A tire pressure monitoring method based on big data analysis

Abstract

The present invention belongs to the technical field of pressure monitoring, and discloses a tire pressure monitoring method for big data analysis, comprising: collecting multimodal data and corresponding pressure labels of the same time series, preprocessing the multimodal data, and obtaining optimized modal data; performing environmental stripping on pressure data based on environmental data to obtain pressure purification data, and dynamically decoupling the pressure purification data based on vehicle dynamic data to obtain pressure compensation data; performing multi-dimensional feature extraction on the pressure compensation data to obtain pressure features; constructing a pressure monitoring model based on the pressure features and the pressure labels, and realizing accurate monitoring of tire pressure based on the pressure monitoring model; and greatly improving the driving safety of the vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of pressure monitoring, and more specifically, to a tire pressure monitoring method based on big data analysis. Background Art

[0002] With the continuous development of the automobile industry and the improvement of its intelligence level, tire pressure monitoring systems have become an important part of modern automobile safety technology; the main function of a tire pressure monitoring system is to monitor the air pressure of vehicle tires in real time, detect abnormal tire pressure in time, and reduce traffic accidents caused by tire pressure problems; although the existing tire pressure monitoring system plays an important role in ensuring driving safety, it still has many limitations and cannot meet the increasingly complex driving environment and the demand for intelligent and high-precision monitoring.

[0003] Traditional tire pressure monitoring systems focus on real-time monitoring of current pressure status and lack the ability to predict pressure change trends in the long term. Changes in tire pressure are often gradual, and small abnormal changes in the early stages may not immediately cause safety issues, but over time, such changes may lead to tire failures. If the system cannot identify this trend, the driver may miss the best time to maintain the tire. Tire pressure is not only affected by internal gas, but also by external environmental factors (such as temperature, humidity, etc.). Existing tire pressure monitoring systems usually fail to take into account the impact of these environmental factors on tire pressure, resulting in inaccurate monitoring results from the tire pressure monitoring system. The dynamic state of the vehicle will have a direct impact on tire pressure. Existing systems cannot fully consider these factors under dynamic driving conditions, which often results in false alarms or missed reports, and the accuracy of monitoring cannot be guaranteed.

[0004] In view of this, the present invention proposes a tire pressure monitoring method based on big data analysis to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a tire pressure monitoring method for big data analysis, comprising:

[0006] S1. Collect multimodal data and corresponding pressure labels of the same time series, where the multimodal data includes: environmental data, vehicle dynamic data and pressure data, and pre-process the multimodal data to obtain optimized modal data;

[0007] S2, performing environmental stripping on the pressure data based on the environmental data to obtain pressure purification data, and dynamically decoupling the pressure purification data based on the vehicle dynamic data to obtain pressure compensation data;

[0008] S3, extracting multi-dimensional features from the pressure compensation data to obtain pressure features;

[0009] S4. Build a pressure monitoring model based on pressure characteristics and pressure labels, and accurately monitor tire pressure based on the pressure monitoring model.

[0010] Furthermore, the environmental data include: external temperature, external humidity and atmospheric pressure; the vehicle dynamic data include: acceleration and tilt angle; and the pressure data include: tire pressure.

[0011] Furthermore, the method of preprocessing the multimodal data includes:

[0012] Each type of data in the multimodal data is taken as a particle set, and each element in the particle set is taken as a particle state. The particle state is clustered using a clustering analysis algorithm to obtain a particle state cluster. The initial state probability corresponding to each particle state is calculated based on the particle state cluster. The initial state probability is the total number of particle states in the particle state cluster where the particle state is located divided by the total number of particle states in the particle set.

[0013] Preset the disturbance factor and filter value, predict the particle state based on the disturbance factor to obtain the particle disturbance, use the density clustering algorithm to cluster the particle disturbance and obtain the particle disturbance cluster; calculate the disturbance state probability corresponding to each particle disturbance based on the particle disturbance cluster, the disturbance state probability is the total number of particle disturbances in the particle disturbance cluster where the particle disturbance is located divided by the total number of particle states in the particle set; update the particle state disturbance weight based on the disturbance state probability to obtain the disturbance weight; normalize the disturbance weight by the minimum-maximum normalization algorithm to obtain Get the normalized weight; map each particle state to a selection wheel based on the normalized weight, divide each particle state into the corresponding arc area in the selection wheel based on the corresponding normalized weight, resample the particle state based on the selection wheel, rotate the selection wheel to get the sampled particles, the number of sampled particles for each resampling is the number of arc areas minus one, repeat until the number of sampled particles is equal to the filter value, and output the disturbance weight at this time as the update sequence; update the particle state based on the update sequence to get the filtered particles, and all the filtered particles constitute the optimized modal data.

[0014] Furthermore, the formula for updating the disturbance weight is: Among them, w t represents the perturbation weight of the t-th particle state, a represents the scale parameter, which is used to control the fluctuation of the perturbation weight, and L t represents the perturbation state probability corresponding to the particle perturbation of the t-th particle state, V t represents the state of the tth particle, X t represents the particle perturbation of the state of the tth particle, P tRepresents the initial state probability of the tth particle state.

