A machine learning based non-contact tire load prediction method and system
By integrating mechanical deformation and vehicle characteristics through a deep gradient boosting architecture, a high-precision tire load prediction model is established, which solves the problems of insufficient accuracy and generalization ability in existing technologies and achieves efficient and low-cost tire load prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 浙江交投高速公路运营管理有限公司
- Filing Date
- 2026-03-09
- Publication Date
- 2026-07-03
AI Technical Summary
Existing non-contact vehicle tire load prediction technologies lack accuracy and generalization ability when faced with diverse tire types and complex working conditions, making it difficult to meet the application requirements for high precision and real-time performance.
We employ a Deep Gradient Boosting architecture (DeepGBM) that integrates mechanical deformation, sidewall attributes, and vehicle classification features. We process sparse features through a deep neural network module to establish a comprehensive tire feature dataset and use a Bayesian optimization algorithm to optimize hyperparameters, thereby constructing a high-precision tire load prediction model.
It achieves high-precision tire load prediction for different vehicle models and operating conditions, improves the model's fitting and generalization capabilities, reduces system installation and maintenance costs, and adapts to real-world engineering scenarios.
Smart Images
Figure CN122333014A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method and system for predicting vehicle tire load, and more particularly to a non-contact method and system for predicting vehicle tire load based on machine learning. Background Technology
[0002] Currently, the rapid development of highway transportation networks and the continuous high-load operation of the logistics and transportation industry have made vehicle overloading a more prominent problem. Vehicle overloading not only drastically accelerates fatigue damage to road surfaces, bridges, and other infrastructure, shortening their service life and increasing social maintenance costs, but also poses a significant safety hazard, leading to serious traffic accidents. Therefore, developing an efficient, accurate, and large-scale deployable vehicle load identification technology is of crucial strategic importance for ensuring traffic network security, extending infrastructure lifespan, and achieving intelligent traffic management.
[0003] Current non-contact weighing methods infer vehicle weight by capturing the geometric deformation information of tires under load, offering significant advantages such as low cost and non-invasiveness. Existing technologies largely rely on classical mechanical models based on physical parameters to predict the contact force between the tire and the road surface. While these models have clear physical meanings, they suffer from cumbersome parameter calibration and insufficient generalization ability when faced with diverse tire types, complex vehicle operating conditions, and environmental changes in the real world, making it difficult to meet the demands of high-precision, real-time applications. Meanwhile, some purely data-driven machine learning methods, while avoiding complex physical modeling, often utilize only single deformation features, failing to encompass specific vehicle and tire type information. This results in the model's inability to distinguish the different responses of tires with different structures and tire pressures under the same load, thus limiting prediction accuracy and model generalization ability. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a non-contact tire load prediction method and system based on machine learning. This non-contact tire load prediction method and system integrates multiple features such as mechanical deformation, sidewall attributes and vehicle classification through the Deep Gradient Boosting (DeepGBM) architecture, and achieves high-precision prediction of tire load under different vehicle models and different working conditions.
[0005] To solve the above-mentioned technical problems, the first technical solution adopted by the present invention is: a non-contact tire load prediction method based on machine learning, comprising the following steps: Step 1: Establish a tire load prediction model. The tire load prediction model adopts a deep gradient boosting architecture, which includes a lightweight gradient boosting module for processing dense features and a deep neural network module for processing sparse features. Step 2: Establish a comprehensive tire feature dataset for training the tire load prediction model. The comprehensive tire feature dataset includes tire feature vectors and corresponding real load values for light vehicles and heavy trucks under laboratory static load, outdoor static load, and outdoor dynamic conditions, respectively. The tire feature vectors include mechanical deformation feature parameters, sidewall identifier feature parameters, and vehicle type classification feature parameters. Step 3: Train the tire load prediction model based on the comprehensive tire feature dataset to obtain the trained tire load prediction model. The trained tire load prediction model takes tire features as input and the predicted tire load value as output. Step 4: Set up a tire load prediction area, collect tire sidewall image data of light vehicles and / or heavy trucks passing through the tire load prediction area in real time without contact, and extract the corresponding real-time tire feature vectors. Step 5: Input the real-time tire features into the trained tire load prediction model and output the corresponding predicted tire load value; Step 6: Display the prediction results in real time through a data visualization device, and issue an alarm for loads exceeding a preset threshold.
