V~16V), min-max normalization formula is used: x'=(x-x_min) / (x_max-x_min), which is mapped to the interval [0,1]; or maximum absolute value normalization method can also be used. Standardization: For parameters without fixed range such as signal strength RSRP (-140dBm~-44dBm) and memory occupancy, Z-score standardization is used: x'=(x-μ) / σ, so that the data conforms to the normal distribution with mean 0 and standard deviation 1. Discrete data encoding: For discrete label data such as vehicle locking state and ignition state, One-Hot Encoding is used to convert it into a binary vector, such as ignition state [OFF, ON, START] encoded as [1, 0, 0], [0, 1, 0], [0, 0, 1]. S1022、extracting a plurality of candidate vehicle operation feature representations from the pre-processed plurality of operation data.

[0032] In this step, key candidate vehicle operation feature representations can be extracted from the pre-processed plurality of operation data based on domain knowledge and data correlation analysis.

[0033] S1023、selecting the plurality of vehicle operation feature representations from the plurality of candidate vehicle operation feature representations according to the relevance of the candidate feature representations to fault prediction.

[0034] To reduce data dimension and improve model efficiency, a filtering and embedding combined method is used to screen the feature representations. In a specific implementation, the screening method in step S1023 can include at least one of the following: determining the variance of each candidate vehicle operation feature representation within a predetermined time period, and selecting the candidate vehicle operation feature representation with a variance greater than a preset variance threshold as the vehicle operation feature representation; for example, removing low-differentiation features with a variance <0.01, such as a fixed and unchanged software version number (without iteration).

[0035] determining the mutual information value between each candidate vehicle operation feature representation and the historical fault label of the vehicle-mounted T-BOX, and selecting the candidate vehicle operation feature representation with a mutual information value greater than a preset information threshold as the vehicle operation feature representation; for example, retaining strongly correlated features with a mutual information value >0.3. Based on the random forest model algorithm, each candidate vehicle operation feature representation is scored for feature importance, and candidate vehicle operation feature representations with a contribution lower than a preset threshold are removed step by step in a recursive manner to obtain vehicle operation feature representations. For example, the features are sorted by weight from high to low, and features with a contribution lower than 5% of the total weight are removed step by step. The screened feature representations are marked by a feature importance weight vector, supporting dynamic updating of feature priority. XGBoost and other algorithms can also be used for feature importance evaluation.

[0036] It should be noted that each of the above screening methods can be used alone or in combination. After pre-processing, the proportion of outliers is ≤0.5%, the data integrity is ≥99.8%, and the dimension of the screened feature set can be controlled within 50 dimensions, meeting the real-time and accuracy requirements of subsequent model training.

[0037] In one specific example, the vehicle operation feature representations extracted from the plurality of operation data can include: For the chip continuous running time length and the positioning accuracy correlation characteristics, the mean change rate of the positioning drift error is determined after the chip continuous power-on time length accumulates a predetermined time length, and the correlation between the chip continuous running time length and the positioning accuracy is quantified by Pearson correlation coefficient.

[0038] For example, the mean change rate of the positioning drift error is calculated when the chip continuous power-on time length accumulates 24 hours, the correlation between the two is quantified by Pearson correlation coefficient (r), and when r≥0.7, it is marked as a strong correlation feature, and the feature value is represented as a time sequence array of (continuous running time length, correlation coefficient value). Alternatively, the correlation can also be quantified by Spearman correlation.

[0039] For the eMMC instruction use intensity feature, based on the total number of CMD8 instruction executions within a predetermined time length and the vehicle power-on time length within the predetermined time length, the average use intensity is determined.

[0040] For example, the total number of CMD8 instruction executions per day is counted, the daily average execution frequency and the standard deviation are calculated, and the formula is: daily average intensity=number of executions on the day / power-on time length (h) on the day, which can reflect the access pressure of the storage controller. For the power cycle interval feature, according to the time stamp of each power cycle, the time length distribution of the interval between adjacent power-ons is calculated, and the statistical value of the time length distribution of the power-on interval is extracted.

[0041] For example, the time stamp of each power cycle is recorded, the time length distribution of the interval between adjacent power-ons is calculated, the mean, median and 90th percentile of the interval are extracted, and a power stability evaluation feature vector is formed. For the LVD trigger and eMMC write amount association feature, a time correlation window of LVD trigger time and eMMC write operation is established, and the ratio of the number of LVD triggers to the amount of eMMC write data in a unit time in the time correlation window is determined.

