Self-adaptive voiceprint denoising method and system for main equipment of power grid
By performing feature fusion and weight optimization of the voiceprint signals of the main equipment of the power grid, and combining convolution operations, adaptive noise removal is achieved, and the diagnostic misjudgment and misjudgment problems under the influence of noise in the prior art are solved, and the robustness of the model and the characterization ability of the voiceprint characteristics are improved.
Patent Information
- Application Number
- CN202411647077.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-02-25
AI Technical Summary
The voiceprint signals of the main equipment of the power grid are easily affected by noise during the acquisition process, resulting in diagnostic misjudgment and misjudgment. The existing technology denoising methods have poor adaptability and high computational complexity, and the effect depends on the threshold selected by experience.
Adaptive denoising method for soundprints of main equipment of the power grid is adopted. By acquiring the energy spectrum characteristics and phase spectrum characteristics of the soundprint signals of the main equipment of the power grid, feature fusion and weight optimization are performed, combined with convolution operations, the noise type is automatically detected and the noise is removed targeted to achieve clean soundprint signals acquisition.
It realizes accurate noise denoising of the voiceprint signal of the main equipment of the power grid, enhances the robustness of the model and the characterization ability of the voiceprint characteristics, and helps to realize the precise classification of the noise-containing types of the voiceprint signal of the main equipment of the power grid.
Smart Images
Figure CN119517065A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of voiceprint processing of power grid main equipment, and in particular relates to a method and system for adaptively denoising voiceprints of power grid main equipment. Background Art
[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.
[0003] The health status of main grid equipment such as transformers directly affects the safe and stable operation of substations. Its health monitoring methods include sensors, visual recognition, voiceprint diagnosis and other methods. Voiceprint diagnosis technology for main grid equipment is a technology that uses the voiceprint information generated by power equipment in different operating states (especially when it fails) to diagnose the health status of equipment. The vibration generated by power equipment during operation is closely related to its mechanical structure. These voiceprint signals contain a large amount of status information and have identification characteristics. When the state of parts or components of the equipment changes due to wear, aging, etc., the characteristics of its voiceprint signal will also change accordingly. By installing contact or non-contact sensors, the sounds generated inside the equipment are collected in real time and the features of these sounds are extracted and identified, so as to determine whether the equipment has faults, vibrations, leakages, electrical partial discharges, etc.
[0004] The voiceprint collection process of the main power grid equipment is often affected by noise. The noise comes from the surrounding equipment environment (wind, rain, traffic noise, human voice, etc.), electromagnetic interference (related to the electrical characteristics of the equipment and the electromagnetic field distribution in the surrounding environment), and the inside of the equipment (related to factors such as the manufacturing process, material selection and circuit design of the equipment). Voiceprint signals affected by noise can lead to misjudgment and missed judgment when diagnosing power equipment. Therefore, denoising the noisy voiceprint signals and enhancing the equipment voiceprint characteristics are important links in the voiceprint diagnosis of the main power grid equipment.
[0005] The filter-based voiceprint denoising technology designs corresponding filters (such as low-pass, high-pass, and band-pass filters) according to the frequency characteristics of specific noise to remove noise within a specific frequency range. However, the performance of the filter depends largely on its design parameters and has poor adaptability.
[0006] Voiceprint denoising technology based on wavelet transform can analyze signals in both time domain and frequency domain, has multi-resolution characteristics, and can better handle non-stationary signals. However, the computational complexity of wavelet transform is high, and the computational resources required are high. The effect of wavelet denoising depends largely on the choice of threshold, and the determination of threshold often needs to be determined based on experience or trial and error, and lacks a unified standard. Summary of the invention
[0007] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a method and system for adaptive denoising of the voiceprint of the main power grid equipment, which can not only autonomously detect the noise type of the voiceprint signal of the main power grid equipment, and remove the noise in a targeted and adaptive manner to achieve the acquisition of the voiceprint signal of the main power grid equipment; but also through two weight optimization and convolution operations, the energy spectrum features and the phase spectrum features are weighted and weighted. Compared with the simple weighted fusion method, the obtained fusion feature integrates the rich information of the phase spectrum and the amplitude spectrum, which can better characterize the voiceprint characteristics, and help to achieve accurate classification of the noise type of the voiceprint signal of the main power grid equipment.
