Tire pressure anomaly prediction method and system based on machine learning

Abstract

The invention discloses a tire pressure anomaly prediction method and system based on machine learning, and belongs to the field of tire pressure anomaly prediction.The tire pressure anomaly prediction method comprises the steps that according to extracted tire pressure features, a support vector machine or a random forest algorithm is adopted, a tire pressure anomaly prediction model is trained, and tire pressure data collected in real time are input; predicting whether tire pressure abnormity occurs in the future 30 minutes, and predicting the type and severity of the abnormity; the tire pressure abnormity early warning information is combined with a vehicle control system, when the tire pressure abnormity risk is predicted, a warning prompt is sent to a driver through a man-machine interaction interface, and meanwhile, a corresponding vehicle control strategy is formulated according to the severity degree of tire pressure abnormity and the predicted evolution trend.

Description

Technical Field

[0001] The present invention belongs to the field of abnormal tire pressure prediction, and in particular relates to a method and system for predicting abnormal tire pressure based on machine learning. Background Art

[0002] During the driving process, the tire pressure will change due to factors such as road conditions and weather. If the tire pressure is abnormal, it will not only affect driving safety, but also accelerate tire wear and shorten tire service life. Therefore, real-time monitoring of tire pressure changes and timely warning of abnormal tire pressure are of great significance to ensure driving safety and reduce vehicle use costs.

[0003] However, due to the complex and changeable environment during vehicle driving, tire pressure data is easily interfered by various factors, resulting in a large number of invalid values, outliers, and impossible data combinations in the data, which brings difficulties to subsequent data analysis and modeling. In addition, when the vehicle is parked and dormant, tire pressure data collection will be interrupted, resulting in large-scale missing data and unable to truly reflect the tire pressure changes during this time period. When establishing a tire pressure abnormality prediction model, it is also necessary to consider the impact of different leakage conditions on the prediction results. The data features corresponding to slow leakage and fast leakage are quite different, and feature selection and model training need to be performed separately to improve prediction accuracy. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a tire pressure abnormality prediction method based on machine learning, comprising:

[0005] Acquire tire pressure data collected during vehicle driving, pre-process the tire pressure data, and obtain a tire pressure data set;

[0006] Completing the tire pressure data set based on a polynomial regression method and ambient temperature change data to obtain a tire pressure time series data set;

[0007] Performing time series analysis on the tire pressure time series data set to obtain a change pattern of the tire pressure data, identifying the change pattern of the tire pressure data based on a combined anomaly detection algorithm, and obtaining abnormal fluctuations of the tire pressure data;

[0008] Preliminarily classifying the abnormal fluctuation type based on the abnormal fluctuation of the tire pressure data to obtain a feature subset;

[0009] Constructing a random forest algorithm, inputting the feature subset into the random forest algorithm for training, and obtaining a tire pressure abnormality prediction model;

[0010] The real-time tire pressure abnormality is predicted based on the tire pressure abnormality prediction model.

[0011] Preferably, the process of obtaining the tire pressure data set includes:

[0012] Obtaining the original tire pressure data collected during vehicle driving and preprocessing the tire pressure data;

[0013] Eliminating invalid values ​​in the tire pressure data according to a preset invalid value threshold range to obtain first processed tire pressure data;

[0014] Correcting abnormal values ​​in the first processed tire pressure data to obtain second processed tire pressure data;

[0015] Eliminating abnormal data in the tire pressure data after the second processing to obtain tire pressure data after the third processing;

[0016] Calculate the statistical features of the tire pressure data after the third processing, perform cluster analysis on the statistical features based on a density clustering algorithm, divide the tire pressure data into different clusters, regard data that does not belong to any normal cluster as noise data and remove it, and obtain a cleaned tire pressure data set.

[0017] Preferably, the process of obtaining the tire pressure time series data set includes:

[0018] Establishing a polynomial regression model based on the tire pressure data before and after the vehicle is parked and the ambient temperature change data in the corresponding time period in the tire pressure data set, estimating the tire pressure change trend during the parking hibernation period based on the polynomial regression model, and generating estimated tire pressure data;

[0019] splicing the estimated tire pressure data and the tire pressure data set to obtain complete sequence data;

[0020] The complete sequence data is smoothed to obtain the tire pressure time series data set.

[0021] Preferably, the process of generating estimated tire pressure data comprises:

[0022] Obtaining the time period information and available tire pressure monitoring data during the vehicle's parking and dormancy period;

[0023] Determine the time point when the tire pressure data is missing based on the acquired time period information and tire pressure monitoring data;

[0024] A polynomial regression algorithm was used to establish a regression model with time as the independent variable and available tire pressure monitoring data as the dependent variable;

[0025] The time points with missing tire pressure data were input into the established polynomial regression model to obtain the estimated tire pressure values ​​at the corresponding time points.

[0026] Preferably, the process of obtaining abnormal fluctuations in tire pressure data includes:

[0027] Performing sliding window segmentation on the tire pressure time series data set to obtain data subsets of multiple time segments;

[0028] constructing a feature vector based on statistical features of the data subsets of the multiple time segments;

[0029] The feature vector is calculated based on the isolation forest algorithm, and multiple isolation trees are constructed by randomly selecting features recursively and dividing the data space according to the feature values ​​to obtain the abnormality degree score of the abnormal point;

[0030] The changing pattern of the tire pressure data is identified based on the abnormality degree score of the abnormal point, and the abnormal fluctuation of the tire pressure data is obtained.

