Brake performance evaluation method and system based on machine learning

Abstract

The invention provides a brake performance evaluation method and system based on machine learning, and belongs to the technical field of computers.The method comprises the steps that first brake data and second brake data are obtained, and the first brake data comprise brake abrasion degree data, brake execution state data and brake identification data; performing feature extraction and fusion on the brake execution state data and the brake wear degree data to generate a brake feature vector; selecting a pre-trained brake performance evaluation model corresponding to the brake identification data based on the brake identification data to obtain a target brake performance evaluation model; inputting the braking feature vector into a target braking performance evaluation model to obtain a primary evaluation result and confidence; and comparing the confidence coefficient with a preset first threshold value to obtain a comparison result, and constructing a multi-source information fusion correction factor library with the second braking data to calibrate the primary evaluation result to obtain a braking performance evaluation result. According to the invention, the efficiency and precision of brake performance evaluation are improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method and system for evaluating braking performance based on machine learning. Background Technology

[0002] With the development of intelligent and connected vehicles, the accuracy and real-time performance evaluation of vehicle braking systems, as core safety components, has become a key requirement for traffic safety management. Traditional braking performance testing relies heavily on offline static bench tests, which not only fail to cover complex dynamic conditions but also suffer from limitations such as long testing cycles and limited data dimensions. Furthermore, the performance baselines of different brake models vary significantly, making universal evaluation models prone to inadequate adaptability. Moreover, braking performance is also influenced by multiple factors, including vehicle load, driver operating habits, and road surface slippage. Evaluation methods that solely focus on the brake's own condition are insufficient to reflect braking performance under real-world conditions.

[0003] Against this backdrop, there is an urgent need for a braking performance evaluation method and system based on machine learning. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a braking performance evaluation method and system based on machine learning.

[0005] A first aspect of this application provides a machine learning-based braking performance evaluation method, comprising: Acquire first braking data and second braking data, wherein the first braking data includes brake wear data, braking execution status data and brake identification data; The braking execution status data and the brake wear data are subjected to feature extraction and fusion to generate a braking feature vector; Based on the brake identification data, select the corresponding pre-trained brake performance evaluation model to obtain the target brake performance evaluation model; The braking feature vector is input into the target braking performance evaluation model to obtain the preliminary evaluation results and confidence level. The confidence level is compared with a preset first threshold to obtain the comparison result; A correction factor library based on the second braking data is constructed using multi-source information fusion. The primary evaluation results are calibrated based on the comparison results and the correction factor library to obtain the braking performance evaluation results.

[0006] A second aspect of this application provides a machine learning-based braking performance evaluation system, comprising: The data acquisition module is used to acquire first braking data and second braking data. The first braking data includes brake wear data, braking execution status data and brake identification data. The feature fusion module is used to extract and fuse features from the braking execution state data and the brake wear data to generate a braking feature vector. The model matching module is used to select the corresponding pre-trained braking performance evaluation model based on the brake identification data to obtain the target braking performance evaluation model. The preliminary evaluation module is used to input the braking feature vector into the target braking performance evaluation model to obtain the preliminary evaluation results and confidence level. The confidence comparison module is used to compare the confidence level with a preset first threshold to obtain a comparison result; The result calibration module is used to construct a correction factor library based on the second braking data through multi-source information fusion; and to calibrate the primary evaluation result based on the comparison result and the correction factor library to obtain the braking performance evaluation result.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described machine learning-based braking performance evaluation method.

[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described machine learning-based braking performance evaluation method.

[0009] The beneficial effects of the machine learning-based braking performance evaluation method and system provided in this application are as follows: The hierarchical data acquisition-dedicated model evaluation-multi-source factor calibration braking performance evaluation system constructed in this application achieves a more precise and intelligent upgrade in braking safety monitoring. At the data level, by separating the brake's own state data from external related data such as vehicle, personnel, and environment, it focuses on both the core performance indicators of the braking system and the impact of complex operating conditions on braking performance. In the evaluation stage, a dedicated braking performance evaluation model is matched to the brake identifier, and the braking feature vector is used as input to obtain the initial evaluation result, ensuring the professionalism and adaptability of the evaluation; at the same time, based on the confidence threshold judgment mechanism and the calibration of the initial result based on the multi-source information correction factor library, the evaluation bias of a single data dimension is effectively eliminated, and the reliability of the evaluation result is improved. This application improves the efficiency and accuracy of braking performance evaluation. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a machine learning-based braking performance evaluation method provided in an embodiment of this application; Figure 2 A structural block diagram of a machine learning-based braking performance evaluation system provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application.

[0011] Explanation of reference numerals in the attached figures: 20. Machine Learning-Based Braking Performance Evaluation System; 21. Data Acquisition Module; 22. Feature Fusion Module; 23. Model Matching Module; 24. Preliminary Evaluation Module; 25. Confidence Comparison Module; 26. Result Calibration Module; 300. Electronic device; 301. Processor; 302. Input device; 303. Output device; 304. Memory; 305. Communication bus. Detailed Implementation

[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0013] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0014] Please refer to Figure 1 , Figure 1 A flowchart illustrating a machine learning-based braking performance evaluation method provided in an embodiment of this application is shown. The method includes: S101: Obtain first braking data and second braking data. The first braking data includes brake wear data, braking execution status data and brake identification data.

[0015] In this embodiment, the first braking data is the core status information of the brake body, including three types of core data: First, brake wear data is collected through the brake's built-in wear sensor or external detection device, including indicators such as brake pad thickness attenuation, brake disc surface scratch depth and area, and brake pad wear uniformity, representing the degree of wear of key friction components of the brake; second, braking execution status data is obtained in real time through the vehicle's electronic control system's controller area network bus, including data such as brake pedal travel, brake hydraulic / air pressure values, brake response delay time, and brake force distribution ratio, representing the real-time response capability and execution efficiency of the braking system; third, brake identification data includes static data such as brake model, manufacturer, compatible vehicle model, rated service life, and calibrated performance parameters. The second braking data includes vehicle status data, driver operation data, and external environment data. Among them, vehicle status data includes motion signals and load signals; external environment data includes road surface adhesion coefficient, ambient temperature and humidity, etc.; driver operation data includes brake pedal depressing force and depressing speed, etc. Motion signals include vehicle speed, vehicle posture, and brake disc rotation speed, etc., and load signals are vehicle load.

[0016] S102: Extract and fuse features from braking execution status data and brake wear data to generate a braking feature vector.

