Transformer substation noise matching and separating method
Through real-time tracking of microphone array system and high-order noise semantic characterization data, combined with the generative adversarial network and Monte Carlo method, efficient matching and separation of substationary noise is achieved, solving the non-stationary nature of noise frequency components and energy distribution and equipment aging. The ambiguity problems caused by environmental changes improve the accuracy of noise separation and signal quality.
Patent Information
- Application Number
- CN202410926056.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-05
AI Technical Summary
The existing noise matching separation methods of substationary stations are difficult to overcome the non-stationary nature of noise frequency components and energy distribution, as well as the fuzzy problem of noise matching separation caused by equipment aging and environmental changes.
The microphone array system is used to collect noise signals, and the independent noise source component separation is performed parallel at the edge through wave direction estimation, 3D scene analysis and real-time tracking of high-order noise semantic representation data.
It improves the accuracy and efficiency of noise separation, can adapt to noise changes, provide purer signal components, improve signal quality, enhance the processing effect of non-stationary component data, and reduce noise interference.
Smart Images

Figure CN118896684A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of substation noise signal measurement, and in particular to a substation noise matching and separation method. Background Art
[0002] Substation noise matching separation is a technology used to deal with substation noise. Substations are devices used to transform voltage or current, and often generate noise, such as noise caused by the operation of transformers, electromagnetic interference of electrical equipment, etc. Noise matching separation technology aims to separate noise from other signals by analyzing and processing substation noise in order to better understand and control the impact of noise. This technology usually includes the following steps: Data acquisition: The noise of the substation is collected using devices such as sensors or microphones. Signal processing: Digital signal processing is performed on the collected noise signal, including operations such as filtering, noise reduction and enhancement to improve the quality and clarity of the signal. Noise feature extraction: Extract the characteristics of the noise from the processed signal, such as spectral characteristics, time domain characteristics, and statistical characteristics. Matching separation algorithm: Use algorithms such as machine learning, signal processing, or pattern recognition to separate noise from other signals. This can be achieved by building a noise model, training a classifier, or applying deep learning methods. Result evaluation: The separated noise is evaluated to verify the accuracy and effectiveness of the separation effect.
[0003] However, the current substation noise matching and separation methods often have the following problems: First, the frequency component and energy distribution of the noise will change under different operating conditions, which has obvious non-stationarity, making matching and separation difficult. How to overcome non-stationarity is a major challenge; second, due to equipment aging, environmental changes and other reasons, there are differences between the actual collected noise and the training data during modeling, resulting in ambiguity in matching and separation. Summary of the invention
[0004] The purpose of the present invention is to provide a substation noise matching separation method to overcome the difficulties caused by the non-stationarity of the frequency components and energy distribution of noise and the aging of equipment, environmental changes and other reasons.
[0005] The purpose of the present invention can be achieved by the following technical solutions:
[0006] A substation noise matching and separation method comprises the following steps:
[0007] Step S1: using a microphone array system to collect noise signals at the substation site to obtain original time domain array data; performing direction of arrival estimation on the original time domain array data to obtain direction of arrival estimation data;
[0008] Step S2: obtaining non-stationary component data and residual mixed signal data according to the array structure of the microphone array system and the direction of arrival estimation data;
[0009] Step S3: Perform 3D scene analysis on the non-stationary component data and the residual mixed signal data to obtain spatiotemporal distribution feature data; perform modeling analysis on the spatiotemporal distribution feature data according to the on-site environment and equipment layout information of the substation to obtain high-order noise semantic representation data;
[0010] Step S4: Track the noise change trajectory in real time based on the high-order noise semantic representation data to obtain collaborative tracking scene data;
[0011] Step S5: Generate a noise spectrum based on the collaborative tracking scene data to obtain a confidence noise spectrum; perform edge-parallel separation of independent noise source components based on the confidence noise spectrum to obtain final separation result data.
[0012] As a preferred technical solution, the step S1 includes the following steps:
[0013] Integrate and deploy MEMS microphone chip pairs according to a preset array structure to form a microphone array system;
[0014] The noise signal at the substation site is collected using a microphone array system to obtain the original time domain array data;
[0015] The impact of non-ideal scenes on the microphone array system is estimated according to preset ideal scene data, and array response tests are performed to obtain scene impact estimation data;
[0016] Perform array response offset analysis based on scene impact estimation data to obtain array response drift data;
[0017] Performing periodic drift calibration on original time-domain array data according to array response drift data to obtain time-domain array calibration data;
[0018] Direction of arrival estimation is performed on the time domain array calibration data to obtain direction of arrival estimation data.
[0019] As a preferred technical solution, obtaining the scene impact estimation data includes the following steps:
[0020] Obtain actual scenario parameter data of substation;
[0021] Extracting deviations from actual scene parameter data according to preset ideal scene data to obtain scene parameter deviation data, wherein the scene parameter deviation data includes electromagnetic field interference data, medium non-uniformity data, and array structure deviation data;
[0022] Conduct electromagnetic compatibility tests on the connection lines of the microphone array system, establish a response model of the circuit under different electromagnetic field intensities, and obtain a magnetic field interference response model;
[0023] The impact is estimated based on the electromagnetic field interference data and the magnetic field interference response model to obtain the electromagnetic field interference impact data;
[0024] The propagation process of environmental parameter disturbance sound is simulated according to the pre-acquired environmental parameter data to obtain disturbance degree data;
[0025] The influence of medium non-uniformity data is estimated according to the disturbance degree data to obtain medium influence data;
[0026] The audio phase detection technology is used to calibrate the deviation between the microphone phase center and the theoretical position of the array structure deviation data, and the influence on the wave arrival angle and wavefront curvature is calculated to obtain the array deviation influence data;
[0027] The electromagnetic field interference impact data, medium impact data and array deviation impact data are combined into scene impact estimation data.
[0028] As a preferred technical solution, the step of performing direction of arrival estimation on the time domain array calibration data to obtain direction of arrival estimation data comprises:
[0029] Divide the time domain array calibration data into blocks of fixed time length to obtain array time block data, wherein the data blocks overlap with each other for a fixed time length;
[0030] Transforming the array time block data from the time domain to the frequency domain, and obtaining the complex spectrum value of each frequency component in the data block to obtain component complex spectrum value data;
[0031] Arranging the complex spectral values of the microphones in the same time block into a matrix according to the component complex spectral value data to obtain an array complex spectral value matrix, wherein the rows of the matrix correspond to different microphones and the columns correspond to different frequency components;
[0032] According to the orientation of each array element in the microphone array system, the manifold vector is calculated at different elevation angles and azimuth angles to obtain manifold vector data;
[0033] The MUSIC subspace algorithm is used to estimate the direction of each frequency component in the time block based on the manifold vector data and the array complex spectrum value matrix to obtain the wave direction estimation data.
[0034] As a preferred technical solution, the step S2 comprises the following steps:
[0035] Adaptive multi-layer tensor iteration is performed according to the geometric structure parameters of the deployed microphone array system and the wave direction estimation data to obtain the refined array manifold data;
[0036] Divide the frequency band covered by the fine array manifold data into sub-bands to obtain a sub-band set;
[0037] The filter level is designed according to the sub-band set to obtain a reconfigurable hierarchical filter grid;
[0038] Using hierarchical filter grids to filter and denoise the fine array manifold data, and obtain filtering result data;
[0039] The filtered result data is subjected to compressed sensing mesh coding to obtain non-stationary component data and residual mixed signal data.
[0040] As a preferred technical solution, the method of performing compressed sensing mesh coding on the filtering result data to obtain non-stationary component data and residual mixed signal data includes the following steps:
[0041] Perform wavelet transform on the filtering result data to obtain coefficient representation data in the wavelet time-frequency domain;
[0042] Perform characteristic statistics of probability distribution and energy concentration degree on coefficient representation data to obtain key coefficient data;
[0043] Designing a mesh coding structure according to key coefficient data to obtain mesh coding structure data;
[0044] Encoding and compressing the coefficient representation data according to the mesh coding structure data to obtain compressed coding data;
[0045] The compressed coded data is separated to obtain non-stationary component data and residual mixed signal data, wherein the non-stationary component data retains the transient noise part, and the residual mixed signal data retains the stationary background noise part.
[0046] As a preferred technical solution, the step S3 comprises the following steps:
[0047] Performing time standardization on the non-stationary component data and the residual mixed signal data to obtain the non-stationary component data and the residual mixed signal data with consistent time steps;
[0048] Performing short-time Fourier transform on the residual mixed signal data to obtain mixed signal spectrum feature data;
[0049] Perform wavelet time-frequency domain conversion on the non-stationary component data to obtain the non-stationary component time domain data;
[0050] The background noise pattern is extracted based on frequency and space from the mixed signal spectrum feature data, and the stable and slowly changing noise parts are assigned high weights through the attention mechanism to obtain the background noise weighted data;
[0051] The transient spike pattern of the non-stationary component time domain data is captured based on the time dimension to obtain the transient spike pattern data; according to the transient spike pattern data, the sudden and paroxysmal noise components are assigned high weights to obtain the transient noise weighted data;
[0052] According to the pre-acquired scene prior data, the transient noise weighted data and the background noise weighted data are embedded with scene semantics to obtain the spatiotemporal distribution characteristic data of the substation site noise;
[0053] According to the on-site environment and equipment layout information of the substation, the spatiotemporal distribution characteristic data is modeled and analyzed to obtain high-order noise semantic representation data.
