A method for analyzing and judging the noise compliance of transformers

Abstract

This invention discloses a method for analyzing and judging the compliance of transformer noise, comprising the following steps: energizing the transformer to be tested and inputting test power to form a noise test environment; collecting noise environment data from different locations and sending it to a cloud center; the cloud center preprocessing the noise environment data and, based on mutually independent pre-trained convolutional neural network models, identifying whether the noise environment data collected from different locations meets product performance requirements; and fusing the identification results of the noise environment data from different locations to obtain the final predicted fusion result. Implementing this invention enables automatic testing and analysis of transformer noise, quickly determining its compliance, and improving work efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of transformer testing, and in particular to a method for analyzing and judging the compliance of transformer noise. Background Technology

[0002] As one of the main pieces of equipment in urban power supply systems, reducing urban noise is a primary task for transformer manufacturers and a new hot spot for competition in the transformer industry. High noise levels are a significant factor restricting the widespread application of dry-type distribution transformers. This not only affects the quality of life for residents near transformers but also increases losses during transformer operation. Therefore, strictly controlling transformer noise requires rigorous testing of manufactured products to ensure they meet market demands, making this crucial.

[0003] However, in the existing technology, transformer noise testing and analysis are mostly done manually, which is inefficient and inaccurate. Summary of the Invention

[0004] The technical problem solved by this invention is to propose a method for analyzing and judging the compliance of transformer noise, which can realize automatic testing and analysis of transformer noise, thereby improving work efficiency and accuracy.

[0005] As one aspect of the present invention, embodiments of the present invention provide a method for analyzing and judging the compliance of transformer noise, which includes at least the following steps:

[0006] Step S1: Power on the transformer to be tested, input the test power within a preset time, and create a noise test environment according to the test power;

[0007] Step S2: Collect noise environment data at different locations using a first data acquisition module and a second data acquisition device located at different positions on the transformer. The noise environment data includes: temperature value, transformer parameters, acquisition module parameters, and noise value.

[0008] Step S3: The noise environment data collected by the first data acquisition module and the second data acquisition module are wirelessly transmitted to the cloud center through the control device.

[0009] Step S4: The cloud center preprocesses the noise environment data collected by the first data acquisition module and the second data acquisition module to generate a parameter format that conforms to the input of the neural network model.

[0010] Step S5: Based on mutually independent pre-trained convolutional neural network models, identify whether the noise environment data collected by the first data acquisition module and the second data acquisition module meet the product performance requirements.

[0011] Step S6: The identification results of the noise environment data collected by the first data acquisition module and the identification results of the noise environment data collected by the second data acquisition module are fused together to obtain the final prediction fusion result, and to determine whether the transformer meets the noise requirements.

[0012] Preferably, step S3 further includes:

[0013] Multiple cloud servers in the cloud center are pre-deployed with the same network operating system and Hadoop distributed file processing system; the distributed cloud storage server in the cloud center stores the parameters of various transformers and compares them with the reported transformer parameters to determine the type of the current transformer.

[0014] Preferably, step S4 further includes:

[0015] Based on the transformer type determined by the transformer parameters, a finite element model corresponding to the transformer type is obtained. Then, based on the finite element model of the transformer and the parameters of the acquisition module, a pre-trained neural network model is obtained.

[0016] The noise values ​​of the noise environment data collected according to the time series are processed to extract characteristic frequency points.

[0017] Preferably, step S5 further includes:

[0018] Step S51: Input the feature frequency points and temperature values ​​of the noise environment data collected in each data acquisition module into a pre-trained neural network model for fitting and feature vector extraction; the neural network model is a hybrid multi-scale convolutional kernel residual network model.

[0019] Step S52: Perform batch normalization on the fitted feature vector; then perform 1-pixel x 1-pixel convolution with global average pooling on the batch normalized feature vector to obtain feature values ​​in the specified output dimension.

[0020] Step S53: The feature values ​​are processed through the Dropout layer of the neural network model to suppress overfitting;

[0021] Step S54: Input the processed feature value data stream into the softmax layer for calculation to determine whether the transformer performance meets the requirements.

