A method for locating low-frequency noise sources in substations

Abstract

The present invention relates to the technical field of power systems, and in particular to a method for locating low-frequency noise sources in substations. The method comprises the following steps: obtaining a three-dimensional environmental model of each structure in the substation, obtaining a geometric topological relationship matrix of the three-dimensional environmental model; performing on-site measurement of the substation to obtain noise measurement data; performing outer convolution fusion on the noise images between adjacent units according to the geometric topological relationship matrix and local noise image data, and iteratively propagating them to the entire substation area to generate initial sound source distribution estimation data for the substation as a whole. The present invention utilizes acoustic energy flow analysis to identify the source position, and combines direction estimation and trajectory reconstruction to separate and track stationary sources and mobile sources. It can accurately locate the stationary source, and at the same time obtain the motion trajectory of the mobile source, thereby improving separation efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method for locating a low-frequency noise source in a substation. Background Art

[0002] Substations are a crucial component of the power system, converting high-voltage electricity to low-voltage energy and vice versa to meet the needs of power transmission, distribution, and supply. They are a critical link in the power system, enabling the transmission, distribution, and conversion of electrical energy. In a power system, electrical energy must be transported from power plants to users via transmission lines. Due to the long distances over which electricity must be transmitted, it is typically boosted to a higher voltage during transmission to reduce energy loss and improve transmission efficiency. This requires substations to transform the voltage and current of the electricity.

[0003] Low-frequency noise in substations refers to low-frequency noise signals generated during substation operation. Compared to high-frequency noise, low-frequency noise has a lower frequency range, typically below 20 Hz. This noise has a longer wavelength and strong propagation ability, allowing it to penetrate buildings and obstacles. Low-frequency noise in substations can impact the surrounding environment and personnel, causing discomfort and disrupting the lives and work of nearby residents. Localizing low-frequency noise sources in substations is the process of determining the specific location of low-frequency noise within or around the substation. However, substations contain a variety of power equipment, and noise sources are distributed across multiple locations. This creates a complex situation with multiple, dispersed sound sources, often leading to source dispersion issues. Furthermore, the complex structure of the substation area generates significant echoes and multipath effects, which can easily confuse direct sound signals, resulting in significant echo and interference issues for noise source location. Furthermore, some equipment, such as circuit breakers, generates mobile sound sources during operation. Their transient and ringing characteristics make noise source location difficult to capture and track. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a method for locating a low-frequency noise source in a substation to solve at least one of the above technical problems.

[0005] To achieve the above object, a method for locating a low-frequency noise source in a substation includes the following steps:

[0006] Acquire a three-dimensional environmental model of each structure in the substation and obtain a topological relationship matrix of geometric bodies in the three-dimensional environmental model; perform on-site measurement of the substation to obtain noise measurement data; and map the noise measurement data to the three-dimensional environmental model based on the measurement location, perform registration, and obtain local noise image data;

[0007] Performing outer convolution fusion on the noise images between adjacent units according to the geometric topological relationship matrix and the local noise image data, and iteratively propagating them to the entire substation area to generate initial sound source distribution estimation data for the entire substation;

[0008] Separate the direct sound signal and multipath echo components based on the initial sound source distribution estimation data to generate echo decoupling data;

[0009] Perform dimensionality compression on the echo decoupling data and extract the eigenmodes of the low-order scattered field to obtain the eigenmode set of the scattered field; perform parameter estimation on the field mode of the eigenmode set of the scattered field to obtain the eigenscattering cloud map data;

[0010] The intrinsic scattering cloud image data is used to identify the acoustic energy flow around the center and to match the noise source position to obtain the potential noise source position data. Based on the potential noise source position data, the instantaneous motion direction of the source point is locked and the motion trajectory of the source point is reconstructed to obtain the noise source motion trajectory data. Based on the noise source motion trajectory data and the potential noise source position data, the position information of the stationary source and the moving source is distinguished and integrated to obtain the moving source path data.

[0011] The modal contribution analysis of the dynamic source path data is performed based on the vibration modes of the substation equipment, and the radiation modes of the stationary and moving noise sources are separated and projected into the scalar sound pressure field to obtain the separated noise source distribution map.

[0012] The present invention obtains a three-dimensional environmental model of each structure in the substation using laser scanning or photogrammetry. These technologies can capture the building's geometric shape and structural information, constructing an accurate three-dimensional model. Next, the topological relationships and boundary conditions of the geometric bodies are extracted from the 3D environmental model, which helps establish the connectivity and boundary constraints between the structures. Using the geometric body topological relationship matrix and local noise image data, the noise images of adjacent units are fused using an outer product convolution fusion method. This process can be viewed as propagating and integrating local noise information, ultimately generating an initial sound source distribution estimate for the entire substation area. This estimate serves as the starting point for subsequent noise source location and isolation. Using the generalized coordinated filtering algorithm, the direct sound signal and multipath echo components are separated based on the initial sound source distribution estimate. Simultaneously, the influence of multipath noise is removed using a manifold learning method to generate echo decoupling data. This step aims to reduce the interference of multipath echoes on noise source location and isolation, thereby improving the accuracy of location and isolation. Dimensionality reduction techniques such as principal component analysis (PCA) can be used to reduce the dimensionality of the echo decoupling data. Next, the low-order eigenmodes of the scattered field are extracted from the compressed data. These eigenmodes reflect the characteristics of the scattered field. Parameters of the scattered field eigenmode set are estimated using a nonlinear regression model to generate eigenscattering cloud data. This step aims to extract characteristic information of the scattered field, providing a more accurate data basis for locating and isolating noise sources. The eigenscattering cloud data is used to identify the center of the acoustic energy flow and to correlate the noise source location with the potential noise source location. Based on the potential noise source location data, the instantaneous motion direction of the source point is determined, and the source trajectory is reconstructed to obtain the noise source motion trajectory data. Furthermore, based on the noise source motion trajectory data and the potential noise source location data, the location information of stationary and mobile sources is distinguished and integrated to obtain the moving source path data. This step aims to accurately locate and track the location and trajectory of the noise source, providing detailed dynamic information. Based on the vibration modal analysis of the substation equipment, modal contribution analysis is performed on the moving source path data. By analyzing the contribution of the moving source path data in different vibration modes, the radiation patterns of stationary and mobile sources can be separated. These modes are then projected into the scalar sound pressure field to obtain a distribution map of separated sources. The purpose of this step is to distinguish the radiation patterns of stationary and moving sources based on the characteristics of the vibration modes, and to present their distribution in space, providing intuitive noise source location and separation results. In summary, the detailed explanation of these steps includes obtaining a three-dimensional environmental model, establishing geometric topological relationships, fusing local noise images, separating direct sound and echo components, removing the influence of multipath noise, dimensionality compression and eigenmode extraction, noise source location identification and motion trajectory reconstruction, dynamic source path differentiation, modal contribution analysis, and noise source distribution maps after separation.The combined application of these steps can achieve precise positioning and effective separation of noise sources, providing useful information and decision support for noise control and management.

[0013] The present invention also provides a low-frequency noise source locating device, comprising:

[0014] The 3D environment modeling module is used to obtain a 3D environment model of each structure in the substation and obtain a topological relationship matrix of geometric bodies in the 3D environment model; perform on-site measurements of the substation to obtain noise measurement data; and map the noise measurement data to the 3D environment model based on the measurement location and perform registration to obtain local noise image data.

[0015] A noise measurement module is used to perform outer convolution fusion on the noise images between adjacent units based on the geometric topological relationship matrix and the local noise image data, and iteratively propagate them to the entire substation area to generate initial sound source distribution estimation data for the entire substation;

[0016] A noise distribution estimation module is used to separate the direct sound signal and the multipath echo components based on the initial sound source distribution estimation data to generate echo decoupling data;

[0017] The echo decoupling module is used to perform dimensionality compression on the echo decoupling data and extract the eigenmodes of the low-order scattered field to obtain the eigenmode set of the scattered field; the parameters of the field mode of the eigenmode set of the scattered field are estimated to obtain the eigenscattering cloud map data;

[0018] The dynamic source path analysis module is used to identify the acoustic energy flow around the center of the intrinsic scattering cloud data and match the noise source position to obtain potential noise source position data; based on the potential noise source position data, the instantaneous motion direction of the source point is locked and the motion trajectory of the source point is reconstructed to obtain noise source motion trajectory data; based on the noise source motion trajectory data and the potential noise source position data, the position information of the stationary source and the moving source is distinguished and integrated to obtain dynamic source path data;

[0019] The modal contribution analysis module is used to perform modal contribution analysis on the dynamic source path data based on the vibration modes of the substation equipment, separate the radiation modes of the stationary and moving noise sources, project them into the scalar sound pressure field, and obtain the separated noise source distribution map. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0021] Figure 1 Schematic diagram of the steps of the method for locating the low-frequency noise source of a substation according to the present invention;

[0022] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0023] Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0024] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative work are within the scope of protection of the present invention.

[0025] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0026] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0027] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for locating a low-frequency noise source in a substation, the method comprising the following steps:

[0028] Acquire a three-dimensional environmental model of each structure in the substation and obtain a topological relationship matrix of geometric bodies in the three-dimensional environmental model; perform on-site measurement of the substation to obtain noise measurement data; and map the noise measurement data to the three-dimensional environmental model based on the measurement location, perform registration, and obtain local noise image data;

[0029] Performing outer convolution fusion on the noise images between adjacent units according to the geometric topological relationship matrix and the local noise image data, and iteratively propagating them to the entire substation area to generate initial sound source distribution estimation data for the entire substation;

[0030] Separate the direct sound signal and multipath echo components based on the initial sound source distribution estimation data to generate echo decoupling data;

[0031] Perform dimensionality compression on the echo decoupling data and extract the eigenmodes of the low-order scattered field to obtain the eigenmode set of the scattered field; perform parameter estimation on the field mode of the eigenmode set of the scattered field to obtain the eigenscattering cloud map data;

[0032] The intrinsic scattering cloud image data is used to identify the acoustic energy flow around the center and to match the noise source position to obtain the potential noise source position data. Based on the potential noise source position data, the instantaneous motion direction of the source point is locked and the motion trajectory of the source point is reconstructed to obtain the noise source motion trajectory data. Based on the noise source motion trajectory data and the potential noise source position data, the position information of the stationary source and the moving source is distinguished and integrated to obtain the moving source path data.

[0033] The modal contribution analysis of the dynamic source path data is performed based on the vibration modes of the substation equipment, and the radiation modes of the stationary and moving noise sources are separated and projected into the scalar sound pressure field to obtain the separated noise source distribution map.

[0034] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of a method for locating a low-frequency noise source in a substation according to the present invention. In this example, the method for locating a low-frequency noise source in a substation includes the following steps:

[0035] Obtaining a three-dimensional environmental model of each structure in the substation, obtaining a geometric topological relationship matrix in the model by obtaining the three-dimensional environmental model; performing on-site measurement of the substation to obtain noise measurement data; mapping the noise measurement data into blocks based on the measurement location in the three-dimensional environmental model, and performing registration to obtain local noise image data;

[0036] In embodiments of the present invention, 3D model data of substation structures can be acquired using tools such as laser scanners, photogrammetry, or CAD software. The 3D environmental model is processed to extract geometric topological relationships and boundary condition information, such as the connectivity and spatial positional relationships between structures. Noise measurements are then performed in the actual substation scene to acquire noise data. The measurement locations are then registered with the 3D environmental model, and the noise data is correlated with the model at the corresponding location to obtain local noise image data.