[0017] In this embodiment, before feature extraction and fusion of braking execution status data and brake wear data, preprocessing of the braking execution status data and brake wear data is required. Specifically, an outlier detection algorithm is used to identify and remove invalid data caused by sensor failure, missing data generated during transmission, and abnormal fluctuation data under extreme conditions. At the same time, for time-series braking execution status data, linear interpolation is used to fill in missing values ​​for short periods, and discrete outlier points in the wear data are smoothed by filtering.

[0018] Secondly, feature extraction is performed on the braking execution status data and brake wear data. For example, for the braking execution status data, a two-layer extraction of temporal and statistical features is first performed. At the temporal feature level, the response period from the triggering of the braking command to the stable output of braking force is captured, and dynamic temporal indicators such as the rise slope of the brake hydraulic / air pressure, the rate of change of the pedal travel, and the fluctuation amplitude of the braking force distribution are extracted during this period to characterize the real-time response characteristics of the braking system. At the statistical feature level, multiple sets of continuously collected braking execution data are summarized and analyzed to calculate the mean and variance of the braking response delay, the deviation rate of the braking force output, and the smoothness coefficient of the braking execution process, representing the stability of the long-term execution state of the braking system.

[0019] For brake wear data, the focus is on extracting wear characteristics and uniformity characteristics. Wear characteristics include core indicators such as the percentage of remaining brake pad thickness, the cumulative value of brake disc wear depth, and the wear rate of key friction components, representing the degree of physical wear of the brake. Uniformity characteristics include parameters such as the wear difference between the left and right sides of the brake pad and the deviation of wear in the circumferential direction of the brake disc, used to determine whether there is an abnormal state of non-uniform wear in the brake.

[0020] After feature extraction from both types of data, a multi-dimensional feature fusion algorithm was used for integration. The fusion process, with the correlation between braking performance and the data as its core logic, first assigned weights to the extracted features. The weight values ​​were determined through training and optimization using historical braking performance samples, with features having a greater impact on braking effectiveness receiving higher weights. Subsequently, through feature concatenation and normalization, the temporal and statistical features of braking execution status were integrated with the wear and uniformity features of brake wear into a vector of a unified dimension. Finally, a braking feature vector was generated that comprehensively represents the braking system's execution capability and physical wear state.

[0021] S103: Select the corresponding pre-trained braking performance evaluation model based on the brake identification data to obtain the target braking performance evaluation model.

[0022] In this embodiment, to achieve matching of performance evaluation models for different brake models, the brake identification data is first matched against a pre-established brake model-model mapping table. Specifically, if a match is successful, the corresponding brake performance evaluation model is used as the target brake performance evaluation model, ensuring strong compatibility between the brake performance evaluation model and the brake model. If a match fails, i.e., the brake identification is not entered into the mapping table, a matching process based on the similarity of key attribute feature vectors is executed. Specifically, firstly, core attribute parameters such as brake type, rated power, adapted friction coefficient, operating temperature range, and design service life are parsed and extracted from the brake identification data. After standardization, a fixed-dimensional key attribute feature vector is generated. Then, the similarity between this feature vector and the key attribute feature vectors of all pre-trained brake performance evaluation models associated with the brakes in the mapping table is calculated. Finally, the brake performance evaluation model corresponding to the feature vector with the highest similarity is selected as the target brake performance evaluation model.

[0023] S104: Input the braking feature vector into the target braking performance evaluation model to obtain the preliminary evaluation results and confidence level.

[0024] In this embodiment, since the target braking performance evaluation model has fixed requirements on the dimension and data format of the input vector, it is first verified whether the dimension of the braking feature vector is consistent with the dimension of the input layer of the target braking performance evaluation model. For example, if they are inconsistent, feature dimension expansion or dimensionality reduction is performed. Feature dimension expansion includes expanding the dimension of low-dimensional vectors using zero-padding or feature derivation; it includes reducing the dimension of high-dimensional vectors using principal component analysis; and simultaneously converting the feature vectors into tensor or matrix formats supported by the target braking performance evaluation model to eliminate inference errors caused by data format differences.

[0025] The target braking performance evaluation model utilizes its pre-trained parameter weights to perform hierarchical calculations on features across various dimensions of the braking feature vector. Specifically, for features related to braking execution state, such as the brake hydraulic pressure rise slope and the mean response delay, the model uses a feature extraction layer to uncover their correlation with braking performance degradation. For features related to brake wear, it performs quantitative comparative analysis based on the wear threshold parameters for that brake model built into the model. The calculation results of each feature are then fused and judged through the fully connected layer or output layer of the target braking performance evaluation model to generate a preliminary evaluation result.

[0026] The initial assessment results are output in the form of performance level + key issue prompts. The performance level is divided into five levels: excellent, good, qualified, need maintenance, and faulty, which correspond to different health states of the braking system. The key issue prompts point out the core factors affecting performance, such as excessive brake pad wear leading to braking force deviation or excessive braking response delay.

[0027] The confidence score generated simultaneously with the initial assessment results in this embodiment is a reliability index calculated by the target braking performance assessment model based on the completeness of the input features and the matching degree between the features and historical samples. Its calculation logic is as follows: by statistically analyzing the average feature similarity between the current braking feature vector and samples of the same performance level in the training set of the target braking performance assessment model, and simultaneously, based on the historical prediction accuracy of the target braking performance assessment model within this similarity interval, a weighted calculation method is used to obtain the confidence score value. The confidence score is represented by a value between 0 and 1; the closer the value is to 1, the higher the reliability of the initial assessment result.

[0028] S105: Compare the confidence level with the preset first threshold to obtain the comparison result.

[0029] In this embodiment, the first threshold is a reliability critical value set based on historical data statistics from a large number of braking performance evaluation samples and the evaluation accuracy requirements in engineering practice. Its value ranges from 0 to 1, and the specific value is customized according to the brake model and evaluation scenario. For example, for scenarios with extremely high evaluation accuracy requirements, such as heavy-duty braking systems for commercial vehicles, the first threshold is typically set to 0.85; for conventional braking scenarios for passenger vehicles, the first threshold can be appropriately lowered to 0.75. The first threshold is stored in the system threshold configuration library and supports iterative optimization based on evaluation results.

[0030] Secondly, the confidence scores are validated to remove invalid confidence scores caused by abnormal feature vectors or inference failures in the target braking performance evaluation model. If the validation fails, the process of regenerating the primary evaluation results is triggered directly. After the validation passes, the valid confidence scores are compared with the first threshold.