[0022] Preferably, step S51 further includes:

[0023] The feature frequency point parameters and temperature values ​​are input into the first-level convolutional network of the pre-trained neural network model, and the feature frequency point parameters and temperature values ​​are subjected to first-level fitting processing based on the first-level convolutional network. The first-level convolutional network includes three layers of convolutional networks, each of which is equipped with 6 2x2 convolutional kernels, 12 1x1 convolutional kernels, and 6 5x5 convolutional kernels.

[0024] The feature frequency point parameters and temperature values ​​after the first-level fitting process are input into the second-level convolutional network of the pre-trained neural network model, and the feature frequency point parameters and temperature values ​​are subjected to a second-level fitting process based on the second-level convolutional network. The second-level convolutional network includes three layers of convolutional networks, each layer of which is equipped with 16 1x1 convolutional kernels, 32 3x3 pixel convolutional kernels, and 16 5x5 convolutional kernels.

[0025] The feature frequency point parameters and temperature values ​​after the second-level fitting process are input into the third-level convolutional network of the pre-trained neural network model, and the parameters of the transformer are subjected to the third-level fitting process based on the third-level convolutional network. The third-level convolutional network includes three layers of convolutional networks, each layer of which is equipped with 8 1x1 convolutional kernels, 16 3x3 convolutional kernels, and 16 5x5 convolutional kernels.

[0026] Preferably, step S52 further includes:

[0027] Calculate the mean of the set of historical characteristic frequency point parameters and temperature values ​​after fitting;

[0028] Calculate the variance of the characteristic frequency point parameters and temperature values ​​after fitting;

[0029] The fitted feature frequency point parameters and temperature values ​​were normalized using the calculated mean and variance.

[0030] The fitted characteristic frequency point parameters and temperature values ​​are subjected to scaling transformation and offset processing.

[0031] Preferably, step S6 further includes:

[0032] The results obtained from the data collected in the first data acquisition module are fused with the results obtained from the data collected in the second data acquisition module. If the outputs of both neural network models are that the product is compliant, the product is judged to be qualified. If the outputs of both neural network models are that the product is non-compliant, the product is judged to be defective. If the outputs of the two neural network models are inconsistent, the product is judged to be non-compliant.

[0033] Implementing the embodiments of the present invention has the following beneficial effects:

[0034] This invention proposes a method for analyzing and judging the compliance of transformer noise. By collecting noise from different locations during transformer operation, feature parameters are generated from the noise features, the feature parameters are standardized in a unified data format, and then a neural network is used for identification and prediction. Ultimately, an objective and accurate prediction result can be achieved, enabling the prediction result to detect the compliance of the transformer.

[0035] In this invention, the data input of the convolutional network is realized through the first data acquisition module and the second data acquisition module. The models of the neural networks are independently used to perform deep learning on the corresponding product compliance results. By fusing the data, the bias caused by the randomness of the deep learning model is reduced, thereby improving the accuracy of product compliance.

[0036] In summary, the embodiments of the present invention can realize automatic testing and analysis of transformer noise, improving work efficiency and accuracy. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of the main flow of an embodiment of a method for analyzing and judging the compliance of transformer noise provided by the present invention;

[0039] Figure 2 This is a schematic diagram illustrating the application environment of the present invention;

[0040] Figure 3 This is a schematic diagram of the first convolutional network structure involved in this invention;

[0041] Figure 4 This is a schematic diagram of the second convolutional network structure involved in this invention;

[0042] Figure 5 This is a schematic diagram of the third convolutional network structure involved in this invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0044] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0045] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0046] like Figure 1 The diagram shown illustrates the main flow of an embodiment of a method for analyzing and judging transformer noise compliance provided by the present invention; in conjunction with... Figures 2 to 5 As shown, in this embodiment, the method includes at least the following steps:

[0047] Step S1: Power on the transformer to be tested, input the test power within a preset time, and create a noise test environment according to the test power; the transformer to be tested can be a dry-type transformer product to be shipped.

