Model parameter determination method, vehicle abnormal sound classification method, device and equipment

Abstract

This application discloses a method for determining model parameters, a method for classifying vehicle abnormal noises, an apparatus, and a device. The method includes: acquiring a training dataset of vehicle abnormal noise features; generating an initial population based on the training dataset; the population containing multiple individuals; constructing a kernel principal component analysis (KPC) model based on preset model parameters; the KPC model is used to extract target abnormal noise feature vectors from the vehicle's abnormal noise features for vehicle abnormal noise classification; determining an objective function based on the KPC model; iteratively updating the initial population based on the objective function to obtain an updated population; determining the optimal individual from the updated population; and determining the target model parameters of the KPC model based on the optimal individual and a preset parameter range. This method can determine the target model parameters of the KPC model most suitable for the current abnormal noise environment, thereby improving the classification accuracy of vehicle abnormal noise classification.

Description

Technical Field

[0001] This application belongs to the field of vehicles, specifically relating to a method for determining model parameters, a method for classifying vehicle abnormal noises, and devices and equipment. Background Technology

[0002] Against the backdrop of the automotive industry's intelligent transformation, vehicle noise detection technology is undergoing a paradigm shift from traditional "experience-based" to intelligent "data-driven." Noise classification technology is a core component for achieving accurate diagnosis and efficient repair of vehicle noises. In the process of noise identification and classification, the selection of acoustic signal features has a decisive impact on the detection results. Since original acoustic features are often high-dimensional and redundant, dimensionality reduction methods are typically needed to remove irrelevant features and reduce computational complexity.

[0003] While commonly used acoustic signal dimensionality reduction algorithms can achieve a certain degree of feature compression in related technologies, they still suffer from insufficient generalization ability and difficulty in parameter tuning in practical applications. This can easily affect the accuracy of abnormal noise classification and limit the performance of abnormal noise detection. Summary of the Invention

[0004] The purpose of this application is to provide a method for determining model parameters, a method for classifying vehicle abnormal noises, an apparatus, and a device that can solve the problems in related technologies such as insufficient generalization ability of sound signal dimensionality reduction algorithms, difficulty in parameter tuning, and thus affecting the accuracy of vehicle abnormal noise classification.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows: In a first aspect, embodiments of this application provide a method for determining model parameters, the method comprising: Obtain a training dataset for the characteristics of abnormal noises from vehicles; An initial population is generated based on the training dataset of the abnormal noise characteristics; the population contains multiple individuals. A kernel principal component analysis model is constructed based on preset model parameters; the kernel principal component analysis model is used to extract target abnormal noise feature vectors from the abnormal noise features of the vehicle, so as to classify the abnormal noise of the vehicle based on the target abnormal noise feature vectors. The objective function is determined based on the kernel principal component analysis model. The initial population is iteratively updated according to the objective function to obtain the updated population; Determine the optimal population individual from the population individuals of the updated population; The target model parameters of the kernel principal component analysis model are determined based on the optimal population individuals and the preset parameter range.

[0006] Optionally, generating an initial population based on the abnormal noise feature training dataset includes: Generate corresponding mapping feature data based on the abnormal noise feature training dataset; Based on a preset population size, mapping feature data that meets preset conditions are randomly selected as individuals in the population to generate an initial population; the preset condition is that the individuals in the population are located within a preset search space.

[0007] Optionally, the step of iteratively updating the initial population according to the objective function to obtain the updated population includes: The initial population is determined as the current population, and corresponding population information is generated; the population information includes the current iteration number; For each individual in the current population, one method is randomly selected from the preset update methods to update the population, thus obtaining the updated population individual. The current population is updated based on the updated population individuals; If the current iteration count is less than the preset maximum iteration count, return to the step of randomly selecting one of the preset update methods for each individual in the current population to update the population, thereby obtaining the updated population individuals; When the current iteration number equals the preset maximum iteration number, the updated population is obtained.

[0008] Optionally, the update method includes communication behavior update, wherein for each individual in the current population, one method is randomly selected from a preset update method for update to obtain an updated population individual, including: From the current population, determine a first target population individual and a second target population individual; the second target population individual is a random population individual from the current population other than the first target population individual; The adaptive iteration factor is determined based on the current iteration number and the preset maximum iteration number. The first updated population individual corresponding to the first target population individual is determined based on the adaptive iteration factor and the second target population individual; The fitness score of the first updated population individual is determined according to the objective function; If the fitness score of the first updated population individual is greater than the fitness score of the first target population individual, the first target population individual is updated to the first updated population individual.

[0009] Optionally, the update method includes foraging behavior update, wherein for each individual in the current population, one method is randomly selected from a preset update method for updating to obtain an updated population individual, including: The fitness score of each individual in the contemporary population is determined according to the objective function. The optimal individuals in the current population are determined based on the fitness scores. The target search range is determined based on the maximum number of iterations. The second updated population individual is determined based on each of the population individuals in the current population, the optimal population individual, and the target search range; The fitness score of the individuals in the second updated population is determined according to the objective function; If the fitness score of the second updated population individual is greater than the fitness score of the population individual, the population individual is updated to the second updated population individual.

[0010] Optionally, determining the target model parameters of the kernel principal component analysis model based on the optimal population individuals and the preset parameter range includes: Candidate model parameters are determined from the preset parameter range according to the preset value step size; The fitness score of the optimal population individual is determined based on the candidate model parameters and the objective function; The candidate model parameters with the highest fitness scores among the individuals in the optimal population are determined as the target model parameters.

[0011] This application also provides a method for classifying vehicle abnormal noises, the method comprising: Obtain the original abnormal noise signal of the vehicle; Feature extraction is performed on the original abnormal noise signal to obtain abnormal noise feature data; The abnormal noise feature data is input into the kernel principal component analysis model to obtain the target abnormal noise feature vector; The abnormal noise feature vector of the target is classified to obtain the abnormal noise classification result; The kernel principal component analysis model is constructed based on the target model parameters.

[0012] Secondly, embodiments of this application provide a model parameter determination apparatus, the apparatus comprising: The first acquisition module is used to acquire the training dataset of abnormal noise features of the vehicle; The population generation module is used to generate an initial population based on the abnormal noise feature training dataset; the population contains multiple population individuals; The model building module is used to build a kernel principal component analysis model based on preset model parameters; the kernel principal component analysis model is used to extract target abnormal noise feature vectors from the abnormal noise features of the vehicle, so as to classify the abnormal noise of the vehicle based on the target abnormal noise feature vectors. The objective function determination module is used to determine the objective function based on the kernel principal component analysis model. The population update module is used to iteratively update the initial population according to the objective function to obtain an updated population. The optimal population individual determination module is used to determine the optimal population individual from the population individuals of the updated population. The target model parameter determination module is used to determine the target model parameters of the kernel principal component analysis model based on the optimal population individuals and the preset parameter range.

[0013] This application embodiment also provides a vehicle abnormal noise classification device, the device comprising: The second acquisition module is used to acquire the original abnormal noise signal of the vehicle; The feature extraction module is used to extract features from the original abnormal noise signal to obtain abnormal noise feature data; The input module is used to input the abnormal noise feature data into the kernel principal component analysis model to obtain the target abnormal noise feature vector; The classification module is used to classify the target abnormal noise feature vector to obtain the abnormal noise classification result; The kernel principal component analysis model is constructed based on the target model parameters.

[0014] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0015] The model parameter determination method provided in this application can obtain a training dataset of vehicle abnormal noise features; then generate an initial population based on the abnormal noise feature training dataset, the population containing multiple individuals; construct a kernel principal component analysis model based on preset model parameters, the kernel principal component analysis model is used to extract target abnormal noise feature vectors from the vehicle's abnormal noise features, so as to classify the vehicle's abnormal noises based on the target abnormal noise feature vectors; determine the objective function based on the kernel principal component analysis model; then iteratively update the initial population according to the objective function until an updated population is reached; determine the optimal individual from the updated population; finally, determine the target model parameters of the kernel principal component analysis model based on the optimal individual and the preset parameter range. In this embodiment, by constructing a kernel principal component analysis model that can be used for dimensionality reduction of sound signals, and obtaining the objective function for iterative updating of the initial population corresponding to the training data of abnormal noise features based on the kernel principal component analysis model, the optimal population individuals obtained by updating the initial population can be used to determine the target model parameters of the kernel principal component analysis model most suitable for the current abnormal noise environment, thereby improving the parameter tuning and generalization ability of the kernel principal component analysis model and ensuring the accuracy of vehicle abnormal noise classification.

