Tire pressure automatic calibration method and system based on single-point reference and scene self-learning
By adopting an automatic tire pressure calibration method based on single-point benchmark and scenario self-learning, and combining the principles of resonance frequency and rolling radius, the fusion strategy is dynamically adjusted to solve the problems of high hardware cost and poor adaptability of existing tire pressure monitoring systems. This achieves high-precision and reliable tire pressure monitoring, adapts to diverse driving scenarios, and self-optimizes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN JIANGXIA CHUNENG AUTOMOBILE TECHNOLOGY R&D CO LTD
- Filing Date
- 2026-03-03
- Publication Date
- 2026-04-17
AI Technical Summary
Existing tire pressure monitoring systems suffer from high hardware costs, calibration logic that is difficult to adapt to diverse driving scenarios, limited accuracy under boundary conditions, concentrated patent protection for core algorithms, and insufficient space for innovative applications.
An automatic tire pressure calibration method based on single-point benchmark and scenario self-learning is adopted. By collecting vehicle signal data, driving scenarios are judged in real time. The tire pressure estimate is calculated by combining the principles of resonance frequency and rolling radius. The fusion weight is dynamically adjusted according to scenario mode and signal quality, and the deviation is used for calibration to achieve continuous rolling update of parameters.
Reduce system costs, improve accuracy and reliability, automatically adapt to a wide variety of driving scenarios, break the limitations of fixed parameter modes, enhance estimation robustness in complex environments, and ensure accuracy for long-term use.
Smart Images
Figure CN121871308A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of automotive electronics and safety control technology, specifically to a method and system for automatic tire pressure calibration based on a single-point reference and scenario self-learning. Background Technology
[0002] Tire pressure monitoring is a key technology in automotive electronics and safety control, and its accuracy directly affects vehicle safety, economy, and stability. Currently, mainstream tire pressure monitoring systems are mainly divided into two categories: direct and indirect. Direct TPMS obtains real-time tire pressure by installing pressure sensors in each tire, offering high measurement accuracy, but suffers from high hardware costs, limited sensor battery life, and susceptibility to wireless signal interference. Indirect TPMS utilizes existing wheel speed sensors and other components to infer tire pressure by analyzing signals such as wheel speed differences. While offering cost advantages, its accuracy and reliability are significantly affected by driving conditions, and it is prone to deviations in complex scenarios. To balance cost and accuracy, hybrid solutions combining direct and indirect sensing technologies have emerged in the industry, attempting to integrate the advantages of both systems; however, there is still room for further optimization in these solutions.
[0003] Existing technologies still face numerous challenges in practical applications: most hybrid solutions require the deployment of multiple direct sensors, which not only increases hardware costs but also presents significant patent barriers; calibration logic largely relies on preset fixed compensation parameter tables or specific filtering algorithms, making it difficult to adapt to the ever-changing driving scenarios in actual vehicle use, such as differences in conditions like highway cruising versus urban congestion, full load versus empty load, and different temperature ranges, resulting in inconsistent calibration accuracy under boundary conditions; furthermore, patent protection for core algorithms is relatively concentrated, focusing primarily on single-frequency extraction or wheel speed comparison logic, limiting innovative applications. To address these challenges, we propose an automatic tire pressure calibration method and system based on a single-point benchmark and scenario self-learning. Summary of the Invention
[0004] To address the aforementioned technical issues, this paper provides an automatic tire pressure calibration method and system based on a single-point benchmark and scenario self-learning. This technical solution solves the problems of most of the above solutions relying on multiple direct sensors, resulting in high hardware costs and patent barriers; calibration logic often using fixed parameters or algorithms, making it difficult to adapt to diverse driving scenarios and limiting accuracy under boundary conditions; and the concentrated patent protection of core algorithms, making them susceptible to existing patent restrictions and limiting the space for innovative applications.
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The automatic tire pressure calibration method based on single-point reference and scenario self-learning includes the following steps: S1. Collect vehicle raw signals, including: direct tire pressure signal of a single designated tire of the vehicle, wheel speed signal of each wheel, triaxial acceleration signal and ambient temperature signal; S2. Extract multi-dimensional features based on the collected signal data to determine the current driving scenario mode of the vehicle in real time; S3. Based on the principle of resonance frequency analysis and the principle of rolling radius calculation, the relative tire pressure estimates of each tire are calculated in parallel. S4. Based on the identified scenario pattern and real-time signal quality, dynamically determine the fusion weight, and perform weighted fusion on the tire pressure estimates calculated based on the resonant frequency analysis principle and the rolling radius calculation principle to obtain the fused tire pressure estimates for all tires except for a single specified tire. S5. Based on the direct measurement of a single specified tire, calculate the systematic deviation between it and the fusion tire pressure estimate of the other tires excluding the single specified tire, and use the deviation to calibrate the fusion estimate and output the final tire pressure value. S6. Under the condition that the vehicle can drive stably and the reference data is reliable, the calibration data under the current scenario is used as a sample to continuously update and optimize the calibration parameters corresponding to the scenario.