[0006] Furthermore, the tire features in step 2 include mechanical deformation feature parameters reflecting the tire's compressive state, sidewall identifier feature parameters reflecting the tire's properties, and vehicle type classification feature parameters.
[0007] Furthermore, the specific process of feature extraction in step 4 is as follows: Step 4.1: Use computer vision algorithms to identify and calculate mechanical deformation characteristic parameters from the vehicle tire sidewall image data. The mechanical deformation characteristic parameters include: tire pixel radius R, rim pixel radius r, tire vertical deflection δ, tire center to ground distance C, tire-road contact length L, theoretical segmentation length D based on sphere-plane contact model, and tire sidewall projected area S. Step 4.2: Use optical character recognition technology to identify and calculate the sidewall identifier feature parameters from the vehicle tire sidewall image data. The sidewall identifier feature parameters include: tire section height H, tire section width W, and tire inflation pressure P. Step 4.3: Determine the vehicle type based on the recognition results of the optical character recognition technology in Step 4.2, and encode the vehicle type classification feature parameters as numerical labels as vehicle type classification feature parameters T: set cars and SUVs as label 0, light trucks as label 1, and heavy trucks and buses as label 2; Step 4.4: Collect the feature parameters obtained in steps 4.1 to 4.3 respectively to form a real-time tire feature vector.
[0008] Furthermore, the specific process for establishing the comprehensive tire feature dataset in step 2 is as follows: Step 2.1: Laboratory light vehicle data acquisition: Apply static load using actuators, and collect tire feature vector data of multiple light vehicle tires under loading levels in conjunction with a dynamic weighing system and linear displacement sensors; Step 2.2: Outdoor heavy-duty truck data collection: This is divided into two parts: static and dynamic. The static part collects tire feature vector data of different truck models under empty, half-loaded, fully loaded and different tire pressure combinations. The dynamic part uses a dynamic weighing system in toll station lanes to obtain tire feature vector data of low-speed passing vehicles. Step 2.3: Clean and merge the data obtained in Step 2.1 and Step 2.2 to form the comprehensive tire feature dataset.
[0009] Furthermore, the specific process in step 5 is as follows: Step 5.1: Input the mechanical deformation feature parameters and sidewall identifier feature parameters in the real-time tire features as dense numerical features into the lightweight gradient boosting module in the trained tire load prediction model, and learn nonlinear feature transformation through gradient boosting decision tree; Step 5.2: The vehicle classification feature parameters in the real-time tire features are used as sparse category features and input into the deep neural network module of the trained tire load prediction model. High-dimensional feature interactions are learned through the embedding layer and the fully connected layer. Step 5.3: The outputs of the lightweight gradient boosting module and the deep neural network module are weighted and fused to obtain the final predicted tire load value.
[0010] Furthermore, the model training process in step 3 is as follows: The hyperparameters of the tire load prediction model are iteratively optimized using a Bayesian optimization algorithm, and the tire load prediction model with the optimal hyperparameters is generated based on a cross-validation strategy as the trained tire load prediction model.
[0011] Furthermore, the Bayesian optimization algorithm's optimization process specifically includes: Step 3.1: Divide the comprehensive tire feature dataset into a training set and a test set, and use a k-fold cross-validation strategy on the training set; Step 3.2: Set the objective function as the average coefficient of determination of k-fold cross-validation; Step 3.3: Use the Bayesian optimization algorithm to jointly optimize the number of trees, maximum depth, and learning rate of the lightweight gradient boosting module, as well as the embedding dimension, learning rate, and hidden layer structure of the deep neural network module. Step 3.4: When the objective function converges or reaches the preset number of iterations, output the optimal hyperparameter combination and solidify it into the tire load prediction model with the optimal hyperparameters.
[0012] To solve the above-mentioned technical problems, the second technical solution proposed by the present invention is: a non-contact tire load prediction system based on machine learning, including an image acquisition component, a data processing device, and a data visualization device.
[0013] The image acquisition component includes an industrial camera unit and a trigger control unit; The data processing device includes a storage unit, a feature extraction program, and a load prediction program. The storage unit is used to store the side image of the tire to be detected captured by the image acquisition component and the pre-trained optimal tire load prediction model; the feature extraction program extracts feature vectors according to the method described in claim 3 and transmits them to the load prediction program; the load prediction program calls the optimal model to perform calculations according to the method described in claim 5, calculates the predicted tire load value, and transmits it to the data visualization device.