[0016] The vehicle abnormal noise classification method provided in this application can acquire the original abnormal noise signal of a vehicle; then extract features from the original abnormal noise signal to obtain abnormal noise feature data; then input the abnormal noise feature data into a kernel principal component analysis model to obtain a target abnormal noise feature vector; and finally classify the target abnormal noise feature vector to obtain the abnormal noise classification result. The kernel principal component analysis model is constructed based on the target model parameters of claim 1. In this application embodiment, by using the kernel principal component analysis model constructed with the target model parameters to perform dimensionality reduction processing on the abnormal noise feature data extracted from the original abnormal noise signal, it is possible to effectively extract the most relevant and representative features in the abnormal noise feature data that are most relevant to the fault category, thereby improving the stability and reliability of subsequent abnormal noise classification. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the steps of an embodiment of a model parameter determination method according to this application; Figure 2 This is a flowchart illustrating the steps of another embodiment of the model parameter determination method of this application; Figure 3 This is a flowchart illustrating the steps of another embodiment of the model parameter determination method of this application; Figure 4 This is a flowchart illustrating the steps of another embodiment of the model parameter determination method of this application; Figure 5 This is a flowchart illustrating the steps of an embodiment of a vehicle abnormal noise classification method according to this application; Figure 6 This is a flowchart illustrating the steps of another embodiment of the vehicle abnormal noise classification method of this application; Figure 7 This is a structural block diagram of an embodiment of a model parameter determination device according to this application; Figure 8 This is a structural block diagram of an embodiment of a vehicle abnormal noise classification device according to this application. Detailed Implementation

[0019] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0021] The method for determining model parameters provided in this application will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0022] Against the backdrop of the automotive industry's intelligent transformation, and facing the core bottlenecks of traditional manual diagnostic methods such as low efficiency and high subjectivity, developing highly automated and precise abnormal noise detection technology has become a key path to improve after-sales service efficiency and ensure driving safety. Abnormal noise classification technology, as a core component for achieving accurate diagnosis and efficient repair, determines the accuracy of fault location and the generation of repair strategies.

[0023] Abnormal noise detection, as a key technology for product quality assessment and fault early warning, can identify and locate abnormal sounds generated during equipment operation, thereby effectively revealing potential hidden quality problems such as assembly defects, structural loosening, component wear, or design flaws within the vehicle. In the process of abnormal noise identification and classification, since the original acoustic features are often high-dimensional and redundant, dimensionality reduction methods are typically needed to remove irrelevant features and reduce computational complexity, thereby improving the performance and generalization ability of the classification model.

[0024] Currently, commonly used acoustic signal dimensionality reduction algorithms, such as the t-distribution random neighborhood embedding algorithm and the local linear embedding algorithm, can achieve feature compression to a certain extent, but in practical applications, they still have significant problems such as insufficient generalization ability and difficulty in parameter tuning. This can easily lead to excessive loss of effective discriminative information and excessive residual redundant noise features in the dimensionality-reduced features, ultimately resulting in a decline in feature representation ability, which in turn affects the accuracy of vehicle abnormal noise classification and restricts the performance of the entire abnormal noise detection system.

[0025] To address the aforementioned issues, this application provides a method for determining model parameters that can identify the target model parameters of the kernel principal component analysis model best suited for the current abnormal noise environment. This method enables parameter tuning of the kernel principal component analysis model and improves its generalization ability, thereby ensuring the accuracy of vehicle abnormal noise classification.

[0026] Reference Figure 1 This is a flowchart illustrating the steps of an embodiment of a model parameter determination method according to this application. The method includes the following steps: Step 101: Obtain the training dataset of abnormal noise features of the vehicle; The training dataset for vehicle abnormal noise features can be a feature set obtained by collecting internal audio signals of the vehicle during operation, obtaining abnormal noise signals from the collected audio signals, and then extracting features from the abnormal noise signals. Alternatively, it can be a pre-prepared training dataset that includes one or more representative vehicle abnormal noise features. This application does not impose any specific restrictions on this.

[0027] Step 102: Generate an initial population based on the abnormal noise feature training dataset; the population contains multiple individuals. The initial population can refer to the population that has not yet started iterative updates.

[0028] Step 103: Construct a kernel principal component analysis model based on preset model parameters; the kernel principal component analysis model is used to extract target abnormal noise feature vectors from the abnormal noise features of the vehicle, so as to classify the abnormal noise of the vehicle based on the target abnormal noise feature vectors; The kernel principal component analysis model can refer to the kernel principal component analysis (KPCA) function model. This model can transform the data in the original input space to a high-dimensional feature space through nonlinear mapping, and then perform linear principal component analysis in this space. KPCA can avoid explicit computation of high-dimensional mapping by using kernel functions, significantly improving computational efficiency. Common kernel functions include, but are not limited to, polynomial kernel functions, sigmoid kernel functions, and Gaussian radial basis function kernel functions. For example, since the Gaussian radial basis function kernel function has simple parameters and strong nonlinear mapping capabilities, this application can choose the Gaussian radial basis function kernel function as the kernel function corresponding to the kernel principal component analysis model, and the preset model parameters can include the kernel bandwidth of the Gaussian radial basis function kernel function. The target abnormal noise feature vector can be the most relevant and representative feature vector of the abnormal noise category obtained after the kernel principal component analysis model performs dimensionality reduction on the abnormal noise features of the vehicle.

[0029] Step 104: Determine the objective function based on the kernel principal component analysis model; The objective function can be used to guide the iterative evolution direction of the initial population, thereby enabling the kernel principal component analysis model to find the optimal projection direction to achieve accurate classification of vehicle noises.

[0030] Step 105: Iteratively update the initial population according to the objective function to obtain the updated population; The initial population can be iteratively updated using intelligent optimization algorithms, including the Parrot Optimizer (PO). The Parrot Optimizer (PO) is an emerging heuristic optimization algorithm that simulates the foraging, lingering, communication, and fear behaviors of parrots. Alternatively, other intelligent optimization algorithms such as Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO) can also be used to update the initial population; this application does not impose specific limitations on this approach.

[0031] Step 106: Determine the optimal population individual from the population individuals of the updated population; Step 107: Determine the target model parameters of the kernel principal component analysis model based on the optimal population individuals and the preset parameter range.

[0032] The preset parameter range can be set based on historical experience, or it can be set according to the mean μ and standard deviation σ corresponding to the abnormal noise features in the abnormal noise feature training dataset. For example, it can be set to [σ / 10, 10σ] or [μ [2σ,μ+2σ], etc., or first determine a relatively broad initial parameter range, and then find the interval where the numerical change trend of the objective function is obvious, and use the interval as the preset parameter range. Alternatively, it can be set separately according to actual needs. This application does not impose specific restrictions on this.

[0033] In this embodiment, a training dataset of vehicle noise features can be obtained. An initial population is then generated based on this dataset, containing multiple individuals. A kernel principal component analysis (KPCA) model, such as a KPCA model based on a Gaussian radial basis function kernel, can be constructed using preset model parameters. This KPCA model can extract target noise feature vectors from the vehicle's noise features, enabling noise classification based on these vectors. An objective function can be determined using the KPCA model. The objective function then guides the iterative evolution of the initial population, updating it until an updated population is reached. The optimal individual is then selected from the updated population. Finally, the target model parameters of the KPCA model are determined based on the optimal individual and a preset parameter range. Through the above implementation process, a kernel principal component analysis model capable of performing dimensionality reduction of sound signals is constructed. Based on the kernel principal component analysis model, the objective function for iterative updating of the initial population corresponding to the abnormal noise feature training data is obtained. By updating the initial population to obtain the optimal population individuals, the target model parameters of the kernel principal component analysis model most suitable for the current abnormal noise environment can be determined, thereby improving the parameter tuning and generalization ability of the kernel principal component analysis model and ensuring the accuracy of vehicle abnormal noise classification.