[0042] For example, the time correlation window is between ±5s, the ratio of the number of LVD triggers to the amount of eMMC write data (MB) in a unit time is counted, and the influence degree of power supply fluctuation on storage operation is quantified. For the user vehicle use time period distribution feature, based on the time stamp of the ignition state signal, the clustering algorithm is used to divide the vehicle use time period into multiple time periods; the power-on frequency, average running time length and communication data volume proportion of each time period are counted, and a user vehicle use feature matrix is formed.

[0043] For example, the use of K-means clustering (K=3) divides the use time into early peak, late peak and non-peak period, and the power-on frequency, average running time and communication data volume ratio of each period are counted to form a user behavior feature matrix. The feature can be used for subsequent optimization of the self-learning time window of the non-inductive power-on control unit to avoid interfering with the user's normal use of the vehicle.

[0044] S103, inputting the plurality of operation feature representations into a pre-trained fault prediction set model to obtain a fault prediction result of the vehicle-mounted T-BOX.

[0045] S104, inputting the plurality of operation data into a pre-constructed health state evaluation model to determine a health state evaluation result of the vehicle-mounted T-BOX.

[0046] For S103 and S104, the fault prediction set model and the health state evaluation model can be a multi-dimensional analysis model constructed based on machine learning and deep learning algorithms, which realizes the forward-looking prediction of the fault risk and the dynamic evaluation of the health state of the vehicle-mounted T-BOX. Based on the fault prediction set model and the health state evaluation model, the embodiment of the present application realizes a "prediction-evaluation" dual-engine architecture, which outputs the quantitative fault probability, type judgment and health state score through the cooperative working mode of historical data training and real-time data reasoning.

[0047] For S103, the fault prediction set model adopts an ensemble learning framework, which integrates the advantages of deep neural networks and support vector machine (SVM) algorithms, and constructs a special prediction model for typical fault types of the T-BOX, specifically including a fault prediction overall model and at least one fault prediction sub-model for a specific fault type of the vehicle-mounted T-BOX. For example, communication module attenuation, positioning chip drift, eMMC life attenuation.

[0048] In one possible implementation, the fault prediction sub-model includes at least one of the following: a communication module chip performance attenuation prediction sub-model, a positioning chip drift fault prediction sub-model, and an eMMC life attenuation prediction sub-model. Correspondingly, step S103 can include: For the fault prediction overall model, the plurality of operation feature representations are input into the fault prediction overall model, the fault prediction overall model uses an improved long short-term memory network as a core feature extractor to perform time sequence feature coding on the plurality of operation feature representations, combines an attention mechanism to strengthen the key time step feature weight, and uses a hybrid density network structure in the output layer to obtain the prediction result of the fault probability distribution and the fault type, and triggers a fault warning signal according to a threshold.

[0049] The fault prediction model in this embodiment uses an improved Long Short-Term Memory (LSTM) network as the core feature extractor. The input layer receives a preprocessed 50-dimensional feature vector and encodes temporal features through three LSTM units (with 128, 64, and 32 hidden nodes, respectively), combining an attention mechanism to strengthen the feature weights of key time steps. The output layer adopts a hybrid density network (MDN) structure, mapping to the fault probability distribution and type prediction results through three fully connected layers. During training, the Adam optimizer and a hybrid loss function are used for model training. The loss function is a weighted sum of cross-entropy loss and mean squared error (weight ratio 7:3). For the communication module chip performance degradation prediction sub-model, the performance-related features of the communication module chip are input into the communication module chip performance degradation prediction sub-model. The performance degradation curve is constructed through the SVM regressor. Based on the performance degradation curve, the probability that the communication performance will be lower than the threshold within a predetermined period of time is predicted, and a fault warning signal is triggered according to the threshold.

[0050] This application embodiment addresses faults such as signal strength attenuation and data transmission rate reduction. The input features focus on the signal strength RSRP time sequence, packet loss rate change rate, and chip temperature trend in the communication data. A performance degradation curve is constructed using an SVM regressor (kernel function RBF, γ=0.1, penalty coefficient C=10) to predict the probability that the communication performance will be lower than the threshold (RSRP≤-110dBm for 10 minutes) within the next 30 days, with a prediction error ≤5%. For the positioning chip drift fault prediction sub-model, the performance-related features of the positioning chip are input into the positioning chip drift fault prediction sub-model. A bidirectional long short-term memory network is used to determine the correlation between the positioning drift trend and the past and future. A sliding time window is set to extract the statistical value of the drift error, and a fault warning signal is triggered according to the threshold.

