Electromechanical equipment noise reduction method based on machine learning

Abstract

The invention discloses an electromechanical equipment noise reduction method based on machine learning, and relates to the field of electromechanical equipment noise reduction, and the method comprises the steps: collecting sound pressure and vibration signals during the operation of electromechanical equipment, and collecting the operation state data of the electromechanical equipment; sparse reconstruction is carried out on the noise features; carrying out sound source localization on the reconstructed noise features by using a beam forming algorithm to obtain a noise distribution diagram; constructing a noise reduction model based on deep learning to obtain noise signals after noise reduction; extracting the frequency of the noise signal in a preset frequency range through a peak search algorithm; analyzing the acquired running state data by using a time-varying linear predictive coding algorithm TVLPC to obtain a noise frequency offset caused by the abnormal running state of the electromechanical equipment; generating an optimal solution by using a particle swarm algorithm; and according to the optimal solution of noise control, noise dominant frequency components are extracted for separation, and a plurality of phase-reversal noise reduction control signals are obtained. Aiming at low noise reduction precision of electromechanical equipment in the prior art, the noise reduction control precision is improved.

Description

Technical Field

[0001] The present application relates to the field of electromechanical equipment noise reduction, and in particular to an electromechanical equipment noise reduction method based on machine learning. Background Art

[0002] In recent years, with the increasing requirements of modern buildings for indoor environmental comfort, the noise problem generated by building mechanical and electrical equipment such as HVAC, elevators, water pumps, etc. during operation has become increasingly prominent. The noise of these equipment will not only reduce the quality of the indoor sound environment and affect people's work, study and life, but may also cause the decline of building sound insulation and sound absorption performance, and even lead to noise nuisance and other problems. Therefore, how to effectively control the noise of building mechanical and electrical equipment and improve the acoustic performance of buildings has become one of the key issues to be solved in the field of modern building acoustic environment design and control.

[0003] Traditional noise control of building electromechanical equipment mainly adopts passive noise reduction technology, such as installing soundproof covers, soundproof barriers, and silencers around the equipment. However, these passive noise reduction measures are mainly aimed at medium and high frequency noise, and the effect of reducing low frequency noise is limited. In addition, the noise of building electromechanical equipment usually has time-varying and non-stationary characteristics, and its spectral components and amplitudes will change dynamically with changes in equipment working conditions and external interference. Traditional passive noise reduction technology is difficult to adapt to this dynamic noise characteristic, and the noise reduction effect is limited.

[0004] In order to overcome the above shortcomings, active noise control (ANC) technology has received more and more attention. ANC technology uses the destructive principle of sound waves to reduce the original noise by applying secondary sound waves with the same frequency, equal amplitude but opposite phase as the original noise in the noise field. Compared with passive noise reduction technology, ANC technology has the advantages of good low-frequency noise reduction effect and adaptability to dynamic noise. However, to achieve efficient and accurate ANC noise reduction, it is necessary to extract and analyze the characteristics of the equipment noise in real time and accurately, and dynamically adjust the control parameters according to the noise characteristics, which puts high demands on noise signal processing and control algorithms.

[0005] In addition, there are many types of mechanical and electrical equipment in buildings, and the noise spectrum is complex and changeable, which poses a challenge to the adaptability and robustness of the noise control system. At the same time, the fluctuation of equipment operating conditions will also introduce noise frequency deviation, further increasing the difficulty of ANC system design. Traditional ANC methods are difficult to meet the requirements of high-precision, real-time, and adaptive control in complex noise environments. Summary of the invention

[0006] In response to the problem of low noise reduction accuracy of electromechanical equipment in the prior art, the present application provides a noise reduction method for electromechanical equipment based on machine learning. By generating a spatial distribution map of the noise source and constructing a deep learning model for noise reduction processing, it focuses on analyzing the noise frequency offset caused by abnormal operating status of the electromechanical equipment, and obtains the optimal control solution through particle swarm optimization, thereby improving the noise reduction control accuracy.

[0007] The purpose of this application is achieved through the following technical solutions.

[0008] The present application provides a method for reducing noise of electromechanical equipment based on machine learning, comprising: collecting sound pressure and vibration signals of electromechanical equipment during operation by means of an acoustic sensor and a fiber Bragg grating sensor, and collecting operation status data of the electromechanical equipment; wherein the operation status data includes rotation speed, vibration and temperature; the acoustic sensor includes a microphone; extracting features of the collected sound pressure and vibration signals to obtain noise features; sparsely reconstructing the noise features; and locating the sound source of the reconstructed noise features by means of a beamforming algorithm to obtain a noise distribution map; constructing a noise reduction model based on deep learning, wherein the noise reduction model takes the noise distribution map as input and the noise-free distribution map as output; and using the trained The denoising model performs denoising on the noise distribution diagram to obtain the denoised noise signal; the denoised noise signal is subjected to spectrum analysis, and the frequency within a preset frequency range of the noise signal is extracted through a peak search algorithm as the main frequency component of the noise; the collected operating status data is analyzed using the time-varying linear predictive coding algorithm TVLPC to obtain the noise frequency offset caused by the abnormal operating status of the electromechanical equipment; the noise main frequency component and the noise frequency offset are used as input, and the particle swarm algorithm is used to generate the optimal solution for the noise control of the electromechanical equipment; according to the optimal solution for noise control, the noise main frequency component is extracted and separated to obtain multiple phase-inverted noise reduction control signals for noise reduction control.

[0009] Among them, the fiber Bragg grating (FBG) sensor is a sensor based on fiber Bragg grating technology. It uses the Bragg grating inside the optical fiber to reflect light of a specific wavelength, and senses the change of external physical quantities by detecting the change in the wavelength of the reflected light. In this application, the fiber Bragg grating sensor is used to collect vibration signals during the operation of electromechanical equipment. When the equipment vibrates, the optical fiber is strained, causing the grating period to change and the wavelength of the reflected light to shift. By demodulating the wavelength displacement of the reflected light, high-sensitivity measurement of the vibration of the electromechanical equipment can be achieved.

[0010] Beamforming is a spatial filtering technology that adjusts the amplitude and phase of the signal received by the array sensor so that the main lobe of the array points to the target direction and the side lobe points to the noise direction, thereby increasing the gain of the target signal and suppressing noise interference. In this application, the beamforming algorithm is used to locate the sound source of the reconstructed noise characteristics. By using the noise signal received by the acoustic sensor array, the amplitude and phase of each channel signal are adjusted by designing an appropriate weight vector so that the response of the array is focused in the direction of the target noise source, thereby achieving spatial positioning of the noise source. Commonly used beamforming algorithms include delayed sum, Capon, MUSIC, etc.

[0011] The peak search algorithm is an algorithm for finding local maximum points in a data sequence. In this application, the peak search algorithm is used to extract the main frequency components from the spectrum of the noise signal after noise reduction. By setting a certain search range and threshold conditions, the local peak points whose amplitude exceeds the threshold are searched in the spectrum data, and the corresponding frequency values ​​are extracted as the main frequency components of the noise. Common peak search algorithms include derivative-based methods, sliding window methods, wavelet transform methods, etc. Peak search can quickly locate important components in the spectrum and provide frequency domain features for subsequent noise control.

[0012] Time-Varying Linear Predictive Coding (TVLPC) is an adaptive speech coding algorithm that achieves signal compression and parameter extraction by linearly predicting the short-term stationary characteristics of the signal. In this application, the TVLPC algorithm is used to analyze the operating status data of electromechanical equipment and extract the noise frequency offset caused by abnormal state. By performing autoregressive modeling on state data such as speed, vibration and temperature, the time-varying linear prediction coefficient is obtained, and then the noise frequency offset caused by abnormal state is estimated based on the change of the prediction coefficient. The TVLPC algorithm can track the time-varying characteristics of the signal, adaptively extract abnormal state characteristics, and provide a compensation basis for noise control.

[0013] Furthermore, feature extraction is performed on the collected sound pressure and vibration signals to obtain noise features, including: using edge computing to perform time-frequency domain analysis on the sound pressure and vibration signals through short-time Fourier transform to extract statistical features; wherein the statistical features include mean, variance and kurtosis; using edge computing, a wavelet packet decomposition algorithm is used to perform multi-scale decomposition of the sound pressure and vibration signals to obtain wavelet packet coefficients; the energy of each wavelet packet coefficient is calculated to obtain wavelet packet energy features; using edge computing, the frequency spectrum of the sound pressure and vibration signals is filtered through a Mel frequency filter group, and the logarithm of the filtered spectrum is taken, and then a discrete cosine transform is performed to obtain Mel frequency cepstrum coefficients MFCC as spectrum envelope features; a feature-level fusion algorithm is used to splice the statistical features, wavelet packet energy features and spectrum envelope features to obtain noise features.

[0014] Furthermore, to generate a noise distribution map, firstly, noise feature classification is performed, and the collected noise feature data is preprocessed to remove abnormal points and invalid data, and a standardized noise feature sample set X = {x1, x2, ..., x N}, where x i is the ith noise feature sample, and N is the total number of samples. The K-means clustering algorithm is used to classify the noise feature samples. First, K samples are randomly selected as the initial cluster centers, and then the sum of the squares of the distances from the samples to the cluster centers is minimized through iterative optimization, and the cluster centers and the category labels of the samples are updated until convergence. The clustering result can be expressed as C = {C1, C2, ..., C K}, where C k is the kth noise feature category, k = 1, 2, ..., K. Each category C k Contains a set of noise feature samples, denoted as Where N k is the number of samples of the k-th type of noise feature.

[0015] Learn the category dictionary, for each noise feature category C k , construct an overcomplete dictionary D k The initial dictionary can be obtained by randomly selecting N k samples as dictionary atoms, or use basic dictionaries such as DCT. Use K-SVD algorithm to k K-SVD minimizes the reconstruction error by alternating the steps of sparse coding and dictionary updating. Sparse coding: Fixed dictionary D k , for each sample x ki Perform sparse representation, that is, solve Constraints, ||a ki ||0≤T, get the sparse coefficient a kiCommonly used sparse representation algorithms include OMP, LARS, etc. Dictionary update: fixed sparse coefficient a ki , for each atom d in the dictionary kj Update. Decompose the residual matrix by SVD Solving for d kj Make Minimum, Updated kj and a kj (j) Repeat sparse coding and dictionary updating until the reconstruction error converges or the maximum number of iterations is reached. The optimal dictionary D for the k-th noise feature is obtained. k and the sparse coefficient A k ={a k1 ,a k2 ,.....,a kNk}.

