Optimization method and system based on Internet of Vehicles SIM card signal monitoring

Abstract

The invention provides an optimization method and system based on Internet of Vehicles SIM card signal monitoring. The method comprises the following steps: collecting signal data of an Internet of Vehicles SIM card; constructing a support vector machine model based on the collected data to distinguish a normal state and an abnormal state of the SIM card signal; the support vector machine model is used to carry out real-time abnormity identification on the Internet of Vehicles SIM card signal; calculating a model evaluation index according to the identification result and obtaining a comprehensive evaluation score; and performing optimization adjustment on the support vector machine model based on the comprehensive evaluation score to realize closed-loop optimization. According to the technical scheme, efficient monitoring and dynamic optimization of SIM card signal abnormity are realized, so that the fault early warning accuracy is improved, and the maintenance cost is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle data application, and in particular relates to an optimization method and system based on vehicle networking SIM card signal monitoring. Background Art

[0002] With the development of connected vehicles, in-vehicle SIM cards have become a key module for vehicles to achieve remote communication, navigation, and vehicle condition monitoring. Once the SIM card encounters problems such as communication interruption, weak signal, or binding abnormality, it often causes the vehicle's remote functions to fail, which not only affects the user's driving experience, but also poses challenges to safety assurance such as emergency rescue and fleet dispatch. Currently, the comprehensive repair and compensation costs caused by SIM card abnormalities in the market are usually in the range of 1,000-3,000 yuan. When users encounter abnormal communication or in-vehicle system functions while driving, they need to report the problem through customer service, and the service station will then bring the vehicle in for inspection and repair. This process will incur high expenses such as first aid fees, labor costs, and material costs. In addition, due to the failure to monitor in time before the failure occurs, the repair efficiency is low and the cost is high.

[0003] The industry has proposed several SIM card signal monitoring and exception handling solutions: CN117412329A discloses a real-time network monitoring method based on SIM card status, network registration status and dial-up status, which restores communication through strategies such as resetting the modem and switching the flight mode.

[0004] CN110807847A proposes to determine whether the terminal is offline or abnormal by reading the calibration value uploaded by the Internet of Vehicles module and comparing it with the preset threshold.

[0005] CN110995823A monitors the data card status and flow information of the offline terminal and executes the corresponding offline processing strategy.

[0006] The aforementioned technical solutions primarily rely on comparing the communication parameters of a single or a few SIM cards against preset thresholds. Improper threshold settings can easily lead to false alarms caused by signal fluctuations or missed alarms due to actual anomalies exceeding the threshold range. Furthermore, these solutions lack a mechanism for optimizing monitoring models using back-end maintenance feedback data, resulting in an inability to dynamically close the loop between monitoring results and actual fault conditions. The resulting technical issues include: 1. The monitoring method lacks adaptability. The fixed threshold strategy cannot take into account the differences in signal characteristics under changing operating environments, which easily leads to low accuracy. 2. Fault warnings are not timely enough; maintenance can only be carried out after users provide feedback, making it impossible to detect faults in advance and respond quickly. 3. The model cannot be continuously optimized. It lacks a closed-loop feedback mechanism that uses actual failure and compensation data from maintenance stations to calibrate monitoring strategies, making it difficult to improve monitoring reliability. Summary of the Invention

[0007] The purpose of the present invention is to address the deficiencies of the above-mentioned background technology and provide an optimization method and system based on vehicle network SIM card signal monitoring, so as to realize a technical solution of efficient monitoring and dynamic optimization of SIM card signal anomalies, thereby improving the accuracy of fault warning and reducing maintenance costs.

[0008] The technical solution adopted by the present invention is: an optimization method based on vehicle network SIM card signal monitoring, comprising the following steps: Collect signal data from the Internet of Vehicles SIM card; Building a support vector machine model based on the collected data to distinguish between a normal state and an abnormal state of the SIM card signal; Using the support vector machine model to perform real-time anomaly recognition on the vehicle networking SIM card signal; Calculate the model evaluation index based on the recognition results and obtain the comprehensive evaluation score; The support vector machine model is optimized and adjusted based on the comprehensive evaluation score to achieve closed-loop optimization.

[0009] In the above technical solution, in the step of collecting vehicle network SIM card signal data, the collected data is preprocessed, including unique-hot encoding of vehicle operating area information, normalizing the data transmission rate, and calculating the product of the signal strength and communication quality as the input feature of the input support vector machine model.

[0010] In the above technical solution, the calculation process of the comprehensive evaluation score includes: The identified SIM card signal anomaly is compared with the vehicle maintenance information to calculate the hit rate and coverage of the support vector machine model prediction, and a comprehensive evaluation score is calculated based on the hit rate, coverage and the value assessment coefficient related to the vehicle failure compensation amount, which is used to comprehensively evaluate the model prediction performance.

[0011] In the above technical solution, in the closed-loop optimization step, a correction coefficient is set for the comprehensive evaluation score, and the correction coefficient is dynamically adjusted based on the fluctuation of the vehicle failure compensation amount to correct the comprehensive evaluation score.

[0012] In the above technical solution, in the closed-loop optimization step, the vehicle data that is misjudged as abnormal by the support vector machine model and the vehicle data with SIM card failure that is not recognized by the model are added to the training samples of the support vector machine model, and the support vector machine model is retrained.

[0013] In the above technical solution, the closed-loop optimization step also includes: calculating and generating an urgency feature value representing the fault priority based on the vehicle fault level, user urgency and estimated repair time information, and adding the urgency feature value as an additional input feature to the training sample of the support vector machine model, and retraining the support vector machine model.