[0015] Furthermore, the method of performing environment stripping on the pressure data based on the environment data includes:

[0016] An initial environment stripping architecture model is constructed, and the initial environment stripping architecture model includes an input layer, a feature extraction layer and a regression layer; the input layer takes environment data and pressure data as input, the feature extraction layer extracts features from the environment data, and types each data type in the environment data to obtain a type code; dimension mapping is performed on each type of data in the environment data based on the type code to obtain a mapping value; N layers of convolution layers are preset, the mapping values ​​are vectorized to obtain a mapping vector, and feature convolution is performed on the mapping vector based on the convolution layer to obtain high-dimensional environment features, and the regression layer takes the predicted pressure data as output; the mean square error function is used as the loss function, and the initial environment stripping architecture model is trained by the least squares method. When the loss function converges, the initial environment stripping architecture model at this time is output as the environment stripping architecture model; the environment data is predicted based on the constructed environment stripping architecture model to obtain the predicted pressure, and the pressure data is subtracted from the predicted pressure data to obtain the pressure purification data.

[0017] Furthermore, the formula for dimension mapping is: Among them, K ij Represents the mapping value of the jth data in the i-th category of environmental data, Val i Represents the benchmark data of the i-th category of environmental data, Val' ij represents the jth data in the i-th category data, and δ represents the mapping bandwidth parameter.

[0018] Furthermore, the method of dynamically decoupling the pressure purification data based on the vehicle dynamic data includes:

[0019] Collect historical static pressure data and historical dynamic state data. The historical static pressure data is the pressure data when the vehicle is at rest. The historical dynamic state data includes: historical dynamic pressure data, historical acceleration and historical tilt angle. Construct a pressure coupling function based on the historical static pressure data and historical dynamic state data. The formula of the pressure coupling function is:

[0020] Among them, CP represents the pressure coupling coefficient, ΔP represents the pressure difference between the historical dynamic pressure data and the historical static pressure data, g represents the gravitational constant, m represents the mass of the vehicle, add represents the historical acceleration, v represents the vehicle speed, and θ represents the historical tilt angle; μ1 represents the acceleration coupling coefficient, and μ2 represents the tilt coupling coefficient; based on the optimal acceleration coupling coefficient, the optimal tilt coupling coefficient and the optimal pressure coupling coefficient, the vehicle dynamic data is used as the input of the pressure coupling function to solve the function and obtain the pressure difference, and the pressure purification data is summed with the pressure difference to obtain the pressure compensation data.

[0021] Furthermore, the optimal acceleration coupling coefficient, the optimal tilt coupling coefficient and the optimal pressure coupling coefficient are obtained by:

[0022] Initialize M groups of coupling coefficient combinations, the coupling coefficient combinations include acceleration coupling coefficients and tilt coupling coefficients, initialize the best coupling group to be empty, initialize the best fitness to be infinitesimal, preset the selection body value, take each group of coupling coefficient combinations as a chromosome, and the genes in the chromosome correspond to the acceleration coupling coefficients and the tilt coupling coefficients; for each chromosome, use the historical dynamic state data as the coefficient training set, calculate the pressure coupling coefficient of each group of historical dynamic state data according to the pressure coupling function, take the standard deviation of the pressure coupling coefficient as the fitness of the chromosome, randomly select the number of chromosomes that are consistent with the selection body value based on the selection body value, compare the fitness of the selected chromosomes, take the chromosome with the smallest fitness value as the parent chromosome, stop when the number of parent chromosomes is consistent with the selection body value, and all parent chromosomes constitute the parent chromosome set;

[0023] The permutation and combination algorithm is used to pair the chromosomes in the parent chromosome set in pairs to obtain the parent combination; a gene in the parent combination is randomly selected as the exchange gene, and the exchange gene in the parent combination is exchanged to obtain the daughter chromosome, and the fitness of each daughter chromosome is calculated based on the coefficient training set and the pressure coupling function, and the daughter chromosome with a fitness less than the parent combination is taken as the preferred generation; the permutation and combination algorithm is used to pair the preferred generation in pairs to obtain the second generation combination;

[0024] Preset the mutation factor, the disturbance factor is a minimum value less than zero, randomly select a gene in the second-generation combination as the mutant gene, increase or decrease the value of A mutation factors for the mutant gene with random probability, obtain the mutant chromosome, calculate the fitness of each mutant chromosome based on the coefficient training set and the pressure coupling function, take the mutant chromosome with a fitness less than the second-generation combination as the preferred mutant, select the preferred mutant with the largest fitness as the candidate individual, when the fitness of the candidate individual is less than the optimal fitness, take the candidate individual as the new optimal coupling, take the fitness of the candidate individual as the new optimal fitness, take the preferred mutant as the new parent chromosome, repeat until the value of the optimal fitness no longer changes, output the optimal coupling group and optimal fitness at this time, the data in the optimal coupling group is the optimal acceleration coupling coefficient and the optimal tilt coupling coefficient, and the optimal fitness is the optimal pressure coupling coefficient.

