Tire pressure health state prediction method based on multi-source data

Abstract

The invention discloses a tire pressure health state prediction method based on multi-source data, comprising the following steps: acquiring multi-source data including tire pressure sensor data, vehicle driving data, environment temperature data and historical tire pressure data; the obtained tire pressure sensor data, the vehicle driving data, the environment temperature data and the historical tire pressure data are preprocessed; feature vectors in the tire pressure sensor data, the vehicle driving data, the environment temperature data and the historical tire pressure data are extracted and fused; and training the prediction model by adopting a machine learning algorithm, and outputting a tire pressure health state prediction result. The method has the following advantages and effects that the tire pressure state can be monitored in real time, the environment temperature, the vehicle driving condition and historical data can be comprehensively considered, multi-source data fusion and an intelligent algorithm are applied to tire pressure health state prediction, the adaptability and stability of a prediction model are improved, and the prediction efficiency is improved. And comprehensive safety guarantee is provided for the driver.

Description

Technical Field

[0001] The present invention relates to the field of automobile tire pressure safety technology, and in particular to a tire pressure health status prediction method based on multi-source data. Background Art

[0002] With the rapid development of intelligent transportation systems around the world, vehicle safety and driving experience have gradually become the focus of attention. As an important part of the vehicle safety system, the tire pressure monitoring system (TPMS) has been widely used in many modern cars. The tire pressure monitoring system monitors tire pressure in real time and detects abnormal tire pressure in a timely manner, thereby effectively reducing traffic accidents caused by improper tire pressure and ensuring the safety of drivers and passengers. However, traditional tire pressure monitoring systems only rely on single sensor data and have certain limitations, and cannot fully and accurately judge the health status of tire pressure. However, with the increasing demand for vehicle intelligence, existing technologies have gradually exposed the following limitations: 1. Traditional tire pressure monitoring systems rely solely on tire pressure sensors to detect tire pressure changes. This approach fails to take into account other factors such as the vehicle's driving status, ambient temperature, and historical changes in tire pressure, all of which may affect the accurate reading of tire pressure. For example, in cold weather, low temperatures can cause tire pressure to drop, while the vehicle's load and driving speed can also affect tire pressure. Therefore, relying solely on tire pressure sensor data cannot fully reflect the changing trend of tire pressure.

[0003] 2. Traditional systems usually only issue an alarm when tire pressure is abnormal, and do not have the ability to predict future changes in tire pressure. Especially in a dynamic driving environment, tire pressure fluctuates with time and driving conditions. The system fails to predict and adapt to these changes in time, resulting in some potential risks not being warned in time.

[0004] 3. Traditional tire pressure monitoring systems have difficulty identifying abnormal driving behaviors or other environmental factors related to tire pressure changes, such as changes in driving habits such as sudden acceleration and sudden braking. Through intelligent identification of these abnormal data, the health status of tire pressure can be better predicted and unnecessary alarms can be reduced. Therefore, how to improve the accuracy and reliability of tire pressure health status prediction has become the focus of research. In recent years, with the rise of big data technology and machine learning methods, more and more studies have begun to apply multi-source data fusion and intelligent algorithms to tire pressure health status prediction to solve problems in traditional systems. Summary of the invention

[0005] The purpose of the present invention is to provide a tire pressure health status prediction method based on multi-source data to solve the problems raised in the background technology.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions: The tire pressure health status prediction method based on multi-source data includes the following steps: S100, acquiring multi-source data, where the multi-source data includes tire pressure sensor data, vehicle driving data, ambient temperature data, and historical tire pressure data; S200, pre-processing the acquired tire pressure sensor data, vehicle driving data, ambient temperature data, and historical tire pressure data respectively; S300: After the preprocessing is completed, feature vectors are extracted from the tire pressure sensor data, the vehicle driving data, the ambient temperature data, and the historical tire pressure data, and are fused to obtain a unified feature vector; S400, using a machine learning algorithm to train a prediction model based on a unified feature vector obtained by fusion, inputting real-time tire pressure sensor data, vehicle driving data, and ambient temperature data into the trained prediction model, and outputting a tire pressure health status prediction result.