[0031] If the confidence level is greater than or equal to the first threshold, the comparison result is determined to be a reliable preliminary evaluation result, indicating that the current braking feature vector has a high degree of matching with the training samples of the target braking performance evaluation model. Conversely, if the confidence level is less than the first threshold, the comparison result is determined to be a preliminary evaluation result requiring calibration, indicating that the current feature vector has a low degree of sample matching. Furthermore, the comparison result is stored in association with the confidence level and the preliminary evaluation result.

[0032] S106: Construct a correction factor library based on the second braking data for multi-source information fusion.

[0033] The initial evaluation results are calibrated based on the comparison results and the correction factor library to obtain the braking performance evaluation results.

[0034] In this embodiment, the second braking data includes vehicle state data, driver operation data, and external environment data; the preset hierarchical correction rule set includes a first-level rule subset, a second-level rule subset, and a third-level rule subset. Next, based on the motion signals in the vehicle state data and the external environment data, the rules in the first-level rule subset are matched and executed to obtain the basic physical layer correction factor; based on the load signals in the vehicle state data, the driver operation data, and the basic physical layer correction factor, the rules in the second-level rule subset are matched and executed to obtain the system response layer correction factor; based on the basic physical layer correction factor and the system response layer correction factor, the rules in the third-level rule subset are matched and executed to obtain the comprehensive performance layer correction factor; the basic physical layer correction factor, the system response layer correction factor, and the comprehensive performance layer correction factor are integrated to obtain a multi-source information fusion correction factor library.

[0035] If the comparison result shows a confidence level greater than or equal to the first threshold, the primary evaluation result is used as the main correction benchmark, and the comprehensive performance layer correction factor in the correction factor library is used as the weight adjustment factor to perform weighted correction on the main correction benchmark to obtain the braking performance evaluation result. If the comparison result shows a confidence level less than the first threshold, the primary evaluation result is used as auxiliary reference information, and its decision weight is reduced to below the preset fourth threshold. The correction factor library is used as the main decision basis to perform comprehensive derivation to obtain the braking performance evaluation result.

[0036] As can be concluded from the above, the hierarchical data acquisition-dedicated model evaluation-multi-source factor calibration braking performance evaluation system constructed in this application achieves a precise and intelligent upgrade in braking safety monitoring. At the data level, by separating the brake's own state data from external related data such as vehicle, personnel, and environment, it focuses on both the core performance indicators of the braking system and the impact of complex operating conditions on braking performance. In the evaluation stage, a dedicated braking performance evaluation model is matched to the brake identifier, and a preliminary evaluation result is output based on the fused feature vector of braking execution state data and brake wear data, ensuring the professionalism and adaptability of the evaluation. Simultaneously, based on a confidence threshold judgment mechanism and a multi-source information correction factor library, the preliminary results are calibrated, effectively eliminating evaluation bias from a single data dimension and improving the reliability of the evaluation results. This application improves the efficiency and accuracy of braking performance evaluation.

[0037] In one embodiment of this application, a target braking performance evaluation model is obtained by selecting a pre-trained braking performance evaluation model based on brake identification data, including: Match the brake identification data with the pre-established brake model-model mapping table; If a match is successful, the corresponding braking performance evaluation model will be used as the target braking performance evaluation model. If the matching fails, a similarity matching process based on key attribute feature vectors is executed to select the braking performance evaluation model with the highest similarity as the target braking performance evaluation model.

[0038] In this embodiment, the brake model-model mapping table is the core association carrier built through a large number of engineering tests and brake performance evaluation model training in the early stage. The brake model-model mapping table not only stores the identifier of each brake, but also binds it to a pre-trained brake performance evaluation model customized for that model, as shown in Table 1.

[0039] Table 1 Brake Model-Model Mapping Table Brake markings Model Number Brake compatibility details Disc brake Model-DZ-120A-V2.1 Compatible with DZ-120A universal disc brakes for passenger vehicles, including compact cars and small SUVs. Drum brake Model-GS-85B-V1.8 Compatible with GS-85B mini-vehicle drum brakes, suitable for mini vans and low-speed cargo tricycles. Model-YT-90D-V1.9 Compatible with dual-shoe drum brakes for YT-90D pickup trucks, suitable for both gasoline-powered and off-road modified pickup trucks. The Model-DZ-120A-V2.1 is a universal disc brake model for passenger vehicles. This braking performance evaluation model employs a two-layer hybrid network architecture. The CNN convolutional layers use three different sizes (3×1, 5×1, 7×1) of one-dimensional convolutional kernels to extract multi-scale spatial features in parallel, with 32 filters configured for each size. The ReLU activation function is used, with a stride of 1 and zero padding to maintain temporal length. The LSTM layers use a two-layer stacked structure, with 64 hidden units per layer and a dropout rate of 0.2 to prevent overfitting. The number of neurons in the fully connected layers decreases from 512 to 256 to 128 to 5, ultimately outputting a 5-class probability distribution using Softmax. During the training phase, the batch size was set to 64, the initial learning rate was 0.001 with cosine annealing decay, and the AdamW optimizer was selected. The cross-entropy loss function was weighted by 1.5 times for the maintenance-needed and fault categories to improve the ability to identify critical states. The braking performance evaluation model was trained for 200 rounds, with an early stop patience value set to 15 rounds. The input to the braking performance evaluation model was a standardized 12-dimensional braking feature vector, and the output included two parts: first, a 5-category braking performance level (Excellent-Good-Quality-Needs Maintenance-Fault); and second, an evaluation confidence score in the 0-1 range, used to characterize the reliability of the evaluation results.

[0040] Model-GS-85B-V1.8 is a model for drum brakes in microcars. This braking performance evaluation model uses an ensemble decision tree hierarchical architecture, consisting of 100 base decision trees forming a random forest model in parallel. The maximum depth of each tree is limited to 15 layers to prevent overfitting, and the minimum number of split samples per node is set to 5 to preserve detailed features. The feature selection layer employs a weighted random subspace strategy, where the selection probability of brake gap-related features during feature sampling is increased to 1.8 times that of standard features. The node splitting criterion uses the Gini coefficient, and forced splitting is enforced when the proportion of fault classes in the node samples exceeds 40% to enhance anomaly detection. During training, a bootstrap sampling method is used to generate training subsets, and fault samples with excessively large gaps are given a 2.0 times sample weight. Model performance is monitored using out-of-bag error estimation. Its input is a 10-dimensional standardized braking feature vector, and the output is a 5-class braking performance level and its corresponding confidence score, with additional specific question prompts for drum brake-specific faults.