[0031] Preferably, the process of obtaining the feature subset includes:

[0032] constructing a decision tree model, and calculating the abnormal fluctuation of the tire pressure data based on the decision tree model to obtain a classification result;

[0033] Feature engineering is constructed based on the classification results, and feature subsets that can effectively distinguish categories are extracted by analyzing the change rate and change amplitude characteristics of the tire pressure data.

[0034] Preferably, the process of obtaining the abnormal tire pressure prediction model includes:

[0035] Constructing a random forest model, and training the random forest model based on the feature subset to obtain a prediction model;

[0036] The prediction model is cross-validated, and the hyperparameters of the model are optimized based on the validation results to obtain the abnormal tire pressure prediction model; wherein the abnormal tire pressure prediction model takes the tire pressure feature vector as input and whether the tire pressure abnormality occurs as output.

[0037] On the other hand, the present invention also provides a tire pressure abnormality prediction system based on machine learning, comprising:

[0038] A data acquisition module, used to acquire tire pressure data collected during vehicle driving, pre-process the tire pressure data, and obtain a tire pressure data set;

[0039] A data processing module, used to complete the tire pressure data set based on a polynomial regression method and ambient temperature change data to obtain a tire pressure time series data set;

[0040] A fluctuation acquisition module, used to perform time series analysis on the tire pressure time series data set to obtain the variation pattern of the tire pressure data, identify the variation pattern of the tire pressure data based on a combined anomaly detection algorithm, and obtain abnormal fluctuations of the tire pressure data;

[0041] A feature acquisition module, configured to preliminarily classify the abnormal fluctuation type based on the abnormal fluctuation of the tire pressure data to obtain a feature subset;

[0042] A training module, used for constructing a random forest algorithm, inputting the feature subset into the random forest algorithm for training, and obtaining a tire pressure abnormality prediction model;

[0043] The prediction module is used to predict the real-time tire pressure abnormality based on the tire pressure abnormality prediction model.

[0044] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method described in the computer program.

[0045] On the other hand, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the method is implemented when the computer program is executed by a processor.

[0046] Compared with the prior art, the present invention has the following advantages and technical effects:

[0047] The present invention discloses a method for predicting abnormal tire pressure based on machine learning. The method first cleans and interpolates tire pressure data collected during vehicle driving to obtain complete tire pressure time series data. Then, time series analysis is performed based on the data to identify abnormal tire pressure fluctuations. For possible slow air leakage and fast air leakage, feature engineering is constructed and feature subsets that effectively distinguish the two situations are extracted. Using these features to train a tire pressure abnormality prediction model, it is possible to predict whether tire pressure abnormality will occur within the next 30 minutes and its type and severity. Finally, the warning information is combined with the vehicle control system to issue a warning to the driver and formulate a corresponding control strategy. The present invention effectively solves the problems of noise and missing in tire pressure data through technologies such as data cleaning, interpolation and feature engineering, and improves the accuracy of tire pressure abnormality prediction. At the same time, combined with the vehicle control system, timely warning and processing of tire pressure abnormality are achieved, thereby improving driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0049] Figure 1 is a flowchart of a method for predicting abnormal tire pressure based on machine learning according to an embodiment of the present invention;

[0050] Figure 2 Schematic diagram of a tire pressure abnormality prediction method based on machine learning according to an embodiment of the present invention.

[0051] Figure 3 This is another schematic diagram of the abnormal tire pressure prediction method and system based on machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0053] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0054] Embodiment 1

[0055] like Figure 1-3 As shown, this embodiment provides a tire pressure abnormality prediction method based on machine learning, including:

[0056] S101. Acquire tire pressure data collected during vehicle driving, and use data cleaning technology to eliminate or correct noise data such as invalid values, abnormal values, and impossible data combinations in the tire pressure data by setting a predefined threshold range to obtain a cleaned tire pressure data set.

[0057] The original tire pressure data collected during the driving process of the vehicle is obtained, and the tire pressure data is preprocessed. According to the preset invalid value threshold range, it is judged whether there are invalid values ​​in the tire pressure data. If there are invalid values, they are eliminated to obtain the first processed tire pressure data. According to the preset abnormal value threshold range, it is judged whether there are abnormal values ​​in the tire pressure data after the first processing. If there are abnormal values, they are corrected to values ​​within the normal range to obtain the second processed tire pressure data. According to the preset data combination rule, it is judged whether there are impossible data combinations in the tire pressure data after the second processing. If there are impossible data combinations, they are eliminated to obtain the third processed tire pressure data. The tire pressure data after the third processing is statistically analyzed, and the statistical characteristics such as the mean and variance of the tire pressure data are calculated to obtain the statistical distribution of the tire pressure data. According to the statistical distribution of the tire pressure data, the density clustering algorithm is used to perform cluster analysis on the tire pressure data, and the tire pressure data is divided into different clusters, each cluster representing a normal tire pressure state. According to the results of the cluster analysis, the final tire pressure data set is determined, and the data that does not belong to any normal cluster is regarded as noise data and eliminated to obtain the cleaned tire pressure data set.