[0034] In some embodiments of this application, generating an initial population based on the abnormal noise feature training dataset includes: Generate corresponding mapping feature data based on the abnormal noise feature training dataset; Based on a preset population size, mapping feature data that meets preset conditions are randomly selected as individuals in the population to generate an initial population; the preset condition is that the individuals in the population are located within a preset search space.

[0035] In related technologies, population generation typically involves random initialization of individuals. However, this random initialization method can lead to high similarity and a lack of diversity among the initial population individuals, making optimization algorithms prone to getting trapped in local optima. In some embodiments of this application, chaotic mapping can be applied to the abnormal noise feature data contained in the training dataset to generate corresponding mapped feature data. Chaotic mapping exhibits high randomness and uncertainty and can include Tent chaotic mapping (piecewise linear mapping), Logistic chaotic mapping (one-dimensional discrete chaos), Circle chaotic mapping (a nonlinear mapping), Henon chaotic mapping (a two-dimensional discrete chaotic mapping), and so on. For example, a relatively stable Circle chaotic mapping, which generates a relatively uniformly distributed population, can be used to map the abnormal noise feature data. A conventional Circle chaotic mapping can be expressed by the following formula:

[0036] in, It can represent the mapping feature data obtained from the mapping; This can represent abnormal noise characteristic data. The result is generated through Circle chaotic mapping. It always falls within the interval [0,1); n can represent the dimension of the solution.

[0037] However, conventional Circle chaotic mapping pulls the mapped values ​​towards a certain interval with a specific probability, resulting in a relatively dense distribution of the generated initial population within a certain value range, such as [0.2, 0.5]. The improved Circle chaotic mapping used in this application can be expressed by the following formula:

[0038] in, It can represent the mapping feature data obtained from the mapping; This can represent abnormal noise characteristic data. The result is generated through Circle chaotic mapping. It always falls within the interval [0,1). In the improved Circle chaotic mapping described above, by increasing... By adjusting the coefficients of the sine function and increasing the initial offset, the nonlinear effects in the iteration process can be made stronger, thereby enhancing the randomness and unpredictability of the chaotic sequence and making the distribution of the mapped feature data more uniform.

[0039] The initial population can be generated by randomly selecting mapped feature data within a preset search space range based on a preset population size. The population size represents the number of individuals in the population, reflecting the population's size; it is typically set to 30, but can be set differently depending on actual needs. This application does not impose specific limitations on this. The search space range reflects the allowed value range of individuals in the population during its iterative evolution. The boundaries of the search space range include an upper and lower limit, used to define the critical values ​​corresponding to the individual population values. The search space range can be set based on experience or actual needs; this application does not impose specific limitations on this.

[0040] For example, the abnormal noise feature data can be subjected to Circle chaotic mapping to obtain mapped feature data, and then the mapped feature data that meets the requirements of population size and search space range can be selected from the mapped feature data to form an initial population; alternatively, feature data that meets the requirements of population size and search space range can be selected from the abnormal noise feature data first, and then the selected abnormal noise feature data can be subjected to Circle chaotic mapping to obtain mapped feature data. If the obtained mapped feature data does not meet the requirements of search space range, features that meet the conditions can be selected from the remaining abnormal noise feature data to fill in the gaps. This application does not impose specific restrictions on this.

[0041] Through the above implementation process, a more uniformly distributed and more diverse initial population can be created using chaotic mapping, thus preventing the population from getting too quickly into local optima during subsequent updates and iterations.

[0042] For example, when the training dataset for abnormal noise features includes two classes of abnormal noise features, the objective function can be an FC discriminant function constructed based on the Fisher Criteria (FC). The objective function can be expressed by the following formula:

[0043] in, This can be expressed as the objective function when the training dataset for abnormal noise features includes two types of abnormal noise features. It can represent the sum of squared errors of the total within-class dispersion of the contemporary population; It can represent the total intra-class dispersion of the current population. Intra-class dispersion can be used to measure the degree of dispersion among samples (individuals) belonging to the same class in the current population. The smaller the intra-class dispersion, the more compact and clustered the samples of the same class are after projection. It can represent the sum of squared errors of the inter-class dispersion of the contemporary population; It can represent the inter-class dispersion of a contemporary population. Inter-class dispersion can be used to measure the degree of separation between the means of samples (individuals) of different classes. The greater the inter-class dispersion, the easier it is to distinguish between classes; T can represent matrix transpose. It can indicate the projection direction. N can represent the population size of the current population.

[0044] The inter-class dispersion of the corresponding contemporary population can be expressed by the following formula:

[0045] in, It can represent the inter-class dispersion of a contemporary population; This can represent the sample mean corresponding to the first type of abnormal noise feature. T can represent the sample mean corresponding to the second type of abnormal sound feature; T can represent the matrix transpose, which means that the column vectors of the matrix are transformed into row vectors through the transpose operation, so that the two vectors can be externally producted.

[0046] The intra-class dispersion of samples corresponding to the i-th type of anomalous feature in the contemporary population can be expressed by the following formula:

[0047] in, can represent the intra-class dispersion of samples corresponding to the i-th type of abnormal sound feature; N can represent the population size of the current population. It can represent the number of individuals in the contemporary population that belong to the i-th type of aberrant trait; This can represent the j-th individual in the contemporary population that belongs to the i-th type of anomalous trait; It can represent the sample mean corresponding to the i-th type of abnormal sound feature; T can represent the matrix transpose.

[0048] When the training dataset for abnormal noise features includes two classes of abnormal noise features, the total intra-class scatter of the corresponding contemporary population can be expressed by the following formula:

[0049] in, It can represent the total intra-class dispersion of the contemporary population; It can represent the intra-class dispersion of samples corresponding to the first type of abnormal noise feature; it can represent the intra-class dispersion of samples corresponding to the second type of abnormal noise feature.

[0050] The sum of squared errors of the total within-class dispersion of the two types of abnormal noise features can be expressed by the following formula:

[0051] in, It can represent the sum of squared errors of the total within-class dispersion of the contemporary population; It can represent the p-th sample vector in the current population; N can represent the population size of the current population; It can represent the mean of the sample corresponding to the i-th type of abnormal noise feature in the projection direction; T can represent the sample mean vector corresponding to the i-th type of abnormal sound feature; T can represent the matrix transpose. It can indicate the projection direction; It can represent the total intraclass dispersion of the contemporary population.

[0052] When the training dataset for abnormal noise features includes two classes of abnormal noise features, the sum of squared errors of the total within-class scatter of the corresponding two classes of abnormal noise features can be expressed by the following formula:

[0053] in, It can represent the sum of squared errors of the inter-class dispersion of the contemporary population; It can represent the mean value of the sample corresponding to the first type of abnormal noise feature in the projection direction; It can represent the mean value of the sample corresponding to the first type of abnormal noise feature in the projection direction; This can represent the sample mean vector corresponding to the first type of abnormal noise feature; It can represent the sample mean vector corresponding to the second type of abnormal noise feature; T can represent the matrix transpose; It can indicate the projection direction; It can represent the inter-class dispersion of a contemporary population.

[0054] For example, when the training dataset for abnormal noise features includes more than two classes of abnormal noise features, instead of calculating the projection difference in a single direction, the global dispersion of the entire data distribution in the feature space can be calculated. The corresponding intra-class dispersion matrix of the current population can be expressed by the following formula:

[0055] in, It can represent the intra-class scatter matrix of the current population; N can represent the population size of the current population; M can represent the number of abnormal noise categories included in the abnormal noise feature training dataset; It can represent the p-th sample vector in the contemporary population; T can represent the sample mean vector corresponding to the i-th type of abnormal sound feature; T can represent the matrix transpose.