[0025] Furthermore, the method of extracting multi-dimensional features from the pressure compensation data includes:

[0026] The time series of pressure compensation data is plotted using the Lagrange interpolation method to obtain a pressure-time curve graph, the pressure-time curve graph is Fourier transformed to obtain a Fourier pressure curve graph, the pressure-time curve graph is wavelet transformed to obtain a wavelet pressure curve graph, a high-frequency threshold and a low-frequency threshold are preset, and based on the low-frequency threshold, the curves in the Fourier pressure curve graph that are greater than the low-frequency threshold are removed to obtain a low-frequency incomplete curve, and based on the high-frequency threshold, the curves in the wavelet pressure curve graph that are less than the high-frequency threshold are removed to obtain a high-frequency incomplete curve;

[0027] Integrate each curve in the low-frequency defective curve over time to obtain the total pressure in the low-frequency area, and divide the total pressure in the low-frequency area by the time span of the corresponding curve to obtain the low-frequency average pressure; average the total pressure in the low-frequency area to obtain the low-frequency average total pressure, average the low-frequency average pressure to obtain the low-frequency segmented pressure, and subtract the total pressure in the low-frequency area of ​​the previous curve segment from the next curve segment in the low-frequency defective curve to obtain the low-frequency pressure change, calculate the standard deviation of the low-frequency pressure change, and obtain the low-frequency change standard deviation; integrate each curve in the high-frequency defective curve over time to obtain the high-frequency area Total pressure: divide the total pressure in the high-frequency area by the time span of the corresponding curve to obtain the average pressure of the high-frequency curve; average the total pressure in the high-frequency area to obtain the high-frequency average total pressure; average the average pressure of the high-frequency curve to obtain the high-frequency segmented pressure; subtract the total pressure in the high-frequency area of ​​the previous curve segment from the next curve segment in the high-frequency incomplete curve to obtain the high-frequency pressure change; calculate the standard deviation of the high-frequency pressure change to obtain the high-frequency transformation standard deviation; the low-frequency average total pressure, low-frequency segmented pressure, low-frequency change standard deviation, high-frequency average total pressure, high-frequency segmented pressure, and high-frequency change standard deviation constitute the pressure characteristics.

[0028] Furthermore, the pressure monitoring model is constructed in the following manner:

[0029] The pressure features and pressure labels are used as training sample sets, the linear regression model is used as the initial model, the linear regression model is trained using the training sample set, the pressure features are used as the input data of the pressure monitoring model, and the predicted pressure labels are used as the output data of the pressure monitoring model; minimizing the error between the actual pressure label and the predicted pressure label is used as the training goal, and the recall rate function is used as the loss function of the pressure monitoring model. When the loss function converges, the training is stopped to obtain the pressure monitoring model.

[0030] Technical effects and advantages of a tire pressure monitoring method based on big data analysis of the present invention:

[0031] The present invention can effectively track the changing trend of tire pressure and respond quickly in a dynamically changing environment through the updating and resampling mechanism of particle states, so that tire pressure monitoring can still maintain high accuracy in a complex or incomplete information environment; by introducing environmental data to perform environmental stripping on pressure data, the interference of temperature, humidity changes and atmospheric pressure fluctuations on tire pressure can be effectively eliminated, which not only ensures the adaptability of the monitoring system to environmental factors, but also improves the ability to accurately capture pure tire pressure changes; through the establishment and optimization of the pressure coupling function, not only can the pressure difference be accurately captured when the vehicle body changes dynamically, but also the pressure data can be accurately compensated based on the change of the vehicle body state, thereby eliminating the interference of the vehicle body dynamics on pressure monitoring; by constructing a pressure monitoring model, the tire pressure state is comprehensively evaluated, and the source of pressure change is accurately identified, avoiding the monitoring deviation caused by environmental changes or vehicle body dynamics in traditional methods, which enables the system to provide more accurate tire pressure monitoring services in practical applications, thereby effectively reducing the safety hazards caused by abnormal tire pressure. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic diagram of a tire pressure monitoring method for big data analysis of the present invention;

[0033] Figure 2 A schematic diagram of a tire pressure monitoring system for big data analysis according to the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] Example 1

[0036] See also Figure 1 As shown, the tire pressure monitoring method of this embodiment based on big data analysis includes:

[0037] S1. Collect multimodal data and corresponding pressure labels of the same time series, where the multimodal data includes: environmental data, vehicle dynamic data and pressure data, and pre-process the multimodal data to obtain optimized modal data;

[0038] S2, performing environmental stripping on the pressure data based on the environmental data to obtain pressure purification data, and dynamically decoupling the pressure purification data based on the vehicle dynamic data to obtain pressure compensation data;

[0039] S3, extracting multi-dimensional features from the pressure compensation data to obtain pressure features;

[0040] S4. Build a pressure monitoring model based on pressure characteristics and pressure labels, and accurately monitor tire pressure based on the pressure monitoring model;

[0041] Environmental data include: outside temperature, outside humidity and atmospheric pressure; vehicle dynamic data include: acceleration and tilt angle; pressure data include: tire pressure; pressure labels are descriptions of different tire pressures, common ones are low pressure and high pressure, etc.; outside temperature is obtained through temperature sensors, and outside humidity is obtained through humidity sensors. Temperature and humidity data are generally collected once a minute, but in cases of large weather changes, the collection frequency needs to be dynamically adjusted. Atmospheric pressure is obtained through pressure sensors, and the collection frequency of atmospheric pressure is usually low, generally once per minute to once every 5 minutes. When the atmospheric pressure changes greatly, the collection frequency is increased; acceleration data is obtained through acceleration sensors, and the tilt angle is obtained through gyroscope measurements. The tilt angle of the vehicle body, especially in turning, uphill, downhill, etc., will affect the load distribution and pressure of the tire; tire pressure is obtained through pressure sensors.