[0016] Joint dictionary sparse representation, concatenate K optimal dictionaries into a joint dictionary D = [D1, D2, ..., D K ] is used to represent the newly collected noise features. For the newly collected noise features x, the OMP algorithm is used to perform sparse representation under the joint dictionary D to solve Constraint: ||α||0≤T, get the sparse coefficient α. Use the sparse coefficient α and the joint dictionary D to reconstruct the noise feature x and get the reconstructed noise feature x r =Dα.

[0017] The noise source is spatially located according to the spatial coordinates (x m ,y m ,z m ),m=1,2,.....,M, construct the element position matrix A and direction vector of the microphone array in is the pitch angle and azimuth of the target direction. The direction matrix H and steering vector d of the beamforming algorithm are constructed using the geometric information of the microphone array. The direction matrix H is designed according to the desired spatial response. Commonly used methods include delay summation and minimum variance distortion. The steering vector d is set according to the target direction and is used to point to the area of ​​interest. r Perform spatial filtering. r Framing, for each frame of data x r (t) is multiplied by the direction matrix H and the steering vector d to obtain the filtered output y(t) = d H Hx r(t), t is the frame index. y(t) is the noise signal in the target direction, which enhances the component in the target direction and suppresses the noise in other directions. The noise source is located on the spatial filtering output y(t). The GCC algorithm is used to calculate the generalized cross-correlation function of y(t) between each microphone pair, and the TDOA of each microphone pair is obtained by searching for the cross-correlation peak. Based on the TDOA observation, a group of localization equations for the spatial position of the noise source is constructed. Assume that the noise source is located at (x s ,y s ,z s ), the TDOA of microphones i and j is τ ij , then it satisfies

[0018] Where c is the speed of sound. Substituting the TDOA of each microphone pair into (x s ,y s ,z s ) is a nonlinear equation system with unknown variables. The least squares LS algorithm is used to solve the positioning equation system. The nonlinear equation system is transformed into a linear equation system, and (x s ,y s ,z s ), which is the spatial coordinate of the noise source.

[0019] Generate a noise distribution map and locate the noise source coordinates (x s ,y s ,z s ) is matched with the three-dimensional model of the electromechanical equipment to mark the location of the noise source. A spatial interpolation algorithm, such as Kriging interpolation or radial basis function interpolation, is used to map the discrete noise source position coordinates to the continuous space of the three-dimensional model. Based on the interpolation results, an isosurface or cloud map of the noise distribution is generated to intuitively display the distribution of noise in different areas of the equipment.

[0020] Furthermore, the frequency within the preset frequency range of the noise signal is extracted by the peak search algorithm as the main frequency component of the noise. Specifically, the noise signal after noise reduction is processed by framing. Assume that the sampling frequency is f s , the frame length is N, the frame shift is M, then the noise signal after framing can be expressed as x(n,m), where n=0,1,......,N-1 is the time index within the frame, and m=0,1,......,M-1 is the frame index. Add a Hamming window to the noise signal after framing. The Hamming window function can be expressed as The noise signal after windowing is

[0021] x w(n,m)=x(n,m)×w(n). Perform fast Fourier transform (FFT) on the windowed noise signal. The number of FFT points is N, and the spectrum of the noise signal is X(k,m), where k=0,1,.....,N-1 is the frequency index. The frequency resolution is f s / N.

[0022] Search for the peak value, based on the speed f of the electromechanical equipment r , set the search range of the main frequency component of the noise signal spectrum. r As the center frequency of the search range, let the frequency multiplication order be L, then the center frequency of the lth frequency multiplication is

[0023] lf r ,l=1,2,.....,L, with center frequency lf r As a benchmark, set the search bandwidth to B, then the search range for the lth frequency doubling is

[0024] In each search range, a peak search algorithm is used to detect local peak points. Commonly used peak search algorithms include first-order difference method, parabola interpolation method, etc. Suppose the peak point frequency index obtained by the search is

[0025] k l (p), p=1,2,.....,P0, P0 is the peak point number, k l (p) corresponds to the peak amplitude |X(k l (p),m). According to the peak amplitude, select the first P0 peak points and extract the corresponding frequency value As a candidate for the main frequency component of noise.

[0026] Combine the frequencies and calculate the frequency difference between the two main frequency components of the candidate noise to obtain the frequency difference matrix D(p,q)=|f l (p)-f l (q),p,q=1,2,.....,P0. Set the frequency difference threshold T. When D(p,q)<T, it is considered that f l (p) and f l (q) The corresponding candidate main frequency components can be merged. For the candidate main frequency components that can be merged, take their average frequency as the merged main frequency, and update the frequency value and number of the candidate main frequency components. The number of updated candidate main frequency components is recorded as P1, and the frequency value is recorded as f c (p),p=1,2,.....,P1.

[0027] Estimate the ESPRIT frequency and use the spectral peaks corresponding to the updated candidate main frequency components to construct the signal subspace of the ESPRIT algorithm. Assume that the dimension of the signal subspace is equal to P1, and the corresponding spectral peak vector is

[0028] s=[X(k c (1),m),X(k c (2),m),......,X(k c (P1),m)] T , where k c (p) is f c (p) The frequency index corresponding to the candidate main frequency component is taken as the noise subspace. According to the ESPRIT algorithm principle, the corrected frequency value of the candidate main frequency component can be estimated by solving the rotational invariance of the signal subspace. The corrected frequency value is recorded as f r (p),p=1,2,.....,P1.

[0029] Select the main frequency component and adjust the corrected frequency value f r (p) are sorted, and the P2 frequencies with the largest values ​​are selected as the final noise main frequency components. The value of P2 can be flexibly set according to the spectrum characteristics of the noise signal and the noise reduction requirements, and should generally be less than or equal to P1. The final noise main frequency component frequency is recorded as f f (p), p=1,2,.....,P2. At the same time, the amplitude and phase parameters of the corresponding frequency components estimated by the ESPRIT algorithm are output for subsequent reconstruction of the noise reduction control signal. In this application, peak search and frequency merging can preliminarily determine the candidate main frequency components, reduce the amount of data, and reduce the complexity of calculation. The ESPRIT algorithm further improves the frequency estimation accuracy based on the candidate main frequency. Finally, the most significant main frequency component in the noise signal is obtained through sorting and screening.

[0030] Furthermore, the noise frequency offset caused by the abnormal operation state of the electromechanical equipment is obtained, including: using the time-varying linear prediction coding algorithm TVLPC to model the collected operation state data, using the AR model to describe the time-varying characteristics of the state data, and estimating the time-varying coefficients of the AR model through the Levinson-Durbin recursive algorithm to obtain the TVLPC model of the state data; wherein the AR model is the abbreviation of the Auto-Regressive model, which is a commonly used time series prediction model. In this application, the AR model is used to describe the time-varying characteristics of the operation state data of the electromechanical equipment. The AR model assumes that the state data at the current moment can be obtained by linearly combining the state data at the previous moments and adding noise, and its mathematical expression is: Where x(n) is the state data at the nth moment, a i (n) is the ith AR coefficient, p is the model order, and e(n) is the prediction error. The AR model characterizes the temporal correlation and evolution law of state data by estimating the optimal coefficients, thereby realizing the prediction of future states.

[0031] Among them, the Levinson-Durbin recursive algorithm

[0032] The Levinson-Durbin recursive algorithm is an efficient method for estimating the parameters of the AR model. In this application, the algorithm is used to estimate the AR model coefficients that describe the time-varying characteristics of the operating status data of electromechanical equipment. The Levinson-Durbin algorithm uses the Yule-Walker equation and recursive structure of the AR model to obtain the optimal AR coefficient estimate through iterative calculation. The basic steps are:

[0033] Calculate the autocorrelation function R of the state data xx (i,n)=E[x(n)x(ni)],i=0,1,...,p;

[0034] Initialize recursive variables: a0(n)=1, E0(n)=R xx (0,n), k=1; recursive calculation:

[0035]

[0036] a k (n) = [a k-1,1 (n-1),......,a k-1,k-1 (n-1),0]-k k (n)[0,a k-1,k-1 (n-1),......,a k-1,1 (n-1)]; update k=k+1, and repeat the steps until k=p+1.

[0037] The final result is a p (n)=[a1(n),a2(n),.....,a p (n)] is the p-order AR model coefficient of the state data at the nth moment.

[0038] Among them, time-varying characteristics refer to the property that the statistical characteristics of a signal change over time. In this application, time-varying characteristics are used to characterize the non-stationarity and dynamic change law of the operating state data of electromechanical equipment. The statistical parameters of state data with time-varying characteristics, such as mean, variance, autocorrelation function, etc., will change over time, showing a certain trend or periodicity. The AR model introduces the time-varying coefficient a i (n) is used to describe the time-varying characteristics of the state data, where the time dependence of the coefficient reflects the dynamic association and evolution law of the state data at different times. The time-varying AR model can be used to capture the time-varying patterns in the state data and realize dynamic monitoring and prediction of the equipment operation status.

[0039] Using the TVLPC model, the one-step prediction value and multi-step prediction value of the state data are calculated, and the prediction value is used as the future change trend of the state data; based on the collected operating status data and the future change trend obtained by the TVLPC model, the prediction error of the state data is calculated, and by setting the prediction error threshold, it is determined whether the operating state of the electromechanical equipment is abnormal; for the operating state determined to be abnormal, the frequency spectrum and power spectral density PSD of the prediction error sequence are calculated to obtain the characteristic frequency components reflecting the abnormal operation of the electromechanical equipment; wherein, the power spectral density (PSD) is a physical quantity that describes the distribution of signal power in the frequency domain. In this application, PSD is used to analyze the frequency domain characteristics of the prediction error sequence when the electromechanical equipment operates abnormally. PSD represents the average power of the prediction error sequence in the unit frequency band, reflecting the contribution of different frequency components to the error energy. For the discrete time series e(n), n=1,2,.....,N, its PSD can be estimated by the periodogram method: in, is the discrete Fourier transform of e(n), k=0,1,.....,N-1 is the normalized frequency. The role of PSD in anomaly detection is to identify the significant frequency components related to the abnormal state by observing the spectral distribution of the prediction error, and then extract the abnormal characteristic frequency for subsequent fault diagnosis and noise control.

