A train traction motor-gearbox coupling noise prediction method and system based on small sample data of vibration

Abstract

This invention discloses a method and system for predicting coupled noise of train traction motors and gearboxes based on small sample vibration data. The steps include: analyzing the high correlation between coupled noise and vibration signals using a combined squared coherence-regression coefficient analysis; establishing a mapping relationship between vibration signals and coupled noise using a feedforward neural network; finding the optimal configuration of the number of hidden layers, nodes, and learning rate using a tree-based Bayesian optimization algorithm to extract the mapping relationship between vibration data and coupled noise data; calculating the weights of feature data using an attention mechanism, constructing a Bayesian-attention-feedforward neural network model, and iteratively strengthening and updating it to output the prediction result. This invention, by constructing a Bayesian-attention-feedforward neural network coupled noise prediction model, accurately predicts coupled noise using a small number of vibration signals, solving the technical problems of poor stability and low accuracy in predicting coupled noise of train traction motors and gearboxes with small sample data.

Description

Technical Field

[0001] This invention relates to the field of noise prediction and design technology for traction drive systems of rail vehicles, and more specifically, to a method and system for predicting the coupled noise of a train traction motor-gearbox based on small sample vibration data. Background Technology

[0002] The coupling noise between the traction motor and gearbox is a major source of noise in rail vehicles. The strong coupling between the two exacerbates the noise, making it extremely difficult to control. Accurately predicting the coupling noise between the traction motor and gearbox and considering its correlation characteristics has always been a challenge in train noise design. A certain correlation exists between vibration signals and coupling noise, and this correlation serves as the basis for guiding the entire train noise design. Therefore, a comprehensive study of the coupling noise between the traction motor and gearbox is needed, along with the development of intelligent prediction methods and systems for coupling noise based on small sample vibration data.

[0003] Existing solutions primarily rely on hardware devices and software algorithms to predict noise in traction drive systems. Hardware devices mainly include sensors and data acquisition units to collect operational data from the traction drive system. Software algorithms, including signal processing, feature extraction, and pattern recognition, analyze and process the collected data to control noise and diagnose faults. While this technology offers simple and rapid processing, several issues remain in practical applications. First, the traction motor and gearbox operate in complex geometric environments with multi-source interference, resulting in a limited and incomplete range of usable noise samples. Second, existing hardware and software algorithms often only predict single types of noise, failing to effectively decouple and predict complex electromechanical coupling noise. Finally, current technologies often lack high accuracy and efficiency in noise prediction, posing risks to vehicle reliability and safety.

[0004] CN202110154558.5 discloses a gearbox fault diagnosis method based on multi-source data fusion. The steps include: first, acquiring vibration signals, noise signals, temperature signals at the gearbox bearings, displacement signals of the gear shaft, and oil level data of the gearbox lubricating oil; then, preprocessing the acquired multi-source data separately; and finally, using multi-sensor data fusion technology to diagnose gearbox faults. This patent, by acquiring multi-source data signals such as vibration, temperature, noise, and displacement of the gearbox and using multi-sensor data fusion technology for fault diagnosis, fully leverages the redundant and complementary functions of each sensor and data source, enabling a more complete and clear expression of gearbox fault information and improving the ability to locate gearbox faults and diagnose complex faults. However, while this method integrates multi-source signals for fault diagnosis, it also has shortcomings. First, multiple acquisitions of multi-source signals require precise time synchronization of each sensor, and the acquisition process often requires acquiring a large number of data samples to support subsequent diagnostic work. Second, this method only uses multi-source data from the gearbox for fault diagnosis and does not consider the coupling effects of other electromechanical devices. Finally, this method uses a traditional backpropagation (BP) neural network for feature extraction. However, the traditional backpropagation neural network used in this patent has a simple structure, but its hyperparameters are difficult to adjust, increasing diagnostic complexity and computational cost. Therefore, how to effectively, cost-effectively, efficiently, and accurately predict traction motor-gearbox coupling noise with small sample data is an urgent problem to be solved in the current development of rail vehicles. Summary of the Invention