[0041] Model-YT-90D-V1.9 is a model for a pickup truck's dual-shoe drum brake. This braking performance evaluation model uses a kernel function mapping hierarchical architecture, belonging to the support vector machine model. The hierarchy includes a feature kernel mapping layer, an optimal hyperplane partitioning layer, and a result output layer. The kernel mapping layer solves the nonlinear separability problem through radial basis functions. The core parameter γ of this function is determined to be 0.1 through grid search, which effectively balances the mapping complexity of the feature space with the model's generalization ability. The penalty factor C is set to 10, achieving an optimal balance between maximizing the classification margin and the tolerance for misclassification. The dual-shoe braking force distribution feature is assigned a weight coefficient of 1.5 in the kernel function calculation. During the training phase, a one-to-one strategy is used to construct 10 binary classifiers to handle the 5-class problem. A misclassification penalty weight of 1.8 is set for samples with uneven braking force distribution under wet road conditions. The optimal parameter combination was determined through 5-fold cross-validation, and the decision function values ​​were converted to 0-1 confidence levels using a probabilistic calibration method. The final model achieved 92.0% accuracy on the retained test set, and the detection sensitivity for braking force imbalance faults under wet and slippery conditions reached 86.3%. The input of this braking performance evaluation model is an 11-dimensional standardized braking feature vector, and the output is a 5-category braking performance level and an evaluation confidence level.

[0042] During matching, the collected brake identification data is compared with the identification information in the mapping table. If a completely matching entry exists, its associated brake performance evaluation model is directly retrieved as the target brake performance evaluation model.

[0043] If a match fails, the key attribute feature vector similarity matching process is immediately triggered. First, core attribute parameters are parsed and extracted from the brake identification data. Then, these parameters are standardized to eliminate dimensional differences, generating fixed-dimensional key attribute feature vectors. After feature vector construction, the similarity between the key attribute feature vectors and the corresponding brake feature vectors of all brake performance evaluation models in the mapping table is calculated. Finally, the brake performance evaluation model with the highest similarity is selected as the target brake performance evaluation model.

[0044] As can be seen from the above, the braking performance evaluation model matching mechanism in this embodiment not only realizes the calling of the exclusive braking performance evaluation model based on the brake model-model mapping table, ensuring the strong adaptability between the evaluation model and the brake model, but also solves the problem of new models and niche customized brakes not being recorded in the mapping table through the key attribute feature vector similarity matching process. This not only improves the accuracy and efficiency of braking performance evaluation, but also enhances the compatibility and fault tolerance of the evaluation system, providing reliable and comprehensive model support for the performance evaluation of different types of brakes.

[0045] In one embodiment of this application, if matching fails, a matching process based on the similarity of key attribute feature vectors is executed to select the braking performance evaluation model with the highest similarity as the target braking performance evaluation model, including: Extract the attribute feature vectors from the brake identification data to obtain the key attribute feature vectors; Obtain the standard attribute feature vectors of all brakes corresponding to the brake performance evaluation models in the brake model-model mapping table; Calculate the similarity between the key attribute feature vector and the standard attribute feature vector; The braking performance evaluation model corresponding to the feature vector with the highest similarity is selected as the target braking performance evaluation model.

[0046] In this embodiment, the attribute parameters that have a core impact on the braking performance evaluation are parsed from the brake identification data. The attribute parameters include core dimensions such as brake structure type, rated torque, suitable friction coefficient range, operating temperature range, design service life, and brake pipeline interface specifications.

[0047] Differentiated coding and standardization are applied to different types of attribute parameters. For example, discrete attributes are encoded using binary or multivariate one-hot coding, while continuous attributes are standardized using Min-Max mapping to the 0-1 range. Finally, the processed parameters from each dimension are integrated into a fixed-length key attribute feature vector. Discrete attributes include brake structure type and brake pipeline interface specifications; these attributes exhibit discontinuous categorical characteristics with no clear numerical relationship. Continuous attributes include rated torque, suitable friction coefficient range, operating temperature range, and design service life; these attributes have continuously changing values ​​that can be described by quantitative indicators. Next, the standardized attribute feature vectors are retrieved and preprocessed. Standard attribute feature vectors corresponding to all brake performance evaluation models are obtained from the brake model-model mapping table. These standard attribute feature vectors are baseline vectors generated from the design parameters and performance baselines of each brake model using the same coding and standardization rules.

[0048] After extracting and standardizing the key attribute feature vectors and standard attribute feature vectors, an appropriate similarity calculation strategy is selected based on the type characteristics of the attribute parameters. Specifically, for discrete attributes, the Jaccard similarity coefficient is used to measure the degree of matching, i.e., the ratio of the number of identical attribute categories in the two vectors to the total number of attribute categories determines the fit of discrete attributes. For continuous attributes, a weighted algorithm of Euclidean distance and cosine similarity is used. Cosine similarity is used to judge the consistency of the overall direction of the vectors, focusing on the matching logic of core performance parameters, while Euclidean distance is used to measure the absolute deviation of parameter values, compensating for the insensitivity of cosine similarity to numerical differences. Simultaneously, weights are assigned based on core attributes, such as a 30% weight for brake structure type and a 25% weight for rated torque, while the weight of non-core attributes is controlled within 10%. A comprehensive similarity value is obtained through weighted fusion; the closer the comprehensive similarity value is to 1, the higher the matching degree between the two sets of vectors.

[0049] After obtaining the comprehensive similarity value of the standard vectors corresponding to all braking performance evaluation models, the braking performance evaluation model corresponding to the feature vector with the highest comprehensive similarity value is selected as the target braking performance evaluation model. Furthermore, if multiple braking performance evaluation models have a similarity value difference of less than or equal to 0.05, a secondary screening based on scenario adaptability is performed. Specifically, based on the actual application scenario of the brake to be evaluated and the coverage of the training sample working conditions of the braking performance evaluation model, models with high scenario matching are prioritized. For example, if the brake to be evaluated is a drum brake adapted for a pickup truck, even if the attribute similarity of a certain disc brake model is slightly higher, the model corresponding to the YT-90D model optimized for pickup truck off-road scenarios will be selected first.