[0006] Preferably, the method for extracting multi-dimensional features based on the collected signal data is as follows: Based on the collected raw signal data, the wheel speed signal, triaxial acceleration signal and ambient temperature signal are preprocessed and segmented, including: dividing the data segments using a fixed-duration sliding window mechanism, and filtering, compensating and handling missing values for each signal; Primary statistical features are extracted from the preprocessed signal, including the mean and standard deviation of the vehicle speed profile, the statistics of lateral and longitudinal acceleration, the vibration energy and dominant frequency of the wheel speed signal in a specific frequency band, and the average value and rate of change of the ambient temperature. The extracted primary statistical features are nonlinearly combined to generate advanced fusion features, including the stationarity index, curve compliance factor, pavement excitation level, and thermal state coefficient. By integrating the aforementioned primary statistical features and advanced fusion features, a multidimensional feature vector representing the driving scenario is constructed for each time window.
[0007] Preferably, the real-time determination of the current driving scenario mode of the vehicle specifically includes: Based on the multidimensional feature vector generated in the current time window, calculate the weighted distance between it and the anchor points of each existing scenario feature vector in the scenario fingerprint database; The comparison between the minimum weighted distance and the preset distance threshold is used as the initial discrimination criterion: if the minimum distance is less than the threshold, it is determined that the corresponding existing scenario is matched, and the calibration parameter set of that scenario is directly called; if the minimum distance is greater than or equal to the threshold, the fine discrimination process is triggered. The detailed discrimination process is as follows: first, the current feature vector is classified into the macro driving category according to the preset rules, and then the density of historical data points in the local neighborhood of the current feature vector is counted in the feature space. The final determination is made based on the comparison between the local density and the preset density threshold: if the density is not lower than the threshold, it is determined to be a transitional state of a known scenario, and the parameters of neighboring scenarios are fused for this calculation; if the density is lower than the threshold, it is determined to be a completely new driving scenario. For a newly identified driving scenario, a new entry is created in the scenario fingerprint database, initialized with the current feature vector as the anchor point, and assigned default or calibration parameters based on macro-class inheritance.
[0008] Preferably, S3 includes: The collected wheel speed signals of each wheel were transformed by time and frequency to extract the resonant peak frequency that characterizes the radial vibration of the tire. Based on the resonant peak frequency, current vehicle speed, and pre-calibrated tire equivalent mass and stiffness coefficient, the estimated tire pressure for each tire based on the resonant frequency principle is calculated. Based on the collected wheel speed signals and triaxial acceleration signals, compensation is made for the differential speed between the left and right wheels caused by vehicle steering. Calculate the wheel speed ratio between coaxial or focused wheels, and deduce the relative change in the corresponding tire rolling radius based on the wheel speed ratio; Based on the relative change in rolling radius and the physical parameters of the tire carcass, the estimated tire pressure for each tire based on the rolling radius principle is calculated.
[0009] Preferably, the relative tire pressure estimate of each tire calculated in parallel in S3 is the ratio of the tire pressure of each tire to the average tire pressure of all tires.
[0010] Preferably, the dynamic determination of the fusion weights specifically involves: Obtain the weight mapping function parameters corresponding to the current recognition scenario pattern from the scenario fingerprint database; Real-time calculation of signal quality vectors including signal-to-noise ratio of each wheel speed signal, lateral acceleration, ambient temperature, and road vibration level; The real-time signal quality vector is input into the weight mapping function of the current scenario, and the real-time fusion weight between the resonant frequency model and the rolling radius model is output.
[0011] Preferably, obtaining the fusion tire pressure estimate for all tires except the single specified tire specifically involves: Obtain tire pressure estimates for each tire calculated based on the principle of resonant frequency, and tire pressure estimates for each tire calculated based on the principle of rolling radius; Obtain the fusion weights dynamically determined based on the current scenario mode and real-time signal quality; The fusion weights are used as the weighting coefficients of the first model, and the value of one minus the fusion weights is used as the weighting coefficients of the second model. Multiply the tire pressure estimate of the first model by its weighting coefficient, multiply the tire pressure estimate of the second model by its weighting coefficient, and add the two products together to obtain the combined tire pressure estimate for the corresponding tire.
[0012] Preferably, the calculation of the systematic deviation between the calculated tire pressure and the estimated fusion tire pressure of all tires except the single specified tire specifically refers to: Calculate the ratio of the tire pressure of each tire obtained based on the fusion algorithm to the average tire pressure of all tires; Based on the direct measurement of a single specified tire, the theoretical tire pressure values of the remaining tires are calculated according to the ratio. The difference between the estimated fusion tire pressure of each tire and its corresponding theoretical tire pressure is calculated as the instantaneous system deviation. The instantaneous system deviation is subjected to exponential smoothing to obtain the smoothed system deviation, which is then stored in the parameter set corresponding to the current scenario. The process of using this deviation to calibrate the fusion estimate and outputting the final tire pressure value is as follows: Obtain the smoothed system bias corresponding to the current scenario; A comprehensive coefficient for determining the validity of the deviation is dynamically calculated. This coefficient is determined based on the health of the benchmark sensor, the confidence level of the scenario matching, and the timeliness of the deviation itself. Multiply the smoothing system deviation by the comprehensive coefficient to obtain the calibration amount for practical application; The final calibrated tire pressure value is obtained by adding the estimated combined tire pressure of each tire to its corresponding actual calibration value.