[0014] In the above technical solution, When the comprehensive evaluation score is greater than or equal to a first threshold, keeping the model parameters unchanged; When the comprehensive evaluation score is less than a first threshold, adding vehicle data misjudged and unrecognized by the support vector machine model to the training samples of the support vector machine model to retrain the model; When the comprehensive evaluation score is less than a second threshold, dynamically adjusting the correction coefficient, and adding the urgency feature value as an additional input feature to the training sample of the support vector machine model to retrain the model; The second threshold is smaller than the first threshold.

[0015] In the above technical solution, During the preset monitoring period T, A={vehicle ID_j|the support vector machine model determines that the SIM card is abnormal within period T}; B={Vehicle ID_j|Confirmed by the repair station as SIM card failure and compensation generated within period T}; C=A∩B.

[0016] HitRate=|C| / |A|; CoverageRate=|C| / |B|; ValueCoefficient=(Σj∈CCost_j) / (Σj∈BCost_j), P=ValueCoefficient×HitRate+(1−ValueCoefficient)×CoverageRate+CorrectionFactor; Where Cost_j is the compensation amount incurred by vehicle j due to SIM card failure within period T; HitRate represents the hit rate, CoverageRate represents the coverage rate, P represents the comprehensive evaluation score; CorrectionFactor represents the correction factor.

[0017] In the above technical solution, The correction coefficient is dynamically determined as follows: The volatility indicator Volatility is calculated using the following formula: Volatility = σ_{VC} / μ_{VC}, Where σ_{VC} is the standard deviation of the compensation amounts for all vehicles with SIM card failures within the preset monitoring period T, and μ_{VC} is the mean of the compensation amounts during the same period; Candidate interval setting: If Volatility is greater than the set threshold, set the candidate α∈[0.3, 0.5]; otherwise, set the candidate α∈[0.1, 0.3); RMSE evaluation: For each α_k in the candidate interval, calculate CorrectionFactor_k=α_k×Volatility, Based on this, the revised comprehensive evaluation score sequence P_k is obtained, and the root mean square error is calculated with the preset target score; Take the α_k with the smallest root mean square error as the adjustment parameter α0 of this cycle, And let CorrectionFactor = α0×Volatility.

[0018] In the above technical solution, The urgency characteristic value is determined according to the following steps: (1) For each vehicle sample, obtain a) Fault level L∈{1=minor, 2=general, 3=serious}, b) User urgency U∈{1=non-urgent, 2=priority, 3=immediate}, c) Estimated maintenance time T, in hours; (2) Calculate linear scores for the three emergency levels: mild, moderate, and severe: ; in, represents the weight coefficient; b i represents a constant term; The linear score z is used as the three types of Softmax input and mapped to P (light), P (medium), and P (heavy): The category corresponding to the maximum probability is taken and coded as “mild=0, medium=1, severe=2” to form the urgency characteristic value.

[0019] The present invention also provides an optimization system based on vehicle networking SIM card signal monitoring, which is used in the method described in the above technical solution, including: Data acquisition module, used to collect signal data from the Internet of Vehicles SIM card; A model training module is used to build a support vector machine model based on the collected data to distinguish between the normal state and the abnormal state of the SIM card signal; A state recognition module, used to use the support vector machine model to perform real-time abnormality recognition on the vehicle network SIM card signal; Model evaluation module, used to calculate model evaluation indicators based on recognition results and obtain comprehensive evaluation scores; A model optimization module is used to optimize and adjust the support vector machine model based on the comprehensive evaluation score to achieve closed-loop optimization.

[0020] The present invention also provides a non-transitory computer-readable storage medium storing a computer-executable program, wherein when the program stored thereon is executed by a processor, the processor executes the method described in the above technical solution.

[0021] The present invention also provides a data processing device configured in a server, comprising: a processor, a memory and a program thereon, wherein the program enables the device to execute the method described in the above technical solution.

[0022] The beneficial effects of the present invention are: the present invention integrates and automatically operates the entire process from data collection to real-time identification, and then to comprehensive evaluation and closed-loop optimization, realizing end-to-end intelligent monitoring and self-optimization of SIM card signal anomalies. It can continuously improve the accuracy of fault identification and response speed without human intervention, and significantly reduce the cost of manual inspections.

[0023] Furthermore, the present invention uniquely encodes the operating area, normalizes the rate, and mines new features of signal strength × communication quality to improve the characterization ability of input features, eliminate the dimensional differences of the original data, and enhance key signal information, so that the SVM model can more accurately capture abnormal patterns, thereby reducing false positives and missed positives.

[0024] Furthermore, the present invention introduces three evaluation indicators, namely "hit rate", "coverage rate" and "value assessment coefficient", to conduct multi-dimensional quantitative evaluation of model performance. It not only measures the accuracy of model alarms, but also takes into account the risk of missed detection and the importance of economic losses, providing a comprehensive and quantifiable decision-making basis for subsequent optimization.

[0025] Furthermore, the present invention introduces an adaptive correction factor based on compensation fluctuations in the comprehensive score, automatically balancing the impact of fluctuations in the value assessment weights. This maintains the stability of the scoring system even in the face of significant fluctuations in compensation amounts, avoids distortions in the model assessment caused by extreme compensation events, and ensures the reliability of continuous optimization.