[0008] In order to achieve the above object, the present invention adopts the following technical solution:
[0009] A first aspect of the present invention provides a method for adaptively denoising a voiceprint of a main power grid device, comprising:
[0010] Acquire the voiceprint signal of the main power grid equipment, extract the energy spectrum features and the phase spectrum features through feature extraction, fuse the energy spectrum features and the phase spectrum features to obtain the fusion features, classify based on the fusion features, and obtain the noise type of the voiceprint signal of the main power grid equipment;
[0011] Based on the noise type, a denoising method is selected to denoise the voiceprint signal of the main equipment of the power grid;
[0012] Among them, the feature fusion step includes: splicing the energy spectrum feature and the phase spectrum feature to obtain the first splicing feature, and then optimizing the weight to obtain the first weight and the second weight; performing a convolution operation on the first weight and the energy spectrum feature to obtain the first weighted energy spectrum feature; performing a convolution operation on the second weight and the phase spectrum feature to obtain the first weighted phase spectrum feature; after splicing the first weighted energy spectrum feature and the first weighted phase spectrum feature to obtain the initial fusion feature, optimizing the weight to obtain the third weight and the fourth weight; performing a convolution operation on the third weight and the energy spectrum feature to obtain the second weighted energy spectrum feature; performing a convolution operation on the fourth weight and the phase spectrum feature to obtain the second weighted phase spectrum feature; the second weighted energy spectrum feature and the second weighted phase spectrum feature are spliced to obtain the fusion feature.
[0013] Furthermore, the step of feature optimization includes:
[0014] For the input features, global pooling, point-by-point convolution, normalization, activation function, point-by-point convolution and normalization are performed in sequence to obtain the first result;
[0015] For the input features, point-by-point convolution, batch normalization, activation function, point-by-point convolution and normalization are performed in sequence to obtain the second result;
[0016] After the first result and the second result are concatenated, two weights are obtained through the Sigmoid function.
[0017] Furthermore, the classification is achieved by sequentially connecting multiple convolutional layers, maximum pooling layers and multiple fully connected layers.
[0018] Furthermore, the step of extracting the energy spectrum feature and the phase spectrum feature includes:
[0019] Window and frame the voiceprint signal of the main equipment of the power grid to obtain a short-term stable signal;
[0020] For short-time stationary signals, energy spectrum characteristics and phase spectrum characteristics are obtained through short-time Fourier transform.
[0021] Furthermore, if the noise type is continuous interference noise, whether the fan is running is determined according to the sound pressure level or the energy of the specified frequency band, and a real-time spectrum subtraction operation is performed to remove the continuous interference noise.
[0022] Furthermore, if the noise type is instantaneous interference noise, the steps of the denoising method include:
[0023] (1) For the voiceprint signals of the main power grid equipment, perform background voiceprint assessment;
[0024] (2) Based on the background voiceprint estimation results, the signal-to-noise ratio of each frame of data is calculated, and the noise period is determined based on the signal-to-noise ratio;
[0025] (3) Based on the amplitude spectrum information of the background voiceprint signal and the phase spectrum information of the signal during the noise period, the voiceprint signal of the main power grid equipment with the instantaneous interference noise removed is calculated through inverse fast Fourier transform.
[0026] Furthermore, the noise type includes no interference, instantaneous interference and continuous interference.