[0037] Performing outer convolution fusion on the noise images between adjacent units according to the geometric topological relationship matrix and the local noise image data, and iteratively propagating them to the entire substation area to generate initial sound source distribution estimation data for the entire substation;

[0038] This embodiment of the present invention utilizes a geometric topological relationship matrix to perform outer convolution on the local noise images between adjacent cells, fusing the noise data from different cells to obtain an initial sound source distribution estimate. This initial sound source distribution estimate is propagated throughout the substation area through an iterative propagation algorithm, updating the sound source estimate for each cell until convergence, resulting in an estimate of the sound source distribution for the entire substation.

[0039] Separate the direct sound signal and multipath echo components based on the initial sound source distribution estimation data to generate echo decoupling data;

[0040] This embodiment of the present invention utilizes a generalized coordinated filtering algorithm to process initial sound source distribution estimation data, separating the direct sound signal from the multipath echo components. A manifold learning method is then applied to model and learn the multipath echo components, removing the influence of multipath noise and generating echo decoupling data.

[0041] Perform dimensionality compression on the echo decoupling data and extract the eigenmodes of the low-order scattered field to obtain the eigenmode set of the scattered field; perform parameter estimation on the field mode of the eigenmode set of the scattered field to obtain the eigenscattering cloud map data;

[0042] This embodiment of the present invention performs dimensionality compression on echo decoupling data, reducing the data dimension and reducing redundant information. Appropriate methods are applied to extract the eigenmodes of the low-order scattered field, generating a set of scattered field eigenmodes. A nonlinear regression model is used to estimate the parameters of the scattered field eigenmode set, obtaining the parameters of the field modes. Based on the parameter estimation results, eigenscattering cloud data is generated for subsequent analysis and processing.

[0043] The intrinsic scattering cloud image data is used to identify the acoustic energy flow around the center and to match the noise source position to obtain the potential noise source position data. Based on the potential noise source position data, the instantaneous motion direction of the source point is locked and the motion trajectory of the source point is reconstructed to obtain the noise source motion trajectory data. Based on the noise source motion trajectory data and the potential noise source position data, the position information of the stationary source and the moving source is distinguished and integrated to obtain the moving source path data.

[0044] The present invention analyzes intrinsic scattering cloud data to identify the central region surrounding the acoustic energy flow, which may be a potential noise source. The potential noise source's location is aligned with the center of the acoustic energy flow, determining its position data. Based on the potential noise source's location data, the source's instantaneous motion direction is inferred. The source's motion trajectory data is reconstructed based on the source's instantaneous motion direction and time information. Based on the noise source's motion trajectory data and the potential noise source's location data, the location information of stationary and moving sources is distinguished and integrated to obtain moving source path data.

[0045] The modal contribution analysis of the dynamic source path data is performed based on the vibration modes of the substation equipment, and the radiation modes of the stationary and moving noise sources are separated and projected into the scalar sound pressure field to obtain the separated noise source distribution map.

[0046] This embodiment of the present invention analyzes the vibration modes of substation equipment and determines the contribution of each mode. Dynamic source path data is correlated with the contributions of the vibration modes to determine the contribution of the dynamic source path to the vibration modes. Based on the dynamic source path data and modal contributions, the radiation patterns of stationary and moving noise sources are separated. The separated noise source radiation patterns are projected onto the scalar sound pressure field to obtain a source distribution map of the noise sources.

[0047] The present invention obtains a three-dimensional environmental model of each structure in the substation using laser scanning or photogrammetry. These technologies can capture the building's geometric shape and structural information, constructing an accurate three-dimensional model. Next, the topological relationships and boundary conditions of the geometric bodies are extracted from the 3D environmental model, which helps establish the connectivity and boundary constraints between the structures. Using the geometric body topological relationship matrix and local noise image data, the noise images of adjacent units are fused using an outer product convolution fusion method. This process can be viewed as propagating and integrating local noise information, ultimately generating an initial sound source distribution estimate for the entire substation area. This estimate serves as the starting point for subsequent noise source location and isolation. Using the generalized coordinated filtering algorithm, the direct sound signal and multipath echo components are separated based on the initial sound source distribution estimate. Simultaneously, the influence of multipath noise is removed using a manifold learning method to generate echo decoupling data. This step aims to reduce the interference of multipath echoes on noise source location and isolation, thereby improving the accuracy of location and isolation. Dimensionality reduction techniques such as principal component analysis (PCA) can be used to reduce the dimensionality of the echo decoupling data. Next, the low-order eigenmodes of the scattered field are extracted from the compressed data. These eigenmodes reflect the characteristics of the scattered field. Parameters of the scattered field eigenmode set are estimated using a nonlinear regression model to generate eigenscattering cloud data. This step aims to extract characteristic information of the scattered field, providing a more accurate data basis for locating and isolating noise sources. The eigenscattering cloud data is used to identify the center of the acoustic energy flow and to correlate the noise source location with the potential noise source location. Based on the potential noise source location data, the instantaneous motion direction of the source point is determined, and the source trajectory is reconstructed to obtain the noise source motion trajectory data. Furthermore, based on the noise source motion trajectory data and the potential noise source location data, the location information of stationary and mobile sources is distinguished and integrated to obtain the moving source path data. This step aims to accurately locate and track the location and trajectory of the noise source, providing detailed dynamic information. Based on the vibration modal analysis of the substation equipment, modal contribution analysis is performed on the moving source path data. By analyzing the contribution of the moving source path data in different vibration modes, the radiation patterns of stationary and mobile sources can be separated. These modes are then projected into the scalar sound pressure field to obtain a distribution map of separated sources. The purpose of this step is to distinguish the radiation patterns of stationary and moving sources based on the characteristics of the vibration modes, and to present their distribution in space, providing intuitive noise source location and separation results. In summary, the detailed explanation of these steps includes obtaining a three-dimensional environmental model, establishing geometric topological relationships, fusing local noise images, separating direct sound and echo components, removing the influence of multipath noise, dimensionality compression and eigenmode extraction, noise source location identification and motion trajectory reconstruction, dynamic source path differentiation, modal contribution analysis, and noise source distribution maps after separation.The combined application of these steps can achieve precise positioning and effective separation of noise sources, providing useful information and decision support for noise control and management.

[0048] Preferably, a three-dimensional environmental model of each structure in the substation is obtained, and a topological relationship matrix of geometric bodies in the three-dimensional environmental model is obtained; on-site measurement of the substation is performed to obtain noise measurement data; and the noise measurement data is mapped to the three-dimensional environmental model in blocks based on the measurement position, and registration is performed to obtain local noise image data, including:

[0049] Obtain 3D environmental model data of each structure in the substation;

[0050] Extract and represent the geometry of the three-dimensional environment model, and the geometry represents the model data;

[0051] Extract the topological relationship and boundary conditions of the geometric bodies in the model based on the 3D environment model data and the geometric body representation model data to obtain the geometric body topological relationship matrix;

[0052] Conduct on-site measurements of substations to obtain noise measurement data, including noise decibel data and measurement point location data;

[0053] Perform wavelet transform filtering on the noise decibel data to obtain noise image data;

[0054] According to the measurement point position data, the noise image data is divided into blocks and mapped to the three-dimensional environment model to obtain block noise data;

[0055] Image registration is performed based on the block noise data and the model units in the three-dimensional environment model to obtain local noise image data.

[0056] 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:

[0057] Obtain 3D environmental model data of each structure in the substation;

[0058] In this embodiment of the present invention, a laser scanner is used to scan the substation to obtain point cloud data. Aerial or ground photography is performed using drones or other photogrammetry equipment to obtain image data of the substation. Based on the substation design drawings and measurement data, a 3D model is constructed using computer-aided design software.

[0059] Extract and represent the geometry of the three-dimensional environment model, and the geometry represents the model data;

[0060] This embodiment of the present invention processes point cloud data acquired by a laser scanner to extract geometric information, such as walls, pillars, and equipment. It also processes image data acquired by photogrammetry and uses computer vision algorithms to extract geometric information from the images. Furthermore, it imports 3D model data through CAD software to extract geometric information from the data.

[0061] Extract the topological relationship and boundary conditions of the geometric bodies in the model based on the 3D environment model data and the geometric body representation model data to obtain the geometric body topological relationship matrix;

[0062] This embodiment of the present invention segments the geometric bodies within the model data based on geometric representations, dividing them into different units or components. Based on the relative positions and connections between the geometric bodies, topological relationships between them, such as adjacency and connectivity, are extracted. Based on the boundary features of the geometric bodies, boundary condition information, such as sound-absorbing material on walls and sound insulation pads on floors, is extracted. Based on the extracted topological relationships and boundary conditions, a geometric topological relationship matrix is constructed for subsequent processing of noise image data.

[0063] Conduct on-site measurements of substations to obtain noise measurement data, including noise decibel data and measurement point location data;

[0064] In this embodiment, a noise measurement instrument is used to measure noise decibels at different locations in the substation, recording the noise level at each measurement point. A total station, GPS, or other measurement equipment is used to measure the coordinates of each measurement point and record the spatial location information of the measurement point.

[0065] Perform wavelet transform filtering on the noise decibel data to obtain noise image data;

[0066] In this embodiment of the present invention, a wavelet transform is applied to noise decibel data to convert the time-domain signal into a frequency-domain signal. The frequency-domain signal after the wavelet transform is filtered to remove unnecessary frequency components while retaining the frequency information related to the noise. An inverse wavelet transform is then performed on the filtered frequency-domain signal to convert it back into a time-domain signal, yielding noise image data.

[0067] According to the measurement point position data, the noise image data is divided into blocks and mapped to the three-dimensional environment model to obtain block noise data;

[0068] This embodiment of the present invention associates measurement point position data with geometric entities in a three-dimensional environment model. Based on the spatial position information of each measurement point, the corresponding geometric entity or model unit in the three-dimensional environment model is determined. Noise image data is then divided into blocks based on the corresponding relationship between the measurement point positions. Based on the geometric entity or model unit where the measurement point is located, the noise image data is divided into corresponding blocks, with each block corresponding to a geometric entity or model unit.

[0069] Image registration is performed based on the block noise data and the model units in the three-dimensional environment model to obtain local noise image data.

[0070] In this embodiment of the present invention, block noise data is registered with model elements in a 3D environment model, aligning them in the same coordinate system. Image processing algorithms, such as feature point matching and transformation models, can be used for this registration. Through this registration operation, the block noise data is aligned with the corresponding model element to obtain local noise image data. Each local noise image data corresponds to a model element and reflects the noise level within that model element.

