A new energy automobile tire burst probability real-time prediction method

Abstract

This invention discloses a real-time prediction method for tire blowout probability in new energy vehicles. First, it collects the global state of the vehicle and the independent states of its four tires. After cleaning and denoising, it extracts temporal and physical coupling features to construct a global comprehensive feature vector. Then, it trains an initial prediction model in the cloud and distributes it. The vehicle generates the comprehensive feature vector online in real time and inputs it into the model, outputting the real-time blowout probability of the four tires, and then performs risk classification and alarm reminders. Simultaneously, the vehicle transmits data according to a trigger mechanism. The cloud extracts the transmitted data and performs sample equalization with historical scarce blowout data before retraining the model. The retrained model then replaces the vehicle-side model. This invention effectively overcomes the problem of extremely sparse blowout samples, enabling the system to continuously evolve and significantly improving prediction accuracy and driving safety.

Description

Technical Field

[0001] This invention relates to the field of automotive safety, and in particular to a method for real-time prediction of the probability of tire blowout in new energy vehicles. Background Technology

[0002] Current real-time prediction technologies for tire blowout probability in new energy vehicles primarily use high-sensitivity sensors to collect tire operating data, attempting to identify "hidden sidewall damage" and "asymmetric wear" through long-term data accumulation. Essentially, it's a "health" monitoring (i.e., equipment check-up mode) based on physical defects in the tire structure, focusing on detecting physical lesions caused by aging or damage to the tire itself.

[0003] In selecting monitoring parameters, we mainly focus on the long-term effects of new energy vehicles' unique "weight, torque, heat source" on tires, and emphasize looking for abnormal signals from the perspective of the structural health degradation of the tire itself.

[0004] Regarding the output and interaction mechanisms, the output focuses on generating "risk tracing reports" and "personalized maintenance plans." This mechanism is a non-real-time post-processing output, primarily intended for vehicle maintenance personnel or as data reference for post-event review.

[0005] This method has the following drawbacks:

[0006] 1. Lack of predictive capabilities for tire blowouts caused by "sudden driving conditions":

[0007] Existing patents primarily focus on the unique characteristics of new energy vehicles, such as "weight, torque, and heat source," as well as "asymmetric wear and hidden sidewall damage." This is essentially a monitoring system based on "health." It excels at detecting physical defects in the tire itself (such as tires being worn out or having damage). However, many tire blowouts occur when the tire itself may be healthy, but rather due to aggressive driving (such as high-speed sharp turns, sudden steering, or overloading over bumps) causing the tire to be subjected to instantaneous forces exceeding its limits. Existing patents focus on finding "abnormal signals" and struggle to predict tire blowouts caused by "normal tires under extreme dynamic conditions."

[0008] 2. Lack of intuitive "human-computer interaction" and "real-time driving guidance":

[0009] Existing patents primarily output "risk tracing reports" and "personalized maintenance plans." These are typically reports reviewed afterward or suggestions for maintenance personnel. In critical moments when a vehicle is traveling at high speed, the driver cannot access these reports or maintenance plans. Existing technologies lack an immediate, intuitive feedback mechanism that can directly intervene in the driver's current actions.

[0010] 3. Data dimension focus bias (structure vs. force):

[0011] Existing patents attempt to identify "hidden sidewall damage" and "asymmetric wear" through data. This is "fault diagnosis," which requires extremely sensitive sensors and a very long period of data accumulation to identify minute damage, and is easily affected by road surface interference, resulting in false alarms (for example, road bumps may be misinterpreted as abnormal signals). Summary of the Invention

[0012] The technical problem to be solved by the present invention is to provide a real-time prediction method for the probability of tire blowout in new energy vehicles, addressing the deficiencies mentioned in the background art.

[0013] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0014] A method for real-time prediction of tire blowout probability in new energy vehicles includes the following steps:

[0015] Step 1) Import the routine driving test record sheets collected manually and the real tire blowout failure record sheets obtained through controlled field and bench puncture tests into the cloud server to build a multi-source raw data stream;

[0016] The multi-source raw data stream includes vehicle status data, tire status data, and force and environmental data.