[0027] A second aspect of the present invention provides a power grid master device voiceprint adaptive denoising system, comprising:
[0028] A classification module is configured to: obtain the voiceprint signal of the main power grid device, obtain energy spectrum features and phase spectrum features through feature extraction, fuse the energy spectrum features and phase spectrum features to obtain fusion features, and classify based on the fusion features to obtain the noise type of the voiceprint signal of the main power grid device;
[0029] A denoising module is configured to: select a denoising method based on the noise type to denoise the voiceprint signal of the main equipment of the power grid;
[0030] Among them, the feature fusion step includes: splicing the energy spectrum feature and the phase spectrum feature to obtain the first splicing feature, and then optimizing the weight to obtain the first weight and the second weight; performing a convolution operation on the first weight and the energy spectrum feature to obtain the first weighted energy spectrum feature; performing a convolution operation on the second weight and the phase spectrum feature to obtain the first weighted phase spectrum feature; after splicing the first weighted energy spectrum feature and the first weighted phase spectrum feature to obtain the initial fusion feature, optimizing the weight to obtain the third weight and the fourth weight; performing a convolution operation on the third weight and the energy spectrum feature to obtain the second weighted energy spectrum feature; performing a convolution operation on the fourth weight and the phase spectrum feature to obtain the second weighted phase spectrum feature; the second weighted energy spectrum feature and the second weighted phase spectrum feature are spliced to obtain the fusion feature.
[0031] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for adaptively denoising voiceprints of a main equipment of a power grid as described above.
[0032] The fourth aspect of the present invention provides a computer device, including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, wherein when the processor executes the program, the steps in the method for adaptive denoising of voiceprint of a main power grid device as described above are implemented.
[0033] Compared with the prior art, the present invention has the following beneficial effects:
[0034] The present invention obtains and weights the energy spectrum features and the phase spectrum features through two weight optimization and convolution operations. Compared with the simple weighted fusion method, the obtained fusion features integrate the rich information of the phase spectrum and the amplitude spectrum, which can better characterize the voiceprint characteristics. This helps the model to understand the voiceprint data of the main power grid equipment more accurately, while enhancing the robustness of the model, and helping the model to accurately classify the noise type of the voiceprint signal of the main power grid equipment.
[0035] The present invention can autonomously detect the noise type of the voiceprint signal of the main power grid equipment, and remove the noise in a targeted and adaptive manner, thereby achieving the acquisition of the clean voiceprint signal of the main power grid equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0037] Figure 1 This is a flow chart of a method for adaptively denoising a voiceprint of a power grid master device according to a first embodiment of the present invention;
[0038] Figure 2 is a noise classification flow chart of the first embodiment of the present invention;
[0039] Figure 3 This is a feature fusion flow chart of Embodiment 1 of the present invention;
[0040] Figure 4 is a flow chart of instantaneous interference noise denoising according to the first embodiment of the present invention;
[0041] Figure 5 It is a structural diagram of a computer device according to a fourth embodiment of the present invention. DETAILED DESCRIPTION
[0042] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0043] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.
[0044] Embodiment 1
[0045] This embodiment provides a method for adaptively denoising voiceprints of a power grid master device.
[0046] This embodiment provides a method for adaptively denoising the voiceprint of a main power grid device. Figure 1 As shown, the noisy voiceprint signal of the main power grid equipment is classified into no interference, instantaneous interference, and continuous interference through the noise classification model of the main power grid equipment. The noise of the instantaneous interference and continuous interference is adaptively removed, and finally a clean voiceprint signal of the main power grid equipment is obtained.
[0047] This embodiment provides a method for adaptively denoising a voiceprint of a power grid master device, comprising the following steps:
[0048] Step 1: Obtain the voiceprint signal of the main power grid equipment, obtain a short-time stable signal by windowing and framing the voiceprint signal, and obtain the energy spectrum characteristics and phase spectrum characteristics of the spectrum through short-time Fourier transform.
[0049] Step 2: Based on the energy spectrum characteristics and phase spectrum characteristics, the noise type of the power grid main equipment voiceprint signal is obtained through the power grid main equipment noise classification model.