[0005] The main technical problem this invention aims to solve is addressing the shortcomings of existing technologies, such as the difficulty in obtaining reliable data on the traction motor-gearbox, the difficulty in decoupling electromechanical coupling noise, and the difficulty in improving the accuracy of noise prediction models. This invention provides a method for predicting train traction motor-gearbox coupling noise based on small sample vibration data. By utilizing the strong correlation between coupling noise and vibration signals, and the excellent decoupling capability of the Bayesian attention-feedforward neural network model, this method uses easily obtainable vibration signals to predict traction motor-gearbox coupling noise, achieving accurate prediction of complex electromechanical coupling noise under different spatial locations. This provides an important foundation and technology for the forward design of vehicle noise.

[0006] Another technical problem solved by the present invention is to provide a train traction motor-gearbox coupling noise prediction system based on small sample data of vibration.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] A method for predicting train traction motor-gearbox coupling noise based on small sample vibration data, comprising the following steps:

[0009] S1. Based on the joint analysis of squared coherence-regression coefficients, the high correlation between coupled noise and vibration signal is analyzed, and the complex data is transformed into a two-dimensional matrix in the frequency domain. f ;

[0010] S2. Learning two-dimensional matrices using a feedforward neural network within a multi-task learning framework. f Based on the characteristics, establish the mapping relationship between vibration signals in the horizontal, vertical, and axial directions and coupled noise;

[0011] S3. A tree-based Bayesian optimization algorithm is used to find the optimal configuration of the number of hidden layers, nodes, and learning rate in a feedforward neural network, and to extract the mapping relationship between vibration data and coupled noise data.

[0012] S31. Selecting the number of hidden layers in a feedforward neural network L Number of hidden layer nodes N Learning rate or To optimize the hyperparameters of a feedforward neural network, the set of hyperparameters for the feedforward neural network is defined as follows: x Define the update hyperparameter set x objective function value y for: Define the configuration space as H={[ L,N,n ] 1 , [ L,N,n ] 2 , ..., [ L,N,n ] n};

[0013] S32. Calculate the probability density function of the hyperparameters given the objective function value based on the current hyperparameter sample points. P ( x | y Construct a Bayesian optimization model based on a tree structure;

[0014] S33. Using the desired improvement as the acquisition function IE y* ( x Identify the next sample point with the highest expected value.

[0015] S34. When the acquisition function IE y* ( x When the hyperparameter set reaches its maximum value, it is considered the optimal hyperparameter set. x*， Represented as:

[0016]

[0017] In the formula L* To determine the optimal number of hidden layers, N* The optimal number of hidden layer nodes. or* The optimal learning rate for the feedforward neural network;

[0018] S35. Optimal Hyperparameter Set x* The feedback is fed back into the feedforward neural network for further training and optimization;

[0019] S4. The attention mechanism calculates the weights of key feature data, constructs a Bayesian-attention-feedforward neural network model, and accurately captures the characteristics of coupling noise.

[0020] S41. In a feedforward neural network, the attention mechanism is used to calculate the feature relationship between vibration signals and coupled noise in the input data. A Bayesian-attention-feedforward neural network model is constructed to accurately capture the characteristics of coupled noise, and the attention mechanism calculates the features. s i The process can be represented as:

[0021]

[0022] Where tan h For activation function, Yes To pay attention to the weight of the mechanism, b For deviation, This is the output of the i-th hidden layer;

[0023] S42. Will s i Convert to exponential form and calculate the current s i The attention score is obtained by dividing the sum of all attention score indices by the ratio of the attention score to the sum of all attention score indices. β i The result of the attention scoring function β i It can be represented as:

[0024]

[0025] Where n is the number of input feature data;

[0026] S5. By iteratively strengthening and updating the parameters and weight coefficients of the Bayesian attention-feedforward neural network model, the prediction result of the traction motor-gearbox coupling noise is output, as follows:

[0027]

[0028] In the formula Ŷ The coupled noise sound pressure level predicted by the Bayesian optimization-attention mechanism-feedforward neural network model. ReLu For activation function, w and b 'These represent the weights and biases of the output layer, respectively;

[0029] S6. Based on the coupling noise prediction results, perform signal anomaly detection and predict multiple related attributes of the traction motor-gearbox, including fault type, fault location, and severity.