[0042] Methods for preprocessing multimodal data include:

[0043] Each type of data in the multimodal data is regarded as a particle set, and each element in the particle set is regarded as a particle state. The particle states are clustered using a clustering analysis algorithm to obtain a particle state cluster. Common clustering analysis algorithms include hierarchical clustering algorithm and density clustering algorithm. The initial state probability corresponding to each particle state is calculated based on the particle state cluster. The initial state probability is the total number of particle states in the particle state cluster where the particle state is located divided by the total number of particle states in the particle set. Clustering analysis helps to cluster similar particle states together, so that the probability distribution of each particle can be better evaluated in the initial state. The calculation of the initial state probability can reflect the relative importance of the particle state in the particle set, thereby providing a basis for subsequent disturbance prediction and particle update.

[0044] Preset the disturbance factor and filter value. The disturbance factor is an integer greater than zero, and the filter value is an integer greater than zero. The particle state is predicted based on the disturbance factor. The formula for disturbance prediction is: X t =V t-1 +P t ×α; where X t represents the particle perturbation of the state of the tth particle, V t-1 represents the state of the t-1th particle, P t Represents the initial state probability of the t-th particle state, and α represents the perturbation factor; the density clustering algorithm is used to cluster the particle perturbations to obtain the particle perturbation clusters; the perturbation state probability corresponding to each particle perturbation is calculated based on the particle perturbation clusters, and the perturbation state probability is the total number of particle perturbations in the particle perturbation cluster where the particle perturbation is located divided by the total number of particle states in the particle set; the particle state is perturbation weighted based on the perturbation state probability, and the formula for updating the perturbation weight is: Among them, w t represents the perturbation weight of the t-th particle state, a represents the scale parameter, which is used to control the fluctuation of the perturbation weight, and L t represents the perturbation state probability corresponding to the particle perturbation of the t-th particle state, V tRepresents the state of the tth particle; normalize the disturbance weight by the minimum-maximum normalization algorithm to obtain the normalized weight; map each particle state to a selection wheel based on the normalized weight, divide the corresponding arc area in the selection wheel based on the corresponding normalized weight, resample the particle state based on the selection wheel, rotate the selection wheel to obtain the sampled particles, the number of sampled particles for each resampling is the number of arc areas minus one, repeat until the number of sampled particles is equal to the filter value, and output the disturbance weight at this time as the update sequence; by introducing the disturbance factor to predict the particle state, the dynamic changes and uncertainty of the system can be taken into account. The disturbance prediction can provide a reasonable estimate of the evolution of the particle state in an incomplete or uncertain environment, so that the particle filter can still maintain a high accuracy in complex situations; update the particle state based on the update sequence, and the formula for particle update is: Among them, Ch b represents the bth filter particle, Wei b represents the perturbation weight of the b-th particle state, D b represents the bth particle state, Num represents the number of particle states, and all filtered particles constitute the optimized modal data.

[0045] The methods of environmental stripping of pressure data based on environmental data include:

[0046] Construct an initial environment stripping architecture model, which includes an input layer, a feature extraction layer, and a regression layer. The input layer takes environmental data and pressure data as input, and the feature extraction layer extracts features from the environmental data, numbers each data type in the environmental data, and obtains the type code. Based on the type code, each type of data in the environmental data is dimensionally mapped, and the formula for dimension mapping is: Among them, K ij Represents the mapping value of the jth data in the i-th category of environmental data, Val i Represents the benchmark data of the i-th category of environmental data, Val' ij Represents the jth data in the i-th category of data, δ represents the mapping bandwidth parameter, which is used to smooth the generalization ability and performance of dimensional mapping. The smaller the mapping bandwidth parameter, the higher the degree of smoothness, and the lower the tolerance to noise and local fluctuations, which may lead to overfitting. The larger the mapping bandwidth parameter, the lower the degree of smoothness, and the higher the tolerance to local fluctuations and noise, which may lead to underfitting. N layers of convolution layers are preset, the mapping value is vectorized to obtain the mapping vector, and the mapping vector is feature convolved based on the convolution layer. The formula for feature convolution is: Among them, End represents the high-dimensional environment features, N represents the total number of convolutional layers, and Juan c Represents the convolution operation of the cth convolutional layer, Inc represents the input of the cth convolutional layer, σ represents the activation function, and the common activation function is the ReLU function. The regression layer uses the predicted pressure data as the output; the mean square error function is used as the loss function, and the initial environment stripping architecture model is trained by the least squares method. When the loss function converges, the initial environment stripping architecture model at this time is output as the environment stripping architecture model; the environment data is predicted based on the constructed environment stripping architecture model to obtain the predicted pressure, and the predicted pressure data is subtracted from the pressure data to obtain the pressure purification data;

[0047] Environmental stripping of pressure data can effectively eliminate the interference of temperature, humidity changes and atmospheric pressure fluctuations on tire pressure. This stripping strategy based on environmental data not only ensures the adaptability of the monitoring system to environmental factors, but also improves the ability to accurately capture pure tire pressure changes. By building an environmental stripping architecture model, this method can accurately predict pressure changes caused by environmental factors, thereby obtaining "pressure purification data", which provides a reliable basis for further dynamic decoupling.