[0014] Furthermore, the functions of the data visualization device include: Real-time display of high-definition images of the tire sidewalls of the vehicle under test; The predicted tire load values are displayed in numerical form, and the extracted key mechanical deformation parameters and sidewall identifier parameters are listed simultaneously. Historical data query and report generation, recording the time of each detection, predicted load, and vehicle image; When the detected load value exceeds the legal or preset limit, an audible and visual alarm is automatically triggered.
[0015] This invention offers the following advantages: 1. It innovatively integrates tire mechanical deformation characteristics, sidewall identifier characteristics, and vehicle classification characteristics. Through a DeepGBM hybrid architecture, a Deep Neural Network (DNN) module is used to specifically handle sparse features such as vehicle label (0, 1, 2), effectively capturing the differences in tire stiffness characteristics across different vehicle models and significantly improving the fitting ability and prediction accuracy for complex nonlinear relationships (R² can reach 0.964). 2. This invention constructs a comprehensive dataset containing high-precision laboratory data, outdoor static heavy-load data, and outdoor dynamic measured data, covering a wide range of vehicle models from light vehicles to heavy trucks and various operating conditions. This results in a trained model with strong generalization ability and robustness, adapting to real-world engineering scenarios. 3. The non-contact tire load prediction method and system based on machine learning proposed in this invention achieves fully automated and non-contact detection, eliminating the need for pre-embedded sensors on the road surface, greatly reducing system installation and maintenance costs, and not affecting normal traffic, thus demonstrating promising engineering application prospects. Attached Figure Description
[0016] Figure 1 This is a flowchart of the non-contact tire load prediction method in an embodiment of the present invention; Figure 2 This is a schematic diagram of tire mechanical deformation and sidewall identifier features in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the composition and construction of a comprehensive dataset in an embodiment of the present invention; Figure 4 This is a schematic diagram of the laboratory static load test setup and measurement in an embodiment of the present invention; Figure 5 This is a schematic diagram of the test setup for dynamic data acquisition of trucks at toll stations in an embodiment of the present invention; Figure 6 This is a hyperparameter optimization process diagram of the lightweight gradient boosting module in the tire load prediction model in this embodiment of the invention; Figure 7 This is a diagram illustrating the hyperparameter optimization process of the deep neural network module in the tire load prediction model in this embodiment of the invention. Figure 8 This is a training loss convergence curve of the deep neural network module in this embodiment of the invention; Figure 9 This is a scatter plot of the prediction performance of the tire load prediction model trained in this embodiment of the invention. Figure 10 This is a diagram of the user interface for system data visualization in an embodiment of the present invention. Detailed Implementation
[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific preferred embodiments.
[0018] In the description of this invention, it should be understood that the terms "left side," "right side," "upper part," "lower part," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. "First," "second," etc., do not indicate the importance of the components, and therefore should not be construed as a limitation of this invention. The specific dimensions used in this embodiment are only for illustrating the technical solution and do not limit the scope of protection of this invention.
[0019] like Figure 1 As shown, a non-contact tire load prediction method based on machine learning includes the following steps: Step 1: Establish a tire load prediction model. The tire load prediction model adopts a deep gradient boosting architecture, which includes a lightweight gradient boosting module for handling dense features and a deep neural network module for handling sparse features. Step 2: Establish a comprehensive tire feature dataset for training the tire load prediction model. The comprehensive tire feature dataset includes tire feature vectors and corresponding real load values for light vehicles and heavy trucks under laboratory static load, outdoor static load, and outdoor dynamic conditions, respectively. The tire feature vectors include mechanical deformation feature parameters reflecting the tire's compression state, sidewall identifier feature parameters reflecting tire attributes, and vehicle type classification feature parameters.
[0020] The specific process for establishing a comprehensive tire feature dataset is as follows: Step 2.1: Laboratory light vehicle data acquisition: Apply static load using actuators, and collect tire feature vector data of multiple light vehicle tires under loading levels in conjunction with a dynamic weighing system and linear displacement sensors; Step 2.2: Outdoor heavy-duty truck data collection: This is divided into two parts: static and dynamic. The static part collects tire feature vector data of different truck models under empty, half-loaded, fully loaded and different tire pressure combinations. The dynamic part uses a dynamic weighing system in toll station lanes to obtain tire feature vector data of low-speed passing vehicles. Step 2.3: Clean and merge the data obtained in Step 2.1 and Step 2.2 to form a comprehensive tire feature dataset.