[0030] Furthermore, the squared coherence-regression coefficient is expressed as:

[0031]

[0032] in P Q_N ( f () is a vibration signal Q and coupled noise signal N cross power spectral density, P Q_Q ( f ) and P N_N ( f () are vibration signals Q and coupled noise signal N The self-power spectral density, Q i For the i-th vibration amplitude, N i Let i be the sound pressure level of the i-th coupled noise. This represents the average sound pressure level of the coupled noise. t 1+ t 2 = 1.

[0033] Furthermore, two-dimensional matrix f Includes time T, frequency F, vibration signal V1 in the horizontal direction, vibration signal V2 in the vertical direction, and vibration signal V3 in the axial direction, in a two-dimensional matrix. f Represented as:

[0034]

[0035] in l 1 ~ l 5 represent the datasets T, F, V1, V2, and V3, respectively, with superscripts 1 to 1. n For the input data set.

[0036] Furthermore, the forward propagation process of a feedforward neural network can be represented as follows:

[0037]

[0038] In the formula, a i Indicates the hidden layer number 1 i The output of the layer, f iIndicates the first hidden layer i Layer input, W ij and b i These represent the th hidden layer, respectively. i Layer weight matrix and bias, q This represents the activation function of the hidden layer. L This represents the number of hidden layers in the feedforward neural network. N This represents the number of hidden layer nodes in the feedforward neural network.

[0039] Furthermore, the probability density function of the hyperparameters p ( x | y ) is represented as:

[0040]

[0041] In the formula, y* Indicates the threshold. l ( x )and g ( x ) respectively represent Less than or greater than or equal to The density estimate.

[0042] Furthermore, the acquisition function IE y* ( x ) is represented as:

[0043]

[0044] p ( y | x Let ) be the posterior probability, expressed as:

[0045]

[0046] p ( x ) is the marginal likelihood function. p ( y ) is the distribution of the objective function values.

[0047] Furthermore, obtain the optimal hyperparameter set. x* back ， The forward propagation process of the feedforward neural network is updated as follows:

[0048]

[0049] Furthermore, the loss function for the forward propagation of the feedforward neural network is: LOSS The formula for updating the parameters of the loss function is expressed as:

[0050]

[0051] In the formula Ŵ ij This represents the weight of the j-th node in the i-th hidden layer of the updated feedforward neural network.

[0052] Furthermore, signal anomaly detection is performed on the predicted results of the traction motor-gearbox coupling noise. When the predicted coupling noise is an abnormal signal, multi-task learning is performed to predict the fault type, fault location, and severity of the traction motor-gearbox. Targeted maintenance is carried out according to different types, locations, and severity of faults to achieve intelligent early warning.

[0053] A train traction motor-gearbox coupling noise prediction system based on small vibration sample data includes:

[0054] The raw data acquisition and analysis module collects raw data and performs data processing and analysis.

[0055] The feedforward neural network module includes one input layer, L hidden layers and one output layer. It learns the features of the input data and establishes the mapping relationship between the vibration signal and the coupled noise.

[0056] The Bayesian optimization module, based on the tree-structured Parzen Estimator, optimizes the hyperparameters of the feedforward neural network model, finds the optimal hyperparameter set, and extracts the mapping relationship between vibration data and coupled noise data.

[0057] The attention mechanism module calculates the weights of key feature data and accurately captures the characteristics of coupling noise.

[0058] The multi-dimensional visualization module visualizes the coupled noise sound pressure values ​​predicted from vibration data at different sensor locations, including two-dimensional visualization, three-dimensional visualization, and modal visualization, clearly showing the fitting relationship between the two.