[0056] The inter-class scatter matrix of the corresponding contemporary population can be expressed by the following formula:

[0057] in, M can represent the inter-class scatter matrix of the contemporary population; M can represent the number of abnormal noise categories included in the abnormal noise feature training dataset. It can represent the sample mean vector corresponding to the i-th type of abnormal noise feature; T can represent the total mean vector of all samples in the contemporary population; T can represent the matrix transpose.

[0058] When the abnormal noise feature training dataset includes more than two types of abnormal noise features, the corresponding objective function can be expressed by the following formula:

[0059] in, It can represent the objective function when the training dataset for abnormal noise features includes two or more types of abnormal noise features; This can represent calculating the trace of the matrix; It can represent the intra-class scatter matrix of the contemporary population; It can represent the inter-class scatter matrix of the contemporary population.

[0060] Reference Figure 2 This is a flowchart illustrating the steps of another embodiment of the model parameter determination method of this application. In some embodiments of this application, the step of iteratively updating the initial population according to the objective function to obtain the updated population includes: Step 201: Determine the initial population as the current population and generate corresponding population information; the population information includes the current iteration number; Here, the current population represents the composition of the population currently used for update iteration. The current iteration number represents the number of update iterations the current population has undergone. When iterative evolution begins for the first time, the initial population can be used as the current population, at which point the current iteration number corresponding to the current population is 0.

[0061] Step 202: For each individual in the current population, randomly select one of the preset update methods to update it, and obtain the updated population individual; Among them, there can be a variety of preset update methods, including at least the foraging update method that simulates the process of a parrot looking for food, the staying update method that simulates a parrot suddenly flying onto its owner and remaining still for a period of time, the communication update method that imitates information sharing within a parrot group, and the fear update method that imitates a parrot's avoidance reaction to unfamiliar individuals.

[0062] Step 203: Update the current population based on the individuals in the updated population; Step 204: If the current iteration number is less than the preset maximum iteration number, return to the step of randomly selecting one of the preset update methods for each individual in the current population to update the population and obtain the updated individuals. The maximum number of iterations can be used as a termination condition for the update iteration, limiting the number of updates to the population. For example, the maximum number of iterations can be set to 500, or it can be dynamically adjusted according to the iteration update situation, or it can be set separately according to actual needs. This application does not impose specific restrictions on this.

[0063] Step 205: If the current iteration number is equal to the preset maximum iteration number, the updated population is obtained.

[0064] The initial population can be defined as the current population, and corresponding population information can be generated. This information includes the current iteration count and the individual information of the best individual. The best individual's information includes its eigenvalues ​​and the current iteration count at which it was first obtained. For each individual in the current population, a pre-defined update method can be randomly selected to update it, resulting in an updated individual. The current population can then be updated based on this updated individual. If the current iteration count is less than the pre-defined maximum iteration count, the current iteration count can be incremented by one to update the current iteration count. Then, for each individual in the current population, a pre-defined update method can be randomly selected to update it, resulting in an updated individual. This process continues for the next iteration. If the current iteration count equals the pre-defined maximum iteration count, the updated population is obtained. Through the above implementation process, in each iteration of an individual in the population, one of the following methods can be randomly selected for its own iterative update: foraging update, dwelling update, communication update, and fear update. The current population is continuously updated based on the resulting updated individuals, leading to continuous optimization of the individuals within the current population. Ultimately, this results in an updated population that determines the optimal parameters of the kernel principal component analysis model, i.e., the target model parameters. The kernel bandwidth of the Gaussian radial basis function affects the distribution of data after mapping to a higher-dimensional space. A large kernel bandwidth results in an overly compact data distribution; a small kernel bandwidth results in an overly dispersed data distribution. Reconstructing the kernel principal component analysis model using the target model parameters can effectively improve the accuracy of subsequent abnormal sound recognition.

[0065] Reference Figure 3 This is a flowchart illustrating the steps of another embodiment of the model parameter determination method of this application. In some embodiments of this application, the update method includes communication behavior update. The step of randomly selecting one update method from a preset set of update methods for each individual in the current population to obtain the updated population individual includes the following steps: Step 301: Determine individuals from the first target population and individuals from the second target population from the current population; The first target population individual is the individual in the population to be iteratively updated. The second target population individual is a random individual in the current population other than the first target population individual.

[0066] Step 302: Determine the adaptive iteration factor based on the current iteration number and the preset maximum iteration number; The adaptive iteration factor can be represented as a value that can dynamically adjust its value with the number of iterations, thereby adjusting the direction of the population's iterative evolution.

[0067] Step 303: Determine the first updated population individual corresponding to the first target population individual based on the adaptive iteration factor and the second target population individual; Among them, the first updated population individual is the new population individual obtained after updating the first updated population individual; Step 304: Determine the fitness score of the first updated population individual according to the objective function; Step 305: If the fitness score of the first updated population individual is greater than the fitness score of the first target population individual, update the first target population individual to the first updated population individual.

[0068] Individuals from the current population that will be used for iterative updates, including the first target population individuals and randomly selected second target population individuals, can be identified. An adaptive iteration factor can be determined based on the current iteration count and a preset maximum iteration count. This adaptive iteration factor can be expressed by the following formula:

[0069] Where H can represent the adaptive iteration factor; It can represent a random number whose value ranges from [0, 1]. It can represent the maximum number of iterations for a population; H can represent the current iteration number corresponding to the current population. In the early stages of population iteration, the range of H values ​​is wider, which ensures that a strategy with larger step sizes and more varied directions is adopted to update the position information (corresponding to the value of an individual in the population). As the number of iterations increases, the range of H values ​​narrows, and the algorithm adopts a strategy that is more inclined towards local fine-tuning to update the position information, mainly by searching for the better regions explored in the early stages, so as to determine the global optimal position more quickly.

[0070] Then, the first updated population individual corresponding to the first target population individual can be determined based on the adaptive iteration factor corresponding to the current iteration number and the second target population individual. The first updated population individual can be represented by the following formula: ) in, It can represent the first updated population individual obtained through communication behavior, that is, the updated position of the r-th population individual in the initial population on the s-th dimension after the (t+1)-th iteration. H can represent the first target population individual, that is, the updated position of the r-th population individual in the initial population on the s-th dimension after the t-th iteration; H can represent the adaptive iteration factor. It can represent an individual in the second target population; It can represent the current iteration number corresponding to the contemporary population.

[0071] Then, the first target population individual in the current population can be replaced with the first updated population individual. The objective function value is then calculated based on the replaced current population, yielding the fitness score of the first updated population individual. This fitness score is then compared to the fitness score of the first target population individual. If the fitness score of the first updated population individual is greater than that of the first target population individual, the first target population individual is updated to become the first updated population individual. If the fitness score of the first updated population individual is less than or equal to that of the first target population individual, the original first target population individual is retained and awaits the next round of iteration.

[0072] For example, if the first updated population individuals obtained by updating the first target population individuals through communication behavior exceed the preset search space range, the first updated population individuals can be updated according to the upper limit or lower limit of the search space. For instance, if the first updated population individuals are less than the preset lower limit of the search space, the lower limit of the search space is used as the value of the first updated population individuals; if the first updated population individuals are greater than the preset upper limit of the search space, the upper limit of the search space is used as the value of the first updated population individuals.

[0073] In traditional optimization algorithms, the location parameters are randomly generated during iteration, which can lead to insufficient exploration capabilities in the early stages and difficulty in achieving local convergence in the later stages. The above implementation process uses an adaptive iteration factor that dynamically changes with the number of iterations to adjust the update method, effectively balancing the global and local search capabilities of the update method during population iteration. This ensures that in the early stages of population iteration, a strategy with larger step sizes and more varied directions can be used to update location information. As the number of iterations increases, a strategy more focused on local fine-tuning is adopted to update location information, searching for previously explored optimal regions, thereby determining the global optimum more quickly. This, in turn, allows for faster determination of the optimal target model parameters, improving parameter tuning efficiency.

