Transformer fault detection method based on acoustic emission and related equipment

Abstract

The embodiment of the invention discloses a transformer fault detection method based on acoustic emission and related equipment, and the method comprises the steps: training a machine learning model through employing an acoustic signal and a corresponding fault type, and enabling the trained machine learning model to automatically extract features from the acoustic signal obtained in an equipment operation process, and the fault type of the equipment is accurately judged based on the characteristics, so that the fault detection precision and efficiency are effectively improved, the downtime and maintenance cost of the equipment are reduced, and intelligent management and maintenance of the mechanical equipment are realized.

Description

[Technical field]

[0001] The present invention relates to the technical field of power system fault analysis, and in particular to a transformer fault detection method based on acoustic emission and related equipment. [Background technology]

[0002] As a key equipment for power transmission and conversion, the reliable operation of transformers is crucial to the stability and efficiency of power systems. Transformer failure may not only cause power outages, but also cause major safety accidents. Therefore, effective methods are needed to timely identify and diagnose potential faults of transformers so that corresponding maintenance measures can be taken. Traditional transformer fault diagnosis technologies mainly include oil sample analysis, vibration analysis, and infrared imaging, which often have certain limitations. Although oil sample analysis can provide rich information, it requires power outages for oil sample collection, which limits its application in real-time monitoring; vibration analysis relies on sensors installed on the equipment, but due to the complex structure of the transformer, some key parts cannot be effectively monitored; infrared imaging is mainly suitable for the detection of external faults, and it is difficult to find internal hidden dangers. [Summary of the invention]

[0003] In view of this, the present invention provides a transformer fault detection method based on acoustic emission and related equipment.

[0004] The specific technical solution of the first embodiment of the present invention is: a transformer fault detection method based on acoustic emission, the method comprising: constructing a three-dimensional model of the transformer; using the three-dimensional model to continuously simulate the operating status of the transformer under different working conditions, and obtaining acoustic signal data of the transformer under different operating conditions; the operating status includes a normal operating status and a fault status corresponding to different fault types; extracting the sound features of the acoustic signal data, and using the sound features as input variables, and outputting the operating status corresponding to the sound features as target variables to train a preset machine learning model, and obtain the training accuracy of the preset machine learning model, until the training accuracy is greater than a preset threshold, completing the training of the preset machine learning model, and obtaining the trained machine learning model; obtaining the acoustic signal data to be identified of the faulty transformer; inputting the acoustic signal data to be identified into the trained machine learning model to obtain the fault status corresponding to the acoustic signal to be identified.

[0005] Preferably, constructing the three-dimensional model of the transformer includes: setting physical field parameters and material properties in finite element software according to the structural dimensions and physical characteristics of the transformer to establish the three-dimensional model of the transformer.

[0006] Preferably, the extraction of sound features of the sound signal data includes: framing and windowing the sound signal data to obtain multiple short-time frames; performing Fourier transform on each short-time frame to obtain spectrum information of the short-time frame; constructing a filter group according to the Mel scale, and filtering the spectrum signal through the Mel filter group; calculating the logarithmic energy of each filter output for the spectrum signal processed by the Mel filter group; performing discrete cosine transform on the logarithmic energy to obtain cepstral coefficients; and extracting cepstral coefficients of preset values ​​to constitute the sound features.

[0007] Preferably, the training method for training the preset machine learning model includes: evaluating the correlation between the sound features and the transformer fault state, selecting the sound features with the highest correlation to train the preset machine learning model; and using the distance correlation method to evaluate the relationship between the sound features and the target variable output.

[0008] Preferably, during the training process of the preset machine learning model, a population crossover mutation algorithm is used to train the preset machine learning model.

[0009] Preferably, the population crossover mutation algorithm also includes: constructing a decision space for multiple dynamic decision variables brought about by environmental changes during the crossover mutation process, identifying the correlation of decision variables at different time points by analyzing historical data of sound features; and selecting decision variables in the new environment through a correlation selection optimization algorithm.