[0058] Obtaining the original tire pressure data during vehicle driving is an important means of monitoring tire status. First, the original data needs to be preprocessed to remove invalid values ​​and abnormal values. For example, the effective range of tire pressure is set to 100-400kPa, and data below 100kPa or above 400kPa is considered invalid and removed. For the remaining data, 180-280kPa can be set as the normal range, and data outside this range can be corrected to the nearest boundary value. This can remove obvious erroneous data caused by sensor failure and other reasons. Next, judge the rationality of the data combination. For example, the tire pressure difference of the four tires is usually not too large. If the tire pressure of the three tires is around 220kPa at a certain moment, and the tire pressure of the other tire is only 150kPa, this combination is likely to be unreasonable and should be removed. This step can remove some abnormal data that is not easily detected by a single threshold. Perform statistical analysis on the processed data and calculate features such as the mean and variance. Assume that the tire pressure mean is 230kPa and the standard deviation is 10kPa, which indicates that most of the data is concentrated in the range of 220-240kPa. These statistical features can reflect the overall state of the tire and the degree of discreteness of the data. Based on the statistical distribution, a density clustering algorithm such as DBSCAN is used to cluster the data. The advantage of the DBSCAN algorithm is that it can discover clusters of any shape and is insensitive to noisy data. For example, three main clusters may be obtained: 220-225kPa, 230-235kPa, and 240-245kPa, representing normal tire pressures under light load, normal load, and heavy load conditions, respectively. Data that does not belong to these clusters may be abnormal or noise and should be eliminated. Through this series of processing, the final cleaned data set can more accurately reflect the changes in tire pressure under normal driving conditions of the vehicle.

[0059] S102. To address the problem of missing tire pressure data during the period when the vehicle is parked and dormant, data interpolation technology is used. Based on the tire pressure data before and after the vehicle is parked in the cleaned tire pressure data set, combined with the ambient temperature change data, a polynomial regression method is used to estimate the tire pressure change trend during the parking period, complete the missing tire pressure data, and obtain complete tire pressure time series data.

[0060] The tire pressure data and ambient temperature change data before and after the vehicle is parked are obtained. The data interpolation technology is used to deal with the problem of missing tire pressure data during the parking hibernation period. The obtained tire pressure data is preprocessed to remove outliers and noise data to obtain the cleaned tire pressure data set. According to the tire pressure data before and after the vehicle is parked in the cleaned tire pressure data set, as well as the ambient temperature change data in the corresponding time period, a polynomial regression model is established to estimate the tire pressure change trend during the parking hibernation period. Through the established polynomial regression model, the missing tire pressure data during the parking hibernation period of the vehicle is interpolated and estimated to obtain the estimated tire pressure data. The estimated tire pressure data during the parking hibernation period is spliced ​​with the tire pressure data before and after the vehicle is parked to obtain the complete tire pressure time series data. The obtained complete tire pressure time series data is smoothed to eliminate the mutations and discontinuities that may be caused by the interpolation estimation, and to improve the continuity and smoothness of the data.

[0061] Obtaining tire pressure data and ambient temperature change data before and after the vehicle is parked is a key step in monitoring tire status. For example, the tire pressure recorded before parking is 32psi and the ambient temperature is 25℃; after parking for 12 hours, the tire pressure is 31psi and the ambient temperature is 20℃. These data provide a basis for analyzing tire pressure changes during parking. In order to solve the problem of missing tire pressure data during parking dormancy, data interpolation technology is used to deal with it. Common interpolation methods include linear interpolation, polynomial interpolation, and spline interpolation. Taking linear interpolation as an example, it can be assumed that the tire pressure changes linearly over time, and the tire pressure data in the middle period can be estimated based on the tire pressure values ​​before and after parking. Preprocessing the acquired tire pressure data and removing outliers and noise data are important steps to ensure data quality. Outliers may be caused by sensor failure or external interference. For example, a tire pressure reading of 90psi that suddenly appears is obviously unreasonable and should be removed. Noise data may appear as a sharp fluctuation in a short period of time, such as the tire pressure jumping from 32psi to 25psi and then back to 32psi within 1 minute. This situation should also be regarded as noise and smoothed. Establishing a polynomial regression model to estimate the tire pressure change trend during parking and dormancy can more accurately reflect the relationship between tire pressure and temperature. For example, a quadratic polynomial model can be used: P = a + bT + cT2, where P is the tire pressure, T is the temperature, and a, b, and c are undetermined coefficients. The tire pressure and temperature data before and after parking are fitted by the least squares method to obtain the model parameters. The established polynomial regression model is used to interpolate and estimate the missing tire pressure data during the parking and dormant period of the vehicle. Assuming that the relationship obtained by the model is P = 30 + 0.1T-0.002T2, the corresponding tire pressure value can be estimated for the temperature at any time during the dormancy period. The estimated tire pressure data during parking and dormancy are spliced ​​with the measured tire pressure data before and after the vehicle is parked to form a complete tire pressure time series data. This step fills the data gap, allowing tire pressure monitoring to cover all-weather scenarios. The obtained complete tire pressure time series data can be smoothed using the moving average method or exponential smoothing method. For example, using a 5-point moving average, that is, taking the average of the two points before and after each data point and itself, can effectively eliminate short-term fluctuations and highlight long-term trends. The processed complete tire pressure time series data is stored in the database to provide a basis for subsequent analysis.