[0048] The type code is the code of the data type in the environmental data. Assuming that the code of the outside temperature in the environmental data is 1, the code of the outside humidity is 2, and the code of the atmospheric pressure is 3, the value set of i in the formula for dimensional mapping is: {1,2,3}; the benchmark data represents the benchmark value of each type of data in the environmental data. Taking atmospheric pressure as an example, the benchmark data of atmospheric pressure is 101.325 kPa.

[0049] The methods for dynamically decoupling pressure purification data based on vehicle dynamic data include:

[0050] Collect historical static pressure data and historical dynamic state data. The historical static pressure data is the pressure data when the vehicle is at rest. The historical dynamic state data includes: historical dynamic pressure data, historical acceleration and historical tilt angle. Construct a pressure coupling function based on the historical static pressure data and historical dynamic state data. The formula of the pressure coupling function is:

[0051] Among them, CP represents the pressure coupling coefficient, ΔP represents the pressure difference between the historical dynamic pressure data and the historical static pressure data, g represents the gravitational constant, m represents the mass of the vehicle, add represents the historical acceleration, v represents the vehicle speed, and θ represents the historical tilt angle; μ1 represents the acceleration coupling coefficient, and μ2 represents the tilt coupling coefficient;

[0052] Initialize M groups of coupling coefficient combinations, where M is an integer greater than zero, and the coupling coefficient combination includes an acceleration coupling coefficient and an inclination coupling coefficient. Initialize the optimal coupling group to be empty, initialize the optimal fitness to be infinitesimal, preset the selection body value, and take each group of coupling coefficient combinations as a chromosome, where the genes in the chromosome correspond to the acceleration coupling coefficient and the inclination coupling coefficient. For each chromosome, use the historical dynamic state data as the coefficient training set, calculate the pressure coupling coefficient of each group of historical dynamic state data according to the pressure coupling function, take the standard deviation of the pressure coupling coefficient as the fitness of the chromosome, randomly select the number of chromosomes that are consistent with the selection body value based on the selection body value, compare the fitness of the selected chromosomes, and take the chromosome with the smallest fitness value as the parent chromosome. Stop when the number of parent chromosomes is consistent with the selection body value, and all parent chromosomes constitute the parent chromosome set.

[0053] The permutation and combination algorithm is used to pair the chromosomes in the parent chromosome set in pairs to obtain the parent combination; a gene in the parent combination is randomly selected as the exchange gene, and the exchange gene in the parent combination is exchanged to obtain the daughter chromosome, and the fitness of each daughter chromosome is calculated based on the coefficient training set and the pressure coupling function, and the daughter chromosome with a fitness less than the parent combination is taken as the preferred generation; the permutation and combination algorithm is used to pair the preferred generation in pairs to obtain the second generation combination;

[0054] Preset the mutation factor, which is a minimum value less than zero. Randomly select a gene in the second-generation combination as the mutant gene, increase or decrease the value of A mutation factors for the mutant gene with random probability, obtain the mutant chromosome, calculate the fitness of each mutant chromosome based on the coefficient training set and the pressure coupling function, and take the mutant chromosome with a fitness less than the second-generation combination as the preferred mutant. Select the preferred mutant with the largest fitness as the candidate individual. When the fitness of the candidate individual is less than the optimal fitness, take the candidate individual as the new optimal coupling, take the fitness of the candidate individual as the new optimal fitness, and take the preferred mutant as the new parent chromosome. Repeat until the value of the optimal fitness no longer changes, output the optimal coupling group and optimal fitness at this time, the data in the optimal coupling group is the optimal acceleration coupling coefficient and the optimal tilt coupling coefficient, and the optimal fitness is the optimal pressure coupling coefficient;

[0055] Substituting the optimal pressure coupling coefficient, the optimal acceleration coupling coefficient and the optimal tilt coupling coefficient into the pressure coupling function, solving the function with the vehicle dynamic data as the input of the pressure coupling function, obtaining the pressure difference, performing pressure correction on the pressure purification data with the pressure difference, summing the pressure purification data with the pressure difference, and obtaining the pressure compensation data;

[0056] Based on the historical static pressure data and vehicle dynamic data, a pressure coupling function is constructed to perform dynamic decoupling processing, accurately simulating and compensating for the impact of vehicle body dynamics on tire pressure. Through the establishment and optimization of the pressure coupling function, not only can the pressure difference be accurately captured when the vehicle body changes dynamically, but the pressure data can also be accurately compensated based on changes in the vehicle body state, thereby eliminating the interference of vehicle body dynamics on pressure monitoring.