[0007] By adopting the above technical scheme, the present invention significantly improves the accuracy of tire pressure health status prediction by combining tire pressure sensor data, vehicle driving data, ambient temperature data and historical tire pressure data. The traditional tire pressure monitoring system only relies on a single tire pressure sensor data, and has the problem of poor adaptability to changes in complex driving environments; while the present invention can comprehensively consider the impact of different factors on tire pressure by integrating multiple data sources, making the prediction result more accurate. The present invention can not only make judgments based on real-time monitoring of tire pressure data, but also dynamically predict tire pressure change trends based on historical tire pressure data and environmental factors, thereby avoiding misjudgments based only on instantaneous data, especially in the case of large temperature changes and complex driving environments, the traditional system cannot effectively predict tire pressure fluctuations, and the present invention improves the reliability of prediction through historical data analysis and real-time dynamic monitoring. In addition, compared with the prior art, the present invention can consider external factors including ambient temperature and vehicle driving status, and can perceive the impact of abnormal driving behaviors such as sudden acceleration and sudden braking on tire pressure, thereby improving the accuracy of tire pressure health prediction, and reducing the deficiency that the traditional system can only alarm under abnormal conditions.

[0008] Further configuration is that the step S200 is specifically: S201, filtering tire pressure sensor data using wavelet transform to remove noise; S202, using an isolation forest algorithm to perform anomaly detection on vehicle driving data; S203, normalization processing of ambient temperature data; S204. Historical tire pressure data can be used for time series modeling to analyze tire pressure change trends and predict future tire pressure conditions.

[0009] By adopting the above technical solution and filtering the tire pressure sensor data with wavelet transform, high-frequency noise in the data can be effectively removed, ensuring the accuracy of the data collected by the tire pressure sensor; wavelet transform can not only process non-stationary signals, but also accurately capture potential important information in the data, ensuring the input quality of the subsequent prediction model.

[0010] The use of the isolation forest algorithm to perform anomaly detection on vehicle driving data can effectively identify abnormal driving behaviors or system failures. Such anomaly detection can not only improve the accuracy of tire pressure health status prediction, but also promptly alarm when abnormal situations occur to ensure driving safety. Through this anomaly detection mechanism, the actual impact of different driving behaviors on tire pressure can be compared, avoiding the traditional system's neglect of abnormal data.

[0011] The preprocessing of ambient temperature data is done through the minimum-maximum normalization method to unify the data, providing more standardized input data for subsequent fusion analysis and avoiding data inconsistency problems caused by different dimensions. Especially in an environment with extreme temperature changes, this preprocessing method can make temperature data more accurate and meaningful, avoiding false alarms of abnormal tire pressure due to abnormal temperature fluctuations.

[0012] Further configuration is that the step S201 is specifically as follows: S2011, selecting a wavelet basis function adapted to the characteristics of tire pressure sensor data; S2012, performing multi-scale wavelet decomposition on tire pressure sensor data to obtain wavelet coefficients of different scales; S2013, performing threshold processing on the wavelet coefficients to remove high-frequency noise components; S2014, reconstructing the wavelet coefficients after the threshold processing to obtain filtered tire pressure sensor data; S2015. Use signal-to-noise ratio and mean square error indicators to perform filtering evaluation on the filtered tire pressure sensor data to ensure that the filtered tire pressure sensor data can accurately reflect the actual state of the tire pressure.

[0013] By adopting the above technical solution, selecting the wavelet basis function adapted to the characteristics of the tire pressure sensor data, and performing multi-scale wavelet decomposition, the multi-frequency components in the tire pressure signal can be effectively processed, and high-frequency noise can be removed, making the tire pressure signal more stable and enhancing the accuracy of the prediction model; through the evaluation methods of signal-to-noise ratio and mean square error, it is ensured that the filtered data reflects the actual state of the tire pressure, avoiding the accumulation of errors, thereby improving the prediction effect of the system.

[0014] Further configuration is that the step S202 is specifically: S2021. Randomly sample a certain number of samples from vehicle driving data; S2022. For each sample, randomly select features and split points, and recursively build an isolation tree until all samples are isolated or the maximum depth of the tree is reached; S2023. For each data point, calculate its average path length in all isolated trees. The shorter the path length, the higher the anomaly score. S2024. According to the set threshold, mark the data points with anomaly scores higher than the threshold as abnormal; S2025. For the detected abnormal data, use the interpolation method or the mean filling method to process it.