[0059] Compared with existing technologies, the beneficial effects are:

[0060] This invention analyzes vibration and coupled noise data of a train traction motor-gearbox to extract key features. A feedforward neural network model is established using the correlation between vibration and coupled noise data. This model accurately predicts the coupled noise generated by the traction motor and gearbox at different spatial locations using readily available vibration signals. In the feedforward neural network, Bayesian optimization is used to pre-train the hyperparameters to approximate the optimal parameters. An attention mechanism is introduced during data training to further focus on and compute the characteristics of complex noise, thereby achieving accurate noise prediction.

[0061] The method described in this invention improves the accuracy of coupled noise prediction on small vibration sample datasets. By using time-frequency domain vibration signals to predict coupled noise of key train components, it provides an effective technical approach for early control and reduction of train noise. Furthermore, this invention provides an important tool for evaluating noise performance during the selection of train components, overall train design, and manufacturing processes. Attached Figure Description

[0062] Figure 1 The flowchart of a train traction motor-gearbox coupling noise prediction method based on small vibration sample data is provided in Example 1.

[0063] Figure 2 This is a schematic diagram of a train traction motor-gearbox coupled noise prediction system based on small sample vibration data, provided in Example 2.

[0064] Figure 3 This is a block diagram of a train traction motor-gearbox coupled noise prediction system based on small vibration sample data, as provided in Example 2.

[0065] Figure 4 This is a schematic diagram of the experimental data provided in Example 3.

[0066] Figure 5 This is a schematic diagram of the train traction motor-gearbox coupling noise prediction results based on vibration small sample data provided in Example 3.

[0067] Figure 6 This is a schematic diagram of the train traction motor-gearbox coupling noise prediction error based on vibration small sample data provided in Example 3. Detailed Implementation

[0068] The following examples further explain and clarify the invention, but the specific examples do not limit the invention in any way.

[0069] Example 1

[0070] This embodiment provides a method for predicting train traction motor-gearbox coupling noise based on small sample vibration data, such as... Figure 1The steps include:

[0071] A method for predicting train traction motor-gearbox coupling noise based on small sample vibration data, comprising the following steps:

[0072] S1. Based on the joint analysis of squared coherence-regression coefficients, the high correlation between coupled noise and vibration signal is analyzed, and the complex data is transformed into a two-dimensional matrix in the frequency domain. f ;

[0073] S11. Using the squared coherence-regression coefficient, coherence analysis is performed on the processed vibration and coupling noise sample data to determine the vibration signal as the main correlation parameter for subsequent coupling noise prediction. The squared coherence-regression coefficient is expressed as:

[0074]

[0075] in P Q_N ( f () is a vibration signal Q and coupled noise signal N cross power spectral density, P Q_Q ( f ) and P N_N ( f () are vibration signals Q and coupled noise signal N The self-power spectral density, Q i For the i-th vibration amplitude, N i Let i be the sound pressure level of the i-th coupled noise. This represents the average sound pressure level of the coupled noise. t 1+ t 2 = 1.

[0076] S12. After the joint analysis of squared coherence and regression coefficients, preliminary data screening is performed on the original vibration signals, and... M Q_N Highly coherent vibration data with a correlation coefficient greater than 0.8 are used to transform the coupled noise signal and vibration signal into a two-dimensional matrix φ in the frequency domain. This matrix includes time T, frequency F, vibration signal V1 in the horizontal direction, vibration signal V2 in the vertical direction, and vibration signal V3 in the axial direction. f Represented as:

[0077]

[0078] in l 1 ~ l 5 represent the datasets T, F, V1, V2, and V3, respectively, with superscripts 1 to 1.n For the input data set.

