A power equipment sound recognition method and system
By fusing MFCC and GFCC features and performing dimensionality reduction, combined with SVM support vector machine, the problem of accuracy in sound recognition of power equipment in high-noise environments was solved, enabling accurate identification and early warning of power equipment faults.
Patent Information
- Application Number
- CN202211344134.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-31
- Publication Date
- 2025-12-05
- Estimated Expiration
- 2042-10-31
AI Technical Summary
In high-noise environments, existing technologies struggle to accurately identify the sounds of power equipment, leading to a decrease in the accuracy of power equipment fault identification.
We employ MFCC and GFCC feature extraction methods, combined with principal component analysis for dimensionality reduction, and use SVM (Support Vector Machine) for classification and recognition, fusing features to improve recognition accuracy.
Accurately identifying the operating sounds of power equipment in high-noise environments provides a theoretical basis for fault early warning and identification, and improves the accuracy of identification.
Smart Images
Figure CN115691508B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of power equipment sound recognition, in particular to a power equipment sound recognition method and system. BACKGROUND
[0002] With the increasing scale of China's power system, the installed capacity and scale of the power grid are continuously improved. On the one hand, it brings considerable economic benefits to production, and on the other hand, the possibility of failure of power equipment and the loss caused thereby also increase. Comprehensive, timely and accurate perception of the state of power equipment is a prerequisite for ensuring the safety of the equipment. Therefore, a large number of researchers have carried out systematic research on various power equipment from the aspects of images, temperature, time sequence, vibration signals and the like of the equipment, aiming to extract effective features that can timely reflect the running state of the equipment, and to transfer human experience to computers by combining pattern recognition and other machine learning methods.
[0003] However, in the field of speech recognition with a large number of research results, mel frequency cepstral coefficients (MFCC) and gamma tone frequency cepstral coefficients (GFCC) are widely used as frequency domain features. MFCC features have a high recognition rate, and GFCC has better robustness. For different research objects, the two features show different recognition effects. In addition, the dimension of the two features is determined by the number of filters. When the dimension is large, there are redundant features, which interfere with the final recognition result. Moreover, in the actual environment, the sound of power equipment is greatly affected by environmental noise, and effective features are difficult to extract, resulting in a decrease in the accuracy of sound recognition.
[0004] In view of the problem that the sound of actual power equipment is disturbed by complex environmental sound, it is urgent to propose a feature fusion device sound recognition method or system to provide a theoretical basis for power equipment fault warning and fault recognition. SUMMARY
[0005] The purpose of the present application is to provide a power equipment sound recognition method and system, which can accurately recognize the running sound of different types of power equipment in a high-noise environment of a power plant.
[0006] To achieve the above-mentioned purpose, the present application provides the following solutions:
[0007] A power equipment sound recognition method, comprising:
[0008] acquiring sound signals of different types of power equipment in a running state; the different types of power equipment include an oil supply pump, a connecting shaft, a coal mill, a coal mill box, a condensate pump and an induced draft fan;
[0009] preprocessing the sound signal;
[0010] extracting MFCC features and GFCC features based on an audio signal feature extraction method according to the preprocessed sound signal;
[0011] determining reduced-dimension MFCC features and reduced-dimension GFCC features based on a principal component analysis method according to the MFCC features and the GFCC features; and combining the reduced-dimension MFCC features and the reduced-dimension GFCC features to determine fusion features;
[0012] classifying and recognizing the fusion features by using a trained SVM support vector machine.
[0013] Optionally, the preprocessing of the sound signal specifically includes:
[0014] segmenting the sound signal;
[0015] pre-emphasizing the segmented sound signal by using a high-pass filter;
[0016] normalizing the pre-emphasized sound signal;
[0017] segmenting the normalized sound signal by using an overlapping segmentation method;
[0018] windowing the segmented sound signal by using a Hamming window;
[0019] determining the preprocessed sound signal by performing fast Fourier transform on the windowed sound signal.