[0026] Furthermore, this method adds samples that were misjudged (false positives / missed negatives) back into the training set and retrains the SVM model. This continuous replenishment of the training set with historically weak samples makes the model more adaptable to edge cases and further improves overall detection coverage.

[0027] Furthermore, the present invention calculates an urgency feature based on fault level, user urgency, and repair duration, and incorporates it into the model as an additional input. This allows the model to automatically perceive fault priority, assigning higher response weights to high-urgency events, and improving the recall rate and early warning time for serious faults.

[0028] Furthermore, the present invention sets a two-level threshold, combines the sample addition and correction coefficient strategies, and selectively executes different optimization actions according to the scores, realizing the "score trigger, strategy grading" optimization mechanism, saving computing overhead when the score is good, and specifically supplementing samples, optimizing parameters or adding emergency features when the score is poor, taking into account both efficiency and effect.

[0029] Furthermore, the present invention clearly defines the hit rate, coverage rate and value assessment coefficient based on the vehicle sets A and B and the compensation amount within a preset period, and uniformly calculates the comprehensive score under the same period caliber, eliminating the ambiguity of statistical caliber and dimension, ensuring that the evaluation indicators are comparable in different time windows and different vehicle clusters, and providing a stable and repeatable evaluation standard for model performance tracking and cross-period optimization.

[0030] Furthermore, the present invention first calculates the volatility of compensation amounts and then adaptively selects α within a candidate interval based on the principle of minimizing the root mean square error (RMSE) to determine the final correction factor. With the accuracy of approaching the target score as the optimization goal, the optimal α is dynamically determined, ensuring that the correction factor is responsive to compensation fluctuations while maximizing the stability of the model's overall score and prediction accuracy.

[0031] Furthermore, this invention assigns independent weights and biases to the three levels of "mild," "moderate," and "severe," generating three linear scores that are mapped to probabilities via Softmax. The maximum value is then encoded as the urgency feature. This combination of multi-dimensional linear combinations and probabilistic mapping allows for precise differentiation of different urgency levels, improving the ability to provide graded early warnings for faults of varying severity and enabling more targeted resource scheduling and response strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but they do not constitute a limitation to the present invention.

[0034] Example 1 like Figure 1 As shown, the present invention provides an optimization method based on vehicle network SIM card signal monitoring, comprising the following steps: Collect signal data from the Internet of Vehicles SIM card; Building a support vector machine model based on the collected data to distinguish between a normal state and an abnormal state of the SIM card signal; Using the support vector machine model to perform real-time anomaly recognition on the vehicle networking SIM card signal; Calculate the model evaluation index based on the recognition results and obtain the comprehensive evaluation score; The support vector machine model is optimized and adjusted based on the comprehensive evaluation score to achieve closed-loop optimization.

[0035] This embodiment demonstrates a SIM card signal anomaly detection and optimization process that is collaboratively deployed between an in-vehicle terminal and a back-end server. Combining sample data from at least three vehicles, it details the entire process of data processing, model training, evaluation index calculation, and model optimization, enabling those skilled in the art to replicate the process.

[0036] The entire system adopts a collaborative architecture between an onboard terminal (T-Box) and a backend server. The onboard terminal is mainly responsible for real-time collection and preliminary processing of vehicle data; the backend server is responsible for model training, predictive analysis, and comprehensive evaluation. Specifically: A SIM card signal monitoring device is installed on the vehicle terminal. It collects real-time SIM card signal data (such as signal strength and communication quality) and vehicle location information during vehicle operation, and uploads this data to the backend via the vehicle's communication module. If the terminal detects a significant anomaly (such as a prolonged period of no signal), it can also record the event locally and notify the driver.

[0037] The backend server receives signal data from multiple vehicle terminals and stores it on a big data platform. The server aggregates and preprocesses the data, trains a SIM card anomaly detection model (such as an SVM classification model), and analyzes the data uploaded by each vehicle in real time to determine if the SIM card is abnormal. If a suspected anomaly is detected in a vehicle, the server notifies customer service or maintenance personnel and records actual repair feedback to continuously optimize the model.

[0038] Through the above division of labor, the terminal side realizes front-end data collection and alarm prompts, and the server side realizes centralized modeling and evaluation optimization, forming a closed-loop collaborative architecture of detection and feedback.

[0039] Table 1 constructs historical signal sample data of three vehicles, including signal strength, communication quality, location, data transmission rate, and corresponding real repair compensation amount, user urgency, and other information, which is used to demonstrate the model training and evaluation process.

[0040] Table 1 Sample data structure Combined with the above sampling data, the entire process of this embodiment is described in detail, including the following steps: Data collection (terminal side): After the vehicle is started, the on-board terminal continuously collects SIM card-related signal data and metadata, including: SIM card connection status, vehicle operating area, signal strength, communication quality, and data transmission rate. In this embodiment, Car1 and Car3 are located in an urban area, and Car2 is located in a suburban area. Each vehicle regularly records the current signal strength (such as -70 dBm, etc.) and communication quality (such as 60%, etc.), as well as the real-time mobile data transmission rate (such as 50 KB / s, etc.) during driving. The terminal adds a timestamp to this raw data and packages it, then uploads it to the backend via the mobile network. For offline scenarios (such as no SIM card signal), the terminal can cache the data locally and retransmit it when the connection is restored.

[0041] Data preprocessing (server side): After receiving the data stream uploaded by the terminal, the backend server cleans and pre-processes the data before storing it in the database. The main operations include: Data filtering: Eliminate data records with missing values ​​or abnormal errors (such as abnormal signal strength values).

