Aircraft noise source positioning method

Abstract

The invention discloses an aircraft noise source positioning method comprising the following steps: sound field signals are collected, and the sound field signals comprise sound pressure signals and three-dimensional particle vibration velocity signals; preprocessing the sound pressure signal and the three-dimensional particle vibration velocity signal to obtain frequency spectrum data; inputting the frequency spectrum data into a trained noise recognition model, and judging a noise event; when the noise event is judged to be aircraft noise, noise source positioning is carried out according to corresponding spectrum data; according to the invention, noise event identification can be carried out firstly, so that the influence of the sound source arrival direction and other noise sources is eliminated, and errors are avoided; and the sound pressure and the three-dimensional particle vibration velocity are collected for preprocessing, effective sound intensity signals are obtained for positioning, and the positioning precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of noise analysis, and more particularly to a method for locating an aircraft noise source. Background Art

[0002] At present, with the continuous development of civil aviation transportation, the airport noise problem is becoming increasingly serious. Using modern testing technology to accurately identify the main noise sources and analyze their sound field characteristics is the key to solving the problem of moving sound source noise. At present, the detection of moving sound source noise is mainly concentrated on traditional measurement methods such as sound pressure method and sound intensity method. It is difficult to objectively reflect the actual movement trajectory of moving sound source noise. The monitoring method of traditional microphone array has low resolution for identifying medium and low frequency noise sources, and has high requirements for array form. The use and maintenance cost of long-term outdoor monitoring is high.

[0003] In order to better describe the sound field and obtain more sound field information, scientists have proposed the idea of ​​sound field energy and developed sound vector sensors. Compared with traditional microphone arrays that only obtain sound pressure, sound vector sensors can simultaneously collect sound pressure information and particle vibration velocity information in the sound field environment.

[0004] Therefore, how to combine the particle vibration velocity to solve the current problems of low accuracy of maneuvering sound source azimuth estimation, large amount of calculation, and unsuitability for real-time processing is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention provides a method for locating aircraft noise sources, which can first identify noise events and then eliminate the influence of the sound source arrival direction and other noise sources to avoid errors; it also collects sound pressure and three-dimensional particle velocity for preprocessing to obtain effective sound intensity signals for positioning, thereby improving positioning accuracy.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A method for locating an aircraft noise source comprises the following steps:

[0008] Collecting sound field signals, wherein the sound field signals include sound pressure signals and three-dimensional particle velocity signals;

[0009] Preprocessing the sound pressure signal and the three-dimensional particle velocity signal to obtain frequency spectrum data;

[0010] Inputting the spectrum data into a trained noise recognition model to determine noise events;

[0011] When a noise event is judged to be aircraft noise, the noise source is located based on the corresponding spectrum data.

[0012] Preferably, the pretreatment step comprises:

[0013] The sound pressure signal and the three-dimensional particle velocity signal are subjected to Fourier transformation and then octave segmentation to form multiple frequency bands, and the sound pressure spectrum and the three-dimensional velocity spectrum are obtained accordingly.

[0014] Preferably, the preprocessing step further includes: extracting characteristic vectors of the sound pressure spectrum and the three-dimensional velocity spectrum by maximum energy gradient method and performing pre-whitening processing.

[0015] Preferably, the sound pressure spectrum feature further includes an A-weighted sound pressure level, and the A-weighted sound pressure level is calculated based on the multiple frequency bands corresponding to the sound pressure signal.

[0016] Preferably, the preprocessing step further includes: normalizing the multiple frequency bands to remove interference noise.

[0017] Preferably, the training step of the noise recognition model includes:

[0018] Construct a training data set to obtain audio data of approaching aircraft noise, departing aircraft noise, cruising aircraft noise, and airport background noise;

[0019] Preprocessing the training data set and inputting it into the noise recognition model to generate a category matrix;

[0020] The model loss is calculated according to a preset loss function, and the parameters of the noise recognition model are optimized.

[0021] Preferably, the loss parameter is a cross entropy loss function, and the specific calculation formula is:

[0022]

[0023] Where M is the number of categories, y c represents a variable (0 or 1), if the category is the same as the sample category, it is 1, otherwise it is 0, p c Represents the predicted probability that the observed sample belongs to category c.