[0010] Preferably, the population crossover mutation algorithm also includes: using a tournament selection method to select individuals with the highest fitness from the current population for reproduction; performing crossover and mutation operations on the selected individuals to generate new individuals; the crossover is a single-point crossover, and the mutation is a Gaussian mutation.

[0011] The specific technical scheme of the second embodiment of the present invention is: a transformer fault detection system based on acoustic emission, the system comprising: a model building module, an acoustic signal acquisition module, a training module, a signal acquisition module to be identified and a fault analysis module; the model building module is used to build a three-dimensional model of the transformer; the acoustic signal acquisition module is used to use the three-dimensional model to continuously simulate the operating state of the transformer under different working conditions, and obtain the acoustic signal data of the transformer under different operating states; the operating state includes a normal operating state and a fault state corresponding to different fault types; the training module is used to extract the sound features of the acoustic signal data, and use the sound features as input variables, and use the operating state corresponding to the sound features as a target variable output to train a preset machine learning model, and obtain the training accuracy of the preset machine learning model, until the training accuracy is greater than a preset threshold, the training of the preset machine learning model is completed, and the trained machine learning model is obtained; the signal acquisition module to be identified is used to obtain the acoustic signal data to be identified of the faulty transformer; the fault analysis module is used to input the acoustic signal data to be identified into the trained machine learning model to obtain the fault state corresponding to the acoustic signal to be identified.

[0012] The specific technical solution of the third embodiment of the present invention is: a transformer fault detection device based on acoustic emission, including a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method described in the first embodiment of the present application.

[0013] The specific technical solution of the fourth embodiment of the present invention is: a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the method described in the first embodiment of the present application.

[0014] Implementing the embodiments of the present invention will have the following beneficial effects:

[0015] The present invention trains a machine learning model by using acoustic signals and corresponding fault types. The trained machine learning model can automatically extract features from the acoustic signals obtained during the operation of the equipment, and accurately determine the type of fault in the equipment based on these features, thereby effectively improving the accuracy and efficiency of fault detection, reducing equipment downtime and maintenance costs, and realizing intelligent management and maintenance of mechanical equipment.

Brief Description of the Drawings

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 is a flowchart of the steps of a transformer fault detection method based on acoustic emission;

[0018] Figure 2 It is a structural diagram of a transformer fault detection system based on acoustic emission;

[0019] Figure 3 A diagram of the internal structure of a computer device;

[0020] Among them, 201 is a model building module; 202 is an acoustic signal acquisition module; 203 is a training module; 204 is a signal acquisition module to be identified; 205 is a fault analysis module. [Specific implementation method]

[0021] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] The terms "first", "second", etc. in the specification, claims and drawings of this application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not limited to the listed steps or modules, but optionally includes steps or modules that are not listed, or optionally includes other steps or modules that are inherent to these processes, methods, products or devices.

[0023] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0024] See also Figure 1, is a flowchart of a transformer fault detection method based on acoustic emission in the first embodiment of the present application, which realizes intelligent management and maintenance of mechanical equipment. The method includes:

[0025] Step 101: construct a three-dimensional model of the transformer;

[0026] Step 102: Continuously simulate the operating state of the transformer under different working conditions using the three-dimensional model to obtain acoustic signal data of the transformer under different operating states; the operating state includes a normal operating state and a fault state corresponding to different fault types;

[0027] Step 103: extract the sound features of the sound signal data, and use the sound features as input variables, and use the operating conditions corresponding to the sound features as target variables to output a preset machine learning model, and obtain the training accuracy of the preset machine learning model, until the training accuracy is greater than a preset threshold, the training of the preset machine learning model is completed, and the trained machine learning model is obtained;

[0028] Step 104, obtaining the acoustic signal data to be identified of the faulty transformer;

[0029] Step 105: input the acoustic signal data to be identified into the trained machine learning model to obtain the fault state corresponding to the acoustic signal to be identified.