[0048] Step S2: Collect noise environment data at different locations on the transformer using a first data acquisition module and a second data acquisition device. The noise environment data includes temperature value, transformer parameters, acquisition module parameters, and noise value. It is understood that the environmental noise value will be different depending on the location of the first data acquisition module and the second data acquisition module on the transformer.

[0049] Step S3: The noise environment data collected by the first data acquisition module and the second data acquisition module are wirelessly transmitted to the cloud center through the control device; wherein, the control device and the cloud center can communicate wirelessly based on communication protocols such as 3G, GPRS, 4G, and 5G; the control device can be a separate communication module or integrated into a computer such as a mobile terminal, platform, or industrial control computer.

[0050] More specifically, in one example, step S3 further includes:

[0051] Multiple cloud servers in the cloud center are pre-deployed with the same network operating system and Hadoop distributed file processing system; the distributed cloud storage server in the cloud center stores the parameters of various transformers and compares them with the reported transformer parameters to determine the type of the current transformer.

[0052] Understandably, the cloud center resides at the network layer of the entire system and possesses powerful data processing capabilities. Utilizing cloud computing technology, the cloud center integrates the three core elements of a data center: computing, networking, and storage. Cloud servers, based on cluster server technology, virtualize multiple components resembling independent servers, offering high security and stability.

[0053] Step S4: The cloud center preprocesses the noise environment data collected by the first data acquisition module and the second data acquisition module to generate a parameter format that conforms to the input of the neural network model.

[0054] More specifically, in one example, step S4 further includes:

[0055] Based on the transformer type determined by the transformer parameters, a finite element model corresponding to the transformer type is obtained. Then, based on the finite element model of the transformer and the parameters of the acquisition module, a pre-trained neural network model is obtained.

[0056] Since the noise values ​​in the noise environment data are determined with reference to the operating environment at the test time, in the embodiments of the present invention, it is necessary to extract the characteristic frequency points of the noise values ​​of the noise environment data collected according to the time series.

[0057] Step S5: Based on mutually independent pre-trained convolutional neural network models, identify whether the noise environment data collected by the first data acquisition module and the second data acquisition module meet the product performance requirements.

[0058] It is understood that, in the embodiments of the present invention, it is necessary to collect noise from each acquisition location of different types of transformers in advance to form a training set and train it, so as to obtain a pre-trained convolutional neural network model corresponding to each location in advance.

[0059] Specifically, in one example, step S5 further includes:

[0060] Step S51: Input the feature frequency points and temperature values ​​of the noise environment data collected in each data acquisition module into a pre-trained neural network model for fitting and feature vector extraction; the neural network model is a hybrid multi-scale convolutional kernel residual network model.

[0061] Step S51 employs a three-level convolutional network, which further includes:

[0062] The feature frequency point parameters and temperature values ​​are input into the first-level convolutional network of the pre-trained neural network model, and the feature frequency point parameters and temperature values ​​are subjected to first-level fitting processing based on the first-level convolutional network. The first-level convolutional network includes three layers of convolutional networks, each of which is equipped with 6 2x2 convolutional kernels, 12 1x1 convolutional kernels, and 6 5x5 convolutional kernels.

[0063] The feature frequency point parameters and temperature values ​​after the first-level fitting process are input into the second-level convolutional network of the pre-trained neural network model, and the feature frequency point parameters and temperature values ​​are subjected to a second-level fitting process based on the second-level convolutional network. The second-level convolutional network includes three layers of convolutional networks, each layer of which is equipped with 16 1x1 convolutional kernels, 32 3x3 pixel convolutional kernels, and 16 5x5 convolutional kernels.

[0064] The feature frequency point parameters and temperature values ​​after the second-level fitting process are input into the third-level convolutional network of the pre-trained neural network model, and the parameters of the transformer are subjected to the third-level fitting process based on the third-level convolutional network. The third-level convolutional network includes three layers of convolutional networks, each layer of which is equipped with 8 1x1 convolutional kernels, 16 3x3 convolutional kernels, and 16 5x5 convolutional kernels.

[0065] Step S52: Perform batch normalization on the fitted feature vector; then perform 1-pixel x 1-pixel convolution with global average pooling on the batch normalized feature vector to obtain feature values ​​in the specified output dimension.

