Vehicle noise prediction method, electronic equipment and medium
By acquiring multi-source data of rail vehicles, performing correlation analysis, and building a deep learning network model, combining attention mechanism and Bayesian optimization method, the problem of difficult to decouple and predict complex electromechanical coupled noise in the existing technology is solved, and high-precision noise prediction is achieved.
Patent Information
- Application Number
- CN202510107504.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-23
- Publication Date
- 2025-06-13
AI Technical Summary
The prior art is difficult to effectively decouple and predict complex electromechanical coupled noise in rail vehicles, and cannot efficiently fuse and utilize these data when processing multimodal heterogeneous data, resulting in low accuracy and efficiency of noise diagnosis.
By acquiring multi-source data, including traction motor vibration data, traction motor noise data, gearbox vibration data, etc., perform correlation analysis, select data with a correlation coefficient greater than the threshold as input, build a deep learning network model, combine attention mechanism and Bayesian optimization method, optimize model hyperparameters to achieve noise prediction.
The noise dynamic characteristics of the traction motor-gear box are effectively extracted, which improves the accuracy of noise prediction, and can capture important data under a small amount of sample data for analysis and learning, and accurately capture the coupled noise characteristics.
Smart Images
Figure CN120145291A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of vehicle noise prediction, and particularly relates to a vehicle noise prediction method, an electronic device, and a medium. Background Art
[0002] With the rapid development of modern means of transportation, especially rail vehicles, their operating status directly affects the safety and riding comfort of trains. Therefore, how to effectively predict the vehicle noise is an urgent problem to be solved in the current development of rail vehicles.
[0003] The existing solutions mainly achieve noise monitoring and fault diagnosis of the traction drive system through hardware devices and software algorithms. The hardware devices mainly include sensors, data collectors, etc., which are used to collect the operating data of the traction drive system. The software algorithms mainly include signal processing, feature extraction, pattern recognition, etc., which are used to analyze and process the collected data to achieve noise control and fault diagnosis.
[0004] However, there are still some problems in the existing technology in practical applications. First, the existing hardware devices and software algorithms often can only perform predictive analysis on a single type of noise, and for complex electromechanical coupling noise, the existing technology often cannot effectively decouple and predict it. Second, when dealing with multi-modal heterogeneous data under the electromechanical coupling of rail vehicles, the existing technology often cannot effectively fuse and utilize these data, resulting in low accuracy and efficiency of noise diagnosis. Finally, in terms of noise prediction, the existing technology often cannot achieve high-precision prediction, which brings certain risks to the reliability and safety of vehicles.
[0005] The traction motor - gearbox is a non-linear dynamic system with a complex sound field environment, and its noise varies complexly with different speeds. Therefore, the noise sample collection of the rail vehicle traction motor - gearbox is limited, showing the characteristics of few and uneven quantities. Traditional machine learning methods require a large number of samples for learning and training to achieve the best prediction effect. However, due to the limitations of few samples and strong interference in the traction motor - gearbox noise prediction, the traditional machine learning methods have low accuracy in vehicle noise prediction. Summary of the Invention
[0006] The purpose of the present invention is to provide a vehicle noise prediction method, an electronic device, and a medium to improve the accuracy of vehicle noise prediction in view of the deficiencies of the existing technology.
[0007] To achieve the above purpose, the technical solution adopted by the present invention is:
[0008] A vehicle noise prediction method includes the following steps:
[0009] S1: Obtain multi-source data of the vehicle, where the multi-source data includes traction motor vibration data, traction motor noise data, traction motor current data, gearbox vibration data, gearbox torque data, and gearbox noise data;
[0010] S2: Conduct correlation analysis on the traction motor noise data and traction motor vibration data, traction motor noise data and traction motor current data, gearbox noise data and gearbox vibration data, and gearbox noise data and gearbox torque data respectively, and select the traction motor vibration data, traction motor current data, gearbox vibration data, and gearbox torque data with a correlation coefficient greater than the threshold as the data set;
[0011] S3: Use the data set as the input of the deep learning network model and train the deep learning network model;
[0012] The deep learning network model includes multiple parallel neural networks and an output layer. The neural network includes a hidden layer and an attention mechanism layer connected in sequence, and the attention mechanism layer is connected to the output layer;
[0013] Input the data set into the hidden layer to obtain the first feature;
[0014] Input the first feature into the attention mechanism layer to obtain the second feature;
[0015] The second feature is processed by the output layer to obtain the coupled noise;
[0016] Optimize the hyperparameters of the deep learning network model through the Bayesian optimization method; use the loss function to iteratively optimize and train the deep learning network model to obtain the noise prediction model.