[0079] S2. Learning two-dimensional matrices using a feedforward neural network within a multi-task learning framework. f Based on the characteristics, a mapping relationship between horizontal, vertical, and axial vibration signals and coupled noise is established. The forward propagation process of the feedforward neural network is represented as follows:

[0080]

[0081] In the formula, a i Indicates the hidden layer number 1 i The output of the layer, f i Indicates the first hidden layer i Layer input, W ij and b i These represent the th hidden layer, respectively. i Layer weight matrix and bias, q This represents the activation function of the hidden layer. L This represents the number of hidden layers in the feedforward neural network. N This represents the number of hidden layer nodes in the feedforward neural network.

[0082] S3. A tree-based Bayesian optimization algorithm is used to find the optimal configuration of the number of hidden layers, nodes, and learning rate in a feedforward neural network, and to extract the mapping relationship between vibration data and coupled noise data.

[0083] S31. Selecting the number of hidden layers in a feedforward neural network L Number of hidden layer nodes N Learning rate or To optimize the hyperparameters of a feedforward neural network, the set of hyperparameters for the feedforward neural network is defined as follows: x Define the update hyperparameter set x objective function value y for: Define the configuration space as H={[ L,N,n ] 1 , [ L,N,n ] 2 , ..., [ L,N,n ] n};

[0084] S32. Calculate the probability density function of the hyperparameters given the objective function value based on the current hyperparameter sample points. P ( x | y A tree-based Bayesian optimization model is constructed, wherein the probability density function of the hyperparameters is... p (x | y ) is represented as:

[0085]

[0086] In the formula, y* Indicates the threshold. l ( x )and g ( x ) respectively represent Less than or greater than or equal to The density estimate.

[0087] S33. Using the desired improvement as the acquisition function IE y* ( x Identify the next sample point with the highest expected acquisition value; acquisition function IE y* ( x ) is represented as:

[0088]

[0089] p ( y | x Let ) be the posterior probability, expressed as:

[0090]

[0091] p ( x ) is the marginal likelihood function. p ( y () is the distribution of the objective function values. Definition c This is a constructor error adjustment function. c = p ( y < y* ), p ( x )= γl ( x )+(1- c ) g ( x Acquisition function IE y* ( x Further expressed as:

[0092] .

[0093] This formula is an optimization. g ( x ) / l ( x ) and maximizationg ( x ) / l ( x The iterative process of ) when IE y* ( x The closer the hyperparameter set is to its maximum value, the better its performance. ∝ indicates that the expression on the left side of the equation is proportional to the expression on the right side.

[0094] S34. When the acquisition function IE y* ( x When the hyperparameter set reaches its maximum value, it is considered the optimal hyperparameter set. x*， Represented as:

[0095]

[0096] In the formula L* To determine the optimal number of hidden layers, N* The optimal number of hidden layer nodes. or* This represents the optimal learning rate for the feedforward neural network.

[0097] Obtain the optimal hyperparameter set x* Then, the forward propagation process of the feedforward neural network is updated as follows:

[0098] .

[0099] A loss function is established using the output value obtained from forward propagation and the actual sample labels. LOSS The formula for updating the parameters of the loss function is expressed as:

[0100]

[0101] In the formula Ŵ ij This represents the weight of the j-th node in the i-th hidden layer of the updated feedforward neural network.

[0102] S4. The attention mechanism calculates the weights of key feature data, constructs a Bayesian-attention-feedforward neural network model, and accurately captures the characteristics of coupling noise.

[0103] S41. In a feedforward neural network, the attention mechanism is used to calculate the feature relationship between vibration signals and coupled noise in the input data. A Bayesian-attention-feedforward neural network model is constructed to accurately capture the characteristics of coupled noise, and the attention mechanism calculates the features. s i The process can be represented as:

[0104]

[0105] Where tanh For activation function, Yes To pay attention to the weight of the mechanism, b For deviation, This is the output of the i-th hidden layer;

[0106] S42. Will s i Convert to exponential form and calculate the current s i The attention score is obtained by dividing the sum of all attention score indices by the ratio of the attention score to the sum of all attention score indices. β i The result of the attention scoring function β i It can be represented as:

[0107]

[0108] Where n is the number of input feature data;