[0040] The characteristic frequency components of the abnormal operation are correlated with the extracted main frequency components of the noise, the frequency offset of the characteristic frequency components of the abnormal operation relative to the main frequency components of the noise is calculated, the mapping relationship between the abnormal operation state of the electromechanical equipment and the noise frequency offset is established, and the noise frequency offset caused by the abnormal state of the electromechanical equipment is obtained.

[0041] Furthermore, the noise main frequency component and the noise frequency offset are used as inputs, and the particle swarm algorithm is used to generate the optimal solution for the electromechanical equipment noise control, including: establishing the optimization problem of electromechanical equipment noise control, taking the noise reduction effect as the optimization goal, constructing the objective function of the noise attenuation, taking the controller parameters as the optimization variables, and establishing the mapping relationship between the optimization variables and the optimization goals, wherein the controller parameters include the control gain, the filter order and the damping coefficient; taking the extracted noise main frequency component as the frequency constraint condition of the objective function, and taking the calculated noise frequency offset as the correction coefficient of the objective function; using the particle swarm optimization algorithm PSO to solve the optimal solution set of the controller parameters; wherein the controller parameters are taken as the position vector of the particles, and the noise attenuation is taken as the fitness function of the particles, and the optimal solution is searched in the parameter space; according to the user's noise reduction requirements, the solution with the largest noise attenuation is selected from the optimal solution set as the optimal solution of the controller parameters.

[0042] Furthermore, according to the optimal solution of noise control, the main frequency component of the noise is extracted and separated to obtain multiple phase-inverted noise reduction control signals, and noise reduction control is performed, including: according to the optimal solution, the parameters of the controller are set; the extracted main frequency component of the noise is tracked by a joint spectrum analysis algorithm, and the instantaneous frequency and amplitude of the main frequency component of the noise is estimated by short-time Fourier transform to obtain the dynamic spectrum of the main frequency of the noise; according to the dynamic spectrum of the main frequency of the noise, the parameters of each main frequency component are extracted by a matching pursuit algorithm, and the parameters of the main frequency component include frequency, amplitude and phase; according to the extracted parameters of each main frequency component, a control signal with equal amplitude, same frequency and opposite phase to each main frequency component is synthesized by frequency modulation and phase inversion algorithm; the synthesized control signal is input into the controller after setting the parameters to perform noise reduction control of electromechanical equipment.

[0043] Among them, the joint spectrum analysis algorithm is an algorithm for analyzing the correlation between two signals in the time-frequency domain. In the present application, the joint spectrum analysis algorithm is used to track the time-varying characteristics of the main frequency component of the noise. The algorithm obtains their time-frequency representation by calculating the short-time Fourier transform of the noise signal and the reference signal, and then calculates their cross-power spectrum and cross-correlation coefficient at each time-frequency point to obtain the joint spectrum. The joint spectrum is a two-dimensional complex matrix, whose amplitude reflects the degree of correlation between the two signals at different time-frequency positions, and the phase reflects their phase difference. By analyzing the peak value and continuity of the joint spectrum, the trajectory and parameters of the main frequency component of the noise in the time-frequency domain can be extracted.

[0044] Dynamic spectrum refers to the spectrum that changes with time, reflecting the time-varying characteristics of the signal frequency components. In this application, the dynamic spectrum is used to represent the changes in the instantaneous frequency and amplitude of the main frequency component of the noise. By performing a short-time Fourier transform on the noise signal, the spectrum at different times is obtained to form a two-dimensional time-frequency representation, which is the dynamic spectrum of the noise signal. The horizontal axis of the dynamic spectrum is time, the vertical axis is frequency, and the color or grayscale represents the amplitude of the frequency component. By analyzing the energy distribution and peak trajectory in the dynamic spectrum, the frequency and amplitude of the main frequency component of the noise can be tracked over time, providing a basis for subsequent parameter extraction and control signal synthesis.

[0045] The matching pursuit algorithm is an algorithm for tracking changes in target signal parameters. In the present application, the matching pursuit algorithm is used to extract the parameters of each main frequency component, including frequency, amplitude and phase, from the dynamic spectrum of the noise main frequency. The algorithm realizes continuous tracking of the main frequency component by searching for the most matching peak point between adjacent time frames of the dynamic spectrum. The specific steps include: searching for local peak points in the dynamic spectrum of the current time frame as candidate main frequency components; finding the component closest to the frequency and amplitude of the current candidate component in the main frequency component set of the previous time frame as a matching reference; calculating the similarity score between the candidate component and the reference component according to the matching degree of frequency and amplitude; selecting the candidate component with the highest similarity score as the main frequency component parameter at the current moment. By iteratively executing the above steps, continuous tracking and extraction of the main frequency component parameters are achieved.

[0046] The phase inverse algorithm is an algorithm for generating a control signal with a phase opposite to that of the target signal. In this application, the phase inverse algorithm is used to synthesize a control signal with the same amplitude, frequency and phase opposite to the main frequency component of the noise for active noise reduction control. Specifically, for a sinusoidal noise signal If a control signal with the same amplitude and frequency but a phase difference of π is generated The two signals will cancel each other out after being superimposed, thus weakening or eliminating the noise signal. In practical applications, the phase inverse algorithm needs to generate a corresponding control signal in real time based on the extracted parameters of the main frequency component of the noise. The specific steps include: for each main frequency component, a sinusoidal signal with the same frequency is generated according to its frequency parameter; according to the amplitude parameter, the amplitude of the sinusoidal signal is adjusted to be equal to the amplitude of the main frequency component; according to the phase parameter, the sinusoidal signal is phase shifted by π to obtain a control signal component with opposite phase; the control signals of each main frequency component are superimposed to obtain a total noise reduction control signal. The control signal generated by the phase inverse algorithm can achieve precise interference and cancellation at the noise source to achieve the purpose of noise reduction.

[0047] Furthermore, a noise reduction model based on deep learning is constructed, wherein the noise reduction model takes the noise distribution map as input and the noise-free distribution map as output; the noise distribution map is subjected to noise reduction processing using the trained noise reduction model to obtain the noise signal after noise reduction, including: collecting noise signals of electromechanical equipment under different working conditions, and performing time-frequency analysis and sound source positioning on the collected noise signals to obtain a noise distribution map corresponding to the noise signal; collecting a noise-free signal corresponding to the noise signal, and performing time-frequency analysis and sound source positioning on the noise-free signal to obtain a noise-free distribution map corresponding to the noise-free signal; and constructing a noise reduction model based on an encoder-decoder structure, wherein the encoder extracts multi-scale features of the noise distribution map through a convolutional neural network CNN, and the decoder extracts multi-scale features of the noise distribution map through a deconvolution neural network CNN. The extracted multi-scale features are reconstructed into a noise-free distribution map through the network and jump connections; the weighted combination of reconstruction loss and adversarial loss is used to construct the loss function of the denoising model; the noise distribution map is used as the input sample of the training data, and the corresponding noise-free distribution map is used as the output sample of the training data. The constructed denoising model is trained using the training data, the loss function is calculated through forward propagation, and the model parameters are updated through back propagation; the noise distribution map corresponding to the newly collected noise signal is used as input, and the trained denoising model is used to extract the features of the noise distribution map through the encoder, and the noise-free distribution map is reconstructed by the decoder to obtain the denoised noise distribution map; the denoised noise distribution map is converted into a denoised noise signal through inverse Fourier transform.

[0048] Compared with the prior art, the advantages of this application are:

[0049] A noise feature dimensionality reduction and reconstruction method based on dictionary learning and sparse representation is proposed. Through cluster analysis and K-SVD dictionary training, an over-complete dictionary of noise features is adaptively constructed, and sparse coding is used to achieve low-dimensional representation of noise features, effectively removing redundant information of noise features. The reconstructed noise features are combined with a beamforming algorithm to locate the sound source, and the spatial distribution map of the noise source is obtained, which intuitively shows the spatial distribution characteristics of equipment noise and provides a basis for noise source identification and location.

[0050] Aiming at the problem of noise frequency deviation caused by the change of the operating state of electromechanical equipment, a noise frequency deviation estimation method based on time-varying linear predictive coding (TVLPC) is proposed. By modeling the equipment operating state data with TVLPC, the future change trend of the equipment state is predicted, and the spectrum characteristics of the state prediction error are calculated to extract the characteristic frequency components reflecting the abnormality of the equipment. By correlating and analyzing the main frequency of the noise and the abnormal characteristic frequency, the mapping relationship between the equipment state and the noise frequency deviation is established, and the adaptive estimation of the noise frequency deviation is realized, which provides a basis for frequency deviation compensation for subsequent control decisions.