[0030] Specifically, a continuous simulation is performed on the operating state of the transformer under different working conditions, and the transformer is tested by a continuous periodic frequency response method to obtain the acoustic signal data of the transformer under multiple working conditions. The internal conditions of the transformer, the deformation of the windings, and the occurrence of mechanical failures are determined according to the mechanical stress and hot spot temperature in the simulation, and the sound data and corresponding failure conditions contained in the test are recorded. Among them, the sound data collected when the transformer is working normally and when a short circuit occurs is collected using a high-sensitivity microphone or acoustic emission sensor. Sound signals are collected under the conditions of normal operation of the transformer, minor failures (such as partial discharge), and serious failures (such as insulation aging). At least 10 minutes of sound data are collected under each state to ensure the representativeness of the sample. A comprehensive comparison is made with the transformer data obtained by simulation to confirm the correspondence between the simulation parameters and the actual operating data, and the setting of the transformer resistance and inductance data correction value in the simulation model is confirmed based on the data difference.

[0031] The method in this embodiment trains a machine learning model by using acoustic signals and corresponding fault types. The trained machine learning model can automatically extract features from the acoustic signals obtained during the operation of the equipment, and accurately determine the type of fault in the equipment based on these features, thereby effectively improving the accuracy and efficiency of fault detection, reducing equipment downtime and maintenance costs, and realizing intelligent management and maintenance of mechanical equipment.

[0032] In a specific embodiment, the construction of the three-dimensional model of the transformer includes: according to the structural dimensions and physical characteristics of the transformer, setting the physical field parameters and material properties in the finite element software to establish the three-dimensional model of the transformer. Specifically, according to the structural dimensions and physical characteristics of the actual power transformer, the physical field parameters and material properties are set in the finite element software to establish a complete and detailed three-dimensional model. Compare the voltage and current data of the simulation model under different feeding modes to determine the optimal voltage and current feeding mode under different working conditions; for the transformer accessory structure, establish a simplified model with actual physical functions by studying the distribution relationship of the structural magnetic field under different working conditions.

[0033] In a specific embodiment, the method of extracting the sound features of the sound signal data includes: framing and windowing the sound signal data to obtain multiple short-time frames; performing Fourier transform on each short-time frame to obtain spectrum information of the short-time frame; constructing a filter group according to the Mel scale, and filtering the spectrum signal through the Mel filter group; calculating the logarithmic energy of each filter output for the spectrum signal processed by the Mel filter group; performing discrete cosine transform on the logarithmic energy to obtain cepstral coefficients; and extracting cepstral coefficients of preset values ​​to constitute the sound features.

[0034] Specifically, the sound signal is processed using the Mel Frequency Cepstral Coefficient (MFCC) method. The detailed steps are as follows: signal preprocessing, using filters (such as low-pass filters) to remove high-frequency noise, and using spectral subtraction and other methods to suppress noise. The amplitude of the signal is normalized to reduce the impact of environmental noise. The formula is Where x is the input sound signal. The signal is framed and windowed to divide the continuous sound signal into small blocks (frames). Each frame is usually 20-40 milliseconds with an overlap of 50%. Assuming the frame length is (N), the signal of each frame is: [x m =[x(mR),x(mR+1),...,x(mR+N-1)], where R is the frame shift, usually half the frame length, and m is the frame index. Perform FFT on each frame signal and use the formula Get the spectrum. Convert the frequency to Mel scale. The relationship between Mel frequency and linear frequency is: Design a set of triangular filters on the Mel frequency scale to cover the required frequency range and calculate the Mel energy Em = ∑f|X(f)|2 H m (f), and take the logarithm to obtain the logarithmic Mel energy Lm = log(Em+∈), and perform discrete cosine transform (DCT) to convert the logarithmic Mel energy into MFCC, remove the correlation and compress the feature dimension. The specific formula is:

[0035] The final output is the MFCC features of each frame. Usually the first 12 to 13 MFCC coefficients are taken as the feature vector MFCC = [c0, c1, ..., c K-1 ].