[0017] The present invention combines the correlation feature analysis of multi-source and multi-modal data of the traction drive system and the attention mechanism, and effectively extracts the noise dynamic characteristics of the traction motor - gearbox; through the attention mechanism, the model can help capture important data for analysis and learning under a small amount of sample data. The attention mechanism calculates the weights of key feature data, accurately captures the characteristics of the coupled noise, and improves the accuracy of noise prediction.
[0018] Furthermore, input the traction motor vibration data, traction motor current data, gearbox vibration data, and gearbox torque data into the noise prediction model to obtain the vehicle coupled noise.
[0019] Furthermore, the expression of the first feature a i is as follows:
[0020]
[0021] where q(·) is the activation function, and W ijis the weight of the j-th node in the i-th hidden layer, and b i is the bias of the i-th hidden layer, is the input data of the i-th hidden layer, L is the number of hidden layers, and N is the number of nodes in the hidden layer.
[0022] Furthermore, the second feature s i has the following expression:
[0023] s i = tanh(W i ·a i + b)
[0024] where W i is the weight of the i-th attention mechanism layer, b is the bias of the attention mechanism layer, and tanh(·) is the activation function.
[0025] Furthermore, the expression of the coupling noise is as follows:
[0026]
[0027] where is the sound pressure value of the coupling noise, w is the weight of the output layer, b' is the bias of the output layer, ReLu(·) is the activation function, and n is the number of input data.
[0028] Preferably, the threshold is 0.8.
[0029] Based on the same inventive concept, the present invention also provides an electronic device, including:
[0030] One or more processors;
[0031] A memory storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the steps of the vehicle noise prediction method.
[0032] Based on the same inventive concept, the present invention also provides a computer-readable storage medium storing a computer program, which when executed by a processor, implements the steps of the vehicle noise prediction method.
[0033] Compared with the prior art, the beneficial effects of the present invention:
[0034] The present invention combines the correlation feature analysis of multi-source and multi-modal data of the traction drive system and the attention mechanism, effectively extracts the noise dynamic characteristics of the traction motor - gearbox; through the attention mechanism, the model can help capture important data for analysis and learning with a small amount of sample data. The attention mechanism calculates the weights of key feature data, accurately captures the characteristics of the coupling noise, and improves the accuracy of noise prediction. Brief Description of the Drawings
[0035] Figure 1 It is a schematic flowchart of the vehicle noise prediction method of the present invention;
[0036] Figure 2 It is a schematic diagram of the deep learning network model of the present invention. Detailed Description of the Invention
[0037] The present invention will be described in detail below in conjunction with embodiments. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.
[0038] Embodiment
[0039] As Figure 1 , the vehicle noise prediction method of this embodiment includes the following processes:
[0040] 1. Obtain the traction motor vibration data, traction motor noise data, traction motor current data, gearbox vibration data, gearbox torque data, and gearbox noise data through the LMS vibration and noise data collector and the sensors on the vehicle to obtain the initial data set.
[0041] 2. Use the mscohere function of MATLAB to perform correlation analysis on the traction motor noise data and traction motor vibration data, traction motor noise data and traction motor current data, gearbox noise data and gearbox vibration data, and gearbox noise data and gearbox torque data, and select the traction motor vibration data, traction motor current data, gearbox vibration data, and gearbox torque data with a correlation coefficient greater than 0.8 as the data set; the correlation analysis is such as Pearson correlation analysis, and the correlation coefficient is such as Pearson correlation coefficient.
[0042] The formula for the Pearson correlation coefficient is as follows:
[0043]
[0044] Among them, r is the Pearson correlation coefficient, X i is the traction motor noise data or gearbox noise data, Y i is the traction motor vibration data or traction motor current data or gearbox vibration data or gearbox torque data, and are the means of the corresponding data respectively.
[0045] By performing correlation feature analysis on the vibration, current, gearbox torque and system coupling noise of the input traction drive system, when the data correlation coefficient is high, it indicates that the correlation between the current, gearbox torque, vibration and other signals and the noise is high, and the current, gearbox torque, vibration and other signals can be used as the input for subsequent noise prediction.
[0046] 3. As Figure 2 , construct a deep learning network model. The deep learning network model includes multiple parallel neural networks and an output layer. The neural network includes a hidden layer and an attention mechanism layer connected in sequence, and the attention mechanism layer is connected to the output layer.
[0047] 1) Input time (T), vibration frequency (F), the coupled vibration energy of the traction motor and the gearbox (V 1 ), the traction motor current data (V 2 ), and the gearbox torque data (V 3 ) as parameters into the deep learning network model. The initial data of the deep learning network model is denoted as matrix
[0048]
[0049] where λ1 to λ5 are the data sets of T, F, V 1 , V 2 , V 3 respectively, and the superscripts 1 - m of T, F, V 1 , V 2 , V 3 are the input data groups. The extracted data is divided into a training set and a test set. The training set is used to repeatedly train the weight parameters of the model, and the test set is used to evaluate the final performance of the model.