[0051] Based on the timing data of positioning drift error and the continuous operation time characteristics of the chip, this application uses a bidirectional LSTM (Bi-LSTM) network to capture the correlation between drift trends. A sliding time window (window size 24 hours) is set to extract the mean, variance and maximum deviation features of the drift error. A fault warning is triggered by threshold logic (drift error > 2m and lasts for 5 minutes), and the prediction lead time is ≥ 24 hours.

[0052] For the eMMC lifetime degradation prediction sub-model, the eMMC performance-related features are input into the eMMC lifetime degradation prediction sub-model. The Transformer time-series prediction model and the self-attention mechanism are used to capture the relationship between the growth of bad blocks in eMMC and the intensity features of eMMC instruction usage, determine the probability of storage failure risk, and trigger a fault warning signal based on the threshold.

[0053] The eMMC lifetime degradation prediction sub-model, as the core sub-model set up for eMMC in this application embodiment, focuses on inputting the number of bad blocks in eMMC, the average daily execution intensity of CMD8 instructions, and the correlation characteristics between LVD triggering and write volume to construct a Transformer-based time-series prediction model. It captures the long-term dependency between bad block growth and instruction usage intensity through a self-attention mechanism, and uses the Cox proportional hazards model in Survival Analysis to calculate the probability of storage failure risk. An alert is triggered when the predicted bad block rate is ≥5% within 3 months or the probability of a single write failure is >1%. The model's prediction accuracy is ≥92%.

[0054] It should be noted that the overall fault prediction model and each fault prediction sub-model can be trained based on historical operating data and fault information. It also supports an incremental learning mechanism based on new fault samples, and ensures that the prediction performance is continuously optimized as the amount of data increases by dynamically fine-tuning the model parameters.

[0055] In one possible implementation, the health status assessment model in step S104 is based on a multi-indicator weighted fusion algorithm to calculate the T-BOX health status score in real time and classify it into levels. Specifically, step S104 may include: S1041. Input the various operating data into the health status assessment model to determine at least one of the following sub-item scores: chip performance health sub-item score, eMMC storage health sub-item score, communication link health sub-item score, and power system health sub-item score.

[0056] For example, the chip performance health sub-score is calculated using a fuzzy comprehensive evaluation method based on indicators such as chip temperature stability (30%), voltage fluctuation amplitude (25%), correlation between continuous running time and positioning accuracy (25%), and memory utilization trend (20%). When the chip temperature is ≥75℃ or the memory utilization is ≥90% for 5 consecutive minutes, this score is reduced by 30%. For the eMMC storage health sub-score, core evaluation indicators include bad block rate (40%), average daily CMD8 instruction execution intensity (25%), number of write failures (20%), and LVD trigger associated risk (15%). 8 points are deducted for every 1% increase in bad block rate, and points are deducted proportionally when the average daily instruction intensity exceeds the threshold (100 times / h), with a maximum score of 100 points. For the communication link health sub-score, the score is calculated based on indicators such as communication stability and communication latency; for the power system health sub-score, the score is calculated based on indicators such as power consumption rate and charging time.

[0057] S1042. Based on the scores of each sub-item and their corresponding weights, the overall health status score of the vehicle-mounted T-BOX is obtained.

[0058] In this step, the health status assessment model constructs a health scoring system: the total health score is out of 100 points, and the weight of each sub-item is determined using the Analytic Hierarchy Process (AHP) or the entropy weight method. The chip performance health sub-item accounts for 35%, the eMMC storage health sub-item accounts for 30%, the communication link health sub-item accounts for 20%, and the power system health sub-item accounts for 15%. Each sub-item score is mapped to the [0,100] interval through a membership function. The total score is calculated as: Total Score = Σ(Sub-item Score × Weight).

[0059] S105. Based on the fault prediction results of the vehicle-mounted T-BOX and the health status assessment results, execute the fault handling action.

[0060] In a first possible implementation, step S105 may include: When the overall health status score belongs to the first overall interval, all sub-item scores belong to the first sub-interval, and there is no fault warning signal, the health status level is determined to be normal, and routine maintenance suggestions are generated.

[0061] When the overall health status score belongs to the second overall interval, all sub-item scores belong to the second sub-interval, and there is at least one fault warning signal, the health status level is determined to be a warning level, potential fault causes and solutions are generated and prompted to the user.