[0050] like Figure 2As shown in the figure, the noise classification model of the main equipment of the power grid consists of a feature fusion module, 6 convolutional layers, a maximum pooling layer, 4 fully connected layers and a Softmax function. Random inactivation is performed between the first convolutional layer, the second convolutional layer, and the third convolutional layer to reduce overfitting and improve the generalization ability of the model. The LeakyReLU activation function is set after each convolutional layer. A normalization layer and a maximum pooling layer are set after the first and second convolutional layers.
[0051] Among them, the Softmax function will classify the noise of the main equipment of the power grid and give the probability value of the category.
[0052] Among them, normalization can significantly accelerate the convergence speed of the model. Since all features are scaled to the same scale, the parameter space can be traversed more efficiently when searching for the optimal solution, reducing unnecessary iterations and thus speeding up the training process. At the same time, the relationship between features can be captured more accurately, thereby improving prediction accuracy and generalization ability.
[0053] Among them, LeakyReLU activation function:
[0054]
[0055] α is the leakage coefficient, which usually takes a value between 0 and 1.
[0056] Among them, the activation function LeakyReLU can solve the neuron death problem of ReLU and alleviate the gradient vanishing problem, which can improve the training effect of the neural network. The neural network using LeakyReLU as the activation function is more stable during the training process, can converge to the optimal solution faster, and improve the performance of the noise classification model of the main equipment of the power grid.
[0057] Among them, maximum pooling: Maximum pooling reduces the spatial dimension (i.e. width and height) of the feature map by taking the maximum value in the local area without changing the number of channels. Mathematically, it means taking out the maximum number in the input array each time the pooling kernel is matched to store the input data features. It has strong ability to retain significant features, is robust to small translations or deformations of the input feature map, and reduces computational complexity and memory consumption.
[0058] like Figure 3As shown, in the feature fusion module: the energy spectrum feature and the phase spectrum feature are first spliced to obtain the first spliced feature; after the first spliced feature passes through the weight optimization module, the first weight (weight 1) and the second weight (weight 2) are obtained; the first weight and the energy spectrum feature are jointly input into the first convolution (convolution 1), and the convolution operation is performed to obtain the first weighted energy spectrum feature; the second weight and the phase spectrum feature are jointly input into the second convolution (convolution 2), and the convolution operation is performed to obtain the first weighted phase spectrum feature; the first weighted energy spectrum feature and the first weighted phase spectrum feature are spliced to obtain the initial fusion feature; after the initial fusion feature passes through the weight optimization module, the third weight (weight 3) and the fourth weight (weight 4) are obtained; the third weight and the energy spectrum feature are jointly input into the third convolution (convolution 3), and the convolution operation is performed to obtain the second weighted energy spectrum feature; the fourth weight and the phase spectrum feature are jointly input into the fourth convolution (convolution 4), and the convolution operation is performed to obtain the second weighted phase spectrum feature; the second weighted energy spectrum feature and the second weighted phase spectrum feature are spliced to obtain the final fusion feature.
[0059] The weight optimization module includes two branches, a splicing function and a Sigmoid function. The first branch includes global pooling, point-by-point convolution, normalization, LeakyReLU activation function, point-by-point convolution and normalization connected in sequence; the second branch includes point-by-point convolution, BN layer, LeakyReLU activation function, point-by-point convolution and normalization connected in sequence; both branches are connected to the splicing function and the Sigmoid function.
[0060] Compared with fusion methods such as simple addition, the rich information of the characteristic comprehensive phase spectrum and amplitude spectrum obtained by this fusion method can better characterize the voiceprint characteristics, which helps the model to understand the voiceprint data of the main power grid equipment more accurately, while enhancing the robustness of the model and helping the model to accurately classify the noise type of the voiceprint signal of the main power grid equipment.
[0061] The training process of the grid main equipment noise classification model includes:
[0062] (1) Data preparation. The voiceprint sample library of the main power grid equipment is constructed. The data comes from the voiceprint signal data of the equipment in the production environment and the voiceprint signal of the equipment without interference. The noisy voiceprint signal is obtained by superimposing different types of noise such as environmental noise.