[0042] Feature normalization: Normalize or standardize numerical features. For example, we can normalize the data transfer rate of different dimensions to a value between 0 and 1 to eliminate dimensionality effects. Suppose the normalized data transfer rates of Car1, Car2, and Car3 are 0.5, 0.3, and 1.0, respectively (corresponding to relative rates of 50, 30, and 100 KB / s).

[0043] Categorical encoding: Categorical features are represented numerically. For example, "Operating Region" uses one-hot encoding: Categories such as "Urban" and "Suburban" are assigned a binary feature. For example, the urban region is coded as (region_city=1, region_suburban=0), while the suburban region is (0,1). Car1 and Car3 belong to the urban region, so the region field in the feature vector can be represented as (1,0); Car2 belongs to the suburban region as (0,1).

[0044] After preprocessing, the data of each car is converted into a feature vector in a unified format. For example, the feature vector of Car1 after preprocessing can be expressed as: X Car1 =[Connection OK=1, Region_City=1, Region_Suburban=0, Signal Strength=−70, Communication Quality=0.60, Transmission Rate Normalized=0.5, ...] (Only some features are listed here. "Connection Normal" = 1 indicates that the SIM card was connected during the acquisition. Actual features also include the comprehensive signal quality calculated below.) After the fault level and user urgency are entered into the maintenance / customer service system, they are connected to the vehicle data platform through the back-end system (such as CRM / work order system). The VIN + timestamp is associated with the signal data in the database, as follows: In the repair work order system, maintenance technicians score SIM card failure events based on their impact (1 = minor, 2 = general, 3 = severe). The backend extracts the VIN, repair timestamp, and fault level fields from the work order table into the database via a standardized API.

[0045] When receiving a breakdown call, customer service records the driver's urgent request (1 = non-urgent, 2 = priority, 3 = immediate) and stores it in the same work order table or a dedicated dispatch table. The backend also pulls this field via the API and associates it with the same VIN and timestamp.

[0046] When a maintenance work order is generated, the system will fill in or automatically estimate the estimated work hours required for the repair. This field is named "Estimated Work Hours." This field is exported to the data warehouse through the Job Scheduling module.

[0047] Perform a left join of the three types of business data and the vehicle signal data using the same VIN + timestamp. If no corresponding work order is found at a certain time point, the vehicle is assumed to be fault-free at that time and the default values ​​(L=1, U=1, T=∞) are used. If multiple work orders exist, the one closest to the signal data time is used.

[0048] Eliminate obviously erroneous business data, such as records where T<=0, L, or U is outside the range of 1–3.

[0049] Convert the fault level L and user urgency U directly into integers; retain one decimal place for the estimated working time T (hours); and calculate its reciprocal 1 / T for subsequent model input.

[0050] Integer L and U are directly input as numerical features; if one-hot representation is required, three-dimensional one-hot encoding is performed on L and U respectively to obtain 6-dimensional binary features.

[0051] Perform Min–Max normalization on the 1 / T feature across the full sample range and map it to [0,1]. For example, if the maximum 1 / T = 1 / 0.5h = 2 and the minimum 1 / T = 1 / 10h = 0.1, then the normalized value is (1 / T–0.1) / (2–0.1).

[0052] For normal samples that have not experienced any failure, 1 / T can be filled with 0 (indicating that no maintenance is required) or the average normalized value of all samples.

[0053] When subsequent steps require model parameter optimization using newly added label features, the cleaned and encoded L, U, 1 / T_normalized, and their one-hot vectors are concatenated with the aforementioned signal features (signal strength, communication quality, integrated signal quality, regional one-hot encoding, rate normalization, etc.) in the same feature matrix.

[0054] Finally, the complete training / prediction input X = […, L, U, 1 / T_norm, …] is obtained, where L, U, and 1 / T_norm can directly affect the sensitivity of the subsequent SVM model to different fault priorities.

[0055] Through the above ETL process, the business information in the maintenance work order and the on-board signal data are seamlessly connected and standardized pre-processed on the back-end platform, laying a reliable data foundation for subsequent urgency feature generation and model training.

[0056] Feature construction (server side): Based on the preprocessed data, new derived features are created to enhance the information content of the model input. In this embodiment, the "comprehensive signal quality" feature is introduced, and the product of signal strength and communication quality is used to represent the overall stability of the signal. The formula is: Comprehensive signal quality = signal strength value × communication quality value.

[0057] For Car 1, the signal strength is -70 dBm (which can be converted to an absolute value of 70 or other proportional representation) and the communication quality is 60% (0.60). The overall signal quality is 70 × 0.60 = 42 (normalized). Car 2's signal strength is 80, and the quality is 0.50, multiplying the product to 40. Car 3's is 60 × 0.80 = 48. This new feature reflects the combined strength and quality of the signal; higher values ​​indicate better overall signal quality.

[0058] Target variable definition (server side): To perform supervised learning, it's necessary to define a label for each record. This example uses the SIM card failure status as the target variable (binary label): normal operation is recorded as 0, and signal anomalies are recorded as 1. In the training samples, this label is determined by historical maintenance data: for example, Car1 and Car2 are labeled 1 because they experienced actual SIM card failures; Car3, which did not experience an anomaly, is labeled 0. In this training dataset, each sample contains a feature vector and a corresponding failure label. The number of training samples in this example is significantly greater than 3, and the three sample data points above are provided for illustrative purposes only.