[0013] Preferably, the continuous rolling update and optimization specifically refers to: Parameter updates are triggered when the vehicle is in a stable straight-line, constant-speed driving state, the reference sensor data is reliable, and the calibration is fully effective. Using the current calibration error as the loss, the parameters of the weight mapping function for the current scenario are optimized by backpropagation using the gradient descent method; Based on the latest instantaneous system bias, the smoothed system bias stored in the scenario parameter set is updated using an incremental learning approach; Increase the sample count for this scenario and recalculate the confidence level for this scenario accordingly.
[0014] The tire pressure automatic calibration system based on single-point reference and scenario self-learning includes: The signal acquisition module is used to acquire the direct tire pressure signal of a single specified tire, the wheel speed signal of each wheel, the triaxial acceleration signal of the vehicle inertial measurement unit, and the ambient temperature signal. The scenario recognition module is used to extract multi-dimensional feature vectors based on the collected signal data and determine the current driving scenario mode of the vehicle in real time. A scenario fingerprint database is used to store a set of calibration parameters that correspond one-to-one with different driving scenario modes; The core processing module includes a dual-model parallel computing unit and an adaptive fusion unit. The dual-model parallel computing unit is used to calculate the relative tire pressure estimates of each tire based on the resonant frequency principle and the rolling radius principle, respectively. The adaptive fusion unit is used to dynamically determine the fusion weight based on the identified current scenario mode and real-time signal quality, and to perform weighted fusion of the dual-model outputs to obtain the fused tire pressure estimates of the tires other than the reference wheel and the single specified tire. The reference calibration and output unit is used to calculate and apply systematic deviations to calibrate the fused tire pressure estimate based on the direct measurement value of the reference wheel, and output the final tire pressure value. The self-learning engine is used to continuously update and optimize the corresponding calibration parameter set in the scenario fingerprint database using the current calibration data when conditions are met.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The tire pressure automatic calibration method and system proposed in this invention, based on single-point benchmark and scenario self-learning, achieves high-precision and high-reliability tire pressure monitoring while significantly reducing system costs through innovative hardware and software co-design. It can automatically identify and adapt to a wide variety of real-world driving scenarios, such as different road conditions, loads, and temperature conditions, and establish a unique calibration fingerprint for each recurring scenario, thus breaking the limitations of traditional fixed-parameter calibration modes. By fusing a dual physical model of resonant frequency and rolling radius, and dynamically adjusting the fusion strategy based on real-time signal quality, it effectively improves the estimation robustness in complex environments. The unique self-learning mechanism enables the system to continuously evolve with use, automatically compensating for characteristic changes caused by tire wear and replacement, ensuring long-term accuracy. Attached Figure Description
[0016] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart illustrating the driving scenario matching and parameter calling process of the present invention; Figure 3 This is a schematic diagram illustrating the principle of the dual-model adaptive weight fusion algorithm of the present invention; Figure 4 This is a flowchart illustrating the interaction between the benchmark calibration and the rolling update of self-learning parameters in this invention. Figure 5 This is a flowchart of the core process of scene recognition and self-learning calibration of the present invention; Figure 6 This is a flowchart of the vehicle stable driving state determination logic of the present invention; Figure 7 Flowchart of a three-level reliability assessment architecture for direct tire pressure sensor data; Figure 8 This is a system framework diagram of the present invention. Detailed Implementation
[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0018] Reference Figure 1 As shown, the automatic tire pressure calibration method based on single-point reference and scenario self-learning includes the following steps: Collect raw vehicle signals, including: direct tire pressure signal of a single designated tire (right front tire). Wheel speed signals for each wheel to The three-axis acceleration signal and ambient temperature signal from the vehicle inertial measurement unit (IMU).