[0074] Reference Figure 4 This is a flowchart illustrating the steps of another embodiment of the model parameter determination method of this application. In some embodiments of this application, the update method includes foraging behavior update. The step of randomly selecting one method from a preset update method for each individual in the current population to update the population and obtain the updated individual includes the following steps: Step 401: Determine the fitness score of each individual in the current population according to the objective function; Fitness scores can be used to evaluate the quality of individuals within a population.

[0075] Step 402: Determine the optimal individuals of the current generation population based on the fitness score; The optimal individual in the population can be the individual with the highest fitness score in the current population.

[0076] Step 403: Determine the target search range based on the maximum number of iterations; Step 404: Determine the second updated population individual based on each of the population individuals in the current population, the optimal population individual, and the target search range; Step 405: Determine the fitness score of the second updated population individuals according to the objective function; Step 406: If the fitness score of the second updated population individual is greater than the fitness score of the population individual, update the population individual to the second updated population individual.

[0077] The fitness score of each individual in the current population can be determined based on the objective function. The individual with the highest fitness score in the current population can be identified as the optimal individual. Then, the individual information of the optimal individual can be added to the population information of the current population to update it. The target search range can be determined based on the maximum number of iterations. The target search range can be expressed by the following formula:

[0078]

[0079] Where a and b represent the search range of the golden ratio. It can represent the current iteration number corresponding to the current population; It can represent the total number of iterations of the population, i.e., the preset maximum number of iterations.

[0080] Then, by combining the golden sine strategy, the second updated population individual can be determined based on each individual in the current population, the optimal population individual, and the target search range. The second updated population individual can be represented by the following formula:

[0081]

[0082]

[0083] in, It can represent the second updated population individual obtained through foraging behavior, that is, the updated position of the r-th population individual in the initial population on the s-th dimension after the (t+1)-th iteration. It can represent the original population individual, that is, the updated position of the r-th population individual in the initial population in the s-th dimension after the t-th iteration; The range can be represented as random numbers, The range can be represented as Random numbers; It can represent the best individual in the current population; This can represent the current iteration number corresponding to the contemporary population; a and b are the search range of the golden ratio; It can represent a value The constant is the negative of the golden ratio.

[0084] Then, individuals from the second-updated population can replace existing individuals in the current population. The objective function is then calculated based on the replaced individuals to obtain the fitness score of the second-updated population. This fitness score is then compared to the fitness scores of the existing individuals. If the fitness score of the second-updated population is greater than that of the existing individuals, the existing individuals are updated to become second-updated individuals. If the fitness score of the second-updated population is less than or equal to that of the existing individuals, the existing individuals are retained for the next iteration.

[0085] Through the above implementation process, the optimal individual in the current population can be determined, and the target search range can be determined based on the relative positional relationship between the current iteration number and the maximum iteration number. Then, by utilizing the volatility of the sine function and the characteristics of the golden ratio, a reasonable search direction and search step size can be determined based on the target search range. This allows for the updating of the individuals in the current population, bringing them closer to the optimal individual, thus obtaining the second updated population. Compared to the foraging behavior update method in the traditional parrot algorithm, the golden sine strategy for updating the foraging behavior of individuals in the population requires fewer parameters, simplifying the parameter tuning process, reducing the complexity of the update, and maintaining a relatively stable convergence speed. It avoids skipping the optimal solution due to an excessively large step size, and also avoids getting trapped in local optima due to an excessively small step size. This allows for faster determination of the optimal target model parameters and improves the quality of parameter tuning.

[0086] For example, the update method also includes dwell behavior update, whereby for each individual in the current population, one method is randomly selected from a preset set of update methods to update the population, resulting in an updated individual, including: The third update population individuals are determined based on the population individuals in the current population and the best population individual.

[0087] The fitness score of individuals in the third update population is determined based on the objective function; If the fitness score of an individual in the third update population is greater than the fitness score of an individual in the population, the individual in the population is updated to the third update population.

[0088] The third update of individuals in the population can be represented by the following formula:

[0089] in, It can represent the third updated population individual obtained through dwell behavior, that is, the updated position of the r-th population individual in the initial population after the (t+1)-th iteration; It can represent an existing individual in the population, that is, the updated position of the r-th individual in the initial population after the t-th iteration; It can represent the best individual in the current population; This indicates the parrot's flight behavior, and dim represents the number of parameters in the function being sought, which corresponds to the population size. It can represent a random number whose value ranges from [0, 1]. The dimension is A vector of all 1s.

[0090] After determining the third-update population individuals, the existing population individuals in the current population can be replaced with these third-update individuals. Then, the objective function value is calculated based on the replaced population, yielding the fitness score of the third-update individual. This fitness score is then compared to the fitness scores of the existing population individuals. If the fitness score of the third-update individual is greater than that of the existing population individuals, the existing population individuals are updated to become third-update individuals. If the fitness score of the third-update individual is less than or equal to that of the existing population individuals, the existing population individuals are retained for the next iteration. Through this process, the population individuals in the current population can be updated using the dwell behavior update method. This allows for a thorough search in the region where the optimal population individual was found, thereby improving the convergence quality of the population.

[0091] For example, the update method also includes fear behavior update, whereby for each individual in the current population, one method is randomly selected from a preset set of update methods to update the population, resulting in an updated individual, including: Determine the current iteration number corresponding to the current population based on population information; The fourth updated population individual is obtained by updating the population individual based on the maximum number of iterations in the current population, the best population individual, and the current number of iterations. The fitness score of the fourth update population individual is determined based on the objective function; If the fitness score of an individual in the fourth update population is greater than the fitness score of an individual in the population, the individual in the population is updated to the fourth update population.

[0092] The fourth update of the population can be represented by the following formula:

[0093] in, It can represent the fourth updated population individual obtained through fear behavior, that is, the updated position of the r-th population individual in the initial population after the (t+1)-th iteration. It can represent an existing individual in the population, that is, the updated position of the r-th individual in the initial population after the t-th iteration; It can represent the best individual in the current population; It can represent a random number whose value ranges from [0, 1]. It can represent the current iteration number corresponding to the current population; It can represent the maximum number of iterations corresponding to the current population.

[0094] After determining the fourth update population individuals, the existing population individuals in the current population can be replaced with these fourth update individuals. Then, the objective function value is calculated based on the replaced population, yielding the fitness score of the fourth update individual. This fitness score is then compared to the fitness scores of the existing population individuals. If the fourth update individual's fitness score is higher than the existing population individual's fitness score, the existing population individual is updated to the fourth update population individual. If the fourth update individual's fitness score is less than or equal to the existing population individual's fitness score, the existing population individual is retained for the next iteration. Through this process, the fear behavior update method can be used to update the population individuals in the current population, preventing populations from clustering in the same area for too long, thus avoiding optimization stagnation, breaking the homogeneity trend of the population, improving global search capabilities, and ultimately enhancing the algorithm's generalization ability.

[0095] For example, for each individual in the current population, one method is randomly selected from a preset set of update methods for updating. After obtaining the updated population individuals, an adaptive Cauchy-Gaussian mutation perturbation can be used to mutate the best individual in the updated population to obtain the mutated optimal individual. The mutated optimal individual can be expressed by the following formula:

[0096] in, It can represent the optimal individual in the mutation; It can represent the best individual in the updated population; and Let represent random variables that satisfy Cauchy and Gaussian distributions, respectively; and This is a dynamic coefficient.

[0097]

[0098]

[0099] in, It can represent the current iteration number corresponding to the current population; It can represent the maximum number of iterations corresponding to the current population.