[0051] The particle swarm optimization algorithm is used to build a multi-objective optimization model for electromechanical equipment noise control, with noise reduction effect as the goal, controller parameters as optimization variables, noise main frequency as frequency constraint, and noise frequency deviation as correction coefficient. Through global optimization, the optimal solution set of noise controller parameters is obtained, and the parameter self-tuning of the noise control system is realized, avoiding the blindness and uncertainty of manual parameter adjustment, and improving the accuracy and efficiency of noise control. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The present application will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0053] Figure 1 is an exemplary flow chart of a method for reducing noise of electromechanical equipment based on machine learning according to some embodiments of the present application;

[0054] Figure 2 is an exemplary flow chart of generating noise characteristics according to some embodiments of the present application;

[0055] Figure 3 is an exemplary flow chart of generating a noise distribution map according to some embodiments of the present application;

[0056] Figure 4 This is an exemplary flow chart for obtaining P2 noise main frequency components according to some embodiments of the present application. DETAILED DESCRIPTION

[0057] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0058] like Figure 1As shown in the figure, the acoustic pressure and vibration signals of the electromechanical equipment during operation are collected by acoustic sensors and fiber Bragg grating sensors, and the operating status data of the electromechanical equipment is collected; wherein the operating status data includes rotation speed, vibration and temperature; the acoustic sensor includes a microphone; feature extraction is performed on the collected sound pressure and vibration signals to obtain noise features; the noise features are sparsely reconstructed; and the beamforming algorithm is used to locate the sound source of the reconstructed noise features to obtain a noise distribution map; a noise reduction model based on deep learning is constructed, the noise reduction model takes the noise distribution map as input and the noise-free distribution map as output; the noise distribution map is denoised using the trained noise reduction model Perform noise reduction processing to obtain a noise signal after noise reduction; perform spectrum analysis on the noise signal after noise reduction, and extract the frequency within a preset frequency range of the noise signal through a peak search algorithm as the main frequency component of the noise; use the time-varying linear prediction coding algorithm TVLPC to analyze the collected operating status data to obtain the noise frequency offset caused by the abnormal operating status of the electromechanical equipment; use the main frequency component of the noise and the noise frequency offset as input, and use the particle swarm algorithm to generate the optimal solution for the noise control of the electromechanical equipment; according to the optimal solution for noise control, extract the main frequency component of the noise for separation, obtain multiple phase-inverted noise reduction control signals, and perform noise reduction control.

[0059] like Figure 2 As shown in the figure, an acoustic sensor array and a fiber Bragg grating (FBG) sensor array are arranged, wherein the acoustic sensor array is composed of multiple microphones for collecting sound pressure signals when the electromechanical equipment is running; the FBG sensor array is composed of multiple FBG sensors for collecting vibration signals when the electromechanical equipment is running. At the same time, a speed sensor, a vibration sensor and a temperature sensor are installed at key positions of the electromechanical equipment to collect the operating status data of the electromechanical equipment, including speed, vibration and temperature. The sound pressure signal output by the acoustic sensor array and the vibration signal output by the FBG sensor array are collected in real time using a data acquisition card and data acquisition software, and the collected analog signals are converted into digital signals. For the sound pressure signal, a 24-bit analog-to-digital converter (ADC) is used to ensure sufficient dynamic range and signal-to-noise ratio; for the vibration signal, a 16-bit ADC is used to meet the acquisition accuracy requirements of the vibration signal. According to the noise frequency range and vibration frequency range of the electromechanical equipment, a suitable sampling frequency, such as 44.1kHz or 48kHz, is set to avoid frequency aliasing and information loss.

[0060] The collected sound pressure and vibration signals are feature extracted to obtain noise features, including: using edge computing nodes to perform short-time Fourier transform (STFT) on the collected sound pressure and vibration signals to convert the time domain signals into time-frequency domain representations. The time domain signals are frame segmented using overlapping sliding windows, and each frame of the signal is multiplied by a Hamming window or a Hamming window to reduce the impact of spectrum leakage. According to the frequency range of the electromechanical equipment noise, a suitable window length and overlap ratio are selected, such as a window length of 1024 sampling points and an overlap ratio of 50%. Each frame of the signal is subjected to a fast Fourier transform (FFT) to obtain a short-time spectrum, i.e., a time-frequency domain representation. Statistical features are extracted from the short-time spectrum, including mean, variance, and kurtosis. The mean reflects the energy distribution of the spectrum, the variance reflects the degree of discreteness of the spectrum, and the kurtosis reflects the degree of peaks of the spectrum. The statistical feature vector of the spectrum is obtained by calculating the amplitude mean, variance, and kurtosis of each frequency point. The statistical feature vectors of all frames are spliced ​​to obtain a statistical feature sequence of the sound pressure signal and the vibration signal.

[0061] Using edge computing nodes, wavelet packet decomposition (WPD) is performed on the collected sound pressure signal and vibration signal. Select appropriate wavelet basis functions, such as db4 or sym4, to decompose the signal at multiple scales. At each scale, the wavelet coefficients are further binary decomposed to obtain wavelet packet coefficients. According to the frequency range and time resolution requirements of electromechanical equipment noise, select an appropriate number of decomposition layers, such as 4 or 5 layers. Calculate the energy characteristics of the coefficients of each wavelet packet node. Use the bi-norm or variance to measure the energy of the wavelet packet coefficients to obtain the energy value of the wavelet packet node. The energy values ​​of all wavelet packet nodes are combined into a feature vector to represent the energy distribution of the sound pressure signal and vibration signal in different frequency bands and time scales, that is, the wavelet packet energy characteristics.

[0062] The obtained short-time spectrum is subjected to Mel frequency filtering by using edge computing nodes. According to the auditory characteristics of the human ear, a Mel frequency filter group is designed to filter and weight the spectrum. The center frequency and bandwidth of the filter group are distributed according to the Mel frequency scale, with dense filters in the low-frequency part and sparse filters in the high-frequency part. The logarithm of the filtered spectrum is taken to obtain the Mel frequency logarithmic power spectrum. The Mel frequency logarithmic power spectrum is subjected to discrete cosine transform (DCT) to obtain the Mel frequency cepstrum coefficients (MFCC). DCT concentrates the energy of the spectrum on low-order coefficients, which is conducive to extracting the envelope characteristics of the spectrum. The first 12th to 20th order MFCCs are selected as the spectrum envelope features to represent the spectrum shape and dynamic changes of the sound pressure signal and vibration signal. The feature-level fusion algorithm is used to splice the statistical features, wavelet packet energy features and spectrum envelope features to obtain the noise feature vectors of the sound pressure signal and vibration signal. The feature vector contains multi-scale and multi-dimensional information in the time domain, frequency domain and time-frequency domain, and comprehensively describes the characteristics of electromechanical equipment noise.

[0063] In this application, by using time-frequency analysis methods such as STFT and WPD, the time-frequency joint features of the sound pressure signal and vibration signal are fully explored, overcoming the limitations of traditional time-domain or frequency-domain analysis; by using multi-domain features such as statistical features, wavelet packet energy features, and spectral envelope features, the noise characteristics are described from multiple perspectives such as energy distribution, frequency components, and time scales, improving the representation ability and discrimination of noise features; through the feature-level fusion algorithm, the multi-domain features are optimally combined, overcoming the deficiencies of single features, making full use of the complementarity of different features, and forming a joint representation of noise features.

[0064] As Figure 3 shown, the noise features are sparsely reconstructed; and the beamforming algorithm is used to localize the sound source of the reconstructed noise features to obtain a noise distribution map, including: based on the extracted noise features, the K-means clustering algorithm is used to classify the noise features. The noise feature vector is used as the input of the clustering. Through iterative optimization, the within-class distance is minimized and the between-class distance is maximized to obtain K noise feature classes. The value of K is determined according to the complexity of the electromechanical equipment noise and prior knowledge, usually taking values from 3 to 10. For each noise feature class, an overcomplete dictionary is constructed. The number of atoms in the overcomplete dictionary is greater than the dimension of the noise features, and it has redundancy and sparsity. The K-SVD algorithm is used to iteratively train the overcomplete dictionary. By alternately performing sparse coding and dictionary update, the reconstruction error is minimized to obtain the optimal dictionary corresponding to the noise feature class. The number of atoms in the dictionary and the number of iterations are balanced according to the reconstruction accuracy and computational efficiency.

[0065] Construct a joint dictionary. For the K noise feature classes, the respective optimal dictionaries D k , k = 1, 2,....., K are obtained through the K-SVD dictionary learning algorithm, where the number of columns of D k is the number of dictionary atoms and the number of rows is the dimension of the noise features. The K optimal dictionaries are concatenated by columns to form a joint dictionary D = [D1, D2,......, D K . The number of columns of the joint dictionary is the sum of all dictionary atoms, and the number of rows is the same as that of a single dictionary. Use the joint dictionary for sparse representation. For the newly collected noise feature y, use the joint dictionary D for sparse representation to obtain the sparse coefficient α, such that y ≈ Da and the number of non-zero elements of α is as small as possible. The orthogonal matching pursuit (OMP) algorithm is used to solve the sparse representation problem: initialize the residual vector r = y, the sparse coefficient vector α = 0, the number of iterations t = 0, and the support set When t < T (T is the maximum number of iterations), obtain ||r|| 2 > ε (ε is the reconstruction error threshold): calculate the inner product of the residual r and each column in the joint dictionary D, and select the column index with the largest absolute value of the inner product Add it to the support set Λ = Λ ∪ {j}. Using the least squares method, solve the non-zero elements of α on the dictionary column corresponding to the support set Λ, so that Minimum. Update residual r = yD Λ a Λ , the number of iterations t = t + 1. Output the sparse coefficient α, where the index of the non-zero element is the element in the support set Λ, and the value of the non-zero element is the least squares solution. The number of non-zero elements in the sparse coefficient α is much smaller than the dimension of the noise feature y, achieving a low-dimensional representation of the noise feature.

[0066] Use the sparse coefficient to reconstruct the noise feature. Use the joint dictionary D and the sparse coefficient α to reconstruct the noise feature y and obtain the reconstructed noise feature y rec The reconstruction formula is: rec =Da, that is, multiply the sparse coefficient α by the joint dictionary D to obtain the reconstructed noise feature y rec The reconstructed noise feature y rec The redundancy and noise components in the original noise feature y are removed, and the quality and robustness of the feature are improved. This application uses K-means clustering and K-SVD dictionary learning to classify and sparsely reconstruct the noise features, remove the redundancy and noise components in the noise features, extract the essential patterns of the noise features, and improve the quality of the noise features. Specifically, first, the K-SVD dictionary learning algorithm is used to learn the optimal dictionary for each noise feature category, and then all the optimal dictionaries are spliced ​​by column to form a joint dictionary. For the newly collected noise features, the joint dictionary is used for sparse representation, and the OMP algorithm is used to solve the sparse representation problem, and the sparse coefficients of the noise features under the joint dictionary are obtained to achieve a low-dimensional representation of the noise features. Finally, the noise features are reconstructed using the sparse coefficients and the joint dictionary, and the redundancy and noise components in the noise features are removed, improving the feature quality and robustness. This application effectively mines the intrinsic structure and pattern of the noise features through dictionary learning and sparse representation, and reduces the dimension of the noise features.