[0050] 2) Input the initial data into the hidden layer, and the forward propagation process can be expressed as;
[0051]
[0052] where a i is the output of the i-th hidden layer, q(·) is the activation function, W ij is the weight of the j-th node in the i-th hidden layer, b i is the bias of the i-th hidden layer, is the input of the i-th hidden layer, L is the number of hidden layers, and N is the number of hidden layer nodes.
[0053] 3) Use the attention mechanism to calculate the feature relationship between vibration and coupled noise in the input data. The process of calculating the feature s i by the attention mechanism can be expressed as:
[0054] s i = tanh(W i · a i + b) (4)
[0055] where W i$W_i$ is the weight of the $i$-th layer of the attention mechanism layer, $b$ is the bias of the attention mechanism layer, and $\tanh(\cdot)$ is the activation function.
[0056] Convert $s$ i into exponential form, and calculate the ratio of the current $s$ i to the sum of all attention score exponents to obtain the attention score $\beta$ i . The result $\beta$ of the attention scoring function i can be expressed as:
[0057]
[0058] where $n$ is the number of input data. Subsequently, the deep learning network model minimizes the loss function through backpropagation, reduces the model prediction error, and continuously adjusts and optimizes the model hyperparameters to achieve convergence.
[0059] The prediction results of the traction motor - gearbox coupling noise are as follows:
[0060]
[0061] where $p$ is the sound pressure value of the coupling noise, $W$ is the weight of the output layer, $b'$ is the bias of the output layer, and $\text{ReLu}(\cdot)$ is the activation function.
[0062] 4) Find the optimal hyperparameter set through the Bayesian optimization algorithm and extract the mapping relationship between vibration data and coupling noise data; The Bayesian feedforward neural network, as the basic architecture of the noise prediction model, can utilize its powerful nonlinear mapping ability to capture the dynamic characteristics of complex systems. Integrate the Bayesian optimization algorithm for hyperparameter optimization and maximize the prediction accuracy by automatically adjusting the model parameters.
[0063] Let $x$ be the set of hyperparameters in the hidden layer, $x = [L, N, \eta]$. $\eta$ represents the learning rate for updating the parameters in the hidden layer during backpropagation. The objective function value $y$ for updating the hyperparameter set $x$ can be defined as:
[0064] $y = f(L, N, \eta)$
[0065] $x$ * $= B[f(x)] = B[f(L, N, \eta)]\ (7)$
[0066] $B[\cdot]$ is the Bayesian optimization process. The prior probability distribution $p(y)$ and the conditional probability distribution $p(x|y)$ are established. $P(x|y)$ is the probability density function of the hyperparameters given the objective function value, and $P(y)$ is the distribution of the objective function value. In this case, $\sum p(y) = 1$, and the expression for $p(x|y)$ is:
[0067]
[0068] where y * represents the threshold value, and l(x) and g(x) represent the density estimation values. Subsequently, the Expected Improvement (EI) is used as the acquisition function to identify the next sample point with the highest acquisition expectation value. The calculation equation of the acquisition function can be expressed as:
[0069]
[0070] where the posterior probability p(y|x) can be expressed as:
[0071]
[0072] where γ is the constructor error adjustment function, γ = p(y < y*). p(x) is the marginal likelihood function, p(x) = γl(x) + (1 - γ)g(x). Equation (9) can be further expressed as:
[0073]
[0074] Equation (11) is an iterative process for optimizing g(x) / l(x) and maximizing g(x) / l(x). When is closer to the maximum value, the performance of the hyperparameter set is better. ∝ indicates that the expression on the left side of the equation is proportional to the expression on the right side.
[0075] When reaches the maximum value, this hyperparameter set is regarded as the optimal hyperparameter set x * .
[0076]
[0077] where L * is the optimal number of hidden layers, N * is the optimal number of nodes in the hidden layer, and η * is the learning rate of the optimal deep learning network model.
[0078] 5) Establish a loss function using the output value obtained from forward propagation and the actual sample label, defined as LOSS. Update the parameters using the loss function, and the formula is as follows:
[0079]
[0080] where represents the weight value of the j-th node in the i-th updated hidden layer.
[0081] Iteratively optimize and train the deep learning network model using the loss function to obtain the noise prediction model.
[0082] 6) Among the evaluation metrics of the model, considering that the Mean Absolute Error (MAE) and the Mean Squared Error (MSE) measure the gap between the predicted value and the true value, the Root Mean Squared Error (RMSE) reflects the error magnitude and generalization ability of the model, and the Mean Absolute Percentage Error (MAPE) is a further standardization of MAE. Therefore, MAE, MSE, RMSE, and MAPE are used as the evaluation metrics for the model accuracy and effectiveness, and their calculation formulas are as follows:
[0083]
[0084] Among them, Y represents the true value of the traction motor - gearbox noise sample.