[0071] By acquiring three-dimensional environmental model data of a substation, the present invention can accurately represent the substation's geometry and structural information. This is crucial for subsequent noise source location and isolation, as noise propagation and reflection can be affected by buildings. By processing the three-dimensional environmental model, geometric bodies are extracted and represented. This approach allows a complex three-dimensional environmental model to be converted into a set of geometric bodies, simplifying subsequent processing. Based on the three-dimensional environmental model data and the geometric representation model data, the topological relationships and boundary conditions between the geometric bodies are extracted. This information is crucial for establishing the connectivity and boundary constraints between the geometric bodies, facilitating subsequent noise source location and isolation. Actual noise measurement data can be obtained by conducting on-site measurements at the substation. This data includes noise intensity (in decibels) and the location of the measurement points. This data can be used for subsequent analysis and processing. The noise decibel data is then subjected to wavelet transform filtering. Wavelet transform filtering smoothes and denoises the noise data, producing clearer and more accurate noise image data. This noise image data can be used to subsequently generate local noise image data. Using the measurement point location data, the noise image data is divided into blocks, and each block is mapped to a corresponding location in the 3D environmental model. This aligns the noise data with the actual geometry of the substation, providing an accurate data foundation for subsequent noise source location and isolation. Image registration is performed using the block noise data and the model elements in the 3D environmental model. Image registration aligns the block noise data with the corresponding model elements, generating local noise image data for each model element. This local noise image data can be used for subsequent noise source location and isolation, providing more accurate local noise information. In summary, 3D environmental model data for each structure in a substation is obtained, and geometric entities are extracted and represented. Then, by extracting the topological relationships and boundary conditions of the geometric entities, a geometric entity topology relationship matrix is obtained. Simultaneously, on-site measurements are conducted at the substation to obtain noise measurement data, including noise decibel data and measurement point location data. Noise image data is obtained by wavelet transform filtering of the noise decibel data. The noise image data is then mapped into blocks within the 3D environmental model, and image registration is performed based on the model elements to obtain local noise image data. The above steps establish an accurate 3D environment model, extract geometric topological relationships, obtain actual noise measurement data, denoise and smooth the noise data, and obtain accurate local noise image data. These results provide the foundation and support for tasks such as noise analysis, noise source location, and separation.

[0072] Preferably, the topological relationship and boundary conditions of the geometric bodies in the model are extracted based on the three-dimensional environment model data and the geometric body representation model data to obtain a geometric body topological relationship matrix, including:

[0073] Performing basic shape type recognition on the geometric body representation model data to obtain geometric body type data;

[0074] In this embodiment of the present invention, geometric analysis software, such as CAD software or specialized geometric processing tools, is used to import geometric representation model data. A geometric type recognition algorithm is then applied to classify geometric objects into basic shape types, such as cubes, cylinders, and spheres, by analyzing their shape, boundary characteristics, curvature, and other properties. Automatic geometric recognition tools are then used to automatically identify and classify the geometric representation model data through techniques such as pattern matching and feature extraction.

[0075] Extract acoustic impedance and absorption coefficient based on material information based on 3D environment model data and geometric representation model data to obtain geometric boundary condition data;

[0076] This embodiment of the present invention obtains material information for each geometric object based on the 3D environmental model data, including surface material, density, and roughness. Based on this material information, acoustic engineering calculation methods are used to calculate the acoustic impedance of each geometric object, which characterizes the ability of the geometric object's interface to reflect sound waves. Based on this material information, acoustic engineering calculation methods are also used to calculate the absorption coefficient of each geometric object, which characterizes the geometric object's ability to absorb sound waves.

[0077] Performing topological relationship analysis of adjacency, inclusion, separation, and intersection between geometric bodies based on geometric body representation model data and geometric body type data to obtain geometric body topological relationship data;

[0078] Embodiments of the present invention extract spatial relationships between geometric bodies, such as distance and relative position, based on geometric body representation model data. The type of each geometric body is determined based on the geometric body type data. Topological relationship analysis algorithms, such as Euclidean space analysis and bounding box analysis, are applied to determine the adjacency, containment, separation, and intersection relationships between geometric bodies.

[0079] Based on the geometric body topological relationship data and the geometric body boundary condition data, the topological relationship and boundary conditions are integrated into a matrix form to obtain the geometric body topological relationship matrix.

[0080] The embodiment of the present invention constructs a matrix to represent the topological relationship between geometric bodies based on geometric body topological relationship data. The rows and columns of the matrix correspond to the various geometric bodies, and the matrix elements represent the topological relationship between the corresponding geometric bodies, such as adjacent, contained, separated, intersecting, etc. According to the geometric body boundary condition data, a matrix is constructed to represent the boundary conditions of the geometric bodies. The rows of the matrix correspond to the various geometric bodies, and the columns correspond to the various boundary conditions. The matrix elements represent the properties of the corresponding geometric bodies under the boundary conditions, such as acoustic impedance, absorption coefficient, etc. By integrating the topological relationship matrix and the boundary condition matrix, the two matrices can be merged or combined according to certain rules to form a comprehensive geometric body topological relationship matrix, which contains the topological relationship between the geometric bodies and the boundary condition information of each geometric body.

[0081] By identifying basic shape types in geometric model data, embodiments of the present invention can determine the type of each geometric object, such as a sphere, cube, or cylinder. This geometric type data can be used for subsequent analysis and processing. Based on the identification and classification of the geometric model data, the basic characteristics and properties of the geometric objects can be extracted, providing a foundation for further acoustic, topological, and boundary condition analysis. By extracting acoustic impedance and absorption coefficients based on material information, the boundary conditions of each geometric object—that is, the acoustic characteristics between it and its surrounding environment—can be determined. This boundary condition data is crucial for acoustic simulation and analysis. By extracting the acoustic properties of geometric objects, such as reflection, absorption, and scattering, the behavior and impact of the geometric object in the acoustic environment can be evaluated. This facilitates the design and optimization of acoustic systems, improving sound propagation and reducing noise. By performing topological relationship analysis on the geometric model data, relationships between geometric objects, such as adjacency, inclusion, separation, and intersection, can be determined. This topological relationship data is crucial for model structural analysis, collision detection, and object interaction. By analyzing the topological relationships between geometric objects, we can understand their organizational structure and connectivity, helping us understand the object's shape, layout, and function. By integrating the topological relationships and boundary conditions of geometric bodies into a matrix form, the relationships between individual geometric bodies can be represented in a structured manner. This matrix representation facilitates further analysis and processing, such as topology optimization, graph theory analysis, and network analysis. Integrating topological relationships and boundary conditions into a matrix helps organize and manage large amounts of geometric body-related data, improving the efficiency of data processing and calculation. At the same time, the matrix representation can also support the application of algorithms and technologies based on matrix operations. In summary, the above steps can achieve the identification and classification of geometric body representation model data, acoustic property extraction, topological relationship analysis, and data integration and structuring. These can provide detailed information about geometric body types, boundary conditions, acoustic properties, and topological relationships, providing a foundation and support for subsequent tasks such as geometric analysis, acoustic simulation, structural analysis, and optimization.

[0082] Preferably, outer convolution fusion is performed on the noise images between adjacent units according to the geometric topological relationship matrix and the local noise image data, and iteratively propagated to the entire substation area to generate initial sound source distribution estimation data for the entire substation, including:

[0083] The substation space is discretized into small space unit grids according to the geometric topological relationship matrix to obtain space grid unit data;

[0084] Assigning local noise image data to spatial units of spatial grid unit data according to measurement point position data to obtain initial grid noise distribution data;

[0085] The initial grid noise distribution data and the geometric topology relationship matrix perform outer convolution on the noise of adjacent grid cells based on boundary conditions to obtain the fusion result data;

[0086] The fusion result data is used as the initial value, and the outer convolution operation based on the boundary conditions is repeated until it propagates to the entire substation area to obtain the grid noise distribution data;

[0087] The grid noise distribution data is interpolated and denoised to obtain the initial sound source distribution estimation data.

[0088] 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:

[0089] The substation space is discretized into small space unit grids according to the geometric topological relationship matrix to obtain space grid unit data;

[0090] This embodiment of the present invention divides the substation space into small, discrete grid cells based on a geometric topological relationship matrix. A grid generation algorithm, such as the finite element method or voxel grid method, can be used to segment the space into discrete grid cells. For each grid cell, relevant data such as its location, size, and boundary information is recorded. This data can be used for subsequent calculations and analysis.

[0091] Assigning local noise image data to spatial units of spatial grid unit data according to measurement point position data to obtain initial grid noise distribution data;

[0092] Embodiments of the present invention obtain location data of measurement points located within or around a substation for measuring noise levels. Based on the measurement point location data, local noise image data at the corresponding locations is obtained. Noise data can be acquired using a sensor or noise measurement device, or by analog or numerical simulation methods. The local noise image data corresponding to the measurement point location data is assigned to the corresponding spatial grid cell. Interpolation methods, such as nearest neighbor interpolation or bilinear interpolation, can be used to assign the local noise value to the grid cell closest to the measurement point.

[0093] The initial grid noise distribution data and the geometric topology relationship matrix perform outer convolution on the noise of adjacent grid cells based on boundary conditions to obtain the fusion result data;

[0094] The embodiment of the present invention is based on the initial grid noise distribution data obtained in the previous step, and each grid unit contains local noise information. According to the geometric body topological relationship matrix, an outer product convolution operation based on boundary conditions is performed between adjacent grid units. Outer product convolution can be used to simulate the propagation and interaction process of sound waves. According to the geometric body topological relationship matrix and boundary condition data, the boundary conditions between adjacent grid units, such as acoustic impedance, absorption coefficient, etc., are determined. The results of the outer product convolution are fused with the boundary conditions to obtain new grid noise distribution data. These data reflect the sound wave propagation and interaction effects between adjacent grid units.

[0095] The fusion result data is used as the initial value, and the outer convolution operation based on the boundary conditions is repeated until it propagates to the entire substation area to obtain the grid noise distribution data;

[0096] In this embodiment of the present invention, the fused data obtained above is used as the initial value. The outer convolution operation is repeated, and the fused data is used as the new initial value. The number of iterations and convergence conditions can be set according to specific circumstances. The outer convolution operation is repeated until the fused data is propagated throughout the substation area. This ensures that noise propagation and interaction effects are considered throughout the entire area, resulting in more accurate grid noise distribution data.

[0097] The grid noise distribution data is interpolated and denoised to obtain the initial sound source distribution estimation data.

[0098] In an embodiment of the present invention, interpolation processing is performed on discretized grid noise distribution data to convert it into continuous sound source distribution data. Common interpolation methods include linear interpolation and spline interpolation, and the appropriate interpolation method is selected based on the needs. The interpolated sound source distribution data is then denoised to eliminate the effects of noise or outliers. Denoising methods can include filters, wavelet transforms, statistical methods, etc., and the appropriate denoising method is selected based on the characteristics of the noise.