[0036] In a specific embodiment, the training method for training a preset machine learning model includes: evaluating the correlation between sound features and transformer fault status, selecting the sound features with the highest correlation to train the preset machine learning model; and using a distance correlation method to evaluate the relationship between the sound features and the target variable output.

[0037] Specifically, the extracted sound features (MFCC feature vectors) are used as input variables, and the corresponding transformer states are used as target outputs. 1. Use MFCC feature vectors to define transformer state labels, such as "normal", "overload", "fault", etc. The data set is divided into training set, validation set and test set, usually with a ratio of 70% training, 15% validation, and 15% testing. 2. Model construction, using the IDP method to evaluate the correlation between MFCC features and transformer states, and select the most significant features for training. Use the distance correlation (DCor) method to evaluate the relationship between features and target variables. Select a suitable machine learning model (such as a neural network) as a prediction model. The model structure may include an input layer, a hidden layer, and an output layer. Among them, the input layer is the input of the selected MFCC feature vector. The hidden layer uses multiple hidden layers to capture complex nonlinear relationships. The output layer outputs the state of the transformer (usually using a softmax activation function for multi-classification).

[0038] In a specific embodiment, during the training of a preset machine learning model, a population crossover mutation algorithm is used to train the preset machine learning model. The population crossover mutation algorithm can perform a global search in the entire parameter space by simulating the process of natural selection and genetic variation, thereby finding a better combination of model parameters. Compared with traditional local search algorithms such as gradient descent, the population crossover mutation algorithm is more likely to find the global optimal solution and avoid falling into the local optimum.

[0039] In a specific embodiment, the population crossover mutation algorithm further includes: constructing a decision space for multiple dynamic decision variables brought about by environmental changes during the crossover mutation process, identifying the correlation of decision variables at different time points by analyzing historical data of sound features; and selecting decision variables in a new environment by using a correlation selection optimization algorithm. Specifically, according to the analysis target, decision variables related to sound features are defined, and these decision variables may represent a certain specific sound state, environmental attributes, or user behavior, etc. Using historical data, the correlation between decision variables and sound features is evaluated. This can be achieved by calculating correlation coefficients, building regression models, or using classification algorithms. Analyzing the changes in the correlation of decision variables at different time points to reveal the law of their change over time helps to understand the performance of decision variables in different environments and provide guidance for subsequent optimization algorithms. According to the correlation of decision variables and the results of time dynamic analysis, a suitable optimization algorithm is selected. In a new environment, a specific decision variable is selected according to the requirements of the optimization algorithm and the correlation of decision variables. Among them, a decision space is constructed for multiple dynamic decision variables brought about by environmental changes, and the correlation of decision variables at different time points is identified by analyzing historical data. A more suitable decision variable is selected in a new environment by using a correlation selection optimization algorithm. Using time series analysis techniques, the changing trends of decision variables are modeled to identify potential periodic or sudden changes. This improves the accuracy of predictions. Based on the results of variables dynamically extracted from the decision space, the population is updated to ensure that the population can adapt to the new environment. After the environmental change occurs, new individuals are generated using the prediction model. These individuals are generated based on the prediction of the optimal solution under the current environmental conditions, aiming to improve the adaptability of the population. Individuals with low fitness or no longer adapting to the new environment in the screening model are replaced, and some excellent existing individuals are selected to cross or mutate with the newly generated individuals to achieve population fusion to maintain the diversity of the population. After the population is updated, the performance of the new population is continuously evaluated to ensure the quality and diversity of its solutions under the new environment. If necessary, the population is further adjusted to optimize performance. Historical population data, including individual decision variables and corresponding objective function values, are collected and stored during the algorithm operation. These data are used to train the prediction model to identify the dynamic trend of the solution. Feature selection techniques are used to evaluate the correlation between input variables and target variables. Variables that have a significant impact on the optimization results are identified, and redundant and irrelevant variables are removed to improve the accuracy of the prediction model. The performance of the next generation of population is predicted based on the selected features. The model analyzes historical data and learns the evolution trend of solutions to predict the optimal solution in the new environment.