[0015] By adopting the above technical solution, the isolation forest algorithm is used to detect anomalies in vehicle driving data, which not only improves the accuracy of tire pressure health status prediction, but also can detect abnormal driving behavior or vehicle system failure. This is a complex problem that traditional methods cannot handle. Through the isolation forest algorithm, the system can identify potential abnormal samples in the data and take appropriate measures to avoid inaccurate predictions caused by abnormal data. The isolation forest algorithm can effectively identify data that does not conform to the normal pattern by recursively constructing an isolated tree and calculating the anomaly score, reducing manual intervention and misjudgment. In particular, through random sampling and the selection of split points, it can better handle abnormal points in high-dimensional data and ensure the accuracy of subsequent training and prediction. By setting a reasonable threshold and using interpolation or mean filling method to process the detected abnormal data, the system can correct the data in time when an abnormal situation occurs. In this way, not only the accuracy of the system is improved, but also the false alarm and omission problems that may exist in traditional methods are avoided, ensuring the reliability of tire pressure monitoring results.

[0016] Further configuration is that the step S203 is specifically as follows: Since the dimension and value range of the ambient temperature data are inconsistent with other data, minimum-maximum normalization is used to ensure that the ambient temperature data are on a uniform scale; ; Among them, T is the original ambient temperature data, T min is the minimum value in the ambient temperature data, T max is the maximum value in the ambient temperature data, T norm is the normalized temperature value, and its range is limited to [0, 1].

[0017] By adopting the above technical solution, the value range of ambient temperature data is quite different from the dimension of other data sources. Through minimum-maximum normalization, all data sources can be processed at a unified scale, so that data from different sources can be effectively fused, ensuring that the fused feature vector has good consistency and effectiveness; when processing extreme temperature data, normalization is used to ensure that the physical meaning of temperature is retained, and the prediction accuracy of the system will not be affected by extreme temperature values, thereby improving the system's adaptability to environmental factors.

[0018] Further configuration is that the step S204 is specifically as follows: The autoregressive integrated moving average model is used to establish the trend model of tire pressure change; ; Among them, Y t represents the tire pressure at time t, c is a constant term, ϕ i is the autoregression coefficient, Y t−i represents the tire pressure value at time ti, θ j represents the coefficient of the moving average part, ε t−j represents the error term at time t−j, ε t represents the prediction error at time t.

[0019] By adopting the above technical solution and using the autoregressive integrated moving average model to establish a trend model for tire pressure changes, it is not only possible to perform real-time data analysis based on traditional tire pressure monitoring, but also to predict future tire pressure changes based on historical data. This is very important for predicting whether tire pressure will continue to drop or rise, especially when abnormal tire pressure is about to occur, and it can give early warning.

[0020] Further configuration is that the step S300 is specifically: S3001, extract the instantaneous pressure value, fluctuation amplitude, and pressure change rate characteristics from the tire pressure sensor data, and form a unified feature vector X tpms ; S3002, extracting vehicle speed, acceleration, and road type features from vehicle driving data, and forming a unified feature vector X vehicle ; S3003, extract the temperature difference between day and night and the temperature change rate characteristics from the ambient temperature data, and form a unified feature vector X temp ; S3004, extracting historical pressure values ​​and pressure fluctuation trend characteristics from historical tire pressure data, and forming a unified feature vector X history ; S3005, a fusion method is used to perform weighted averaging on the extracted feature vectors to generate a unified feature vector. The fusion process takes into account the weight of each data source and dynamically adjusts the weight coefficient according to the credibility of each data source. The formula is as follows: X fused =w1X tpms +w2X vehicle +w3X temp +w4X history； Among them, X fused represents the unified feature vector after fusion, w1, w2, w3, w4 are the weight coefficients of tire pressure sensor data, vehicle driving data, ambient temperature data and historical tire pressure data, respectively, X tpms is the feature vector of tire pressure sensor data, X vehicle is the feature vector of vehicle driving data, X temp is the characteristic vector of ambient temperature data, X history is the feature vector of historical tire pressure data.