[0050] As can be seen from the above, this embodiment achieves accurate model adaptation for brakes not included in the mapping table through standardized vector extraction and similarity calculation. The unified benchmark based on the standard attribute vectors in the mapping table ensures the rigor of the matching logic, effectively fills the blind spot of accurate mapping of exclusive models, and improves the compatibility of the evaluation system with new models and customized brakes, providing an efficient and reliable model selection path for the performance evaluation of various brakes.

[0051] In one embodiment of this application, after selecting the braking performance evaluation model with the highest similarity as the target braking performance evaluation model, the method further includes: Record this matching event and generate a mapping record including brake identification data, the identification of the selected target braking performance evaluation model, and similarity. When the cumulative number of mapping records is greater than or equal to the preset second threshold, or when the number of times the evaluation results output by the target braking performance evaluation model are verified as valid is greater than or equal to the preset third threshold, the mapping records are converted into permanent mapping relationships and updated to the brake model-model mapping table.

[0052] In this embodiment, after determining the target braking performance evaluation model through key attribute feature vector similarity matching, it is necessary to record the matching event and generate a standardized mapping record.

[0053] The mapping record contains three main parts: first, brake identification data; second, the identification of the target braking performance evaluation model, including model number, model architecture type, training sample coverage conditions, and other information; and third, similarity-related quantitative data, including the specific dimensional values ​​of key attribute feature vectors, the comprehensive similarity value with the standard vectors of each braking performance evaluation model, the specific value of the highest similarity, and the weight allocation rules used in the calculation process.

[0054] Next, the generated mapping records are stored in a temporary mapping database for accumulation and verification. When the preset upgrade conditions are met, they are converted into permanent mapping relationships and updated to the brake model-model mapping table. The upgrade conditions are divided into two categories, and an update can be triggered if either one is met: The first category is the quantity accumulation condition. When the accumulated number of mapping records for the same brake identifier data is greater than or equal to the preset second threshold, it indicates that the similarity matching results of the brake have a certain frequency of reuse and practical basis. The setting of the second threshold needs to be based on the application popularity of the brake. For example, it is set to 5 for general brakes and 3 for niche customized brakes. The second category is the result verification condition. When the number of times the evaluation results output by the target brake performance evaluation model in the mapping record are confirmed to be valid is greater than or equal to the preset third threshold, it indicates that the adaptability of the brake performance evaluation model and the brake has been verified. The third threshold is determined based on the complexity of the braking scenario.

[0055] In this embodiment, after triggering the update process, a consistency check is performed on the mapping records. For example, it verifies whether the correspondence between the brake identifier and the target braking performance evaluation model is consistent in multiple matches. If a conflict exists, a manual review is initiated. After the verification passes, the mapping records are converted into standard permanent mapping entries and added to the brake model-model mapping table according to the association format of brake unique identifier-target model number. The model status in the table is also updated synchronously to "verified". At the same time, the system automatically deletes the corresponding mapping records in the temporary database to avoid data redundancy.

[0056] From the above, it can be concluded that this embodiment achieves traceability and verifiability of the matching process of non-registered brake models by retaining core information such as brake identification data, target brake performance evaluation model identification, and similarity. By setting dual thresholds for cumulative quantity and effective verification times, the reliability and practicality of newly added permanent mapping relationships are ensured. This avoids disorderly expansion of the mapping table and continuously incorporates the adaptation relationships verified in practice into the mapping table, gradually expanding the scope of accurate matching of the brake performance evaluation model.

[0057] In one embodiment of this application, it further includes: Based on vehicle identification code sequences, permanent geographical location information, or historical driving pattern data, clustering algorithms are used to divide vehicles into different vehicle clusters. The success rate of converting the mapping records of the vehicle cluster into permanent mapping relationships is statistically analyzed. Adjust the second and third thresholds based on the update success rate.

[0058] In this embodiment, core feature data of the vehicles is first collected. This core feature data includes inherent attributes extracted from the vehicle identification number sequence, such as brand, model, and braking system configuration; the vehicle's permanent geographical location information, including cold and temperate zones; and historical driving pattern data, including average daily mileage, heavy load ratio, and emergency braking ratio. Based on this multi-dimensional data, the K-means clustering algorithm is used for cluster division. Specifically, the number of clusters is first determined using the silhouette coefficient method; then, vehicles are grouped into different clusters based on feature similarity to obtain the initial clusters.

[0059] After determining the number of clusters and dividing the initial clusters, the characteristics of the initial clusters are checked for consistency. By calculating the variance and coefficient of variation of the vehicle feature data within the cluster, abnormal vehicles with excessive feature deviations are removed and reassigned to suitable clusters. This ensures that the braking system usage scenarios and operating conditions within each vehicle cluster are highly homogeneous, such as the heavy-duty commercial vehicle cluster in cold mountainous areas, the commuter passenger vehicle cluster in temperate cities, and the new energy vehicle cluster in subtropical coastal areas.

[0060] Secondly, the update success rate of successfully converting the mapping records of the vehicle cluster into permanent mapping relationships is statistically analyzed. When calculating the update success rate, differentiated statistical dimensions are set based on the operating characteristics of the vehicle cluster. For example, for heavy-duty commercial vehicle clusters, the focus is on the verification pass rate of evaluation results under mountainous slope conditions; for new energy vehicle clusters, the focus is on the verification effectiveness of regenerative braking coordination scenarios. Finally, the update success rate of the vehicle cluster is calculated using the core formula: the number of records successfully converted to permanent mappings within the cluster / the total number of mapping records participating in the verification within the cluster.

[0061] Subsequently, the second and third thresholds are adjusted based on the update success rate. Specifically, when the update success rate is consistently lower than the first success rate threshold and reaches a preset first cycle, the second and third thresholds are increased by a preset first increment; when the update success rate is consistently higher than the second success rate threshold and reaches a preset first cycle, the second and third thresholds are decreased by a preset second increment; wherein, the first success rate threshold is lower than the second success rate threshold. The formula for calculating the first success rate is: R1=min(R x ×0.8,0.6) R1 represents the highest success rate; R x 0.8 represents the historical average success rate of vehicle cluster mapping record conversion; 0.8 represents the industry fault tolerance standard for braking performance evaluation, while 0.6 represents the lower limit constraint value of the success rate derived from the industry safety standards for braking performance evaluation, engineering practice experience, and the reliability baseline of mapping relationship updates; the calculation result takes the lower value of the two as the initial benchmark.