[0020] Optionally, the extracting of the MFCC features and the GFCC features based on the audio signal feature extraction method according to the preprocessed sound signal specifically includes:
[0021] filtering the preprocessed sound signal based on a Mel filter to obtain log energy; and performing discrete cosine transform on the log energy to obtain the MFCC features;
[0022] decomposing the preprocessed sound signal based on a gamma filter; and extracting the GFCC features by using the decomposed sound signal.
[0023] Optionally, the determining of the reduced-dimension MFCC features and the reduced-dimension GFCC features based on the principal component analysis method according to the MFCC features and the GFCC features specifically includes:
[0024] centering the MFCC features and the GFCC features;
[0025] determine a covariance matrix according to the centralized MFCC feature and the centralized GFCC feature;
[0026] determine eigenvalues and corresponding eigenvectors of the covariance matrix according to the covariance matrix;
[0027] project the MFCC feature and the GFCC feature onto the corresponding eigenvectors to determine a reduced-dimension MFCC feature and a reduced-dimension GFCC feature.
[0028] Optionally, the reduced-dimension MFCC feature and the reduced-dimension GFCC feature are combined to determine a fusion feature, specifically including:
[0029] determine a contribution rate of the reduced-dimension MFCC feature and the reduced-dimension GFCC feature in different feature dimensions respectively; and select the first four dimensions with an accumulated contribution rate greater than 95% to determine an 8-dimensional fusion feature.
[0030] Optionally, the fusion feature is classified and recognized by using the trained SVM support vector machine, and the method further includes:
[0031] obtain a sound signal data set; the sound signal data set includes sound signals of different types of power equipment;
[0032] add Gaussian white noise with different signal-to-noise ratios to the sound signals in the sound signal data set to obtain an updated sound signal data set;
[0033] train the SVM support vector machine according to the fusion feature corresponding to each sound signal in the updated sound signal data set to determine a trained SVM support vector machine.
[0034] Optionally, the trained SVM support vector machine uses an RBF Gaussian radial basis kernel function.
[0035] A power equipment sound recognition system includes:
[0036] a sound signal acquisition module configured to acquire sound signals of different types of power equipment in a running state; the different types of power equipment include an oil supply pump, a connecting shaft, a coal mill, a coal mill box, a condensate water pump, and an induced draft fan;
[0037] a sound signal preprocessing module configured to preprocess the sound signals;
[0038] a feature extraction module configured to extract MFCC features and GFCC features based on an audio signal feature extraction method according to the preprocessed sound signals;
[0039] The feature dimension reduction module is configured to determine reduced dimension MFCC features and reduced dimension GFCC features based on a principal component analysis method according to the MFCC features and the GFCC features, and combine the reduced dimension MFCC features and the reduced dimension GFCC features to determine the fusion features.
[0040] The classification and recognition module is configured to classify and recognize the fusion features by using a trained SVM support vector machine.
[0041] A power equipment sound recognition system comprises at least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the power equipment sound recognition method.
[0042] According to the specific embodiments of the present application, the following technical effects are achieved.
[0043] The power equipment sound recognition method and system provided by the present application combine MFCC features and GFCC features in sound signals of different types of power equipment in a running state to determine fusion features, and classify and recognize the fusion features by using a trained SVM support vector machine. The present application can accurately recognize the running sound of different types of power equipment in a high-noise environment of a power plant, which solves the problem of sound interference of actual power equipment by complex environments and the like. Furthermore, the present application provides a theoretical basis for power equipment fault early warning and fault recognition. BRIEF DESCRIPTION OF DRAWINGS
[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.
[0045] Figure 1 A flowchart of the power equipment sound recognition method provided by the present application is shown in the figure.
[0046] Figure 2 A software flowchart corresponding to the present application is shown in the figure.
[0047] Figure 3 A whole flowchart of the present application is shown in the figure. DETAILED DESCRIPTION
[0048] With reference to the accompanying drawings: clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of the present application.
[0049] The purpose of the present application is to provide a power equipment sound recognition method and system, which can accurately identify the running sound of different types of power equipment in a high-noise environment of a power plant.