[0063] The original long voiceprint of the main power grid equipment is sliced and labeled with three labels: no interference, instantaneous interference, and continuous interference. The labeling example is as follows: "XXX.wav": "no interference".
[0064] (2) By windowing and framing the voiceprint signal, a short-time stationary signal is obtained, and the energy spectrum feature and phase spectrum feature are obtained through short-time Fourier transform. The energy spectrum feature and phase spectrum feature are used as the input features of the model training network, and finally the classification of the voiceprint noise type (no interference, instantaneous interference, and continuous interference) is realized. Based on the data of the voiceprint sample library of the main equipment of the power grid, the network architecture is continuously trained until the model classification is optimal.
[0065] Step 3: Voiceprint adaptive denoising.
[0066] The voiceprint signals of the main power grid equipment are divided into non-interference, instantaneous interference, and continuous interference noise by the main power grid equipment noise classification model, and denoising is performed using a targeted denoising method.
[0067] Regarding continuous interference noise, fan noise is one of the typical signals for transformer soundprint signals. Since the soundprint signals generated by transformer operation and fan operation are both stable noise, it is possible to collect the soundprint of a single fan during operation, determine whether the fan is running based on the sound pressure level or the energy of the specified frequency band (fan rotation frequency band), and perform real-time spectral subtraction operations to remove continuous interference noise.
[0068] For instantaneous interference noise, such as Figure 4 As shown in the figure, voiceprint signal denoising is achieved through the following four steps:
[0069] (1) Background voiceprint assessment. It is not exactly the same as the speech signal spectral subtraction method. The speech signal spectral subtraction uses the noise statistical characteristics of the speech intermittent segment to replace the noise statistical characteristics of the superimposed noise period, and then performs spectral subtraction to obtain the noise-suppressed speech signal. According to the stable characteristics of the transformer voiceprint, by finding the signal of the non-noise interference period as the target signal, and then performing spectral replacement on the signal of the noise interference period, retaining the original phase information of the signal, and replacing it with the amplitude information of the frame with the minimum energy (the weakest interference frame) in the period, the noise suppression of any instantaneous interference is achieved. Specifically, according to the stable characteristics of the transformer voiceprint, by finding the signal of the non-noise interference period as the target signal, it can be regarded as the background voiceprint signal.
[0070] (2) Noise period estimation. Based on the background voiceprint estimation results in (1), the signal-to-noise ratio of each frame of data is calculated, and the sequence number of the frame to be denoised is determined according to the set noise consistency threshold. That is, the frame with a signal-to-noise ratio greater than the threshold value is determined as a noise period;
[0071] (3) Spectral substitution calculation. This process is similar to conventional spectral subtraction. The data is fused through the amplitude spectrum information of the background voiceprint signal obtained in (1) and the phase spectrum information of the noise segment signal in (2).
[0072] Among them, the steps of data fusion processing include: the voiceprint containing instantaneous noise (i.e., the voiceprint during the noise period) is transformed through fast Fourier transform to obtain the amplitude spectrum A1 and the phase spectrum P1; compared with the amplitude spectrum, the phase spectrum contains more voiceprint information that can be retained, so the amplitude replacement method is used to remove the instantaneous noise; the signal voiceprint in the adjacent non-noise interference period is transformed through fast Fourier transform to obtain the amplitude spectrum A2 and the phase spectrum P2.
[0073] Among them, the time of instantaneous noise is very short, and the signal voiceprint of the adjacent non-noise interference period of the same time length can be regarded as the background voiceprint signal.
[0074] (4) Data restoration: After replacing the spectrum in (3), the full spectrum information is fused and inverse fast Fourier transform is performed to calculate the de-noised time domain signal.