[0019] Multi-dimensional features are extracted from the collected signal data to determine the current driving scenario mode of the vehicle in real time; The method for extracting multi-dimensional features based on the collected signal data is as follows: Based on the acquired raw signal data, the wheel speed signal, triaxial acceleration signal, and ambient temperature signal are preprocessed and segmented, including: dividing the data into segments using a fixed-duration sliding window mechanism, with the sliding window length... seconds, sliding step To ensure data continuity and complete driving segments, each signal is filtered, compensated, and missing value processed. The wheel speed signal is filtered with median to remove abnormal pulses, and zero-filling is used to handle data loss caused by CAN bus frame dropping. The IMU signal has the gravitational acceleration component removed and the coordinate system is transformed to the vehicle coordinate system. The temperature signal is filtered with a first-order low-pass filter to smooth instantaneous fluctuations. Primary statistical features are extracted from the preprocessed signal, including the average vehicle speed of the vehicle speed profile. (Unit: km / h) Standard deviation of vehicle speed Maximum acceleration (unit m / s) 2 Maximum deceleration Mean absolute value of lateral acceleration (unit m / s) 2 ), maximum value Standard deviation Longitudinal acceleration standard deviation ,energy The total energy of wheel speed signal in the 8-25Hz frequency band , main frequency (Unit: Hz) Average ambient temperature (Unit: °C) Temperature change rate (Unit: °C / min) These characteristics describe the effects of driving intensity and road conditions, number of curves and steering habits, frequency of start-stop, road surface roughness, and environment on tire physical properties, respectively. The extracted primary statistical features are nonlinearly combined to generate advanced fusion features, including the stability index, curve load factor, pavement excitation level, and thermal state coefficient. A value close to 1 indicates a very stable condition (such as cruising), while a value close to 0 indicates drastic changes (such as traffic congestion). (Curve load factor) It comprehensively characterizes the strength and speed of curves, distinguishes between high-speed curves and low-speed U-turns, and assesses the road surface excitation level. By normalizing the vibration energy to vehicle speed, the increased vibration caused by high vehicle speed alone is eliminated, providing a purer reflection of road surface quality and thermal state coefficient. A comprehensive assessment of the potential impact of both the absolute value and rate of temperature change on tire pressure is conducted. Integrating the aforementioned primary statistical features and advanced fusion features, a 12-dimensional feature vector representing the driving scenario is constructed for each time window: .
[0020] The real-time determination of the vehicle's current driving scenario mode specifically refers to: Reference Figure 2 As shown, the multidimensional feature vector generated based on the current time window Calculate its anchor points with each existing scenario feature vector in the scenario fingerprint database. The weighted Euclidean distance between them, the formula for weighted Euclidean distance is: in The feature weights are set based on the importance of the features. , , High weight, Let i be the i-th dimension of the current feature vector. Let i be the i-th feature of the j-th scenario anchor point; Using the minimum weighted distance and a preset distance threshold The comparison results serve as the initial discrimination criteria: if the minimum distance is less than the threshold, it is determined that a match has been found with the corresponding existing scenario, and the calibration parameter set of that scenario is directly called; if the minimum distance is greater than or equal to the threshold, the fine discrimination process is triggered. The detailed discrimination process is as follows: First, the current feature vector is classified into the macro-driving category according to preset rules; if... and Then it is marked as a highway. and If the value is greater than 0.5, it is marked as urban congestion. If the value exceeds a high threshold, it is marked as a severe road condition. This step provides context for subsequent analysis. Then, the density of historical data points in the local neighborhood of the current feature vector is statistically analyzed within the feature space, with the current feature vector as the center and a radius of... The number of statistical historical feature points within the hypersphere ; Based on local density and preset density threshold The comparison results are used for final determination: if the density is not lower than the threshold, it is determined to be a transitional state of a known scenario, and the parameters of the K nearest neighbor scenarios in the region are fused for this calculation; if the density is lower than the threshold, it is determined to be a completely new driving scenario. For each newly identified driving scenario, a new entry is created in the scenario fingerprint database, using the current feature vector as the anchor point. Initialize the weight function parameters. Initialize the parameters to those of a mature scenario with a similar macroscopic category or the global default parameters, with an offset. Initially a zero vector, confidence level Sample size It assigns default or macro-category-based calibration parameters to them.
[0021] Based on the principles of resonance frequency analysis and rolling radius calculation, the relative tire pressure estimates for each tire are calculated in parallel. Reference Figure 3 As shown, the steps include: Short-time Fourier transform was performed on the collected wheel speed signals of each wheel to extract the 13-18Hz resonant peak frequency characterizing the radial vibration of the tire. ; Based on the resonant peak frequency, current vehicle speed, and pre-calibrated tire equivalent mass and stiffness coefficient, the estimated tire pressure for each tire based on the resonant frequency principle is calculated. The resonant frequency model formula is as follows: ,in This refers to the equivalent mass of the tires and suspension. This is the tire stiffness coefficient (related to temperature and wear). For vehicle speed, this model is sensitive to minor air leaks and responds quickly, but is greatly affected by road surface excitation (roughness) and suspension condition. Based on the collected wheel speed signals and triaxial acceleration signals, the IMU is used to compensate for the differential speed of the left and right wheels caused by the vehicle's steering. Calculate the wheel speed ratio between coaxial or diagonally opposite wheels, and deduce the relative change in the corresponding tire rolling radius based on the wheel speed ratio; Based on the relative change in rolling radius and the physical parameters of the tire carcass, the estimated tire pressure for each tire based on the rolling radius principle is calculated. The rolling radius model formula is as follows: in For reference wheel speed, For the target wheel speed, The tire rolling radius at standard tire pressure. This refers to the Young's modulus of the tire carcass. The model is characterized by its tread thickness. It is less affected by road surface disturbances and is stable when driving in a straight line, but it is sensitive to load distribution, tire wear, and slippage.