[0054] As a preferred technical solution, the modeling and analysis of the spatiotemporal distribution feature data according to the on-site environment and equipment layout information of the substation to obtain high-order noise semantic representation data includes the following steps:
[0055] Encode the scene prior data into structured semantic representation to obtain scene semantic encoding data;
[0056] The temporal and spatial distribution feature data is modeled according to the scene semantic coding data, and the key parameters of the noise burst, persistence and attenuation in time are extracted to obtain the time domain characteristic data;
[0057] The spatiotemporal distribution feature data is modeled in the frequency domain according to the scene semantic coding data, and the frequency components of the noise and the characteristics of the energy distribution are identified to obtain the frequency domain characteristic data;
[0058] Based on the scene semantic coding data, the spatiotemporal distribution feature data is spatially modeled, and the propagation and diffusion of noise in space and the spatial pattern of hot spots are analyzed to obtain spatial characteristic data;
[0059] High-order semantic analysis is performed on the time domain characteristic data, frequency domain characteristic data and spatial characteristic data to obtain high-order noise semantic representation data.
[0060] As a preferred technical solution, the step S4 comprises the following steps:
[0061] An unsupervised generative adversarial network framework is designed based on high-order noise semantic representation data, in which the generator learns to generate high-order noise semantic sequences, and the discriminator determines the authenticity of high-order noise semantic sequences compared with high-order noise semantic representation data;
[0062] Performing semantic latent encoding on the high-order noise semantic representation data to obtain noise semantic latent data;
[0063] The noise semantic latent data is used as the starting condition to train the generator and the discriminator to obtain the noise semantic time series trajectory data;
[0064] The noisy semantic temporal trajectory data is decoded in actual scenes to obtain collaborative tracking scene data.
[0065] As a preferred technical solution, step S5 includes the following steps:
[0066] The Monte Carlo method is used to simulate the propagation of noise in the three-dimensional scene of the substation based on the collaborative tracking scene data to obtain fuzzy noise simulation data;
[0067] The fuzziness degree of noise propagation is quantified and classified according to the fuzzy noise simulation data to obtain the noise fuzziness degree data;
[0068] According to the noise fuzziness degree data, the fuzzy noise simulation data is subjected to fuzzy set theory-based fuzzy degree reduction to obtain a confidence noise spectrum;
[0069] The confidence noise map is deployed to the edge device, and the independent noise source components are separated in parallel at the edge to obtain the final separation result data.
[0070] Compared with the prior art, the present invention has the following beneficial effects:
[0071] The use of the sensor integrated chip deployment and microphone array system of the present invention can accurately collect noise signals at the substation site and provide raw data for subsequent processing. Drift calibration and wave direction estimation can eliminate the influence of non-ideal fields and provide accurate wave direction estimation data. This helps to locate the position and direction of the noise source and provide important information for subsequent noise separation. The acquisition of adaptive multi-layer tensor iteration and fine array manifold data can better represent the noise signal and capture the spatiotemporal characteristics of the noise. This helps to improve the accuracy and effect of noise separation. Designing a reconfigurable hierarchical filter grid and performing filtering and denoising can reduce noise interference, extract the required signal components, and improve the quality of the signal. This provides a more reliable data basis for subsequent noise separation and signal analysis. By performing 3D scene analysis on non-stationary component data and residual mixed signal data, the spatiotemporal distribution feature data of the noise can be extracted. This helps to have a deeper understanding of the generation and propagation laws of noise. Modeling and analysis based on the on-site environment and equipment layout information of the substation can more accurately interpret and analyze the spatiotemporal distribution feature data and obtain high-order noise semantic representation data. This provides richer information for subsequent noise separation and analysis. Noise tracking based on high-order noise semantic representation data can capture the changes of noise in real time and accurately model the dynamic characteristics of noise. This helps to better understand the time-varying characteristics of noise and provide a more accurate basis for subsequent noise separation and removal. The adaptive update of the noise tracking model can adjust and optimize the model parameters according to the actual noise changes, providing real-time and accurate noise estimation results. This enables the noise tracking model to adapt to the changes of noise and provide more reliable noise estimation. The application of Monte Carlo ambiguity reduction and noise spectrum generation can further enhance the processing effect of non-stationary component data, reduce the interference of noise on the signal, and improve the quality of separation results. The edge-parallel independent noise source component separation algorithm can separate noise and signal more accurately and provide purer signal components. This is of great significance for subsequent signal analysis, fault detection and other tasks, and provides a more reliable signal basis. In summary, the detailed description of the above steps shows the specific value and significance of the substation noise matching separation method in each step. These steps include accurately collecting noise signals, locating noise sources, improving signal quality, capturing the temporal and spatial characteristics of noise, accurately modeling the dynamic characteristics of noise, and separating noise and signal components. These effects help to improve the accuracy and efficiency of noise processing and provide a more reliable data basis for subsequent analysis and processing tasks. BRIEF DESCRIPTION OF THE DRAWINGS
[0072] Figure 1 A schematic flow chart of the steps of the substation noise matching and separation method of the present invention;
[0073] Figure 2Detailed step flow diagram of step S1 in the substation noise matching and separation method of the present invention;
[0074] Figure 3 The figure is a detailed flowchart of step S2 in the substation noise matching and separation method of the present invention. DETAILED DESCRIPTION
[0075] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.
[0076] Example 1
[0077] As one embodiment of the present invention, the present invention provides a noise matching and separation method for a substation. Figure 1 As shown, in this example, the substation noise matching and separation method includes the following steps:
[0078] Step S1: deploy sensor integrated chips in the substation to form a microphone array system, and use the microphone array system to collect noise signals at the substation site to obtain original time domain array data; perform drift calibration based on non-ideal field offset on the original time domain array data, and perform wave direction estimation to obtain wave direction estimation data;
[0079] The embodiment of the present invention installs the sensor integrated chip to the appropriate position in the substation and ensures its normal operation; connects multiple microphones together to form a microphone array system, and the layout of the microphones can adopt linear, planar or other suitable structures. Start the microphone array system to start receiving noise signals at the substation site; collect noise signals through the microphone array system and convert them into electrical signals. Perform drift calibration for non-ideal field offset on the collected raw time domain array data; this can be achieved by performing time delay calibration on each microphone signal in the array to eliminate offsets caused by inaccurate microphone position or environmental influences; perform wave direction estimation, and estimate the wave direction of the noise signal by performing signal processing and beamforming algorithm on the calibrated array data.
[0080] Step S2: performing adaptive multi-layer tensor iteration according to the array structure of the microphone array system and the wave direction estimation data, so as to obtain fine array manifold data; designing a reconfigurable hierarchical filter grid according to the fine array manifold data, and filtering and denoising the fine array manifold data using the hierarchical filter grid, so as to obtain filtering result data; performing compressed sensing mesh coding on the filtering result data, so as to obtain non-stationary component data and residual mixed signal data;
[0081] The embodiment of the present invention constructs a multi-layer tensor model based on the structure of the microphone array and the wave direction estimation data; the multi-layer tensor model is optimized using an adaptive iterative algorithm to extract fine array manifold data. Based on the fine array manifold data, a hierarchical filter grid is designed, and the grid has different filtering characteristics in space; the design of the grid can be optimized according to the spectral characteristics and spatial distribution of the noise to maximize the denoising effect. The fine array manifold data is passed to the designed hierarchical filter grid for filtering and denoising operations; filtering and denoising can be achieved by applying appropriate filtering algorithms to each grid element, such as wavelet transform, adaptive filtering, etc. The filtered result data is compressed sensing encoded to reduce the storage and transmission overhead of the data; using a suitable coding algorithm, such as sparse representation, wavelet coding, etc., the data is encoded into non-stationary component data and residual mixed signal data.
[0082] Step S3: Perform 3D scene analysis on the non-stationary component data and the residual mixed signal data to obtain spatiotemporal distribution feature data; perform modeling analysis on the spatiotemporal distribution feature data according to the on-site environment and equipment layout information of the substation to obtain high-order noise semantic representation data;
[0083] The embodiment of the present invention performs 3D scene analysis on non-stationary component data and residual mixed signal data to obtain the spatial and temporal distribution characteristics of noise; signal processing and image processing techniques, such as time-frequency analysis, spatial filtering, clustering, etc., can be used to analyze data and extract relevant features. Based on the on-site environment and equipment layout information of the substation, the spatiotemporal distribution feature data is modeled and analyzed; statistical analysis, machine learning and other methods can be used to associate the data with the structure and noise source of the substation to obtain high-order noise semantic representation data.
[0084] Step S4: Based on the high-order noise semantic representation data, a generative adversarial network is used to track the noise change trajectory in real time, and the noise tracking model in the generative adversarial network is adaptively updated to obtain collaborative tracking scene data;
[0085] The embodiment of the present invention uses a generative adversarial network (GAN) to track the noise change trajectory in real time based on high-order noise semantic representation data; trains a noise tracking model, which receives the current high-order noise semantic representation data and previous tracking results, and outputs a predicted noise change trajectory. According to the results of real-time tracking, the noise tracking model is adaptively updated to improve the accuracy and stability of tracking; feedback control algorithms or incremental learning methods can be used to update model parameters according to actual tracking errors. The real-time tracked noise change trajectory is processed in coordination with other related scene data; the noise change trajectory can be integrated with the monitoring data of the substation, equipment status, etc. to obtain more comprehensive scene information.
[0086] Step S5: Perform Monte Carlo ambiguity reduction on the collaborative tracking scene data and generate a noise spectrum to obtain a confidence noise spectrum; perform edge-parallel separation of independent noise source components based on the confidence noise spectrum to obtain final separation result data.