[0109] S5. By iteratively strengthening and updating the parameters and weight coefficients of the Bayesian attention-feedforward neural network model, the prediction result of the traction motor-gearbox coupling noise is output, as follows:

[0110]

[0111] In the formula Ŷ The coupled noise sound pressure level predicted by the Bayesian optimization-attention mechanism-feedforward neural network model. ReLu For activation function, w and b 'These represent the weights and biases of the output layer, respectively;

[0112] S6. Based on the coupling noise prediction results, perform signal anomaly detection and predict multiple related attributes of the traction motor-gearbox, including fault type, fault location, and severity. Specifically, perform signal anomaly detection based on the coupling noise prediction results of the traction motor-gearbox. When the predicted coupling noise is an abnormal signal, perform multi-task learning to predict the fault type, fault location, and severity of the traction motor-gearbox. Perform targeted maintenance based on different types, locations, and severity of faults to achieve intelligent early warning.

[0113] Example 2

[0114] This embodiment provides a train traction motor-gearbox coupling noise prediction system based on small sample vibration data, such as... Figure 2 , 3 ,include:

[0115] The raw data acquisition and analysis module collects raw data and performs data processing and analysis.

[0116] The feedforward neural network module consists of one input layer, L hidden layers, and one output layer. It learns the features of the input data and establishes a mapping relationship between the vibration signal and the coupled noise.

[0117] The Bayesian optimization module uses a tree-structured Parzen Estimator to optimize the hyperparameters of the feedforward neural network model, finds the optimal set of hyperparameters, and extracts the mapping relationship between vibration data and coupled noise data.

[0118] The attention mechanism module calculates the weights of key feature data and accurately captures the characteristics of coupling noise.

[0119] The multi-dimensional visualization module visualizes the coupled noise sound pressure values ​​predicted from vibration data at different sensor locations, including two-dimensional visualization, three-dimensional visualization, and modal visualization, clearly showing the fitting relationship between the two.

[0120] This embodiment can also be used for traction motor-gearbox coupling noise prediction under other variables.

[0121] Example 3

[0122] This embodiment provides a specific example of a train traction motor-gearbox coupling noise prediction method based on small sample vibration data, such as... Figure 4-6 ,include:

[0123] The experiment was conducted on a traction motor-gearbox of a certain model. The experimental setup included eight accelerometers and nine noise acquisition microphones. Accelerometers were placed at various vibration points of the traction motor-gearbox, and the locations of the accelerometers corresponding to different vibrations were labeled V1-V8. Noise acquisition microphones were placed at various locations near the traction motor-gearbox, designated N1-N9. The accelerometers and noise acquisition microphones collected vibration signals and coupled noise signals in the time domain, respectively. These signals were then processed and analyzed using Fourier transform on the central control platform and LMS Test. Lab 14A software to obtain the frequency domain data of the vibration signals and coupled noise. The frequency domain variation curves of the coupled noise and vibration signals are shown below. Figure 4 As shown in (a) and (b).

[0124] By downsampling, horizontal (X), vertical (Y), and axial (Z) vibration data from eight acceleration sensors of the traction motor and gearbox were extracted at equal intervals as prediction inputs, resulting in a small dataset of 1600 sets of data on the correlation between frequency, vibration data, and coupled noise data. A Bayesian attention-feedforward neural network noise prediction model was trained using 95% of the data, with the remaining 5% used for prediction. The overall steps of this embodiment are the same as in Embodiment 1, and after multiple iterations, the coupled noise prediction result is output.

[0125] Figure 5 Predicted sound pressure levels for coupled noise from the V1-V8 accelerometer sensors of the traction motor-gearbox at noise acquisition point N1. Figure 5 As can be seen from the evaluation of 80 sets of data, the Bayesian optimization-attention-feedforward neural network has high accuracy in predicting the coupling noise between the traction motor and the gearbox.