[0059] Model training (server side): A binary classification model for SIM card anomaly detection is trained using the support vector machine (SVM) algorithm. SVM separates "normal" (0) and "abnormal" (1) samples to the greatest extent by finding an optimal hyperplane in the high-dimensional feature space. Its optimization goal can be expressed as: Minimize: The constraints are: Among them, x i is the feature vector of the i-th sample, is the corresponding label (positive abnormality = 1, negative normal = 0), w is the model weight vector, b is the bias term, The slack variable allows a certain degree of classification error, and C is the penalty coefficient that controls the model's tolerance to misclassification.

[0060] In order to improve the nonlinear discrimination ability of the model, the Gaussian kernel function (RBF kernel) can be selected to map the features to a higher dimension; for example, setting the RBF kernel parameters , with a penalty coefficient C = 1 as an example. During training, w and b are continuously adjusted to maximize the margin between classification boundaries and minimize misclassifications. After training on sample data (and more historical data), the SVM model achieves optimal parameters. This model is able to predict whether a SIM card is abnormal based on the input features.

[0061] The weight vector w represents the importance of each feature in the discrimination (for example, in this embodiment, if the component of w corresponding to the "comprehensive signal quality" is large, it means that the feature has a more significant impact on fault judgment); the bias b determines the threshold position of the classification plane, which can balance the overall proportion of positive and negative classes; the larger the penalty parameter C, the more "zero tolerance" the model has for training errors, and the more complex the model (possibly overfitting); the smaller C is, the more it tolerates some errors to obtain a simpler boundary.

[0062] 6. Fault prediction (server side): The trained SVM model is deployed on the backend server for predictive analysis of each vehicle's real-time data (it can also be deployed to the terminal for local judgment as needed). When new vehicle data arrives, the server performs the same preprocessing and feature extraction as in the training, and then inputs it into the SVM model to calculate the output value. Output Indicates that the model determines that the SIM card may be abnormal, and 0 indicates normal. For the three cars in the example, if their data is used as new input: Car1: The overall signal quality is low (the signal is weak and the quality is average). The model may output an abnormal $1.

[0063] Car 2: Both the signal and quality are low. However, because the area is suburban (possibly with fewer base stations and more low normal signal conditions), the model may not classify it as an abnormality of 0 under the current parameters, which means a false positive (missed detection).

[0064] Car3: The signal strength and quality are high and should be considered normal, but the model is affected by noise and incorrectly marks it as an anomaly 1.

[0065] Assume that the model's judgment on this sample data is as follows: Car1 is abnormal (1), Car2 is normal (0), and Car3 is abnormal (1). For vehicles judged to be abnormal (such as Car1 and Car3), the backend server immediately issues an alarm to notify customer service or contact the owner of the nearest service station for pre-inspection, and marks the vehicle as requiring special monitoring. After receiving the server instruction, the on-board terminal can prompt "SIM card communication abnormality, recommended for prompt repair" on the vehicle interface. For vehicles judged to be normal (such as Car2), regular monitoring continues.

[0066] 7. Model evaluation index calculation After the model has been deployed for a period of time, its performance needs to be evaluated based on actual results. By comparing the model predictions with subsequent real-world maintenance data, the following metrics can be calculated (using sample data as an example): Hit Rate: The proportion of vehicles predicted by the model to be abnormal that actually have faults. Formula: A is the set of vehicles that are judged to be abnormal by the model, and B is the set of vehicles that actually have failures and need to be repaired. is the number of vehicles that the model hits (predicted abnormal and actually abnormal). In this example, A = Car1, Car3 (cars predicted abnormal by the model), B = Car1, Car2 (actual faulty vehicles). , hit one car. Hit rate .

[0067] Coverage Rate: The proportion of vehicles that actually need repair that are successfully predicted by the model. Formula: Coverage Rate = |A∩B| / |B| In the above example, of the two cars Car1 and Car2 that actually had faults, the model only caught Car1, resulting in a coverage rate of 1 / 2 (50%). The hit rate and coverage rate reflect the model's precision and recall, respectively: a 50% hit rate means half of the model's alerts are correct, while a 50% coverage rate means half of the actual faults are caught early.

[0068] Value Coefficient: This is a measure of the effectiveness of the model in capturing the economic value of faults. It takes into account the compensation amount (repair cost) for each fault and calculates the weighted proportion of compensation amounts for the faults captured by the model relative to the total compensation amounts for all faults. It can be defined as the ratio of the sum of compensation amounts for all faults captured by the model to the total compensation amounts for all faults: Cost j is the compensation amount incurred by vehicle j due to SIM card failure during the monitoring period.

[0069] In this example, Car1 is a hit and pays 2,000 yuan in compensation, while Car2 is a miss and pays 1,200 yuan in compensation. The total fault compensation = 2,000 + 1,200 = 3,200 yuan. The value assessment coefficient = 2,000 / 3,200 = 0.625 (62.5%). Compared to a coverage rate of 50%, the value coefficient of 62.5% is higher because the model captures faults with higher compensation costs (Car1 pays more than Car2), demonstrating the model's relative value in mitigating high-cost faults.

[0070] Comprehensive Score: The above three indicators are weighted and summarized with the correction coefficient to give the overall performance score of the model. The formula is: Comprehensive score = value coefficient × hit rate + (1 − value coefficient) × coverage rate + correction coefficient Substituting the data in this example: hit rate 0.5, coverage rate 0.5, and value coefficient 0.625, the basic comprehensive score = 0.625 × 0.5 + (1-0.625) × 0.5 = 0.3125 + 0.1875 = 0.50. Without the correction factor, the score is 50 (a full score of 1 corresponds to 100 on a 100-point scale). The correction factor is used to dynamically adjust the score, as detailed below.