[0087] The embodiment of the present invention performs Monte Carlo ambiguity reduction on the collaborative tracking scene data to reduce uncertainty and improve the confidence of the noise spectrum; the Monte Carlo method is used to sample and reconstruct the tracking results, and the probability distribution of the noise is estimated through multiple sampling. Based on the data after Monte Carlo reduction, a noise spectrum is generated; statistical modeling methods, such as Gaussian mixture model, hidden Markov model, etc., can be used to model the data to obtain a representation of the noise spectrum. Edge-parallel independent noise source components are separated according to the noise spectrum; blind source separation algorithms, such as independent component analysis (ICA), non-negative matrix decomposition (NMF), etc., are used to decompose the noise spectrum into independent noise source components. The final separation result data is obtained, in which each component represents an independent noise source; the separation results can be further processed and analyzed according to application requirements, such as noise removal, noise source location, etc.
[0088] The use of the sensor integrated chip deployment and microphone array system of the present invention can accurately collect noise signals at the substation site and provide raw data for subsequent processing. Drift calibration and wave direction estimation can eliminate the influence of non-ideal fields and provide accurate wave direction estimation data. This helps to locate the position and direction of the noise source and provide important information for subsequent noise separation. The acquisition of adaptive multi-layer tensor iteration and fine array manifold data can better represent the noise signal and capture the spatiotemporal characteristics of the noise. This helps to improve the accuracy and effect of noise separation. Designing a reconfigurable hierarchical filter grid and performing filtering and denoising can reduce noise interference, extract the required signal components, and improve the quality of the signal. This provides a more reliable data basis for subsequent noise separation and signal analysis. By performing 3D scene analysis on non-stationary component data and residual mixed signal data, the spatiotemporal distribution feature data of the noise can be extracted. This helps to have a deeper understanding of the generation and propagation laws of noise. Modeling and analysis based on the on-site environment and equipment layout information of the substation can more accurately interpret and analyze the spatiotemporal distribution feature data and obtain high-order noise semantic representation data. This provides richer information for subsequent noise separation and analysis. Noise tracking based on high-order noise semantic representation data can capture the changes of noise in real time and accurately model the dynamic characteristics of noise. This helps to better understand the time-varying characteristics of noise and provide a more accurate basis for subsequent noise separation and removal. The adaptive update of the noise tracking model can adjust and optimize the model parameters according to the actual noise changes, providing real-time and accurate noise estimation results. This enables the noise tracking model to adapt to the changes of noise and provide more reliable noise estimation. The application of Monte Carlo ambiguity reduction and noise spectrum generation can further enhance the processing effect of non-stationary component data, reduce the interference of noise on the signal, and improve the quality of separation results. The edge-parallel independent noise source component separation algorithm can separate noise and signal more accurately and provide purer signal components. This is of great significance for subsequent signal analysis, fault detection and other tasks, and provides a more reliable signal basis. In summary, the detailed description of the above steps shows the specific value and significance of the substation noise matching separation method in each step. These steps include accurately collecting noise signals, locating noise sources, improving signal quality, capturing the temporal and spatial characteristics of noise, accurately modeling the dynamic characteristics of noise, and separating noise and signal components. These effects help to improve the accuracy and efficiency of noise processing and provide a more reliable data basis for subsequent analysis and processing tasks.
[0089] Preferably, step S1 comprises the following steps:
[0090] Step S11: integrating and deploying the MEMS microphone chip pairs according to a preset array structure to form a microphone array system;
[0091] Step S12: using a microphone array system to collect noise signals at the substation site, thereby obtaining original time domain array data;
[0092] Step S13: estimating the impact of a non-ideal scene on the microphone array system according to preset ideal scene data, and performing an array response test, thereby obtaining scene impact estimation data;
[0093] Step S14: performing an array response offset analysis based on the scene impact estimation data, thereby obtaining array response drift data;
[0094] Step S15: performing periodic and small batch drift calibration on the original time domain array data according to the array response drift data, thereby obtaining time domain array calibration data;
[0095] Step S16: performing direction of arrival estimation on the time domain array calibration data, thereby obtaining direction of arrival estimation data.
[0096] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S1 in the embodiment of the present invention, step S1 includes the following steps:
[0097] Step S11: integrating and deploying the MEMS microphone chip pairs according to a preset array structure to form a microphone array system;
[0098] The embodiment of the present invention selects a MEMS microphone chip suitable for a microphone array and ensures that it has the required performance and characteristics; designs the structure of the microphone array, including the number, arrangement and layout of the microphones, according to the application requirements and scene characteristics. According to the designed array structure, the selected MEMS microphone chips are installed one by one in appropriate positions and connected to corresponding circuits and interfaces; the array is calibrated to ensure that the response characteristics and sensitivity of each microphone are consistent. This can be achieved by sending a known signal or using a calibration algorithm.
[0099] Step S12: using a microphone array system to collect noise signals at the substation site, thereby obtaining original time domain array data;
[0100] The embodiment of the present invention places the microphone array system at an appropriate location in the substation to ensure that the required scene area is covered; the microphone array system is started to start receiving noise signals at the substation site. The microphone array converts the signal into an electrical signal and transmits it to a data acquisition device through an interface; the data acquisition device is used to record the original time domain array data received from the microphone array system. The data acquisition device can be a digital audio interface, a data acquisition card, etc.
[0101] Step S13: estimating the impact of a non-ideal scene on the microphone array system according to preset ideal scene data, and performing an array response test, thereby obtaining scene impact estimation data;
[0102] The embodiment of the present invention establishes an ideal scene model based on preset ideal scene characteristics, including expected noise distribution, sound source direction, etc.; and estimates the impact of non-ideal scenes on the microphone array system based on environmental factors and interference sources in the actual scene. This can be done through field observation, measurement or simulation analysis. An array response test is performed to measure the response characteristics of the microphone array system in non-ideal scenes. This can be achieved by sending a known signal and recording the microphone output. Based on the array response test results and the non-ideal scene impact estimation, scene impact estimation data is obtained to describe the degree and characteristics of the impact of non-ideal scenes on the microphone array.
[0103] Step S14: performing an array response offset analysis based on the scene impact estimation data, thereby obtaining array response drift data;
[0104] The embodiment of the present invention analyzes the frequency and spatial offset of the array response based on the scene impact estimation data. This can be achieved by comparing the array response characteristics in ideal scenes and non-ideal scenes. Based on the analysis results, the drift data of the array response is extracted, including information such as the offset of the frequency response and the sensitivity change.
[0105] Step S15: performing periodic and small batch drift calibration on the original time domain array data according to the array response drift data, thereby obtaining time domain array calibration data;
[0106] The embodiment of the present invention selects an appropriate drift calibration method according to the characteristics of the array response drift data. Common methods include batch calibration, online calibration, etc. The array is calibrated using the drift calibration method. This can be done by sending a known signal and recording the microphone output to collect calibration data. Based on the collected calibration data, the original time domain array data is drift calibrated. This can be achieved by performing gain adjustment, phase calibration, and other processing on the signal of each microphone channel. The time domain array calibration data after drift calibration is obtained to reduce the impact of non-ideal scenarios and improve data quality.
[0107] Step S16: performing direction of arrival estimation on the time domain array calibration data, thereby obtaining direction of arrival estimation data.
[0108] The embodiment of the present invention selects an appropriate DOA estimation method according to application requirements and data characteristics. Common methods include beamforming, maximum likelihood estimation, etc. The time domain array calibration data is preprocessed, including denoising, filtering and other operations to improve the accuracy of DOA estimation. The preprocessed data is processed using the selected DOA estimation method to obtain an estimation result of the DOA. This can be a single direction estimation or a multi-direction estimation, and the specific method depends on the application requirements. The DOA estimation data is obtained to describe the direction and position information of the sound source in a given scenario.
[0109] The present invention can form a microphone array system by integrating and deploying MEMS microphone chips according to a preset array structure. This array structure can provide the ability of multiple microphones to collect noise signals at the same time, thereby capturing more details and spatiotemporal features. The original time domain array data can be obtained by collecting noise signals at the substation site through the microphone array system. These data provide a comprehensive observation of substation noise and provide a basis for subsequent processing and analysis. By using the preset ideal scene data to estimate the impact of the non-ideal scene on the site where the microphone array system is located, the impact of the environment on the array system can be understood. Performing an array response test and obtaining scene impact estimation data helps to accurately evaluate the performance of the array system and provide a reference basis for subsequent processing. By analyzing the scene impact estimation data, the offset information of the array response can be obtained. These drift data reflect the response changes of the array system in the field environment, providing a basis for subsequent drift calibration. By using the array response drift data to perform drift calibration on the original time domain array data, the non-ideal effect caused by the array response change can be eliminated. The time domain array calibration data obtained in this way has better accuracy and reliability, providing a more reliable basis for subsequent wave direction estimation and noise separation. By performing DOA estimation on the time domain array calibration data, the location and direction of the noise source can be determined. The DOA estimation data provides spatial characteristic information about the noise, which provides an important positioning basis for subsequent noise separation and removal.
[0110] Preferably, step S13 comprises the following steps:
[0111] Step S131: Acquire actual scenario parameter data of the substation;
[0112] The embodiment of the present invention uses an electromagnetic field measuring instrument to measure the electromagnetic field in the substation and record the measurement results. Common electromagnetic field measuring instruments include magnetic field measuring instruments, electric field measuring instruments, etc. The environmental parameter data of the substation, such as temperature, humidity, etc., are obtained through environmental parameter measuring instruments. These parameters have an impact on the non-uniformity of the medium. The actual structure of the microphone array system is measured using a measuring instrument, including information such as the position and layout of the microphone. Common measuring instruments include laser rangefinders, three-dimensional scanners, etc.