[0062] The formula for calculating the second success rate is: R² = min(Ry × 0.9, 0.85), where 0.85 is the upper limit constraint value for the success rate; Ry is the historical excellent success rate, and its calculation formula is: , The success rate of the i-th sample within the top 30% quantile. Let be the weight coefficient of the cluster corresponding to the i-th sample. This represents the number of samples within the top 30% quantile. The weighting coefficients are set based on typical core clusters and special operating condition clusters in braking scenarios. For example, the cluster of heavy-duty commercial vehicles in cold mountainous areas, due to their harsh operating conditions and extremely high requirements for braking reliability, is defined as a core and critical scenario and is assigned the highest weight. =1.5; The temperate city commuter passenger vehicle cluster, due to its widespread application and largest data volume, is assigned a basic weight as a standard reference scenario. =1.0; while the subtropical coastal new energy vehicle cluster, due to the special nature of regenerative braking scenarios and the fact that data samples are still accumulating, is currently considered a special operating condition cluster awaiting observation and is assigned a lower weight, for example... =0.7.

[0063] The first amplitude is set based on the deviation rate between the update success rate and the first success rate threshold. For example, when the deviation rate is less than or equal to 20%, the first amplitude is set to 10%-15%; when the deviation rate is 20%-40%, the first amplitude is set to 15%-20%; when the deviation rate is greater than 40%, the first amplitude is set to 20%-25%. An additional 5%-10% amplitude is added for high-safety clusters such as passenger transport and hazardous chemical transport. Furthermore, the adjusted second threshold is less than 150% of the initial second threshold, and the third threshold is less than 200% of the initial third threshold. The second amplitude is referenced to the update... The success rate and the excess rate of the second success rate threshold are set as the core basis. For example, when the excess rate is less than or equal to 20%, the second range is set to 10%-15%; when the excess rate is 20%-40%, the second range is set to 15%-20%; when the excess rate is greater than 40%, the second range is set to 20%. If the accuracy of the braking performance evaluation model is greater than or equal to 95%, an additional 5% range is added, and the second threshold after the adjustment is greater than 60% of the initial second threshold, and the third threshold is greater than 50% of the initial third threshold. The first cycle is set based on the mapping record generation frequency.

[0064] This embodiment constructs a clustering feature space based on historical driving pattern data of vehicles, including at least historical driver operation data feature dimensions and historical external environment feature dimensions; obtains an initial clustering parameter set for the target vehicle cluster, which includes at least one of the following: number of clusters, neighborhood radius, or minimum number of samples; and adjusts the initial clustering parameter set based on the update success rate: when the update success rate is less than the lower limit constraint value, the number of clusters is increased or the neighborhood radius is decreased by a preset first step length; when the update success rate is greater than the upper limit constraint value, the number of clusters is decreased or the neighborhood radius is increased by the first step length. The first step length is the quantization step size for adjusting the clustering parameters, and is set differentially based on the number of clusters, neighborhood radius, and cluster characteristics.

[0065] From the above, it can be concluded that this embodiment achieves the management of vehicles in the same braking scenario by dividing the vehicle clusters, making the adjustment of the second and third thresholds more in line with the working characteristics and adaptation requirements of different clusters. Based on the statistics of update success rate, it not only avoids the adaptation deviation of a single threshold for different clusters, but also improves the targeting, flexibility and overall adaptability of the brake model-model mapping table update mechanism by optimizing the efficiency and reliability of the balance mapping relationship update between the second and third thresholds.

[0066] In one embodiment of this application, a correction factor library for multi-source information fusion is constructed based on the second braking data, including: The second braking data includes vehicle status data, driver operation data, and external environment data; A pre-defined hierarchical correction rule set is provided, which includes a first-level rule subset, a second-level rule subset, and a third-level rule subset. Based on motion signals in vehicle state data and external environment data, the rules in the first-level rule subset are matched and executed to obtain the basic physical layer correction factor. Based on the load signal, driver operation data and basic physical layer correction factor in the vehicle status data, the rules in the second-layer rule subset are matched and executed to obtain the system response layer correction factor. Based on the basic physical layer correction factor and the system response layer correction factor, the rules in the third layer rule subset are matched and executed to obtain the comprehensive performance layer correction factor. By integrating the basic physical layer correction factor, the system response layer correction factor, and the comprehensive performance layer correction factor, a correction factor library for multi-source information fusion is obtained.

[0067] In this embodiment, the second braking data includes vehicle status data, driver operation data, and external environment data. Vehicle status data includes vehicle speed, brake disc temperature, vehicle load, and suspension status; driver operation data includes brake pedal travel, pedal force, and brake trigger frequency; external environment data includes road surface adhesion coefficient, ambient temperature and humidity, weather conditions such as rain / snow, and road slope and curvature.

[0068] The hierarchical correction rule set comprises a first-level rule subset, a second-level rule subset, and a third-level rule subset. This hierarchical correction rule set is divided into three levels based on the physical meaning and applicable scope of the correction logic. The first-level rule subset focuses on the fundamental logic of the environment-vehicle physical coupling, including rules relating road surface adhesion coefficient to braking deceleration, correction rules for the effect of ambient temperature on brake disc friction coefficient, and adjustment rules for the effect of road gradient on brake load distribution. The second-level rule subset emphasizes the coordinated response of load, operation, and basic correction, including rules for compensating for changes in vehicle load on brake pressure distribution, adaptation rules for pedal operation intensity and basic physical correction factors, and rules for thermal fade superposition correction under continuous braking scenarios. The third-level rule subset addresses the comprehensive adaptation of correction factors across all dimensions, including rules for resolving multi-factor conflicts, rules for prioritizing correction weights under extreme conditions, and factor fusion rules oriented towards braking performance evaluation goals.

[0069] In this embodiment, motion signals from vehicle status data and external environment data are matched and the corresponding rules in the first-level rule subset are executed. The motion signals include vehicle speed, vehicle posture, and brake disc rotation speed.

[0070] Secondly, based on the load signal and driver operation data in the vehicle status data, and based on the basic physical layer correction factor, a subset of second-layer rules is matched and executed. The load signal includes the vehicle body load. Based on the basic physical layer factor, the braking pressure distribution coefficient is adjusted upwards by 10%. For example, if more than three consecutive high-frequency braking events are detected, the system response layer's thermal fatigue superposition rule is activated based on the basic physical layer's thermal attenuation factor, generating a collaborative correction factor for braking performance attenuation, ultimately yielding the system response layer correction factor.