[0071] In summary, these indicators comprehensively evaluate the effectiveness of the model: the hit rate / coverage rate focuses on the accuracy of quantity, the value coefficient emphasizes quality (economic benefits), and the comprehensive score serves as the basis for decision-making.

[0072] When calculating the comprehensive score, a correction factor is introduced to improve the stability and rationality of the score. This is because the compensation amounts for different failure cases can vary significantly, causing fluctuations in the value assessment coefficient, which in turn affects the reliability of the comprehensive score. To this end, an adaptive mechanism is designed to dynamically adjust the correction factor, as follows: The coefficient of variation of all fault compensation amounts is used to quantify the degree of fluctuation in compensation amounts. The formula is: Volatility = σ_{VC} / μ_{VC}, Where σ_{VC} is the standard deviation of the compensation amounts for all vehicles with SIM card failures within the preset monitoring period T, and μ_{VC} is the mean of the compensation amounts during the same period; In this example, the actual breakdown compensation amount set is {2000, 1200}, with a mean of 1600 yuan and a standard deviation of approximately 400. The volatility is then 400 / 1600 = 0.25 (25%). If the breakdown compensation amounts for different vehicles vary greatly within a certain period of time (e.g., there are both minor breakdowns of a few hundred yuan and major breakdowns of several thousand yuan), will be higher; otherwise it will be lower.

[0073] Correction coefficient calculation: A correction value is assigned to the comprehensive score based on the volatility to balance the impact of high-value faults on the score. The correction coefficient is defined as: Correction coefficient = α × volatility; Where α is an adjustment parameter (ranging from 0 to 1). The strategy for selecting α is to increase the adjustment when the fluctuation is large and decrease it when the fluctuation is small. For example, when the compensation amount fluctuates greatly, α can be set between 0.3 and 0.5; when the fluctuation is small, α can be set between 0.1 and 0.3.

[0074] During the first run, since there is no historical optimal α0, an initial value (such as the middle value 0.3) can be selected in the interval [0.1, 0.5] as α to calculate the first correction coefficient.

[0075] In each subsequent cycle, the parameter optimization method described below will be used: Reselect the “optimal α0 of this cycle” using the candidate interval (0.1–0.3 or 0.3–0.5) + RMSE minimization criterion; Then, use α0 as the current adjustment parameter and substitute it into the correction coefficient = α0 × Volatility formula.

[0076] That is, each time the correction coefficient is calculated, α is the optimal value obtained from the previous "parameter optimization" step. Only before the first optimization, a default reasonable intermediate value is used as a guide; after that, the α0 output by the automatic optimization is completely updated during each cycle. In this example, the current α0 = 0.3, and the correction coefficient is 0.075 The correction factor is added back to the overall score: Overall score = 0.50 + 0.075 = 0.575 (i.e., 57.5 points). This correction mechanism can smooth out the overall score from being too low or too high if a high proportion of failures with high compensation costs occur within an assessment cycle (which may result in a higher value coefficient but with a small sample size). This also ensures the stability of the model's scoring across different scenarios.

[0077] 8. Model optimization and incremental learning strategy Based on the comprehensive evaluation score, the system decides whether to trigger the model optimization process. This embodiment presets two thresholds: 0.8 (80 points) and 0.6 (60 points). The optimization strategy is as follows: When the comprehensive score P is greater than or equal to 0.8: the model performs well and no adjustment is required.

[0078] When 0.6≤P<0.8: the model is average, triggering two strategies: adding sample data and parameter optimization.

[0079] When P<0.6: the model is not effective and requires comprehensive optimization. The three strategies of increasing sample data, adding new labels, and optimizing parameters are all triggered.

[0080] Based on the above calculations, the overall score in this case is approximately 0.575, which is lower than 0.6, so all three optimization measures should be implemented: Add sample data (server side): Include cases that the model misjudged in the training set and supplement with new fault samples to improve the model's learning of edge cases. Specifically, this includes: Add missed samples: For vehicles that actually experienced a fault but were not predicted by the model (e.g., Car2), their data (including the fault label = 1) is added to the training set. This allows the model to identify previously unrecognized patterns with weaker signals, such as Car2, in the future. Assuming that the training set previously lacked abnormal samples with weak signals, such as those in the suburbs, adding Car2 enriches the distribution.

[0081] Correcting false positives: For vehicles that the model predicts are abnormal but are actually normal (such as Car3), analyze the reasons for the misclassification and, if appropriate, add them to the training set as counterexamples (labeled 0). For example, if Car3 is misclassified despite having high signal quality, this may be because the training data lacks normal examples of high signal strength in urban areas. Adding Car3 data allows the model to learn that this feature combination corresponds to normality.

[0082] After adding these samples, the size and diversity of the new training set are improved, and the SVM model can be retrained to obtain updated w and b parameters, thereby reducing the above-mentioned misjudgments.