[0113] Step S132: extracting deviations from actual scene parameter data according to preset ideal scene data, thereby obtaining scene parameter deviation data, wherein the scene parameter deviation data includes electromagnetic field interference data, medium non-uniformity data, and array structure deviation data;
[0114] The embodiment of the present invention sets the ideal scene data according to the application requirements, including parameters such as ideal electromagnetic field interference, medium uniformity, and array structure. By comparing the actual scene parameter data with the ideal scene data, the deviation data of electromagnetic field interference, medium non-uniformity, and array structure are extracted. This can be achieved by calculating the difference, statistical analysis, and other methods.
[0115] Step S133: performing an electromagnetic compatibility test on the connection lines of the microphone array system, and establishing a response model of the circuit under different electromagnetic field intensities, thereby obtaining a magnetic field interference response model;
[0116] The embodiment of the present invention uses electromagnetic compatibility testing equipment to test the connection line of the microphone array system to evaluate its performance and stability in an electromagnetic environment. This can include tests on aspects such as conducted interference and radiated interference. Based on the test results, a response model of the connection line under different electromagnetic field strengths is established. This can use mathematical modeling methods, such as curve fitting, regression analysis, etc., to obtain the response function of the connection line.
[0117] Step S134: performing impact estimation according to the electromagnetic field interference data and the magnetic field interference response model, thereby obtaining electromagnetic field interference impact data;
[0118] The embodiment of the present invention preprocesses the electromagnetic field interference data, such as denoising and filtering, to extract effective interference signals. Using the magnetic field interference response model, the preprocessed electromagnetic field interference data and the response model are calculated to obtain the actual impact of the electromagnetic field interference on the system. This can include estimates of the signal-to-noise ratio drop and distortion caused by interference.
[0119] Step S135: simulating the propagation process of the environmental parameter disturbance sound according to the pre-acquired environmental parameter data, so as to obtain disturbance degree data;
[0120] The embodiment of the present invention uses the pre-acquired environmental parameter data to establish an environmental parameter simulation model to simulate the changes of environmental parameters during the sound propagation process. For example, acoustic simulation software can be used to perform simulation calculations. Based on the environmental parameter simulation results, the impact of environmental parameter disturbances on the sound propagation process is extracted. This may include data on sound attenuation, propagation delay, etc.
[0121] Step S136: estimating the influence of the medium non-uniformity data according to the disturbance degree data, thereby obtaining medium influence data;
[0122] The embodiment of the present invention analyzes the disturbance degree data to understand the influence of the disturbance on the medium heterogeneity. For example, by statistical analysis, correlation analysis and other methods. According to the analysis results, the influence of the medium heterogeneity data is estimated. This may include the estimation of the changes in medium parameters, acoustic characteristics and other aspects.
[0123] Step S137: using audio phase detection technology to calibrate the deviation between the microphone phase center and the theoretical position of the array structure deviation data, and calculate the impact on the wave arrival angle and wavefront curvature, so as to obtain array deviation impact data;
[0124] The embodiment of the present invention uses audio phase detection technology to obtain the phase information of each microphone in the microphone array. Common methods include cross-correlation method, phase difference measurement method, etc. According to the phase information, the array structure deviation is calibrated, and the microphone phase center is aligned with the theoretical position. Then, based on the calibrated data, the influence of the array deviation on the angle of arrival and wavefront curvature is calculated. This can be achieved by analyzing phase difference, beamforming and other methods. According to the calculation results, the influence data of the array deviation on the angle of arrival and wavefront curvature are obtained. These data can be used to evaluate the influence of the array structure on sound reception, as well as correction and compensation in signal processing.
[0125] Step S138: combining the electromagnetic field interference impact data, the medium impact data and the array deviation impact data into scene impact estimation data.
[0126] The embodiment of the present invention determines the weights of electromagnetic field interference, medium non-uniformity and array deviation in the overall impact according to actual application requirements and scenario characteristics. This can be determined by expert evaluation, experimental data analysis and other methods. The electromagnetic field interference impact data, medium impact data and array deviation impact data are integrated according to the weights to obtain the overall impact data. This can be achieved by using weighted average, multi-dimensional data fusion and other methods.
[0127] The present invention can obtain the specific characteristics and parameters of the on-site environment by obtaining the actual scene parameter data of the substation. These data provide a practical reference for subsequent impact estimation and calibration. By comparing the actual scene parameter data with the preset ideal scene data, the deviation information of the scene parameters can be extracted. These deviation data include electromagnetic field interference data, medium non-uniformity data and array structure deviation data, which provide a basis for subsequent impact estimation and calibration. By performing electromagnetic compatibility testing on the connection lines of the microphone array system, the response of the circuit under different electromagnetic field intensities can be understood. Establishing a magnetic field interference response model helps to predict and estimate the impact of the electromagnetic field on the system, and provides a reference for subsequent impact estimation and calibration. By using electromagnetic field interference data and magnetic field interference response models, the impact of electromagnetic field interference can be estimated. Electromagnetic field interference impact data provides information about the impact of electromagnetic field interference on system performance, providing a basis for subsequent impact calibration and processing. By simulating the pre-acquired environmental parameter data, the degree of disturbance of environmental parameters on the sound propagation process can be understood. The disturbance degree data provides quantitative information about environmental parameter disturbance, which helps subsequent impact estimation and calibration. By using the disturbance degree data to estimate the impact of medium non-uniformity data, the degree of impact of medium non-uniformity on system performance can be understood. The medium impact data provides information about the medium's impact on system performance, providing a basis for subsequent impact calibration and processing. By calibrating the array structure deviation data through audio phase detection technology, the deviation between the microphone phase center and the theoretical position can be accurately determined. Such calibration helps to improve the positioning accuracy of the array system and calculate the impact on the wave arrival angle and wavefront curvature. The array deviation impact data provides information about the array structure deviation on system performance, providing a basis for subsequent impact estimation and calibration. By combining the electromagnetic field interference impact data, medium impact data and array deviation impact data, comprehensive scene impact estimation data can be obtained. These data provide comprehensive information about the impact of scene parameter deviation on system performance, which helps to analyze and optimize the performance of the system and provide a basis for subsequent decision-making and improvement.
[0128] Preferably, step S16 comprises the following steps:
[0129] Step S161: dividing the time domain array calibration data into blocks of fixed time length, thereby obtaining array time block data, wherein the data blocks overlap with each other for a fixed time length;
[0130] The embodiment of the present invention obtains time domain array calibration data, that is, the output signal of the microphone array after calibration. The time domain array calibration data is divided into blocks according to a fixed time length, and a sliding window method can be used. For example, a time window length of T is selected, and a fixed overlap length of T is set. overlapBy sliding the window, multiple data blocks of fixed time length are extracted from the time domain array calibration data. The length of each data block is T, and there is TT between adjacent data blocks. overlap overlapping part.
[0131] Step S162: transform the array time block data from the time domain to the frequency domain, and obtain the complex spectrum value of each frequency component in the data block, thereby obtaining component complex spectrum value data;
[0132] The embodiment of the present invention performs a fast Fourier transform (FFT) on each array time block data to convert the time domain signal into a frequency domain signal. For each data block, the complex spectrum value of each frequency component in its frequency domain representation is extracted. The complex spectrum value includes amplitude and phase information, which can represent the amplitude and phase of the frequency component. Component complex spectrum value data is obtained, wherein the complex spectrum value of each frequency component in each data block can be represented in the form of a matrix or array.
[0133] Step S163: Arranging the complex spectral values of the microphones in the same time block into a matrix according to the component complex spectral value data, thereby obtaining an array complex spectral value matrix, wherein the rows of the matrix correspond to different microphones, and the columns correspond to different frequency components;
[0134] For the component complex spectral value data of each time block, the embodiment of the present invention arranges the complex spectral values of different microphones in the same time block into a matrix in a certain order. Assuming that there are N microphones in the array and the number of frequency components is M, an array complex spectral value matrix of size N×M is obtained. In the matrix, each row corresponds to a microphone, each column corresponds to a frequency component, and each element in the matrix is the complex spectral value of the corresponding microphone on the corresponding frequency component.
[0135] Step S164: performing manifold vector calculation at different elevation angles and azimuth angles according to the orientation of each array element in the microphone array system, thereby obtaining manifold vector data;
[0136] The embodiment of the present invention determines the orientation information of each array element in the microphone array system. This can be determined by the position and direction arrangement when designing and arranging the array, or obtained by additional sensors or calibration methods. Within the range of different elevation angles and azimuth angles, a set of discrete direction angles are selected and used as the target direction for manifold vector calculation. For each target direction, the corresponding manifold vector is calculated based on the orientation information of each array element in the microphone array system. The manifold vector is a set of weight vectors related to the target direction, which is used to describe the distribution of sound sources in the sound field.
[0137] Step S165: using the MUSIC subspace algorithm to estimate the direction of each frequency component in the time block according to the manifold vector data and the array complex spectral value matrix, thereby obtaining the direction of arrival estimation data.
[0138] The embodiment of the present invention inputs the obtained array complex spectral value matrix and manifold vector data into the MUSIC (Multiple Signal Classification) subspace algorithm. The MUSIC algorithm is a spectrum estimation method based on subspace decomposition, which is used to estimate the direction of the signal. The MUSIC algorithm obtains the signal subspace and the noise subspace by calculating the eigendecomposition of the array complex spectral value matrix. The pseudo-spectral function is constructed using the manifold vector data, and the direction of arrival of each frequency component in each time block can be estimated by finding the minimum point in the pseudo-spectral function. For each frequency component in each time block, the corresponding direction of arrival estimation data is obtained by the MUSIC algorithm, which can be expressed in the form of an angle or a direction vector.