[0075] The phase spectrum P1 of the voiceprint containing instantaneous noise and the amplitude spectrum A2 of the signal voiceprint in the adjacent non-noise interference period are used for inverse fast Fourier transform to obtain the time domain signal of the voiceprint of the main power grid equipment with instantaneous noise removed, thereby achieving fusion.
[0076] The present embodiment provides a method for adaptively denoising the voiceprint of a power grid master device, which can autonomously detect the noise type of the voiceprint signal of the power grid master device, and remove the noise in a targeted and adaptive manner, thereby achieving the acquisition of a clean voiceprint signal of the power grid master device.
[0077] Embodiment 2
[0078] This embodiment provides a system for adaptively removing noise from voiceprints of a power grid master device, which specifically includes:
[0079] A classification module is configured to: obtain the voiceprint signal of the main power grid device, obtain energy spectrum features and phase spectrum features through feature extraction, fuse the energy spectrum features and phase spectrum features to obtain fusion features, and classify based on the fusion features to obtain the noise type of the voiceprint signal of the main power grid device;
[0080] A denoising module is configured to: select a denoising method based on the noise type to denoise the voiceprint signal of the main equipment of the power grid;
[0081] Among them, the feature fusion step includes: splicing the energy spectrum feature and the phase spectrum feature to obtain the first splicing feature, and then optimizing the weight to obtain the first weight and the second weight; performing a convolution operation on the first weight and the energy spectrum feature to obtain the first weighted energy spectrum feature; performing a convolution operation on the second weight and the phase spectrum feature to obtain the first weighted phase spectrum feature; after splicing the first weighted energy spectrum feature and the first weighted phase spectrum feature to obtain the initial fusion feature, optimizing the weight to obtain the third weight and the fourth weight; performing a convolution operation on the third weight and the energy spectrum feature to obtain the second weighted energy spectrum feature; performing a convolution operation on the fourth weight and the phase spectrum feature to obtain the second weighted phase spectrum feature; the second weighted energy spectrum feature and the second weighted phase spectrum feature are spliced to obtain the fusion feature.
[0082] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.
[0083] Embodiment 3
[0084] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method for adaptively denoising a voiceprint of a power grid master device as described in the first embodiment above are implemented.
[0085] Embodiment 4
[0086] This embodiment provides a computer device, such as Figure 5 As shown, it includes a computer-readable storage medium 1003, a processor 1001, a communication interface 1002, and a computer program stored on the computer-readable storage medium 1003 and executable on the processor 1001, wherein the processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 may be connected via a bus or other means. The communication interface 1002 is used to receive and send data, and when the processor 1001 executes the program, the steps in the method for adaptively denoising the voiceprint of a main device of a power grid as described in the first embodiment above are implemented.
[0087] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0088] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0089] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for adaptively denoising the voiceprint of a main power grid device, characterized in that: include: Acquire the voiceprint signal of the main power grid equipment, extract the energy spectrum features and the phase spectrum features through feature extraction, fuse the energy spectrum features and the phase spectrum features to obtain the fusion features, classify based on the fusion features, and obtain the noise type of the voiceprint signal of the main power grid equipment; Based on the noise type, a denoising method is selected to denoise the voiceprint signal of the main equipment of the power grid; Among them, the feature fusion step includes: splicing the energy spectrum feature and the phase spectrum feature to obtain the first splicing feature, and then optimizing the weight to obtain the first weight and the second weight; performing a convolution operation on the first weight and the energy spectrum feature to obtain the first weighted energy spectrum feature; performing a convolution operation on the second weight and the phase spectrum feature to obtain the first weighted phase spectrum feature; after splicing the first weighted energy spectrum feature and the first weighted phase spectrum feature to obtain the initial fusion feature, optimizing the weight to obtain the third weight and the fourth weight; performing a convolution operation on the third weight and the energy spectrum feature to obtain the second weighted energy spectrum feature; performing a convolution operation on the fourth weight and the phase spectrum feature to obtain the second weighted phase spectrum feature; the second weighted energy spectrum feature and the second weighted phase spectrum feature are spliced to obtain the fusion feature.