[0071] Finally, based on the basic physical layer correction factors, system response layer correction factors, adaptation scenarios, and conflict states, a subset of third-layer rules is matched and executed. Specifically, for multi-factor adaptation conflict scenarios, such as the contradiction between modulated deceleration under low road adhesion requirements and modulated pressure under high load requirements, the factor weight priority is determined through conflict resolution logic in the rule set; for special operating conditions such as emergency braking, performance-priority fusion rules are activated to strengthen the correction ratio of key factors; for regular commuting operating conditions, balanced fusion rules are adopted to ensure the universality and stability of correction factors. Through the processing of the third-layer rule subset, the comprehensive performance layer correction factor is finally obtained.

[0072] After completing the hierarchical calculation of the three-layer correction factors, the basic physical layer correction factors, system response layer correction factors, and comprehensive performance layer correction factors are classified and integrated according to the dimensions of scene label + factor type + adaptation range to construct a multi-source information fusion correction factor library.

[0073] As can be seen from the above, the correction factor library construction process in this embodiment integrates multi-dimensional braking data of vehicle status, driver operation, and external environment. It uses a hierarchical rule set to progressively generate correction factors. The first-level rule subset captures the basic physical coupling law between the environment and the vehicle, providing an objective underlying basis for correction. The second-level rule subset integrates load and operation data and basic factors to achieve system response adaptation. The third-level rule subset completes the comprehensive optimization and conflict resolution of all-dimensional factors. The final integrated correction factor library not only includes all influencing factors of braking scenarios, ensuring the scenario-specificity and logical rigor of correction factors, but also provides dynamic and accurate correction support for braking performance evaluation models, effectively improving the accuracy and reliability of evaluation results under different working conditions.

[0074] In one embodiment of this application, the initial evaluation results are calibrated based on the comparison results and the correction factor library to obtain the braking performance evaluation results, including: If the comparison result is a confidence level greater than or equal to the first threshold, then the first calibration mode is executed; If the comparison result shows a confidence level less than the first threshold, then the second calibration mode is executed; The first calibration mode includes: using the primary assessment results as the main correction benchmark, using the comprehensive performance layer correction factors in the correction factor library as weight adjustment factors, and performing weighted correction on the main correction benchmark to obtain the braking performance assessment results. The second calibration mode includes: using the primary assessment results as auxiliary reference information and using the correction factor library as the main decision information; The braking performance evaluation results are obtained by comprehensively deriving auxiliary reference information and main decision-making information.

[0075] In this embodiment, if the comparison result has a confidence level greater than or equal to the first threshold, it indicates that the initial assessment result has a high basic credibility, and the first calibration mode will be activated. This first calibration mode uses the initial assessment result as the core correction benchmark, retaining its judgment logic for the core dimensions of braking performance. Simultaneously, it uses the comprehensive performance layer correction factor from the correction factor library as a weight adjustment factor to calibrate the benchmark result for different braking scenarios. For example, when the vehicle is in a heavy-load braking scenario on a slippery road surface, the initial assessment result gives a basic judgment of good braking performance. The low adhesion + high load synergy factor in the comprehensive performance layer correction factor will adjust the weight of this result downwards, ultimately outputting an assessment result that the braking performance basically meets the requirements, suggesting a reduction in vehicle speed and a decrease in continuous braking. This retains the core judgment of the initial assessment while using scenario-based correction factors to compensate for the general assessment model's insufficient adaptation to special working conditions.

[0076] If the comparison result shows a confidence level less than the first threshold, it indicates that the initial assessment result is affected by factors such as the complexity of the operating conditions, missing data, or model adaptation bias, resulting in insufficient basic reliability. In this case, the system will switch to the second calibration mode. In this mode, the initial assessment result is no longer used as the core basis but as auxiliary reference information. Instead, the correction factor library serves as the main decision-making information, guiding the calibration process. The braking performance is assessed through the combination and derivation of multi-dimensional factors.

[0077] Specifically, the comprehensive derivation process consists of three steps: The first step is factor selection and scenario matching. Based on the core characteristics of the current braking scenario, corresponding basic physical layer, system response layer, and comprehensive performance layer correction factors are extracted from the correction factor library to form a specific correction factor set for the scenario. At the same time, factors irrelevant to the current scenario are removed to avoid interference. The second step is auxiliary information fusion and conflict verification. The auxiliary reference information in the primary evaluation results is compared with the specific factor set. If there is a conflict between the normal braking response in the primary evaluation results and the correction logic in the specific factor set that low temperature causes an increase in brake fluid viscosity and a 15% response delay, the physical law derivation results of the correction factor library shall prevail. Only the response time value of the primary evaluation is used as the basic calculation parameter, and the correction and adjustment are made based on the specific factor set. For example, if the core trends of the two are consistent, the primary evaluation data is used as a supplementary basis for factor derivation. The third step is multi-factor weighted calculation and result output. Based on the explicit weight priority in the comprehensive performance layer correction factors, for example, in cold-weather heavy load scenarios, the weight of the heat fade factor is 0.35, the weight of the load correction factor is 0.3, the weight of the road adhesion factor is 0.25, and the weight of other auxiliary factors is 0.1. The correction values ​​of each factor are weighted and summed to obtain the calibration values ​​of each dimension of braking performance. Finally, the calibration values ​​are transformed into concrete evaluation results based on industry evaluation standards, such as braking performance meeting the standards but with the risk of low-temperature response delay, and recommendations to check the brake fluid grade and shorten the brake gap. At the same time, the credibility of the results is marked. The credibility is calculated based on the data integrity of the exclusive correction factor set.

[0078] As can be seen from the above, the flexible switching between the two calibration modes in this embodiment not only ensures the consistency of evaluation efficiency and basic judgment in high-confidence scenarios, but also solves the problem of reliability of evaluation results in low-confidence scenarios. By dynamically allocating the weights of the primary evaluation results and the correction factor library, a braking performance evaluation and calibration system with both universality and scenario adaptability is constructed.

[0079] Corresponding to the machine learning-based braking performance evaluation method in the above embodiments, Figure 2 This is a structural block diagram of a machine learning-based braking performance evaluation system provided in one embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The machine learning-based braking performance evaluation system 20 includes: a data acquisition module 21, a feature fusion module 22, a model matching module 23, a preliminary evaluation module 24, a confidence comparison module 25, and a result calibration module 26.