[0139] The present invention divides the time domain array calibration data into blocks of fixed time length, thereby obtaining array time block data. Through this step, the accuracy and stability of data processing can be improved, and more data samples can be provided for subsequent analysis and processing. The array time block data is transformed from the time domain to the frequency domain, and the complex spectrum values of each frequency component in the data block are obtained, thereby obtaining component complex spectrum value data. This conversion operation can analyze the data from the frequency domain perspective, and obtain the amplitude and phase information of each frequency component, providing a basis for subsequent direction estimation. According to the component complex spectrum value data, the complex spectrum values of the microphones in the same time block are arranged into a matrix, thereby obtaining an array complex spectrum value matrix. By integrating the data into a matrix form, the relationship between different microphones and frequency components can be more conveniently represented, thereby providing a more convenient operation and analysis method for direction estimation and data processing. According to the orientation of each array element in the microphone array system, the manifold vector is calculated at different elevation angles and azimuth angles, thereby obtaining manifold vector data. By calculating the manifold vector, the response characteristics of the microphone array in different directions can be described, providing important information for subsequent wave arrival direction estimation. The MUSIC subspace algorithm is used to estimate the direction of each frequency component in the time block based on the manifold vector data and the array complex spectrum value matrix, thereby obtaining the direction of arrival estimation data. This step provides an estimation result about the direction of the sound source, providing important data basis for sound source positioning and tracking.
[0140] Preferably, step S2 comprises the following steps:
[0141] Step S21: performing adaptive multi-layer tensor iteration according to the geometric structure parameters of the deployed microphone array system and the direction of arrival estimation data, so as to obtain fine array manifold data;
[0142] Step S22: dividing the frequency band covered by the fine array manifold data into sub-bands, thereby obtaining a sub-band set;
[0143] Step S23: Designing filter levels according to the sub-band set, thereby obtaining a reconfigurable hierarchical filter grid;
[0144] Step S24: filtering and denoising the fine array manifold data using a hierarchical filter grid, thereby obtaining filtering result data;
[0145] Step S25: performing compressed sensing mesh coding on the filtering result data to obtain non-stationary component data and residual mixed signal data.
[0146] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 Detailed step flow diagram of step S2 in the embodiment of the present invention, step S2 includes the following steps:
[0147] Step S21: performing adaptive multi-layer tensor iteration according to the geometric structure parameters of the deployed microphone array system and the direction of arrival estimation data, so as to obtain fine array manifold data;
[0148] The embodiment of the present invention obtains the geometric parameters of the deployed microphone array system, including information such as the position, orientation and array shape of the microphone. The wave direction estimation data obtained in step S165 is combined with the geometric parameters of the microphone array to perform adaptive multi-layer tensor iteration. Adaptive multi-layer tensor iteration is an optimization method for estimating the weight coefficient of each microphone in the array to minimize the error between the estimated result and the observed data. During the iteration process, the value of the weight coefficient is adjusted according to the difference between the observed data and the estimated result until the convergence condition is reached. The result obtained by adaptive multi-layer tensor iteration, that is, the refined array manifold data, can more accurately describe the distribution of sound sources in the sound field.
[0149] Step S22: dividing the frequency band covered by the fine array manifold data into sub-bands, thereby obtaining a sub-band set;
[0150] The embodiment of the present invention performs frequency domain analysis on the fine array manifold data, for example, using a fast Fourier transform (FFT) to convert the data into the frequency domain. The data in the frequency domain is divided into sub-bands according to application requirements and signal characteristics. The division can be based on frequency range, bandwidth or other related indicators. The frequency domain data is divided into multiple sub-bands, each sub-band corresponds to a certain frequency range, and a sub-band set is formed.
[0151] Step S23: Designing filter levels according to the sub-band set, thereby obtaining a reconfigurable hierarchical filter grid;
[0152] The embodiment of the present invention designs a corresponding number of filter levels according to the number of sub-band sets. Each filter level corresponds to a sub-band in the sub-band set. The filter level can adopt different filter types and parameters, such as bandpass filters, high-pass filters or other filter structures, and is selected and designed according to the characteristics of the sub-bands. The designed filter levels are combined to form a reconfigurable hierarchical filter grid. The filter grid can be used for subsequent filtering processing operations.
[0153] Step S24: filtering and denoising the fine array manifold data using a hierarchical filter grid, thereby obtaining filtering result data;
[0154] The embodiment of the present invention inputs the fine array manifold data into the hierarchical filter grid. According to the structure and parameters of the filter grid, the input data is filtered. Convolution operation or other filtering methods can be used. The filtering process will filter and denoise the signals in different frequency ranges or sub-bands according to the design of the filter grid, remove noise or unnecessary signal components, and obtain the filtering result data, that is, the data after filtering, the noise component of which is significantly reduced and the signal quality is improved.
[0155] Step S25: performing compressed sensing mesh coding on the filtering result data to obtain non-stationary component data and residual mixed signal data.
[0156] The embodiment of the present invention performs compressed sensing processing on the filtering result data to reduce the storage space or transmission bandwidth of the data; compressed sensing is a compression method based on signal sparsity, which uses the sparsity of the signal in a certain representation domain to achieve efficient compression through sampling and reconstruction processes; in compressed sensing, a sparse representation method (such as wavelet transform or sparse dictionary) can be used to represent the filtering result data; the sampling process downsamples the represented data to reduce the sampling rate; through the decoding and reconstruction process, non-stationary component data and residual mixed signal data are obtained; the non-stationary component data represents the important signal components in the filtering result, and the residual mixed signal data represents the compressed or lost signal components.
[0157] The present invention improves the sensitivity and accuracy of the microphone array system to the direction of the sound source through adaptive multi-layer tensor iteration and acquisition of fine array manifold data. Secondly, through sub-band division and hierarchical filter design, the signal processing process is optimized, making it possible to optimize the characteristics of different frequency bands. This further improves the efficiency and accuracy of the processing. In addition, the application of filtering denoising and compressed sensing coding brings about the reduction of noise and interference, improves signal quality and clarity, and realizes efficient data compression and transmission. In summary, the beneficial effects of the entire step include enhancing the accuracy of sound source positioning, optimizing the efficiency and accuracy of signal processing, improving signal quality and clarity, and realizing efficient data compression and transmission.
[0158] Preferably, step S25 comprises the following steps:
[0159] Step S251: performing wavelet transform on the filtering result data, thereby obtaining coefficient representation data in the wavelet time-frequency domain;
[0160] In an embodiment of the present invention, the filtering result data is subjected to wavelet transform to convert it into the wavelet time-frequency domain. Wavelet transform is a time-frequency analysis method that decomposes the signal into wavelet coefficients of different scales and frequencies. Wavelet transform tools and libraries, such as PyWavelets, Wavelet Toolbox, etc., can be used to select appropriate wavelet basis functions and transformation methods for transformation.
[0161] Step S252: Performing characteristic statistics of probability distribution and energy concentration degree on the coefficient representation data, thereby obtaining key coefficient data;
[0162] The embodiment of the present invention performs statistical analysis on the coefficient representation data obtained by wavelet transform, including probability distribution and energy concentration degree, etc.; the mean, variance, skewness, kurtosis and other statistical quantities of the coefficient representation data can be calculated to describe the probability distribution characteristics; considering the characteristics of the coefficient representation data in the time and frequency domain, the coefficient data of different scales and frequency bands can be analyzed to extract key coefficients.
[0163] Step S253: designing a mesh coding structure according to the key coefficient data, thereby obtaining mesh coding structure data;
[0164] The embodiment of the present invention designs a mesh coding structure according to the characteristics and importance of key coefficient data; the mesh coding structure may include coding nodes and connection relationships at different levels for storing and representing coefficient data; and an appropriate coding structure, such as a tree structure, a Huffman tree, etc., may be designed according to the distribution characteristics and correlation of the coefficient data.
[0165] Step S254: encoding and compressing the coefficient representation data according to the mesh coding structure data, thereby obtaining compressed coding data;
[0166] The embodiments of the present invention utilize a designed mesh coding structure to encode and compress coefficient representation data; entropy coding methods, such as Huffman coding, arithmetic coding, etc., can be used to map coefficient data into a shorter coding sequence; based on the coding structure and coding algorithm, key coefficient data is encoded and compressed to reduce data storage space or transmission bandwidth.
[0167] Step S255: Separate the compressed coded data to obtain non-stationary component data and residual mixed signal data, wherein the non-stationary component data retains the transient noise part, and the residual mixed signal data retains the stationary background noise part.
[0168] The embodiment of the present invention decodes the compressed coded data and restores it to the representation form before encoding; according to the encoding structure and the decoding algorithm, the compressed coded data is separated into non-stationary component data and residual mixed signal data; the non-stationary component data represents the transient noise part in the filtering result, which can be used for subsequent processing or analysis; the residual mixed signal data represents the stationary background noise part in the filtering result, which can also be used for specific applications or subsequent processing.
[0169] The present invention converts the filtering result data into the wavelet time-frequency domain through wavelet transform, and can extract the time-frequency characteristics of the signal, so as to better describe the local changes and frequency distribution of the signal. Secondly, by performing characteristic statistics on the coefficient representation data, such as analysis of probability distribution and energy concentration, the key coefficient data, i.e., the coefficient carrying the main information, can be determined. This helps to reduce redundant information and improve the compression efficiency of data. Further, through the design of the mesh coding structure, according to the characteristics of the key coefficient data, a more effective data encoding method is achieved. This helps to further reduce the redundancy of the data and improve the efficiency of compression coding. Finally, by separating the compressed coding data, the non-stationary component data and the residual mixed signal data are separated, and the decomposition and separation of the signal are achieved. In summary, the whole step includes extracting the time-frequency characteristics of the signal, reducing redundant information, improving the data compression efficiency, and achieving the decomposition and separation of the signal. These effects help to improve the efficiency and accuracy of signal processing and analysis, while reducing the storage and transmission overhead of the data.