[0017] The vehicle status data includes the vehicle's longitudinal speed. Lateral acceleration yaw rate Steering wheel angle Brake line pressure ESC / ABS intervention flag ,in Represents the current discrete timestamp;

[0018] The tire condition data includes the real-time tire pressure, real-time tire temperature, and wheel speed of each tire, making... For tire position index, The first, second, third, and fourth tires are the front left, front right, rear left, and rear right tires of the vehicle, respectively. , , Representing vehicle number Real-time tire pressure, real-time tire temperature, and wheel speed of each tire;

[0019] The stress and environmental data include estimates of the vertical load at the wheel ends of each tire and the ambient temperature. , among which, the The estimated wheel-end vertical load of each tire is ;

[0020] Step 2) Perform data alignment and cleaning on the multi-source raw data streams to obtain the denoised initial standard time series matrix;

[0021] Step 2.1) For missing data points caused by transmission delays or packet loss, linear interpolation is used to fill in the missing data points, forcibly mapping vehicle state data sequences, tire state data, and force and environment data with different sampling frequencies to a unified timestamp. On the sequence, the formula is as follows:

[0022] ;

[0023] in, Indicates the first after filling The tires at the timestamp Data points at that location, The timestamp of the most recent valid data before the missing point. The timestamp of the most recent valid data after the missing point. , The first The tires at the timestamp , The following data points;

[0024] Step 2.2) Perform moving average noise reduction on the imputed data, setting the moving average time window size to [value missing]. The mean filter function is called to calculate the smoothed eigenvalues. ,in This represents the offset of historical moments within the time window. This yields the denoised initial standard time series matrix;

[0025] Step 3) On the cloud server, perform multi-dimensional feature vector calculation and concatenation based on the initial standard time series matrix to obtain the initial global comprehensive feature vector. ;

[0026] Step 3.1): Read the initial standard time series matrix column by column and enter the sliding time window. The statistical function is called to calculate the variance of each tire pressure within the window. ,in, The mean tire pressure within the calculated window was used to calculate the tire pressure change rate. ;

[0027] Step 3.2), based on the vehicle's constant wheelbase. with understeer gradient Combined with the matrix , , The yaw rate residual is calculated using a four-arithmetic-operation unit. ;

[0028] Step 3.3), read the wheel speed of each tire. Calculate the absolute value of the wheel speed difference between the front wheels on the same axle. Wheel speed difference with coaxial rear wheel ;

[0029] Step 3.4) Perform array concatenation operation, then... , , , , spliced ​​together as the first Independent feature vectors of each tire ,Will , , , , , , , , , Concatenate into a global vehicle data vector ;

[0030] Step 3.5) Perform one-dimensional array concatenation to generate an initial global comprehensive feature vector of fixed size. ;

[0031] Step 4) Generate a training sample set in the cloud based on the global comprehensive feature vector;

[0032] Step 4.1), set the early warning window duration for tire blowout to [specify duration]. Retrieve the physical blowout timestamps recorded in the actual tire blowout failure record table. For each physical blowout timestamp:

[0033] Step 4.1.1), assign it to the variable. And extract the corresponding tire position index. ;

[0034] Step 4.1.2), the global integrated feature vector In the time interval The data within it is treated as a positive sample and its label is set to 1;

[0035] Step 4.2), for the global integrated feature vector The remaining data after removing all positive samples are divided into negative samples according to the duration of the early warning window for tire blowout, and the label of the negative samples is set to 0 to form a training sample set.

[0036] The training sample set includes There are 1 sample; the label corresponding to sample m is . When sample m is a positive sample When sample m is a negative sample, the label ;

[0037] Step 5) Build a tire blowout prediction model in the cloud and train and calibrate it using the training sample set;

[0038] Step 5.1), Mapping function based on gradient boosting tree Establish a tire blowout prediction model, where, This is the data for the m-th sample; its input is the duration. Global comprehensive feature vector The output is the probability of a tire blowout for each of the four tires.

