A method for optimizing design of a soundproof cover for low-frequency noise of a transformer

Abstract

The present application relates to the technical field of noise control design, in particular to a kind of transformer low-frequency noise sound shield optimization design method, comprising: acquisition transformer rated operation under low-frequency noise spectrum data set, including octave sound pressure level, phase angle sequence and box vibration acceleration data.Improved principal component analysis algorithm based on eigenvalue cumulative contribution rate dynamically determines the number of principal components, and the data set is dimensionally compressed to generate noise source characteristic vector set.Characteristic vector is input into pre-training sound shield parameter optimization model to solve candidate structure parameter combination, and low-frequency fitting effect avoidance correction is completed relying on phase angle sequence and panel bending wave wave number matching relationship, to generate optimization structure parameter configuration table, according to which the sound shield manufacturing process guidance data containing panel layering sequence and reinforcing rib arrangement spacing is output.The method realizes the fine processing of low-frequency noise data and the intelligent optimization of sound shield parameters, and improves the low-frequency noise suppression capability and the design construction adaptability.

Description

Technical Field

[0001] This invention relates to the field of noise control design technology, and in particular to an optimized design method for a soundproof enclosure for low-frequency noise in transformers. Background Technology

[0002] Transformers generate continuous low-frequency noise under long-term rated operation. Soundproof enclosures are the main structural form for controlling the propagation of low-frequency noise in engineering. Traditional soundproof enclosure designs only collect noise sound pressure level data, without simultaneously collecting octave band center frequency sound pressure level, frequency phase angle sequence, and transformer tank vibration acceleration related monitoring data, resulting in incomplete noise source information. Conventional principal component analysis algorithms use a fixed number of principal components and cannot adaptively select effective dimensions based on the data's own distribution characteristics, making it impossible to effectively reduce dimensionality and extract features from high-dimensional noise spectrum data.

[0003] The selection of structural parameters for existing soundproof enclosures relies on empirical design and lacks an intelligent parameter optimization model with a multi-objective solution process. Consequently, the structural parameters cannot serve as the optimal matching solution under numerous quiet acoustic constraints. The design process neglects the impact of low-frequency matching effects on the sound insulation performance of the enclosure and fails to utilize the matching correlation between the phase angle sequence and the bending wave number of the enclosure panel for parameter correction, making low-frequency acoustic mismatches in the structural parameters prone to occur.

[0004] Traditional design processes lack process guidance data that can directly guide production, and there are no standardized process parameters related to panel layer stacking sequence and stiffener spacing. The overall design mode remains at the level of experience-based design and simple fitting verification, with parameter optimization disconnected from actual manufacturing processes. It cannot achieve integrated processing of multi-dimensional noise data analysis, intelligent parameter optimization, low-frequency effect correction, and process data generation. The industry urgently needs to establish a multi-dimensional noise and vibration synchronous acquisition and processing model, and construct a parameter optimization and correction mechanism adapted to low-frequency acoustic characteristics to meet the application requirements of high-precision design, acoustic performance compliance, and direct manufacturing capability for transformer low-frequency noise soundproof enclosures. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an optimized design method for a soundproof enclosure to reduce low-frequency noise in transformers.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for optimizing the design of a soundproof enclosure for low-frequency noise in transformers, comprising:

[0007] The low-frequency noise spectrum data set of the target transformer under rated operating conditions is collected. The low-frequency noise spectrum data set includes sound pressure level measurements corresponding to multiple octave band center frequencies, phase angle sequences of each frequency component, and vibration acceleration monitoring data of the transformer tank.

[0008] An improved principal component analysis algorithm is applied to the low-frequency noise spectrum data set for dimensionality reduction and compression to generate a set of noise source feature vectors for the target transformer. The improved principal component analysis algorithm dynamically adjusts the number of retained principal components based on the cumulative contribution rate of the feature values.

[0009] The noise source feature vector set is input into a pre-trained soundproof enclosure parameter optimization model for multi-objective optimization to generate candidate structural parameter combinations for the soundproof enclosure.

[0010] The candidate structural parameter combinations are subjected to low-frequency coincidence effect avoidance correction processing to generate an optimized structural parameter configuration table.

[0011] Based on the optimized structural parameter configuration table, a soundproof enclosure manufacturing process guidance data set is generated. The soundproof enclosure manufacturing process guidance data set includes the stacking sequence scheme of each panel and the arrangement spacing scheme of the internal reinforcing ribs.

[0012] As a further aspect of the present invention, an improved principal component analysis algorithm is applied to the low-frequency noise spectrum data set for dimensionality reduction and compression to generate a set of noise source feature vectors for the target transformer, including:

[0013] The sound pressure level measurements corresponding to each octave center frequency in the low-frequency noise spectrum data set are arranged in ascending order of frequency to form the original sound pressure level sequence. The phase angle sequence is then aligned and arranged in the same order as the octave center frequency to form the original phase angle sequence.

[0014] Perform complex domain mapping processing on the original sound pressure level sequence and the original phase angle sequence, and combine the sound pressure level measurement value of each frequency point with the corresponding phase angle to construct a data point in complex form, thereby obtaining a complex domain noise data matrix;

[0015] Calculate the covariance matrix of the complex domain noise data matrix to obtain the complex covariance matrix. Each element of the complex covariance matrix is ​​composed of the complex correlation coefficient between the corresponding frequency points.

[0016] The complex covariance matrix is ​​subjected to eigenvalue decomposition to obtain an eigenvalue sequence and an eigenvector corresponding to each eigenvalue. All eigenvalues ​​in the eigenvalue sequence are extracted and sorted in descending order of value.

[0017] The cumulative contribution rate is recursively calculated for the sorted feature value sequence. Starting from the largest feature value, the feature values ​​are added sequentially and divided by the sum of the feature values ​​to obtain the cumulative contribution rate corresponding to each accumulation step. The accumulation step corresponding to the first time the cumulative contribution rate reaches the preset contribution rate threshold is determined as the number of principal components to be retained.

[0018] The eigenvectors corresponding to the eigenvalues ​​located at the first retained principal component position in the sorted eigenvalue sequence are selected to form a projection transformation matrix. The complex domain noise data matrix and the projection transformation matrix are multiplied to obtain the dimension-reduced complex feature matrix.

[0019] The real part of each complex element is extracted from the dimensionality-reduced complex feature matrix as the principal component amplitude and the imaginary part as the principal component phase. All principal component amplitudes and principal component phases are concatenated row by row to generate the noise source feature vector set.

[0020] As a further aspect of the present invention, the covariance matrix of the complex domain noise data matrix is ​​calculated to obtain a complex covariance matrix, wherein each element of the complex covariance matrix is ​​composed of complex correlation coefficients between corresponding frequency points, including:

[0021] Extract all row vectors from the complex domain noise data matrix. Each row vector represents a complex observation sequence corresponding to a frequency point. The complex observation sequence consists of complex sound pressure values ​​corresponding to multiple time sampling points.

[0022] For any two different frequency points corresponding to complex observation sequences, calculate the complex correlation coefficient between the two complex observation sequences. The formula for calculating the complex correlation coefficient is: multiply the conjugate sequences of the first and second complex observation sequences point by point and sum them up, then divide by the square root of the product of the sum of the magnitudes of the first and second complex observation sequences.

[0023] Traverse all frequency pairs in the complex domain noise data matrix, perform complex correlation coefficient calculation for each frequency pair, and generate a complex correlation coefficient value.

[0024] All complex correlation coefficient values ​​are filled into a symmetric matrix according to the row and column order of frequency point pairs. The row index and column index of the symmetric matrix correspond to the frequency point numbers in the complex domain noise data matrix, thus forming a complex covariance matrix.

[0025] As a further aspect of the present invention, before performing eigenvalue decomposition on the complex covariance matrix, the method further includes:

[0026] Calculate the magnitude of each row vector in the complex domain noise data matrix, mark the row vectors with a magnitude less than a preset noise floor threshold as low-energy row vectors, and mark the row vectors with a magnitude greater than or equal to the preset noise floor threshold as high-energy row vectors.

[0027] Extract all high-energy row vectors to form a high-energy submatrix, and extract all low-energy row vectors to form a low-energy submatrix.

[0028] Random truncation is performed on the low-energy submatrix. Row vectors with a preset retention ratio are randomly extracted from the low-energy submatrix. The extracted row vectors are then merged with the high-energy submatrix row by row to form a filtered complex domain noise data matrix.

[0029] The filtered complex domain noise data matrix is ​​used to replace the complex domain noise data matrix for subsequent calculation of the covariance matrix.

[0030] Calculate the total number of row vectors in the filtered complex domain noise data matrix. If the total number of row vectors is less than the preset minimum number of samples, cancel the random truncation process and directly use the original complex domain noise data matrix.

[0031] As a further aspect of the present invention, the noise source feature vector set is input into a pre-trained soundproof enclosure parameter optimization model for multi-objective optimization to generate candidate structural parameter combinations for the soundproof enclosure, including:

[0032] The candidate structural parameter combination includes the thickness distribution parameters of each panel of the soundproof enclosure, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material.