[0100] After determining the optimal mutant individual, the original optimal individual in the updated population can be replaced with the optimal mutant individual. Then, the objective function is calculated based on the replaced population to obtain the fitness score of the optimal mutant individual. The fitness score of the optimal mutant individual is then compared with the fitness score of the original optimal individual in the updated population. If the fitness score of the optimal mutant individual is greater than that of the original optimal individual, the original optimal individual is updated to the optimal mutant individual. If the fitness score of the optimal mutant individual is less than or equal to that of the original optimal individual, the original optimal individual is retained for the next round of iteration.

[0101] In the later stages of population iteration, the PO algorithm is prone to getting trapped in local optima due to the decrease in individual diversity within the population. Through the aforementioned implementation process, Cauchy-Gaussian mutation perturbation can be used to perturb the optimal individuals in the updated population during each iteration. In the early stages of iteration, Cauchy mutation is emphasized to maintain population diversity and enable rapid global search; in the later stages, Gaussian mutation is emphasized to improve the algorithm's ability to discover local optima and quickly escape local optima.

[0102] In some embodiments of this application, determining the target model parameters of the kernel principal component analysis model based on the optimal population individuals and a preset parameter range includes: Candidate model parameters are determined from the preset parameter range according to the preset value step size; The fitness score of the optimal population individual is determined based on the candidate model parameters and the objective function; The candidate model parameters with the highest fitness scores among the individuals in the optimal population are determined as the target model parameters.

[0103] The step size defines the interval for selecting candidate model parameters within the parameter range. This step size can be set according to actual needs, and this application does not impose specific restrictions on it. Candidate model parameters can be determined from a preset parameter range based on the preset step size. Then, each selected candidate model parameter can be iterated over, and the fitness score corresponding to the optimal individual in the updated population after iterative updates can be calculated for each candidate model parameter. The candidate model parameter with the highest fitness score among the optimal individuals is determined as the target model parameter. Through the above implementation process, the target model parameters of the kernel principal component analysis model most suitable for the current abnormal noise environment can be determined from a preset parameter range based on the optimal individuals in the updated population. This achieves parameter tuning of the kernel principal component analysis model and improves its generalization ability, thereby ensuring the accuracy of vehicle abnormal noise classification.

[0104] Reference Figure 5 This is a flowchart illustrating the steps of an embodiment of a vehicle abnormal noise classification method according to this application. The method includes the following steps: Step 501: Obtain the original abnormal noise signal of the vehicle; Among them, the original abnormal noise signal of the vehicle can be an audio signal including abnormal noise generated inside the vehicle during operation.

[0105] Step 502: Extract features from the original abnormal noise signal to obtain abnormal noise feature data; The abnormal noise feature data includes, but is not limited to, Mel spectrum features, zero-crossing rate features, power spectral density, Fourier transform features, etc., and can be extracted using the corresponding audio feature extraction method according to actual needs. This application does not impose specific restrictions on this.

[0106] Step 503: Input the abnormal noise feature data into the kernel principal component analysis model to obtain the target abnormal noise feature vector; The target abnormal noise feature vector can be obtained by dimensionality reduction of the abnormal noise feature data using a kernel principal component analysis model, and is the most relevant and representative feature vector to the abnormal noise category. Step 504: Classify the target abnormal noise feature vector to obtain the abnormal noise classification result; The kernel principal component analysis model is constructed based on the target model parameters.

[0107] The original abnormal noise signal generated during vehicle operation can be obtained from a vehicle undergoing fault testing or repair. Then, feature extraction is performed on the original abnormal noise signal to obtain abnormal noise feature data. For example, feature extraction can be performed based on Mel frequency cepstral coefficients, or it can be performed using a pre-trained deep learning model; this application does not impose specific limitations in this regard. Inputting the abnormal noise feature data into a kernel principal component analysis model constructed based on the target model parameters yields the target abnormal noise feature vector.

[0108] For example, after inputting the abnormal noise feature data into the kernel principal component analysis model constructed based on the target model parameters, the model first converts the abnormal noise feature data from the low-dimensional space. Mapping to higher-dimensional space For example, the dataset of abnormal noise features Through nonlinear mapping Obtain the mapped sample dataset , Then you can map the sample dataset. Decentralization is performed to obtain the target sample data. The mapped sample dataset is then processed. Decentralization can be represented by the following formula:

[0109] Where n can represent the number of features in the abnormal noise feature dataset; It can represent the i-th feature data in the abnormal noise feature dataset; It can represent the i-th feature data in the mapped sample dataset; .

[0110] The covariance matrix of the mapped sample data can then be determined, and it can be expressed by the following formula:

[0111] in, It can represent the covariance matrix of the mapped sample data, and n can represent the number of features in the abnormal feature dataset; It can represent the k-th feature data in the abnormal noise feature dataset; It can represent the k-th feature data in the mapped sample dataset; It can represent a mapped sample dataset; T can represent the matrix transpose; .

[0112] The covariance matrix of the mapped sample data can be decomposed using eigenvalues. The result of the decomposition can be expressed by the following formula:

[0113] in, It can represent the feature values ​​corresponding to the mapped sample data; It can represent The corresponding feature vector, n can represent the number of features in the abnormal noise feature dataset; T can represent a mapped sample dataset; T can represent the matrix transpose.

[0114] Suppose there exists a set of coefficients ,make Corresponding feature vectors and mapping sample datasets The following relationship must be satisfied:

[0115] The eigenvalue decomposition result of the covariance matrix of the mapped sample data can then be expressed by the following formula:

[0116]

[0117] Furthermore, a target kernel function can be introduced, which is a Gaussian radial basis function constructed based on the target model parameters, and a kernel function can be defined. kernel matrix K To simplify calculations, the array elements of this kernel matrix can be represented by the following formula:

[0118] in, It can represent the array element in the i-th row and j-th column of the kernel matrix. , ; It can represent the Gaussian radial basis kernel function constructed based on the parameters of the target model; It can represent the i-th feature data in the mapped sample dataset; It can represent the j-th feature data in the mapped sample dataset.

[0119] Based on the target model parameters Constructed Gaussian radial basis kernel function This can be expressed by the following formula:

[0120] Kernel matrix Decentralization yields a decentralized kernel function matrix. Centralized kernel function matrix This can be expressed by the following formula:

[0121]

[0122] After introducing kernel functions, they can effectively handle nonlinear data, eliminating the need for nonlinear mappings. Therefore, the eigenvalue decomposition result of the covariance matrix of the mapped sample data can be expressed as follows:

[0123] Further results were obtained:

[0124] in, It can represent the eigenvalues ​​corresponding to the kernel matrix K, and satisfy the following: ; It can represent eigenvalues The corresponding feature vector.

[0125] The feature data in the mapped sample dataset can be used to map the k-th feature vector. The projection onto the surface is used as the candidate fraction vector, which can be expressed by the following formula:

[0126] in, It can represent a candidate score vector. It can represent the k-th eigenvector. It can represent the feature data corresponding to the candidate score vector. It can represent the input mapping sample data.

[0127] After determining the candidate component vectors, the contribution rate of each candidate component vector can be calculated. The contribution rate of a candidate component vector can be expressed by the following formula:

[0128] in, It can represent the contribution rate of the k-th candidate vector. It can represent the k-th eigenvalue; n can represent the total number of candidate subvectors.

[0129] Then, the contribution rates of the candidate sub-vectors can be sorted from largest to smallest to obtain a sequence of candidate sub-vector contribution rates. These contribution rates can be accumulated from largest to smallest. If the ratio of the total contribution rate of the top k candidate sub-vectors to the total contribution rate of all candidate sub-vectors is greater than a preset total contribution rate threshold, then the current accumulated number k can be determined as the number of target abnormal noise feature vectors. The total contribution rate threshold can be set to 0.9, or it can be set according to actual needs; this application does not impose specific restrictions on this. The number of target abnormal noise feature vectors can be determined by the following formula:

[0130] in, can represent the i-th feature value; N can represent the total number of candidate feature vectors; k can represent the number of target abnormal feature vectors; It can represent a preset total contribution rate threshold.