[0039] Step 5.2) Based on the training sample set, the blowout prediction model is trained under supervision using a weighted binary cross-entropy loss function. The calculation formula is as follows: ,in, The pre-defined positive sample penalty weights, The sigmoid activation function; mapping function This indicates that the tire blowout prediction model is effective for the first... The original risk score of each tire output ;

[0040] Step 5.3) Use the Platt Scaling algorithm to perform probability calibration on the risk score. The calculation formula is as follows: , Preset weights, These are the preset correction parameters;

[0041] Step 6) Send and deploy the trained and calibrated tire blowout prediction model to the vehicle.

[0042] Step 7) Acquire real-time data at the vehicle end to form a multi-source real-time data stream, wherein the format of the multi-source real-time data stream is the same as that of the multi-source original data stream;

[0043] Then, the multi-source real-time data streams are aligned and cleaned to obtain a denoised real-time standard time series matrix;

[0044] Next, multi-dimensional feature vectors are calculated and concatenated based on the real-time standard time series matrix to obtain the real-time global comprehensive feature vector. And stored on the vehicle side;

[0045] Step 8), integrate the real-time global feature vector. By duration After being divided, the data is input into the tire blowout prediction model deployed on the vehicle to obtain the real-time blowout probability of the four tires at the current moment, where the probability of the blowout is... Real-time tire blowout probability denoted as a multidimensional probability vector ;

[0046] Step 9), conduct risk assessment and provide alerts;

[0047] Step 9.1): Calculate the dynamic early warning threshold in real time according to the following formula. :

[0048] ;

[0049] In the formula, The preset speed sensitivity coefficient, The preset baseline speed, The preset baseline warning threshold;

[0050] Step 9.2), calculate the dynamic emergency threshold in real time according to the following formula. :

[0051] ;

[0052] In the formula, The preset baseline emergency threshold;

[0053] Step 9.3), sequentially calculate the probability of tire blowout for each of the four tires and... , Comparison, if there exists a real-time probability of a tire blowout. satisfy The system determines the situation as an "emergency" and alerts the driver; if there is a real-time probability of a tire blowout... satisfy If so, it is determined to be a "warning" state and the driver is alerted;

[0054] Step 10), when the real-time global comprehensive feature vector stored on the vehicle side... If the size exceeds the preset data capacity threshold, the stored real-time global comprehensive feature vector will be... Upload to the cloud and simultaneously clear the real-time global comprehensive feature vector stored on the vehicle. ;

[0055] Step 11), the cloud receives the real-time global comprehensive feature vector sent by the vehicle. After processing the data, the vehicle will send the real-time global comprehensive feature vector. The data is merged into the current sample training set, and then the current tire blowout prediction model is retrained based on the merged sample training set. The retrained tire blowout prediction model is then sent and deployed to the vehicle.

[0056] As a further optimization of the real-time prediction method for tire blowout probability of new energy vehicles according to the present invention, the duration of the early warning window for tire blowout is... The range is 10s-120s.

[0057] As a further optimization of the real-time prediction method for tire blowout probability of new energy vehicles according to the present invention, the duration of the early warning window for tire blowout is... Set the time to 60 seconds.

[0058] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0059] 1. By integrating vehicle dynamics data, it can accurately predict tire blowouts caused by "sudden driving conditions" and has stronger anti-interference capabilities;

[0060] Unlike existing technologies, which rely solely on high-sensitivity sensors to identify minute physical damage to the tire sidewall, this invention not only collects the tire's own state data but also innovatively integrates the vehicle's three-dimensional dynamics data, such as XYZ directional velocity, acceleration, and yaw rate, as well as force / load information.