[0099] The present invention discretizes the substation space based on a geometric topological relationship matrix, dividing it into small spatial grids. These spatial grid cell data provide a structured description of the substation area, laying the foundation for subsequent noise distribution analysis and processing. By gridding the substation space, complex geometric structures can be converted into regular grid structures, simplifying problem processing and calculation. Gridding also facilitates subsequent operations such as interpolation, noise propagation, and sound source distribution estimation. By assigning measurement point location data and local noise image data to the spatial cells of the spatial grid cell data, initial grid noise distribution data can be obtained. This data describes the noise levels in different spatial cells of the substation and provides an initial state for subsequent noise propagation and analysis. By associating the measurement point location data with the spatial grid cell data, the actual measured noise data can be mapped to the spatial grid cells, achieving a spatial representation of the noise data and facilitating noise analysis and processing. By performing an outer product convolution based on boundary conditions on the initial grid noise distribution data, the noise between adjacent grid cells can be fused. This takes into account the mutual influence and propagation patterns between grid cells, resulting in more accurate noise distribution data. Outer convolution based on boundary conditions simulates the propagation of noise between adjacent grid cells. This helps understand the spatial diffusion of noise within the substation and provides a basis for subsequent noise control and optimization. By repeatedly performing outer convolution based on boundary conditions, noise can be gradually propagated to grid cells throughout the substation area, resulting in more accurate grid noise distribution data. This data provides detailed noise distribution information for use in noise assessment and the development of control measures. By repeatedly performing outer convolution based on boundary conditions and using the resulting data as the initial value for the next iteration, the accuracy and precision of the noise distribution data can be gradually improved. This iterative process better simulates noise propagation and distribution, enhancing model reliability. Interpolation and denoising of the grid noise distribution data yields initial sound source distribution estimates. This data describes the location and intensity distribution of sound sources within the substation, providing a basis for sound source control and optimization. Interpolation and denoising fill in gaps and noise interference in the grid noise distribution data, improving the data's completeness and accuracy. This helps better understand and analyze the distribution of sound sources within the substation. In summary, the above steps include providing spatial grid cell data, implementing data association, simulating noise propagation and distribution, optimizing the precision and accuracy of noise distribution data, and obtaining initial noise source distribution estimates. These results help us gain a deeper understanding of substation noise characteristics and provide a scientific basis for noise control and optimization.

[0100] Preferably, separating the direct sound signal and the multipath echo components according to the initial sound source distribution estimation data to generate echo decoupling data includes:

[0101] The generalized coordinated filtering algorithm is used to perform parameterized modeling on the initial sound source distribution estimation data. The model is divided into two parts: the direct sound component and the multipath sound component, thus obtaining the direct sound model and the multipath sound model.

[0102] The generalized coordinated filtering of the embodiment of the present invention is a statistical model estimation method for separating signal components. By performing parameterized modeling on the sound source distribution data, the sound source model is decomposed into a direct sound component and a multipath sound component. Based on the initial sound source distribution estimation data, the sound source is parameterized using an appropriate mathematical model. Common methods include Gaussian mixture models and sparse representation models. Parametric modeling can decompose the sound source into different components, such as direct sound and multipath sound. Through parametric modeling, a model of the direct sound component is extracted. The direct sound model describes the sound wave component that propagates directly from the sound source to the receiving point. Through parametric modeling, a model of the multipath sound component is extracted. The multipath sound model describes the sound wave component that propagates from the sound source to the receiving point through multiple paths such as reflection and diffraction.

[0103] A joint optimization objective function is constructed based on the direct sound model and the multipath sound model. The two models are weighed by a coordination factor, and the objective function is solved using the generalized least squares method to obtain the separated direct sound component data and multipath sound component data.

[0104] This embodiment of the present invention constructs a joint optimization objective function based on the direct sound model and the multipath sound model, which includes constraints for separating direct and multipath sounds. The specific form of the objective function can be defined based on the specific sound source model and separation requirements. A coordination factor is introduced to balance the importance of direct and multipath sounds. The coordination factor can be a parameter whose value can be adjusted to control the contribution of direct and multipath sounds to the separation results. The objective function is solved using the generalized least squares method to minimize the objective function and obtain a solution that optimizes the separation of direct and multipath sounds. The generalized least squares method is a commonly used parameter estimation method that can be solved using optimization algorithms such as gradient descent or Newton's method.

[0105] A random state space model of the multipath echo path is constructed based on the multipath acoustic component data. Each path is filtered and tracked based on the Kalman particle filter framework. The path delay, attenuation, and direction are statistically analyzed to obtain preliminary multipath path parameter data.

[0106] In an embodiment of the present invention, multipath acoustic component data is modeled as a random state-space model, where the state represents parameters such as the delay, attenuation, and direction of the multipath echo path. Based on acoustic characteristics and signal propagation theory, a suitable state-space model, such as a linear dynamic system model, can be selected. The multipath echo path is tracked and estimated using the Kalman particle filter framework. The Kalman particle filter is a recursive filtering algorithm that can estimate the state variables of a dynamic system. By continuously iterating and updating the particle weights, an estimate of the multipath path parameters can be obtained. Based on the tracking results of the Kalman particle filter, statistical analysis of the multipath path parameters such as delay, attenuation, and direction is performed to obtain preliminary multipath path parameter data.

[0107] The multipath acoustic component data is sparsely represented on the wavelet basis to obtain a noise sparse representation model;

[0108] This embodiment of the present invention performs a wavelet transform on multipath acoustic component data, converting it from the time domain to the wavelet domain. The wavelet transform can represent a signal as a linear combination of wavelet basis functions of different scales and frequencies. In the wavelet domain, a sparse representation method is used to represent the multipath acoustic component data. The goal of sparse representation is to approximate the original signal using a small number of wavelet basis function coefficients, ensuring that most coefficients are zero or near zero. Using this sparse representation, a noise sparse representation model for the multipath acoustic component data is obtained. This noise sparse representation model describes the sparsity characteristics of the multipath acoustic component data on the wavelet basis.

[0109] Adaptively model the multipath noise based on the noise sparse representation model and preliminary multipath path parameter data, and subtract the multipath noise component from the direct sound component data to obtain the multipath noise-removed field data;

[0110] Based on a sparse representation model and multipath path parameter data, the present invention designs an iterative algorithm to decompose and optimize multipath acoustic component data. The iterative algorithm aims to separate multipath echo path data from background noise data and continuously update the optimization results. Through the iterative algorithm, the multipath acoustic component data is decomposed into multipath echo path data and background noise data. This decomposition process leverages the sparsity of sparse representation, representing the multipath echo path data as a small number of non-zero wavelet basis function coefficients and the background noise data as a sparse vector with most coefficients close to zero. During the iterative process, the optimization results are updated based on the decomposition results and multipath path parameter data, ensuring that the separated multipath echo path data and background noise data are more accurate and reliable.

[0111] The residual noise manifold is constructed for the residual noise of the multipath denoised field data using manifold learning technology, and the multipath denoised field data is mapped onto the noise manifold to obtain the reconstructed residual multipath noise component data.

[0112] In this embodiment of the present invention, the multipath-denoised field data is preprocessed, for example, to remove direct sound components and multipath noise components, retaining only the residual noise. Manifold learning tools (such as t-SNE or Isomap) are used to perform manifold learning on the residual noise data to construct a residual noise manifold. Appropriate parameters, such as neighborhood size and dimensionality reduction, are selected based on the specific situation. The multipath-denoised field data is preprocessed to remove the direct sound components, retaining only the multipath-denoised components. Manifold learning mapping methods (such as LLE and Isomap) are used to map the multipath-denoised field data onto the constructed noise manifold. This results in the corresponding position of the multipath-denoised field data on the noise manifold.

[0113] The reconstructed residual multipath noise component data is subtracted from the multipath denoised field data to obtain the echo decoupling data.

[0114] In this embodiment of the present invention, wavelet transform is applied to multipath-denoised field data to convert it into wavelet coefficients. Appropriate wavelet basis functions and scales are selected to suit the characteristics of the multipath-denoised field data. A sparse representation method is used to sparsely represent the wavelet coefficients of the multipath-denoised field data. An L1 norm minimization algorithm (such as the Lasso algorithm) or a dictionary learning algorithm can be used to find as few nonzero coefficients as possible to represent the multipath-denoised field data. Based on the sparse representation results, a noise sparse representation model for the multipath-denoised field data is obtained, which contains a relatively small number of nonzero coefficients, representing the noise component.

[0115] The present invention uses a generalized coordinated filtering algorithm to perform parameterized modeling on the initial sound source distribution estimation data. This represents the sound source distribution data as a combination of two models: a direct sound component and a multipath sound component, better describing the propagation and reflection characteristics of sound in the substation. The sound source is divided into two components: the direct sound component and the multipath sound component, and corresponding models are established for each. The direct sound model describes the path of sound propagating directly from the source to the receiving point, while the multipath sound model describes the path of sound reaching the receiving point after multiple reflections, diffractions, and reflections. This decomposition helps to better understand the sound propagation mechanism. By combining the direct sound model and the multipath sound model to construct a joint optimization objective function, the trade-off between the two can be balanced. This helps accurately separate the direct sound component and the multipath sound component from the mixed sound source distribution data, improving the interpretability and accuracy of the data. By applying the generalized least squares method to solve the joint optimization objective function, the optimized direct sound component data and multipath sound component data can be obtained. This solution method can effectively solve the parameter estimation problem and obtain relatively accurate separation results. By constructing a random state-space model of the multipath echo path based on multipath acoustic component data, the propagation path of multipath sound can be modeled. This helps understand the multiple reflections and propagation characteristics of sound in substations and provides a parameterized description of the path. The multipath echo path is tracked and estimated using the Kalman particle filter framework. This allows for the extraction of path statistical information such as delay, attenuation, and direction from the observed data, providing a foundation for subsequent multipath denoising and separation. By sparsely representing the multipath acoustic component data on a wavelet basis, the noise component can be separated from the signal and a sparse representation model of the noise can be extracted. This facilitates further analysis and processing of the noise component, providing a foundation for subsequent multipath denoising and noise separation. Based on the noise sparse representation model and preliminary multipath path parameter data, multipath noise is adaptively modeled. This enables more accurate modeling and estimation of multipath noise, helping to more effectively mitigate its impact. By subtracting the multipath noise component from the direct acoustic component data, multipath-de-noised field data is obtained. This removes the impact of multipath noise, resulting in a cleaner and more distinct acoustic signal and improving the accuracy of subsequent processing tasks. Manifold learning technology enables modeling and analysis of residual noise in multipath-denoised field data. This helps better understand the inherent structure and characteristics of the residual noise, providing more accurate information for subsequent processing. By mapping the multipath-denoised field data onto the noise manifold, reconstructed residual multipath noise component data can be obtained. This extracts information about the noise components, further reducing multipath noise interference and improving signal quality. By subtracting the reconstructed residual multipath noise component data from the multipath-denoised field data, echo decoupling data can be obtained. This effectively eliminates the impact of echoes, retaining only components related to the direct sound in the signal, improving signal clarity and intelligibility.By removing the reconstructed residual multipath noise component, the interference of multipath noise on the signal can be effectively reduced, the quality and accuracy of the signal can be improved, and the subsequent processing tasks can be made more reliable and stable.