[0062] When the overall health status score belongs to the third overall interval, all sub-item scores belong to the third sub-interval, and there are at least two fault warning signals, the health status level is determined to be a fault level, a fault diagnosis report and maintenance guidance are generated, and the user is notified.

[0063] Among them, from the first total interval to the second total interval and the third total interval, and from the first sub-interval to the second sub-interval and the third sub-interval, the interval scores gradually decrease, indicating a worse health status.

[0064] In one example, the following levels are defined: Normal level: Total score ≥ 85 points, all sub-item scores ≥ 70 points, no warning signals; Warning level: Total score 60-84 points, or any sub-item score 50-69 points, triggering 1-2 warning signals; Fault level: Total score < 60 points, or any sub-item score < 50 points, or triggering 3 or more warning signals. The health status level determination results are updated in real time with the latest data, with an update cycle of ≤ 5 minutes. Alternatively, fuzzy C-means clustering can be used to replace thresholds for classifying health levels, in order to better adapt to personalized data distributions.

[0065] Furthermore, based on the results of fault prediction and health assessment, corresponding decision-making suggestions are provided to users. When the T-BOX is in normal condition, routine maintenance suggestions are provided; when the T-BOX is in a warning state, users are promptly reminded to perform inspection and maintenance, and potential fault causes and solutions are provided; when the T-BOX malfunctions, a fault diagnosis report and repair guidance are quickly provided and users are notified to help them quickly troubleshoot the fault. This fault information can be sent to the vehicle's infotainment display via Ethernet messages.

[0066] In practical implementation, the outputs of the fault prediction ensemble model and the health status assessment model adopt a standardized data format, including: Fault prediction results: Fault probability distribution table for the next 7 / 15 / 30 days (accurate to 0.1%), TOP3 high-risk fault types and confidence levels; Health assessment report: Overall health score, detailed scores for each sub-item, level identifier and analysis of deduction reasons; Warning signal list: Includes warning type, risk level (high / medium / low), recommended handling measures and timeliness requirements. The output interface supports CAN bus message (100ms cycle) and Ethernet JSON format push, meeting automotive-grade real-time and reliability requirements (MTBF≥10000 hours).

[0067] In a second possible implementation, step S105 may further include: If at least one fault warning signal is present, or if the chip is continuously powered on for more than a preset time, and if the vehicle has been turned off and locked, or has been in sleep mode for more than a predetermined time and has not received a remote command, it is determined that the vehicle-mounted T-BOX needs to perform a power-on operation. The system learns the user's non-use time window based on the user's vehicle usage time distribution characteristics, and performs a power-on operation on the vehicle T-BOX within the non-use time window.

[0068] This application's embodiments design a multi-dimensional combination of power-on trigger conditions, such as when the chip is continuously powered on for more than a threshold (which can be adjusted according to the vehicle configuration), or when the performance of the communication module chip or the positioning accuracy drops to a warning threshold, and at the same time the vehicle is turned off, locked, and in sleep mode for more than 90 minutes, or there are no diagnostic / remote commands. Combined with a self-learning time window (using the extracted feature representation to learn the user's vehicle usage time period value to avoid peak user usage times), the T-BOX core chip is automatically re-powered during the vehicle's locked and sleep mode, achieving seamless power-on control and recording the changes in performance parameters before and after power-on.

[0069] This application embodiment also found that storage modules such as the eMMC (embedded multimedia card) of T-BOX have a lifespan problem, and their premature failure is mainly due to excessive use of CMD8 instructions (initialization instructions) and abnormal power supply behavior (such as frequent triggering of low voltage detection LVD). Therefore, in the third possible implementation, step S105 may further include: When any of the following is detected: CMD8 instruction execution frequency exceeds the limit, LVD trigger count is abnormal, or eMMC storage health sub-item score is below the threshold, an instruction rate limiting policy is implemented to restrict CMD8 instruction execution, and / or power protection parameters are adjusted to optimize the LVD trigger threshold.

[0070] Furthermore, the management system in this embodiment may also include a data storage and management module, responsible for storing and managing the collected raw data, preprocessed data, model training data, seamless power-on execution records and performance optimization logs, eMMC lifespan degradation data, instruction execution logs, power supply anomaly event records, user vehicle usage history data, and fault prediction and health assessment results. Distributed database technology is employed to ensure data security, reliability, and scalability. Data is periodically uploaded to the vehicle manufacturer's TSP platform via a data reporting module. Simultaneously, data query and statistical analysis functions are provided to facilitate further data mining and utilization by users.