[0021] In this embodiment, as Figure 3 As shown, this embodiment obtained a total of 2312 sets of valid sample data, specifically including: Data acquisition from a light vehicle in the laboratory. (Example) Figure 4 As shown, nine different tire specifications covering car and SUV models were selected. A vertical load was applied using an actuator, the true value of the contact force was measured by WIMs, the displacement was measured by LVDT, and 648 high-precision light vehicle samples were finally obtained through synchronous shooting.
[0022] Outdoor heavy-duty truck data collection. The static part selected 10 different models of two-axle trucks, combining 1 / 4 load, 1 / 2 load, and full load conditions, along with multiple tire pressure levels, setting 12 test conditions; the dynamic part... Figure 5 As shown, the test was conducted in a specific lane at the Nanjing South toll station. Test trucks passed at a constant speed of 2-6 km / h, and dynamic axle loads were obtained using WIMs. After data cleaning and filtering, a dynamic load dataset for trucks containing 866 valid samples was formed. Through the above steps, a comprehensive dataset of 2312 sets was constructed, covering a wide range of data.
[0023] Step 3: Train the tire load prediction model based on the comprehensive tire feature dataset to obtain the trained tire load prediction model. The trained tire load prediction model takes tire features as input and the predicted tire load value as output. The model training process is as follows: The hyperparameters of the tire load prediction model are iteratively optimized using a Bayesian optimization algorithm. The tire load prediction model with the optimal hyperparameters is generated based on a cross-validation strategy and used as the trained tire load prediction model.
[0024] The Bayesian optimization algorithm's optimization process specifically includes: Step 3.1: Divide the comprehensive tire feature dataset into a training set and a test set, and use a k-fold cross-validation strategy on the training set; Step 3.2: Set the objective function as the average coefficient of determination of k-fold cross-validation; Step 3.3: Use the Bayesian optimization algorithm to jointly optimize the number of trees, maximum depth, and learning rate of the lightweight gradient boosting module, as well as the embedding dimension, learning rate, and hidden layer structure of the deep neural network module. Step 3.4: When the objective function converges or reaches the preset number of iterations, output the optimal hyperparameter combination and solidify it into the tire load prediction model with the optimal hyperparameters.
[0025] In this embodiment, a 5-fold cross-validation strategy is employed on the training set. A Bayesian optimization method is used to optimize the hyperparameters, with the objective function being the average coefficient of determination (R²) of the 5-fold cross-validation. For the LightGBM module, the optimized hyperparameters include the number of trees, maximum depth, and learning rate, and the optimization process is as follows: Figure 6 As shown; for the DNN module, the optimized hyperparameters include embedding dimension, learning rate, and hidden layer structure, and the optimization process is as follows. Figure 7 As shown. After sufficient iterative training, the training and testing loss curves of the DNN module are as follows. Figure 8 As shown, it exhibits good convergence. The prediction performance of the final optimal model on the test set is as follows: Figure 9 As shown, the coefficient of determination R² reaches 0.964, proving the high accuracy of the model.
[0026] Step 4: Set up the tire load prediction area, collect the tire sidewall image data of light vehicles and / or heavy trucks passing through the tire load prediction area in real time without contact, and extract the corresponding real-time tire feature vector. The specific process of feature extraction is as follows: Step 4.1: Use computer vision algorithms to identify and calculate mechanical deformation feature parameters from vehicle tire sidewall image data. The mechanical deformation feature parameters include: tire pixel radius R, rim pixel radius r, tire vertical deflection δ, tire center to ground distance C, tire-road contact length L, theoretical segmentation length D based on sphere-plane contact model, and tire sidewall projected area S. Step 4.2: Use optical character recognition technology to identify and calculate the sidewall identifier feature parameters from the vehicle tire sidewall image data. The sidewall identifier feature parameters include: tire section height H, tire section width W, and tire inflation pressure P. Step 4.3: Determine the vehicle type based on the recognition results of the optical character recognition technology in Step 4.2, and encode the vehicle type classification feature parameters as numerical labels as vehicle type classification feature parameters T: set cars and SUVs as label 0, light trucks as label 1, and heavy trucks and buses as label 2; Step 4.4: Collect the feature parameters obtained from Steps 4.1 to 4.3 to form a real-time tire feature vector.