[0021] By adopting the above technical solution, by extracting the features of tire pressure sensor data, vehicle driving data, ambient temperature data and historical tire pressure data, the present invention ensures that information is obtained from multiple dimensions, especially by combining the features of different data sources, the tire pressure health status can be comprehensively evaluated. This multi-dimensional information fusion effectively overcomes the one-sided judgment caused by relying only on a single data source in traditional methods. The weighted average fusion method is used to dynamically adjust the weight coefficient of each feature, and the fusion strategy can be adjusted according to the credibility of each data source; for example, when the ambient temperature has a greater impact on the tire pressure, the system can automatically increase the weight of the ambient temperature data. Unlike the traditional equal-weight fusion, this weighted fusion method can significantly improve the adaptability of the prediction model to different environmental conditions and enhance the intelligence level of the system. The feature fusion process effectively eliminates redundant information and improves the efficiency of subsequent training; traditional methods often use simple weighting or directly use the original data, which is easy to cause interference from redundant information, while the feature extraction and weighted fusion scheme of the present invention makes the features more concentrated and effectively improves the accuracy of model prediction.

[0022] Further configuration is that step S400 is specifically: S4001, the unified feature vector X generated after step S3005 fused Input into the support vector machine model as training data; S4002. Since the prediction of tire pressure health status involves nonlinear data, the RBF kernel function is used in the support vector machine model to map the data to a higher dimensional space, so that the originally linearly inseparable situation becomes linearly separable; S4003: After the training is completed, the tire pressure sensor data, vehicle driving data, and ambient temperature data acquired in real time are preprocessed and used as a unified feature vector X fused The data is input into the support vector machine model, which then outputs the tire pressure health status prediction result.

[0023] By adopting the above technical solution, using support vector machines combined with RBF kernel functions to process nonlinear data, the complex nonlinear relationship problem in tire pressure health status prediction can be effectively solved. By mapping the data to a higher-dimensional space, the support vector machine can find an optimal hyperplane to separate samples of different categories (such as normal tire pressure and abnormal tire pressure); the use of RBF kernel functions provides the support vector machine with powerful nonlinear mapping capabilities, so that the high-dimensional data in the tire pressure prediction problem can become linearly separable in the new space. The advantage of this method lies in its efficiency and accuracy in processing high-dimensional data, especially in dynamic driving environments. The support vector machine can quickly respond to real-time changes and improve the system's ability to ensure driver safety.

[0024] Further configuration is that, in step S4003, the tire pressure health status prediction result includes but is not limited to: Is the tire pressure within the normal range? Fluctuation of tire pressure, whether it is too high, too low or fluctuates abnormally; Output the prediction of future tire pressure changes based on historical tire pressure.

[0025] By adopting the above technical solution, through the prediction of real-time data, it can provide multiple information such as whether the tire pressure is within the normal range, whether there are abnormal fluctuations, and future change trends; this comprehensive health status assessment can not only issue an alarm when a problem occurs, but also help the driver predict future tire pressure changes, so as to make adjustments or maintenance in advance.

[0026] Through the above-mentioned innovations, the present invention can fully solve the problems existing in the prior art, provide a more accurate, flexible and intelligent tire pressure health status prediction method, and greatly improve the adaptability and safety of the tire pressure monitoring system in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of the process of the embodiment. DETAILED DESCRIPTION

[0028] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0029] As attached Figure 1 As shown; This embodiment discloses a tire pressure health status prediction method based on multi-source data, including the following steps: S100, obtaining multi-source data, the multi-source data including tire pressure sensor data, vehicle driving data, ambient temperature data and historical tire pressure data; the multi-source data includes historical multi-source data, with a large number of samples, each of which includes multiple features, and each sample has a corresponding label for training a subsequent prediction model; the historical multi-source data is used to train a support vector machine model. In this process, the support vector machine model learns to obtain the uniform feature vector X by minimizing the error. fused to the output prediction.