[0066] Specifically, step S52 further includes:

[0067] Calculate the mean of the set of historical characteristic frequency point parameters and temperature values ​​after fitting;

[0068] Calculate the variance of the characteristic frequency point parameters and temperature values ​​after fitting;

[0069] The fitted feature frequency point parameters and temperature values ​​were normalized using the calculated mean and variance.

[0070] The fitted characteristic frequency point parameters and temperature values ​​are subjected to scaling transformation and offset processing.

[0071] Understandably, the normalization of the fitted feature vectors by taking the mean and variance in this invention accelerates the convergence speed of the entire network model, alleviates the gradient vanishing problem in deep networks to a certain extent, and makes training deep network models easier and more stable.

[0072] Step S53: The feature values ​​are processed through the Dropout layer of the neural network model to suppress overfitting;

[0073] Step S54: Input the processed feature value data stream into the softmax layer for calculation to determine whether the transformer performance meets the requirements.

[0074] Step S6: The identification results of the noise environment data collected by the first data acquisition module and the identification results of the noise environment data collected by the second data acquisition module are fused together to obtain the final prediction fusion result, and to determine whether the transformer meets the noise requirements.

[0075] Specifically, step S6 further includes:

[0076] The results obtained from the data collected in the first data acquisition module are fused with the results obtained from the data collected in the second data acquisition module. If both neural network models output that the product is compliant, the product is determined to be qualified. If both neural network models output that the product is non-compliant, the product is determined to be defective. If the outputs of the two neural network models are inconsistent, the product is determined to be non-compliant. It is understood that in this invention, the first and second data acquisition modules are used to input data into the convolutional network. The neural network models independently perform deep learning corresponding to the product compliance results. The fusion of these modules reduces the bias caused by the randomness of the deep learning models and improves the accuracy of product compliance.

[0077] Understandably, in this invention, a sample data set of characteristic frequency point parameters and temperature values ​​is obtained; the sample data set is then input into a hybrid multi-scale convolutional kernel residual network in batches according to different characteristic frequency point parameters and temperature values ​​to fit the waveform characteristics of the characteristic frequency point parameters and temperature values ​​at different magnitudes and extract feature vectors; the fitted feature vectors are then subjected to batch normalization; the batch-normalized feature vectors are then subjected to convolutional processing with global average pooling to obtain a feature map of a specified output dimension; the feature map is then processed through the Dropout layer of the neural network to suppress overfitting; the processed data stream is then input into the softmax layer for computation, ultimately achieving an objective and accurate prediction result, enabling the prediction result to meet the compliance testing requirements of transformers.

[0078] Implementing the embodiments of the present invention has the following beneficial effects:

[0079] This invention proposes a method for analyzing and judging the compliance of transformer noise. By collecting noise from different locations during transformer operation, feature parameters are generated from the noise features, the feature parameters are standardized in a unified data format, and then a neural network is used for identification and prediction. Ultimately, an objective and accurate prediction result can be achieved, enabling the prediction result to detect the compliance of the transformer.

[0080] In this invention, the data input of the convolutional network is realized through the first data acquisition module and the second data acquisition module. The models of the neural networks are independently used to perform deep learning on the corresponding product compliance results. By fusing the data, the bias caused by the randomness of the deep learning model is reduced, thereby improving the accuracy of product compliance.

[0081] In summary, the embodiments of the present invention can realize automatic testing and analysis of transformer noise, improving work efficiency and accuracy.

[0082] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the claims of the present invention. Therefore, any equivalent changes or modifications made without departing from the spirit disclosed in the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method of analyzing and judging transformer noise compliance, characterized by, At least comprising the following steps: Step S1, energize the transformer to be detected, input test power within a preset time, and form a noise test environment according to the test power; Step S2, collect noise environment data at the position through the first data acquisition module and the second data acquisition device arranged at different positions on the transformer, wherein the noise environment data includes temperature value, transformer parameter, acquisition module parameter and noise value; Step S3, send the noise environment data collected by the first data acquisition module and the second data acquisition module to the cloud center through the wireless mode through the control equipment respectively; Step S4, the cloud center pre-processes the noise environment data collected by the first data acquisition module and the second data acquisition module to generate a parameter format conforming to the input of the neural network model; Step S5, based on the pre-trained convolutional neural network model, whether the noise environment data collected by the first data acquisition module and the second data acquisition module meets the product performance requirement is identified respectively; Step S6, fuse the identification results of the noise environment data collected by the first data acquisition module and the noise environment data collected by the second data acquisition module to obtain the final prediction fusion result, and determine whether the transformer noise meets the requirements.