[0057] Methods for extracting multi-dimensional features from pressure compensation data include:

[0058] The time series of pressure compensation data is plotted using the Lagrange interpolation method to obtain a pressure-time curve graph, the pressure-time curve graph is Fourier transformed to obtain a Fourier pressure curve graph, the pressure-time curve graph is wavelet transformed to obtain a wavelet pressure curve graph, a high-frequency threshold and a low-frequency threshold are preset, and based on the low-frequency threshold, the curves in the Fourier pressure curve graph that are greater than the low-frequency threshold are removed to obtain a low-frequency incomplete curve, and based on the high-frequency threshold, the curves in the wavelet pressure curve graph that are less than the high-frequency threshold are removed to obtain a high-frequency incomplete curve;

[0059] Integrate each curve segment in the low-frequency incomplete curve in time to obtain the total pressure in the low-frequency area, divide the total pressure in the low-frequency area by the time span of the corresponding curve to obtain the low-frequency average pressure; average the total pressure in the low-frequency area to obtain the low-frequency average total pressure, average the low-frequency average pressure to obtain the low-frequency segmented pressure, subtract the total pressure in the low-frequency area of ​​the previous curve segment from the next curve segment in the low-frequency incomplete curve to obtain the low-frequency pressure change, calculate the standard deviation of the low-frequency pressure change, and obtain the low-frequency change standard deviation;

[0060] Integrate each curve segment in the high-frequency incomplete curve in time to obtain the total pressure in the high-frequency area, and divide the total pressure in the high-frequency area by the time span of the corresponding curve to obtain the average pressure of the high-frequency curve; average the total pressure in the high-frequency area to obtain the high-frequency average total pressure, average the average pressure of the high-frequency curve to obtain the high-frequency segmented pressure, subtract the total pressure in the high-frequency area of ​​the previous curve segment from the next curve segment in the high-frequency incomplete curve to obtain the high-frequency pressure change, calculate the standard deviation of the high-frequency pressure change, and obtain the high-frequency transformation standard deviation;

[0061] The pressure characteristics are composed of low-frequency average total pressure, low-frequency segmented pressure, low-frequency change standard deviation, high-frequency average total pressure, high-frequency segmented pressure, and high-frequency change standard deviation.

[0062] The construction methods of the pressure monitoring model include:

[0063] The pressure features and pressure labels are used as training sample sets, the linear regression model is used as the initial model, the linear regression model is trained using the training sample set, the pressure features are used as the input data of the pressure monitoring model, and the predicted pressure labels are used as the output data of the pressure monitoring model; minimizing the error between the actual pressure label and the predicted pressure label is used as the training goal, and the recall rate function is used as the loss function of the pressure monitoring model. When the loss function converges, the training is stopped to obtain the pressure monitoring model, and accurate monitoring of tire pressure is achieved based on the constructed pressure monitoring model.

[0064] This embodiment can effectively track the changing trend of tire pressure and respond quickly in a dynamically changing environment through the update and resampling mechanism of particle state, so that tire pressure monitoring can still maintain high accuracy in a complex or incomplete information environment; by introducing environmental data to perform environmental stripping on pressure data, it can effectively eliminate the interference of temperature, humidity changes and atmospheric pressure fluctuations on tire pressure, which not only ensures the adaptability of the monitoring system to environmental factors, but also improves the ability to accurately capture pure tire pressure changes; through the establishment and optimization of the pressure coupling function, it can not only accurately capture the pressure difference when the vehicle body changes dynamically, but also accurately compensate for the pressure data based on the change of the vehicle body state, thereby eliminating the interference of the vehicle body dynamics on pressure monitoring; by constructing a pressure monitoring model, comprehensively evaluating the pressure state of the tire, accurately identifying the source of pressure changes, and avoiding the monitoring deviation caused by environmental changes or vehicle body dynamics in traditional methods, which enables the system to provide more accurate tire pressure monitoring services in actual applications, thereby effectively reducing the safety hazards caused by abnormal tire pressure.

[0065] Example 2

[0066] See also Figure 2 As shown, the part not described in detail in this embodiment can be seen from the description of embodiment 1, and a tire pressure monitoring system for big data analysis is provided, including:

[0067] Data acquisition and processing module: collects multimodal data and corresponding pressure labels of the same time series. The multimodal data includes: environmental data, vehicle dynamic data and pressure data. Preprocesses the multimodal data to obtain optimized modal data.

[0068] Influencing factor module: based on the environmental data, the pressure data is stripped of the environment to obtain the pressure purification data; based on the vehicle dynamic data, the pressure purification data is dynamically decoupled to obtain the pressure compensation data;

[0069] Feature extraction module: extract multi-dimensional features from pressure compensation data to obtain pressure features;

[0070] Model building module: builds a pressure monitoring model based on pressure features and pressure labels, and accurately monitors tire pressure based on the pressure monitoring model;

[0071] The modules are connected to each other via wired and / or wireless means to achieve data transmission between modules.

[0072] Example 3

[0073] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the tire pressure monitoring method for big data analysis provided above is implemented.

[0074] Since the electronic device introduced in this embodiment is an electronic device used to implement a tire pressure monitoring method for big data analysis in the embodiment of this application, based on the tire pressure monitoring method for big data analysis introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the tire pressure monitoring method for big data analysis in the embodiment of this application, it belongs to the scope of protection of this application.