[0030] S200, pre-processing the acquired tire pressure sensor data, vehicle driving data, ambient temperature data, and historical tire pressure data respectively; S300: After the preprocessing is completed, feature vectors are extracted from the tire pressure sensor data, the vehicle driving data, the ambient temperature data, and the historical tire pressure data, and are fused to obtain a unified feature vector; S400, using a machine learning algorithm to train a prediction model based on a unified feature vector obtained by fusion, inputting real-time tire pressure sensor data, vehicle driving data, and ambient temperature data into the trained prediction model, and outputting a tire pressure health status prediction result.

[0031] It should be added that the real-time tire pressure sensor data is obtained by arranging sensors in the tire pressure monitoring system; the real-time vehicle driving data is obtained through the vehicle's on-board system or external sensors; the real-time ambient temperature data is obtained through environmental monitoring equipment or the vehicle's own temperature sensor; the historical tire pressure data is obtained through existing tire pressure monitoring equipment or the cloud platform historical data system.

[0032] Among them, step S200 is specifically as follows: S201, filtering tire pressure sensor data using wavelet transform to remove noise; S202, using an isolation forest algorithm to perform anomaly detection on vehicle driving data; S203, normalization processing of ambient temperature data; S204. Historical tire pressure data can be used for time series modeling to analyze tire pressure change trends and predict future tire pressure conditions.

[0033] In order to improve the accuracy of the prediction model, it is necessary to integrate tire pressure sensor data, vehicle driving data, ambient temperature data and historical tire pressure data to improve the prediction ability of tire pressure health status.

[0034] Among them, step S201 is specifically as follows: S2011, selecting a wavelet basis function adapted to the characteristics of the tire pressure sensor data, and in this embodiment, the Daubechies wavelet (db4) is selected; S2012, performing multi-scale wavelet decomposition on tire pressure sensor data to obtain wavelet coefficients of different scales; S2013, performing threshold processing on the wavelet coefficients to remove high-frequency noise components; S2014, reconstructing the wavelet coefficients after the threshold processing to obtain filtered tire pressure sensor data; S2015. Use signal-to-noise ratio and mean square error indicators to perform filtering evaluation on the filtered tire pressure sensor data to ensure that the filtered tire pressure sensor data can accurately reflect the actual state of the tire pressure.

[0035] It should be added that the filtered data should have a high signal-to-noise ratio and a low mean square error to ensure that the filtered data can accurately reflect the true state of the tire pressure. In this embodiment, by setting a clear threshold standard (SNR ≥ 25dB, MSE ≤ 0.5 kPa²), it is ensured that the filtered data can accurately reflect the true state of the tire pressure, thereby laying a reliable data foundation for subsequent feature extraction and health prediction.

[0036] Among them, step S202 is specifically as follows: S2021. Randomly sample a certain number of samples from vehicle driving data; S2022. For each sample, randomly select features and split points, and recursively build an isolation tree until all samples are isolated or the maximum depth of the tree is reached; S2023. For each data point, calculate its average path length in all isolated trees. The shorter the path length, the higher the anomaly score. S2024. According to the set threshold, mark the data points with anomaly scores higher than the threshold as abnormal; S2025. For the detected abnormal data, use the interpolation method or the mean filling method to process it.

[0037] Among them, step S203 is specifically as follows: Since the dimension and value range of the ambient temperature data are inconsistent with other data, minimum-maximum normalization is used to ensure that the ambient temperature data are on a uniform scale; ; Among them, T is the original ambient temperature data, T min is the minimum value in the ambient temperature data, T max is the maximum value in the ambient temperature data, T norm is the normalized temperature value, and its range is limited to [0, 1].

[0038] In this embodiment, if the ambient temperature data is normalized using the Z-score: Tstd = (T − μ) / σ; Among them, μ is the mean and σ is the standard deviation. Although Z-score can eliminate the dimension, it has the following limitations: This may cause the data to exceed the [0, 1] interval (for example, extreme temperature values ​​will cause T std Too large); The value after Z-score cannot intuitively reflect the relative position of temperature in the data set; Therefore, this embodiment uses minimum-maximum normalization for the ambient temperature data, which can not only retain the physical meaning of the data (for example, 0 represents the lowest temperature and 1 represents the highest temperature), but also strictly limit the output range to avoid the influence of abnormal values ​​on the model.