[0024] Preferably, the steps of locating the noise source according to the corresponding spectrum data include:

[0025] Get spectrum data X (t) , and obtain the spatial spectrum;

[0026] X (t) =A(s(t))+n(t)

[0027] Among them, X (t)are the vectors of each received signal, A is the direction vector matrix composed of the propagation paths from the sound source to each signal receiving part element, s(t) is the aircraft sound source signal vector, and n(t) is the environmental noise vector;

[0028] According to the spectrum data X (t) Calculate the covariance matrix, and perform iterative search in various directions through spatial spectrum maximum likelihood estimation to obtain the value of the objective function in each direction;

[0029] The objective function is:

[0030] Where u is the unit vector in the direction of the sound source, r pv is the covariance matrix of the sound pressure signal and the particle velocity signal, Q VV It is the autocovariance matrix of the particle velocity signal on the diagonal of the covariance matrix Q.

[0031] The signal subspace eigenvector, the noise subspace eigenvector, the covariance matrix and the spatial spectrum are input into a trained sound source localization network model to obtain a sound source direction angle.

[0032] Preferably, the noise source is located according to the corresponding spectrum data, and the step further includes: before performing the iterative search, using an average sound intensity device to estimate and calculate the initial angle range of the sound source, and when performing the iterative search, the search range is set within the search initial angle range.

[0033] Preferably, the step of locating the noise source according to the corresponding spectrum data further includes:

[0034] Performing eigendecomposition on the covariance matrix through subspace analysis to obtain a signal subspace vector and a noise subspace vector;

[0035] The spatial spectrum, the covariance matrix, the signal subspace vector and the noise subspace vector are input into a trained sound source localization network model to obtain a sound source direction angle.

[0036] One or more fully connected layers are used to infer the azimuth of the sound source based on the extracted features.

[0037] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for locating an aircraft noise source, which can first identify noise events, and then eliminate the influence of the sound source arrival direction and other noise sources to avoid errors; it also collects sound pressure and three-dimensional particle velocity for preprocessing to obtain an effective sound intensity signal for positioning, thereby improving positioning accuracy; the present invention performs four-channel data calculation on sound pressure and three-dimensional velocity, and in this way a single vector microphone can be used to realize data collection. Therefore, compared with the use of multiple vector microphones, the complexity of data processing can be reduced and the amount of calculation can be reduced; the present invention improves the response speed by reducing the amount of calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0039] Figure 1 The accompanying drawing is a schematic flow chart of a method for locating an aircraft noise source provided by the present invention.

[0040] Figure 2 The accompanying drawing is a schematic diagram of the principle of an average sound pressure device in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] The embodiment of the present invention discloses a method for locating an aircraft noise source, comprising the following steps:

[0043] S1: Collecting sound field signals, which include sound pressure signals and three-dimensional particle velocity signals. Specifically, the sound pressure information and particle velocity information of the noise source of the incoming and outgoing aircraft are collected through the vector microphone. Based on the suppression effect of the acoustic vector sensor on the isotropic noise source, the influence of the background noise on the target noise source can be reduced, and the aircraft positioning ambiguity effect caused by the environmental noise can be effectively overcome. In addition, since the particle velocity itself has directionality, the aircraft far-field positioning technology has stronger robustness and accuracy. Therefore, this embodiment can use a single vector microphone to collect far-field aircraft noise data.

[0044] In another embodiment, the timing of collecting the sound field signal is when the aircraft is approaching and departing. Among the airport noise impacts, the aircraft noise during the cruising phase cannot cause great acoustic pollution due to the attenuation effect in the air, while the aircraft noise during approaching and departing phases mostly affects the acoustic environment of surrounding residents. Therefore, the aircraft noise signal during approaching and departing phases is used as the analysis object, and the sound source sound pressure information and particle vibration velocity information are collected through the acoustic vector sensor.

[0045] In order to further implement the above technical solution, noise monitoring equipment is arranged at noise-sensitive points around the airport runway. The noise monitoring equipment includes acoustic vector sensors, vector conditioners, data acquisition cards, windshields, power supplies, etc. To ensure the accuracy of the experimental data, the experiment is carried out in a weather environment without strong winds and rain. The data acquisition time is 15s-20s and the acquisition frequency is 25600HZ. The acoustic vector sensor can simultaneously collect the sound pressure information at the same position and the particle velocity information in the three-dimensional direction. Through the vector conditioner and data acquisition card, the audio information is converted into a digital signal for subsequent data analysis and processing.