[0116] Preferably, the echo decoupling data is dimensional compressed, and the eigenmodes of the low-order scattered field are extracted to obtain a scattered field eigenmode set; and the parameters of the scattered field eigenmode set are estimated to obtain eigenscattering cloud map data, including:

[0117] The echo decoupling data is discretely sampled in the spatiotemporal domain and arranged into a kernel tensor form to obtain the kernel tensor representation data;

[0118] In this embodiment of the present invention, echo decoupling data is discretely sampled in the spatiotemporal domain, selecting an appropriate sampling method and sampling rate to obtain discrete data points. The discretely sampled data is then arranged in a specific order to form a kernel tensor. A corresponding kernel tensor is constructed based on the sampling dimension and the number of sampling points. For example, for three-dimensional data, a three-dimensional kernel tensor can be constructed.

[0119] The supersymmetric matrix decomposition technology is used to perform low-rank decomposition approximation on the core tensor representation data to obtain low-dimensional embedded vector data;

[0120] This embodiment of the present invention uses supersymmetric matrix decomposition (SMF) technology to perform a low-rank decomposition of the core tensor representation data. An appropriate SMF method is selected and parameterized based on the nature and dimensionality of the data. After decomposition, low-dimensional embedded vector data is obtained.

[0121] The low-dimensional embedded vector data is mapped onto the differential manifold, and the scattering field eigenmode is reconstructed to obtain the scattering field eigenmode set;

[0122] In this embodiment of the present invention, manifold learning tools are used to map low-dimensional embedded vector data onto a differential manifold, obtaining a representation in manifold space. Appropriate manifold learning algorithms and parameter settings are selected. Eigenmode reconstruction methods are used to process the mapped data, for example, using techniques such as PCA or ICA to extract the eigenmodes of the scattered field. Eigenmodes are statistically significant characteristic patterns that can reflect important structures in the data.

[0123] The parameters of the eigenmode set of the scattered field are estimated through a nonlinear regression model to obtain the intrinsic scattering cloud map data.

[0124] The embodiment of the present invention uses a set of eigenmodes of the scattered field as input data and prepares corresponding target data, such as actual field pattern parameters. The prepared input data and target data are used to train a selected nonlinear regression model. By iteratively optimizing the parameters of the model, the model is able to estimate the parameters of the field pattern for the input data; the trained nonlinear regression model is used to estimate the parameters of the field pattern for the set of eigenmodes of the scattered field. Taking each eigenmode as input, the model will output the corresponding field pattern parameter value. The field pattern parameters are combined with the eigenmode to generate eigenscattering cloud map data. Depending on the specific application requirements, the parameter values can be used to reconstruct the scattered field or perform other subsequent processing.

[0125] By discretely sampling echo decoupling data in the spatiotemporal domain, the present invention can convert continuous signal data into discrete sampling points, facilitating subsequent processing and calculation. The data obtained by discrete sampling is arranged into a core tensor form, converting the multidimensional data structure into a higher-order tensor representation. This can better capture the multidimensional characteristics and structural information of the data, providing a richer data representation for subsequent processing. Using supersymmetric matrix decomposition technology, the core tensor representation data is approximated by low-rank decomposition. Low-rank decomposition can reduce high-dimensional data to lower dimensions, reducing redundant information and extracting the data's key features. The low-dimensional embedded vector data obtained by the low-rank decomposition can be used to map the original data into a lower-dimensional space. This reduces data storage and computational costs while preserving the data's key features, facilitating subsequent processing and analysis. By mapping the low-dimensional embedded vector data onto a differential manifold, the data can be represented as points on the manifold. The structure and geometric properties of the manifold can be utilized to analyze the data and extract its implicit features. By mapping the low-dimensional embedded vector data onto the differential manifold, the eigenmodes of the scattered field can be reconstructed. Eigenmodes are characteristic patterns that describe the scattering field. By reconstructing the eigenmode set, the primary modes and characteristics of the scattering field can be extracted. Parameters of the eigenmode set can be estimated by applying a nonlinear regression model. The nonlinear regression model can estimate the model parameters by fitting the relationship between actual data and the model. Estimating the parameters of the eigenmode set using the nonlinear regression model yields the field pattern parameters of the scattering field. These parameters reflect the characteristics and properties of the scattering field, providing a deeper understanding and description of the scattering field. Field pattern parameter estimation can generate intrinsic scattering cloud data.

[0126] Preferably, the parameters of the field mode are estimated for the eigenmode set of the scattered field by a nonlinear regression model to obtain the intrinsic scattering cloud map data, including:

[0127] The nonlinear scattering model of eigenmode theory is established based on the eigenmode set of the scattered field and the topological relationship matrix of the geometric body;

[0128] The embodiments of the present invention construct a geometric model based on the shape and position of the actual scatterer and establish a geometric topological relationship matrix. Using the scattered field eigenmode set and the geometric topological relationship matrix, a nonlinear scattering model is established. Numerical methods, physical models, or statistical methods can be used, depending on the actual problem.

[0129] Identify the free parameters of wave number, attenuation rate and phase in the nonlinear scattering model and obtain the free parameter set;

[0130] The present invention selects an appropriate parameter identification method based on the specific nonlinear scattering model and data characteristics. Using this method, free parameters such as wave number, decay rate, and phase in the nonlinear scattering model are identified. Estimated values of the free parameters are obtained based on the degree of fit with the observed data.

[0131] According to the preset measured pattern data, the eigenmode set of the scattered field and the difference of the nonlinear scattering model are constructed as an optimization objective function to obtain a nonlinear regression objective function;

[0132] This embodiment of the present invention collects or generates a set of measured scattering pattern data as preset target data. By comparing the differences between the scattered field eigenmode set and the nonlinear scattering model in the measured pattern data, an optimization objective function is constructed to measure the differences. Specific methods for measuring these differences include mean square error (MSE) and correlation coefficient.

[0133] Solve the model parameters of the nonlinear regression objective function according to the free parameter set to obtain the free parameter estimation data;

[0134] The embodiment of the present invention sets the initial parameter values of the nonlinear regression method according to the characteristics of the problem, and uses the selected nonlinear regression method to solve the nonlinear regression objective function to obtain the optimal free parameter estimation value data.

[0135] Substitute the free parameter estimation data into the nonlinear scattering model to generate intrinsic scattering cloud data.

[0136] In this embodiment of the present invention, the free parameter estimation data obtained in step S444 is substituted into the nonlinear scattering model to replace the original free parameters. The nonlinear scattering model after substitution is used to calculate and generate intrinsic scattering cloud data, which reflects the characteristics of the scattering field.

[0137] The present invention establishes a nonlinear scattering model based on eigenmode theory, which can more accurately describe the physical processes and phenomena of the scattered field. By considering the eigenmode set of the scattered field and the topological relationship between geometric bodies, the model can capture more scattering phenomena and provide a more precise description. The nonlinear scattering model can provide explanatory power for various physical phenomena in the scattered field. By establishing this model, we can understand the wave behavior in the scattering process, the interaction between scatterers, and other aspects, and further study the characteristics of the scattered field. By identifying parameters such as wave number, decay rate, and phase in the nonlinear scattering model, we can determine the free parameters in the model. These parameters control the behavior and characteristics of the model. Identifying them allows for better adjustment and optimization of the model to better match actual observed data. Once the free parameter set is obtained, it can be used in subsequent parameter estimation and optimization processes. This parameter set provides a preliminary estimate of the model parameters, providing a basis for further analysis and research. By constructing the difference between the eigenmode set of the scattered field and the nonlinear scattering model into an optimization objective function, the gap between the model and actual observed data can be quantified. This objective function can be used as the basis for nonlinear regression optimization to help determine the optimal model parameters. By optimizing the objective function, the nonlinear regression model can be made to better fit the measured data. This improves the model's accuracy and predictive power, enabling it to better describe and predict the actual scattering field. Solving the nonlinear regression objective function yields model parameter estimates. These parameter estimates are the optimal values for the model to fit the measured data, subject to the constraints of the optimization objective function. They more accurately describe the characteristics and behavior of the scattering field. Obtaining free parameter estimates can be used to represent model parameters. These estimates provide specific numerical values for the model parameters, facilitating quantitative analysis and investigation of scattering field characteristics. Substituting the free parameter estimates into the nonlinear scattering model generates intrinsic scattering contours. These contours, calculated based on the estimated model parameters, reflect the characteristics and behavior of the scattering field. They provide a visual representation of the scattering field, facilitating understanding and analysis of scattering phenomena. The generated intrinsic scattering contours can be used to further investigate the characteristics of the scattering field. Analysis of the contours reveals patterns, structures, and distributions in the scattering field, providing a deeper understanding of the nature of scattering phenomena. This is of great significance for the design and optimization of scattering field applications.

[0138] Preferably, the intrinsic scattering cloud map data is subjected to acoustic energy flow around the center identification, and the noise source position is matched to obtain potential noise source position data; the source point instantaneous motion direction is locked according to the potential noise source position data, and the motion trajectory of the source point is reconstructed to obtain noise source motion trajectory data; the position information of the stationary source and the moving source is distinguished and integrated according to the noise source motion trajectory data and the potential noise source position data to obtain moving source path data, including:

[0139] The acoustic energy flow around the center is identified on the intrinsic scattering cloud image data, and the noise source position is matched to obtain the potential noise source position data;

[0140] This embodiment of the present invention performs preprocessing operations such as noise removal and filtering on intrinsic scattering cloud image data to improve the signal-to-noise ratio and data quality. Image processing techniques and object recognition algorithms are used to identify the center of the acoustic energy flow in the cloud image. The center of the acoustic energy flow is then matched with the location of the actual scatterer to obtain the location data of the potential noise source.

[0141] According to the potential noise source position data and the filtered acoustic energy flow field data, the acoustic energy flow gradient vector based on the source position is calculated to obtain the instantaneous motion direction data of the source point;

[0142] This embodiment of the present invention applies a suitable filtering algorithm, such as a Gaussian filter, to the acoustic energy flow field data to smooth the data and remove noise. Based on the potential noise source location data and the filtered acoustic energy flow field data, the acoustic energy flow gradient vector at the source location is calculated to determine the instantaneous direction of motion of the source.

[0143] The instantaneous motion direction data of the source point is reconstructed based on the time series analysis of the source point motion trajectory to obtain the noise source motion trajectory data;

[0144] The present invention performs smoothing and filtering on the instantaneous motion direction data of the source point to remove noise and unnecessary fluctuations. Time series analysis methods, such as Kalman filtering and spline interpolation, are then applied to analyze and reconstruct the processed data to obtain the motion trajectory data of the noise source.

[0145] According to the preset trajectory classification standard and the potential noise source location data, the noise source motion trajectory data is separated into stationary sources and mobile sources to obtain mobile source trajectory data and stationary source location data;

[0146] The present invention extracts features from noise source trajectory data, calculates the velocity and acceleration of each trajectory point, and records changes in the direction of motion. Using a noise source classification algorithm, the extracted trajectory features are input into a classification model to classify stationary and moving sources.

[0147] The moving source trajectory data and the stationary source position data are merged to obtain the moving source path data.