[0061] The direct benefits include: overcoming the limitation of existing technologies that can only perform "slow damage detection" and achieving dynamic "situational awareness"; even when the tire itself is in a healthy state, this model can accurately capture situations where the tire is subjected to instantaneous force close to its physical limit due to the driver's violent operation (such as high-speed sharp turns, sudden steering, overloading and going over bumps); this multi-source data fusion significantly improves the prediction accuracy and system robustness, effectively avoids false alarms caused by the interference of road bumps from a single sensor, and adapts to more complex working conditions and driving style differences;

[0062] 2. Upgraded from traditional "single threshold alarm" to "probabilistic prediction", providing interpretable risk quantification indicators.

[0063] Unlike existing technologies, which typically use fixed tire pressure / temperature thresholds for a binary judgment of "normal / abnormal", this invention introduces machine learning and time-series feature modeling to calculate and output a specific quantified "burst probability" in real time.

[0064] The direct beneficial effects are: it changes the shortcomings of traditional threshold alarms, which are prone to delays or false alarms; the probabilistic and interpretable risk output mechanism enables the system to implement multi-level risk management (such as mild prompts, moderate warnings, and emergency intervention) according to different probability ranges; this not only provides drivers with more detailed decision-making basis, but also provides a quantitative data interface for subsequent vehicle active safety takeover systems;

[0065] 3. Fine-grained status monitoring for single wheel positions significantly improves risk location accuracy.

[0066] Technical features that differ from existing technologies: The method and model of this invention support independent data extraction and inference for four-wheeled or multi-axle multi-wheeled vehicles;

[0067] The direct beneficial effects are: it can accurately output the probability of a tire blowout for each individual tire position, and the driver can intuitively grasp the dynamic trend of the relative risk of each tire through the central control vehicle tire topology map; compared with the general alarm for the whole vehicle, fine-grained single tire prediction helps the driver to more accurately judge the source of risk and take targeted avoidance or maintenance measures. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

[0069] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings:

[0070] This invention can be implemented in many different forms and should not be considered as limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully express the scope of the invention to those skilled in the art.

[0071] like Figure 1 As shown, this invention discloses a method for real-time prediction of tire blowout probability in new energy vehicles, comprising the following steps:

[0072] Step 1) Import the routine driving test record sheets collected manually and the real tire blowout failure record sheets obtained through controlled field and bench puncture tests into the cloud server to build a multi-source raw data stream;

[0073] The multi-source raw data stream includes vehicle status data, tire status data, and force and environmental data.

[0074] The vehicle status data includes the vehicle's longitudinal speed. Lateral acceleration yaw rate Steering wheel angle Brake line pressure ESC / ABS intervention flag ,in Represents the current discrete timestamp;

[0075] The tire condition data includes the real-time tire pressure, real-time tire temperature, and wheel speed of each tire, making... For tire position index, The first, second, third, and fourth tires are the front left, front right, rear left, and rear right tires of the vehicle, respectively. , , Representing vehicle number Real-time tire pressure, real-time tire temperature, and wheel speed of each tire;

[0076] The stress and environmental data include estimates of the vertical load at the wheel ends of each tire and the ambient temperature. , among which, the The estimated wheel-end vertical load of each tire is ;

[0077] Step 2) Perform data alignment and cleaning on the multi-source raw data streams to obtain the denoised initial standard time series matrix;

[0078] Step 2.1) For missing data points caused by transmission delays or packet loss, linear interpolation is used to fill in the missing data points, forcibly mapping vehicle state data sequences, tire state data, and force and environment data with different sampling frequencies to a unified timestamp. On the sequence, the formula is as follows:

[0079] ;

[0080] in, Indicates the first after filling The tires at the timestamp Data points at that location, The timestamp of the most recent valid data before the missing point. The timestamp of the most recent valid data after the missing point. , The first The tires at the timestamp , The following data points;

[0081] Step 2.2) Perform moving average noise reduction on the imputed data, setting the moving average time window size to [value missing]. The mean filter function is called to calculate the smoothed eigenvalues. ,in This represents the offset of historical moments within the time window. This yields the denoised initial standard time series matrix;

[0082] Step 3) On the cloud server, perform multi-dimensional feature vector calculation and concatenation based on the initial standard time series matrix to obtain the initial global comprehensive feature vector. ;