[0170] Preferably, step S3 comprises the following steps:
[0171] Step S31: performing time standardization on the non-stationary component data and the residual mixed signal data, so as to obtain the non-stationary component data and the residual mixed signal data with consistent time steps;
[0172] The embodiment of the present invention performs time alignment and sampling rate adjustment on the non-stationary component data and the residual mixed signal data so that they have the same time step and sampling rate; signal processing tools and libraries, such as NumPy, SciPy, etc., can be used to perform interpolation or resampling operations so that the data has consistent time resolution.
[0173] Step S32: performing short-time Fourier transform on the residual mixed signal data to obtain mixed signal spectrum feature data;
[0174] In an embodiment of the present invention, a short-time Fourier transform (STFT) is applied to the residual mixed signal data to convert the signal into the time-frequency domain; the STFT divides the signal into short-time windows, performs a Fourier transform on each window, and obtains spectrum information; signal processing tools and libraries, such as NumPy, SciPy, etc., can be used to perform STFT calculations and select appropriate window sizes and overlap rates.
[0175] Step S33: performing wavelet time-frequency domain conversion on the non-stationary component data, thereby obtaining non-stationary component time domain data;
[0176] The embodiment of the present invention applies wavelet transform to non-stationary component data to convert it into wavelet time-frequency domain; wavelet transform decomposes the signal into wavelet coefficients of different scales and frequencies, revealing the changing characteristics of the signal in the time-frequency domain; wavelet library or related signal processing tools, such as PyWavelets, Wavelet Toolbox, etc., can be used to select appropriate wavelet basis functions and transformation methods for transformation.
[0177] Step S34: extracting the background noise pattern based on frequency and space from the mixed signal spectrum feature data, and assigning high weights to the stable and slowly changing noise parts through the attention mechanism, so as to obtain background noise weighted data;
[0178] The embodiments of the present invention perform frequency and spatial statistical analysis on the mixed signal spectrum feature data to extract the background noise pattern; statistics such as the mean and variance in the frequency and spatial dimensions can be calculated, or other background noise estimation methods can be applied; through the attention mechanism, a high weight is assigned to the smooth and slowly changing noise parts to highlight the influence of the background noise pattern.
[0179] Step S35: capturing the transient spike pattern based on the time dimension of the non-stationary component time domain data, thereby obtaining transient spike pattern data; assigning high weights to the sudden and paroxysmal noise components according to the transient spike pattern data, thereby obtaining transient noise weighted data;
[0180] The embodiments of the present invention analyze the time domain data of non-stationary components in the time dimension to capture transient spike patterns. Peak detection algorithms, waveform analysis methods, etc. can be used to identify and extract transient spike patterns. Based on the transient spike pattern data, high weights are assigned to sudden and paroxysmal noise components to highlight the impact of transient noise.
[0181] Step S36: embedding the transient noise weighted data and the background noise weighted data into scene semantics according to the pre-acquired scene priori data, thereby obtaining the spatiotemporal distribution characteristic data of the substation site noise;
[0182] The embodiments of the present invention collect and obtain scene prior data related to the substation site, including environmental information, equipment layout, working status, etc.; combine transient noise weighted data and background noise weighted data with scene prior data to perform scene semantic embedding; and use machine learning techniques, such as deep learning models or traditional feature engineering methods, to fuse data with scene prior information and extract spatiotemporal distribution feature data.
[0183] Step S37: Model and analyze the spatiotemporal distribution feature data according to the on-site environment and equipment layout information of the substation, so as to obtain high-order noise semantic representation data.
[0184] The embodiment of the present invention constructs corresponding models and algorithms according to the on-site environmental information and equipment layout of the substation; machine learning methods such as cluster analysis and principal component analysis can be used to model and analyze spatiotemporal distribution feature data and extract high-order noise semantic representation data; model training and optimization are performed to obtain more accurate noise characterization results.
[0185] The present invention normalizes the non-stationary component data and the residual mixed signal data in time so that they have a consistent time step. This helps to compare and analyze the data and provides temporal consistency. Secondly, the residual mixed signal data is subjected to spectrum analysis by short-time Fourier transform to obtain the spectrum feature data of the mixed signal. This provides information about the signal in the frequency domain, which helps further analysis and processing. Further, the non-stationary component data is subjected to wavelet time-frequency domain conversion to obtain the time domain data of the non-stationary component. This provides information about the signal in the time-frequency domain, which helps to capture the transient characteristics and change patterns of the signal. Background noise weighted data is obtained by extracting the background noise pattern and applying the attention mechanism to the mixed signal spectrum feature data. This makes the characteristics of the background noise prominent, and can assign high weights to the stable and slowly changing noise parts. By capturing the transient spike pattern of the non-stationary component time domain data and generating the transient noise weighted data, the sudden and paroxysmal noise components can be assigned high weights. This helps to capture the transient characteristics and sudden events in the noise. By embedding scene prior data, the transient noise weighted data and background noise weighted data are associated with the spatiotemporal distribution characteristics of the substation site noise. This provides an understanding and description of the spatiotemporal distribution of noise. Finally, based on the substation site environment and equipment layout information, the spatiotemporal distribution feature data is modeled and analyzed to obtain high-order noise semantic representation data. This provides a higher level of semantic understanding and description of noise. In summary, the entire step includes time standardization, frequency domain and time domain feature extraction, highlighting transient and background noise components, understanding of spatiotemporal distribution characteristics, and modeling analysis. These effects help improve the ability to understand and describe noise, and provide a more accurate and comprehensive basis for subsequent noise processing and analysis.
[0186] Preferably, step S37 includes the following steps:
[0187] Step S371: performing structured semantic representation encoding on the scene priori data, thereby obtaining scene semantic encoding data;
[0188] The embodiments of the present invention collect and organize scene prior data related to the substation site, including environmental information, equipment layout, working status, etc.; structure the scene prior data and represent it in a form suitable for processing by machine learning algorithms, such as using data structures such as vectors, matrices, or graphs; and use data processing tools and libraries, such as Pandas, NumPy, etc., to perform data conversion, standardization, and encoding operations.
[0189] Step S372: performing time series modeling on the spatiotemporal distribution feature data according to the scene semantic coding data, and extracting the key parameters of the burst, persistence and attenuation of the noise in time, so as to obtain the time domain characteristic data;
[0190] The embodiment of the present invention uses scene semantic coding data as input to perform time series modeling on spatiotemporal distribution feature data, and can use statistical models, time series models and other methods to extract key parameters of noise in time, such as the duration and decay rate of burst noise. Sliding window, peak detection and other technologies can be used to extract parameters; based on the extracted parameters, data describing the time domain characteristics of noise, such as the time distribution of burst noise, the persistence of noise, etc., can be obtained.
[0191] Step S373: frequency domain modeling is performed on the spatiotemporal distribution feature data according to the scene semantic coding data, and the frequency components and energy distribution characteristics of the noise are identified, thereby obtaining frequency domain characteristic data;
[0192] The embodiments of the present invention utilize scene semantic coding data to perform frequency domain modeling on spatiotemporal distribution feature data, and methods such as Fourier transform and wavelet transform may be used; the frequency components of noise may be identified, and the main frequency components of noise may be extracted through methods such as spectrum analysis and frequency filtering; the distribution characteristics of noise energy in the frequency domain may be analyzed, such as the frequency ranges in which the energy is concentrated, the shape of the spectrum, etc., and methods such as power spectrum density estimation and spectrum feature extraction may be used.
[0193] Step S374: spatially modeling the spatiotemporal distribution feature data according to the scene semantic coding data, and analyzing the propagation and diffusion of noise in space and the spatial pattern of hot spots, thereby obtaining spatial characteristic data;
[0194] The embodiments of the present invention utilize scene semantic coding data to perform spatial modeling on spatiotemporal distribution feature data, and methods such as spatial interpolation and cluster analysis may be used; the propagation and diffusion characteristics of noise in space may be analyzed, and methods such as propagation models and distance attenuation models may be used to understand the laws of noise propagation; the spatial patterns of noise hot spots may be analyzed, and methods such as spatial clustering, heat maps, and spatial density estimation may be used to find high-density areas of noise.
[0195] Step S375: Perform high-order semantic analysis on the time domain characteristic data, the frequency domain characteristic data, and the spatial characteristic data, so as to obtain high-order noise semantic representation data.
[0196] The embodiments of the present invention integrate time domain characteristic data, frequency domain characteristic data and spatial characteristic data, and conduct a comprehensive analysis in combination with scene semantic coding data; machine learning methods, such as deep learning models, clustering analysis, etc., can be used to train and analyze the characteristic data to obtain high-order noise semantic representation data; high-order semantic analysis can include pattern recognition, association analysis, feature extraction and other technologies to reveal deeper semantic information in the noise data; combined with domain expertise and experience, the high-order noise semantic representation data is interpreted and interpreted to obtain a deeper understanding of noise characteristics.