[0085] This embodiment proposes to identify the preferred vibration data, traction motor current data, and gearbox torque data for predicting the coupled noise of the traction motor and gearbox based on the correlation analysis of multi - source data; and adopt the small - sample noise prediction method of the traction motor - gearbox based on Bayesian optimization - attention - neural network to train the preferred vibration - coupling data of the traction motor and gearbox, traction motor current data, and gearbox torque data. The Bayesian optimization - attention - neural network method combines the correlation feature analysis of multi - source and multi - modal data of the traction drive system, and combines the advantages of the neural network and the attention mechanism to effectively extract the noise dynamic characteristics of the traction motor - gearbox. The attention mechanism helps the model to capture important knowledge and learn quickly with a small number of samples. According to the small - sample noise prediction model of the traction motor - gearbox based on Bayesian optimization - attention - neural network, an effective prediction of the noise of the rail vehicle traction motor - gearbox is realized.
[0086] The method of this embodiment not only reduces the computational cost and model complexity of the model, but also improves the prediction accuracy of the model; by real - time monitoring the coupled vibration data of the traction motor and gearbox, the traction motor current harmonic data, and the gearbox torque data under vehicle operation, it can effectively and real - time predict the noise of the rail vehicle traction drive system, providing a new idea for the noise identification and abnormal noise prediction of rail vehicles.
[0087] Another embodiment of the present invention provides an electronic device, including:
[0088] One or more processors;
[0089] A memory, on which one or more programs are stored. When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the vehicle noise prediction method.
[0090] In some implementations, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk memory.
[0091] In other implementations, the processor may be various types of general-purpose processors such as a central processing unit (CPU) or a digital signal processor (DSP), which is not limited herein.
[0092] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the vehicle noise prediction method are implemented.
[0093] The content clarified in the above embodiments should be understood that these embodiments are only used to more clearly illustrate the present invention, rather than limiting the scope of the present invention. After reading the present invention, various equivalent forms of modification of the present invention by those skilled in the art all fall within the scope defined by the appended claims of this application.
Claims
1. A vehicle noise prediction method, characterized in that: The following steps are involved: S1: Acquire multi-source data of the vehicle, wherein the multi-source data includes traction motor vibration data, traction motor noise data, traction motor current data, gearbox vibration data, gearbox torque data, and gearbox noise data; S2: performing correlation analysis on the traction motor noise data and the traction motor vibration data, the traction motor noise data and the traction motor current data, the gearbox noise data and the gearbox vibration data, and the gearbox noise data and the gearbox torque data, respectively, and selecting the traction motor vibration data, the traction motor current data, the gearbox vibration data, and the gearbox torque data with correlation coefficients greater than a threshold as the data set; S3: using the data set as input of a deep learning network model to train the deep learning network model; The deep learning network model includes a multi-layer parallel neural network and an output layer, the neural network includes a hidden layer and an attention mechanism layer connected in sequence, and the attention mechanism layer is connected to the output layer; Inputting the data set into the hidden layer to obtain a first feature; Inputting the first feature into the attention mechanism layer to obtain a second feature; The second feature is processed by the output layer to obtain coupling noise; The hyperparameters of the deep learning network model are optimized by using a Bayesian optimization method; the deep learning network model is trained by iterative optimization using a loss function to obtain a noise prediction model.
2. The vehicle noise prediction method according to claim 1, characterized in that: The traction motor vibration data, traction motor current data, gearbox vibration data, and gearbox torque data are input into the noise prediction model to obtain the vehicle coupling noise.
3. The vehicle noise prediction method according to claim 1, characterized in that: The first feature a i The expression is as follows: Among them, q(·) is the activation function, W ij is the weight of the jth node in the i-th hidden layer, b i is the bias of the i-th hidden layer, is the input data of the i-th hidden layer, L is the number of hidden layers, and N is the number of hidden layer nodes.
4. The vehicle noise prediction method according to claim 3, characterized in that: Second feature i The expression is as follows: si=tanh(W i ·a i +b) Among them, W i is the weight of the i-th attention mechanism layer, b is the bias of the attention mechanism layer, and tanh(·) is the activation function.
5. The vehicle noise prediction method according to claim 4, characterized in that: The expression of coupled noise is as follows: in, is the coupling noise sound pressure value, w is the weight of the output layer, b' is the bias of the output layer, ReLu(·) is the activation function, and n is the number of input data.
6. An electronic device, characterized in that: include: one or more processors; A memory having one or more programs stored thereon, which, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the steps of the method according to any one of claims 1 to 5.
7. A computer-readable storage medium, characterized in that: The computer program is stored therein, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.