[0126] In model evaluation metrics, considering that Mean Absolute Error (MAE) measures the difference between predicted and true values, Root Mean Square Error (RMSE) reflects the magnitude of the model's error and its generalization ability, and Mean Absolute Percentage Error (MAPE) is a further standardization of MAE, MAE, RMSE, and MAPE are used as evaluation metrics for model accuracy and performance, and their calculation formulas are as follows:

[0127]

[0128] Where Y represents the true value of the traction motor-gearbox noise sample. This represents the predicted value output by a noise prediction model based on a Bayesian optimization-attention-feedforward neural network. For example... Figure 6 As shown, the coupled noise prediction error value of the traction motor-gearbox V1-V8 acceleration sensors at the N1 noise acquisition point is visualized using a line graph.

[0129] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting train traction motor-gearbox coupling noise based on small sample vibration data, characterized in that the steps are as follows: include: S1. Based on the joint analysis of squared coherence-regression coefficients, the high correlation between coupled noise and vibration signal is analyzed, and the complex small sample data is transformed into a two-dimensional matrix in the frequency domain. φ ; S2. Learning two-dimensional matrices using a feedforward neural network within a multi-task learning framework. φ Based on the characteristics, establish the mapping relationship between vibration signals in the horizontal, vertical, and axial directions and coupled noise; S3. A tree-based Bayesian optimization algorithm is used to find the optimal configuration of the number of hidden layers, nodes, and learning rate in a feedforward neural network, and to extract the mapping relationship between vibration data and coupled noise data. S31. Selecting the number of hidden layers in a feedforward neural network L Number of hidden layer nodes N Learning rate η To optimize the hyperparameters of a feedforward neural network, the set of hyperparameters for the feedforward neural network is defined as follows: x Define the update hyperparameter set x objective function value y for: Define the configuration space as H={[ L,N,η ] 1 , [ L,N,η ] 2 , ..., [ L,N,η ] n }; S32. Calculate the probability density function of the hyperparameters given the objective function value based on the current hyperparameter sample points. P ( x | y Construct a Bayesian optimization model based on a tree structure; S33. Using the desired improvement as the acquisition function EI y* ( x Identify the next sample point with the highest expected value. S34. When the acquisition function EI y* ( x When the hyperparameter set reaches its maximum value, it is considered the optimal hyperparameter set. x*， Represented as: In the formula L* To determine the optimal number of hidden layers, N* The optimal number of hidden layer nodes. η* The optimal learning rate for the feedforward neural network; S35. Optimal Hyperparameter Set x* The feedback is fed back into the feedforward neural network for further training and optimization; S4. The attention mechanism calculates the weights of key feature data, constructs a Bayesian-attention-feedforward neural network model, and accurately captures the characteristics of coupling noise. S41. In a feedforward neural network, the attention mechanism is used to calculate the feature relationship between vibration signals and coupled noise in the input data. A Bayesian-attention-feedforward neural network model is constructed to accurately capture the characteristics of coupled noise, and the attention mechanism calculates the features. s i The process can be represented as: Where tan h For activation function, Ŵi To pay attention to the weight of the mechanism, b For deviation, This is the output of the i-th hidden layer; S42. Will s i Convert to exponential form and calculate the current s i The attention score is obtained by dividing the sum of all attention score indices by the ratio of the attention score to the sum of all attention score indices. β i The result of the attention scoring function β i It can be represented as: Where n is the number of input feature data; S5. By iteratively strengthening and updating the parameters and weight coefficients of the Bayesian attention-feedforward neural network model, the prediction result of the traction motor-gearbox coupling noise is output, as follows: In the formula Ŷ The coupled noise sound pressure level predicted by the Bayesian optimization-attention mechanism-feedforward neural network model. ReLu For activation function, w and b 'These represent the weights and biases of the output layer, respectively; S6. Based on the coupling noise prediction results, perform signal anomaly detection and predict multiple related attributes of the traction motor-gearbox, including fault type, fault location, and severity.