[0033] The principal component amplitudes in the noise source feature vector set are used as the first input group and input to the sound source intensity mapping layer of the soundproof enclosure parameter optimization model. The sound source intensity mapping layer maintains a frequency-amplitude response curve library. The frequency-amplitude response curve library records the equivalent sound source intensity values ​​corresponding to different principal component amplitude intervals. The corresponding equivalent sound source intensity value is read from the frequency-amplitude response curve library according to the principal component amplitude interval in which the principal component amplitude is located.

[0034] The principal component phase in the noise source feature vector set is used as the second input group and input to the sound source distribution and localization layer of the soundproof enclosure parameter optimization model. The sound source distribution and localization layer maintains a phase-position mapping grid. The phase-position mapping grid divides the value range of the principal component phase into multiple sector intervals. Each sector interval is associated with a preset sound source spatial coordinate region. The sound source spatial coordinates corresponding to the principal component phase are determined according to the sector interval into which the principal component phase falls.

[0035] After the sound source intensity mapping layer and the sound source distribution and positioning layer have completed the mapping respectively, the equivalent sound source intensity value corresponding to each frequency point is paired with the sound source spatial coordinates to form a spatial distribution sound source list. Each entry in the spatial distribution sound source list contains a frequency identifier, a sound source intensity value and a spatial coordinate triplet.

[0036] The list of spatially distributed sound sources is input into the cover parameter iterative search layer of the soundproof cover parameter optimization model. The cover parameter iterative search layer maintains a parameter search tree. Each node of the parameter search tree represents a combination of the thickness distribution parameters of the soundproof cover, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material.

[0037] Starting from the root node of the parameter search tree, calculate the estimated insertion loss of the soundproof enclosure parameter combination corresponding to the root node under the excitation of the spatially distributed sound source list. Use the estimated insertion loss obtained in this calculation as the fitness score of the root node. Select a preset number of child nodes from high to low according to the fitness score to enter the next layer of search.

[0038] Repeat the node expansion and fitness score calculation operations until the preset search depth is reached or the optimal fitness score no longer increases after a preset number of iterations. Then, output the combination of soundproof enclosure parameters stored on the node corresponding to the current optimal fitness score as the candidate structure parameter combination.

[0039] As a further aspect of the present invention, the calculation of the insertion loss prediction value of the soundproof enclosure parameter combination corresponding to the root node under the excitation of the spatially distributed sound source list includes:

[0040] Read the thickness distribution parameters, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material from the root node. Determine the surface density and bending stiffness of each panel of the soundproof enclosure based on the thickness distribution parameters. Determine the air layer thickness based on the depth parameters of the internal cavity. Determine the flow resistance and porosity of the sound-absorbing material based on the filling density parameters of the sound-absorbing material.

[0041] Extract the frequency identifier and spatial coordinates of the corresponding sound source of the current entry, compare the spatial coordinates of the sound source corresponding to the current entry with the spatial positions of each preset panel of the soundproof enclosure, and determine the panel number of the soundproof enclosure that is mainly excited by the current sound source.

[0042] For the frequency identifier of the current entry, calculate the sound transmission loss in the quality control area, stiffness control area, and coincidence effect area of ​​the panel excited by the current sound source at the frequency corresponding to the frequency identifier. Then, splice the calculated sound transmission loss in the quality control area, stiffness control area, and coincidence effect area according to frequency segments to form the single-layer sound transmission loss curve of the current excited panel.

[0043] When the panel thickness distribution parameter indicates a multi-layer composite structure, the acoustic impedance of each layer of material is calculated by series superposition to obtain the composite sound transmission loss curve of the multi-layer composite structure.

[0044] The air layer thickness of the internal cavity and the flow resistance of the sound-absorbing material are input into the preset cavity sound absorption calculation module to obtain the sound absorption coefficient of the cavity corresponding to the current entry frequency identifier; the composite transmission loss curve of the multi-layer composite structure is superimposed with the sound absorption coefficient obtained in this calculation by octave band to obtain the estimated sound insulation value of the current excited panel under the current sound source excitation.

[0045] The energy average of the sound insulation estimates for all entries is used as the estimated insertion loss.

[0046] As a further aspect of the present invention, the candidate structural parameter combinations are subjected to low-frequency coincidence effect avoidance correction processing to generate an optimized structural parameter configuration table, including:

[0047] The thickness distribution parameters of each panel are extracted from the candidate structural parameter combination. The bending wave number frequency characteristic curve of each panel is calculated based on the thickness distribution parameters. The coincidence effect initiation frequency value of each panel is extracted from the bending wave number frequency characteristic curve.

[0048] The dominant frequency components whose sound pressure level measurements exceed a preset sound pressure threshold are extracted from the low-frequency noise spectrum data set and a dominant frequency list is formed. The matching effect starting frequency value of each panel is compared with each frequency value in the dominant frequency list one by one.

[0049] When the absolute value of the difference between the starting frequency value of the matching effect of a certain panel and a certain frequency value in the dominant frequency list is less than the preset matching frequency tolerance threshold, the corresponding panel is marked as the target panel whose matching effect needs to be corrected.

[0050] The target panel is subjected to thickness offset adjustment processing. The thickness value in the thickness distribution parameter of the target panel is increased by a preset thickness adjustment step size. The bending wave number frequency characteristic curve after the thickness is increased is recalculated to obtain the adjusted coincidence effect starting frequency value. The thickness offset adjustment processing is repeated until the absolute value of the difference between the adjusted coincidence effect starting frequency value and all dominant frequency values ​​is greater than or equal to the preset coincidence frequency tolerance threshold.

[0051] The thickness distribution parameters of all panels after thickness offset adjustment, the original depth parameters of the internal cavity, and the original filling density parameters of the sound-absorbing material are recombined to obtain the first corrected combination of structural parameters.

[0052] The depth parameter of the internal cavity in the first corrected combination of structural parameters is fine-tuned. The quarter-wavelength resonance depth is calculated based on the lowest frequency value in the list of dominant frequencies. The depth parameter of the internal cavity is adjusted to a value that is close to the quarter-wavelength resonance depth and has the smallest absolute value of deviation from the original depth parameter, thus obtaining the second corrected combination of structural parameters.

[0053] The parameters in the second revised structural parameter combination are organized into a structured parameter table according to the panel number and hierarchical order. Each row of the structured parameter table corresponds to a panel or a cavity layer, and each column corresponds to a parameter type. The structured parameter table serves as the optimized structural parameter configuration table.

[0054] As a further aspect of the present invention, before recombining the thickness distribution parameters of all panels after thickness offset adjustment, the original depth parameters of the internal cavity, and the original filling density parameters of the sound-absorbing material, the method further includes:

[0055] After performing thickness offset adjustment on the target panel, record the total cumulative increase in the thickness of the target panel, and determine whether the total cumulative increase in thickness exceeds the preset maximum allowable thickness increment.

[0056] When the cumulative increase in thickness exceeds the preset maximum allowable thickness increment, stop the thickness offset adjustment process for the target panel and mark the target panel as an uncorrectable panel;

[0057] Collect all panel numbers marked as uncorrectable panels. For each uncorrectable panel, query the thickness distribution parameters of the adjacent panels from the optimized structural parameter configuration table, and copy the thickness distribution parameters of the adjacent panels to the uncorrectable panel as alternative thickness values.

[0058] The replacement thickness value is recombined with the original internal cavity depth parameter and the original sound-absorbing material filling density parameter to form the corrected structural parameters for the location of the uncorrectable panel.

[0059] As a further aspect of the present invention, a set of manufacturing process guidance data for the soundproof enclosure is generated based on the optimized structural parameter configuration table, including:

[0060] Read the thickness distribution parameters of each panel from the optimized structural parameter configuration table, determine the grade and specifications of the board material required for each panel based on the thickness distribution parameters, and group panels with the same grade and specifications into the same board cutting batch to form a board cutting batch list.

[0061] Read the depth parameters of the internal cavity from the optimized structural parameter configuration table, calculate the required height of the support frame for each cavity position based on the depth parameters of the internal cavity, and group the support frames whose height dimensions differ from the preset size tolerance threshold into the same welding fixture group to form a frame welding process grouping table.

[0062] Read the filling density parameters of the sound-absorbing material from the optimized structural parameter configuration table, calculate the required weight and laying thickness of the sound-absorbing material for each cavity area based on the filling density parameters, and group areas with similar sound-absorbing material weight and laying thickness into the same filling operation batch to form a sound-absorbing material filling operation instruction sheet.

[0063] For each batch in the sheet material cutting batch list, extract the total weight and total cutting length of the sheet material in the current batch, calculate the minimum cutting power of the required cutting equipment based on the total weight of the sheet material, estimate the cutting time based on the total cutting length, and append the minimum cutting power and cutting time to the cutting process parameter column of the current batch to generate a sheet material cutting batch list with cutting parameters.

[0064] For each welding fixture group in the frame welding process grouping table, extract the number of supporting frames and the total weld length in the current welding fixture group, determine the welding sequence number based on the number of supporting frames in the current welding fixture group, estimate the welding time based on the total weld length in the current welding fixture group, and append the welding sequence number and welding time to the welding process parameter column of the current welding fixture group to generate a frame welding process grouping table with welding parameters.