[0062] For example, a vehicle was in a parked dormant state from 20:00 on May 1, 2023 to 8:00 on May 2. The system recorded the tire pressure data before and after the dormancy, but the data was missing during the dormancy period. The time point at which the tire pressure data is missing can be determined by comparing the actual recorded data timestamp with the complete time series. The polynomial regression algorithm is a commonly used data fitting method that is suitable for describing nonlinear relationships. In tire pressure estimation, a quadratic or cubic polynomial model can be selected, with time as the independent variable and tire pressure as the dependent variable. For example, assuming that the tire pressure data at three time points of 20:00, 20:30 and 8:00 are known, a regression model of the following form can be established: P = a*t^2+b*t+c, where P is the tire pressure, t is the time, and a, b, and c are the coefficients to be determined. The time point at which the tire pressure data is missing is input into the established polynomial regression model to obtain the estimated tire pressure value at the corresponding time point. For example, to estimate the tire pressure at 2:00 a.m., just substitute t=6 (assuming 20:00 as the starting point) into the model to get the estimated result. Judging the rationality of the tire pressure estimate is an important step to ensure data reliability. Usually, the normal tire pressure range is between 180-280 kPa. If the estimated value falls within this range, it can be used directly. If the estimated value is out of range, such as an abnormally low value of 150 kPa, other methods need to be used to re-estimate. Linear interpolation is a simple and effective alternative. Assuming that there is a reliable tire pressure data point before and after the abnormal estimation point, these two points can be used for linear interpolation. For example, if the tire pressure at 23:00 is 220 kPa and the tire pressure at 5:00 is 210 kPa, then the tire pressure at 2:00 a.m. can be estimated to be 215 kPa. By integrating the estimated tire pressure values ​​at the missing time points with the original tire pressure monitoring data, a complete vehicle tire pressure monitoring data sequence can be obtained.

[0063] S103. Based on the complete tire pressure time series data, time series analysis technology is used to calculate the statistical characteristics of the tire pressure data, including the mean, variance, and slope, to characterize the change pattern of the tire pressure data, and a combined anomaly detection algorithm is used to identify abnormal fluctuations in the tire pressure data and determine whether there is an abnormal tire pressure situation.

[0064] Obtain the tire pressure time series data of vehicle tires and construct a tire pressure time series data set, which contains tire pressure values ​​at multiple time points. Preprocess the tire pressure time series data to remove missing values ​​and outliers to ensure the integrity and accuracy of the data. Use time series analysis technology to calculate the statistical characteristics of the tire pressure time series data, including mean, variance, and slope, to characterize the change pattern of tire pressure data in the time dimension. Based on the calculated statistical characteristics, construct a tire pressure anomaly detection model, which comprehensively considers the characteristics of the tire pressure data such as mean, variance, and slope to determine whether there are abnormal fluctuations in the tire pressure data. Use a combined anomaly detection algorithm, such as the isolation forest algorithm or the local anomaly factor algorithm, to perform anomaly detection on the tire pressure time series data and identify abnormal points or abnormal fragments in the data.

[0065] Specifically, the tire pressure monitoring system realizes real-time monitoring of tire status by collecting tire pressure time series data of vehicle tires. First, the system obtains tire pressure values ​​over a period of time, such as recording tire pressure every 5 minutes, and constructs a data set containing timestamps and corresponding tire pressure values. For example, a vehicle collects 96 sets of tire pressure data in 8 hours. The data preprocessing stage is crucial to ensure the accuracy of subsequent analysis. The system removes obviously abnormal values, such as zero values ​​or values ​​outside the normal range due to sensor failure. At the same time, for a small number of missing data points, interpolation methods can be used to fill them. After preprocessing, a cleaned tire pressure data set is obtained. Next, the system calculates the statistical characteristics of the tire pressure data. The mean reflects the overall level of tire pressure, the variance indicates the degree of tire pressure fluctuation, and the slope reflects the trend of tire pressure change. For example, when a vehicle is in normal driving condition, the mean tire pressure is 220kPa and the variance is 5kPa 2 , the slope is close to 0, indicating that the tire pressure is relatively stable. Based on these statistical features, the system constructs a tire pressure anomaly detection model. The model takes into account multiple factors, such as whether the tire pressure deviates from the normal range, whether the tire pressure changes too much in a short period of time, etc. For example, if the tire pressure suddenly drops by more than 20%, or continues to drop by more than 10kPa within 1 hour, the system will determine it as an abnormal situation. In order to improve the accuracy of anomaly detection, the system will also use more complex algorithms. The isolation forest algorithm identifies anomalies by building a decision tree, while the local anomaly factor algorithm detects anomalies by comparing the density of data points with their neighboring points. These algorithms are able to discover complex anomaly patterns that are difficult to identify with a single feature. When the system detects an abnormal tire pressure, an alarm is triggered immediately. The alarm can be an indicator light on the dashboard or a sound reminder through the vehicle system.

[0066] The time series data collected by the tire pressure sensor of the vehicle is obtained, and the data is preprocessed to remove missing values ​​and outliers to obtain a normalized time series data set. According to the characteristics of the tire pressure time series data, an appropriate time window size is selected, and the data is segmented by sliding windows to obtain data subsets of multiple time segments. For each data subset of the time segment, relevant statistical features such as mean, variance, kurtosis, etc. are extracted to construct a feature vector as the input of the anomaly detection model. The feature vector is trained using the isolation forest algorithm. By recursively randomly selecting features and dividing the data space according to the feature values, multiple isolated trees are constructed to obtain the abnormal degree score of the abnormal point. The local anomaly factor algorithm is used to calculate the local density difference between each data point and its neighboring points. By setting a threshold, it is determined whether the data point is a local anomaly point and the local anomaly degree of the anomaly point is obtained. The anomaly scoring results of the isolation forest algorithm and the local anomaly factor algorithm are combined, and the comprehensive anomaly score of each time segment is obtained through weighted average or voting mechanism, and the abnormal threshold is set to determine the abnormal segment. The detected abnormal points and abnormal fragments are combined with the vehicle driving conditions to analyze the causes of the abnormalities, such as sudden changes in tire pressure, slow air leakage, etc., to provide a decision-making basis for intelligent monitoring and early warning of vehicle tire pressure.