[0039] Among them, step S204 is specifically as follows: The autoregressive integrated moving average model is used to establish the trend model of tire pressure change; ; Among them, Y t represents the tire pressure at time t, c is a constant term, ϕ i is the autoregression coefficient, Y t−i represents the tire pressure value at time ti, θ j represents the coefficient of the moving average part, ε t−j represents the error term at time t−j, ε t represents the prediction error at time t.

[0040] Among them, step S300 is specifically: S3001, extract the instantaneous pressure value, fluctuation amplitude, and pressure change rate characteristics from the tire pressure sensor data, and form a unified feature vector X tpms ; S3002, extracting vehicle speed, acceleration, and road type features from vehicle driving data, and forming a unified feature vector X vehicle ; S3003, extract the temperature difference between day and night and the temperature change rate characteristics from the ambient temperature data, and form a unified feature vector X temp ; S3004, extracting historical pressure values ​​and pressure fluctuation trend characteristics from historical tire pressure data, and forming a unified feature vector X history ; S3005, a fusion method is used to perform weighted averaging on the extracted feature vectors to generate a unified feature vector. The fusion process takes into account the weight of each data source and dynamically adjusts the weight coefficient according to the credibility of each data source. The formula is as follows: Xfused =w1X tpms +w2X vehicle +w3X temp +w4X history； Among them, X fused represents the unified feature vector after fusion, w1, w2, w3, w4 are the weight coefficients of tire pressure sensor data, vehicle driving data, ambient temperature data and historical tire pressure data, respectively, X tpms is the feature vector of tire pressure sensor data, X vehicle is the feature vector of vehicle driving data, X temp is the characteristic vector of ambient temperature data, X history is the feature vector of historical tire pressure data.

[0041] Among them, step S400 is specifically: S4001, the unified feature vector X generated after step S3005 fused Input into the support vector machine model as training data; S4002. Since the prediction of tire pressure health status involves nonlinear data, the RBF kernel function is used in the support vector machine model to map the data to a higher dimensional space, so that the originally linearly inseparable situation becomes linearly separable; S4003: After the training is completed, the tire pressure sensor data, vehicle driving data, and ambient temperature data acquired in real time are preprocessed and used as a unified feature vector X fused The data is input into the support vector machine model, which then outputs the tire pressure health status prediction result.

[0042] Among them, in step S4003, the tire pressure health status prediction results include but are not limited to: Is the tire pressure within the normal range? Fluctuation of tire pressure, whether it is too high, too low or fluctuates abnormally; Output the prediction of future tire pressure changes based on historical tire pressure.

[0043] Example 1 Applied to a certain brand of new energy small sedan; S100,historical multi-source data is obtained through the brand’s cloud, with a time span of 6 months and a time resolution of 1 minute; The real-time tire pressure sensor data is obtained by the Schrader S4000 TPMS sensor, with a sampling frequency of 10Hz and a range of 0-350kPa; left front tire pressure: 240.5kPa (at start) → 248.2kPa (after 2 hours of high-speed driving), noise standard deviation 1.2kPa (electromagnetic interference); The real-time vehicle driving data is obtained from the OBD-II interface and the vehicle CAN bus, with a collection frequency of 1Hz. The average speed of the vehicle on urban roads in the vehicle driving data is 30km / h (standard deviation ±8km / h), and the high-speed cruising speed is 110km / h. The speed during emergency braking is -3.5m / s², and the speed during emergency acceleration is 2.8m / s². Real-time ambient temperature data is provided by the on-board external temperature sensor (Bosch NTC type, accuracy ±0.3°C) and the Meteorological Bureau API. The data range is 10°C (winter) to 45°C (summer), with a maximum daily temperature difference of 15°C.

[0044] Historical tire pressure data is obtained through the brand's cloud, and the natural leakage rate is 0.08kPa / day (caused by tire rubber penetration).

[0045] S201. Wavelet transform is used to filter tire pressure sensor data to remove noise. SNR is increased from 18dB to 32dB, and the high-frequency noise filtering rate is 95%.

[0046] S202. Use the isolation forest algorithm to detect anomalies in vehicle driving data; the parameter configuration is: 200 isolated trees, maximum depth 15, anomaly score threshold 0.65; after detection, the abnormal data accounts for 0.5% (such as CAN signal error when the vehicle speed suddenly changes from 0→60km / h).