[0065] The batch list of sheet metal cutting parameters, the grouping table of frame welding processes with welding parameters, and the work instruction sheet for filling sound-absorbing materials are arranged in chronological order of the manufacturing process. The three documents together constitute the manufacturing process guidance data set for the soundproof enclosure.

[0066] As a further aspect of the present invention, the method further includes:

[0067] After the soundproof enclosure is manufactured according to the soundproof enclosure manufacturing process guidance data set and installed on the target transformer, the actual low-frequency noise spectrum data at the preset measurement points outside the enclosure is collected within the preset verification time window.

[0068] The actual low-frequency noise spectrum data is compared with the sound pressure level measurement value at the corresponding frequency point in the low-frequency noise spectrum data set by calculating the difference at each frequency point to obtain the actual noise reduction sequence at each frequency point.

[0069] The actual noise reduction sequence is compared with the theoretical noise reduction sequence calculated according to the optimized structural parameter configuration table, and the noise reduction deviation value at each frequency point is calculated. Frequency points whose noise reduction deviation value exceeds the preset deviation threshold are marked as deviation frequency points.

[0070] The percentage of the number of deviation frequency points out of the total number of frequency points is counted. When the percentage is greater than the preset re-optimization trigger threshold, the noise reduction deviation value and deviation sign of each deviation frequency point are extracted from the frequency value corresponding to the deviation frequency point.

[0071] Based on the noise reduction deviation value and deviation sign at each deviation frequency point, the panel thickness distribution parameter corresponding to the current deviation frequency point in the optimized structural parameter configuration table is reverse-adjusted. The direction of the reverse adjustment is opposite to the deviation sign, and the magnitude of the reverse adjustment is proportional to the noise reduction deviation value, thus obtaining the re-optimized structural parameter configuration table.

[0072] Using the re-optimized structural parameter configuration table as the new optimized structural parameter configuration table, the step of generating the soundproof enclosure manufacturing process guidance data set is executed again to produce a revised version of the manufacturing process guidance data set.

[0073] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0074] By dynamically adjusting the number of retained principal components based on the cumulative contribution rate of eigenvalues, dimensionality reduction and compression processing is performed on the transformer low-frequency noise spectrum data set. This allows for the autonomous selection of effective feature dimensions based on the inherent distribution patterns of high-dimensional data. Redundant information and invalid interference components are removed from the spectrum data, resulting in a set of feature vectors that characterize the core attributes of the noise source. This overcomes the limitations of fixed principal component selection patterns, fully preserving low-frequency correlation features in sound pressure level, phase angle, and vibration acceleration. While reducing the data dimensionality, the integrity of the original acoustic information is maintained, providing well-structured basic input data for subsequent model optimization.

[0075] The noise source feature vector set is fed into the soundproof enclosure parameter optimization model to complete multi-objective optimization and output multiple sets of candidate structural parameter combinations for the soundproof enclosure. Based on the matching relationship between the phase angle sequence and the wavenumber of the bending wave of the soundproof enclosure panel, low-frequency coincidence effect avoidance correction is performed, which can correct the acoustic performance deviation caused by the candidate parameters in the low-frequency band. The corrected structural parameter configuration table can conform to the low-frequency noise propagation law of transformers and adapt to the actual acoustic working state of the soundproof enclosure.

[0076] Based on the optimized structural parameter configuration table, a manufacturing process guidance data set for the soundproof enclosure is generated, clarifying the panel stacking sequence and internal stiffener spacing scheme. A correspondence is established between the soundproof enclosure structural design parameters and actual manufacturing processes, bridging the gap between the design and manufacturing stages. The process execution basis for soundproof enclosure component fabrication is standardized, forming a complete design chain from multi-dimensional noise data processing, feature reduction, parameter optimization, low-frequency effect correction to process data output. This standardizes the execution logic of the entire transformer soundproof enclosure design process, adapting to standardized engineering design and batch processing application modes. Attached Figure Description

[0077] Figure 1 This is a state diagram of an optimized design method for a soundproof enclosure for low-frequency noise in a transformer, as described in this invention.

[0078] Figure 2 A flowchart for filtering and processing noisy data matrices in the complex domain;

[0079] Figure 3 This is a flowchart for processing uncorrectable panels. Detailed Implementation

[0080] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0081] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0082] See Figure 1 This invention provides an optimized design method for a soundproof enclosure to reduce low-frequency noise in transformers. The specific method includes:

[0083] A set of low-frequency noise spectrum data generated by the target transformer under rated operating conditions was collected. This set included sound pressure level measurements at multiple octave band center frequencies, phase angle sequences corresponding to each frequency component, and vibration acceleration monitoring data of the transformer tank surface. An improved principal component analysis algorithm was then applied to the obtained low-frequency noise spectrum data set to achieve data dimensionality reduction and compression. This algorithm dynamically determines the number of principal components to be retained based on the cumulative contribution rate of eigenvalues. After dimensionality reduction, a set of noise source feature vectors characterizing the noise source properties of the target transformer was generated. This set of noise source feature vectors was then fed into a pre-trained soundproof enclosure parameter optimization model. This model, through its internal multi-objective optimization solution mechanism, output a series of candidate structural parameter combinations for the soundproof enclosure. For these candidate structural parameter combinations, a low-frequency coincidence effect avoidance correction process was performed. This process adjusted the parameters based on the matching relationship between the previously collected phase angle sequence and the bending wave number of the soundproof enclosure panel to avoid unfavorable coincidence effects, ultimately generating an optimized structural parameter configuration table. Based on this optimized structural parameter configuration table, a set of manufacturing process guidelines for soundproof enclosures was derived and generated to guide actual production. This set of guidelines explicitly includes the stacking sequence of each panel and the arrangement spacing of the internal reinforcing ribs.

[0084] In one embodiment of the present invention, the acquired low-frequency noise spectrum data set, in which the sound pressure level measurements corresponding to each octave band center frequency are arranged in ascending order of frequency, forms an original sound pressure level sequence. The phase angle sequence is also aligned and arranged in the same order as the octave band center frequencies, forming an original phase angle sequence. Complex domain mapping processing is performed on the original sound pressure level sequence and the original phase angle sequence. Specifically, the sound pressure level measurement at each frequency point is combined with its corresponding phase angle to construct data points in complex form. All the complex data points at all frequencies constitute a complex domain noise data matrix. See also... Figure 2The covariance matrix of the complex domain noise data matrix is ​​calculated to obtain a complex covariance matrix. Each element of this matrix consists of the complex correlation coefficient between corresponding frequency points. The process of calculating the complex correlation coefficient includes: extracting each row vector from the complex domain noise data matrix. Each row vector represents a complex observation sequence consisting of complex sound pressure values ​​from multiple time sampling points corresponding to a frequency point. For any two complex observation sequences at different frequency points, the complex correlation coefficient between them is calculated by multiplying the first complex observation sequence and the conjugate sequence of the second complex observation sequence point by point and summing the results, then dividing by the square root of the product of the sum of the magnitudes of the first and second complex observation sequences. This process is repeated for all possible frequency pairs, calculating a complex correlation coefficient value for each pair, and filling these values ​​into a symmetric matrix according to the row and column order of the frequency pair, thus forming the final complex covariance matrix. Before performing eigenvalue decomposition on the complex covariance matrix, the complex domain noise data matrix can be filtered. Calculate the magnitude of each row vector in the matrix. Row vectors with magnitudes less than a preset noise floor threshold are marked as low-energy row vectors, while those with magnitudes greater than or equal to the threshold are marked as high-energy row vectors. Extract all high-energy row vectors to form a high-energy submatrix, and all low-energy row vectors to form a low-energy submatrix. Perform random truncation on the low-energy submatrix, i.e., randomly extract a preset retention ratio of row vectors. Merge the extracted row vectors with the high-energy submatrix row by row to form a filtered complex-domain noise data matrix. This filtered complex-domain noise data matrix replaces the original complex-domain noise data matrix for subsequent covariance matrix calculations. If the total number of row vectors in the filtered complex-domain noise data matrix is ​​less than a preset minimum sample size, cancel the random truncation and directly use the original complex-domain noise data matrix. Perform eigenvalue decomposition on the complex covariance matrix to obtain its eigenvalue sequence and the eigenvector corresponding to each eigenvalue. Extract all eigenvalues ​​from the eigenvalue sequence and sort them in descending order of value. The cumulative contribution rate is recursively calculated for the sorted eigenvalue sequence, starting with the largest eigenvalue and accumulating sequentially. After each accumulation, the proportion of the current accumulated value to the total sum of eigenvalues ​​is calculated as the cumulative contribution rate. The number of accumulation steps corresponding to the first time the cumulative contribution rate reaches a preset contribution rate threshold is determined as the number of principal components to be retained. The eigenvectors corresponding to the eigenvalues ​​in the sorted eigenvalue sequence that are among the first few retained principal components are selected and used to form a projection transformation matrix. The complex domain noise data matrix is ​​multiplied by this projection transformation matrix to obtain the dimensionality-reduced complex feature matrix. From the dimensionality-reduced complex feature matrix, the real part of each complex element is extracted as the principal component amplitude, and the imaginary part is extracted as the principal component phase. All principal component amplitudes and phases are concatenated row-wise to generate the noise source feature vector set.