[0050] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.
[0051] Figure 1 A flowchart of a power equipment sound recognition method provided by the present application is shown in Figure 2 A software flowchart corresponding to the present application is shown in Figure 1 and Figure 2 The power equipment sound recognition method provided by the present application comprises:
[0052] S101, acquiring sound signals of different types of power equipment in a running state; the different types of power equipment include: oil supply pump, connecting shaft, coal mill, coal mill box, condensate pump and induced draft fan.
[0053] As a specific embodiment, in a thermal power plant in a running state, 96-channel sound image instruments are used to collect sound signals of six types of equipment on site, including oil supply pump, connecting shaft, coal mill, coal mill box, condensate pump and induced draft fan.
[0054] S102, pre-processing the sound signals.
[0055] As shown in Figure 3 S102 specifically comprises:
[0056] The sound signals are segmented.
[0057] The segmented sound signals are pre-emphasized using a high-pass filter.
[0058] The pre-emphasized sound signals are normalized.
[0059] The normalized sound signals are segmented in units of 160ms, with a frame length of 32ms and a frame shift of 16ms.
[0060] The segmented sound signals are windowed using a Hamming window.
[0061] Performing fast Fourier transform on the windowed sound signal to determine the preprocessed sound signal.
[0062] S103, extracting MFCC features and GFCC features based on the audio signal feature extraction method according to the preprocessed sound signal.
[0063] S103 specifically includes:
[0064] Filtering the preprocessed sound signal based on a Mel filter to obtain a log energy; and performing discrete cosine transform on the log energy to obtain the MFCC features. The Mel filter bank is a kind of bionic filter constructed according to the characteristics of human ear hearing mechanism. The Mel filter bank has the characteristics of low frequency density and high frequency sparseness. The approximate expression of Mel frequency and linear frequency is as follows:
[0065]
[0066] Wherein, f is the linear frequency, unit: Hz.
[0067] Calculating the discrete power spectrum, filtering through a set of Mel filters to obtain a log energy; performing discrete cosine transform (DCT) on the log energy calculated in the previous step to obtain static MFCC features:
[0068]
[0069] Wherein, M is the number of filters, N represents the number of Fourier transform points, n is the order of MFCC features, L is the maximum order of MFCC features, S(m) represents the log energy output by each filter group, m = 1, 2,..., L.
[0070] Based on the preprocessed sound signal, the audio signal is decomposed based on a Gammatone filter; and the GFCC features are extracted using the decomposed sound signal. The Gammatone filter is a set of filter models used to simulate the frequency decomposition characteristics of the cochlea. It can be used for audio signal decomposition, which is convenient for subsequent feature extraction. It is a standard cochlear auditory filter based on the mechanism of human ear hearing. Its time domain expression is as follows:
[0071] h i (t)=ct q-1 exp(-2πb i t)cos(2πf ci t+φ i )U(t), t≥0, 1≤i≤M;
[0072] Wherein, c represents the gain of the filter; q is the filter order; i is the number of the filter; M is the maximum number of the filter; fc is the center frequency of the filter, φ is the offset phase; U(t) represents the step function; b is the decay factor which determines the decay speed of the current filter to the impulse response, the greater b is, the shorter the filtering time is, and it is related to the equivalent rectangular bandwidth ERB(fci) as follows:
[0073] b i =1.019ERB(f ci );
[0074] The equivalent rectangular bandwidth is:
[0075]
[0076] The center frequencies of the filters are equally spaced in the ERB domain.
[0077] In S104, the MFCC features and the GFCC features are determined based on a principal component analysis method, and the MFCC features and the GFCC features are combined to determine fusion features.
[0078] In S104, the MFCC features and the GFCC features are centralized, i.e., de-meaned.
[0079]
[0080] A covariance matrix is determined based on the centralized MFCC features and the centralized GFCC features.