[0197] The present invention associates noise data with semantic information of the scene by generating scene semantic coding data. By performing structured coding on the scene data, important semantic features of the scene can be extracted, providing a basis for subsequent analysis and modeling. This helps to better understand and describe the noise data, and provides contextual information related to the scene in subsequent steps. Through time series modeling and key parameter extraction, the temporal variation pattern and characteristics of the noise can be deeply understood. The acquisition of time domain characteristic data helps to analyze the temporal variation of noise data, including key features such as burstiness, persistence and attenuation. This provides important time domain information for subsequent noise analysis and processing. Frequency domain modeling and feature identification can understand the characteristics of noise data in the frequency domain. By analyzing the frequency components and energy distribution, the frequency domain features in the noise data, such as the main frequency components, the frequency distribution range, etc., can be identified. This helps to perform frequency domain analysis and processing on the noise, providing a basis for subsequent noise feature extraction and identification. Through spatial modeling and analysis, the spatial distribution and propagation characteristics of the noise can be understood. The spatial pattern analysis of the propagation and diffusion of noise and the hot spot area helps to determine the spatial characteristics of the noise, such as the propagation path, the diffusion range and the distribution of the hot spot area. This provides important spatial information for subsequent noise analysis and positioning. Through high-order semantic analysis, more complex patterns and associations in time, frequency, and spatial characteristic data can be deeply explored. The acquisition of high-order noise semantic representation data helps to more comprehensively understand the semantic meaning and feature representation of noise data, thereby providing deeper noise analysis and inference capabilities. In summary, the above steps include association with scene semantics, analysis and extraction of time, frequency, and spatial characteristics, and acquisition of high-order noise semantic representation. These effects provide the basis and support for the in-depth understanding, precise description, and further application of noise data.
[0198] Preferably, step S4 comprises the following steps:
[0199] Step S41: designing an unsupervised generative adversarial network framework based on the high-order noise semantic representation data, wherein the generator learns to generate a high-order noise semantic sequence, and the discriminator discriminates the authenticity of the high-order noise semantic sequence and the high-order noise semantic representation data;
[0200] The embodiment of the present invention designs a framework of a generative adversarial network (GAN), including a generator and a discriminator; the input of the generator is high-order noise semantic representation data, and learns to generate high-order noise semantic sequences, and models such as recurrent neural networks (such as LSTM, GRU) can be used for sequence generation; the role of the discriminator is to judge the difference between the generated high-order noise semantic sequence and the real high-order noise semantic representation data, and models such as convolutional neural networks and multi-layer perceptrons can be used for discrimination; by training the generator and the discriminator, the generator gradually learns to generate realistic high-order noise semantic sequences, and the discriminator is also continuously optimized to better distinguish between real data and generated data.
[0201] Step S42: performing semantic latent encoding on the high-order noise semantic representation data, thereby obtaining noise semantic latent code data;
[0202] The embodiments of the present invention perform semantic latent encoding on high-order noise semantic representation data to convert it into a latent semantic code representation; an encoder model, such as an autoencoder, a variational autoencoder, etc., can be used to map the high-order noise semantic representation data to a low-dimensional latent code space; the latent code can represent a compact, low-dimensional representation of the high-order noise semantic features, and has the function of retaining key semantic information.
[0203] Step S43: using the noise semantic latent code data as the starting condition to train the generator and the discriminator, thereby obtaining the noise semantic time series trajectory data;
[0204] The embodiment of the present invention uses the noise semantic latent code data as the input of the generator, trains the generator model so that it can generate noise semantic time series trajectory data; at the same time, uses the noise semantic latent code data as the input of the discriminator to discriminate the authenticity of the generated noise semantic time series trajectory data; iteratively trains the generator and the discriminator, and continuously optimizes the generation ability of the generator and the discrimination ability of the discriminator by comparing the noise semantic time series trajectory data generated by the generator with the real data.
[0205] Step S44: performing actual scene decoding on the noise semantic time series trajectory data to obtain collaborative tracking scene data.
[0206] The embodiment of the present invention inputs the generated noise semantic time series trajectory data into a decoder model, and the decoder model converts the noise semantic time series trajectory data into corresponding collaborative tracking scene data; the decoder can be a reverse operation model, mapping the generated noise semantic time series trajectory data back to high-order noise semantic representation data, and further decoding it to obtain the collaborative tracking scene data.
[0207] The present invention uses the design of an unsupervised generative adversarial network framework, and the generator can learn to generate high-order noise semantic sequences, simulate and approximate the high-order semantic features of real noise data. The discriminator is used to discriminate the difference between the generated high-order noise semantic sequence and the real high-order noise semantic representation data, prompting the generator to continuously optimize the generated results. Such a framework helps to improve the generation quality and authenticity of noise data, and provides more reliable generation results for noise data processing and application in subsequent steps. By latent encoding of high-order noise semantic representation data, complex semantic features can be converted into more compact and easier to handle latent representations. The generation of noise semantic latent data helps to reduce the dimension and redundancy of data, and improve the expression efficiency and processing effect of noise data. This provides a more convenient and efficient way for data transmission, storage and processing in subsequent steps. By training with noise semantic latent data as the starting condition, the generator can generate time series trajectory data consistent with noise semantic features. This helps to simulate and generate noise semantic time series trajectory data with similar characteristics to actual noise data. The generated noise semantic time series trajectory data can provide richer and more accurate time series information for noise data analysis, processing and application in subsequent steps. By performing actual scene decoding on the noise semantic time series trajectory data, the noise data can be converted into tracking data related to the actual scene. This helps to eliminate the interference of noise on the scene data and improve the ability to understand and describe the scene. The obtained collaborative tracking scene data can be used for subsequent scene analysis, decision-making and application, providing more accurate and reliable data support for solving practical problems. In summary, the above steps include generating high-quality noise data, reducing data dimensions and redundancy, simulating the time series trajectory of noise, and providing data related to the actual scene. These effects provide a more reliable, efficient and accurate method and basis for the generation, processing and application of noise data.
[0208] Preferably, step S5 comprises the following steps:
[0209] Step S51: simulating the propagation of noise in the three-dimensional scene of the substation according to the collaborative tracking scene data by using the Monte Carlo method, thereby obtaining fuzzy noise simulation data;
[0210] The embodiment of the present invention uses the Monte Carlo method to simulate noise propagation based on collaborative tracking scene data and a three-dimensional scene model of a substation; the Monte Carlo method is a random simulation method that estimates the probability distribution of numerical solutions or event occurrences through random sampling and statistical analysis; during the simulation process, factors such as the location, intensity, propagation path, and environmental parameters of the noise source are considered to simulate the propagation process of noise in the three-dimensional scene; based on the simulation results, fuzzy noise simulation data is generated, which includes the propagation characteristics of noise in the three-dimensional scene of the substation.
[0211] Step S52: quantifying and classifying the fuzziness degree of noise propagation according to the fuzzy noise simulation data, thereby obtaining noise fuzziness degree data;
[0212] The embodiments of the present invention quantify and classify the degree of fuzziness of noise propagation based on fuzzy noise simulation data; image processing and signal processing methods, such as fuzziness index, spectrum analysis, time domain feature extraction, etc., can be used to analyze the fuzzy noise data; according to the analysis results, the degree of noise fuzziness can be quantified, for example, using numerical indicators or scores to describe the degree of fuzziness of the noise; the degree of noise fuzziness can also be classified, and the fuzzy noise data can be divided into different fuzziness levels, such as weak fuzziness, medium fuzziness, strong fuzziness, etc.
[0213] Step S53: performing fuzziness reduction based on fuzzy set theory on the fuzzy noise simulation data according to the noise fuzziness degree data to obtain a confidence noise spectrum;
[0214] The embodiment of the present invention uses fuzzy set theory to reduce the fuzziness of fuzzy noise simulation data based on noise fuzziness degree data; fuzzy set theory is a mathematical tool for processing fuzzy information, which can introduce uncertainty and fuzziness into data processing; according to the noise fuzziness degree data, a fuzzy set is constructed and a membership function is defined to describe the fuzziness degree of the noise; the fuzzy noise simulation data is reduced in fuzziness using fuzzy set operations and fuzzy reasoning rules to obtain a confidence noise spectrum; the confidence noise spectrum can represent the confidence or credibility of the noise, reflecting the uncertainty and fuzziness of the noise distribution.
[0215] Step S54: deploy the confidence noise map to the edge device, and perform edge-parallel separation of independent noise source components to obtain the final separation result data.
[0216] The embodiment of the present invention deploys the generated confidence noise spectrum to the edge device, so that it can perform real-time noise source separation processing; the edge device can be an embedded system, server, etc. with certain computing capabilities; on the edge device, the confidence noise spectrum is used to separate independent noise source components to separate different noise sources; edge parallel processing can improve the efficiency and real-time performance of noise source separation by processing multiple noise source components in parallel and performing separation operations at the same time; the final separation result data represents the separation effect of each independent noise source, which can be used for further analysis, processing or application.
[0217] The present invention uses the Monte Carlo method to simulate the noise propagation simulation, which can simulate the noise propagation process in the three-dimensional scene of the substation. This helps to understand the diffusion mode and path of noise in the scene, thereby providing accurate reference data for noise analysis and processing in subsequent steps. The obtained fuzzy noise simulation data can be used to further study the law of noise propagation and quantitatively analyze the noise. By quantifying and classifying the fuzzy noise simulation data, the fuzziness degree of noise propagation can be objectively evaluated and described. This helps to quantitatively analyze the distribution and intensity changes of noise in the scene, and provides a basis for noise processing in subsequent steps. The obtained noise fuzziness degree data can be used for the formulation of noise control strategies and noise impact assessment. Fuzziness reduction based on fuzzy set theory can process fuzzy noise simulation data according to noise fuzziness degree data, reduce or eliminate fuzziness, and obtain a clearer and more reliable confidence noise spectrum. This helps to improve the quality and accuracy of the noise spectrum, and provides a more reliable data basis for noise separation and analysis in subsequent steps. Deploying the confidence noise spectrum to the edge device can achieve real-time noise source separation and analysis, and reduce data transmission and processing delays. By using edge parallelism, the noise source can be separated into components independently, and the noise signal can be separated from other signals. This helps to extract the target signal and reduce the impact of noise on system performance, providing more accurate and reliable results for subsequent signal processing and applications. In summary, the above steps include simulating noise propagation, quantifying and classifying the degree of noise ambiguity, reducing ambiguity, and separating noise sources in real time. These effects provide a more accurate, reliable and efficient method and basis for the analysis, processing and application of noise data.