[0085] In the specific implementation, a low-frequency noise spectrum data set containing four octave center frequencies (63Hz, 125Hz, 250Hz, and 500Hz) generated by the target transformer under rated operating conditions is collected. This low-frequency noise spectrum data set includes the sound pressure level measurement value corresponding to each octave center frequency, the phase angle sequence of each frequency component, and vibration acceleration monitoring data of the transformer tank. In the specific implementation, the sound pressure level measurement values ​​corresponding to each octave center frequency in the low-frequency noise spectrum data set are arranged in ascending order of frequency, forming an original sound pressure level sequence such as [65, 68, 71, 60], in decibels. The phase angle sequence is aligned and arranged in the same order as the octave center frequencies, forming an original phase angle sequence such as [0.12, 0.85, 1.57, 2.41], in radians. Complex domain mapping processing is performed on the original sound pressure level sequence and the original phase angle sequence. In the specific implementation, the sound pressure level measurement value at each frequency point is... With corresponding phase angle Combinatorial construction of data points in complex form ,in Represents the base of the natural exponential function. It is the imaginary unit. Performing this operation on all frequency points yields a complex domain noise data matrix.

[0086] The covariance matrix of the complex domain noise data matrix is ​​calculated to obtain the complex covariance matrix. Each element of the complex covariance matrix consists of the complex correlation coefficient between corresponding frequency points. In some embodiments, all row vectors are extracted from the complex domain noise data matrix. Each row vector represents a complex observation sequence corresponding to a frequency point, and the complex observation sequence consists of complex sound pressure values ​​corresponding to multiple time sampling points. For any two different complex observation sequences corresponding to frequency points... and Calculate the complex correlation coefficient between two complex observation sequences. multiple correlation coefficient The calculation formula is:

[0087]

[0088] in: Represents a sequence The A complex number of elements, Represents a sequence The The conjugate of a complex number of elements express The model, express The modulus, the summation symbol This represents the summation of all elements in the sequence. Iterate through all frequency pairs in the complex-domain noise data matrix, performing the complex correlation coefficient calculation for each frequency pair. Calculate and generate a complex correlation coefficient value. Fill all the complex correlation coefficient values ​​into a symmetric matrix according to the row and column order of the frequency point pairs. The row index and column index of the symmetric matrix correspond to the frequency point numbers in the complex domain noise data matrix, thus forming a complex covariance matrix.

[0089] Before performing eigenvalue decomposition on the complex covariance matrix, a data filtering step is included. In some embodiments, the magnitude of each row vector in the complex domain noise data matrix is ​​calculated. Row vectors with a magnitude less than a preset noise floor threshold are marked as low-energy row vectors, and row vectors with a magnitude greater than or equal to the preset noise floor threshold are marked as high-energy row vectors. All high-energy row vectors are extracted to form a high-energy submatrix, and all low-energy row vectors are extracted to form a low-energy submatrix. Optionally, random truncation is performed on the low-energy submatrix. A preset retention ratio of row vectors is randomly extracted from the low-energy submatrix, and the extracted row vectors are merged with the high-energy submatrix row by row to form a filtered complex domain noise data matrix. The filtered complex domain noise data matrix replaces the original complex domain noise data matrix for subsequent calculation of the covariance matrix. The total number of row vectors in the filtered complex domain noise data matrix is ​​calculated. When the total number of row vectors is less than a preset minimum sample size, the random truncation is canceled, and the original complex domain noise data matrix is ​​used directly.

[0090] In practice, eigenvalue decomposition is performed on the complex covariance matrix to obtain an eigenvalue sequence and an eigenvector corresponding to each eigenvalue. All eigenvalues ​​in the eigenvalue sequence are extracted and sorted in descending order of value. Optionally, a cumulative contribution rate recursive calculation is performed on the sorted eigenvalue sequence. Starting from the largest eigenvalue, the eigenvalues ​​are successively added and divided by the sum of the eigenvalues ​​to obtain the cumulative contribution rate corresponding to each accumulation step. The accumulation step corresponding to the first time the cumulative contribution rate reaches a preset contribution rate threshold is determined as the number of principal components to be retained. The eigenvectors corresponding to the eigenvalues ​​in the sorted eigenvalue sequence that are located at the first retained principal component position are selected to form a projection transformation matrix. The complex domain noise data matrix is ​​multiplied by the projection transformation matrix to obtain the dimensionality-reduced complex feature matrix. The real part of each complex element in the dimensionality-reduced complex feature matrix is ​​extracted as the principal component amplitude, and the imaginary part is extracted as the principal component phase. It can be understood that the principal component amplitude reflects the distribution intensity of noise energy on the principal components, and the principal component phase reflects the spatial or temporal correlation characteristics of the noise components. All principal component amplitudes and principal component phases are concatenated row by row to generate a set of noise source feature vectors.

[0091] In one embodiment of the present invention, a set of noise source feature vectors is input into a pre-trained soundproof enclosure parameter optimization model for multi-objective optimization. The candidate structural parameter combination includes the thickness distribution parameters of each panel of the soundproof enclosure, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material. During the solution process, the principal component amplitudes in the noise source feature vector set are used as the first input group and input into the sound source intensity mapping layer of the soundproof enclosure parameter optimization model. This sound source intensity mapping layer maintains a frequency-amplitude response curve library, which records the equivalent sound source intensity values ​​corresponding to different principal component amplitude intervals. The corresponding equivalent sound source intensity value is read from the library according to the amplitude interval in which the input principal component amplitude falls. Simultaneously, the principal component phases in the noise source feature vector set are used as the second input group and input into the sound source distribution and localization layer of the model. This sound source distribution and localization layer maintains a phase-position mapping grid, which divides the value range of the principal component phase into multiple sector intervals. Each sector interval is associated with a preset sound source spatial coordinate region. The corresponding sound source spatial coordinates are determined according to the sector interval in which the input principal component phase falls. After mapping is completed in the sound source intensity mapping layer and the sound source distribution localization layer, the equivalent sound source intensity value corresponding to each frequency point is paired with the spatial coordinates of the sound source to form a spatially distributed sound source list. Each entry in the list contains a frequency identifier, a sound source intensity value, and a spatial coordinate triplet. This spatially distributed sound source list is input into the model's enclosure parameter iterative search layer. This layer maintains a parameter search tree, where each node represents a combination of parameters for the thickness distribution of the soundproof enclosure, the depth of the internal cavity, and the filling density of the sound-absorbing material. Starting from the root node of the parameter search tree, the insertion loss prediction of the soundproof enclosure parameter combination corresponding to the root node under the excitation of the spatially distributed sound source list is calculated, and this prediction is used as the fitness score of the root node. A preset number of child nodes are selected from high to low fitness scores to enter the next layer of search. The node expansion and fitness score calculation operations are repeated until a preset search depth is reached or the optimal fitness score no longer increases after a preset number of iterations. The soundproof enclosure parameter combination stored at the node corresponding to the current optimal fitness score is output as a candidate structural parameter combination.

[0092] In practice, the noise source feature vector set is input into a pre-trained soundproof enclosure parameter optimization model for multi-objective optimization. The generated candidate structural parameter combination includes the thickness distribution parameters of each panel of the soundproof enclosure, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material. The principal component amplitudes in the noise source feature vector set are used as the first input group and input to the sound source intensity mapping layer of the soundproof enclosure parameter optimization model. In practice, the principal component amplitude can be a sequence such as [5.2, 1.8, 0.6]. The sound source intensity mapping layer maintains a frequency-amplitude response curve library, which records the equivalent sound source intensity values ​​corresponding to different principal component amplitude intervals. For example, the amplitude interval (0, 2.0] corresponds to an intensity value of 70dB, and the amplitude interval (2.0, 4.0] corresponds to an intensity value of 75dB. The corresponding equivalent sound source intensity value is read from the frequency-amplitude response curve library according to the principal component amplitude interval in which the input principal component amplitude is located. Meanwhile, the principal component phase in the noise source feature vector set is used as the second input group to the sound source distribution and localization layer of the soundproof enclosure parameter optimization model. In specific implementation, the principal component phase can be a sequence such as [0.3, 2.1, 5.0]. The sound source distribution and localization layer maintains a phase-position mapping grid, which divides the value range of the principal component phase [0, 2π) into multiple sector intervals. Each sector interval is associated with a preset sound source spatial coordinate region. For example, the phase interval [0, π / 2) is associated with the coordinate region (x>0, y>0). The sound source spatial coordinates corresponding to the principal component phase are determined according to the sector interval into which the input principal component phase falls.