[0046] S2: Preprocess the sound pressure signal and the three-dimensional particle velocity signal to obtain spectrum data. Specifically, the collected sound pressure signal and the three-dimensional pointing velocity signal are subjected to Fourier transform (FFT) and segmented, and then normalized to remove interference noise to improve the signal-to-noise ratio of the characteristic information in the sound source noise signal. The signal frequency is divided into 34 frequency bands using a 1 / 3 octave filtering algorithm. For the i-th frequency band, first, according to the cutoff frequency fl, fu and the number of points of the sampling frequency of the frequency band, the spectrum data in the interval [fl, fu] is intercepted from the array obtained by FFT (the amplitude of other intervals is 0) and restored to a double-sided spectrum; then an IFFT transformation is performed to obtain the time domain signal corresponding to the frequency; for the sound pressure signal, the A-weighted sound pressure level in each frequency band can also be calculated. After octave filtering, the wide-band signal can be decomposed into narrower frequency bands, which is more conducive to improving the signal-to-noise ratio in the signal and making subsequent feature extraction more effective.

[0047] In order to further implement the above technical solution, the sound pressure information and particle velocity information in the recorded aircraft noise signal are spectrally analyzed, and the maximum energy gradient method is used to obtain the characteristic vector of the audio signal of the initial measured aircraft noise source and pre-whiten it, in preparation for the subsequent sound signal analysis.

[0048] S3: Input the spectrum data into the trained noise recognition model to determine the noise event.

[0049] Regarding the judgment of noise events, not all noise events are caused by aircraft during the aircraft positioning process. For example, the sounds of cars and construction sites near the monitoring point are often misjudged as aircraft noise events by the monitoring system due to their high sound pressure levels and long duration. In order to ensure the accuracy of aircraft noise positioning and the completeness of the report, the collected sound signals need to be classified before aircraft positioning.

[0050] In one embodiment, an automated machine learning method based on image classification is proposed to classify noise events using 1 second LAeq events at 1 / 3 octave frequencies.

[0051] Among them, the noise recognition model is a convolutional neural network model. Specifically, the input of the network is a processed one-third octave band image as an input signal, which is input to the input layer of the neural network. Then, the hierarchical features of the image are extracted by stacking several convolutional layers. Each convolutional layer uses a set of learnable filters (or convolution kernels), which perform convolution operations with the input data in the spatial dimension to generate feature maps. The convolution operation maintains the spatial structure of the input data and can capture local dependencies and image statistical characteristics. Immediately after each convolutional layer, a nonlinear activation function is introduced to enhance the expressive power of the model. The present invention uses a rectified linear unit (ReLU) as an activation function, which is defined as f(x) = max(0,x), which helps to alleviate the gradient vanishing problem and accelerate the convergence speed. Pooling layers usually follow the convolutional layers to sample feature maps and reduce their spatial size, thereby reducing the number of parameters and computational complexity. In this model, we adopt a max-pooling strategy to select the maximum value in each local area as a representative while maintaining the significance of the feature. After the convolution and pooling layers, the feature map is flattened into a one-dimensional vector so that it can be used as input to the fully connected layer. The flattening layer converts the two-dimensional feature map to a one-dimensional feature vector, providing the fully connected layer with an input in the appropriate format. The flattened one-dimensional feature vector is fed into one or more fully connected layers (also called dense layers), which are responsible for integrating the extracted features and performing high-level reasoning. In a fully connected layer, each neuron is connected to all neurons in the previous layer. Finally, the output of the fully connected layer is passed to the output layer, which contains the same number of neurons as the number of target categories. In this case, the output layer has two neurons, corresponding to the two categories of aircraft noise and non-aircraft noise.

[0052] The training steps of the noise recognition model include:

[0053] S31: Construct a training data set to obtain audio data of approaching aircraft noise, departing aircraft noise, cruising aircraft noise and airport background noise.

[0054] S32: Preprocess the training data set and input it into the noise recognition model to generate a category matrix.

[0055] S33: Calculate the model loss according to the preset loss function, and optimize the parameters of the noise recognition model. The loss parameter is the cross entropy loss function, and the specific calculation formula is:

[0056]

[0057] Where M is the number of categories, y c represents a variable (0 or 1), if the category is the same as the sample category, it is 1, otherwise it is 0, p c Represents the predicted probability that the observed sample belongs to category c.