[0075] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0076] The above are only preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A tire pressure monitoring method based on big data analysis, characterized in that: include: S1. Collect multimodal data and corresponding pressure labels of the same time series, where the multimodal data includes: environmental data, vehicle dynamic data and pressure data, and pre-process the multimodal data to obtain optimized modal data; S2, performing environmental stripping on the pressure data based on the environmental data to obtain pressure purification data, and dynamically decoupling the pressure purification data based on the vehicle dynamic data to obtain pressure compensation data; S3, extracting multi-dimensional features from the pressure compensation data to obtain pressure features; S4. Build a pressure monitoring model based on pressure characteristics and pressure labels, and accurately monitor tire pressure based on the pressure monitoring model; The method of performing environment stripping on pressure data based on environment data includes: Construct an initial environment stripping architecture model, which includes an input layer, a feature extraction layer and a regression layer; the input layer takes environment data and pressure data as input, the feature extraction layer extracts features from the environment data, types each data type in the environment data, and obtains a type code; dimension mapping is performed on each type of data in the environment data based on the type code to obtain a mapping value; N layers of convolution layers are preset, the mapping values ​​are vectorized to obtain a mapping vector, and feature convolution is performed on the mapping vector based on the convolution layer to obtain high-dimensional environment features, and the regression layer uses the predicted pressure data as output; the mean square error function is used as the loss function, and the initial environment stripping architecture model is trained using the least squares method. When the loss function converges, the initial environment stripping architecture model at this time is output as the environment stripping architecture model; the environment data is predicted based on the constructed environment stripping architecture model to obtain the predicted pressure, and the pressure data is subtracted from the predicted pressure data to obtain the pressure purification data; The method of dynamically decoupling the pressure purification data based on the vehicle dynamic data includes: Collect historical static pressure data and historical dynamic state data. The historical static pressure data is the pressure data when the vehicle is at rest. The historical dynamic state data includes: historical dynamic pressure data, historical acceleration and historical tilt angle. Construct a pressure coupling function based on the historical static pressure data and historical dynamic state data. The formula of the pressure coupling function is: Among them, CP represents the pressure coupling coefficient, ΔP represents the pressure difference between the historical dynamic pressure data and the historical static pressure data, g represents the gravitational constant, m represents the mass of the vehicle, add represents the historical acceleration, v represents the vehicle speed, and θ represents the historical tilt angle; μ1 represents the acceleration coupling coefficient, and μ2 represents the tilt coupling coefficient; based on the optimal acceleration coupling coefficient, the optimal tilt coupling coefficient and the optimal pressure coupling coefficient, the vehicle dynamic data is used as the input of the pressure coupling function to solve the function and obtain the pressure difference, and the pressure purification data is summed with the pressure difference to obtain the pressure compensation data.

2. The tire pressure monitoring method based on big data analysis according to claim 1, characterized in that: The environmental data include: external temperature, external humidity and atmospheric pressure; the vehicle dynamic data include: acceleration and tilt angle; the pressure data include: tire pressure.

3. The tire pressure monitoring method based on big data analysis according to claim 2, characterized in that: The method of preprocessing the multimodal data includes: Each type of data in the multimodal data is taken as a particle set, and each element in the particle set is taken as a particle state. The particle state is clustered using a clustering analysis algorithm to obtain a particle state cluster. The initial state probability corresponding to each particle state is calculated based on the particle state cluster. The initial state probability is the total number of particle states in the particle state cluster where the particle state is located divided by the total number of particle states in the particle set. Preset the disturbance factor and filter value, predict the particle state based on the disturbance factor to obtain the particle disturbance, use the density clustering algorithm to cluster the particle disturbance and obtain the particle disturbance cluster; calculate the disturbance state probability corresponding to each particle disturbance based on the particle disturbance cluster, the disturbance state probability is the total number of particle disturbances in the particle disturbance cluster where the particle disturbance is located divided by the total number of particle states in the particle set; update the particle state disturbance weight based on the disturbance state probability to obtain the disturbance weight; normalize the disturbance weight by the minimum-maximum normalization algorithm to obtain Get the normalized weight; map each particle state to a selection wheel based on the normalized weight, divide each particle state into the corresponding arc area in the selection wheel based on the corresponding normalized weight, resample the particle state based on the selection wheel, rotate the selection wheel to get the sampled particles, the number of sampled particles for each resampling is the number of arc areas minus one, repeat until the number of sampled particles is equal to the filter value, and output the disturbance weight at this time as the update sequence; update the particle state based on the update sequence to get the filtered particles, and all the filtered particles constitute the optimized modal data.

4. The tire pressure monitoring method based on big data analysis according to claim 3, characterized in that: The formula for updating the disturbance weight is: Among them, w t represents the perturbation weight of the t-th particle state, a represents the scale parameter, which is used to control the fluctuation of the perturbation weight, and L t represents the probability of the perturbation state corresponding to the particle perturbation of the t-th particle state, V t represents the state of the tth particle, X t represents the particle perturbation of the state of the tth particle, P t Represents the initial state probability of the tth particle state.

5. The tire pressure monitoring method based on big data analysis according to claim 4, characterized in that: The formula for dimension mapping is: Among them, K ij Represents the mapping value of the jth data in the i-th category of environmental data, Val i Represents the benchmark data of the i-th category of environmental data, Val' ij represents the jth data in the i-th category data, and δ represents the mapping bandwidth parameter.