[0093] After the sound source intensity mapping layer and the sound source distribution localization layer have completed their mapping, the equivalent sound source intensity value corresponding to each frequency point is paired with the spatial coordinates of the sound source to form a spatially distributed sound source list. In some embodiments, each entry in the spatially distributed sound source list contains a frequency identifier, a sound source intensity value, and a spatial coordinate triplet, for example, an entry is (125Hz, 75dB, (1.2m, 0.5m, 0.8m)). The spatially distributed sound source list is input into the cover parameter iterative search layer of the soundproof cover parameter optimization model. The cover parameter iterative search layer maintains a parameter search tree, and each node of the parameter search tree represents a combination of the thickness distribution parameters of the soundproof cover, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material. Starting from the root node of the parameter search tree, calculate the estimated insertion loss of the soundproof enclosure parameter combination corresponding to the root node under the excitation of the spatially distributed sound source list. It can be understood that the estimated insertion loss reflects the theoretical sound insulation performance of the parameter combination. The estimated insertion loss obtained in this calculation is used as the fitness score of the root node. Based on the fitness score, a preset number of child nodes are selected from high to low to enter the next level of search.

[0094] Repeat the node expansion and fitness score calculation operations until a preset search depth is reached or the optimal fitness score no longer improves after a preset number of iterations. In some embodiments, the fitness score... The calculation can follow the formula:

[0095]

[0096] in: This represents the total number of entries in the list of spatially distributed sound sources. This indicates that the first item in the list... The estimated sound insulation value calculated from each sound source item. Indicates the first Each sound source entry corresponds to a weighting coefficient, which can be correlated with the sound source intensity value. The summation sign is... This indicates that all entries are summed. Optionally, the preset search depth can be set to 10 layers, and the preset number of iterations can be set to 5. The soundproof enclosure parameter combination stored at the node corresponding to the current optimal fitness score is output as a candidate structure parameter combination. The candidate structure parameter combination can be a set of parameter values ​​such as {Panel 1 thickness: 4mm, Panel 2 thickness: 6mm, Cavity depth: 100mm, Filling density: 32kg / m³}.

[0097] In one embodiment of the present invention, the specific process for calculating the estimated insertion loss of the soundproof enclosure parameter combination corresponding to the root node under the excitation of a spatially distributed sound source list is as follows: The thickness distribution parameters, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material are read from the root node. Based on the thickness distribution parameters, the areal density and bending stiffness of each panel of the soundproof enclosure are determined; based on the depth parameters of the internal cavity, the air layer thickness is determined; and based on the filling density parameters of the sound-absorbing material, the flow resistance and porosity of the sound-absorbing material are determined. When processing the spatially distributed sound source list, the frequency identifier of the current entry and the spatial coordinates of the corresponding sound source are extracted. The spatial coordinates of the sound source corresponding to the current entry are compared with the spatial positions of each preset panel of the soundproof enclosure to determine the panel number of the soundproof enclosure mainly excited by the current sound source. For the frequency identifier of the current entry, the sound transmission loss in the quality control zone, the sound transmission loss in the stiffness control zone, and the sound transmission loss in the coincidence effect zone of the panel excited by the current sound source at that frequency are calculated. The calculated sound transmission loss in the quality control zone, the sound transmission loss in the stiffness control zone, and the sound transmission loss in the coincidence effect zone are segmented and spliced ​​according to frequency to form the single-layer sound transmission loss curve of the currently excited panel. When the panel thickness distribution parameters indicate that the panel is a multi-layer composite structure, the acoustic impedance of each layer is calculated in series to obtain the composite transmission loss curve of the multi-layer composite structure. The air layer thickness of the internal cavity and the flow resistance of the sound-absorbing material are input into the preset cavity sound absorption calculation module to obtain the sound absorption coefficient of the cavity corresponding to the current entry frequency. The composite transmission loss curve of the multi-layer composite structure is superimposed with the sound absorption coefficient obtained in this calculation by octave band to obtain the estimated sound insulation value of the currently excited panel under the current sound source excitation. All entries in the spatially distributed sound source list are traversed, and the estimated sound insulation value corresponding to each entry is calculated. Finally, the energy average of the estimated sound insulation values ​​of all entries is used as the estimated insertion loss value of the root node parameter combination.

[0098] In practical implementation, the estimated insertion loss of the soundproof enclosure parameter combination corresponding to the root node under the excitation of the spatially distributed sound source list is calculated. First, the thickness distribution parameters, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material are read from the root node. For example, the thickness distribution parameters indicate {panel 1: 3mm steel plate, panel 2: 2mm aluminum plate}, the depth parameter of the internal cavity is 150mm, and the filling density parameter of the sound-absorbing material is 40kg / m³. Based on the thickness distribution parameters, the areal density and bending stiffness of each panel of the soundproof enclosure are determined. Based on the depth parameters of the internal cavity, the air layer thickness is determined. Based on the filling density parameters of the sound-absorbing material, the flow resistance and porosity of the sound-absorbing material are determined. In practical implementation, the frequency identifier and the spatial coordinates of the corresponding sound source of the current entry are extracted. The spatial coordinates of the sound source corresponding to the current entry are compared with the spatial positions of each preset panel of the soundproof enclosure to determine the panel number of the soundproof enclosure that is mainly excited by the current sound source. For example, the sound source coordinates (1.2m, 0.5m, 0.8m) are determined to be the main excitation panel number A-03.

[0099] For the current entry's frequency identifier, calculate the transmission loss in the mass control zone, stiffness control zone, and coincidence effect zone of the panel excited by the current sound source at the corresponding frequency. It can be understood that the calculation of these losses depends on the panel's material parameters and frequency. The calculated transmission losses in these zones are then segmented by frequency and spliced ​​together to form the single-layer transmission loss curve of the currently excited panel. When the panel thickness distribution parameter indicates a multi-layer composite structure, the acoustic impedance of each layer is calculated in series to obtain the composite transmission loss curve of the multi-layer composite structure. The calculation follows the formula:

[0100]

[0101] in: This represents the total acoustic impedance of the multilayer composite structure. Indicates the first Acoustic impedance of layered materials, Indicates the number of material layers, summation symbol This indicates that the acoustic impedance of all layers is summed. In some embodiments, the panel structure parameters are shown in Table 1.

[0102] Table 1: Structure and Material Parameters of Sound Insulation Enclosure Panel

[0103] The air layer thickness of the internal cavity and the flow resistance of the sound-absorbing material are input into a preset cavity sound absorption calculation module to obtain the sound absorption coefficient of the cavity corresponding to the current entry frequency identifier. Optionally, the cavity sound absorption calculation module can output a sound absorption coefficient value between 0 and 1 based on the theoretical model of porous sound-absorbing materials, given the flow resistance, porosity, air layer thickness, and frequency. The composite sound transmission loss curve of the multi-layer composite structure is superimposed with the sound absorption coefficient obtained in this calculation at octave bands to obtain the estimated sound insulation value of the currently excited panel under the current sound source excitation. The superposition calculation can be an energy averaging or logarithmic addition operation. The above process is repeated to traverse all entries in the spatially distributed sound source list and calculate the estimated sound insulation value corresponding to each entry. In some embodiments, the spatially distributed sound source list may contain dozens of entries, each corresponding to a specific frequency and sound source location. The energy average of the estimated sound insulation values ​​of all entries is used as the estimated insertion loss value. The energy average value can be obtained by converting each estimated sound insulation value into a sound energy attenuation factor, calculating the arithmetic mean, and then converting it back to a decibel value.

[0104] In one embodiment of the present invention, a low-frequency coincidence effect avoidance correction process is performed on candidate structural parameter combinations to generate an optimized structural parameter configuration table. Thickness distribution parameters of each panel are extracted from the candidate structural parameter combinations. Based on the thickness distribution parameters, the flexural wavenumber frequency characteristic curve of each panel is calculated, and the coincidence effect initiation frequency value of each panel is extracted from the curve. Dominant frequency components whose sound pressure level measurements exceed a preset sound pressure threshold are extracted from the low-frequency noise spectrum data set to form a dominant frequency list. The coincidence effect initiation frequency value of each panel is compared one by one with each frequency value in the dominant frequency list. When the absolute value of the difference between the coincidence effect initiation frequency value of a certain panel and a certain frequency value in the dominant frequency list is less than a preset coincidence frequency tolerance threshold, that panel is marked as a target panel for which the coincidence effect needs correction. See also... Figure 3The thickness offset adjustment process is performed on the target panel by increasing the thickness value in its thickness distribution parameters by a preset thickness adjustment step. The bending wave number frequency response curve after the thickness increase is recalculated to obtain the adjusted coincidence effect initiation frequency value. This process is repeated until the absolute value of the difference between the adjusted coincidence effect initiation frequency value and all dominant frequency values ​​is greater than or equal to the preset coincidence frequency tolerance threshold. During the thickness offset adjustment process, the cumulative increase in thickness of the target panel is recorded, and it is determined whether this total exceeds the preset maximum allowable thickness increment. When the cumulative increase in thickness exceeds the preset maximum allowable thickness increment, the thickness offset adjustment process for the target panel is stopped, and it is marked as an uncorrectable panel. The panel numbers of all marked uncorrectable panels are collected. For each uncorrectable panel, the thickness distribution parameters of its adjacent panels are retrieved from the optimized structural parameter configuration table, and the thickness distribution parameters of the adjacent panels are copied to the uncorrectable panel as replacement thickness values. Subsequently, the thickness distribution parameters of all correctable panels after thickness offset adjustment, the alternative thickness values ​​determined for uncorrectable panels, the original depth parameters of the internal cavities, and the original filling density parameters of the sound-absorbing material are recombinated to obtain the first corrected structural parameter combination. Depth fine-tuning is then performed on the depth parameters of the internal cavities in the first corrected structural parameter combination. The quarter-wavelength resonance depth is calculated based on the lowest frequency value in the dominant frequency list, and the depth parameters of the internal cavities are adjusted to a value close to this quarter-wavelength resonance depth with the smallest absolute deviation from the original depth parameters, resulting in the second corrected structural parameter combination. The parameters in the second corrected structural parameter combination are then organized into a structured parameter table according to panel number and layer order. Each row of this table corresponds to a panel or a layer of cavities, and each column corresponds to a parameter type. This structured parameter table is the optimized structural parameter configuration table.