[0218] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.
Claims
1. A substation noise matching and separation method, characterized in that: The following steps are involved: Step S1: using a microphone array system to collect noise signals at the substation site to obtain original time domain array data; Performing direction of arrival estimation on the original time-domain array data to obtain direction of arrival estimation data; Step S2: obtaining non-stationary component data and residual mixed signal data according to the array structure of the microphone array system and the direction of arrival estimation data; Step S3: performing 3D scene analysis on the non-stationary component data and the residual mixed signal data to obtain spatiotemporal distribution feature data; Based on the substation's on-site environment and equipment layout information, the temporal and spatial distribution feature data is modeled and analyzed to obtain high-order noise semantic representation data; Step S4: Track the noise change trajectory in real time based on the high-order noise semantic representation data to obtain collaborative tracking scene data; Step S5: Generate a noise spectrum based on the collaborative tracking scene data to obtain a confidence noise spectrum; perform edge-parallel separation of independent noise source components based on the confidence noise spectrum to obtain final separation result data.
2. A substation noise matching and separation method according to claim 1, characterized in that: The step S1 comprises the following steps: Integrate and deploy MEMS microphone chip pairs according to a preset array structure to form a microphone array system; The noise signal at the substation site is collected using a microphone array system to obtain the original time domain array data; The impact of non-ideal scenes on the microphone array system is estimated according to preset ideal scene data, and array response tests are performed to obtain scene impact estimation data; Perform array response offset analysis based on scene impact estimation data to obtain array response drift data; Performing periodic drift calibration on original time-domain array data according to array response drift data to obtain time-domain array calibration data; Direction of arrival estimation is performed on the time domain array calibration data to obtain direction of arrival estimation data.
3. A substation noise matching and separation method according to claim 2, characterized in that: Obtaining the scene impact estimation data includes the following steps: Obtain actual scenario parameter data of substation; Extracting deviations from actual scene parameter data according to preset ideal scene data to obtain scene parameter deviation data, wherein the scene parameter deviation data includes electromagnetic field interference data, medium non-uniformity data, and array structure deviation data; Conduct electromagnetic compatibility tests on the connection lines of the microphone array system, establish a response model of the circuit under different electromagnetic field intensities, and obtain a magnetic field interference response model; The impact is estimated based on the electromagnetic field interference data and the magnetic field interference response model to obtain the electromagnetic field interference impact data; The propagation process of environmental parameter disturbance sound is simulated according to the pre-acquired environmental parameter data to obtain disturbance degree data; The influence of medium non-uniformity data is estimated according to the disturbance degree data to obtain medium influence data; The audio phase detection technology is used to calibrate the deviation between the microphone phase center and the theoretical position of the array structure deviation data, and the influence on the wave arrival angle and wavefront curvature is calculated to obtain the array deviation influence data; The electromagnetic field interference impact data, medium impact data and array deviation impact data are combined into scene impact estimation data.
4. A substation noise matching and separation method according to claim 3, characterized in that: The step of performing direction of arrival estimation on the time domain array calibration data to obtain direction of arrival estimation data comprises: Divide the time domain array calibration data into blocks of fixed time length to obtain array time block data, wherein the data blocks overlap with each other for a fixed time length; Transforming the array time block data from the time domain to the frequency domain, and obtaining the complex spectrum value of each frequency component in the data block to obtain component complex spectrum value data; Arranging the complex spectral values of the microphones in the same time block into a matrix according to the component complex spectral value data to obtain an array complex spectral value matrix, wherein the rows of the matrix correspond to different microphones and the columns correspond to different frequency components; According to the orientation of each array element in the microphone array system, the manifold vector is calculated at different elevation angles and azimuth angles to obtain manifold vector data; The MUSIC subspace algorithm is used to estimate the direction of each frequency component in the time block based on the manifold vector data and the array complex spectrum value matrix to obtain the wave direction estimation data.
5. A substation noise matching and separation method according to claim 1, characterized in that: The step S2 comprises the following steps: Adaptive multi-layer tensor iteration is performed according to the geometric structure parameters of the deployed microphone array system and the wave direction estimation data to obtain the refined array manifold data; Divide the frequency band covered by the fine array manifold data into sub-bands to obtain a sub-band set; The filter level is designed according to the sub-band set to obtain a reconfigurable hierarchical filter grid; Using hierarchical filter grids to filter and denoise the fine array manifold data, and obtain filtering result data; The filtered result data is subjected to compressed sensing mesh coding to obtain non-stationary component data and residual mixed signal data.
6. A substation noise matching and separation method according to claim 5, characterized in that: The method of performing compressed sensing mesh coding on the filtering result data to obtain non-stationary component data and residual mixed signal data includes the following steps: Perform wavelet transform on the filtering result data to obtain coefficient representation data in the wavelet time-frequency domain; Perform characteristic statistics of probability distribution and energy concentration degree on coefficient representation data to obtain key coefficient data; Designing a mesh coding structure according to key coefficient data to obtain mesh coding structure data; Encoding and compressing the coefficient representation data according to the mesh coding structure data to obtain compressed coding data; The compressed coded data is separated to obtain non-stationary component data and residual mixed signal data, wherein the non-stationary component data retains the transient noise part, and the residual mixed signal data retains the stationary background noise part.
7. A substation noise matching and separation method according to claim 1, characterized in that: The step S3 comprises the following steps: Performing time standardization on the non-stationary component data and the residual mixed signal data to obtain the non-stationary component data and the residual mixed signal data with consistent time steps; Performing short-time Fourier transform on the residual mixed signal data to obtain mixed signal spectrum feature data; Perform wavelet time-frequency domain conversion on the non-stationary component data to obtain the non-stationary component time domain data; The background noise pattern is extracted based on frequency and space from the mixed signal spectrum feature data, and the stable and slowly changing noise parts are assigned high weights through the attention mechanism to obtain the background noise weighted data; The transient spike pattern of the non-stationary component time domain data is captured based on the time dimension to obtain the transient spike pattern data; according to the transient spike pattern data, the sudden and paroxysmal noise components are assigned high weights to obtain the transient noise weighted data; According to the pre-acquired scene prior data, the transient noise weighted data and the background noise weighted data are embedded with scene semantics to obtain the spatiotemporal distribution characteristic data of the substation site noise; According to the on-site environment and equipment layout information of the substation, the spatiotemporal distribution characteristic data is modeled and analyzed to obtain high-order noise semantic representation data.
8. A substation noise matching and separation method according to claim 7, characterized in that: The modeling and analysis of the spatiotemporal distribution feature data according to the on-site environment and equipment layout information of the substation to obtain high-order noise semantic representation data includes the following steps: Encode the scene prior data into structured semantic representation to obtain scene semantic encoding data; The temporal and spatial distribution feature data is modeled according to the scene semantic coding data, and the key parameters of the noise burst, persistence and attenuation in time are extracted to obtain the time domain characteristic data; The spatiotemporal distribution feature data is modeled in the frequency domain according to the scene semantic coding data, and the frequency components of the noise and the characteristics of the energy distribution are identified to obtain the frequency domain characteristic data; Based on the scene semantic coding data, the spatiotemporal distribution feature data is spatially modeled, and the propagation and diffusion of noise in space and the spatial pattern of hot spots are analyzed to obtain spatial characteristic data; High-order semantic analysis is performed on the time domain characteristic data, frequency domain characteristic data and spatial characteristic data to obtain high-order noise semantic representation data.
9. A substation noise matching and separation method according to claim 1, characterized in that: The step S4 comprises the following steps: An unsupervised generative adversarial network framework is designed based on high-order noise semantic representation data, in which the generator learns to generate high-order noise semantic sequences, and the discriminator determines the authenticity of high-order noise semantic sequences compared with high-order noise semantic representation data; Performing semantic latent encoding on the high-order noise semantic representation data to obtain noise semantic latent data; The noise semantic latent data is used as the starting condition to train the generator and the discriminator to obtain the noise semantic time series trajectory data; The noisy semantic temporal trajectory data is decoded in actual scenes to obtain collaborative tracking scene data.
10. A substation noise matching and separation method according to claim 1, characterized in that: The step S5 comprises the following steps: The Monte Carlo method is used to simulate the propagation of noise in the three-dimensional scene of the substation based on the collaborative tracking scene data to obtain fuzzy noise simulation data; The fuzzy degree of noise propagation is quantified and classified according to the fuzzy noise simulation data to obtain the noise fuzzy degree data; According to the noise fuzziness degree data, the fuzzy noise simulation data is subjected to fuzzy set theory-based fuzzy degree reduction to obtain a confidence noise spectrum; The confidence noise map is deployed to the edge device, and the independent noise source components are separated in parallel at the edge to obtain the final separation result data.
Citation Information
Cited By
Image noise reduction method and device based on block matching and storage medium
CN119313587A
Noise model construction method based on machine learning method
CN119649841A
Multi-node edge induction sampling system and method based on environmental monitoring
CN119779470A
Primary and secondary fusion centralized DTU equipment intelligent simulation test method and system
CN119782873A
Electromechanical equipment noise reduction method based on machine learning
CN119939123A