[0027] In this embodiment, mechanical deformation characteristics are taken as follows: tire radius R, rim radius r, vertical deflection δ (δ=RC), contact length L, etc. Figure 2 As shown, OCR technology is used to extract sidewall identifier features: cross-sectional width W, cross-sectional height H, tire pressure P, etc. Based on the OCR recognition results, the vehicle type is determined and coded: if determined to be a car or SUV, the feature column T label is set to 0; if determined to be a light truck, the feature column T label is set to 1; if determined to be a heavy truck or bus, the feature column T label is set to 2.
[0028] Step 5: Input the real-time tire features into the trained tire load prediction model and output the corresponding predicted tire load value. The specific process in step 5 is as follows: Step 5.1: Input the mechanical deformation feature parameters and sidewall identifier feature parameters from the real-time tire features into the lightweight gradient boosting module of the trained tire load prediction model as dense numerical features, and learn nonlinear feature transformation through gradient boosting decision tree; Step 5.2: Take the vehicle classification feature parameters in the real-time tire features as sparse category features and input them into the deep neural network module in the trained tire load prediction model. Learn the high-dimensional feature interaction through the embedding layer and the fully connected layer. Step 5.3: Weight and fuse the outputs of the lightweight gradient boosting module and the deep neural network module to obtain the final predicted tire load value.
[0029] Step 6: Display the prediction results in real time through a data visualization device, and issue an alarm for loads exceeding a preset threshold. This embodiment also includes a non-contact tire load prediction system based on machine learning, comprising an image acquisition component, a data processing device, and a data visualization device.
[0030] The image acquisition component includes an industrial camera unit and a trigger control unit; The data processing device includes a storage unit, a feature extraction program, and a load prediction program; The storage unit stores the sidewall image of the tire to be detected captured by the image acquisition component and the pre-trained optimal tire load prediction model; the feature extraction program extracts feature vectors and transmits them to the load prediction program; the load prediction program calls the optimal model to perform calculations, obtains the predicted tire load value, and transmits it to the data visualization device. For example... Figure 10 As shown, the data visualization device has the following functions: real-time display of high-definition images of the tire sidewalls of the vehicle under test; display of predicted tire load values in digital form, and simultaneously list the extracted key mechanical deformation parameters and sidewall identifier parameters; historical data query and report generation, recording the time of each test, predicted load and vehicle image; and automatically triggering an audible and visual alarm when the detected load value exceeds the legal or preset limit.
[0031] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.
Claims
1. A non-contact tire load prediction method based on machine learning, characterized in that: Includes the following steps: Step 1: Establish a tire load prediction model. The tire load prediction model adopts a deep gradient boosting architecture, which includes a lightweight gradient boosting module for processing dense features and a deep neural network module for processing sparse features. Step 2: Establish a comprehensive tire feature dataset for training the tire load prediction model. The comprehensive tire feature dataset includes tire feature vectors and corresponding real load values for light vehicles and heavy trucks under laboratory static load, outdoor static load, and outdoor dynamic conditions, respectively. Step 3: Train the tire load prediction model based on the comprehensive tire feature dataset to obtain the trained tire load prediction model. The trained tire load prediction model takes the tire feature vector as input and the predicted tire load value as output. Step 4: Set up a tire load prediction area, collect tire sidewall image data of light vehicles and / or heavy trucks passing through the tire load prediction area in real time without contact, and extract the corresponding real-time tire feature vectors. Step 5: Input the real-time tire feature vector into the trained tire load prediction model and output the corresponding predicted tire load value; Step 6: Display the prediction results in real time through a data visualization device, and issue an alarm for loads exceeding a preset threshold.
2. The non-contact tire load prediction method according to claim 1, characterized in that: The tire feature vector in step 2 includes mechanical deformation feature parameters reflecting the tire's compression state, sidewall identifier feature parameters reflecting tire attributes, and vehicle type classification feature parameters.