[0067] According to the spatial position coordinates of each microphone in the acoustic sensor array, the array signal processing algorithm is used to calculate the array element position matrix and direction vector of the microphone array. The array element position matrix represents the geometric arrangement of the microphone array, and the direction vector represents the directivity of the microphone array. The direction matrix and steering vector of the beamforming algorithm are constructed using the geometric information of the microphone array for spatial filtering and sound source localization. The reconstructed noise features are used as the input of the beamforming algorithm, and the noise features are spatially filtered using the direction matrix and steering vector. By adjusting the direction matrix and steering vector, the main lobe of the microphone array is aligned with the target direction, interference from other directions is suppressed, and the noise signal in the target direction is extracted to obtain the spatial filtering result. Spatial filtering enhances the target noise signal and provides high-quality input for sound source localization. The beamforming algorithm is used to perform spatial filtering on the reconstructed noise features, which enhances the noise signal in the target direction, suppresses interference from other directions, and improves the accuracy and reliability of sound source localization. Spatial filtering combines the spatial correlation and directionality of the noise signal, fully utilizes the spatial sampling advantages of the acoustic sensor array, and provides high-quality input for noise source localization.

[0068] For the noise signal after spatial filtering, the generalized cross-correlation (GCC) algorithm is used to calculate the cross-correlation function of the noise signal of each channel in the acoustic sensor array. By finding the maximum value of the cross-correlation function, the arrival time difference (TDOA) of the noise signal of each channel is estimated. TDOA reflects the time difference of the sound source arriving at different microphones and contains the information of the spatial position of the sound source. Taking TDOA as the observation quantity and combining it with the direction matrix of the microphone array, the localization equation group of the spatial position of the noise source is constructed. The localization equation group establishes the mathematical relationship between TDOA and the spatial coordinates of the sound source and is a nonlinear equation group. The least squares (LS) algorithm is used to solve the localization equation group. The spatial position coordinates of the noise source are obtained by minimizing the sum of squares of the error between the TDOA observation value and the estimated value. The LS algorithm has the characteristics of fast convergence and global optimization, and is suitable for real-time sound source localization. Through the GCC algorithm and the LS algorithm, the TDOA of the noise signal is estimated, the spatial position coordinates of the noise source are solved, and the accurate localization of the noise source is achieved. Sound source localization reveals the spatial distribution law of the noise source and provides a basis for targeted noise control. Combined with the three-dimensional model of the electromechanical equipment, a noise distribution map is generated, which intuitively shows the location and intensity of the noise source. According to the spatial position coordinates of the noise source, combined with the three-dimensional model of the electromechanical equipment, the noise distribution map is generated by using spatial interpolation and mapping algorithms. The noise distribution map intuitively shows the position and intensity distribution of the noise source in the electromechanical equipment space in the form of color or contour lines.

[0069] A noise reduction model based on deep learning is constructed. The noise reduction model takes the noise distribution map as input and the noise-free distribution map as output. The noise distribution map is subjected to noise reduction processing using the trained noise reduction model to obtain the noise signal after noise reduction, including: collecting noise signals of electromechanical equipment under different working conditions, including working conditions of different speeds, loads, fault types, etc. Time-frequency analysis and sound source location are performed on the collected noise signals to obtain the noise distribution map corresponding to the noise signal. The noise distribution map represents the joint distribution characteristics of the noise signal in the time-frequency domain and the spatial domain, and provides input samples for the training of the noise reduction model. Under the same working conditions, a noise-free signal corresponding to the noise signal is collected as the target of noise reduction. Time-frequency analysis and sound source location are performed on the noise-free signal to obtain the noise-free distribution map corresponding to the noise-free signal. The noise-free distribution map represents the time-frequency domain and spatial domain characteristics of the electromechanical equipment during normal operation, and provides output samples for the training of the noise reduction model.

[0070] A denoising model based on an encoder-decoder structure is constructed. The encoder uses a convolutional neural network (CNN) to extract the multi-scale features of the noise distribution map through multi-layer convolution and pooling operations. The decoder uses a deconvolutional neural network to gradually restore the extracted multi-scale features to a noise-free distribution map through multi-layer deconvolution and upsampling operations. A skip connection is introduced between the encoder and the decoder to splice the shallow features of the encoder with the deep features of the decoder, retaining the detailed information of the noise distribution map and improving the reconstruction quality. An end-to-end denoising model is constructed to directly establish the mapping relationship from the noise distribution map to the noise-free distribution map, avoiding the complex noise modeling and decomposition process in the traditional method, simplifying the denoising process and improving the denoising efficiency. The encoder-decoder structure is adopted, and the convolutional neural network is used to extract the multi-scale features of the noise distribution map, fully mining the joint features of the noise signal in the time-frequency domain and the spatial domain, capturing the local and global patterns of the noise, fusing the shallow features of the encoder with the deep features of the decoder, retaining the detailed information of the noise distribution map, improving the fidelity of the signal after denoising, and avoiding the information loss caused by over-smoothing.

[0071] The loss function of the denoising model is constructed by using a weighted combination of reconstruction loss and adversarial loss. The reconstruction loss measures the pixel-level difference between the noise distribution map after denoising and the noise-free distribution map, and is calculated using mean square error (MSE) or mean absolute error (MAE). The adversarial loss introduces a discriminator network to determine the difference between the generated noise-free distribution map and the true noise-free distribution map, and is calculated using binary cross entropy. By adjusting the weights of the reconstruction loss and adversarial loss, the fidelity and realism of the denoising are balanced. The reconstruction loss ensures that the noise distribution map after denoising is close to the noise-free distribution map at the pixel level, maintaining the detailed characteristics of the signal. The adversarial loss uses the discriminator network to identify the difference between the generated noise-free distribution map and the true noise-free distribution map, making the generated noise-free distribution map more realistic and having better visual quality.

[0072] The noise distribution map is used as the input sample of the training data, and the corresponding noise-free distribution map is used as the output sample of the training data to construct a paired training data set. The constructed denoising model is trained end-to-end using the training data. In the forward propagation process, the noise distribution map is input into the encoder to extract multi-scale features, and then the noise-free distribution map is reconstructed through the decoder to calculate the reconstruction loss and adversarial loss. In the back propagation process, according to the gradient of the loss function, the optimization algorithm (such as Adam) is used to update the parameters of the model and iteratively optimize the denoising performance. The noise distribution map corresponding to the newly collected noise signal is used as input, and the trained denoising model is used for inference. The noise distribution map extracts multi-scale features through the encoder, and then the noise-free distribution map is reconstructed through the decoder to obtain the denoised noise distribution map. The denoised noise distribution map retains the key components of the noise signal, removes noise interference, and achieves effective noise suppression.

[0073] The noise distribution map after denoising is converted into a time domain signal through methods such as inverse Fourier transform or inverse wavelet transform to obtain a noise signal after denoising. Compared with the original noise signal, the noise component of the noise signal after denoising is significantly weakened and the signal quality is significantly improved, providing a high-quality data foundation for subsequent feature extraction and fault diagnosis.

[0074] The spectral analysis of the noise signal after noise reduction is performed, and the frequency within the preset frequency range of the noise signal is extracted by the peak search algorithm as the main frequency component of the noise, including: performing spectral analysis on the obtained noise signal after noise reduction. The noise signal is framed and windowed using a Hamming window with a frame length of N to reduce the influence of spectrum leakage. The windowed noise signal is subjected to a fast Fourier transform (FFT) to convert the time domain signal into a frequency domain signal to obtain a spectral representation of the noise signal. The spectrum represents the amplitude distribution of the noise signal at different frequencies.

[0075] like Figure 4As shown, according to the rotation speed information of the electromechanical equipment, the search range of the main frequency component of the noise signal spectrum is set. Specifically, the rotation speed information of the electromechanical equipment is obtained. The rotation speed data of the electromechanical equipment is collected in real time through a rotation speed sensor installed on the electromechanical equipment, such as an encoder or a Hall sensor. The rotation speed data can be expressed as revolutions per minute (RPM) or revolutions per second (Hz). According to the type and noise characteristics of the electromechanical equipment, the frequency multiplication relationship between the main frequency component and the rotation frequency is determined. Generally, the main frequency component of the noise of the electromechanical equipment appears at an integer multiple frequency of the rotation frequency, such as 1 times, 2 times, 3 times, etc. Different types of electromechanical equipment may have different frequency multiplication characteristics, which need to be determined based on the structure of the equipment, the noise mechanism and empirical knowledge. Calculate the fundamental frequency value corresponding to the rotation frequency. According to the obtained rotation speed data, convert it into a frequency value as the fundamental frequency of the noise signal spectrum. For example, if the rotation speed is 1500RPM, the fundamental frequency is 1500 / 60=25Hz. Determine the frequency multiplication range of the main frequency component of the noise. According to the determined frequency multiplication relationship, set the frequency multiplication range of the main frequency component of the noise, usually between 1 and 10 times. For example, if the frequency multiplication relationship is 1, 2, 3 times, the frequency multiplication range of the main frequency component of the noise is [1, 2, 3]. Calculate the frequency range of the main frequency component of the noise. Multiply the base frequency value by the frequency multiplication range to obtain the frequency range of the main frequency component of the noise. For example, if the base frequency is 25Hz and the frequency multiplication range is [1, 2, 3], the frequency range of the main frequency component of the noise is [25, 50, 75]Hz. Consider the uncertainty of the frequency range. Since the actual speed may fluctuate, there is a certain uncertainty in the frequency estimation of the base frequency and the main frequency component. In order to improve the robustness of the main frequency component search, a certain frequency tolerance can be introduced on the basis of the frequency range. For example, the frequency range can be expanded to [(1-ε)×f, (1+ε)×f], where f is the original frequency value and ε is the frequency tolerance coefficient, which is usually 0.05 to 0.1. The calculated frequency range of the main frequency component of the noise is used as the target area for the subsequent spectrum peak search. The spectrum peak search algorithm will perform local peak point detection within this frequency range to extract the main frequency component of the noise signal.