[0083] Step 3.1): Read the initial standard time series matrix column by column and enter the sliding time window. The statistical function is called to calculate the variance of each tire pressure within the window. ,in, The mean tire pressure within the calculated window was used to calculate the tire pressure change rate. ;

[0084] Step 3.2), based on the vehicle's constant wheelbase. with understeer gradient Combined with the matrix , , The yaw rate residual is calculated using a four-arithmetic-operation unit. ;

[0085] Step 3.3), read the wheel speed of each tire. Calculate the absolute value of the wheel speed difference between the front wheels on the same axle. Wheel speed difference with coaxial rear wheel ;

[0086] Step 3.4) Perform array concatenation operation, then... , , , , spliced ​​together as the first Independent feature vectors of each tire ,Will , , , , , , , , , Concatenate into a global vehicle data vector ;

[0087] Step 3.5) Perform one-dimensional array concatenation to generate an initial global comprehensive feature vector of fixed size. ;

[0088] Step 4) Generate a training sample set in the cloud based on the global comprehensive feature vector;

[0089] Step 4.1), set the early warning window duration for tire blowout to [specify duration]. Retrieve the physical blowout timestamps recorded in the actual tire blowout failure record table. For each physical blowout timestamp:

[0090] Step 4.1.1), assign it to the variable. And extract the corresponding tire position index. ;

[0091] Step 4.1.2), the global integrated feature vector In the time interval The data within it is treated as a positive sample and its label is set to 1;

[0092] Step 4.2), for the global integrated feature vector The remaining data after removing all positive samples are divided into negative samples according to the duration of the early warning window for tire blowout, and the label of the negative samples is set to 0 to form a training sample set.

[0093] The training sample set includes There are 1 sample; the label corresponding to sample m is . When sample m is a positive sample When sample m is a negative sample, the label ;

[0094] Step 5) Build a tire blowout prediction model in the cloud and train and calibrate it using the training sample set;

[0095] Step 5.1), Mapping function based on gradient boosting tree Establish a tire blowout prediction model, where, This is the data for the m-th sample; its input is the duration. Global comprehensive feature vector The output is the probability of a tire blowout for each of the four tires.

[0096] Step 5.2) Based on the training sample set, the blowout prediction model is trained under supervision using a weighted binary cross-entropy loss function. The calculation formula is as follows: ,in, The pre-defined positive sample penalty weights, The sigmoid activation function; mapping function This indicates that the tire blowout prediction model is effective for the first... The original risk score of each tire output ;

[0097] Step 5.3) Use the Platt Scaling algorithm to perform probability calibration on the risk score. The calculation formula is as follows: , Preset weights, These are the preset correction parameters;

[0098] Step 6) Send and deploy the trained and calibrated tire blowout prediction model to the vehicle.

[0099] Step 7) Acquire real-time data at the vehicle end to form a multi-source real-time data stream, wherein the format of the multi-source real-time data stream is the same as that of the multi-source original data stream;

[0100] Then, the multi-source real-time data streams are aligned and cleaned to obtain a denoised real-time standard time series matrix;

[0101] Next, multi-dimensional feature vectors are calculated and concatenated based on the real-time standard time series matrix to obtain the real-time global comprehensive feature vector. And stored on the vehicle side;

[0102] Step 8), integrate the real-time global feature vector. By duration After being divided, the data is input into the tire blowout prediction model deployed on the vehicle to obtain the real-time blowout probability of the four tires at the current moment, where the probability of the blowout is... Real-time tire blowout probability denoted as a multidimensional probability vector ;

[0103] Step 9), conduct risk assessment and provide alerts;

[0104] Step 9.1): Calculate the dynamic early warning threshold in real time according to the following formula. :

[0105] ;

[0106] In the formula, The preset speed sensitivity coefficient, The preset baseline speed, The preset baseline warning threshold;

[0107] Step 9.2), calculate the dynamic emergency threshold in real time according to the following formula. :

[0108] ;

[0109] In the formula, The preset baseline emergency threshold;