[0047] S203, normalization processing of ambient temperature data; S204. Historical tire pressure data can be used for time series modeling to analyze tire pressure change trends and predict future tire pressure conditions; S3005: A fusion method is used to perform weighted averaging on the extracted feature vectors to generate a unified feature vector. The fusion process takes into account the weight of each data source and dynamically adjusts the weight coefficient according to the credibility of each data source. The details are as follows: X fused =0.5X tpms +0.3X vehicle +0.1X temp +0.1X history； S4001, the unified feature vector X generated after step S3005 fused Input into the support vector machine model as training data; S4002. Since the prediction of tire pressure health status involves nonlinear data, the RBF kernel function is used in the support vector machine model, RBF (γ=0.05, C=12), and the hyperparameters are selected by Bayesian optimization; the training set contains 50,000 samples, 80% of which are used for training and the remaining 20% ​​are used for verification; the tire pressure is predicted for the next hour, and MAE=0.7kPa.

[0048] S4003: After the training is completed, the tire pressure sensor data, vehicle driving data, and ambient temperature data acquired in real time are preprocessed and used as a unified feature vector X fused The data is input into the support vector machine model, which then outputs the tire pressure health status prediction result.

[0049] Application Scenario 1 Long-distance high-speed driving: Input data: tire pressure 248kPa (lasting 2 hours), vehicle speed 110km / h, external temperature 38°C; Prediction model output: Health status: Critical high pressure (threshold 250kPa), triggering a level 1 alarm.

[0050] Root cause analysis: Temperature rise causes 85% of the increase in tire pressure, while vehicle speed accounts for 10%.

[0051] Recommended measures: Reduce the vehicle speed to 90km / h. It is predicted that the tire pressure will drop to 245kPa within 30 minutes.

[0052] Application Scenario 2 When the tire pressure is insufficient in low temperature environment: Input data: tire pressure 215kPa (-5°C in the early morning), historical trend shows natural leakage; Prediction model output: Healthy status: lower than the standard value of 220kPa, it is recommended to replenish air to 230kPa.

[0053] Long-term forecast: Tire pressure will drop to 210kPa (confidence interval ±2kPa) in the next 7 days.

[0054] Example 2 After being applied to a certain brand of heavy-duty trucks, the accuracy rate of abnormal tire pressure warning was 99.2%, and tire blowout accidents were reduced by 76%. It can identify tire wear 2 weeks in advance based on the tire pressure fluctuation trend, reducing maintenance costs by 34%.

[0055] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A tire pressure health status prediction method based on multi-source data, characterized in that: The following steps are included: S100, acquiring multi-source data, where the multi-source data includes tire pressure sensor data, vehicle driving data, ambient temperature data, and historical tire pressure data; S200, pre-processing the acquired tire pressure sensor data, vehicle driving data, ambient temperature data, and historical tire pressure data respectively; S300: After the preprocessing is completed, feature vectors are extracted from the tire pressure sensor data, the vehicle driving data, the ambient temperature data, and the historical tire pressure data, and are fused to obtain a unified feature vector; S400, using a machine learning algorithm to train a prediction model based on a unified feature vector obtained by fusion, inputting real-time tire pressure sensor data, vehicle driving data, and ambient temperature data into the trained prediction model, and outputting a tire pressure health status prediction result.

2. The tire pressure health status prediction method based on multi-source data according to claim 1, characterized in that: The step S200 is specifically as follows: S201, filtering tire pressure sensor data using wavelet transform to remove noise; S202, using an isolation forest algorithm to perform anomaly detection on vehicle driving data; S203, normalization processing of ambient temperature data; S204. Historical tire pressure data can be used for time series modeling to analyze tire pressure change trends and predict future tire pressure conditions.

3. The tire pressure health status prediction method based on multi-source data according to claim 2, characterized in that: The step S201 is specifically as follows: S2011, selecting a wavelet basis function adapted to the characteristics of tire pressure sensor data; S2012, performing multi-scale wavelet decomposition on tire pressure sensor data to obtain wavelet coefficients of different scales; S2013, performing threshold processing on the wavelet coefficients to remove high-frequency noise components; S2014, reconstructing the wavelet coefficients after the threshold processing to obtain filtered tire pressure sensor data; S2015. Use signal-to-noise ratio and mean square error indicators to perform filtering evaluation on the filtered tire pressure sensor data to ensure that the filtered tire pressure sensor data can accurately reflect the actual state of the tire pressure.