[0076] In the search range of the main frequency component, the peak search algorithm is used to detect local peak points. By sliding the window on the spectrum and comparing the spectrum amplitude within the window, the local maximum point is found as the candidate noise main frequency component. The P0 local peak points with the largest spectrum amplitude are selected and the corresponding frequency values ​​are extracted as the preliminary screened noise main frequency components.

[0077] Frequency merging operation, calculate the frequency difference of the candidate noise main frequency components. For the obtained P0 candidate noise main frequency components, calculate the frequency difference between them. Suppose the frequency value of the candidate main frequency component is The frequency difference ΔF is calculated as: ΔF(i,j)=|f i -fj |,i,j=1,2,......,P0, and i≠j; set the frequency merging threshold ε. According to the type of electromechanical equipment and the noise characteristics, set a frequency merging threshold ε to determine whether two candidate main frequency components belong to the same main frequency component. The selection of threshold ε needs to balance the frequency resolution and frequency estimation error, usually 1 to 2 times the spectrum resolution. Determine whether the frequency difference is less than the threshold. For each pair of candidate main frequency components, determine whether their frequency difference ΔF(i,j) is less than the frequency merging threshold ε. If ΔF(i,j)<ε, it is considered that the two candidate main frequency components are caused by the same main frequency component and need to be merged. Merge candidate main frequency components. For the candidate main frequency components that meet the merging conditions, take the average of their frequency values ​​as the frequency of the merged main frequency component. Assume that the frequency value of the candidate main frequency component to be merged is Then the frequency of the combined main frequency component is f merge The calculation formula is:

[0078] Update the set of candidate main frequency components. Add the merged main frequency component frequency to the set of candidate main frequency components, and delete the original merged candidate main frequency components. Repeat until all candidate main frequency components that meet the merging conditions have been processed. The frequency merging operation calculates the frequency difference between the candidate main frequency components to determine whether they belong to the same main frequency component, and averages the candidate main frequency components that meet the merging conditions to obtain the merged main frequency component frequency. The frequency merging operation can eliminate the main frequency component frequency estimation deviation caused by the limitation of the spectrum resolution, and improve the accuracy of the main frequency component frequency estimation.

[0079] The ESPRIT algorithm extracts the main frequency component of the noise, and for the obtained noise signal spectrum, constructs the signal autocorrelation matrix R. Assuming the noise signal spectrum is X(k), k = 0, 1, ..., N-1, where N is the number of spectrum points, the calculation formula of the autocorrelation matrix R is: M is the order of the autocorrelation matrix, and * represents the complex conjugate. Perform eigenvalue decomposition on the autocorrelation matrix R to obtain the eigenvalues ​​λ1,λ2,......,λ M and the corresponding eigenvectors v1,v2,......,v M . Arrange the eigenvalues ​​in order from large to small, and the corresponding eigenvectors are also sorted accordingly. According to the size of the eigenvalue, select the eigenvectors corresponding to the first P largest eigenvalues ​​to form the signal subspace Us. The value of P is usually the number of main frequency components of the noise, which can be determined based on prior knowledge or through threshold judgment. Using the rotational invariance of the signal subspace Us, construct the frequency estimation matrix Φ. Assuming the dimension of Us is M×P, the calculation formula of the frequency estimation matrix Φ is: Among them U s1and U s2 The submatrices obtained by removing the last and first rows of Us are: Pseudo-inverse, usually Moore-Penrose Pseudo-inverse, is a concept used to process non-square (or singular square) matrices. Solve the eigenvalues ​​μ1, μ2, ..., μ of the frequency estimation matrix Φ P . Eigenvalue μ i The frequency f of the main frequency component of the noise i The following relationship exists: Where angle(*) represents the phase angle of the complex number in radians. The frequency estimate f i As the frequency estimation result of the main frequency component of the noise, the ESPRIT algorithm achieves high-resolution frequency estimation by constructing the signal subspace and utilizing the subspace rotation invariance, and can accurately extract the frequency value of the main frequency component of the noise.

[0080] The ESPRIT algorithm constructs the signal's autocorrelation matrix, decomposes its eigenvalues, selects the signal subspace, uses the rotational invariance of the subspace to construct a frequency estimation matrix, solves the eigenvalues ​​of the matrix, and obtains the frequency estimation value of the noise main frequency component. The ESPRIT algorithm has high frequency resolution and anti-noise interference capabilities, and can accurately estimate the frequency value of the noise main frequency component, providing important frequency domain feature information for subsequent noise analysis and fault diagnosis. The frequency values ​​estimated by the ESPRIT algorithm are sorted, and the P1 frequency with the largest frequency value is selected as the final noise main frequency component. The value of P1 can be set according to the type of electromechanical equipment and empirical knowledge. Usually, 3 to 5 main frequency components can cover most of the noise energy. The final selected noise main frequency component represents the main frequency characteristics of the noise signal after noise reduction.

[0081] To obtain the noise frequency offset caused by the abnormal operation state of the electromechanical equipment, the operation state data is first modeled, and the collected electromechanical equipment operation state data is preprocessed, including denoising, normalization, etc., to obtain a standardized state data sequence x(n), n = 1, 2, ..., N, where N is the data length. The time-varying linear predictive coding algorithm (TVLPC) is used to model the state data x(n). TVLPC describes the time-varying characteristics of the data through an autoregressive (AR) model. The model order is p and the expression is where a i (n) is the time-varying AR coefficient, and e(n) is the prediction error.

[0082] The Levinson-Durbin recursive algorithm is used to estimate the time-varying AR coefficient of the TVLPC model. The algorithm iteratively calculates the AR coefficient by minimizing the mean square value of the prediction error. xx (i,n) is the autocorrelation function of the state data,

[0083] R xx (i,n)=E[x(n)x(ni)],i=0,1,.....,p, solve the following system of equations using the Levinson-Durbin algorithm:

[0084] According to the estimated time-varying AR coefficient a i (n), establish the TVLPC model of state data:

[0085] Specifically, the TVLPC algorithm proposes to use the time-varying AR model to describe the time-varying characteristics of state data, where the key parameter is the time-varying AR coefficient a i (n). The Levinson-Durbin algorithm is used to estimate the time-varying AR coefficient a in the TVLPC model. i (n) is an effective method to obtain the optimal AR coefficient estimate by recursively solving the Yule-Walker equation. i Substituting (n) into the TVLPC model, we get the time-varying AR model representation of the state data: The TVLPC algorithm provides the framework of the time-varying AR model, while the Levinson-Durbin algorithm is an effective tool for solving the parameters of the model.

[0086] Predict the operating status and use the established TVLPC model to calculate the one-step prediction value of the status data and multi-step forecasts m is the prediction step length. The one-step prediction formula is Multi-step prediction is achieved through iterative calculation: in Represents the predicted value at the n+mith moment.

[0087] 2.2 Predicted value and As the changing trend of status data at future moments, it reflects the expected evolution law of the operating status of electromechanical equipment.

[0088] Detect operational anomalies and calculate the prediction error sequence of status data The prediction error reflects the deviation between the actual collected state data and the predicted value of the TVLPC model. Set the prediction error threshold δ. When |e(n)|>δ, the operating state at the nth moment is judged to be abnormal; otherwise, it is judged to be normal. The threshold δ can be determined based on experience or statistical laws, such as δ=3σ, where σ is the standard deviation of the prediction error. Extract the prediction error data corresponding to the moment of abnormal operating state, and record it as the abnormal prediction error sequence e a (n),n=1,2,.....,N a , where N a is the number of abnormal states.

[0089] The abnormal characteristic frequency is extracted and the abnormal prediction error sequence ea(n) is analyzed in the frequency domain. Fast Fourier transform (FFT) is used to calculate e a Spectrum of (n) The spectrum reflects the amplitude distribution of different frequency components in the forecast error sequence. The power spectral density PSD of the abnormal forecast error sequence is calculated.

[0090] PSD describes the power distribution of the signal at each frequency point and can be estimated by the periodogram method:

[0091] According to the spectrum and PSD, the significant frequency components of the abnormal prediction error sequence are extracted as the characteristic frequencies reflecting the abnormal operation of the electromechanical equipment. By setting the amplitude threshold or statistical test method, the frequency points with concentrated energy and related to the abnormal state are screened out and recorded as the abnormal characteristic frequency set F. a ={f1,f2,.....,f M}, where M is the number of abnormal characteristic frequencies.

[0092] Calculate the noise frequency offset and record the extracted noise main frequency component as F0 = {F 01 ,F 02 ,.....,F 0K}, where K is the number of main frequency components. Using correlation analysis methods, such as cross-correlation function or coherence function, calculate the abnormal characteristic frequency set F a The correlation between each frequency and the main frequency component F0 of the noise. a Each abnormal characteristic frequency f in i ,i=1,2,....,M, find the noise main frequency F with the greatest correlation 0j ,j=1,2,.....,K. Calculate f i Relative to F 0j The frequency offset Δf i =f i -F 0j, get the offset of the ith abnormal characteristic frequency relative to the main frequency of the noise. Establish a mapping relationship table between the abnormal operation status of electromechanical equipment and the noise frequency offset, including the abnormal characteristic frequency f i , the corresponding noise main frequency F 0j and frequency offset Δf i This mapping relationship links the abnormal state with the noise spectrum, providing a basis for subsequent noise control.