[0071] Please see Figure 2 , Figure 2 This is a second flowchart illustrating a management method for a vehicle network control unit, provided as another embodiment of this application. Figure 2As shown in the illustration, the control method provided in this application includes the following steps: First, the data acquisition module collects various operational data from the vehicle-mounted T-BOX, such as hardware status data, software operational data, and vehicle operational status data. Second, the data preprocessing module performs data cleaning, noise reduction, normalization, feature extraction, and selection on the various operational data to obtain various operational feature representations. Third, the fault prediction ensemble model outputs the fault prediction results of the vehicle-mounted T-BOX, and the health status assessment model determines the health status assessment results of the vehicle-mounted T-BOX. Fourth, by combining the fault prediction results and health status assessment results of the vehicle-mounted T-BOX, fault handling actions are executed; the sensorless power-on control unit automatically performs a power-on operation on the T-BOX core chip during vehicle lock-up and sleep periods, and the eMMc optimization control unit executes command current limiting strategies and adjusts power protection parameters. Fifth, all relevant data is stored in a distributed database and uploaded to the vehicle manufacturer's TSP platform.

[0072] This application provides a management method for a vehicle network control unit. Through multi-module collaborative innovation, it solves the problems of delayed fault warning, one-sided health assessment, and low efficiency of passive maintenance in the prior art. Compared with the prior art, it has the following technical effects: First, the comprehensiveness and reliability of multi-source data acquisition have been significantly improved, laying a high-quality data foundation for fault prediction.

[0073] In existing technologies, data acquisition by vehicle-mounted T-BOX is mostly limited to a single hardware status or basic communication parameters, resulting in problems such as insufficient acquisition dimensions, low accuracy, and poor interface reliability, leading to incomplete capture of fault characteristics. This application's embodiments achieve high-precision acquisition of multi-dimensional data through a distributed architecture and standardized interface design. It covers four major categories of data: hardware status (chip temperature, voltage, eMMC bad blocks, etc.), software operation (process status, memory usage, etc.), communication performance (RSRP, packet loss rate, etc.), and vehicle operation (vehicle speed, ignition status, etc.), increasing the acquisition dimensions by more than 60% compared to existing technologies. Employing high-precision sensors (temperature accuracy ±0.5℃, voltage error ≤0.2%), highly reliable interfaces (CAN FD / 1000BASE-T1 Ethernet, end-to-end latency ≤10ms), and 2500Vrms isolation protection, data integrity is improved to 99.8%, with outlier rates ≤0.5%, solving the problems of high data noise and easy data loss in existing technologies. For the first time, a correlation model is established between eMMC lifetime decay and CMD8 instruction strength and LVD triggering behavior, enabling accurate early warning of storage failure risks and filling the gap in existing technologies for monitoring the health status of storage modules.

[0074] Second, more precise data preprocessing significantly improves feature effectiveness and reduces model redundancy.

[0075] Existing data preprocessing technologies often employ simple filtering or single normalization methods, without designing specific processing logic for the temporal and multi-source characteristics of T-BOX data, resulting in low feature discrimination and dimensional redundancy in the input model.

[0076] This application's embodiments achieve dual optimization of data quality and feature effectiveness through a pipelined preprocessing architecture: combining the 3σ principle and Kalman filtering, outliers, missing values, and noise are precisely processed, resulting in a signal-to-noise ratio improvement of ≥15dB after filtering, thus solving the data interference problem in the complex electromagnetic environment of vehicles; innovatively extracting correlation features such as chip runtime and positioning accuracy, eMMC command strength, and power cycle interval, and using mutual information method and recursive feature elimination technology, the feature dimension is controlled to within 50 dimensions (60% reduction compared to the original data), while retaining core fault features, improving model input efficiency by 40%; employing a combination strategy of linear normalization, Z-score standardization, and one-hot encoding, the dimensional differences are eliminated, ensuring that multi-source data can be fused and analyzed, providing high-quality feature vectors for subsequent model training.

[0077] Third, the accuracy and lead time of fault prediction are significantly improved, enabling early warning to replace post-failure maintenance.

[0078] Existing fault prediction technologies mostly rely on simple threshold judgments or single models, which are lagging in early warning of gradual faults (such as eMMC lifespan decay and communication module performance degradation), with a prediction accuracy of less than 70%, and cannot clearly define the fault type and probability.