[0083] New tag features (server side): New label information related to fault urgency is introduced to improve the model's sensitivity to high-risk faults. The specific method is to use information such as fault severity, user urgency, and estimated repair time in actual maintenance data to calculate an urgency score through a linear model. This score is then mapped to classification labels: minor, moderate, or severe (for example, coded as 0 / 1 / 2). The details are as follows: The urgency characteristic value is determined according to the following steps: For each vehicle sample, obtain: a) Fault level L∈{1=minor, 2=general, 3=serious}, b) User urgency U∈{1=non-urgent, 2=priority, 3=immediate}, c) Estimated maintenance time T, in hours; Calculate linear scores for light, moderate, and severe urgency levels: ; in, represents the weight coefficient; b i represents a constant term; The linear score z is used as the three types of Softmax input and mapped to P (light), P (medium), and P (heavy): The category corresponding to the maximum probability is taken and coded as “mild=0, medium=1, severe=2” to form the urgency characteristic value.

[0084] For example, for Car1 (fault level 3 = severe, urgency 3 = immediate, repair time 2 hours), the calculated probability of the severe category is the highest, and the urgency label is assigned 2 (severe); Car2 (level 2, general fault; urgency 2; repair time 1.5 hours) may correspond to the medium urgency label 1.

[0085] These urgency labels are added as additional information to model training samples. One approach is to input them as new features into the main model, allowing the model to understand the importance of certain samples (for example, samples with "heavy" labels can be given greater weight), thereby increasing the model's sensitivity to high-urgency / high-risk failure modes. This enhancement allows the model to focus more on potentially costly failure scenarios during training, improving the recall rate for these scenarios.

[0086] (3) Parameter optimization (server side): For the correction coefficient part of the comprehensive evaluation score, further adjust the parameters or other model hyperparameters to improve the stability of model performance.

[0087] First, recalculate the compensation volatility on the new training data to verify whether the previous α value is reasonable. A grid search combined with metrics such as the root mean square error (RMSE) can be used to select the optimal α. For example, try α = 0.1, 0.2, ..., 0.5 and substitute it into the correction coefficient formula. Observe the fluctuations of the comprehensive score over time, compare the score curve with the target stability level, and select the α that results in the most stable and sufficiently high score, as follows: If Volatility is greater than the set threshold, set candidate α∈[0.3, 0.5]; otherwise, set candidate α∈[0.1,0.3); RMSE evaluation: For each α_k in the candidate interval, calculate CorrectionFactor_k=α_k×Volatility, Based on this, the revised comprehensive evaluation score sequence P_k is obtained, and the root mean square error is calculated with the preset target score; Take the α_k with the smallest root mean square error as the adjustment parameter α0 of this cycle, And let CorrectionFactor = α0×Volatility.

[0088] At the same time, you can also adjust the hyperparameters of the SVM model itself (such as the penalty coefficient or kernel function parameters) to prevent overfitting or underfitting. In this example, due to the addition of new samples and features, C can be slightly increased to 1.5 to enhance the fit of the new complex samples, and the RBF kernel parameters can be fine-tuned to the new data distribution. After parameter optimization is completed, the comprehensive evaluation score is recalculated to ensure that it is improved to a reasonable range (for example, above 0.8).

[0089] After incremental learning and parameter tuning, the model completes a self-improvement phase. For our example, assume the updated model is tested again on the data for Car1, Car2, and Car3: Car1 still predicts an anomaly; Car2, having already included similar samples in its training, successfully predicts an anomaly this time; and the Car3 model recognizes that high-quality signals should be considered normal, eliminating false positives. This improves both the hit rate and coverage (both reaching 100% in this small sample), increases the value assessment coefficient, and significantly boosts the overall score. In actual deployment, after multiple rounds of this feedback optimization, the model will become even more accurate and reliable.

[0090] This embodiment realizes the intelligent monitoring and optimization of SIM card signal anomalies through the terminal-backend collaboration. The on-board terminal collects data in real time, and the back-end server uses the SVM model to predict faults, and combines the maintenance feedback to calculate indicators such as hit rate, coverage, and value coefficient to form a comprehensive score. A dynamic correction coefficient is used to balance the fluctuations in the scoring results, and when the score is low, it automatically triggers strategies such as adding samples, introducing new label features, and parameter tuning to perform incremental optimization of the model. The entire process includes clear data processing steps and formula calculations, and example numerical calculations are given at each key node to verify the feasibility and repeatability of the method. By continuously looping the above process, the system can promptly detect SIM card signal anomalies, reduce false alarm and missed alarm rates, improve fault coverage and economic value, save costs for customers and maintenance parties, and improve the reliability of the vehicle communication system.

[0091] Example 2 The present invention provides an optimization system based on vehicle networking SIM card signal monitoring, which is used in the method described in the above technical solution, including: Data acquisition module, used to collect signal data from the Internet of Vehicles SIM card; A model training module is used to build a support vector machine model based on the collected data to distinguish between the normal state and the abnormal state of the SIM card signal; A state recognition module, used to use the support vector machine model to perform real-time abnormality recognition on the vehicle network SIM card signal; Model evaluation module, used to calculate model evaluation indicators based on recognition results and obtain comprehensive evaluation scores; A model optimization module is used to optimize and adjust the support vector machine model based on the comprehensive evaluation score to achieve closed-loop optimization.

[0092] Example 3 The present invention also provides a non-transitory computer-readable storage medium storing a computer-executable program, wherein when the program stored thereon is executed by a processor, the processor performs the functions / steps described in the above technical solution.

[0093] Example 4 The present invention also provides a data processing device configured in a server, comprising: a processor, a memory and a program thereon, wherein the program enables the device to execute the functions of each module of the system described in the above technical solution.