[0081] The variances of the features x1 and x2 are respectively on the diagonals, and the covariances are on the non-diagonals. If the covariance is greater than 0, one of x1 or x2 increases, and the other also increases; if the covariance is less than 0, one increases and the other decreases; if the covariance is 0, the two are independent. The greater the absolute value of the covariance is, the greater the influence of one on the other is, and vice versa.
[0082]
[0083] According to the covariance calculation formula, the covariance matrix C of the M samples in the N-dimensional features is obtained.
[0084] Eigenvalues and corresponding eigenvectors of the covariance matrix are determined based on the covariance matrix.
[0085] Eigenvalues and corresponding eigenvectors of the covariance matrix C are determined based on the knowledge of matrices.
[0086] Cu=λu;
[0087] The eigenvalues are N, each corresponding to an eigenvector, the eigenvalues are sorted in descending order, the first k largest eigenvalues are selected, and the corresponding k eigenvectors are taken out to obtain a new set of eigenvectors.
[0088] The MFCC features and the GFCC features are projected onto corresponding eigenvectors to determine the reduced MFCC features and the reduced GFCC features.
[0089] The reduced feature calculation formula is as follows:
[0090]
[0091] S104, specifically comprising:
[0092] The contribution rates of the reduced MFCC features and the reduced GFCC features in different feature dimensions are determined respectively; and the first four-dimensional features with a cumulative contribution rate greater than 95% are selected for fusion to determine 8-dimensional fusion features.
[0093] S105, using the trained SVM support vector machine to classify and identify the fusion features.
[0094] S105 further comprises:
[0095] A sound signal data set is obtained; the sound signal data set comprises sound signals of different types of power equipment.
[0096] Gaussian white noise with different signal-to-noise ratios is added to the sound signals in the sound signal data set to obtain an updated sound signal data set.
[0097] The SVM support vector machine is trained according to the fusion features corresponding to each sound signal in the updated sound signal data set to determine the trained SVM support vector machine.
[0098] As a specific embodiment, to further quantitatively evaluate the anti-noise effect of the proposed fusion features, 20 dB and 10 dB Gaussian white noise is added to the original sound signal data set, and the accuracy and robustness of the fusion features are verified based on the SVM algorithm. SVM is a small sample classification algorithm, which can be mapped to a high-dimensional space using a kernel function, and can achieve effective classification under the condition of a small number of training samples. We selected the RBF Gaussian radial basis kernel function, and when the gamma value was set to 100 and the penalty factor c was set to 25, the fitting effect was best.
[0099] For the above method, the application provides a power equipment sound recognition system to further illustrate the beneficial effects of the application. The power equipment sound recognition system comprises:
[0100] The sound signal acquisition module is configured to acquire sound signals of different types of power equipment in a running state, wherein the different types of power equipment include an oil supply pump, a connecting shaft, a coal mill, a coal mill box, a condensate water pump and an induced draft fan.
[0101] The sound signal preprocessing module is configured to preprocess the sound signals.
[0102] The feature extraction module is configured to extract MFCC features and GFCC features based on an audio signal feature extraction method according to the preprocessed sound signals.
[0103] The feature dimension reduction module is configured to determine reduced-dimension MFCC features and reduced-dimension GFCC features based on a principal component analysis method according to the MFCC features and the GFCC features, and combine the reduced-dimension MFCC features and the reduced-dimension GFCC features to determine fusion features.
[0104] The classification and recognition module is configured to classify and recognize the fusion features by using a trained SVM support vector machine.
[0105] In order to implement the method corresponding to the above-mentioned embodiment one and achieve the corresponding functions and technical effects, the present application provides a power equipment sound recognition system, comprising at least one processor, at least one memory and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, the method is implemented.
[0106] In the specification, each embodiment is described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between each embodiment can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.
[0107] The principles and implementation modes of the present application are described by using specific examples in this paper, and the above embodiment description is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In conclusion, the content of the specification should not be understood as a limitation of the present application.