[0079] This application's embodiments achieve accurate prediction and early warning through an integrated architecture of "basic model + specialized sub-models": the basic model adopts an improved LSTM + attention mechanism to effectively capture time-dependent features; for core faults such as communication module attenuation, positioning drift, and eMMC failure, specialized sub-models (such as the Transformer time-series prediction model) are designed, with an eMMC lifetime attenuation prediction accuracy of ≥92% and a communication performance attenuation prediction error of ≤5%; a hybrid density network (MDN) is introduced to output probability distribution, clarifying the fault probability and the top 3 high-risk types for the next 7 / 15 / 30 days, thus solving the rigid judgment defect of existing technologies that are either / or. Fourth, the health assessment system is more comprehensive, enabling dynamic quantitative management of T-BOX status.

[0080] Existing health assessment technologies often focus only on hardware operating parameters, lacking multi-dimensional weighted analysis. The assessment results are one-sided (e.g., presented only as a "normal / fault" dichotomy) and cannot reflect the overall health trend of the equipment.

[0081] This application's embodiments construct a multi-dimensional weighted scoring system to achieve refined management of health status: covering four major sub-items: chip performance (35%), eMMC storage (30%), communication link (20%), and power system (15%). The weights are quantified using the Analytic Hierarchy Process (AHP), and the total score accurately reflects the overall status of the equipment. The system is further subdivided into three levels of status: "normal (≥85 points) - warning (60-84 points) - fault (<60 points)". Combined with the analysis of the reasons for deductions in sub-items, the health status can be traced. The evaluation results are updated in real time (cycle ≤5 minutes), which is more timely than the existing technology (update cycle ≥30 minutes), providing users with dynamic status feedback.

[0082] 5. Added proactive prevention and control logic to reduce the failure rate and extend equipment life.

[0083] Existing technologies mostly adopt a passive response mode (post-failure maintenance) for T-BOX failures, lacking an active intervention mechanism, which leads to repeated failures and shortened equipment lifespan.

[0084] This application's embodiments achieve proactive fault prevention through the innovative design of the decision support module, solving the performance degradation problem caused by long-term operation. Experimental verification shows that it can reduce temporary communication module failures by 30%. The newly added "eMMC optimization control unit" reduces the damage to the storage module caused by excessive CMD8 use and LVD anomalies through strategies such as instruction current limiting and power parameter adjustment, extending the average lifespan of eMMC by more than 20%. Based on the early warning results, it provides targeted maintenance suggestions (such as "high-risk eMMC faults suggest checking power stability"), improving fault handling efficiency by 50% and reducing user downtime.

[0085] VI. Systematized data storage and management system supports full lifecycle traceability and iterative optimization.

[0086] Existing technologies suffer from fragmented and monotonous data storage, lacking systematic management of fault logs and optimization records, making it difficult to support model iteration and fault tracing.

[0087] This application embodiment integrates a distributed database and a TSP platform to achieve full lifecycle management of data: storing all data such as raw data, preprocessing results, fault prediction reports, and contactless power-on records, supporting multi-dimensional queries and statistical analysis; periodically uploading data to the vehicle manufacturer's TSP platform to provide the vehicle manufacturer with a basis for batch equipment status analysis, helping to optimize product iteration; improving data security and scalability, meeting automotive-grade storage requirements (MTBF≥10000 hours).

[0088] In summary, this invention significantly improves the accuracy of fault prediction, the comprehensiveness of health assessment, and the proactivity of fault handling for vehicle-mounted T-BOX through collaborative innovation in multi-source data acquisition, precise preprocessing, intelligent prediction and evaluation, proactive prevention and control, and systematic data management. It effectively reduces the fault incidence rate (experimentally verified to be more than 40% lower), extends equipment life, and enhances user experience, demonstrating significant engineering application value.

[0089] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a vehicle networking control unit management system provided in an embodiment of this application. Figure 3 As shown, the management system 300 includes: The data acquisition module 310 is used to acquire various operating data of the vehicle-mounted T-BOX; the types of operating data include: hardware status data, software operating data and vehicle operating status data; Feature extraction module 320 is used to extract multiple operational feature representations from the multiple operational data; The fault prediction module 330 is used to input the various operational feature representations into a pre-trained fault prediction set model to obtain the fault prediction result of the vehicle-mounted T-BOX; wherein, the fault prediction set model includes an overall fault prediction model and at least one fault prediction sub-model for a specific fault type of the vehicle-mounted T-BOX. The health assessment module 340 is used to input the various operating data into a pre-built health status assessment model to determine the health status assessment result of the vehicle-mounted T-BOX. The fault handling module 350 is used to combine the fault prediction results of the vehicle-mounted T-BOX and the health status assessment results to perform fault handling actions.