[0058] In this embodiment, in order to accurately identify the category of aircraft noise, we use a deep convolutional neural network model and conduct a rigorous training process. Model training aims to minimize the difference between the predicted output and the true label while ensuring that the model has good generalization ability. This embodiment uses an aircraft noise dataset consisting of 200 samples, each of which is a preprocessed one-third octave band map. The dataset is randomly divided into a training set (80%), a validation set (10%), and a test set (10%) to support model training, parameter adjustment, and evaluation. As a criterion for measuring model performance, we use the cross-entropy loss function, which provides an effective method for evaluating the difference between the predicted probability distribution and the true distribution for multi-category classification tasks. In order to optimize the network weights, this embodiment uses the Adam optimizer, which combines the advantages of the momentum method and RMSProp to adaptively adjust the learning rate of each parameter. The Adam optimizer is widely used in the deep learning community for its efficient computing performance and low memory requirements. Model training involves the setting of multiple hyperparameters, including learning rate, batch size, and number of training cycles (epochs). Based on the performance feedback on the validation set, we set the initial value of the learning rate to 1e-3 and adjusted it through the learning rate decay strategy. The batch size was set to 32 to balance the model update speed and memory consumption. The entire network training process was designed to run 50 training cycles, and an early stopping strategy was adopted to avoid overfitting, that is, if the loss of the validation set did not improve for 5 consecutive cycles, the training was terminated. During the training process, we also introduced L2 regularization to reduce the risk of overfitting, and used the dropout technique to randomly drop neurons in the fully connected layer. The model performance was evaluated on the validation set and the test set using evaluation indicators such as accuracy and receiver operating characteristic curve (ROC-AUC). Finally, the real-time noise data was converted into a one-third octave band map and fed into the trained CNN model. The classification result output by the model indicates whether an aircraft noise event is detected at this time.

[0059] S4: When the noise event is judged to be aircraft noise, the noise source is located according to the corresponding spectrum data. The signal collected by the acoustic vector sensor is converted into an omnidirectional sound pressure signal p(x) and three co-point orthogonal pointing velocity signals v x , v y , v z In the aircraft noise signal acquisition, it is usually a non-uniform noise environment, assuming that each channel signal obeys a random process signal with zero mean Gaussian random distribution.

[0060] Positioning method 1: First determine the initial angle range of the direction angle of the sound source through average sound intensity estimation, and then perform accurate spatial spectrum maximum likelihood estimation iterative search.

[0061] The specific steps include:

[0062] Get the spectrum data X(t ) , and obtain the spatial spectrum;

[0063] X(t)=A(s(t))+n(t)

[0064] Where X(t) is the vector of each received signal, A is the direction vector matrix composed of the propagation paths from the sound source to each signal receiving part element, s(t) is the aircraft sound source signal vector, and n(t) is the environmental noise vector;

[0065] According to the spectrum data X(t ) Calculate the covariance matrix, and perform iterative search in various directions through spatial spectrum maximum likelihood estimation to obtain the value of the objective function in each direction;

[0066] The objective function is:

[0067] Among them, u is the unit vector in the direction of the sound source, rpv is the covariance matrix of the sound pressure signal and the particle velocity signal, Q VV is the autocovariance matrix of the particle velocity signal on the diagonal of the covariance matrix Q, α p is the ambient noise power, α v The power of a signal relative to a sound source.

[0068] Among them, regarding the calculation of the covariance matrix: the covariance matrix Q of the signal is the expected outer product of the received signal vector, which can be estimated as:

[0069]

[0070] Where T is the number of samples in the observation time window, X H (t) is the conjugate transpose of X(t); R is the noise covariance matrix, where r pp , r pv , r vp , r vv are the elements of the signal covariance, representing the covariance matrices of sound pressure-sound pressure, sound pressure-particle velocity, particle velocity-sound pressure, and particle velocity-particle velocity, respectively, indicating the correlation between the two signals. A is the propagation vector matrix corresponding to the signal propagation path, and R is the covariance matrix of the noise power.