2. The train traction motor-gearbox coupling noise prediction method based on vibration small sample data according to claim 1, characterized in that, The squared coherence-regression coefficient is expressed as: in P Q_N ( f () is a vibration signal Q and coupled noise signal N cross power spectral density, P Q_Q ( f ) and P N_N ( f () are vibration signals Q and coupled noise signal N The self-power spectral density, Q i For the i-th vibration amplitude, N i Let i be the sound pressure level of the i-th coupled noise. This represents the average sound pressure level of the coupled noise. τ 1+ τ 2 = 1.

3. The train traction motor-gearbox coupling noise prediction method based on vibration small sample data according to claim 1, characterized in that, Two-dimensional matrix φ Includes time T, frequency F, vibration signal V1 in the horizontal direction, vibration signal V2 in the vertical direction, and vibration signal V3 in the axial direction, in a two-dimensional matrix. φ Represented as: in λ 1 ~ λ 5 represent the datasets T, F, V1, V2, and V3, respectively, with superscripts 1 to 1. n For the input data set.

4. The train traction motor-gearbox coupling noise prediction method based on vibration small sample data according to claim 1, characterized in that, The forward propagation process of a feedforward neural network can be represented as follows: In the formula, a i Indicates the hidden layer number 1 i The output of the layer, φ i Indicates the first hidden layer i Layer input, W ij and b i These represent the th hidden layer, respectively. i Layer weight matrix and bias, q This represents the activation function of the hidden layer. L This represents the number of hidden layers in the feedforward neural network. N This represents the number of hidden layer nodes in the feedforward neural network.

5. The train traction motor-gearbox coupling noise prediction method based on vibration small sample data according to claim 1, characterized in that, probability density function of hyperparameters P ( x | y ) is represented as: In the formula, y* Indicates the threshold. l ( x )and g ( x ) respectively represent Less than or greater than or equal to The density estimate.

6. The train traction motor-gearbox coupling noise prediction method based on vibration small sample data according to claim 5, characterized in that, Acquisition function EI y* ( x ) is represented as: p ( y | x Let ) be the posterior probability, expressed as: p ( x ) is the marginal likelihood function. p ( y ) is the distribution of the objective function values.

7. The train traction motor-gearbox coupling noise prediction method based on vibration small sample data according to claim 4, characterized in that, Obtain the optimal hyperparameter set x* back ， The forward propagation process of the feedforward neural network is updated as follows: 。 8. The train traction motor-gearbox coupling noise prediction method based on vibration small sample data according to claim 7, characterized in that, The loss function for the forward propagation of a feedforward neural network is LOSS The formula for updating the parameters of the loss function is expressed as: In the formula Ŵ ij This represents the weight of the j-th node in the i-th hidden layer of the updated feedforward neural network.

9. The method for predicting train traction motor-gearbox coupling noise based on vibration small sample data according to claim 1, characterized in that, The predicted coupling noise of the traction motor-gearbox is used to detect signal anomalies. When the predicted coupling noise is an abnormal signal, multi-task learning is performed to predict the fault type, fault location and severity of the traction motor-gearbox. Targeted maintenance is carried out according to different types, locations and severity of faults to achieve intelligent early warning.

10. The system for predicting train traction motor-gearbox coupling noise based on vibration small sample data according to any one of claims 1 to 9, characterized in that, include: The raw data acquisition and analysis module collects raw data and performs data processing and analysis. The feedforward neural network module includes one input layer, L hidden layers and one output layer. It learns the features of the input data and establishes the mapping relationship between the vibration signal and the coupled noise. The Bayesian optimization module, based on the tree-structured Parzen Estimator, optimizes the hyperparameters of the feedforward neural network model, finds the optimal hyperparameter set, and extracts the mapping relationship between vibration data and coupled noise data. The attention mechanism module calculates the weights of key feature data and accurately captures the characteristics of coupling noise. The multi-dimensional visualization module performs one-dimensional to multi-dimensional visualization processing on the coupling noise sound pressure value predicted from vibration data at different sensor locations, clearly showing the fitting relationship between the two.

Images