[0093] In this application, the collected operating status data is analyzed using the time-varying linear prediction coding algorithm TVLPC, and the noise frequency offset caused by the abnormal operating status of the electromechanical equipment is obtained. The TVLPC algorithm describes the time-varying characteristics of the status data through an autoregressive model, estimates the time-varying AR coefficient using the Levinson-Durbin recursive algorithm, and constructs a dynamic prediction model for the status data. Based on the TVLPC model, the predicted value of the status data is calculated, and whether the operating status of the electromechanical equipment is abnormal is judged by the prediction error. For abnormal states, the frequency domain characteristics of the prediction error sequence are analyzed, and the characteristic frequency components reflecting the abnormal operation are extracted. The abnormal characteristic frequency is correlated with the main frequency component of the noise, the frequency offset is calculated, and a mapping relationship between the abnormal operating status of the electromechanical equipment and the noise frequency offset is established. Using this mapping relationship, the offset of the noise frequency can be estimated according to the operating status prediction error, providing a basis for subsequent fault diagnosis and predictive maintenance.

[0094] The noise main frequency component and noise frequency offset are used as input, and the particle swarm algorithm is used to generate the optimal solution for electromechanical equipment noise control, including: establishing the optimization problem of electromechanical equipment noise control, taking the noise reduction effect as the optimization goal, and defining the objective function J(x) of noise attenuation: Among them A n is the amplitude of the original noise signal, A c is the amplitude of the noise signal after control, and x is the controller parameter vector. The controller parameters are selected as optimization variables, including the control gain K p ,K i ,K d , filter order N and damping coefficient ξ, constitute the optimization variable vector x = [K p ,K i ,K d ,N,ξ]. The mapping relationship between the optimization variable x and the optimization target J(x) is established, which is usually obtained by cascading the controller transfer function and the noise transfer function.

[0095] The main frequency component of the noise and the frequency offset are included in the constraints of the optimization problem. Construct frequency constraint: min(f i )≤f≤max(f i), where i = 1, 2, ....., P1, and f is the main frequency component of the noise signal after control. According to the calculated noise frequency offset Δf1, Δf2, ..., Δf M , construct the correction term of the objective function: Where i = 1, 2, ..., M, w i is the weight coefficient of the frequency offset.

[0096] The particle swarm optimization algorithm PSO is used to solve the optimal solution set of controller parameters, initialize the particle swarm, and randomly generate the position vector x of N particles. i and the velocity vector v i , where i = 1, 2, ..., N. Position vector x i Represents the value of the controller parameter, the velocity vector v i Indicates the search direction and step size of the parameters. Calculate the fitness value of each particle, that is, the objective function The value of represents the noise attenuation corresponding to the particle position. Update the individual optimal position p of each particle i and the global optimal position p g The optimal position of an individual is p i represents the optimal solution found by the i-th particle in the search history, and the global optimal position p g Represents the optimal solution found by the entire particle swarm in the search history. Update the velocity vector and position vector of each particle according to the individual optimal position and the global optimal position:

[0097] Where t is the number of iterations, w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers between [0, 1]. Repeat until the termination condition is met, such as reaching the maximum number of iterations or the change in fitness value is less than the threshold. Output the optimal solution set obtained by the particle swarm search, including the optimal values ​​of the controller parameters and the corresponding noise attenuation.

[0098] According to the user's noise reduction requirements, the optimal solution is selected from the optimal solution set. The solutions in the optimal solution set are sorted and arranged in order from large to small according to the noise attenuation. According to the noise reduction target specified by the user, such as the target value of the noise attenuation or the noise limit, the solution that meets the requirements is selected from the optimal solution set as the optimal solution of the controller parameters. If the user does not specify a specific noise reduction target, the solution with the largest noise attenuation is selected as the optimal solution by default. This application uses the noise main frequency component and the noise frequency offset as input, and uses the particle swarm algorithm to generate the optimal solution for electromechanical equipment noise control. By establishing the optimization problem of noise control, taking the noise reduction effect as the goal, the controller parameters as the optimization variables, the noise main frequency component as the frequency constraint condition, and the noise frequency offset as the correction term of the objective function, a mathematical model of noise control is constructed. The particle swarm optimization algorithm PSO is used to search for the optimal solution set of controller parameters in the parameter space through the position update and velocity update of the particles, and the parameter optimization of noise control is realized. According to the user's noise reduction requirements, the solution with the largest noise attenuation is selected from the optimal solution set as the final controller parameter, and the effective control of electromechanical equipment noise is realized. This application comprehensively utilizes the frequency domain characteristics of the noise signal and the time domain characteristics of the operating status of the electromechanical equipment, and finds the optimal parameters for noise control through an intelligent optimization algorithm.

[0099] According to the optimal solution of noise control, the main frequency component of noise is extracted and separated, and the noise reduction control of electromechanical equipment is performed. First, the controller parameters are set according to the optimal solution of noise control. The optimal solution contains the optimal gain, order, filter type and other parameter information of the controller. The parameter values ​​in the optimal solution are assigned to the controller, and the structure and coefficients of the controller are adjusted to achieve the optimal control performance. The type of controller can be an adaptive filter, a feedforward compensator, a feedback controller, etc., which is selected according to the specific control strategy.

[0100] The joint spectrum analysis algorithm is used to track the main frequency component of the noise, and the extracted main frequency component signal x(t) and the reference signal y(t) are preprocessed. The reference signal can be a delayed version of the noise signal, an approximate model of the noise signal, or other related signals. x(t) and y(t) are centered and normalized to remove the DC component and amplitude influence of the signal, and obtain zero-mean unit variance signals x'(t) and y'(t). The preprocessed signals x'(t) and y'(t) are framed and divided into overlapping frames of length N, with an overlapping length of L between frames. Each frame is represented by x i '(n) and y i '(n), i is the frame number, n is the time index within the frame, n = 0, 1, ....., N-1.

[0101] Calculate the time frame representation, for each frame signal xi '(n) and y i '(n) Perform short-time Fourier transform (STFT) to obtain their time-frequency representation X i (f,t) and Y i (f, t). The calculation formula of STFT is: Where f is the frequency variable, t is the time variable, and w(n) is the analysis window function, such as a Hamming window or a Gaussian window. Based on the STFT results, the power spectrum P of each frame is calculated. xx,i (f,t) and P yy,i (f, t), and the cross-power spectrum P xy,i (f,t). xx,i (f,t)=|X i (f,t)| 2 , P yy,i (f,t)=|Y i (f,t)| 2 , P xy,i (f,t)=X i (f,t)Y i *(f,t), where * represents the complex conjugate operation.

[0102] Estimate the joint spectrum, using the power spectrum and cross-power spectrum to calculate the joint spectrum C of x(t) and y(t) xy (f, t). The calculation formula of the joint spectrum is: Where M is the total number of frames. Joint spectrum C xy (f,t) is a complex matrix whose magnitude |C xy (f, t)| reflects the correlation between x(t) and y(t) at frequency f and time t, and the phase ∠C xy (f, t) reflects their phase difference at frequency f and time t.

[0103] Track the main frequency of the noise, according to the amplitude of the joint spectrum |C xy (f,t)|, determine the time-frequency position of the main frequency component of the noise. |C xy The local peak of (f, t)| corresponds to the area where x(t) and y(t) are highly correlated in the time-frequency domain, representing the time-frequency trajectory of the main frequency component of the noise. xy (f, t)| performs peak search and threshold judgment to extract the significant main frequency component of noise. Set the amplitude threshold T, when |C xyWhen (f, t)|>T, it is considered that there is a significant main frequency component at (f, t). The extracted noise main frequency component is subjected to continuity analysis and trajectory tracking. Since the noise main frequency component is continuous in time, the most likely trajectory connection is searched between adjacent time frames of the joint spectrum to form a complete main frequency trajectory. The tracked noise main frequency trajectory is smoothed to remove isolated points and short-term mutations to obtain a stable main frequency component change curve.

[0104] According to the dynamic spectrum S of the main frequency of the noise k (t), the matching pursuit algorithm is used to extract the parameters of each main frequency component. Matching pursuit achieves continuous tracking of the spectrum peak and parameter estimation by measuring the similarity between the spectrum peak at the current moment and the spectrum peak at the previous moment. For the kth main frequency component, in its dynamic spectrum S k Search for the local peak point in (t) and extract the corresponding frequency f k (t), amplitude A k (t) and phase Parameters. Frequency parameter f k (t) is the peak frequency of the dynamic spectrum at time t; the amplitude parameter A k (t) is the amplitude value at the peak frequency; phase parameter It can be estimated by continuous wavelet transform or Hilbert transform. Smoothing is performed to remove noise and mutation points, and a continuous and smooth main frequency component parameter trajectory is obtained.

[0105] Synthesize the control signal, and use the frequency modulation and phase inversion algorithm to synthesize the control signal with the same amplitude, frequency and opposite phase as the main frequency component according to the extracted parameters of each main frequency component. For the kth main frequency component, synthesize the control signal corresponding to it Frequency modulation uses the instantaneous frequency of the main frequency component as the modulation frequency of the control signal, and the phase is reversed by introducing a phase shift of π based on the phase of the main frequency component to achieve vibration cancellation. The synthesized main frequency control signals are superimposed to obtain the total noise reduction control signal The control signal is synchronized with the main frequency component of the original noise in frequency, equal in amplitude, and opposite in phase, thereby achieving precise noise reduction control.

[0106] Output control signal, input the synthesized noise reduction control signal c(t) to the controller with set parameters. According to the input control signal, the controller generates secondary vibration or sound wave matching the main frequency component of the noise to offset and weaken the original noise. Through the action of the controller, the active noise reduction control of electromechanical equipment is realized, the influence of the main frequency component of the noise is reduced, and the vibration and noise performance of the equipment is improved.

Claims

1. A method for reducing noise of electromechanical equipment based on machine learning, characterized in that: include: Acoustic sensors and fiber Bragg grating sensors are used to collect sound pressure and vibration signals of electromechanical equipment during operation, and to collect operating status data of the electromechanical equipment; wherein the operating status data includes rotation speed, vibration and temperature; the acoustic sensor includes a microphone; Extract features from the collected sound pressure and vibration signals to obtain noise features; The noise features are sparsely reconstructed; and the sound source is located using the beamforming algorithm to obtain a noise distribution map. A denoising model based on deep learning is constructed. The denoising model takes the noise distribution map as input and the noise-free distribution map as output. The noise distribution map is denoised using the trained denoising model to obtain a denoised noise signal. Performing spectrum analysis on the noise signal after noise reduction, and extracting the frequency within a preset frequency range of the noise signal as the main frequency component of the noise through a peak search algorithm; The collected operating status data is analyzed using the time-varying linear predictive coding algorithm TVLPC to obtain the noise frequency offset caused by the abnormal operating status of the electromechanical equipment; Taking the main frequency component of noise and the frequency offset of noise as input, the particle swarm algorithm is used to generate the optimal solution for the noise control of electromechanical equipment. According to the optimal solution for noise control, the main frequency component of the noise is extracted and separated to obtain multiple phase-inverted noise reduction control signals for noise reduction control.