[0022] The relative tire pressure estimates for each tire, calculated in parallel, are the ratios of each tire's tire pressure to the average tire pressure of all tires.
[0023] Based on the identified scenario patterns and real-time signal quality, the fusion weights are dynamically determined, and the tire pressure estimates calculated based on the above two principles are weighted and fused to obtain the fused tire pressure estimates for all tires except for a single specified tire. The dynamic determination of fusion weights specifically refers to: Obtain the weight mapping function parameters corresponding to the current recognition scenario pattern from the scenario fingerprint database. This parameter stores the weight function. Internal parameters (such as neural network weights, polynomial coefficients); Real-time calculation of signal-to-noise ratio including each wheel speed signal , lateral acceleration Ambient temperature Road surface vibration level Including the signal quality vector ; The real-time signal quality vector is input into the weight mapping function of the current scenario. In the middle, the real-time fusion weights between the output resonant frequency model and the rolling radius model are displayed. For example, in a highway cruising scenario, the road surface is smooth and the vibration is low, so the weighting function tends to give a higher weight to the resonant frequency model. In a bumpy urban road scenario, when a high level of vibration is detected, the weight of the resonant frequency model will be automatically reduced, and the rolling radius model will be relied on more.
[0024] Specifically, the method for obtaining the combined tire pressure estimate for all tires except the single specified tire is as follows: Obtain estimated tire pressure values for each tire based on the principle of resonant frequency. And the estimated tire pressure values for each tire calculated based on the rolling radius principle. ; Obtain the fusion weights dynamically determined based on the current scenario and real-time signal quality. ; The fusion weight is used as the weighting coefficient of the first model (resonance frequency model), and the value of one minus the fusion weight is used as the weighting coefficient of the second model (rolling radius model). Multiply the tire pressure estimate from the first model by its weighting coefficient, multiply the tire pressure estimate from the second model by its weighting coefficient, and add the two products together to obtain the combined tire pressure estimate for the corresponding tire. The combined formula is as follows: Based on the direct measurement of a single specified tire, calculate the systematic deviation between it and the fusion tire pressure estimate of the other tires excluding the single specified tire, and use the deviation to calibrate the fusion estimate to output the final tire pressure value. The systematic deviation between the calculated tire pressure estimate and the combined tire pressure estimate of all tires except for the single specified tire is specifically as follows: Calculate the ratio of the tire pressure of each tire obtained based on the fusion algorithm to the average tire pressure of all tires: in to , This step changes the comparison benchmark from "absolute atmospheric pressure" to "average pressure of four wheels," highlighting the differences between tires; Direct measurements of a single specified tire (right front wheel) Based on this ratio, the theoretical tire pressure values for the remaining tires are calculated: in The formula is based on the core physical assumption that, under a certain scenario of stable driving, the relative tire pressure ratio of each tire should be close to its true relative pressure ratio. The difference between the estimated tire pressure of each tire and its corresponding theoretical tire pressure is calculated as the instantaneous system bias. ; The instantaneous system deviation is exponentially smoothed to obtain the smoothed system deviation. ,in The smoothing factor (e.g., 0.9) has a larger value, the greater the historical inertia and the slower the response to new deviations. The smoothed system deviation is stored in the parameter set corresponding to the current scenario and serves as one of the core parameters of the scenario fingerprint.
[0025] Reference Figure 4 As shown, the specific steps for calibrating the fusion estimate using this deviation and outputting the final tire pressure value are as follows: Obtain the smoothed system bias corresponding to the current scenario. ; Dynamically calculate the comprehensive coefficient used to determine the validity of the deviation. This coefficient is based on the health status of the benchmark sensor. Confidence of scenario matching and the timeliness of the deviation itself jointly determined, The health status of the benchmark sensor is checked. The mutation rate and signal loss determination within a short time window are determined by the scene matching confidence score, which is taken from the output of the scene recognition module. The timeliness of the deviation is calculated based on the number of learning samples since the last update. Multiply the smoothing system deviation by the comprehensive coefficient to obtain the calibration amount for practical application; The final calibrated tire pressure output value is obtained by adding the combined estimated tire pressure of each tire to its corresponding actual calibration value. ,when Output high-confidence tire pressure values when Output the medium confidence tire pressure value when When a low-confidence tire pressure value is output and a "system self-check suggestion" prompt is triggered, the system determines that the calibration is unreliable and directly outputs the original fusion value.
[0026] Under the condition that the vehicle can drive stably and the reference data is reliable, the calibration data under the current scenario is used as a sample, and the calibration parameters corresponding to the scenario are continuously updated and optimized.