2. A method for adaptively denoising the voiceprint of a main power grid device according to claim 1, characterized in that: The step of feature optimization comprises: For the input features, global pooling, point-by-point convolution, normalization, activation function, point-by-point convolution and normalization are performed in sequence to obtain the first result; For the input features, point-by-point convolution, batch normalization, activation function, point-by-point convolution and normalization are performed in sequence to obtain the second result; After the first result and the second result are concatenated, two weights are obtained through the Sigmoid function.
3. A method for adaptively denoising the voiceprint of a main power grid device according to claim 1, characterized in that: The classification is achieved by sequentially connecting multiple convolutional layers, maximum pooling layers and multiple fully connected layers.
4. A method for adaptively denoising the voiceprint of a main power grid device according to claim 1, characterized in that: The step of extracting the energy spectrum feature and the phase spectrum feature comprises: Window and frame the voiceprint signal of the main equipment of the power grid to obtain a short-term stable signal; For short-time stationary signals, energy spectrum characteristics and phase spectrum characteristics are obtained through short-time Fourier transform.
5. The method for adaptively denoising the voiceprint of a main power grid device according to claim 1, characterized in that: If the noise type is continuous interference noise, it is determined whether the fan is running according to the sound pressure level or the energy of the specified frequency band, and a real-time spectrum subtraction operation is performed to remove the continuous interference noise.
6. A method for adaptively denoising the voiceprint of a power grid master device according to claim 1, characterized in that: If the noise type is instantaneous interference noise, the steps of the denoising method include: (1) For the voiceprint signals of the main power grid equipment, perform background voiceprint assessment; (2) Based on the background voiceprint estimation results, the signal-to-noise ratio of each frame of data is calculated, and the noise period is determined based on the signal-to-noise ratio; (3) Based on the amplitude spectrum information of the background voiceprint signal and the phase spectrum information of the signal during the noise period, the voiceprint signal of the main power grid equipment with the instantaneous interference noise removed is calculated through inverse fast Fourier transform.
7. A method for adaptively denoising the voiceprint of a main equipment of a power grid according to claim 1, characterized in that: The noise types include no interference, instantaneous interference and continuous interference.
8. A power grid main equipment voiceprint adaptive denoising system, characterized in that: include: A classification module is configured to: obtain the voiceprint signal of the main power grid device, obtain energy spectrum features and phase spectrum features through feature extraction, fuse the energy spectrum features and phase spectrum features to obtain fusion features, and classify based on the fusion features to obtain the noise type of the voiceprint signal of the main power grid device; A denoising module is configured to: select a denoising method based on the noise type to denoise the voiceprint signal of the main equipment of the power grid; Among them, the feature fusion step includes: splicing the energy spectrum feature and the phase spectrum feature to obtain the first splicing feature, and then optimizing the weight to obtain the first weight and the second weight; performing a convolution operation on the first weight and the energy spectrum feature to obtain the first weighted energy spectrum feature; performing a convolution operation on the second weight and the phase spectrum feature to obtain the first weighted phase spectrum feature; after splicing the first weighted energy spectrum feature and the first weighted phase spectrum feature to obtain the initial fusion feature, optimizing the weight to obtain the third weight and the fourth weight; performing a convolution operation on the third weight and the energy spectrum feature to obtain the second weighted energy spectrum feature; performing a convolution operation on the fourth weight and the phase spectrum feature to obtain the second weighted phase spectrum feature; the second weighted energy spectrum feature and the second weighted phase spectrum feature are spliced to obtain the fusion feature.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a method for adaptively denoising voiceprints of a power grid master device as described in any one of claims 1 to 7 are implemented.
10. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored in the computer-readable storage medium and executable on the processor, characterized in that: When the processor executes the program, the steps of the method for adaptively denoising voiceprints of a power grid master device according to any one of claims 1 to 7 are implemented.