[0071] Furthermore, before performing the iterative search, an average sound intensity device is used to estimate and calculate the initial angle range of the sound source, and when performing the iterative search, the search range is set within the search initial angle range, and the steps include:

[0072] First, calculate the average sound energy flow density per unit area perpendicular to the sound intensity flow, that is, the sound intensity vector I.

[0073]

[0074] T is the signal duration, p(t) is the sound pressure time domain signal, and v(t) is the particle velocity time domain signal

[0075] In the calculation process, we usually use Instead of the above formula, the average sound intensity is calculated by combining p(t) with the particle velocity information in the x, y, and z directions to obtain the sound intensity vectors in the x, y, and z directions and output them as ( is the mean value of the time domain signal of the sound intensity in the x direction, is the mean value of the time domain signal of the sound intensity in the y direction, is the mean value of the time domain signal of the sound intensity in the z direction), and finally the direction angle θ is:

[0076]

[0077] The pitch angle δ is:

[0078]

[0079] Positioning method 2: Based on positioning method 1, a neural network is used for final positioning. The covariance matrix is ​​eigen-decomposed by subspace analysis to obtain a signal subspace vector and a noise subspace vector; the spatial spectrum, the covariance matrix, the signal subspace vector and the noise subspace vector are input into a trained sound source localization network model to obtain the sound source direction angle.

[0080] Specifically, the sound source localization network model includes: an input layer for obtaining preprocessed data; one or more convolutional layers for extracting features and generating feature maps; the convolutional layer is composed of a set of learnable convolutional kernels; a pooling layer for reducing feature dimensions and further extracting features; and one or more fully connected layers for inferring the azimuth of the sound source based on the extracted features.

[0081] For the training of the sound source localization network model, first select a loss function suitable for the regression task, the mean square error (MSE). Adjust the network weights to minimize the loss function. Input the training data into the network in batches, and update the weights through back propagation at each iteration. Adjust hyperparameters such as learning rate, batch size, number of network layers, and number of filters to find the optimal configuration. The final fully connected layer output will be a continuous value representing the predicted direction angle of the sound source. Find the direction angle that maximizes the objective function through the convolutional neural network. Finally, calculate the specific direction angle estimate according to the direction vector:

[0082]

[0083] Determine the direction angles when the objective function is optimal. These angles are the estimated DOA directions of the sound source.

[0084] In this embodiment, the signal subspace analysis method includes principal component analysis (PCA) and singular value decomposition (SVD).

[0085] In principal component analysis (PCA), the signal matrix is ​​subjected to eigendecomposition of the covariance matrix to obtain a set of orthogonal principal components (i.e., subspaces), each of which has a different variance. These principal components are sorted by variance and can be used to describe the energy distribution and correlation of the signal.

[0086] In singular value decomposition (SVD), the signal matrix is ​​decomposed into the product of three matrices: U, Σ, and V^T. Among them, U and V^T are orthogonal matrices, Σ is a diagonal matrix, and the diagonal elements are called singular values. The column vectors of U and V^T constitute the left and right subspaces of the signal matrix, and the diagonal elements of Σ represent the energy of the signal in each subspace.

[0087] In another embodiment, the direction range is further determined by using the maximum likelihood estimation of the spatial spectrum. The specific process is as follows:

[0088] The signal collected by the acoustic vector sensor is converted into an omnidirectional sound pressure signal p(x) and three co-point orthogonal pointing velocity signals v x , v y , v z In the aircraft noise signal collection, it is usually a non-uniform noise environment. It is assumed that each channel signal obeys a zero-mean Gaussian random distribution random process signal. Its signal power is α s , the ambient noise power is α p , the signal power α associated with the sound source v , the received signal covariance matrix is:

[0089]

[0090] Where R is the noise covariance matrix, where rpp , r pv , r vp , r vv are the elements of the signal covariance, representing the covariance matrices of sound pressure-sound pressure, sound pressure-particle velocity, particle velocity-sound pressure, and particle velocity-particle velocity, respectively, indicating the correlation between the two signals. A is the propagation vector matrix corresponding to the signal propagation path, and R is the covariance matrix of the noise power.

[0091] The maximum likelihood estimation method is used to estimate the DOA of the sound source signal. In order to obtain the maximum likelihood estimation under DOA, it is necessary to find the most appropriate direction vector u = [u x ,u y ,u z ]T makes the joint probability function as large as possible and obtains the objective function. The objective function is continuously modified through iteration:

[0092]

[0093] Where u is the unit vector in the direction of the sound source, r pv is the covariance matrix of the sound pressure signal and the particle velocity signal, Q VV It is the autocovariance matrix of the particle velocity signal on the diagonal of the covariance matrix Q.