[0110] Step 9.3), sequentially calculate the probability of tire blowout for each of the four tires and... , Comparison, if there exists a real-time probability of a tire blowout. satisfy The system determines the situation as an "emergency" and alerts the driver; if there is a real-time probability of a tire blowout... satisfy If so, it is determined to be a "warning" state and the driver is alerted;

[0111] Step 10), when the real-time global comprehensive feature vector stored on the vehicle side... If the size exceeds the preset data capacity threshold, the stored real-time global comprehensive feature vector will be... Upload to the cloud and simultaneously clear the real-time global comprehensive feature vector stored on the vehicle. ;

[0112] Step 11), the cloud receives the real-time global comprehensive feature vector sent by the vehicle. After processing the data, the vehicle will send the real-time global comprehensive feature vector. The data is merged into the current sample training set, and then the current tire blowout prediction model is retrained based on the merged sample training set. The retrained tire blowout prediction model is then sent and deployed to the vehicle.

[0113] The duration of the early warning window for tire blowout The range is 10s-120s, with 60s preferred.

[0114] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0115] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A new energy vehicle tire burst probability real-time prediction method, characterized in that, Includes the following steps: Step 1) Import the routine driving test record sheets collected manually and the real tire blowout failure record sheets obtained through controlled field and bench puncture tests into the cloud server to build a multi-source raw data stream; The multi-source raw data stream includes vehicle status data, tire status data, and force and environmental data. The vehicle status data includes the vehicle's longitudinal speed. Lateral acceleration yaw rate Steering wheel angle Brake line pressure ESC / ABS intervention flag ,in Represents the current discrete timestamp; The tire condition data includes the real-time tire pressure, real-time tire temperature, and wheel speed of each tire, making... For tire position index, The first, second, third, and fourth tires are the front left, front right, rear left, and rear right tires of the vehicle, respectively. , , Representing vehicle number Real-time tire pressure, real-time tire temperature, and wheel speed of each tire; The force and environment data include wheel end vertical load estimation values of each tire and an ambient environment temperature , wherein the wheel end vertical load estimation value of the first tire is ; Step 2) Perform data alignment and cleaning on the multi-source raw data streams to obtain the denoised initial standard time series matrix; Step 2.1), for the missing data points caused by transmission delay or packet loss, linear interpolation method is used to fill in, and the vehicle state data sequence with different sampling frequencies, tire state data, force and environmental data are forced to map to the unified timestamp In sequence, the formula is as follows: ； in, Indicates the first after filling The tires at the timestamp Data points at that location, The timestamp of the most recent valid data before the missing point. The timestamp of the most recent valid data after the missing point. , The first The tires at the timestamp , The following data points; Step 2.2) Perform moving average noise reduction on the imputed data, setting the moving average time window size to [value missing]. The mean filter function is called to calculate the smoothed eigenvalues. ,in This represents the offset of historical moments within the time window. This yields the denoised initial standard time series matrix; Step 3), the cloud server performs the calculation and splicing of the multi-dimensional feature vectors based on the initial standard time series matrix, to obtain an initial global comprehensive feature vector ; Step 3.1): Read the initial standard time series matrix column by column and enter the sliding time window. The statistical function is called to calculate the variance of each tire pressure within the window. ,in, The mean tire pressure within the calculated window was used to calculate the tire pressure change rate. ; Step 3.2), based on the vehicle's constant wheelbase. with understeer gradient Combined with the matrix , , The yaw rate residual is calculated using a four-arithmetic-operation unit. ; Step 3.3), reading the tire wheel speed of each tire , respectively calculating the absolute value of the difference between the front wheel speed and the rear wheel speed of the same axle ; and the absolute value of the difference between the front wheel speed and the rear wheel speed of the same axle ; Step 3.4) Perform array concatenation operation, then... , , , , spliced ​​together as the first Independent feature vectors of each tire ,Will , , , , , , , , , Concatenate into a global vehicle data vector ; Step 3.5) Perform one-dimensional array concatenation to generate an initial global