2. The method for reducing noise of electromechanical equipment based on machine learning according to claim 1, characterized in that: Extract features from the collected sound pressure and vibration signals to obtain noise features, including: Using edge computing, short-time Fourier transform is used to analyze the sound pressure and vibration signals in the time and frequency domains and extract statistical features, including mean, variance and kurtosis. Using edge computing, the wavelet packet decomposition algorithm is used to perform multi-scale decomposition of the sound pressure and vibration signals to obtain the wavelet packet coefficients; the energy of each wavelet packet coefficient is calculated to obtain the wavelet packet energy characteristics; Using edge computing, the spectrum of the sound pressure and vibration signals is filtered through a Mel frequency filter bank, and the logarithm of the filtered spectrum is taken, and then a discrete cosine transform is performed to obtain the Mel frequency cepstrum coefficient MFCC as the spectrum envelope feature; The feature-level fusion algorithm is used to combine the statistical features, wavelet packet energy features and spectrum envelope features to obtain the noise features.

3. The method for reducing noise of electromechanical equipment based on machine learning according to claim 2, characterized in that: Get the noise distribution diagram, including: Based on the noise features, the noise features are classified by cluster analysis algorithm to obtain K noise feature categories; For each noise feature category, an overcomplete dictionary is constructed, and the overcomplete dictionary is iteratively trained using the K-SVD algorithm to obtain the optimal dictionary corresponding to the noise feature; The newly collected noise features are sparsely represented using a joint dictionary composed of K optimal dictionaries. The sparse representation problem is solved by the orthogonal matching pursuit (OMP) algorithm to obtain the sparse coefficients of the noise features under the joint dictionary. Using the sparse coefficients of the noise features, the noise features are reconstructed through a joint dictionary to obtain the reconstructed noise features; According to the spatial position coordinates of each microphone in the acoustic sensor array, the array signal processing algorithm is used to calculate the array element position matrix and direction vector of the microphone array as geometric information; the geometric information of the acoustic sensor array is used to construct the direction matrix and steering vector of the beamforming algorithm; The reconstructed noise features are spatially filtered using the direction matrix and the steering vector to extract the noise signal in the target direction, and the spatial filtering result is used as the input for noise source positioning. The generalized cross-correlation GCC algorithm is used to calculate the cross-correlation function of the noise signal of each channel in the acoustic sensor array, and the arrival time difference TDOA of the noise signal of each channel is calculated by obtaining the maximum value of the cross-correlation function; Taking the time difference of arrival TDOA as the observation quantity, according to the direction matrix, the positioning equation group of the noise source spatial position is constructed; The least squares LS algorithm is used to solve the positioning equations to obtain the spatial position coordinates of the noise source; According to the spatial position coordinates of the noise source, combined with the three-dimensional model of the electromechanical equipment, the noise distribution map is generated through spatial interpolation and mapping algorithms.

4. The method for reducing noise of electromechanical equipment based on machine learning according to claim 3, characterized in that: The clustering analysis algorithm is the K-means clustering algorithm.

5. The method for reducing noise of electromechanical equipment based on machine learning according to any one of claims 1 to 4, characterized in that: The peak search algorithm is used to extract the frequency within the preset frequency range of the noise signal as the main frequency component of the noise, including: According to the noise signal after denoising, a Hamming window with a frame length of N is used to frame and window the noise signal, and a fast Fourier transform FFT is performed on the windowed noise signal to obtain a noise signal spectrum; According to the rotation speed of the electromechanical equipment, the search range of the main frequency component of the noise signal spectrum is set, and the multiple of the rotation speed frequency is used as the center frequency of the search range; In the main frequency component search range, the peak search algorithm is used to detect the local peak points of the noise signal spectrum, obtain the P0 local peak points with the largest spectrum amplitude, and extract the corresponding frequency values ​​as candidate noise main frequency components; The signal subspace of the ESPRIT algorithm is constructed using the P0 spectrum peaks corresponding to the candidate noise main frequency components, and the spectrum part except the spectrum peaks corresponding to the candidate noise main frequency components is used as the noise subspace; According to the number P0 of candidate noise main frequency components, the order of the ESPRIT algorithm is determined, and the order is equal to P0; The rotation invariance of the signal subspace is solved by the ESPRIT algorithm, and the corrected frequency values ​​of P1 candidate noise main frequency components are obtained; The calculated P1 corrected frequency values ​​are sorted, and the P2 frequencies with the largest corrected frequency values ​​are selected as the final noise main frequency components.

6. The method for reducing noise of electromechanical equipment based on machine learning according to claim 5, characterized in that: Before calculating the frequency value of the main frequency component of the noise, it also includes: Calculate the frequency difference between the candidate main frequency components of noise; When the frequency difference is less than the set threshold, the corresponding two candidate main frequency components are merged, and the average frequency value of the two candidate main frequency components is taken as the merged main frequency.

7. The method for reducing noise of electromechanical equipment based on machine learning according to claim 6, characterized in that: Obtain the noise frequency offset caused by abnormal operation of electromechanical equipment, including: The collected running status data is modeled using the time-varying linear predictive coding algorithm TVLPC, the time-varying characteristics of the status data are described using the AR model, and the time-varying coefficients of the AR model are estimated using the Levinson-Durbin recursive algorithm to obtain the TVLPC model of the status data. Using the TVLPC model, the one-step prediction value and multi-step prediction value of the state data are calculated, and the prediction value is used as the future change trend of the state data; According to the collected operating status data and the future change trend obtained by the TVLPC model, the prediction error of the status data is calculated, and by setting the prediction error threshold, it is determined whether the operating status of the electromechanical equipment is abnormal; For the operation status judged as abnormal, the characteristic frequency components reflecting the abnormal operation of electromechanical equipment are obtained by calculating the spectrum and power spectrum density PSD of the prediction error sequence; The characteristic frequency components of the abnormal operation are correlated with the extracted main frequency components of the noise, the frequency offset of the characteristic frequency components of the abnormal operation relative to the main frequency components of the noise is calculated, the mapping relationship between the abnormal operation state of the electromechanical equipment and the noise frequency offset is established, and the noise frequency offset caused by the abnormal state of the electromechanical equipment is obtained.

8. The method for reducing noise of electromechanical equipment based on machine learning according to claim 1, characterized in that: Generate the optimal solution for electromechanical equipment noise control using particle swarm optimization, including: Establish the optimization problem of electromechanical equipment noise control, take the noise reduction effect as the optimization goal, construct the objective function of noise attenuation, take the controller parameters as the optimization variables, and establish the mapping relationship between the optimization variables and the optimization goals. The controller parameters include control gain, filter order and damping coefficient. The extracted main frequency component of the noise is used as the frequency constraint condition of the objective function, and the calculated noise frequency offset is used as the correction coefficient of the objective function; The particle swarm optimization algorithm PSO is used to solve the optimal solution set of controller parameters. The controller parameters are used as the position vector of the particles, the noise attenuation is used as the fitness function of the particles, and the optimal solution is searched in the parameter space. According to the user's noise reduction requirements, the solution with the largest noise attenuation is selected from the optimal solution set as the optimal solution for the controller parameters.

9. The method for reducing noise of electromechanical equipment based on machine learning according to claim 8, characterized in that: According to the optimal solution of noise control, the main frequency components of the noise are extracted and separated, including: According to the optimal solution, set the parameters of the controller; The extracted main frequency components of noise are tracked by using joint spectrum analysis algorithm, and the instantaneous frequency and amplitude of the main frequency components of noise are estimated by short-time Fourier transform to obtain the dynamic spectrum of the main frequency of noise. According to the dynamic spectrum of the main frequency of the noise, the matching pursuit algorithm is used to extract the parameters of each main frequency component, and the parameters of the main frequency component include frequency, amplitude and phase; According to the extracted parameters of each main frequency component, a control signal having the same amplitude, the same frequency and the opposite phase as each main frequency component is synthesized by using frequency modulation and phase inversion algorithm; The synthesized control signal is input into the controller after setting the parameters to perform noise reduction control on the electromechanical equipment.

10. The method for reducing noise of electromechanical equipment based on machine learning according to claim 9, characterized in that: Build a deep learning-based denoising model, including: Collect noise signals of electromechanical equipment under different working conditions, perform time-frequency analysis and sound source location on the collected noise signals, and obtain the noise distribution map corresponding to the noise signals; Collecting a noise-free signal corresponding to the noise signal, and performing time-frequency analysis and sound source localization on the noise-free signal to obtain a noise-free distribution map corresponding to the noise-free signal; A denoising model based on an encoder-decoder structure is constructed, in which the encoder extracts multi-scale features of the noise distribution map through a convolutional neural network (CNN), and the decoder reconstructs the extracted multi-scale features into a noise-free distribution map through a deconvolutional neural network and skip connections. The loss function of the denoising model is constructed by using a weighted combination of reconstruction loss and adversarial loss. The noise distribution map is used as the input sample of the training data, and the corresponding noise-free distribution map is used as the output sample of the training data. The constructed denoising model is trained using the training data, the loss function is calculated through forward propagation, and the model parameters are updated through back propagation; The noise distribution map corresponding to the newly collected noise signal is used as input, and the trained noise reduction model is used to extract the features of the noise distribution map through the encoder, and the noise-free distribution map is reconstructed by the decoder to obtain the noise distribution map after noise reduction; The denoised noise distribution map is converted into a denoised noise signal through inverse Fourier transform.

Images