[0148] For noise sources classified as stationary, embodiments of the present invention can accurately locate them using traditional positioning methods, such as Doppler positioning and cross-correlation positioning. For noise sources classified as mobile, methods based on the Doppler effect or sound source localization techniques, such as the Extended Kalman Filter (EKF) or particle filter, can be used for real-time or offline positioning estimation.

[0149] The present invention can determine the location of potential noise sources by identifying the acoustic energy flow around the center of intrinsic scattering cloud data. This helps accurately locate the approximate location of the noise source and provides a basis for subsequent analysis. The steps corresponding to the noise source location can preprocess the intrinsic scattering cloud data, converting it into more interpretable and analyzable potential noise source location data. This can reduce data complexity and the computational effort required for subsequent analysis. By calculating the acoustic energy flow gradient vector based on the source location, the instantaneous motion direction of the source point can be determined. This provides information about the source point's motion, helping to understand the noise source's motion behavior and characteristics. The source point's instantaneous motion direction data can be used to analyze the noise source's motion pattern, speed, and trajectory changes. This helps reveal the noise source's motion patterns and characteristics, providing a basis for subsequent trajectory reconstruction and analysis. By performing time series analysis on the source point's instantaneous motion direction data, the noise source's motion trajectory can be reconstructed. This provides information about the noise source's motion path in space, facilitating more in-depth research and analysis of the noise source's motion behavior. The noise source's motion trajectory data can be used to understand the noise source's motion behavior and pattern in space. Trajectory data analysis can reveal the motion patterns, frequency, amplitude, and other characteristics of noise sources, providing a basis for subsequent classification and analysis. Based on pre-set trajectory classification criteria and potential noise source location data, noise source trajectory data can be separated into mobile source trajectory data and stationary source location data. This facilitates classification and categorization of noise sources, making subsequent analysis more accurate and targeted. Mobile source trajectory data provides information on the motion trajectory of mobile noise sources, enabling analysis of their motion patterns, paths, and speed. Stationary source location data provides location information on stationary noise sources, enabling analysis of their distribution and spatial characteristics. By combining mobile source trajectory data and stationary source location data, complete moving source path data can be obtained. This allows for comprehensive analysis of the overall distribution and motion characteristics of noise sources by considering both the location information of mobile and stationary sources. Moving source path data provides a global perspective on the motion paths of noise sources, helping to understand the overall motion patterns, trends, and evolution of noise sources. Furthermore, moving source path data can be correlated with other environmental factors to reveal the formation mechanisms and influencing factors of noise sources. In summary, these steps include noise source location identification, motion direction calculation, motion trajectory reconstruction, source point classification and separation, moving and stationary source analysis, and global perspective analysis. These results enable a more comprehensive and in-depth understanding and analysis of the location, motion behavior, and dynamic characteristics of noise sources, providing valuable data and information support for research and applications in related fields.

[0150] Preferably, the intrinsic scattering cloud image data is subjected to acoustic energy flow around the center identification and noise source position correspondence to obtain potential noise source position data, including:

[0151] Performing corresponding acoustic energy flow vector field calculation on the intrinsic scattering cloud image data according to a preset time step to obtain acoustic energy flow vector field data;

[0152] In an embodiment of the present invention, acquired intrinsic scattering cloud map data is loaded into a computing environment; a suitable time step is determined based on the data acquisition frequency and actual needs; and an acoustic model or an acoustic finite element method is used to calculate and simulate the acoustic energy flow vector field based on the intrinsic scattering cloud map data.

[0153] Filtering the acoustic energy flow vector field data and extracting the effective energy transmission path to obtain filtered acoustic energy flow field data;

[0154] The present invention applies filtering methods to the acoustic energy flow vector field data to remove noise and smooth the data. A threshold or a specific energy transmission path extraction method is used to extract transmission paths with higher energy, which may correspond to the main transmission direction of the acoustic energy flow.

[0155] The curl tensor of the filtered acoustic energy flow field data is calculated, and the area surrounding the acoustic energy flow is identified to obtain the potential noise source location estimation data;

[0156] This embodiment of the present invention uses a curl calculation method to calculate a curl tensor based on filtered acoustic energy flow field data to describe the rotational characteristics of the acoustic energy flow field. Image processing techniques and object recognition algorithms are then applied to identify the central region of the acoustic energy flow within the filtered acoustic energy flow field, which may correspond to the location of a potential noise source.

[0157] A polar sound ray tracing model is constructed based on the potential noise source location estimation data, and the sound path from each spatial point to the potential noise source location is calculated to obtain the sound path field data;

[0158] This embodiment of the present invention prepares the potential noise source location estimation data obtained in step S513, ensuring that the data format and structure are suitable for subsequent processing. Based on acoustic propagation theory and physical parameters, a polar ray tracing model is constructed to simulate the acoustic energy propagation path. This model can take into account properties of the medium, such as sound velocity, density, and attenuation. Using the polar ray tracing model, the acoustic path from each spatial point to the potential noise source location is calculated. The acoustic path represents the length of the path a sound wave travels through a medium.

[0159] The matching degree of the potential source position is evaluated based on the intrinsic scattering cloud map data and the sound path field data to obtain the potential noise source position data.

[0160] This embodiment of the present invention loads the acquired intrinsic scattering cloud data into a computing environment, ensuring that the data format and structure are suitable for subsequent processing. The sound path field data calculated in step S514 is prepared, ensuring that the data format and structure are suitable for subsequent processing. Appropriate methods are used to compare the intrinsic scattering cloud data with the sound path field data to assess the degree of match between spatial points and potential noise source locations. Common methods include correlation analysis, difference metrics, and statistical methods.

[0161] By calculating the acoustic energy flow vector field on intrinsic scattering cloud map data, the present invention can obtain the propagation direction and energy flow of acoustic energy in space. This helps understand the propagation path and distribution characteristics of acoustic energy. The acoustic energy flow vector field data provides information on the flow of acoustic energy in space and can be used to analyze the propagation path, velocity, and concentration area of acoustic energy. This helps identify the propagation range and impact area of noise sources. By filtering the acoustic energy flow vector field data, noise and interference can be removed, and true and effective acoustic energy flow field information can be extracted. This helps improve the quality and reliability of the data. The filtered acoustic energy flow field data can be used to extract effective energy transmission paths, namely the main channels and paths for acoustic energy propagation. This helps determine the main direction and path of acoustic energy propagation, providing a basis for subsequent analysis. By calculating the curl tensor on the filtered acoustic energy flow field data, the curl information of the acoustic energy flow field can be obtained, namely the rotation and vortex distribution of acoustic energy in space. This helps reveal the center-surrounding characteristics and vortex structure of the acoustic energy flow field. By analyzing the curl information of the acoustic energy flow field, the center-surrounding area of acoustic energy flow can be identified and the location of potential noise sources can be inferred. This helps to preliminarily determine the location of the noise source, providing a reference for subsequent location matching and assessment. By constructing a polar ray tracing model based on the estimated potential noise source location data, we can simulate the propagation path and direction of sound rays in space. This helps understand the propagation path and propagation characteristics of sound energy. Using the polar ray tracing model, we can calculate the acoustic path from each spatial point to the potential noise source location—that is, the distance the sound travels. This helps assess the intensity and attenuation of sound propagation, providing a basis for subsequent location matching and assessment. By comparing and analyzing the intrinsic scattering contour data with the acoustic path field data, we can assess the degree of match of the potential noise source location. This helps determine the exact location of the potential noise source and validates the validity of the inferences made in previous steps. This match assessment yields potential noise source location data, specifically determining the specific location of the noise source in space. This helps locate the noise source and provides a basis for subsequent noise control and management. In summary, the above steps include acoustic energy flow vector field calculation, acoustic energy flow analysis, noise filtering, energy transmission path extraction, curl tensor calculation, noise source position estimation, polar sound ray tracing model construction, sound path calculation, position matching evaluation, and noise source position data acquisition. These effects help to understand the characteristics of acoustic energy propagation, determine the location of noise sources, and evaluate the position matching degree, providing an important basis for noise control and management.

[0162] Preferably, modal contribution analysis is performed on the dynamic source path data based on the vibration modes of the substation equipment, and the radiation modes of the stationary and moving noise sources are separated and projected into the scalar sound pressure field to obtain a noise source distribution map after separation, including:

[0163] Perform vibration modal analysis on substation equipment information and design parameters to obtain equipment vibration modal data;

[0164] This embodiment of the present invention collects detailed information and design parameters of substation equipment, including equipment type, dimensions, and material properties. This ensures data accuracy and completeness. Appropriate vibration modal analysis methods, such as finite element analysis (FEA), are selected. A finite element model of the equipment is created, including its geometry, material properties, and boundary conditions. Vibration modal analysis is then performed to calculate the equipment's resonant frequency, mode shape, and vibration modal data under different modes.

[0165] Match the equipment vibration mode with the source path data in the spatial and frequency domains to obtain a matching score matrix;

[0166] Embodiments of the present invention collect substation-related dynamic source path data, such as sensor data and noise source location data. Equipment vibration modal data is matched with the dynamic source path data. Correlation analysis, spectrum analysis, and other methods can be used. Spatially, the positional relationship between the equipment vibration modes and the dynamic source path is compared to determine the degree of match. In the frequency domain, the spectral characteristics of the equipment vibration modes and the dynamic source path are compared to determine the degree of match. Based on the spatial and frequency domain matching results, a matching score matrix is constructed to record the degree of match between the equipment vibration modes and the dynamic source path.

[0167] According to the matching score matrix, the main contributing vibration modes and their contribution coefficients of each noise source are determined to obtain the modal contribution vector data;

[0168] This embodiment of the present invention determines the primary contributing vibration mode of each noise source based on the matching score matrix. The mode with the highest matching score can be selected as the primary contributing vibration mode. A contribution coefficient is calculated for each noise source, reflecting its contribution to the device vibration mode. The contribution coefficient can be calculated based on the values in the matching score matrix, for example, by normalization.

[0169] The radiation field of each noise source is separated into different modal components according to the modal contribution vector data, and the modal component data of the stationary source and the modal component data of the moving source are obtained;

[0170] This embodiment of the present invention separates the radiation field of each noise source into different modal components. Appropriate separation methods, such as modal filtering and modal superposition, can be used to categorize the separated modal component data into stationary and mobile sources. The stationary source modal component data represents the radiation field distribution generated by the device's own vibration. The mobile source modal component data represents the radiation field distribution of external noise sources related to device vibration.

[0171] Perform modal superposition on the modal component data of the stationary source and the modal component data of the mobile source, and synthesize the total radiation field distribution of each radiation source to obtain the total radiation field data of the stationary source and the total radiation field data of the mobile source;

[0172] The embodiment of the present invention sums up the modal component data of the stationary source, that is, adds up the radiation field data of each modal component; this can be achieved by superimposing the amplitude and phase of each modal component. Usually, the amplitude can be added directly, while the phase needs to be reasonably adjusted; for each modal component, the adjustment of the amplitude and phase can be determined based on the contribution coefficient obtained from the modal contribution vector data. The contribution coefficient reflects the contribution degree of each modal component to the total radiation field. The modal component data of the mobile source are summed up, that is, the radiation field data of each modal component are also added; similar to the modal superposition of the stationary source, it can be achieved by superimposing the amplitude and phase of each modal component; the adjustment of the amplitude and phase can also be determined based on the contribution coefficient obtained from the modal contribution vector data.