[0067] Acquiring time series data collected by the vehicle tire pressure sensor is the first step of the tire pressure monitoring system. Assuming that a vehicle records the tire pressure data of four tires every 5 minutes during driving for 24 hours, a tire pressure time series data containing 288 time points is formed. These data may contain missing values ​​and outliers caused by sensor failure or transmission errors. In the preprocessing stage, missing values ​​are filled by interpolation, and outliers are removed using the 3σ principle (that is, data points with a difference of more than 3 times the standard deviation from the mean are considered outliers) to ensure the integrity and accuracy of the data. Selecting an appropriate time window size to perform sliding window segmentation on the data is to capture the local characteristics of tire pressure changes. Assuming that 10 minutes is selected as a time window, the 24-hour data is segmented into 144 time segments. Each time segment contains tire pressure data at 2 time points, forming a data subset. This segmentation method can balance the local details and overall trends of the data, which is convenient for subsequent feature extraction. For the data subset of each time segment, statistical features such as mean, variance and kurtosis are extracted. For example, the tire pressure data of a certain time segment is 2.5 bar and 2.6 bar, and the calculated mean is 2.55 bar, the variance is 0.01, and the kurtosis is 0 (assuming that the data distribution is relatively symmetrical). These features constitute a feature vector, which is used as the input of the anomaly detection model. The feature vector comprehensively reflects the stability and volatility of the tire pressure data, providing a basis for anomaly detection. The isolation forest algorithm is used to train the feature vector, and multiple isolated trees are constructed by recursively randomly selecting features and dividing the data space. Assume that 100 isolated trees are trained, and a feature vector is isolated to a leaf node in 50 trees, and its anomaly score is 0.5. The isolation forest algorithm measures the anomaly by the degree of isolation of the data. The higher the score, the greater the possibility of anomaly. The local anomaly factor algorithm is used to calculate the local density difference of each data point. Assuming that the average distance of the k nearest neighbors of a data point is 0.1, while the average distance of other data points is 0.05, the local anomaly factor of the data point is high and may be regarded as an anomaly point. By setting a threshold (such as 1.5), it is determined whether the data point is a local anomaly point. The local anomaly factor algorithm focuses on local density differences and can capture local anomalies that may be ignored by the isolated forest. The anomaly scoring results of the isolated forest and the local anomaly factor are combined, and the comprehensive anomaly score of each time segment is obtained through a weighted average or voting mechanism. Assuming that the isolated forest score weight is 0.6, the local anomaly factor score weight is 0.4, and the comprehensive anomaly score of a certain time segment is 0.7, which exceeds the preset threshold of 0.6, it is determined to be an abnormal segment. This comprehensive method can improve the accuracy and robustness of anomaly detection. The detected abnormal points and abnormal segments are combined with the vehicle driving conditions to analyze the cause of the abnormality. For example, if the tire pressure drops rapidly within a certain period of time, combined with the driving conditions and vehicle speed, it may be judged as a slow leak caused by the tire being punctured by a sharp object.Another reason why tire pressure suddenly increases over a period of time may be that high temperature causes the gas in the tire to expand. This analysis provides a decision-making basis for intelligent monitoring and early warning of vehicle tire pressure, and promptly reminds drivers to take measures to avoid potential safety risks. Continuously monitoring tire pressure time series data and updating the anomaly detection model in real time are to adapt to the dynamic changes in tire pressure data. As the vehicle is used for a longer time, factors such as tire wear and air pressure changes will affect the distribution characteristics of tire pressure data. Real-time updating of the model can ensure the accuracy and real-time nature of anomaly detection and enhance the practical value of the tire pressure monitoring system.

[0068] S104: Based on the identified abnormal tire pressure fluctuations, the abnormal fluctuation types are preliminarily classified to distinguish possible slow leakage and rapid leakage. For these two different leakage situations, feature engineering is constructed respectively, and by analyzing the change rate and change amplitude characteristics of the tire pressure data, feature subsets that can effectively distinguish slow leakage and rapid leakage are extracted for subsequent tire pressure abnormality type judgment.

[0069] The tire pressure data collected by the tire pressure sensor of the vehicle is obtained, and the tire pressure data is preprocessed to remove abnormal values ​​and noise interference to obtain the filtered tire pressure data. According to the filtered tire pressure data, the change rate and change amplitude of the tire pressure in unit time are calculated to obtain the tire pressure change rate characteristics and change amplitude characteristics. For the slow leakage situation, it is determined whether the tire pressure change rate is less than the preset slow leakage rate threshold, and the duration exceeds the preset slow leakage duration threshold. If so, it is determined to be a slow leakage type. For the fast leakage situation, it is determined whether the tire pressure change amplitude is greater than the preset fast leakage amplitude threshold, and the change rate is greater than the preset fast leakage rate threshold. If so, it is determined to be a fast leakage type. According to the characteristic differences between slow leakage and fast leakage, a decision tree model is constructed, and the tire pressure change rate and change amplitude are selected as the input features of the decision tree. The decision tree model is trained to obtain a tire pressure abnormal type judgment model. When an abnormal fluctuation of tire pressure is identified, the change rate and change amplitude characteristics of the abnormal fluctuation are extracted and input into the tire pressure abnormal type judgment model to obtain an abnormal type judgment result.