comprehensive feature vector of fixed size. ; Step 4) Generate a training sample set in the cloud based on the global comprehensive feature vector; Step 4.1), set the length of the tire burst early warning window as , search the physical tire burst time stamp recorded in the real tire burst failure record table, for each physical tire burst time stamp: Step 4.1.1), and assign it to the variable and extract the corresponding tire position index ; Step 4.1.2), the global integrated feature vector In the time interval The data within it is treated as a positive sample and its label is set to 1; Step 4.2), for the global comprehensive feature vector After removing all positive sample data, the remaining data is divided into various negative samples according to the length of the tire burst early warning window, and the label of the negative sample is set to 0 to form a training sample set. The training sample set comprises samples; a label corresponding to a sample m is , when the sample m is a positive sample , the label of the sample m is ; Step 5) Build a tire blowout prediction model in the cloud and train and calibrate it using the training sample set; Step 5.1), Mapping function based on gradient boosting tree Establish a tire blowout prediction model, where, This is the data for the m-th sample; its input is the duration. Global comprehensive feature vector The output is the probability of a tire blowout for each of the four tires. Step 5.2) Based on the training sample set, the blowout prediction model is trained under supervision using a weighted binary cross-entropy loss function. The calculation formula is as follows: ,in, The pre-defined positive sample penalty weights, The sigmoid activation function; mapping function This indicates that the tire blowout prediction model is effective for the first... The original risk score of each tire output ; Step 5.3) Use the Platt Scaling algorithm to perform probability calibration on the risk score. The calculation formula is as follows: , Preset weights, These are the preset correction parameters; Step 6) Send and deploy the trained and calibrated tire blowout prediction model to the vehicle. Step 7) Acquire real-time data at the vehicle end to form a multi-source real-time data stream, wherein the format of the multi-source real-time data stream is the same as that of the multi-source original data stream; Then, the multi-source real-time data streams are aligned and cleaned to obtain a denoised real-time standard time series matrix; Next, multi-dimensional feature vectors are calculated and concatenated based on the real-time standard time series matrix to obtain the real-time global comprehensive feature vector. And stored on the vehicle side; Step 8), the real-time global comprehensive feature vector According to the length After division, input into the tire burst prediction model deployed at the vehicle end to obtain the real-time tire burst probability of the four tires at the current time, wherein the real-time tire burst probability of the first , denoted as a multi-dimensional probability vector ;​ Step 9), conduct risk assessment and provide alerts; Step 9.1), the dynamic early warning threshold is calculated in real time according to the following formula : ； In the formula, The preset speed sensitivity coefficient, The preset baseline speed, The preset baseline warning threshold; Step 9.2), the dynamic emergency threshold is calculated in real time according to the following formula : ； In the formula, is a preset reference emergency threshold value; Step 9.3), sequentially calculate the probability of tire blowout for each of the four tires and... , Comparison, if there exists a real-time probability of a tire blowout. satisfy The system determines the situation as an "emergency" and alerts the driver; if there is a real-time probability of a tire blowout... satisfy If so, it is determined to be a "warning" state and the driver is alerted; Step 10), when the real-time global comprehensive feature vector stored on the vehicle side... If the size exceeds the preset data capacity threshold, the stored real-time global comprehensive feature vector will be... Upload to the cloud and simultaneously clear the real-time global comprehensive feature vector stored on the vehicle. ; Step 11): The cloud receives the real-time global comprehensive feature vector sent by the vehicle. After processing the data, the vehicle will send the real-time global comprehensive feature vector. The data is merged into the current sample training set, and then the current tire blowout prediction model is retrained based on the merged sample training set. The retrained tire blowout prediction model is then sent and deployed to the vehicle.

2. The method for real-time prediction of tire blowout probability for new energy vehicles according to claim 1, characterized in that, The length of the tire burst early warning window is in the range of 10s-120s.

3. The method for real-time prediction of tire blowout probability for new energy vehicles according to claim 2, characterized in that, The duration of the early warning window for tire blowout Set the time to 60 seconds.

Images