[0173] According to the geometric topological relationship matrix, the total radiation field data of the stationary source and the total radiation field data of the mobile source are projected into the three-dimensional substation environment to obtain the separated noise source distribution map.

[0174] In an embodiment of the present invention, the total radiation field data of stationary sources and the total radiation field data of mobile sources are projected onto a three-dimensional substation environment based on a geometric body topological relationship matrix to obtain a separated noise source distribution map. A matrix is prepared to describe the topological relationship between the various geometric bodies of the substation; the rows and columns of the matrix represent different geometric bodies, and the matrix elements represent the connection relationship between the corresponding geometric bodies (such as adjacent, connected, etc.). Three-dimensional environmental projection of the total radiation field data of stationary sources: Based on the geometric body topological relationship matrix, the position and distribution of the total radiation field data of stationary sources in the substation environment are determined; the total radiation field data of stationary sources are projected into the three-dimensional space of the substation, taking into account the positional relationship and boundary conditions between the geometric bodies; three-dimensional visualization software or programming tools can be used to correspond and project the total radiation field data of stationary sources with the geometric model of the substation to generate a noise source distribution map. Three-dimensional environmental projection of the total radiation field data of mobile sources: Determine the location and distribution of the total radiation field data of mobile sources in the substation environment based on the topological relationship matrix of geometric bodies; project the total radiation field data of mobile sources into the three-dimensional space of the substation, taking into account the positional relationship and boundary conditions between geometric bodies; similarly, use three-dimensional visualization software or programming tools to correspond and project the total radiation field data of mobile sources with the geometric model of the substation to generate a noise source distribution map.

[0175] By analyzing the vibration modes of substation equipment, the present invention can understand the vibration characteristics and response frequencies of the equipment under different vibration modes. This helps determine the equipment's natural vibration frequency and modal morphology, providing a foundation for subsequent matching and separation processes. By matching the equipment vibration modes with the dynamic source path data, the correlation and matching degree between each vibration mode and the corresponding dynamic source path can be determined. This helps determine the correlation between noise sources and equipment vibration modes, providing a basis for subsequent modal contribution calculations. By analyzing the matching score matrix, the primary contributing modes and corresponding contribution coefficients of each noise source to the equipment vibration modes can be determined. This helps understand the extent of each noise source's influence on equipment vibration, providing a basis for subsequent modal separation and radiation field analysis. By separating the radiation field of a noise source into different modal components based on the modal contribution vector data, the noise radiation characteristics can be correlated with the equipment vibration modes. This helps understand the contribution of different modes to the radiation field, providing a basis for subsequent modal superposition and total radiation field analysis. By superimposing the modal component data of each stationary source with the modal component data of each moving source, the total radiation field distribution of each radiation source can be synthesized. This helps to comprehensively consider the contribution of different modes to the radiation field and obtain more accurate total radiation field data. By performing a three-dimensional substation environment projection on the total radiation field data of the stationary source and the total radiation field data of the mobile source according to the geometric topological relationship matrix, the position and distribution of the separated noise sources in the substation environment can be visualized. This helps to intuitively understand the distribution of noise sources and provide guidance for further noise control and optimization. In summary, by executing the above steps, the substation noise sources can be separated and analyzed, and equipment vibration modal data, modal contribution vector data, modal component data of stationary sources and mobile sources, total radiation field data, and noise source distribution maps can be obtained. These results are of great significance for the identification, positioning and control of noise sources, and help to improve the noise problems in the substation environment and improve the reliability and operating efficiency of equipment.

[0176] The present invention also provides a low-frequency noise source locating device, comprising:

[0177] The 3D environment modeling module is used to obtain a 3D environment model of each structure in the substation and obtain a topological relationship matrix of the geometric bodies in the 3D environment model; perform on-site measurements of the substation to obtain noise measurement data; and map the noise measurement data to the 3D environment model based on the measurement location and perform registration to obtain local noise image data. This module includes the following submodules:

[0178] 3D modeling acquisition module, used to obtain 3D environmental model data of each structure in the substation;

[0179] The geometry extraction module is used to extract and represent the geometry of the three-dimensional environment model, and the geometry represents the model data;

[0180] A topological relationship extraction module is used to extract the topological relationship and boundary conditions of the geometric bodies in the model based on the three-dimensional environment model data and the geometric body representation model data, and obtain the geometric body topological relationship matrix;

[0181] On-site measurement module, used to conduct on-site measurement of the substation to obtain noise measurement data, including noise decibel data and measurement point location data;

[0182] A wavelet filtering module is used to perform wavelet transform filtering on the noise decibel data to obtain noise image data;

[0183] A data registration module is used to match the noise image data blocks to the three-dimensional environment model according to the measurement point position data to obtain block noise data;

[0184] An image registration module is used to perform image registration based on the block noise data and the model units in the three-dimensional environment model to obtain local noise image data;

[0185] The noise measurement module is used to perform outer convolution fusion on the noise images between adjacent units based on the geometric topological relationship matrix and the local noise image data, and iteratively propagate them to the entire substation area to generate initial sound source distribution estimation data for the entire substation. This module includes the following submodules:

[0186] The spatial discretization module is used to discretize the substation space into small spatial unit grids according to the geometric topological relationship matrix to obtain spatial grid unit data;

[0187] A noise assignment module is used to assign local noise image data to spatial units of spatial grid unit data according to measurement point position data to obtain initial grid noise distribution data;

[0188] The convolution fusion module is used to perform outer convolution of the initial grid noise distribution data and the geometric topology relationship matrix on the noise of adjacent grid cells based on boundary conditions to obtain the fusion result data;

[0189] The noise propagation module is used to use the fusion result data as the initial value and repeat the outer convolution operation based on the boundary conditions until it propagates to the entire substation area to obtain the grid noise distribution data;

[0190] The interpolation and denoising module is used to interpolate and denoise the grid noise distribution data to obtain the initial sound source distribution estimation data;

[0191] The noise distribution estimation module is used to separate the direct sound signal and multipath echo components based on the initial sound source distribution estimation data to generate echo decoupling data. This module includes the following submodules:

[0192] A parameterized modeling module is used to perform parameterized modeling on the initial sound source distribution estimation data using a generalized coordinated filtering algorithm, and to divide the model into two parts: a direct sound component and a multipath sound component, thereby obtaining a direct sound model and a multipath sound model;

[0193] An optimization solution module is used to construct a joint optimization objective function based on the direct sound model and the multipath sound model, balance the two models through a coordination factor, and solve the objective function using the generalized least squares method to obtain the separated direct sound component data and multipath sound component data;

[0194] The path tracking module is used to construct a random state space model of the multipath echo path based on the multipath acoustic component data, and filter and track each path based on the Kalman particle filter framework, and perform statistics on the path delay, attenuation and direction to obtain preliminary multipath path parameter data;

[0195] The noise sparse representation module is used to perform sparse representation on the multipath sound component data on the wavelet basis to obtain a noise sparse representation model;

[0196] An adaptive modeling module is used to adaptively model the multipath noise based on the noise sparse representation model and the preliminary multipath path parameter data, and to subtract the multipath noise component from the direct sound component data to obtain the multipath noise-removed field data;

[0197] The residual noise manifold module is used to construct a residual noise manifold for the residual noise of the multipath denoised field data using manifold learning technology, and map the multipath denoised field data onto the noise manifold to obtain reconstructed residual multipath noise component data;

[0198] A noise decoupling module is used to subtract the reconstructed residual multipath noise component data from the multipath de-noised field data to obtain echo decoupling data;

[0199] The echo decoupling module is used to perform dimensionality compression on the echo decoupling data and extract the eigenmodes of the low-order scattered field to obtain the eigenmode set of the scattered field; it also estimates the parameters of the field mode of the eigenmode set of the scattered field to obtain the intrinsic scattering cloud map data; this module includes the following submodules:

[0200] The kernel tensor representation module is used to discretely sample the echo decoupling data in the spatiotemporal domain and arrange it into a kernel tensor form to obtain kernel tensor representation data;

[0201] The low-rank decomposition module is used to perform low-rank decomposition approximation on the core tensor representation data using supersymmetric matrix decomposition technology to obtain low-dimensional embedded vector data;

[0202] A pattern reconstruction module is used to map the low-dimensional embedded vector data onto the differential manifold and reconstruct the eigenmode of the scattered field to obtain the eigenmode set of the scattered field;

[0203] The parameter estimation module is used to estimate the parameters of the field mode of the scattering field eigenmode set through a nonlinear regression model to obtain the eigenscattering cloud map data;

[0204] The dynamic source path analysis module is used to identify the acoustic energy flow around the center of the intrinsic scattering cloud data and match the noise source position to obtain potential noise source position data; based on the potential noise source position data, the instantaneous motion direction of the source point is locked and the motion trajectory of the source point is reconstructed to obtain noise source motion trajectory data; based on the noise source motion trajectory data and the potential noise source position data, the position information of the stationary source and the moving source is distinguished and integrated to obtain dynamic source path data; this module includes the following submodules:

[0205] The acoustic energy flow analysis module is used to identify the center of the acoustic energy flow around the intrinsic scattering cloud data and to correspond the noise source position to obtain the potential noise source position data;

[0206] The motion direction locking module is used to calculate the acoustic energy flow gradient vector based on the potential noise source position data and the filtered acoustic energy flow field data to obtain the instantaneous motion direction data of the source point;

[0207] The motion trajectory reconstruction module is used to reconstruct the source point motion trajectory based on the time series analysis of the source point instantaneous motion direction data to obtain the noise source motion trajectory data;

[0208] The static and dynamic source separation module is used to separate the static source and the mobile source of the noise source motion trajectory data according to the preset trajectory classification standard and the potential noise source position data, and obtain the mobile source trajectory data and the static source position data;

[0209] The path integration module is used to merge the moving source trajectory data and the stationary source position data to obtain the moving source path data;

[0210] The modal contribution analysis module is used to perform modal contribution analysis on the dynamic source path data based on the vibration modes of the substation equipment, separate the radiation modes of the stationary and moving noise sources, project them into the scalar sound pressure field, and obtain the separated noise source distribution map. This module includes the following submodules:

[0211] The vibration modal analysis module is used to perform vibration modal analysis on the substation equipment information and design parameters to obtain equipment vibration modal data;

[0212] The modal matching module is used to match the equipment vibration mode with the source path data in the spatial and frequency domains to obtain a matching score matrix;

[0213] The contribution calculation module is used to determine the main contributing vibration mode and its contribution coefficient of each noise source according to the matching score matrix, and obtain the modal contribution vector data;

[0214] The modal separation module is used to separate the radiation field of each noise source into different modal components according to the modal contribution vector data, and obtain the modal component data of the stationary source and the modal component data of the moving source;

[0215] The radiation field synthesis module is used to perform modal superposition on the modal component data of the stationary source and the modal component data of the mobile source, and synthesize the total radiation field distribution of each radiation source to obtain the total radiation field data of the stationary source and the total radiation field data of the mobile source;

[0216] The three-dimensional projection module is used to perform three-dimensional substation environment projection on the total radiation field data of the stationary source and the total radiation field data of the mobile source according to the geometric topological relationship matrix to obtain the separated noise source distribution map.