[0090] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.

[0091] The memory 420 stores machine-readable instructions that can be executed by the processor 410. When the electronic device 400 is running, the processor 410 and the memory 420 communicate via the bus 430. When the machine-readable instructions are executed by the processor 410, the steps of the management method of the vehicle networking control unit as described in the above method embodiment can be executed. For specific implementation details, please refer to the method embodiment, which will not be repeated here.

[0092] This application also provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it can execute the steps of the vehicle networking control unit management method as described in the above method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0093] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0094] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0095] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0096] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0097] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A management method of a vehicle networked control unit, characterized by, The management method comprises: Obtaining a plurality of running data of the vehicle-mounted T-BOX; the types of the running data include: hardware state data, software running data and vehicle running state data; Extracting a plurality of running feature representations from the plurality of running data; Inputting the plurality of running feature representations into a pre-trained fault prediction ensemble model to obtain a fault prediction result of the vehicle-mounted T-BOX; wherein the fault prediction ensemble model comprises a fault prediction overall model and at least one fault prediction sub-model specific to a specific fault type of the vehicle-mounted T-BOX; Inputting the plurality of running data into a pre-constructed health state evaluation model to determine a health state evaluation result of the vehicle-mounted T-BOX; Comprehensively executing a fault handling action based on the fault prediction result and the health state evaluation result of the vehicle-mounted T-BOX.

2. The method of claim 1, wherein, Extracting a plurality of vehicle running feature representations from the plurality of running data, comprising: Preprocessing the plurality of running data; the preprocessing mode comprises data cleaning and normalization; Extracting a plurality of candidate vehicle running feature representations from the preprocessed plurality of running data; According to the relevance of the candidate feature representation and the fault prediction, the plurality of vehicle running feature representations are selected from the plurality of candidate vehicle running feature representations.

3. The method of claim 2, wherein, According to the relevance of the candidate feature representation and the fault prediction, the plurality of vehicle running feature representations are selected from the plurality of candidate vehicle running feature representations, comprising at least one of the following: Determine the variance of each candidate vehicle running feature representation within a predetermined time period, and select the candidate vehicle running feature representation with a variance greater than a preset variance threshold as a vehicle running feature representation; Determine the mutual information value between each candidate vehicle running feature representation and the historical fault label of the vehicle-mounted T-BOX, and select the candidate vehicle running feature representation with a mutual information value greater than a preset information threshold as a vehicle running feature representation; Based on the random forest model algorithm, each candidate vehicle running feature representation is scored for feature importance, and the candidate vehicle running feature representation with a contribution degree lower than a preset threshold is removed step by step in a recursive manner to obtain a vehicle running feature representation.

4. The method of claim 2, wherein, Extracting a plurality of candidate vehicle running feature representations from the preprocessed plurality of running data, comprising: For the chip continuous running time length and positioning accuracy correlation feature, the mean change rate of the positioning drift error is determined after the chip continuous power-on time length accumulates a predetermined time length, and the correlation between the chip continuous running time length and the positioning accuracy is quantified by Pearson correlation; For the eMMC instruction usage intensity feature, based on the total number of CMD8 instruction executions within a predetermined time period and the vehicle power-on time length within the predetermined time period, the average usage intensity is determined; For the power cycle interval feature, according to the time stamp of each power cycle, the time length distribution of the interval between adjacent two power-ons is calculated, and the statistical value of the time length distribution of the power-on interval is extracted; For the LVD trigger and eMMC write amount association feature, a time correlation window of the LVD trigger time and the eMMC write operation is established, and the ratio of the number of LVD triggers to the amount of eMMC write data in a unit time of the time correlation window is determined; For the user vehicle use period distribution feature, based on the timestamp of the ignition state signal, a clustering algorithm is used to divide the vehicle use period into multiple periods; the power-on frequency, average running time and communication data volume ratio of each period are counted to form a user vehicle use feature matrix.