[0094] A more accurate estimate of the direction vector is obtained by searching for possible direction angles (e.g., scanning the space or grid searching) and calculating the objective function value for each angle.

[0095] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0096] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for locating an aircraft noise source, characterized in that: The following steps are involved: Collecting sound field signals, wherein the sound field signals include sound pressure signals and three-dimensional particle velocity signals; Preprocessing the sound pressure signal and the three-dimensional particle velocity signal to obtain frequency spectrum data; Inputting the spectrum data into a trained noise recognition model to determine noise events; When a noise event is judged to be aircraft noise, the noise source is located based on the corresponding spectrum data.

2. The method for locating an aircraft noise source according to claim 1, characterized in that: The pre-processing step comprises: The sound pressure signal and the three-dimensional particle velocity signal are subjected to Fourier transformation and then octave segmentation to form multiple frequency bands, and a sound pressure spectrum and a three-dimensional velocity spectrum are obtained accordingly.

3. The method for locating an aircraft noise source according to claim 2, characterized in that: The preprocessing step also includes: extracting characteristic vectors of the sound pressure spectrum and the three-dimensional velocity spectrum by maximum energy gradient method and performing pre-whitening processing.

4. The method for locating an aircraft noise source according to claim 2, characterized in that: The sound pressure spectrum feature further includes an A-weighted sound pressure level, and the A-weighted sound pressure level is calculated according to the multiple frequency bands corresponding to the sound pressure signal.

5. The method for locating an aircraft noise source according to claim 2, characterized in that: The preprocessing step further includes: normalizing the multiple frequency bands to remove interference noise.

6. The method for locating an aircraft noise source according to claim 1, characterized in that: The training steps of the noise recognition model include: Construct a training data set to obtain audio data of approaching aircraft noise, departing aircraft noise, cruising aircraft noise, and airport background noise; Preprocessing the training data set and inputting it into the noise recognition model to generate a category matrix; The model loss is calculated according to a preset loss function, and the parameters of the noise recognition model are optimized.

7. The method for locating an aircraft noise source according to claim 6, characterized in that: The loss parameter is the cross entropy loss function, and the specific calculation formula is: Where M is the number of categories, y c represents a variable (0 or 1), if the category is the same as the sample category, it is 1, otherwise it is 0, p c Represents the predicted probability that the observed sample belongs to category c.

8. The method for locating an aircraft noise source according to claim 1, characterized in that: The noise source is located according to the corresponding spectrum data, and the steps include: Get spectrum data X (t) , and obtain the spatial spectrum; X (t) =A(s(t))+n(t) Among them, X (t) are the vectors of each received signal, A is the direction vector matrix composed of the propagation paths from the sound source to each signal receiving part element, s(t) is the aircraft sound source signal vector, and n(t) is the environmental noise vector; According to the spectrum data X (t) Calculate the covariance matrix, and perform iterative search in various directions through spatial spectrum maximum likelihood estimation to obtain the value of the objective function in each direction; The objective function is: Where u is the unit vector in the direction of the sound source, r pv is the covariance matrix of the sound pressure signal and the particle velocity signal, Q VV It is the autocovariance matrix of the particle velocity signal on the diagonal of the covariance matrix Q. The signal subspace eigenvector, the noise subspace eigenvector, the covariance matrix and the spatial spectrum are input into a trained sound source localization network model to obtain a sound source direction angle.

9. The method for locating an aircraft noise source according to claim 8, characterized in that: The noise source is located according to the corresponding spectrum data, and the steps also include: before performing the iterative search, using an average sound intensity device to estimate and calculate the initial angle range of the sound source, and when performing the iterative search, the search range is set within the search initial angle range.

10. The method for locating an aircraft noise source according to claim 8, characterized in that: The noise source is located according to the corresponding spectrum data, and the steps further include: Performing eigendecomposition on the covariance matrix through subspace analysis to obtain a signal subspace vector and a noise subspace vector; The spatial spectrum, the covariance matrix, the signal subspace vector and the noise subspace vector are input into a trained sound source localization network model to obtain a sound source direction angle.

Images