[0131] After determining the target abnormal noise feature vector, the target abnormal noise feature vector can be classified to obtain the abnormal noise classification result. Classification of the target abnormal noise feature vector can be achieved through a pre-trained SVM vector machine or other pre-trained classification models; this application does not impose specific limitations on this. Through the above implementation process, a kernel principal component analysis model can be constructed using the target model parameters to quickly and effectively reduce the dimensionality of the abnormal noise feature data, obtaining the most relevant and representative target abnormal noise feature vectors. Then, classification and identification can be performed based on the target abnormal noise feature vectors to determine the abnormal noise type corresponding to the original abnormal noise signal, effectively improving the classification efficiency and accuracy of abnormal noise type classification.

[0132] Reference Figure 6 This is a flowchart illustrating the steps of another embodiment of the vehicle abnormal noise classification method of this application.

[0133] When classifying abnormal noises in vehicles, you can first input a feature dataset containing signal features of abnormal noise signals into the vehicle abnormal noise classification system, and then divide the feature dataset into a training set and a test set. It is possible to establish a classification model for vehicle abnormal noises and initialize the model parameters of the classification model; A KPCA model can be established and its parameters initialized. First, determine that the kernel function to be used is the Gaussian radial basis kernel function and set an initial kernel bandwidth for it. The training set can then be input into the population optimization and update module; The population optimization and update module can initialize the relevant parameters of the Parrot algorithm, including but not limited to population size, maximum number of iterations, and search space range; Then the population optimization and update module can search for the optimal location, that is, the optimal individual in the population, through strategies such as foraging and communication. After each search, the global optimal position and global optimal value are updated; the global optimal position can represent the position of the iteration number updated by the best individual in the current population in the total number of iterations; the global optimal value can correspond to the feature value of the best individual in the current population. After completing one iteration update of the current population, it can be determined whether the current iteration count has reached the maximum number of iterations; If the current iteration count has not reached the maximum iteration count, the parrot's position can be updated based on the updated individual positions in the population. Then, the steps of searching for the optimal position by executing strategies such as foraging and communication can be returned to, and the next round of iteration can be performed. If the maximum number of iterations is reached in the current iteration, the optimal parameter combination of the KPCA model is output based on the best individual in the final updated population. Then the obtained optimal parameters, which are the target model parameters, can be used to reconstruct the KPCA model; After reconstructing the KPCA model, the feature data from the test set can be input into the reconstructed KPCA model to obtain dimensionality-reduced features, which are the target abnormal sound feature vectors. The dimensionality-reduced features obtained from the KPCA model can be input into a classification model for classification. The classification model can identify and output the final classification result based on the dimensionality reduction features.

[0134] Through the above implementation process, effective features related to vehicle abnormal noises can be extracted quickly and accurately from complex acoustic signals. At the same time, redundant information is removed, and the feature dimension is significantly reduced, thereby improving the discrimination accuracy of abnormal noise features and the classification accuracy and efficiency of abnormal noise classification.

[0135] It should be noted that the model parameter determination method and vehicle abnormal noise classification method provided in this application embodiment can be executed by a model parameter determination device and a vehicle abnormal noise classification device, or by a control module in the model parameter determination device and vehicle abnormal noise classification device for executing the loading model parameter determination method and vehicle abnormal noise classification method. This application embodiment uses the execution of the loading model parameter determination method and vehicle abnormal noise classification method by the model parameter determination device and the vehicle abnormal noise classification device as an example to illustrate the model parameter determination method and vehicle abnormal noise classification method provided in this application embodiment.

[0136] Reference Figure 7 This is a structural block diagram of an embodiment of a model parameter determination device according to this application. The device includes the following modules: The first acquisition module 701 is used to acquire the training dataset of abnormal noise features of the vehicle; Population generation module 702 is used to generate an initial population based on the abnormal noise feature training dataset; the population contains multiple population individuals; The model building module 703 is used to build a kernel principal component analysis model based on preset model parameters; the kernel principal component analysis model is used to extract target abnormal noise feature vectors from the abnormal noise features of the vehicle, so as to classify the abnormal noise of the vehicle based on the target abnormal noise feature vectors. The objective function determination module 704 is used to determine the objective function based on the kernel principal component analysis model. The population update module 705 is used to iteratively update the initial population according to the objective function to obtain an updated population. The optimal population individual determination module 706 is used to determine the optimal population individual from the population individuals of the updated population. The target model parameter determination module 707 is used to determine the target model parameters of the kernel principal component analysis model based on the optimal population individuals and the preset parameter range.

[0137] The population generation module 702 includes: The mapping feature data generation submodule is used to generate corresponding mapping feature data based on the abnormal noise feature training dataset. An initial population generation submodule is used to randomly select mapping feature data that meet preset conditions as population individuals according to a preset population size, so as to generate an initial population; the preset condition is that the population individuals are located within a preset search space range.

[0138] The population update module 705 includes: The current population determination submodule is used to determine the initial population as the current population and generate corresponding population information; the population information includes the current iteration number; The population individual update submodule is used to randomly select one of the preset update methods to update each population individual in the current population, thereby obtaining the updated population individual. The current population update submodule is used to update the current population based on the individuals in the update population; The iteration count determination submodule is used to, when the current iteration count is less than the preset maximum iteration count, return the method to be randomly selected from the preset update methods for each individual in the current population to update, and obtain the updated population individual; when the current iteration count is equal to the preset maximum iteration count, obtain the updated population.

[0139] The update method includes communication behavior update, and the population individual update submodule is further used for: From the current population, determine a first target population individual and a second target population individual; the second target population individual is a random population individual from the current population other than the first target population individual; The adaptive iteration factor is determined based on the current iteration number and the preset maximum iteration number. The first updated population individual corresponding to the first target population individual is determined based on the adaptive iteration factor and the second target population individual; The fitness score of the first updated population individual is determined according to the objective function; If the fitness score of the first updated population individual is greater than the fitness score of the first target population individual, the first target population individual is updated to the first updated population individual.

[0140] The update method includes foraging behavior update, and the population individual update submodule is further used for: The fitness score of each individual in the contemporary population is determined according to the objective function. The optimal individuals in the current population are determined based on the fitness scores. The target search range is determined based on the maximum number of iterations. The second updated population individual is determined based on each of the population individuals in the current population, the optimal population individual, and the target search range; The fitness score of the individuals in the second updated population is determined according to the objective function; If the fitness score of the second updated population individual is greater than the fitness score of the population individual, the population individual is updated to the second updated population individual.

[0141] The target model parameter determination module 707 includes: The candidate model parameter determination submodule is used to determine candidate model parameters from the preset parameter range according to a preset value step size; The optimal population individual score determination submodule is used to determine the fitness score of the optimal population individual based on the candidate model parameters and the objective function; The target model parameter determination submodule is used to determine the candidate model parameters with the highest fitness scores of the individuals in the optimal population as the target model parameters.

[0142] Reference Figure 8 This is a structural block diagram of an embodiment of a vehicle abnormal noise classification device according to this application. The device includes the following modules: The second acquisition module 801 is used to acquire the original abnormal noise signal of the vehicle; Feature extraction module 802 is used to extract features from the original abnormal noise signal to obtain abnormal noise feature data; Input module 803 is used to input the abnormal noise feature data into the kernel principal component analysis model to obtain the target abnormal noise feature vector; The classification module 804 is used to classify the target abnormal noise feature vector to obtain the abnormal noise classification result; The kernel principal component analysis model is constructed based on the target model parameters described in claim 1.

[0143] The model parameter determination device and vehicle abnormal noise classification device in the embodiments of this application can be devices, or components, integrated circuits, or chips in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. The embodiments of this application do not impose specific limitations.

[0144] The model parameter determination device and vehicle abnormal noise classification device in this application embodiment can be devices with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit the specific operating system used.

[0145] The model parameter determination device provided in this application embodiment can achieve... Figures 1 to 4 ,and Figure 6 In the method embodiment, the model parameter determination device implements each process, and the vehicle abnormal noise classification device can achieve... Figure 5 The various processes implemented by the vehicle abnormal noise classification device in the method embodiment will not be described again here to avoid repetition.