4. The tire pressure health status prediction method based on multi-source data according to claim 2, characterized in that: The step S202 is specifically as follows: S2021. Randomly sample a certain number of samples from vehicle driving data; S2022. For each sample, randomly select features and split points, and recursively build an isolation tree until all samples are isolated or the maximum depth of the tree is reached; S2023. For each data point, calculate its average path length in all isolated trees. The shorter the path length, the higher the anomaly score. S2024. According to the set threshold, mark the data points with anomaly scores higher than the threshold as abnormal; S2025. For the detected abnormal data, use the interpolation method or the mean filling method to process it.

5. The tire pressure health status prediction method based on multi-source data according to claim 2, characterized in that: The step S203 is specifically as follows: Since the dimension and value range of the ambient temperature data are inconsistent with other data, minimum-maximum normalization is used to ensure that the ambient temperature data are on a uniform scale; ； Among them, T is the original ambient temperature data, T min is the minimum value in the ambient temperature data, T max is the maximum value in the ambient temperature data, T norm is the normalized temperature value, and its range is limited to [0, 1].

6. The tire pressure health status prediction method based on multi-source data according to claim 2, characterized in that: The step S204 is specifically as follows: The autoregressive integrated moving average model is used to establish the trend model of tire pressure change; ； Among them, Y t represents the tire pressure at time t, c is a constant term, ϕ i is the autoregression coefficient, Y t−i represents the tire pressure value at time ti, θ j represents the coefficient of the moving average part, ε t−j represents the error term at time t−j, ε t represents the prediction error at time t.

7. The tire pressure health status prediction method based on multi-source data according to claim 1, characterized in that: The step S300 is specifically as follows: S3001, extract the instantaneous pressure value, fluctuation amplitude, and pressure change rate characteristics from the tire pressure sensor data, and form a unified feature vector X tpms ; S3002, extracting vehicle speed, acceleration, and road type features from vehicle driving data, and forming a unified feature vector X vehicle ; S3003, extract the temperature difference between day and night and the temperature change rate characteristics from the ambient temperature data, and form a unified feature vector X temp ; S3004, extracting historical pressure values ​​and pressure fluctuation trend characteristics from historical tire pressure data, and forming a unified feature vector X history ; S3005, a fusion method is used to perform weighted averaging on the extracted feature vectors to generate a unified feature vector. The fusion process takes into account the weight of each data source and dynamically adjusts the weight coefficient according to the credibility of each data source. The formula is as follows: X fused =w1X tpms +w2X vehicle +w3X temp +w4X history； Among them, X fused represents the unified feature vector after fusion, w1, w2, w3, w4 are the weight coefficients of tire pressure sensor data, vehicle driving data, ambient temperature data and historical tire pressure data, respectively, X tpms is the feature vector of tire pressure sensor data, X vehicle is the feature vector of vehicle driving data, X temp is the characteristic vector of ambient temperature data, X history is the feature vector of historical tire pressure data.

8. The tire pressure health status prediction method based on multi-source data according to claim 7, characterized in that: The step S400 is specifically as follows: S4001, the unified feature vector X generated after step S3005 fused Input into the support vector machine model as training data; S4002. Since the prediction of tire pressure health status involves nonlinear data, the RBF kernel function is used in the support vector machine model to map the data to a higher dimensional space, so that the originally linearly inseparable situation becomes linearly separable; S4003: After the training is completed, the tire pressure sensor data, vehicle driving data, and ambient temperature data acquired in real time are preprocessed and used as a unified feature vector X fused The data is input into the support vector machine model, which then outputs the tire pressure health status prediction result.

9. The tire pressure health status prediction method based on multi-source data according to claim 8, characterized in that: In step S4003, the tire pressure health status prediction results include but are not limited to: Is the tire pressure within the normal range? Fluctuation of tire pressure, whether it is too high, too low or fluctuates abnormally; Output the prediction of future tire pressure changes based on historical tire pressure.

Images