[0027] Reference Figure 5 As shown, the continuous rolling updates and optimizations specifically refer to: Parameter updates are triggered when the vehicle is in a stable straight-line, constant-speed driving state, the reference sensor data is reliable, and the calibration is fully effective. Reference Figure 6 As shown, the vehicle must maintain a stable driving state for more than 20 seconds, and all core conditions must be met (the mean absolute value of lateral acceleration and the mean absolute value of longitudinal acceleration meet the requirements, and the average vehicle speed...). The reference tire pressure signal has no frame drops and the standard deviation of fluctuation is [not specified]. Stability overall score point( Deduction items, Points are awarded for factors including vehicle speed stability, road surface smoothness, steering wheel flexibility, and load fluctuation indication. Reference Figure 7 As shown, the reliability of the reference sensor data must meet the overall reliability score. The score is calculated using a three-tiered progressive architecture: real-time signal quality diagnostics, short-term trend rationality analysis, and long-term consistency verification; calibration is fully effective upon completion. ; Using the current calibration error as the loss The parameters of the weight mapping function for the current scenario are optimized through backpropagation using gradient descent, with parameter update amount... ,in The learning rate is the parameter of the updated weight mapping function. ; Based on the latest instantaneous system bias, the smoothed system bias stored in the scenario parameter set is updated incrementally using the following formula: The constraints are ,in To maximize the single update step size and prevent drastic parameter changes due to a single severe shock, With the number of scenario samples Inversely proportional ( The more samples, the more conservative the updates; Increase the sample count for this scenario. And based on this, the confidence level of the scenario is recalculated. The more samples there are, the higher the confidence level, and the smaller the parameter update amplitude during self-learning, the more stable the result.
[0028] Example Implementation Scenario: Assume a vehicle frequently travels fully loaded on highways during Monday morning rush hour (Scenario A) and empty in urban areas during Friday evening rush hour (Scenario B). In the initial learning phase, after several trips, the system automatically clusters and creates "fingerprints" for Scenario A and Scenario B. For example, it learns that under Scenario A, due to rightward shift caused by load, the rolling radius model weight of the right rear wheel needs to be automatically increased. When re-entering the highway, the system identifies it as Scenario A, immediately retrieves the corresponding parameter set from the fingerprint database for fusion calculation, and outputs accurate tire pressure after calibration using the right front wheel reference value. If tires are changed under Scenario A, the system will initially detect the deviation. As the changes occur, the self-learning engine slowly updates the fingerprint of scenario A using new data, eventually adapting to the new tire characteristics without requiring manual reset.
[0029] Reference Figure 8As shown, the tire pressure automatic calibration system based on single-point reference and scenario self-learning includes: The signal acquisition module is used to acquire the direct tire pressure signal of a single designated tire (right front wheel). Wheel speed signals for each wheel to The three-axis acceleration signal and ambient temperature signal from the vehicle inertial measurement unit; The scene recognition module is used to extract a 12-dimensional feature vector from the collected signal data and calculate the current feature vector. With each existing scenario anchor point in the scenario fingerprint database Weighted Euclidean distance The system performs initial discrimination and triggers a fine discrimination process when no match is found. It determines the current driving scenario mode based on macro-category classification and local density test. This module is the "perception brain" of the system, and its design directly determines the accuracy and intelligence of the system's adaptive calibration. The scenario fingerprint database stores calibration parameter sets that correspond one-to-one with different driving scenario modes. Each parameter set contains the weight mapping function parameters for that scenario. Reference offset Confidence level Sample size Weight mapping function parameters Save weight function Internal parameters, reference offset Record the historical average tire pressure deviation between the indirect measurement wheel and the reference wheel under this scenario for rapid initial calibration. The confidence level and sample size reflect the fingerprint maturity. The core processing module includes a dual-model parallel computing unit and an adaptive fusion unit. The dual-model parallel computing unit is used to calculate the relative tire pressure estimate of each tire based on the resonant frequency principle and the rolling radius principle, respectively. The resonant frequency model extracts the 13-18Hz resonant peak frequency of the wheel speed signal. Combined with the equivalent mass of the tire Stiffness coefficient and vehicle speed calculate The rolling radius model compensates for the effects of steering differential, calculates the wheel speed ratio, and converts the change in rolling radius, combining this with tire physical parameters (rolling radius under standard tire pressure). Young's modulus of the fetus tread thickness )calculate The adaptive fusion unit is used to determine the current scenario pattern and real-time signal quality vector based on the identified current scenario pattern. Dynamically determine fusion weights Through formula The outputs of the two models are weighted and fused to obtain the fused tire pressure estimates for all tires except the reference wheel and the single specified tire. The dynamic weight adjustment logic is the core protection of this algorithm. The reference calibration and output unit is used to measure values directly from the reference wheel. Based on this, the relative tire pressure ratio, theoretical tire pressure value, and instantaneous system deviation of each tire are calculated. After exponential smoothing, a smoothed system deviation is obtained. A comprehensive coefficient for the effectiveness of the deviation is dynamically calculated. The smoothed system deviation is multiplied by the comprehensive coefficient to obtain the calibration value. The merged tire pressure estimate is then calibrated, and the final tire pressure value is output. Different intervals are marked with different confidence levels. This unit is the system's "accuracy arbiter," which directly determines the absolute accuracy and reliability of the system output. A self-learning engine is used to ensure stable vehicle operation. (Points), Reliability of reference sensor data () (points) and calibration is fully effective. When the condition is met, the parameters of the weight mapping function are optimized using the gradient descent method with the current calibration data. ), updating the smoothed system bias using incremental learning ( Increase the number of scenario samples. And recalculate the confidence level The corresponding calibration parameter set in the scenario fingerprint database is continuously updated and optimized to achieve continuous self-optimization of parameters, making the system "more accurate the more it is used".