[0105] In practical implementation, a low-frequency coincidence effect avoidance correction process is applied to the candidate structural parameter combinations. The thickness distribution parameters of each panel are extracted from the candidate structural parameter combinations; for example, a set of parameters might be {panel A-01: 3mm, panel A-02: 5mm, panel B-01: 4mm}. Based on the thickness distribution parameters, the bending wave number frequency response curve for each panel is calculated, and the coincidence effect initiation frequency value for each panel is extracted from the bending wave number frequency response curve. In practical implementation, the coincidence effect initiation frequency value... It can be done through the formula:

[0106]

[0107] in: This indicates the critical frequency of the coincidence effect. This indicates the speed of sound in air. Indicates the areal density of the panel. This indicates the bending stiffness of the panel. Dominant frequency components whose sound pressure level measurements exceed a preset sound pressure threshold are extracted from the low-frequency noise spectrum data set. For example, if the preset sound pressure threshold is 70dB, the extracted dominant frequency list would be [120Hz, 250Hz, 500Hz].

[0108] The initial frequency value of the matching effect for each panel is compared with each frequency value in the dominant frequency list one by one. When the absolute value of the difference between the initial frequency value of the matching effect for a certain panel and a frequency value in the dominant frequency list is less than a preset matching frequency tolerance threshold, for example, the preset matching frequency tolerance threshold is 10Hz, the corresponding panel is marked as the target panel whose matching effect needs to be corrected. In some embodiments, the comparison and marking results are shown in Table 2.

[0109] Table 2: Comparison of Panel Matching Frequencies and Dominant Frequencies

[0110]

[0111] Thickness offset adjustment is performed on the target panel, increasing the thickness value in the thickness distribution parameters by a preset thickness adjustment step, for example, 0.5mm. The bending wave number frequency response curve after the increased thickness is recalculated to obtain the adjusted coincidence effect initiation frequency value. The thickness offset adjustment process is repeated until the absolute value of the difference between the adjusted coincidence effect initiation frequency value and all dominant frequency values ​​is greater than or equal to the preset coincidence frequency tolerance threshold. In specific implementation, the thickness of panel A-01 is adjusted from 3.0mm to 3.5mm and then to 4.0mm. The corresponding coincidence effect initiation frequency changes from 245Hz to approximately 280Hz and 320Hz. Finally, at 4.0mm, the difference between 320Hz and the dominant frequency 250Hz (70Hz) is greater than 10Hz, and the adjustment stops.

[0112] During the thickness offset adjustment process, the cumulative increase in thickness of the target panel is recorded, and it is determined whether the cumulative increase in thickness exceeds the preset maximum allowable thickness increment. For example, the preset maximum allowable thickness increment is set to 3.0 mm. When the cumulative increase in thickness exceeds the preset maximum allowable thickness increment, the thickness offset adjustment process for the target panel is stopped, and the target panel is marked as an uncorrectable panel. All panel numbers marked as uncorrectable panels are collected. For each uncorrectable panel, the thickness distribution parameters of adjacent panels are retrieved from the optimized structural parameter configuration table, and the thickness distribution parameters of the adjacent panels are copied to the uncorrectable panel as replacement thickness values. This step aims to avoid over-adjustment of local parameters by borrowing adjacent structural parameters. The thickness distribution parameters of all panels after thickness offset adjustment, the original depth parameters of the internal cavities, and the original filling density parameters of the sound-absorbing material are recombine to obtain the structural parameter combination after the first correction.

[0113] Optionally, a depth fine-tuning process is performed on the depth parameter of the internal cavity in the first modified structural parameter combination. The quarter-wavelength resonance depth is calculated based on the lowest frequency value in the dominant frequency list. For example, if the lowest frequency in the dominant frequency list is 120Hz and the speed of sound in air is 340m / s, its quarter-wavelength is approximately 0.71m. The depth parameter of the internal cavity is adjusted to a value close to the quarter-wavelength resonance depth with the smallest absolute deviation from the original depth parameter. For example, if the original depth parameter is 150mm, the adjusted depth parameter can be 175mm or 200mm, resulting in the second modified structural parameter combination. In some embodiments, the parameters in the second modified structural parameter combination are organized into a structured parameter table according to panel number and hierarchical order. Each row of the structured parameter table corresponds to a panel or a cavity layer, and each column corresponds to a parameter type. The structured parameter table serves as the optimized structural parameter configuration table.

[0114] In one embodiment of the present invention, a soundproof enclosure manufacturing process guidance data set is generated based on an optimized structural parameter configuration table. The thickness distribution parameters of each panel are read from the optimized structural parameter configuration table, for example, panel thickness distribution parameters are {panel A-01: 4mm steel plate, panel A-02: 2mm damping layer, panel B-01: 6mm steel plate}. The required material grade and specifications for each panel are determined based on the thickness distribution parameters, for example, 4mm steel plate corresponds to grade Q235B. Panels with the same grade and specifications are grouped into the same material cutting batch, forming a material cutting batch list. The depth parameters of the internal cavities are read from the optimized structural parameter configuration table, for example, the depth parameter of the internal cavity is 150mm. The required height dimension of the support frame for each cavity location is calculated based on the internal cavity depth parameter, for example, the height dimension is 145mm. Support frames whose height dimensions differ from a preset size tolerance threshold by less than 5mm are grouped into the same welding fixture group, forming a frame welding process grouping table. Read the sound-absorbing material filling density parameter from the optimized structural parameter configuration table, for example, 40 kg / m³. Calculate the required weight and thickness of sound-absorbing material for each cavity area based on the filling density parameter. For example, if an area has a volume of 0.5 m³, the sound-absorbing material weight is 20 kg and the laying thickness is 100 mm. Group areas with similar sound-absorbing material weights and laying thicknesses into the same filling batch to form a sound-absorbing material filling operation instruction sheet.

[0115] For each batch in the sheet metal cutting batch list, extract the total weight and total cutting length of the sheet metal in the current batch. For example, a batch contains 10 4mm steel plates, with a total weight of 235kg and a total cutting length of 60 meters. Calculate the minimum cutting power required for the cutting equipment based on the total weight of the sheet metal, estimate the cutting time based on the total cutting length, and append the minimum cutting power and cutting time to the cutting process parameter column of the current batch to generate a sheet metal cutting batch list with cutting parameters. In some embodiments, the cutting time... You can use the formula:

[0116]

[0117] in: Indicates the total cutting length. Indicates the average cutting speed. Indicates the number of times a special process needs to be performed. This indicates the average time consumed for each special process. For each welding fixture group in the frame welding process grouping table, the number of supporting frames and the total weld length within the current welding fixture group are extracted. For example, if a fixture group contains 8 supporting frames and a total weld length of 16 meters, the welding sequence number is determined based on the number of supporting frames in the current welding fixture group, and the welding time is estimated based on the total weld length. The welding sequence number and welding time are then appended to the welding process parameter column of the current welding fixture group to generate a frame welding process grouping table with welding parameters. The batch list of plate cutting with cutting parameters, the frame welding process grouping table with welding parameters, and the sound-absorbing material filling work instruction sheet are arranged in chronological order of the manufacturing process. The three documents together constitute the soundproof enclosure manufacturing process guidance data set.

[0118] After the soundproof enclosure is manufactured according to the soundproof enclosure manufacturing process guidance data set and installed on the target transformer, actual low-frequency noise spectrum data at preset measurement points outside the enclosure are collected within a preset verification time window. It can be understood that the preset measurement points can be arranged at multiple locations 1 meter away from the surface of the soundproof enclosure and 1.5 meters high. The actual low-frequency noise spectrum data is compared with the sound pressure level measurement values ​​at the corresponding frequency points in the low-frequency noise spectrum data set, and the difference is calculated point-by-point to obtain the actual noise reduction sequence for each frequency point. The actual noise reduction sequence is compared with the theoretical noise reduction sequence calculated according to the optimized structural parameter configuration table, and the noise reduction deviation value for each frequency point is calculated. Frequency points whose noise reduction deviation value exceeds a preset deviation threshold are marked as deviation frequency points. In some embodiments, the preset deviation threshold can be set to 3dB. The percentage of deviation frequency points out of the total number of frequency points is counted. When the percentage is greater than a preset re-optimization trigger threshold, such as 15%, the noise reduction deviation value and deviation sign for each deviation frequency point are extracted from the frequency value corresponding to the deviation frequency point. Optionally, the deviation sign "+" indicates that the actual noise reduction is lower than the theoretical value, and "-" indicates that the actual noise reduction is higher than the theoretical value.