[0217] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0218] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for locating low-frequency noise sources in a substation, characterized in that: include: Obtaining a three-dimensional environmental model of each structure in the substation, and obtaining a topological relationship matrix of geometric bodies in the three-dimensional environmental model; Conduct on-site measurements of the substation to obtain noise measurement data. Based on the measurement location, the noise measurement data is mapped to the 3D environmental model and registered to obtain local noise image data. Performing outer convolution fusion on the noise images between adjacent units according to the geometric topological relationship matrix and the local noise image data, and iteratively propagating them to the entire substation area to generate initial sound source distribution estimation data for the entire substation; The direct sound signal and multipath echo components are separated based on the initial sound source distribution estimation data to generate echo decoupling data, specifically: The generalized coordinated filtering algorithm is used to perform parameterized modeling on the initial sound source distribution estimation data. The model is divided into two parts: the direct sound component and the multipath sound component, thus obtaining the direct sound model and the multipath sound model. A joint optimization objective function is constructed based on the direct sound model and the multipath sound model. The two models are weighed by a coordination factor, and the objective function is solved using the generalized least squares method to obtain the separated direct sound component data and multipath sound component data. A random state space model of the multipath echo path is constructed based on the multipath acoustic component data. Each path is filtered and tracked based on the Kalman particle filter framework. The path delay, attenuation, and direction are statistically analyzed to obtain preliminary multipath path parameter data. The multipath acoustic component data is sparsely represented on the wavelet basis to obtain a noise sparse representation model; Adaptively model the multipath noise based on the noise sparse representation model and preliminary multipath path parameter data, and subtract the multipath noise component from the direct sound component data to obtain the multipath noise-removed field data; The residual noise manifold is constructed for the residual noise of the multipath denoised field data using manifold learning technology, and the multipath denoised field data is mapped onto the noise manifold to obtain the reconstructed residual multipath noise component data. subtracting the reconstructed residual multipath noise component data from the multipath de-noised field data to obtain echo decoupling data; Perform dimension compression on the echo decoupling data and extract the eigenmodes of the low-order scattered field to obtain the eigenmode set of the scattered field; Estimating the parameters of the field pattern for the eigenmode set of the scattered field to obtain the eigenscattering cloud map data; The intrinsic scattering cloud image data is used to identify the acoustic energy flow around the center and to match the noise source position to obtain the potential noise source position data; based on the potential noise source position data, the instantaneous motion direction of the source point is locked and the motion trajectory of the source point is reconstructed to obtain the noise source motion trajectory data; Based on the noise source motion trajectory data and potential noise source location data, the location information of the stationary source and the moving source are distinguished and integrated to obtain the moving source path data; The modal contribution analysis of the dynamic source path data is performed based on the vibration modes of the substation equipment, and the radiation modes of the stationary and moving noise sources are separated and projected into the scalar sound pressure field to obtain the separated noise source distribution map.

2. The method for locating a low-frequency noise source in a substation according to claim 1, wherein: Obtaining a three-dimensional environmental model of each structure in the substation, and obtaining a topological relationship matrix of geometric bodies in the three-dimensional environmental model; performing on-site measurement of the substation to obtain noise measurement data; Based on the measurement location, the noise measurement data is mapped to the 3D environment model in blocks and registered to obtain local noise image data, including: Obtain 3D environmental model data of each structure in the substation; Extract and represent the geometry of the three-dimensional environment model, and the geometry represents the model data; Extract the topological relationship and boundary conditions of the geometric bodies in the model based on the 3D environment model data and the geometric body representation model data to obtain the geometric body topological relationship matrix; Conduct on-site measurements of substations to obtain noise measurement data, including noise decibel data and measurement point location data; Perform wavelet transform filtering on the noise decibel data to obtain noise image data; According to the measurement point position data, the noise image data is divided into blocks and mapped to the three-dimensional environment model to obtain block noise data; Image registration is performed based on the block noise data and the model units in the three-dimensional environment model to obtain local noise image data.

3. The method for locating a low-frequency noise source in a substation according to claim 2, characterized in that: The topological relationship and boundary conditions of the geometric bodies in the model are extracted based on the 3D environment model data and the geometric body representation model data to obtain the geometric body topological relationship matrix, including: Performing basic shape type recognition on the geometric body representation model data to obtain geometric body type data; Extract acoustic impedance and absorption coefficient based on material information based on 3D environment model data and geometric representation model data to obtain geometric boundary condition data; Performing topological relationship analysis of adjacency, inclusion, separation, and intersection between geometric bodies based on geometric body representation model data and geometric body type data to obtain geometric body topological relationship data; Based on the geometric body topological relationship data and the geometric body boundary condition data, the topological relationship and boundary conditions are integrated into a matrix form to obtain the geometric body topological relationship matrix.

4. The method for locating a low-frequency noise source in a substation according to claim 3, wherein: The noise images between adjacent units are fused by outer convolution according to the geometric topological relationship matrix and the local noise image data, and iteratively propagated to the entire substation area to generate the initial sound source distribution estimation data of the entire substation, including: The substation space is discretized into small space unit grids according to the geometric topological relationship matrix to obtain space grid unit data; Assigning local noise image data to spatial units of spatial grid unit data according to measurement point position data to obtain initial grid noise distribution data; The initial grid noise distribution data and the geometric topology relationship matrix perform outer convolution on the noise of adjacent grid cells based on boundary conditions to obtain the fusion result data; The fusion result data is used as the initial value, and the outer convolution operation based on the boundary conditions is repeated until it propagates to the entire substation area to obtain the grid noise distribution data; The grid noise distribution data is interpolated and denoised to obtain the initial sound source distribution estimation data.

5. The method for locating a low-frequency noise source in a substation according to claim 4, characterized in that: Perform dimension compression on the echo decoupling data and extract the eigenmodes of the low-order scattered field to obtain the eigenmode set of the scattered field; Estimating the parameters of the field pattern of the scattered field eigenmode set to obtain the intrinsic scattering cloud data, including: The echo decoupling data is discretely sampled in the spatiotemporal domain and arranged into a kernel tensor form to obtain the kernel tensor representation data; The supersymmetric matrix decomposition technology is used to perform low-rank decomposition approximation on the core tensor representation data to obtain low-dimensional embedded vector data; Mapping the low-dimensional embedded vector data onto the differential manifold and reconstructing the eigenmode of the scattered field to obtain the eigenmode set of the scattered field; The parameters of the eigenmode set of the scattered field are estimated through a nonlinear regression model to obtain the intrinsic scattering cloud map data.

6. The method for locating a low-frequency noise source in a substation according to claim 5, characterized in that: The parameters of the scattering field eigenmode set are estimated through nonlinear regression model to obtain the intrinsic scattering cloud data, including: The nonlinear scattering model of eigenmode theory is established based on the eigenmode set of the scattered field and the topological relationship matrix of the geometric body; Identify the free parameters of wave number, attenuation rate and phase in the nonlinear scattering model and obtain the free parameter set; According to the preset measured pattern data, the eigenmode set of the scattered field and the difference of the nonlinear scattering model are constructed as an optimization objective function to obtain a nonlinear regression objective function; Solve the model parameters of the nonlinear regression objective function according to the free parameter set to obtain the free parameter estimation data; Substitute the free parameter estimation data into the nonlinear scattering model to generate intrinsic scattering cloud data.

7. The method for locating a low-frequency noise source in a substation according to claim 6, characterized in that: The intrinsic scattering cloud image data is used to identify the acoustic energy flow around the center and to match the noise source position to obtain the potential noise source position data; based on the potential noise source position data, the instantaneous motion direction of the source point is locked and the motion trajectory of the source point is reconstructed to obtain the noise source motion trajectory data; Based on the noise source motion trajectory data and potential noise source location data, the location information of the stationary source and the moving source are distinguished and integrated to obtain the moving source path data, including: The acoustic energy flow around the center is identified on the intrinsic scattering cloud image data, and the noise source position is matched to obtain the potential noise source position data; According to the potential noise source position data and the filtered acoustic energy flow field data, the acoustic energy flow gradient vector based on the source position is calculated to obtain the instantaneous motion direction data of the source point; The instantaneous motion direction data of the source point is reconstructed based on the time series analysis of the source point motion trajectory to obtain the noise source motion trajectory data; According to the preset trajectory classification standard and the potential noise source location data, the noise source motion trajectory data is separated into stationary sources and mobile sources to obtain mobile source trajectory data and stationary source location data; The moving source trajectory data and the stationary source position data are merged to obtain the moving source path data.

8. The method for locating a low-frequency noise source in a substation according to claim 7, characterized in that: The intrinsic scattering cloud image data is used to identify the acoustic energy flow around the center and to correspond to the noise source position to obtain the potential noise source location data, including: Performing corresponding acoustic energy flow vector field calculation on the intrinsic scattering cloud image data according to a preset time step to obtain acoustic energy flow vector field data; Filtering the acoustic energy flow vector field data and extracting the effective energy transmission path to obtain filtered acoustic energy flow field data; The curl tensor of the filtered acoustic energy flow field data is calculated, and the area surrounding the acoustic energy flow is identified to obtain the potential noise source location estimation data; A polar sound ray tracing model is constructed based on the potential noise source location estimation data, and the sound path from each spatial point to the potential noise source location is calculated to obtain the sound path field data; The matching degree of the potential source position is evaluated based on the intrinsic scattering cloud map data and the sound path field data to obtain the potential noise source position data.

9. The method for locating a low-frequency noise source in a substation according to claim 8, wherein: Based on the vibration modes of the substation equipment, the modal contribution analysis of the dynamic source path data is performed, and the radiation modes of the stationary and moving noise sources are separated and projected into the scalar sound pressure field to obtain the separated noise source distribution map, including: Perform vibration modal analysis on substation equipment information and design parameters to obtain equipment vibration modal data; Match the equipment vibration mode with the source path data in the spatial and frequency domains to obtain a matching score matrix; According to the matching score matrix, the main contributing vibration modes and their contribution coefficients of each noise source are determined to obtain the modal contribution vector data; The radiation field of each noise source is separated into different modal components according to the modal contribution vector data, and the modal component data of the stationary source and the modal component data of the moving source are obtained; Perform modal superposition on the modal component data of the stationary source and the modal component data of the mobile source, and synthesize the total radiation field distribution of each radiation source to obtain the total radiation field data of the stationary source and the total radiation field data of the mobile source; According to the geometric topological relationship matrix, the total radiation field data of the stationary source and the total radiation field data of the mobile source are projected into the three-dimensional substation environment to obtain the separated noise source distribution map.

Images