[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. A tire pressure automatic calibration method based on single-point reference and scenario self-learning, characterized in that, Includes the following steps: S1. Collect vehicle raw signals, including: direct tire pressure signal of a single designated tire of the vehicle, wheel speed signal of each wheel, triaxial acceleration signal and ambient temperature signal; S2. Extract multi-dimensional features based on the collected signal data to determine the current driving scenario mode of the vehicle in real time; S3. Based on the principle of resonance frequency analysis and the principle of rolling radius calculation, the relative tire pressure estimates of each tire are calculated in parallel. S4. Based on the identified scenario pattern and real-time signal quality, dynamically determine the fusion weights, and perform weighted fusion on the tire pressure estimates calculated based on the resonant frequency analysis principle and the rolling radius calculation principle to obtain the fused tire pressure estimates for all tires except for a single specified tire. S5. Based on the direct measurement of a single specified tire, calculate the systematic deviation between it and the fusion tire pressure estimate of the other tires excluding the single specified tire, and use the deviation to calibrate the fusion estimate and output the final tire pressure value. S6. Under the condition that the vehicle can drive stably and the reference data is reliable, the calibration data under the current scenario is used as a sample to continuously update and optimize the calibration parameters corresponding to the scenario.
2. The single reference point and context self-learning based tire pressure automatic calibration method according to claim 1, wherein, The method for extracting multi-dimensional features based on the collected signal data is as follows: Based on the collected raw signal data, the wheel speed signal, triaxial acceleration signal and ambient temperature signal are preprocessed and segmented, including: dividing the data into segments using a fixed-duration sliding window mechanism, and filtering, compensating and handling missing values for each signal; Primary statistical features are extracted from the preprocessed signal, including the mean and standard deviation of the vehicle speed profile, the statistics of lateral and longitudinal acceleration, the vibration energy and dominant frequency of the wheel speed signal in a specific frequency band, and the average value and rate of change of the ambient temperature. The extracted primary statistical features are nonlinearly combined to generate advanced fusion features, including the stationarity index, curve compliance factor, pavement excitation level, and thermal state coefficient. By integrating the aforementioned primary statistical features and advanced fusion features, a multidimensional feature vector representing the driving scenario is constructed for each time window.
3. The automatic tire pressure calibration method based on single-point reference and scenario self-learning according to claim 2, characterized in that, The real-time determination of the vehicle's current driving scenario mode specifically refers to: Based on the multidimensional feature vector generated in the current time window, calculate the weighted distance between it and the anchor points of each existing scenario feature vector in the scenario fingerprint database; The comparison between the minimum weighted distance and the preset distance threshold is used as the initial discrimination criterion: if the minimum distance is less than the threshold, it is determined that the corresponding existing scenario is matched, and the calibration parameter set of that scenario is directly called; if the minimum distance is greater than or equal to the threshold, the fine discrimination process is triggered. The detailed discrimination process is as follows: first, the current feature vector is classified into the macro driving category according to the preset rules, and then the density of historical data points in the local neighborhood of the current feature vector is counted in the feature space. The final determination is made based on the comparison between the local density and the preset density threshold: if the density is not lower than the threshold, it is determined to be a transitional state of a known scenario, and the parameters of neighboring scenarios are fused for this calculation; if the density is lower than the threshold, it is determined to be a completely new driving scenario. For a newly identified driving scenario, a new entry is created in the scenario fingerprint database, initialized with the current feature vector as the anchor point, and assigned default or calibration parameters based on macro-class inheritance.
4. The automatic tire pressure calibration method based on single-point reference and scenario self-learning according to claim 3, characterized in that, S3 includes: The collected wheel speed signals of each wheel were transformed by time and frequency to extract the resonant peak frequency that characterizes the radial vibration of the tire. Based on the resonant peak frequency, current vehicle speed, and pre-calibrated tire equivalent mass and stiffness coefficient, the estimated tire pressure for each tire based on the resonant frequency principle is calculated. Based on the collected wheel speed signals and triaxial acceleration signals, compensation is made for the differential speed between the left and right wheels caused by vehicle steering. Calculate the wheel speed ratio between coaxial or focused wheels, and deduce the relative change in the corresponding tire rolling radius based on the wheel speed ratio; Based on the relative change in rolling radius and the physical parameters of the tire carcass, the estimated tire pressure for each tire based on the rolling radius principle is calculated.