3. The non-contact tire load prediction method according to claim 2, characterized in that: The specific process of feature extraction in step 4 is as follows: Step 4.1: Use computer vision algorithms to identify and calculate mechanical deformation characteristic parameters from the vehicle tire sidewall image data. The mechanical deformation characteristic parameters include: tire pixel radius R, rim pixel radius r, tire vertical deflection δ, tire center to ground distance C, tire-road contact length L, theoretical segmentation length D based on sphere-plane contact model, and tire sidewall projected area S. Step 4.2: Use optical character recognition technology to identify and calculate the sidewall identifier feature parameters from the vehicle tire sidewall image data. The sidewall identifier feature parameters include: tire section height H, tire section width W, and tire inflation pressure P. Step 4.3: Determine the vehicle type based on the recognition results of the optical character recognition technology in Step 4.2, and encode the vehicle type classification feature parameters as numerical labels as vehicle type classification feature parameters T: set cars and SUVs as label 0, light trucks as label 1, and heavy trucks and buses as label 2; Step 4.4: Collect the feature parameters obtained in steps 4.1 to 4.3 respectively to form a real-time tire feature vector.
4. The non-contact tire load prediction method according to claim 2, characterized in that: The specific process for establishing the comprehensive tire feature dataset in step 2 is as follows: Step 2.1: Laboratory light vehicle data acquisition: Apply static load using actuators, and collect tire feature vector data of multiple light vehicle tires under loading levels in conjunction with a dynamic weighing system and linear displacement sensors; Step 2.2: Outdoor heavy-duty truck data collection: This is divided into two parts: static and dynamic. The static part collects tire feature vector data of different truck models under empty, half-loaded, fully loaded and different tire pressure combinations. The dynamic part uses a dynamic weighing system in toll station lanes to obtain tire feature vector data of low-speed passing vehicles. Step 2.3: Clean and merge the data obtained in Step 2.1 and Step 2.2 to form the comprehensive tire feature dataset.
5. The non-contact tire load prediction method according to claim 2, characterized in that: The specific process in step 5 is as follows: Step 5.1: Input the mechanical deformation feature parameters and sidewall identifier feature parameters in the real-time tire features as dense numerical features into the lightweight gradient boosting module in the trained tire load prediction model, and learn nonlinear feature transformation through gradient boosting decision tree; Step 5.2: The vehicle classification feature parameters in the real-time tire features are used as sparse category features and input into the deep neural network module of the trained tire load prediction model. High-dimensional feature interactions are learned through the embedding layer and the fully connected layer. Step 5.3: The outputs of the lightweight gradient boosting module and the deep neural network module are weighted and fused to obtain the final predicted tire load value.
6. The non-contact tire load prediction method according to claim 1, characterized in that: The model training process in step 3 is as follows: The hyperparameters of the tire load prediction model are iteratively optimized using a Bayesian optimization algorithm, and the tire load prediction model with the optimal hyperparameters is generated based on a cross-validation strategy as the trained tire load prediction model.
7. The non-contact tire load prediction method according to claim 6, characterized in that: The Bayesian optimization algorithm's optimization process specifically includes: Step 3.1: Divide the comprehensive tire feature dataset into a training set and a test set, and use a k-fold cross-validation strategy on the training set; Step 3.2: Set the objective function as the average coefficient of determination of k-fold cross-validation; Step 3.3: Use the Bayesian optimization algorithm to jointly optimize the number of trees, maximum depth, and learning rate of the lightweight gradient boosting module, as well as the embedding dimension, learning rate, and hidden layer structure of the deep neural network module. Step 3.4: When the objective function converges or reaches the preset number of iterations, output the optimal hyperparameter combination and solidify it into the tire load prediction model with the optimal hyperparameters.
8. A non-contact tire load prediction system based on machine learning, characterized in that: Includes image acquisition components, data processing devices, and data visualization devices: The image acquisition component includes an industrial camera unit and a trigger control unit; The data processing device includes a storage unit, a feature extraction program, and a load prediction program. The storage unit is used to store the side image of the tire to be detected captured by the image acquisition component and the pre-trained optimal tire load prediction model; the feature extraction program extracts feature vectors according to the method described in claim 3 and transmits them to the load prediction program; the load prediction program calls the optimal model to perform calculations according to the method described in claim 5, calculates the predicted tire load value, and transmits it to the data visualization device.
9. The non-contact tire load prediction system according to claim 8, characterized in that: The data visualization device has the following functions: Real-time display of high-definition images of the tire sidewalls of the vehicle under test; The predicted tire load values are displayed in numerical form, and the extracted key mechanical deformation parameters and sidewall identifier parameters are listed simultaneously. Historical data query and report generation, recording the time of each detection, predicted load, and vehicle image; When the load value is detected to exceed the statutory or preset limit, an audible and visual alarm is automatically triggered.