[0070] Calculate the tire pressure change characteristics. Assuming that the tire pressure of a vehicle drops from 36psi to 34psi within one hour, the tire pressure change rate is -2psi / h and the change amplitude is 2psi. These characteristics are the key basis for judging the type of tire pressure abnormality. For slow air leakage, a threshold can be set for judgment. For example, if the tire pressure drop rate is less than 0.5psi / h and lasts for more than 2 hours, there may be a slow air leakage. This slow but continuous pressure drop is usually caused by tiny cracks at the connection between the tire and the wheel hub or loose valve stems, which need to be checked and repaired in time. Rapid air leakage is manifested as a sharp drop in tire pressure. If the tire pressure drops by more than 5psi within 10 minutes and the drop rate is greater than 30psi / h, it may be a rapid air leakage. This situation may be caused by a sharp object puncturing the tire or a serious collision, and it is necessary to stop and check immediately to prevent serious accidents. Based on these characteristics, a decision tree model can be constructed to judge the type of tire pressure abnormality. The root node of the decision tree can be the tire pressure change rate, and the second-level nodes are the change duration and change amplitude. By optimizing the decision tree with training data, a classification model with high accuracy can be obtained. In actual applications, when abnormal tire pressure is detected, the system extracts relevant features and inputs them into the decision tree model. Assuming that the rate of abnormal tire pressure drop is -1psi / h and lasts for 3 hours, the model may judge it as a slow leak. At this time, the system will trigger a slow leak warning to remind the driver to check the tires at the next refueling. In contrast, if the tire pressure is detected to drop from 35psi to 25psi within 5 minutes, the model is likely to judge it as a rapid leak. In this case, the system will immediately issue an emergency alarm, requiring the driver to pull over as soon as possible to ensure safety. This decision tree-based method for judging the type of tire pressure abnormality has high accuracy and interpretability.

[0071] S105. Based on the extracted tire pressure features, a support vector machine or a random forest algorithm is used to train a tire pressure abnormality prediction model, and by inputting the real-time collected tire pressure data, it is predicted whether tire pressure abnormality will occur within the next 30 minutes, as well as the type and severity of the abnormality.

[0072] The vehicle tire pressure data is collected through sensors to obtain multi-dimensional time series data such as tire pressure and temperature, and data cleaning and preprocessing are performed to remove outliers and noise data to obtain a high-quality tire pressure feature data set. According to the tire pressure feature data set, key features reflecting the tire pressure change trend and abnormal pattern are extracted, such as statistical features such as tire pressure mean, variance, and kurtosis, as well as time series features such as tire pressure change rate and abnormal duration, to construct a feature vector for tire pressure abnormality prediction. Machine learning algorithms such as support vector machines and random forests are used to train tire pressure abnormality prediction models with tire pressure feature vectors as input and whether tire pressure abnormality occurs as output. Model hyperparameters are optimized through cross-validation and other methods to improve the prediction accuracy and generalization ability of the model. If it is necessary to predict the type and severity of tire pressure abnormality, the abnormality type and severity are used as the output labels of the model, and multi-classification algorithms such as logistic regression and decision trees are used for training to obtain a tire pressure abnormality prediction model that can predict the type and severity of abnormality. Deploy the trained tire pressure abnormality prediction model to the vehicle system or cloud server, receive the tire pressure data collected by the vehicle sensor in real time, extract and preprocess the data, input it into the prediction model for reasoning, and determine whether tire pressure abnormality will occur within a certain period of time in the future. If the prediction results show that tire pressure abnormality will occur in the future, further determine the type and severity of the abnormality, determine whether an early warning is required based on the preset threshold, and push the early warning information to the driver through the vehicle system or mobile application in a timely manner to remind the driver to take corresponding measures. Continuously monitor the vehicle's tire pressure status, feed back the actual tire pressure abnormality data to the prediction model, use new data to incrementally train and optimize the model, and continuously improve the model's prediction accuracy and adaptability.

[0073] Specifically, the tire pressure monitoring system collects vehicle tire pressure data to achieve real-time monitoring of tire status and abnormal warning. Taking a certain brand of car as an example, its tire pressure sensor collects data every 5 seconds, including tire pressure, temperature and other information. The original data may contain abnormal values, such as zero values ​​or extremely high values ​​caused by sensor failure, which need to be eliminated by methods such as median filtering. The preprocessed data set contains key features that reflect the trend of tire pressure changes, such as statistics such as tire pressure mean and standard deviation, as well as time series features such as tire pressure change rate and abnormal duration. Feature extraction is the key to building an effective prediction model. For slow air leakage, you can pay attention to the long-term change trend of tire pressure, such as calculating the average rate of decrease of tire pressure within 24 hours; for fast air leakage, you need to capture drastic changes in a short period of time, such as the drop in tire pressure within 10 minutes. These features together constitute the feature vector for tire pressure anomaly prediction, which provides input for subsequent machine learning models. Support vector machine (SVM) and random forest are commonly used machine learning algorithms that can be used to train tire pressure anomaly prediction models. SVM classifies data points by finding the optimal hyperplane, which is suitable for processing high-dimensional features; random forest integrates the results of multiple decision trees and has strong noise resistance. When training the model, grid search and other methods can be used to optimize hyperparameters, such as the kernel function type of SVM and the number of trees of random forest, to improve model performance. In order to predict the type and severity of abnormal tire pressure, a multi-classification algorithm can be used. For example, logistic regression is used to classify abnormal tire pressure into three categories: normal, slight leakage, and severe leakage, or a decision tree is used to predict specific abnormal causes, such as slow leakage, rapid leakage, and temperature abnormality. The output of these models can provide drivers with more accurate warning information. When deploying the trained model to actual applications, computing resources and real-time requirements need to be considered. For vehicle systems with limited computing power, lightweight models such as decision trees can be selected; while cloud deployment can use more complex models such as deep learning networks to achieve higher prediction accuracy. After receiving real-time tire pressure data, the model extracts and preprocesses features and inputs them into the prediction model for inference. For example, if the model predicts that tire pressure will drop by more than 20% in the next 2 hours, the system will trigger an early warning to remind the driver to pay attention to the tire condition.