5. The automatic tire pressure calibration method based on single-point reference and scenario self-learning according to claim 4, characterized in that, The relative tire pressure estimates for each tire calculated in parallel in S3 are the ratios of each tire's tire pressure to the average tire pressure of all tires.
6. The automatic tire pressure calibration method based on single-point reference and scenario self-learning according to claim 5, characterized in that, The dynamic determination of fusion weights specifically refers to: Obtain the weight mapping function parameters corresponding to the current recognition scenario pattern from the scenario fingerprint database; Real-time calculation of signal quality vectors including signal-to-noise ratio of each wheel speed signal, lateral acceleration, ambient temperature, and road vibration level; The real-time signal quality vector is input into the weight mapping function of the current scenario, and the real-time fusion weight between the resonant frequency model and the rolling radius model is output.
7. The automatic tire pressure calibration method based on single-point reference and scenario self-learning according to claim 6, characterized in that, Specifically, the method for obtaining the combined tire pressure estimate for all tires except the single specified tire is as follows: Obtain tire pressure estimates for each tire calculated based on the principle of resonant frequency, and tire pressure estimates for each tire calculated based on the principle of rolling radius; Obtain the fusion weights dynamically determined based on the current scenario mode and real-time signal quality; The fusion weights are used as the weighting coefficients of the first model, and the value of one minus the fusion weights is used as the weighting coefficients of the second model. Multiply the tire pressure estimate of the first model by its weighting coefficient, multiply the tire pressure estimate of the second model by its weighting coefficient, and add the two products together to obtain the combined tire pressure estimate for the corresponding tire.
8. The automatic tire pressure calibration method based on single-point reference and scenario self-learning according to claim 7, characterized in that, The systematic deviation between the calculated tire pressure estimate and the combined tire pressure estimate of all tires except for the single specified tire is specifically as follows: Calculate the ratio of the tire pressure of each tire obtained based on the fusion algorithm to the average tire pressure of all tires; Based on the direct measurement of a single specified tire, the theoretical tire pressure values of the remaining tires are calculated according to the ratio. The difference between the estimated fusion tire pressure of each tire and its corresponding theoretical tire pressure is calculated as the instantaneous system deviation. The instantaneous system deviation is subjected to exponential smoothing to obtain the smoothed system deviation, which is then stored in the parameter set corresponding to the current scenario. The process of using this deviation to calibrate the fusion estimate and outputting the final tire pressure value is as follows: Obtain the smoothed system bias corresponding to the current scenario; A comprehensive coefficient for determining the validity of the deviation is dynamically calculated. This coefficient is determined based on the health of the benchmark sensor, the confidence level of the scenario matching, and the timeliness of the deviation itself. Multiply the smoothing system deviation by the comprehensive coefficient to obtain the calibration amount for practical application; The final calibrated tire pressure value is obtained by adding the estimated combined tire pressure of each tire to its corresponding actual calibration value.
9. The automatic tire pressure calibration method based on single-point reference and scenario self-learning according to claim 8, characterized in that, The specific details of continuous rolling updates and optimizations are as follows: Parameter updates are triggered when the vehicle is in a stable straight-line, constant-speed driving state, the reference sensor data is reliable, and the calibration is fully effective. Using the current calibration error as the loss, the parameters of the weight mapping function for the current scenario are optimized by backpropagation using the gradient descent method; Based on the latest instantaneous system bias, the smoothed system bias stored in the scenario parameter set is updated using an incremental learning approach; Increase the sample count for this scenario and recalculate the confidence level for this scenario accordingly.
10. A tire pressure automatic calibration system based on single-point reference and scenario self-learning, characterized in that, include: The signal acquisition module is used to acquire the direct tire pressure signal of a single specified tire, the wheel speed signal of each wheel, the triaxial acceleration signal of the vehicle inertial measurement unit, and the ambient temperature signal. The scenario recognition module is used to extract multi-dimensional feature vectors based on the collected signal data and determine the current driving scenario mode of the vehicle in real time. A scenario fingerprint database is used to store a set of calibration parameters that correspond one-to-one with different driving scenario modes; The core processing module includes a dual-model parallel computing unit and an adaptive fusion unit. The dual-model parallel computing unit is used to calculate the relative tire pressure estimates of each tire based on the resonant frequency principle and the rolling radius principle, respectively. The adaptive fusion unit is used to dynamically determine the fusion weight based on the identified current scenario mode and real-time signal quality, and to perform weighted fusion of the dual-model outputs to obtain the fused tire pressure estimates of the tires other than the reference wheel and the single specified tire. The reference calibration and output unit is used to calculate and apply systematic deviations to calibrate the fused tire pressure estimate based on the direct measurement value of the reference wheel, and output the final tire pressure value. The self-learning engine is used to continuously update and optimize the corresponding calibration parameter set in the scenario fingerprint database using the current calibration data when conditions are met.