2. The method of claim 1, wherein, In step S3, further comprising: The multiple cloud servers in the cloud center are pre-deployed with the same network operating system and Hadoop distributed file processing system; the distributed cloud storage server of the cloud center stores parameters of various transformers, and compares the reported transformer parameters to determine the type of the current transformer.

3. The method of claim 2, wherein, In step S4, further comprising: According to the type of the transformer determined by the transformer parameter, the finite element model corresponding to the type of the transformer is obtained, and then the pre-trained neural network model corresponding to the finite element model of the transformer and the acquisition module parameter is obtained; The noise value of the noise environment data collected in the time sequence is extracted and processed for the characteristic frequency point.

4. The method of claim 3, wherein, In step S5, further comprising: Step S51, input the characteristic frequency point and the temperature value of the noise environment data collected in each data acquisition module into the pre-trained neural network model to fit and extract the feature vector; the neural network model is a hybrid multi-scale convolution kernel residual network model; Step S52, batch normalization processing is performed on the fitted feature vector; the feature value of the specified output dimension is obtained by 1 pixel x 1 pixel convolution and global average pooling processing on the batch normalized feature vector; Step S53, the feature value is processed through the Dropout layer of the neural network model to suppress overfitting; Step S54, input the processed feature value data stream into the softmax layer for operation to determine whether the transformer performance meets the requirements.

5. The method of claim 4, wherein, The step S51 further comprises: The characteristic frequency point parameter and the temperature value are input into the first level convolutional network of the pre-trained neural network model, and the characteristic frequency point parameter and the temperature value are subjected to first level fitting processing based on the first level convolutional network, the first level convolutional network comprising three layers of convolutional networks, each layer of convolutional network being provided with 6 2x2 convolutional kernels, 12 1x1 convolutional kernels and 6 5x5 convolutional kernels; The characteristic frequency point parameter and the temperature value subjected to the first level fitting processing are input into the second level convolutional network of the pre-trained neural network model, and the characteristic frequency point parameter and the temperature value are subjected to second level fitting processing based on the second level convolutional network, the second level convolutional network comprising three layers of convolutional networks, each layer of convolutional network being provided with 16 1x1 convolutional kernels, 32 3x3 pixel convolutional kernels and 16 5x5 convolutional kernels; The characteristic frequency point parameter and the temperature value subjected to the second level fitting processing are input into the third level convolutional network of the pre-trained neural network model, and the characteristic frequency point parameter and the temperature value are subjected to third level fitting processing based on the third level convolutional network, the third level convolutional network comprising three layers of convolutional networks, each layer of convolutional network being provided with 8 1x1 convolutional kernels, 16 3x3 convolutional kernels and 16 5x5 convolutional kernels.

6. The method of claim 5, wherein, The step S52 further comprises: calculating the mean value of the fitted historical characteristic frequency point parameter and temperature value set; calculating the variance of the fitted characteristic frequency point parameter and temperature value; normalizing the fitted characteristic frequency point parameter and temperature value using the calculated mean value and variance; performing scale transformation and offset processing on the fitted characteristic frequency point parameter and temperature value.

7. The method according to any one of claims 1 to 6, wherein, In the step S6, further comprising: fusing the results obtained based on the data collected in the first data collection module and the results obtained based on the data collected in the second data collection module; if the output results of the two neural network models are both product compliance, the result is determined as product qualified; if the output results of the two neural network models are both product non-compliance, the result is determined as product unqualified; if the output results of the two neural network models are inconsistent, the result is determined as product non-compliance.

Images