[0074] S106: combining the tire pressure abnormality warning information with the vehicle control system, and when the risk of tire pressure abnormality is predicted, issuing a warning to the driver through the human-computer interaction interface. At the same time, formulating a corresponding vehicle control strategy according to the severity of the tire pressure abnormality and the predicted evolution trend.

[0075] The vehicle tire pressure data is collected in real time through the on-board sensor, and the collected tire pressure data is transmitted to the on-board controller. After receiving the tire pressure data, the on-board controller calls the pre-trained tire pressure abnormality prediction model to determine whether there is an abnormal risk in the current tire pressure. If the prediction result shows that there is a risk of abnormal tire pressure, the on-board controller sends the tire pressure abnormality warning information to the human-computer interaction interface. After the human-computer interaction interface receives the tire pressure abnormality warning information, it issues a warning reminder to the driver through a display screen, voice broadcast, etc. The on-board controller determines the corresponding vehicle control strategy based on the predicted severity and evolution trend of the tire pressure abnormality. If the severity of the tire pressure abnormality is low and the predicted abnormal evolution trend is slow, mitigation control measures such as reducing the vehicle speed and avoiding sudden acceleration are adopted. If the severity of the tire pressure abnormality is high or the predicted abnormal evolution trend is fast, mandatory control measures such as limiting the vehicle speed and cutting off the power output are adopted to ensure driving safety.

[0076] Specifically, real-time tire pressure data collection by on-board sensors is a key link in ensuring driving safety. For example, a certain model of car is equipped with a high-precision pressure sensor that can collect tire pressure data 10 times per second with an accuracy of ±0.1psi. These data are transmitted to the central controller via the on-board CAN bus for real-time monitoring. The tire pressure abnormality prediction model is the core of the intelligent early warning system. The model uses the support vector machine algorithm to predict the tire pressure status in the next 30 minutes by analyzing the change trend of historical tire pressure data. The model input includes multi-dimensional features such as current tire pressure, temperature, and driving speed, and the output is the probability and severity of tire pressure abnormality. For example, the model may predict that there is an 80% probability of a moderate leak in a certain tire in 30 minutes. The human-computer interaction interface is an important way to convey warning information to the driver. Modern on-board systems usually adopt multimodal interaction methods, including dashboard display, central control screen prompts, and voice broadcasts. When the risk of abnormal tire pressure is detected, the system will adopt different reminder strategies according to the severity. Minor abnormalities may only display a yellow warning icon on the dashboard, while serious abnormalities will trigger an audible and visual alarm and voice broadcast the specific situation. The formulation of vehicle control strategies needs to comprehensively consider the severity and evolution trend of tire pressure anomalies. For minor anomalies, the system may advise the driver to reduce speed and avoid sudden acceleration. For example, when it is predicted that the pressure of a certain tire may drop by 10% within 30 minutes, the system will recommend limiting the speed to below 80km / h. For serious anomalies, such as predicting that the tire pressure may drop by 50% within 10 minutes, the system will take more radical measures, such as automatically limiting the speed to 40km / h or cutting off the power output, and guiding the driver to stop and check as soon as possible. The implementation of this intelligent tire pressure management system can significantly improve driving safety. Through predictive maintenance, the incidence of serious accidents such as tire blowouts can be effectively reduced. Data from a certain automobile manufacturer shows that after adopting this system, the accident rate related to tire pressure has dropped by 30%. In addition, timely detection and handling of tire pressure anomalies can also extend the service life of tires and improve fuel efficiency. In actual applications, the system can save users an average of 5-10% of tire replacement costs and 2-3% of fuel consumption. However, the effectiveness of the system is highly dependent on the accuracy of the sensor and the prediction model. Therefore, it is crucial to calibrate sensors and update prediction models regularly.

[0077] On the other hand, this embodiment also provides a tire pressure abnormality prediction system based on machine learning, including:

[0078] A data acquisition module, used to acquire tire pressure data collected during vehicle driving, pre-process the tire pressure data, and obtain a tire pressure data set;

[0079] A data processing module, used to complete the tire pressure data set based on a polynomial regression method and ambient temperature change data to obtain a tire pressure time series data set;

[0080] A fluctuation acquisition module, used to perform time series analysis on the tire pressure time series data set to obtain the variation pattern of the tire pressure data, identify the variation pattern of the tire pressure data based on a combined anomaly detection algorithm, and obtain abnormal fluctuations of the tire pressure data;

[0081] A feature acquisition module, configured to preliminarily classify the abnormal fluctuation type based on the abnormal fluctuation of the tire pressure data to obtain a feature subset;

[0082] A training module, used for constructing a random forest algorithm, inputting the feature subset into the random forest algorithm for training, and obtaining a tire pressure abnormality prediction model;

[0083] The prediction module is used to predict the real-time tire pressure abnormality based on the tire pressure abnormality prediction model.