[0080] Data acquisition module 21 is used to acquire first braking data and second braking data. The first braking data includes brake wear data, braking execution status data and brake identification data. Feature fusion module 22 is used to extract and fuse features from braking execution state data and brake wear data to generate braking feature vectors; The model matching module 23 is used to select the corresponding pre-trained braking performance evaluation model based on the brake identification data to obtain the target braking performance evaluation model. The preliminary assessment module 24 is used to input the braking feature vector into the target braking performance assessment model to obtain the preliminary assessment results and confidence level; The confidence comparison module 25 is used to compare the confidence level with a preset first threshold to obtain the comparison result; The result calibration module 26 is used to construct a correction factor library based on the second braking data and to calibrate the primary evaluation results based on the comparison results and the correction factor library to obtain the braking performance evaluation results.

[0081] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the data acquisition module 21, feature fusion module 22, model matching module 23, preliminary evaluation module 24, confidence comparison module 25, and result calibration module 26 are shown.

[0082] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0083] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0084] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0085] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the machine learning-based braking performance evaluation method provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0086] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0087] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0088] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0089] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0090] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0091] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0092] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0093] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A machine learning-based braking performance evaluation method, characterized in that, include: Acquire first braking data and second braking data, wherein the first braking data includes brake wear data, braking execution status data and brake identification data; The second braking data includes vehicle status data, driver operation data, and external environment data; The braking execution status data and the brake wear data are subjected to feature extraction and fusion to generate a braking feature vector; Based on the brake identification data, select the corresponding pre-trained brake performance evaluation model to obtain the target brake performance evaluation model; The braking feature vector is input into the target braking performance evaluation model to obtain the preliminary evaluation results and confidence level. The confidence level is compared with a preset first threshold to obtain the comparison result; A correction factor library based on the second braking data is constructed using multi-source information fusion. The primary evaluation results are calibrated based on the comparison results and the correction factor library to obtain the braking performance evaluation results.

2. The braking performance evaluation method based on machine learning according to claim 1, characterized in that, The step of selecting the corresponding pre-trained braking performance evaluation model based on the brake identification data to obtain the target braking performance evaluation model includes: The brake identification data is matched and queried with the pre-established brake model-model mapping table; If a match is successful, the corresponding braking performance evaluation model shall be used as the target braking performance evaluation model. If the matching fails, a similarity matching process based on key attribute feature vectors is executed to select the braking performance evaluation model with the highest similarity as the target braking performance evaluation model.

3. The braking performance evaluation method based on machine learning according to claim 2, characterized in that, If the matching fails, a matching process based on the similarity of key attribute feature vectors is executed to select the braking performance evaluation model with the highest similarity as the target braking performance evaluation model, including: Extract the attribute feature vector from the brake identification data to obtain the key attribute feature vector; Obtain the standard attribute feature vectors of all brakes corresponding to the brake performance evaluation models in the brake model-model mapping table; Calculate the similarity between the key attribute feature vector and the standard attribute feature vector; The braking performance evaluation model corresponding to the feature vector with the highest similarity is selected as the target braking performance evaluation model.

4. The braking performance evaluation method based on machine learning according to claim 3, characterized in that, After selecting the braking performance evaluation model with the highest similarity as the target braking performance evaluation model, the method further includes: Record this matching event and generate a mapping record including the brake identification data, the identifier of the selected target braking performance evaluation model, and the similarity; When the cumulative number of the mapping records is greater than or equal to a preset second threshold, or when the number of times the evaluation results output by the target braking performance evaluation model are verified as valid is greater than or equal to a preset third threshold, the mapping records are converted into permanent mapping relationships and updated to the brake model-model mapping table.

5. The braking performance evaluation method based on machine learning according to claim 4, characterized in that, Also includes: Based on the vehicle identification code sequence, permanent geographical location information or historical driving pattern data, the vehicles are divided into different vehicle clusters using a clustering algorithm. The success rate of converting the mapping records of the vehicle cluster into permanent mapping relationships is statistically analyzed. Based on the update success rate, adjust the second threshold and the third threshold.

6. The braking performance evaluation method based on machine learning according to claim 1, characterized in that, The method of constructing a multi-source information fusion correction factor library based on the second braking data includes: A pre-defined hierarchical correction rule set is provided, which includes a first-level rule subset, a second-level rule subset, and a third-level rule subset. Based on the motion signals in the vehicle state data and the external environment data, the rules in the first layer rule subset are matched and executed to obtain the basic physical layer correction factor. Based on the load signal in the vehicle status data, the driver operation data, and the basic physical layer correction factor, the rules in the second layer rule subset are matched and executed to obtain the system response layer correction factor. Based on the basic physical layer correction factor and the system response layer correction factor, the rules in the third layer rule subset are matched and executed to obtain the comprehensive performance layer correction factor. The correction factor library is obtained by integrating the basic physical layer correction factor, the system response layer correction factor, and the comprehensive performance layer correction factor.

7. The braking performance evaluation method based on machine learning according to claim 6, characterized in that, The calibration of the primary evaluation result based on the comparison result and the correction factor library to obtain the braking performance evaluation result includes: If the comparison result indicates that the confidence level is greater than or equal to the first threshold, then the first calibration mode is executed; If the comparison result indicates that the confidence level is less than the first threshold, then the second calibration mode is executed; The first calibration mode includes: using the primary evaluation result as the main correction benchmark, using the comprehensive performance layer correction factor in the correction factor library as the weight adjustment factor, and performing weighted correction on the main correction benchmark to obtain the braking performance evaluation result; The second calibration mode includes: using the primary evaluation results as auxiliary reference information and using the correction factor library as the main decision information; The braking performance evaluation result is obtained by comprehensively deriving the auxiliary reference information and the main decision information.

8. A braking performance evaluation system based on machine learning, characterized in that, include: The data acquisition module is used to acquire first braking data and second braking data. The first braking data includes brake wear data, braking execution status data and brake identification data. The feature fusion module is used to extract and fuse features from the braking execution state data and the brake wear data to generate a braking feature vector. The model matching module is used to select the corresponding pre-trained braking performance evaluation model based on the brake identification data to obtain the target braking performance evaluation model. The preliminary evaluation module is used to input the braking feature vector into the target braking performance evaluation model to obtain the preliminary evaluation results and confidence level. The confidence comparison module is used to compare the confidence level with a preset first threshold to obtain a comparison result; The result calibration module is used to construct a correction factor library based on the second braking data for multi-source information fusion. The primary evaluation results are calibrated based on the comparison results and the correction factor library to obtain the braking performance evaluation results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images