[0094] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0095] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0096] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0097] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0098] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

[0099] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. An optimization method based on vehicle networking SIM card signal monitoring, characterized in that: The following steps are involved: Collect signal data from the Internet of Vehicles SIM card; Building a support vector machine model based on the collected data to distinguish between a normal state and an abnormal state of the SIM card signal; Using the support vector machine model to perform real-time anomaly recognition on the vehicle networking SIM card signal; Calculate the model evaluation index based on the recognition results and obtain the comprehensive evaluation score; The support vector machine model is optimized and adjusted based on the comprehensive evaluation score to achieve closed-loop optimization.

2. The method according to claim 1, characterized in that In the step of collecting vehicle network SIM card signal data, the collected data is preprocessed, including one-hot encoding of vehicle operating area information, normalizing the data transmission rate, and calculating the product of the signal strength and communication quality as the input feature of the input support vector machine model.

3. The method according to claim 1, characterized in that The calculation process of the comprehensive assessment score includes: The identified SIM card signal anomaly is compared with the vehicle maintenance information to calculate the hit rate and coverage of the support vector machine model prediction, and a comprehensive evaluation score is calculated based on the hit rate, coverage and the value assessment coefficient related to the vehicle failure compensation amount, which is used to comprehensively evaluate the model prediction performance.

4. The method according to claim 3, characterized in that In the closed-loop optimization step, a correction coefficient is set for the comprehensive evaluation score, and the correction coefficient is dynamically adjusted based on fluctuations in the vehicle failure compensation amount to correct the comprehensive evaluation score.

5. The method according to claim 4, characterized in that In the closed-loop optimization step, the vehicle data misjudged as abnormal by the support vector machine model and the vehicle data with SIM card failure not recognized by the model are added to the training samples of the support vector machine model, and the support vector machine model is retrained.

6. The method according to claim 5, characterized in that The closed-loop optimization step further includes: calculating and generating an urgency feature value representing the fault priority based on the vehicle fault level, the user urgency, and the estimated repair time information, adding the urgency feature value as an additional input feature to the training samples of the support vector machine model, and retraining the support vector machine model; When the comprehensive evaluation score is greater than or equal to a first threshold, keeping the model parameters unchanged; When the comprehensive evaluation score is less than a first threshold, adding vehicle data misjudged and unrecognized by the support vector machine model to the training samples of the support vector machine model to retrain the model; When the comprehensive evaluation score is less than a second threshold, dynamically adjusting the correction coefficient, and adding the urgency feature value as an additional input feature to the training sample of the support vector machine model to retrain the model; The second threshold is smaller than the first threshold.

7. The method according to claim 4, characterized in that During the preset monitoring period T, A={vehicle ID_j|the support vector machine model determines that the SIM card is abnormal within period T}; B={Vehicle ID_j|Confirmed by the repair station as SIM card failure and compensation generated within period T}; C=A∩B; HitRate=|C| / |A|; CoverageRate=|C| / |B|; ValueCoefficient=(Σ j∈C Cost_j) / (Σ j∈B Cost_j), P=ValueCoefficient×HitRate+(1−ValueCoefficient)×CoverageRate+CorrectionFactor; Where Cost_j is the compensation amount incurred by vehicle j due to SIM card failure within period T; HitRate represents the hit rate, CoverageRate represents the coverage rate, P represents the comprehensive evaluation score; CorrectionFactor represents the correction factor.

8. The method according to claim 7, characterized in that The correction coefficient is dynamically determined as follows: The volatility indicator Volatility is calculated using the following formula: Volatility = σ_{VC} / μ_{VC}, Where σ_{VC} is the standard deviation of the compensation amounts for all vehicles with SIM card failures within the preset monitoring period T, and μ_{VC} is the mean of the compensation amounts during the same period; Candidate interval setting: If Volatility is greater than the set threshold, set candidate α∈[0.3,0.5]; otherwise, set candidate α∈[0.1,0.3); RMSE evaluation: For each α_k in the candidate interval, calculate CorrectionFactor_k=α_k×Volatility, Based on this, the revised comprehensive evaluation score sequence P_k is obtained, and the root mean square error is calculated with the preset target score; Take the α_k with the smallest root mean square error as the adjustment parameter α0 of this cycle, And let CorrectionFactor = α0×Volatility.

9. The method according to claim 6, wherein: The urgency characteristic value is determined according to the following steps: (1) For each vehicle sample, obtain a) Fault level L∈{1=minor, 2=general, 3=serious}, b) User urgency U∈{1=non-urgent, 2=priority, 3=immediate}, c) Estimated maintenance time T, in hours; (2) Calculate linear scores for the three emergency levels: mild, moderate, and severe: ； in, represents the weight coefficient; b i represents a constant term; The linear score z is used as the three types of Softmax input and mapped to P (light), P (medium), and P (heavy): The category corresponding to the maximum probability is taken and coded as "mild = 0, medium = 1, severe = 2" to form the urgency characteristic value.

10. An optimization system based on vehicle networking SIM card signal monitoring, characterized in that For implementing the method according to any one of claims 1 to 9, comprising: Data acquisition module, used to collect signal data from the Internet of Vehicles SIM card; A model training module is used to build a support vector machine model based on the collected data to distinguish between the normal state and the abnormal state of the SIM card signal; A state recognition module, used to use the support vector machine model to perform real-time abnormality recognition on the vehicle network SIM card signal; Model evaluation module, used to calculate model evaluation indicators based on recognition results and obtain comprehensive evaluation scores; A model optimization module is used to optimize and adjust the support vector machine model based on the comprehensive evaluation score to achieve closed-loop optimization.

Images