[0084] On the other hand, this embodiment further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method described in the computer program.

[0085] On the other hand, this embodiment further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and wherein the method is implemented when the computer program is executed by a processor.

[0086] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A tire pressure abnormality prediction method based on machine learning, characterized in that: include: Acquire tire pressure data collected during vehicle driving, pre-process the tire pressure data, and obtain a tire pressure data set; Completing the tire pressure data set based on a polynomial regression method and ambient temperature change data to obtain a tire pressure time series data set; Performing time series analysis on the tire pressure time series data set to obtain a change pattern of the tire pressure data, identifying the change pattern of the tire pressure data based on a combined anomaly detection algorithm, and obtaining abnormal fluctuations of the tire pressure data; Preliminarily classifying the abnormal fluctuation type based on the abnormal fluctuation of the tire pressure data to obtain a feature subset; Constructing a random forest algorithm, inputting the feature subset into the random forest algorithm for training, and obtaining a tire pressure abnormality prediction model; The real-time tire pressure abnormality is predicted based on the tire pressure abnormality prediction model.

2. The method according to claim 1, characterized in that: The process of obtaining the tire pressure data set includes: Obtaining the original tire pressure data collected during vehicle driving and preprocessing the tire pressure data; Eliminating invalid values ​​in the tire pressure data according to a preset invalid value threshold range to obtain first processed tire pressure data; Correcting abnormal values ​​in the first processed tire pressure data to obtain second processed tire pressure data; Eliminating abnormal data in the tire pressure data after the second processing to obtain tire pressure data after the third processing; Calculate the statistical features of the tire pressure data after the third processing, perform cluster analysis on the statistical features based on a density clustering algorithm, divide the tire pressure data into different clusters, regard data that does not belong to any normal cluster as noise data and remove it, and obtain a cleaned tire pressure data set.

3. The method according to claim 1, characterized in that The process of obtaining the tire pressure time series data set includes: Establishing a polynomial regression model based on the tire pressure data before and after the vehicle is parked and the ambient temperature change data in the corresponding time period in the tire pressure data set, estimating the tire pressure change trend during the parking hibernation period based on the polynomial regression model, and generating estimated tire pressure data; splicing the estimated tire pressure data and the tire pressure data set to obtain complete sequence data; The complete sequence data is smoothed to obtain the tire pressure time series data set.

4. The method according to claim 3, characterized in that The process of generating estimated tire pressure data includes: Obtaining the time period information and available tire pressure monitoring data during the vehicle's parking and dormancy period; Determine the time point when the tire pressure data is missing based on the acquired time period information and tire pressure monitoring data; A polynomial regression algorithm was used to establish a regression model with time as the independent variable and available tire pressure monitoring data as the dependent variable; The time points with missing tire pressure data were input into the established polynomial regression model to obtain the estimated tire pressure values ​​at the corresponding time points.

5. The method according to claim 1, characterized in that The process of obtaining abnormal fluctuations in tire pressure data includes: Performing sliding window segmentation on the tire pressure time series data set to obtain data subsets of multiple time segments; constructing a feature vector based on statistical features of the data subsets of the multiple time segments; The feature vector is calculated based on the isolation forest algorithm, and multiple isolation trees are constructed by randomly selecting features recursively and dividing the data space according to the feature values ​​to obtain the abnormality degree score of the abnormal point; The changing pattern of the tire pressure data is identified based on the abnormality degree score of the abnormal point, and the abnormal fluctuation of the tire pressure data is obtained.

6. The method according to claim 1, characterized in that The process of obtaining the feature subset includes: Building a decision tree model, and calculating the abnormal fluctuation of the tire pressure data based on the decision tree model to obtain a classification result; Feature engineering is constructed based on the classification results, and feature subsets that can effectively distinguish categories are extracted by analyzing the change rate and change amplitude characteristics of the tire pressure data.

7. The method according to claim 1, characterized in that The process of obtaining the abnormal tire pressure prediction model includes: Constructing a random forest model, and training the random forest model based on the feature subset to obtain a prediction model; The prediction model is cross-validated, and the hyperparameters of the model are optimized based on the validation results to obtain the abnormal tire pressure prediction model; wherein the abnormal tire pressure prediction model takes the tire pressure feature vector as input and whether the tire pressure abnormality occurs as output.

8. A tire pressure abnormality prediction system based on machine learning, characterized in that: include: A data acquisition module, used to acquire tire pressure data collected during vehicle driving, pre-process the tire pressure data, and obtain a tire pressure data set; A data processing module, used to complete the tire pressure data set based on a polynomial regression method and ambient temperature change data to obtain a tire pressure time series data set; A fluctuation acquisition module, used to perform time series analysis on the tire pressure time series data set to obtain the variation pattern of the tire pressure data, identify the variation pattern of the tire pressure data based on a combined anomaly detection algorithm, and obtain abnormal fluctuations of the tire pressure data; A feature acquisition module, configured to preliminarily classify the abnormal fluctuation type based on the abnormal fluctuation of the tire pressure data to obtain a feature subset; A training module, used for constructing a random forest algorithm, inputting the feature subset into the random forest algorithm for training, and obtaining a tire pressure abnormality prediction model; The prediction module is used to predict the real-time tire pressure abnormality based on the tire pressure abnormality prediction model.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.

Images