[0119] Based on the noise reduction deviation value and sign at each deviation frequency point, the panel thickness distribution parameters corresponding to the current deviation frequency point in the optimized structural parameter configuration table are adjusted in reverse. The direction of the reverse adjustment is opposite to the deviation sign, and the magnitude of the reverse adjustment is proportional to the noise reduction deviation value. For example, for a frequency of 250Hz, if the deviation value is +4dB and the deviation sign is "+", it means that the actual sound insulation effect is 4dB worse than the theoretical prediction. Therefore, the panel thickness parameters related to this frequency are increased, with the adjustment magnitude proportional to 4dB, resulting in a re-optimized structural parameter configuration table. This re-optimized structural parameter configuration table is then used as the new optimized structural parameter configuration table, and the step of generating the soundproof enclosure manufacturing process guidance data set is repeated to produce a revised version of the manufacturing process guidance data set.

[0120] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for optimizing the design of a soundproof enclosure for low-frequency noise in transformers, characterized in that, include: The low-frequency noise spectrum data set of the target transformer under rated operating conditions is collected. The low-frequency noise spectrum data set includes sound pressure level measurements corresponding to multiple octave band center frequencies, phase angle sequences of each frequency component, and vibration acceleration monitoring data of the transformer tank. An improved principal component analysis algorithm is applied to the low-frequency noise spectrum data set for dimensionality reduction and compression to generate a set of noise source feature vectors for the target transformer. The improved principal component analysis algorithm dynamically adjusts the number of retained principal components based on the cumulative contribution rate of the feature values. The noise source feature vector set is input into a pre-trained soundproof enclosure parameter optimization model for multi-objective optimization, generating candidate structural parameter combinations for the soundproof enclosure, including: The candidate structural parameter combination includes the thickness distribution parameters of each panel of the soundproof enclosure, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material. The principal component amplitudes in the noise source feature vector set are used as the first input group and input to the sound source intensity mapping layer of the soundproof enclosure parameter optimization model. The sound source intensity mapping layer maintains a frequency-amplitude response curve library. The frequency-amplitude response curve library records the equivalent sound source intensity values ​​corresponding to different principal component amplitude intervals. The corresponding equivalent sound source intensity value is read from the frequency-amplitude response curve library according to the principal component amplitude interval in which the principal component amplitude is located. The principal component phase in the noise source feature vector set is used as the second input group and input to the sound source distribution and localization layer of the soundproof enclosure parameter optimization model. The sound source distribution and localization layer maintains a phase-position mapping grid. The phase-position mapping grid divides the value range of the principal component phase into multiple sector intervals. Each sector interval is associated with a preset sound source spatial coordinate region. The sound source spatial coordinates corresponding to the principal component phase are determined according to the sector interval into which the principal component phase falls. After the sound source intensity mapping layer and the sound source distribution and positioning layer have completed the mapping respectively, the equivalent sound source intensity value corresponding to each frequency point is paired with the sound source spatial coordinates to form a spatial distribution sound source list. Each entry in the spatial distribution sound source list contains a frequency identifier, a sound source intensity value and a spatial coordinate triplet. The list of spatially distributed sound sources is input into the cover parameter iterative search layer of the soundproof cover parameter optimization model. The cover parameter iterative search layer maintains a parameter search tree. Each node of the parameter search tree represents a combination of the thickness distribution parameters of the soundproof cover, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material. Starting from the root node of the parameter search tree, calculate the estimated insertion loss of the soundproof enclosure parameter combination corresponding to the root node under the excitation of the spatially distributed sound source list. Use the estimated insertion loss obtained in this calculation as the fitness score of the root node. Select a preset number of child nodes from high to low according to the fitness score to enter the next layer of search. Repeat the node expansion and fitness score calculation operations until the preset search depth is reached or the optimal fitness score no longer increases after a preset number of iterations. Then, output the combination of soundproof enclosure parameters stored on the node corresponding to the current optimal fitness score as the candidate structure parameter combination. The candidate structural parameter combinations are subjected to low-frequency coincidence effect avoidance correction processing to generate an optimized structural parameter configuration table. Based on the optimized structural parameter configuration table, a soundproof enclosure manufacturing process guidance data set is generated. The soundproof enclosure manufacturing process guidance data set includes the stacking sequence scheme of each panel and the arrangement spacing scheme of the internal reinforcing ribs.

2. The method for optimizing the design of a soundproof enclosure for low-frequency noise in transformers according to claim 1, characterized in that, An improved principal component analysis algorithm is applied to the low-frequency noise spectrum data set for dimensionality reduction and compression to generate a set of noise source feature vectors for the target transformer, including: The sound pressure level measurements corresponding to each octave center frequency in the low-frequency noise spectrum data set are arranged in ascending order of frequency to form the original sound pressure level sequence. The phase angle sequence is then aligned and arranged in the same order as the octave center frequency to form the original phase angle sequence. Perform complex domain mapping processing on the original sound pressure level sequence and the original phase angle sequence, and combine the sound pressure level measurement value of each frequency point with the corresponding phase angle to construct a data point in complex form, thereby obtaining a complex domain noise data matrix; Calculate the covariance matrix of the complex domain noise data matrix to obtain the complex covariance matrix. Each element of the complex covariance matrix is ​​composed of the complex correlation coefficient between the corresponding frequency points. The complex covariance matrix is ​​subjected to eigenvalue decomposition to obtain an eigenvalue sequence and an eigenvector corresponding to each eigenvalue. All eigenvalues ​​in the eigenvalue sequence are extracted and sorted in descending order of value. The cumulative contribution rate is recursively calculated for the sorted feature value sequence. Starting from the largest feature value, the feature values ​​are added sequentially and divided by the sum of the feature values ​​to obtain the cumulative contribution rate corresponding to each accumulation step. The accumulation step corresponding to the first time the cumulative contribution rate reaches the preset contribution rate threshold is determined as the number of principal components to be retained. The eigenvectors corresponding to the eigenvalues ​​located at the first retained principal component position in the sorted eigenvalue sequence are selected to form a projection transformation matrix. The complex domain noise data matrix and the projection transformation matrix are multiplied to obtain the dimension-reduced complex feature matrix. The real part of each complex element is extracted from the dimensionality-reduced complex feature matrix as the principal component amplitude and the imaginary part as the principal component phase. All principal component amplitudes and principal component phases are concatenated row by row to generate the noise source feature vector set.

3. The method for optimizing the design of a soundproof enclosure for low-frequency noise in a transformer according to claim 2, characterized in that, Calculate the covariance matrix of the complex domain noise data matrix to obtain a complex covariance matrix. Each element of the complex covariance matrix consists of the complex correlation coefficient between corresponding frequency points, including: Extract all row vectors from the complex domain noise data matrix. Each row vector represents a complex observation sequence corresponding to a frequency point. The complex observation sequence consists of complex sound pressure values ​​corresponding to multiple time sampling points. For any two different frequency points corresponding to complex observation sequences, calculate the complex correlation coefficient between the two complex observation sequences. The formula for calculating the complex correlation coefficient is: multiply the conjugate sequences of the first and second complex observation sequences point by point and sum them up, then divide by the square root of the product of the sum of the magnitudes of the first and second complex observation sequences. Traverse all frequency pairs in the complex domain noise data matrix, perform complex correlation coefficient calculation for each frequency pair, and generate a complex correlation coefficient value. All complex correlation coefficient values ​​are filled into a symmetric matrix according to the row and column order of frequency point pairs. The row index and column index of the symmetric matrix correspond to the frequency point numbers in the complex domain noise data matrix, thus forming a complex covariance matrix.

4. The method for optimizing the design of a soundproof enclosure for low-frequency noise in a transformer according to claim 2, characterized in that, Before performing eigenvalue decomposition on the complex covariance matrix, the process further includes: Calculate the magnitude of each row vector in the complex domain noise data matrix, mark the row vectors with a magnitude less than a preset noise floor threshold as low-energy row vectors, and mark the row vectors with a magnitude greater than or equal to the preset noise floor threshold as high-energy row vectors. Extract all high-energy row vectors to form a high-energy submatrix, and extract all low-energy row vectors to form a low-energy submatrix. Random truncation is performed on the low-energy submatrix. Row vectors with a preset retention ratio are randomly extracted from the low-energy submatrix. The extracted row vectors are then merged with the high-energy submatrix row by row to form a filtered complex domain noise data matrix. The filtered complex domain noise data matrix is ​​used to replace the complex domain noise data matrix for subsequent calculation of the covariance matrix. Calculate the total number of row vectors in the filtered complex domain noise data matrix. If the total number of row vectors is less than the preset minimum number of samples, cancel the random truncation process and directly use the original complex domain noise data matrix.