6. The tire pressure monitoring method based on big data analysis according to claim 5, characterized in that: The optimal acceleration coupling coefficient, the optimal tilt coupling coefficient and the optimal pressure coupling coefficient are obtained by: Initialize M groups of coupling coefficient combinations, the coupling coefficient combinations include acceleration coupling coefficients and tilt coupling coefficients, initialize the best coupling group to be empty, initialize the best fitness to be infinitesimal, preset the selection body value, take each group of coupling coefficient combinations as a chromosome, and the genes in the chromosome correspond to the acceleration coupling coefficients and the tilt coupling coefficients; for each chromosome, use the historical dynamic state data as the coefficient training set, calculate the pressure coupling coefficient of each group of historical dynamic state data according to the pressure coupling function, take the standard deviation of the pressure coupling coefficient as the fitness of the chromosome, randomly select the number of chromosomes that are consistent with the selection body value based on the selection body value, compare the fitness of the selected chromosomes, take the chromosome with the smallest fitness value as the parent chromosome, stop when the number of parent chromosomes is consistent with the selection body value, and all parent chromosomes constitute the parent chromosome set; The permutation and combination algorithm is used to pair the chromosomes in the parent chromosome set in pairs to obtain the parent combination; a gene in the parent combination is randomly selected as the exchange gene, and the exchange gene in the parent combination is exchanged to obtain the daughter chromosome, and the fitness of each daughter chromosome is calculated based on the coefficient training set and the pressure coupling function, and the daughter chromosome with a fitness less than the parent combination is taken as the preferred generation; the permutation and combination algorithm is used to pair the preferred generation in pairs to obtain the second generation combination; Preset the mutation factor, the disturbance factor is a minimum value less than zero, randomly select a gene in the second-generation combination as the mutant gene, increase or decrease the value of A mutation factors for the mutant gene with random probability, obtain the mutant chromosome, calculate the fitness of each mutant chromosome based on the coefficient training set and the pressure coupling function, take the mutant chromosome with a fitness less than the second-generation combination as the preferred mutant, select the preferred mutant with the largest fitness as the candidate individual, when the fitness of the candidate individual is less than the optimal fitness, take the candidate individual as the new optimal coupling, take the fitness of the candidate individual as the new optimal fitness, take the preferred mutant as the new parent chromosome, repeat until the value of the optimal fitness no longer changes, output the optimal coupling group and optimal fitness at this time, the data in the optimal coupling group is the optimal acceleration coupling coefficient and the optimal tilt coupling coefficient, and the optimal fitness is the optimal pressure coupling coefficient.

7. The tire pressure monitoring method based on big data analysis according to claim 6, characterized in that: The method of extracting multi-dimensional features from pressure compensation data includes: The time series of pressure compensation data is plotted using the Lagrange interpolation method to obtain a pressure-time curve graph, the pressure-time curve graph is Fourier transformed to obtain a Fourier pressure curve graph, the pressure-time curve graph is wavelet transformed to obtain a wavelet pressure curve graph, a high-frequency threshold and a low-frequency threshold are preset, and based on the low-frequency threshold, the curves in the Fourier pressure curve graph that are greater than the low-frequency threshold are removed to obtain a low-frequency incomplete curve, and based on the high-frequency threshold, the curves in the wavelet pressure curve graph that are less than the high-frequency threshold are removed to obtain a high-frequency incomplete curve; Integrate each curve in the low-frequency defective curve over time to obtain the total pressure in the low-frequency area, and divide the total pressure in the low-frequency area by the time span of the corresponding curve to obtain the low-frequency average pressure; average the total pressure in the low-frequency area to obtain the low-frequency average total pressure, average the low-frequency average pressure to obtain the low-frequency segmented pressure, and subtract the total pressure in the low-frequency area of ​​the previous curve segment from the next curve segment in the low-frequency defective curve to obtain the low-frequency pressure change, calculate the standard deviation of the low-frequency pressure change, and obtain the low-frequency change standard deviation; integrate each curve in the high-frequency defective curve over time to obtain the high-frequency area Total pressure: divide the total pressure in the high-frequency area by the time span of the corresponding curve to obtain the average pressure of the high-frequency curve; average the total pressure in the high-frequency area to obtain the high-frequency average total pressure; average the average pressure of the high-frequency curve to obtain the high-frequency segmented pressure; subtract the total pressure in the high-frequency area of ​​the previous curve segment from the next curve segment in the high-frequency incomplete curve to obtain the high-frequency pressure change; calculate the standard deviation of the high-frequency pressure change to obtain the high-frequency transformation standard deviation; the low-frequency average total pressure, low-frequency segmented pressure, low-frequency change standard deviation, high-frequency average total pressure, high-frequency segmented pressure, and high-frequency change standard deviation constitute the pressure characteristics.

8. The tire pressure monitoring method based on big data analysis according to claim 7, characterized in that: The construction method of the pressure monitoring model includes: The pressure features and pressure labels are used as training sample sets, the linear regression model is used as the initial model, the linear regression model is trained using the training sample set, the pressure features are used as the input data of the pressure monitoring model, and the predicted pressure labels are used as the output data of the pressure monitoring model; minimizing the error between the actual pressure label and the predicted pressure label is used as the training goal, and the recall rate function is used as the loss function of the pressure monitoring model. When the loss function converges, the training is stopped to obtain the pressure monitoring model.

Images