[0146] The model parameter determination device provided in this application can acquire a training dataset of vehicle abnormal noise features, and then generate an initial population based on the training dataset. This population can contain multiple individuals. A kernel principal component analysis (KPCA) model can be constructed based on preset model parameters, such as a KPCA model based on a Gaussian radial basis function kernel. The KPCA model can be used to extract target abnormal noise feature vectors from the vehicle's abnormal noise features, and then classify the vehicle's abnormal noises based on these target feature vectors. An objective function can be determined based on the KPCA model. Then, the objective function can guide the iterative evolution direction of the initial population, and the initial population can be iteratively updated until an updated population is reached. The optimal individual can then be determined from the updated population. Finally, the target model parameters of the KPCA model can be determined based on the optimal individual and a preset parameter range. Through the above implementation process, a kernel principal component analysis model capable of performing dimensionality reduction of sound signals is constructed. Based on the kernel principal component analysis model, the objective function for iterative updating of the initial population corresponding to the abnormal noise feature training data is obtained. By updating the initial population to obtain the optimal population individuals, the target model parameters of the kernel principal component analysis model most suitable for the current abnormal noise environment can be determined, thereby improving the parameter tuning and generalization ability of the kernel principal component analysis model and ensuring the accuracy of vehicle abnormal noise classification.

[0147] The vehicle abnormal noise classification device provided in this application can acquire the original abnormal noise signals generated during vehicle operation from vehicles undergoing fault testing or repair. Then, it extracts features from the original abnormal noise signals to obtain abnormal noise feature data. Inputting the abnormal noise feature data into a kernel principal component analysis model constructed based on target model parameters yields a target abnormal noise feature vector. After determining the target abnormal noise feature vector, it can be classified to obtain the abnormal noise classification result. Through the above implementation process, the kernel principal component analysis model constructed using target model parameters can perform rapid and effective dimensionality reduction processing on the abnormal noise feature data, obtaining the most relevant and representative target abnormal noise feature vectors. Then, classification and identification are performed based on the target abnormal noise feature vectors to determine the abnormal noise type corresponding to the original abnormal noise signal, effectively improving the classification efficiency and accuracy of abnormal noise type classification.

[0148] Optionally, embodiments of this application also provide an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the various processes of the above-described model parameter determination method and vehicle abnormal noise classification method embodiments, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0149] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0150] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described model parameter determination method and vehicle abnormal noise classification method, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0151] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0152] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above-described model parameter determination method and vehicle abnormal noise classification method embodiments, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0153] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0154] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0155] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0156] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A model parameter determination method characterized by, The method includes: Obtain a training dataset for the characteristics of abnormal noises from vehicles; An initial population is generated based on the training dataset of the abnormal noise characteristics; the population contains multiple individuals. A kernel principal component analysis model is constructed based on preset model parameters; the kernel principal component analysis model is used to extract target abnormal noise feature vectors from the abnormal noise features of the vehicle, so as to classify the abnormal noise of the vehicle based on the target abnormal noise feature vectors. The objective function is determined based on the kernel principal component analysis model. The initial population is iteratively updated according to the objective function to obtain the updated population; Determine the optimal population individual from the population individuals of the updated population; The target model parameters of the kernel principal component analysis model are determined based on the optimal population individuals and the preset parameter range.

2. The method of claim 1, wherein, The step of generating an initial population based on the training dataset of the abnormal noise features includes: Generate corresponding mapping feature data based on the abnormal noise feature training dataset; Based on a preset population size, mapping feature data that meets preset conditions are randomly selected as individuals in the population to generate an initial population; the preset condition is that the individuals in the population are located within a preset search space.

3. The method of claim 1, wherein, The step of iteratively updating the initial population according to the objective function to obtain the updated population includes: The initial population is determined as the current population, and corresponding population information is generated; the population information includes the current iteration number; For each individual in the current population, one method is randomly selected from the preset update methods to update the population, thus obtaining the updated population individual. The current population is updated based on the updated population individuals; If the current iteration count is less than the preset maximum iteration count, return to the step of randomly selecting one of the preset update methods for each individual in the current population to update the population, thereby obtaining the updated population individuals; When the current iteration number equals the preset maximum iteration number, the updated population is obtained.

4. The method of claim 3, wherein, The update method includes communication behavior update, wherein for each individual in the current population, one method is randomly selected from a preset update method for update to obtain an updated population individual, including: From the current population, determine a first target population individual and a second target population individual; the second target population individual is a random population individual from the current population other than the first target population individual; The adaptive iteration factor is determined based on the current iteration number and the preset maximum iteration number. The first updated population individual corresponding to the first target population individual is determined based on the adaptive iteration factor and the second target population individual; The fitness score of the first updated population individual is determined according to the objective function; If the fitness score of the first updated population individual is greater than the fitness score of the first target population individual, the first target population individual is updated to the first updated population individual.

5. The method according to claim 3, characterized in that, The update method includes foraging behavior update, wherein for each individual in the current population, one method is randomly selected from a preset update method for update to obtain an updated population individual, including: The fitness score of each individual in the contemporary population is determined according to the objective function. The optimal individuals in the current population are determined based on the fitness scores. The target search range is determined based on the maximum number of iterations. The second updated population individual is determined based on each of the population individuals in the current population, the optimal population individual, and the target search range; The fitness score of the individuals in the second updated population is determined according to the objective function; If the fitness score of the second updated population individual is greater than the fitness score of the population individual, the population individual is updated to the second updated population individual.

6. The method according to claim 1, characterized in that, The step of determining the target model parameters of the kernel principal component analysis model based on the optimal population individuals and the preset parameter range includes: Candidate model parameters are determined from the preset parameter range according to the preset value step size; The fitness score of the optimal population individual is determined based on the candidate model parameters and the objective function; The candidate model parameters with the highest fitness scores among the individuals in the optimal population are determined as the target model parameters.

7. A method for classifying abnormal noises in vehicles, characterized in that, The method includes: Obtain the original abnormal noise signal of the vehicle; Feature extraction is performed on the original abnormal noise signal to obtain abnormal noise feature data; The abnormal noise feature data is input into the kernel principal component analysis model to obtain the target abnormal noise feature vector; The abnormal noise feature vector of the target is classified to obtain the abnormal noise classification result; The kernel principal component analysis model is constructed based on the target model parameters described in claim 1.

8. A model parameter determination device, characterized in that, The device includes: The first acquisition module is used to acquire the training dataset of abnormal noise features of the vehicle; The population generation module is used to generate an initial population based on the abnormal noise feature training dataset; the population contains multiple population individuals; The model building module is used to build a kernel principal component analysis model based on preset model parameters; the kernel principal component analysis model is used to extract target abnormal noise feature vectors from the abnormal noise features of the vehicle, so as to classify the abnormal noise of the vehicle based on the target abnormal noise feature vectors. The objective function determination module is used to determine the objective function based on the kernel principal component analysis model. The population update module is used to iteratively update the initial population according to the objective function to obtain an updated population. The optimal population individual determination module is used to determine the optimal population individual from the population individuals of the updated population. The target model parameter determination module is used to determine the target model parameters of the kernel principal component analysis model based on the optimal population individuals and the preset parameter range.

9. A vehicle abnormal noise classification device, characterized in that, The device includes: The second acquisition module is used to acquire the original abnormal noise signal of the vehicle; The feature extraction module is used to extract features from the original abnormal noise signal to obtain abnormal noise feature data; The input module is used to input the abnormal noise feature data into the kernel principal component analysis model to obtain the target abnormal noise feature vector; The classification module is used to classify the target abnormal noise feature vector to obtain the abnormal noise classification result; The kernel principal component analysis model is constructed based on the target model parameters described in claim 1.

10. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the steps of the model parameter determination method as described in any one of claims 1-6 or the vehicle abnormal noise classification method as described in claim 7.

Images