5. The method of claim 1, wherein, The fault prediction sub-model includes at least one of the following: a communication module chip performance degradation prediction sub-model, a positioning chip drift fault prediction sub-model, and an eMMC life degradation prediction sub-model; then input the multiple running feature representations into the pre-trained fault prediction ensemble model to obtain the fault prediction result of the vehicle-mounted T-BOX, including: Input the multiple running feature representations into the fault prediction overall model, which uses an improved long short-term memory network as the core feature extractor to perform time series feature coding on the multiple running feature representations, combines with the attention mechanism to strengthen the key time step feature weight, and uses a hybrid density network structure in the output layer to obtain the prediction result of the fault probability distribution and the fault type, and triggers a fault warning signal according to the threshold value; Input the communication module chip performance related feature representation into the communication module chip performance degradation prediction sub-model, construct a performance degradation curve through an SVM regressor, predict the probability that the communication performance is lower than the threshold value within a predetermined time in the future based on the performance degradation curve, and trigger a fault warning signal according to the threshold value; Input the positioning chip performance related feature representation into the positioning chip drift fault prediction sub-model, use a bidirectional long short-term memory network to determine the forward and backward correlation of the positioning drift trend, set a sliding time window to extract the statistical value of the drift error, and trigger a fault warning signal according to the threshold value; Input the eMMC performance related feature representation into the eMMC life degradation prediction sub-model, use a time series prediction model of Transformer and a self-attention mechanism to capture the relationship between the bad block growth of eMMC and the eMMC instruction usage intensity feature, determine the storage failure risk probability, and trigger a fault warning signal according to the threshold value.

6. The method of claim 5, wherein, Input the multiple running data into the pre-constructed health state evaluation model to determine the health state evaluation result of the vehicle-mounted T-BOX, including: Input the multiple running data into the health state evaluation model to determine the score of at least one of the following sub-items: chip performance health sub-item score, eMMC storage health sub-item score, communication link health sub-item score, and power system health sub-item score; According to the sub-item scores and corresponding weights, the health state total score of the vehicle-mounted T-BOX is obtained.

7. The method of claim 6, wherein, Integrate the fault prediction result and the health state evaluation result of the vehicle-mounted T-BOX to perform fault handling actions, including: When the health state total score belongs to the first total interval, each sub-item score belongs to the first sub-interval, and there is no fault warning signal, the health state level is determined to be normal, and a regular maintenance suggestion is generated; When the total health score belongs to a second total interval, each sub-item score belongs to a second sub-interval, and there is at least one fault warning signal, the health state level is determined as a warning level, potential fault causes and solutions are generated, and the user is prompted; When the total health score belongs to a third total interval, each sub-item score belongs to a third sub-interval, and there are at least two fault warning signals, the health state level is determined as a fault level, a fault diagnosis report and maintenance guidance are generated, and the user is prompted.

8. The method of claim 5, wherein, The fault prediction result of the vehicle-mounted T-BOX and the health state evaluation result are comprehensively considered to perform a fault handling action, and the method further includes: When there is at least one fault warning signal or the chip is continuously powered on for more than a preset time length, if the vehicle has been turned off and locked, has been in sleep for more than a predetermined time length, and has not received a remote instruction, it is determined that the vehicle-mounted T-BOX needs to perform a power-on operation; According to the user's vehicle use time window learned from the user's vehicle use time period distribution characteristics, the power-on operation of the vehicle-mounted T-BOX is performed in the vehicle use time window.

9. The method of claim 7, wherein, The fault prediction result of the vehicle-mounted T-BOX and the health state evaluation result are comprehensively considered to perform a fault handling action, and the method further includes: When any one of the following conditions is monitored: the execution frequency of the CMD8 instruction exceeds the standard, the number of LVD triggers is abnormal, and the eMMC storage health sub-item score is lower than the threshold value, an instruction flow limiting strategy is performed to limit the execution of the CMD8 instruction, and / or, the power supply protection parameter is adjusted to optimize the LVD trigger threshold value.

10. A management system of vehicle telematics control units, characterized by The management system includes: A data acquisition module for acquiring a plurality of operating data of the vehicle-mounted T-BOX; the types of operating data include: hardware state data, software operating data, and vehicle operating state data; A feature extraction module for extracting a plurality of operating feature representations from the plurality of operating data; A fault prediction module for inputting the plurality of operating feature representations into a pre-trained fault prediction ensemble model to obtain a fault prediction result of the vehicle-mounted T-BOX; wherein the fault prediction ensemble model includes a fault prediction overall model and at least one fault prediction sub-model for a specific fault type of the vehicle-mounted T-BOX; A health evaluation module for inputting the plurality of operating data into a pre-constructed health state evaluation model to determine a health state evaluation result of the vehicle-mounted T-BOX; A fault handling module for comprehensively considering the fault prediction result of the vehicle-mounted T-BOX and the health state evaluation result to perform a fault handling action.

Images