5. The method for optimizing the design of a soundproof enclosure for low-frequency noise in a transformer according to claim 4, characterized in that, The calculated insertion loss estimate for the combination of soundproof enclosure parameters corresponding to the root node under the excitation of the spatially distributed sound source list includes: Read the thickness distribution parameters, the depth parameters of the internal cavity, and the filling density parameters of the sound-absorbing material from the root node. Determine the surface density and bending stiffness of each panel of the soundproof enclosure based on the thickness distribution parameters. Determine the air layer thickness based on the depth parameters of the internal cavity. Determine the flow resistance and porosity of the sound-absorbing material based on the filling density parameters of the sound-absorbing material. Extract the frequency identifier and spatial coordinates of the corresponding sound source of the current entry, compare the spatial coordinates of the sound source corresponding to the current entry with the spatial positions of each preset panel of the soundproof enclosure, and determine the panel number of the soundproof enclosure that is mainly excited by the current sound source. For the frequency identifier of the current entry, calculate the sound transmission loss in the quality control area, stiffness control area, and coincidence effect area of ​​the panel excited by the current sound source at the frequency corresponding to the frequency identifier. Then, splice the calculated sound transmission loss in the quality control area, stiffness control area, and coincidence effect area according to frequency segments to form the single-layer sound transmission loss curve of the current excited panel. When the panel thickness distribution parameter indicates a multi-layer composite structure, the acoustic impedance of each layer of material is calculated by series superposition to obtain the composite sound transmission loss curve of the multi-layer composite structure. The air layer thickness of the internal cavity and the flow resistance of the sound-absorbing material are input into the preset cavity sound absorption calculation module to obtain the sound absorption coefficient of the cavity corresponding to the current entry frequency identifier; the composite transmission loss curve of the multi-layer composite structure is superimposed with the sound absorption coefficient obtained in this calculation by octave band to obtain the estimated sound insulation value of the current excited panel under the current sound source excitation. The energy average of the sound insulation estimates for all entries is used as the estimated insertion loss.

6. The method for optimizing the design of a soundproof enclosure for low-frequency noise in a transformer according to claim 5, characterized in that, The candidate structural parameter combinations are subjected to low-frequency coincidence effect avoidance correction processing to generate an optimized structural parameter configuration table, including: The thickness distribution parameters of each panel are extracted from the candidate structural parameter combination. The bending wave number frequency characteristic curve of each panel is calculated based on the thickness distribution parameters. The coincidence effect initiation frequency value of each panel is extracted from the bending wave number frequency characteristic curve. The dominant frequency components whose sound pressure level measurements exceed a preset sound pressure threshold are extracted from the low-frequency noise spectrum data set and a dominant frequency list is formed. The matching effect starting frequency value of each panel is compared with each frequency value in the dominant frequency list one by one. When the absolute value of the difference between the starting frequency value of the matching effect of a certain panel and a certain frequency value in the dominant frequency list is less than the preset matching frequency tolerance threshold, the corresponding panel is marked as the target panel whose matching effect needs to be corrected. The target panel is subjected to thickness offset adjustment processing. The thickness value in the thickness distribution parameter of the target panel is increased by a preset thickness adjustment step size. The bending wave number frequency characteristic curve after the thickness is increased is recalculated to obtain the adjusted coincidence effect starting frequency value. The thickness offset adjustment processing is repeated until the absolute value of the difference between the adjusted coincidence effect starting frequency value and all dominant frequency values ​​is greater than or equal to the preset coincidence frequency tolerance threshold. The thickness distribution parameters of all panels after thickness offset adjustment, the original depth parameters of the internal cavity, and the original filling density parameters of the sound-absorbing material are recombined to obtain the first corrected combination of structural parameters. The depth parameter of the internal cavity in the first corrected combination of structural parameters is fine-tuned. The quarter-wavelength resonance depth is calculated based on the lowest frequency value in the list of dominant frequencies. The depth parameter of the internal cavity is adjusted to a value that is close to the quarter-wavelength resonance depth and has the smallest absolute value of deviation from the original depth parameter, thus obtaining the second corrected combination of structural parameters. The parameters in the second revised structural parameter combination are organized into a structured parameter table according to the panel number and hierarchical order. Each row of the structured parameter table corresponds to a panel or a cavity layer, and each column corresponds to a parameter type. The structured parameter table serves as the optimized structural parameter configuration table.

7. The method for optimizing the design of a soundproof enclosure for low-frequency noise in a transformer according to claim 6, characterized in that, Before recombining the thickness distribution parameters of all panels after thickness offset adjustment, the original depth parameters of the internal cavity, and the original filling density parameters of the sound-absorbing material, the process further includes: After performing thickness offset adjustment on the target panel, record the total cumulative increase in the thickness of the target panel, and determine whether the total cumulative increase in thickness exceeds the preset maximum allowable thickness increment. When the cumulative increase in thickness exceeds the preset maximum allowable thickness increment, stop the thickness offset adjustment process for the target panel and mark the target panel as an uncorrectable panel; Collect all panel numbers marked as uncorrectable panels. For each uncorrectable panel, query the thickness distribution parameters of the adjacent panels from the optimized structural parameter configuration table, and copy the thickness distribution parameters of the adjacent panels to the uncorrectable panel as alternative thickness values. The replacement thickness value is recombined with the original internal cavity depth parameter and the original sound-absorbing material filling density parameter to form the corrected structural parameters for the location of the uncorrectable panel.

8. The method for optimizing the design of a soundproof enclosure for low-frequency noise in a transformer according to claim 6, characterized in that, Based on the optimized structural parameter configuration table, a set of manufacturing process guidance data for the soundproof enclosure is generated, including: Read the thickness distribution parameters of each panel from the optimized structural parameter configuration table, determine the grade and specifications of the board material required for each panel based on the thickness distribution parameters, and group panels with the same grade and specifications into the same board cutting batch to form a board cutting batch list. Read the depth parameters of the internal cavity from the optimized structural parameter configuration table, calculate the required height of the support frame for each cavity position based on the depth parameters of the internal cavity, and group the support frames whose height dimensions differ from the preset size tolerance threshold into the same welding fixture group to form a frame welding process grouping table. Read the filling density parameters of the sound-absorbing material from the optimized structural parameter configuration table, calculate the required weight and laying thickness of the sound-absorbing material for each cavity area based on the filling density parameters, and group areas with similar sound-absorbing material weight and laying thickness into the same filling operation batch to form a sound-absorbing material filling operation instruction sheet. For each batch in the sheet material cutting batch list, extract the total weight and total cutting length of the sheet material in the current batch, calculate the minimum cutting power of the required cutting equipment based on the total weight of the sheet material, estimate the cutting time based on the total cutting length, and append the minimum cutting power and cutting time to the cutting process parameter column of the current batch to generate a sheet material cutting batch list with cutting parameters. For each welding fixture group in the frame welding process grouping table, extract the number of supporting frames and the total weld length in the current welding fixture group, determine the welding sequence number based on the number of supporting frames in the current welding fixture group, estimate the welding time based on the total weld length in the current welding fixture group, and append the welding sequence number and welding time to the welding process parameter column of the current welding fixture group to generate a frame welding process grouping table with welding parameters. The batch list of sheet metal cutting parameters, the grouping table of frame welding processes with welding parameters, and the work instruction sheet for filling sound-absorbing materials are arranged in chronological order of the manufacturing process. The three documents together constitute the manufacturing process guidance data set for the soundproof enclosure.

9. The method for optimizing the design of a soundproof enclosure for low-frequency noise in a transformer according to claim 1, characterized in that, The method further includes: After the soundproof enclosure is manufactured according to the soundproof enclosure manufacturing process guidance data set and installed on the target transformer, the actual low-frequency noise spectrum data at the preset measurement points outside the enclosure is collected within the preset verification time window. The actual low-frequency noise spectrum data is compared with the sound pressure level measurement value at the corresponding frequency point in the low-frequency noise spectrum data set by calculating the difference at each frequency point to obtain the actual noise reduction sequence at each frequency point. The actual noise reduction sequence is compared with the theoretical noise reduction sequence calculated according to the optimized structural parameter configuration table, and the noise reduction deviation value at each frequency point is calculated. Frequency points whose noise reduction deviation value exceeds the preset deviation threshold are marked as deviation frequency points. The percentage of the number of deviation frequency points out of the total number of frequency points is counted. When the percentage is greater than the preset re-optimization trigger threshold, the noise reduction deviation value and deviation sign of each deviation frequency point are extracted from the frequency value corresponding to the deviation frequency point. Based on the noise reduction deviation value and deviation sign at each deviation frequency point, the panel thickness distribution parameter corresponding to the current deviation frequency point in the optimized structural parameter configuration table is reverse-adjusted. The direction of the reverse adjustment is opposite to the deviation sign, and the magnitude of the reverse adjustment is proportional to the noise reduction deviation value, thus obtaining the re-optimized structural parameter configuration table. Using the re-optimized structural parameter configuration table as the new optimized structural parameter configuration table, the step of generating the soundproof enclosure manufacturing process guidance data set is executed again to produce a revised version of the manufacturing process guidance data set.

Images