Claims
1. A power equipment sound recognition method, characterized by, The method comprises the following steps: acquiring sound signals of different types of power equipment in a running state; the different types of power equipment include an oil supply pump, a connecting shaft, a coal mill, a coal mill box, a condensate water pump, and an induced draft fan; preprocessing the sound signals; extracting MFCC features and GFCC features based on an audio signal feature extraction method according to the preprocessed sound signals; determining reduced-dimension MFCC features and reduced-dimension GFCC features based on a principal component analysis method according to the MFCC features and the GFCC features; combining the reduced-dimension MFCC features and the reduced-dimension GFCC features to determine fusion features; determining the contribution rates of the reduced-dimension MFCC features and the reduced-dimension GFCC features in different feature dimensions respectively; and selecting the first four features with cumulative contribution rates greater than 95% for fusion to determine 8-dimensional fusion features; classifying and identifying the fusion features by using a trained SVM support vector machine.
2. The power equipment sound recognition method of claim 1, wherein The preprocessing of the sound signals specifically comprises the following steps: segmenting the sound signals; pre-emphasizing the segmented sound signals by using a high-pass filter; normalizing the pre-emphasized sound signals; frame-segmenting the normalized sound signals by using an overlapping segmentation method; windowing the frame-segmented sound signals by using a Hamming window; determining the preprocessed sound signals by performing fast Fourier transform on the windowed sound signals.
3. The power equipment sound recognition method of claim 1, wherein The extraction of the MFCC features and the GFCC features based on the audio signal feature extraction method according to the preprocessed sound signals specifically comprises the following steps: filtering the preprocessed sound signals based on a Mel filter to obtain log energy; and performing discrete cosine transform on the log energy to obtain MFCC features; performing audio signal decomposition on the preprocessed sound signals based on a gamma filter; and extracting GFCC features by using the decomposed sound signals.
4. The power equipment sound recognition method of claim 1, wherein The determination of the reduced-dimension MFCC features and the reduced-dimension GFCC features based on the principal component analysis method according to the MFCC features and the GFCC features specifically comprises the following steps: centralizing the MFCC features and the GFCC features; determining a covariance matrix according to the centralized MFCC features and the centralized GFCC features; determining eigenvalues of the covariance matrix and corresponding eigenvectors according to the covariance matrix; projecting the MFCC features and the GFCC features onto the corresponding eigenvectors to determine the reduced-dimension MFCC features and the reduced-dimension GFCC features.
5. The power equipment sound recognition method of claim 1, wherein The classification and identification of the fusion features by using the trained SVM support vector machine further comprises the following steps: acquiring a sound signal dataset; the sound signal dataset includes sound signals of different types of power equipment; adding Gaussian white noise with different signal-to-noise ratios to the sound signals in the sound signal dataset to obtain an updated sound signal dataset; training the SVM support vector machine according to the fusion features corresponding to each sound signal in the updated sound signal dataset to determine the trained SVM support vector machine.
6. The power equipment sound recognition method of claim 1, wherein The trained SVM support vector machine adopts an RBF Gaussian radial basis kernel function.
7. A power equipment sound recognition system for implementing the power equipment sound recognition method according to any one of claims 1-6; characterized by, The power equipment sound recognition system comprises: A sound signal acquisition module is configured to acquire sound signals of different types of power equipment in a running state, wherein the different types of power equipment include an oil supply pump, a connecting shaft, a coal mill, a coal mill box, a condensate water pump, and an induced draft fan. A sound signal preprocessing module is configured to preprocess the sound signals. A feature extraction module is configured to extract MFCC features and GFCC features based on an audio signal feature extraction method according to the preprocessed sound signals. A feature dimension reduction module is configured to determine reduced MFCC features and reduced GFCC features based on a principal component analysis method according to the MFCC features and the GFCC features, and combine the reduced MFCC features and the reduced GFCC features to determine fusion features. A classification and recognition module is configured to classify and recognize the fusion features by using a trained SVM support vector machine.
8. A power equipment sound recognition system, characterized by, The method comprises: at least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method of any one of claims 1-6.