[0040] In a specific embodiment, the population crossover mutation algorithm further includes: using a tournament selection method to select individuals with the highest fitness from the current population for reproduction; performing crossover and mutation operations on the selected individuals to generate new individuals; the crossover is a single-point crossover, and the mutation is a Gaussian mutation. Specifically, a fitness function is defined during the training process to evaluate the performance of the model, and a cross entropy loss function is usually used: Use the tournament selection method to select individuals with high fitness from the current population for reproduction. Perform crossover and mutation operations on the selected individuals to generate new individuals. Crossover uses single-point crossover and mutation uses Gaussian mutation.

[0041] In a specific embodiment, in an algorithm framework for setting a change threshold based on the standard deviation of historical data, stream processing technology is used to collect and process historical data of the objective function in real time, so as to quickly calculate the standard deviation, analyze historical data, automatically learn the optimal change threshold, and adaptively optimize the change threshold. The threshold setting is optimized based on the characteristics of different environments. At the same time, the statistical significance test is combined to determine the significance of the change of the objective function, and the operator decision is made based on the significance to trigger the population update. A data visualization dashboard is constructed to display the changes in the objective function, and an automated alarm mechanism is established. Through feedback loops and multi-objective optimization strategies, the threshold setting is gradually adjusted to further improve the adaptability of the algorithm. In addition, the system's ability to identify and respond to environmental changes is enhanced by combining ensemble learning and reinforcement learning methods, thereby optimizing the overall algorithm performance.

[0042] In a specific embodiment, the development and application of online monitoring background monitoring software is established to realize the triggering, collection, transmission, storage and data management of monitoring signals, collect the relevant parameters of the acoustic emission of power transformers and perform statistical analysis, and have a visual human-machine interface. Based on the collected information, the database is called for rapid analysis, and important data such as the location of the transformer fault, the aging and wear of the transformer, etc. can be confirmed in real time. It can be evaluated whether the power transformer has functional defects or damage, and its degree of wear can also be evaluated. Such a system will allow for more effective estimation of the life of the equipment and planning of renovations.

[0043] In the specific embodiments, see Figure 2, is a structural diagram of a transformer fault detection system based on acoustic emission in the second embodiment of the present application, the system comprising: a model building module 201, an acoustic signal acquisition module 202, a training module 203, a signal acquisition module 204 to be identified and a fault analysis module 205; the model building module 201 is used to build a three-dimensional model of the transformer; the acoustic signal acquisition module 202 is used to use the three-dimensional model to continuously simulate the operating state of the transformer under different working conditions, and obtain the acoustic signal data of the transformer under different operating states; the operating state includes a normal operating state and a fault state corresponding to different fault types; the training module 203 is used to extract the The sound features of the sound signal data are used, and the sound features are used as input variables, and the operating conditions corresponding to the sound features are used as target variables for output to train the preset machine learning model, and the training accuracy of the preset machine learning model is obtained, until the training accuracy is greater than a preset threshold, the training of the preset machine learning model is completed, and the trained machine learning model is obtained; the signal to be identified acquisition module 204 is used to obtain the sound signal data to be identified of the faulty transformer; the fault analysis module 205 is used to input the sound signal data to be identified into the trained machine learning model to obtain the fault state corresponding to the sound signal to be identified.

[0044] The transformer fault detection system based on acoustic emission mainly uses machine learning to detect the possibility of mechanical defects in power transformers. The proposed diagnostic method uses wavelet decomposition to extract details from the sound signal and then determines the numerical value of each detail. With the help of the original database of sound signal waveforms recorded for defective power transformers, a wide range of mechanical loss conditions are simulated, and the sound signals and their corresponding transformer states are used for learning to create a transformer fault identification system and create a diagnostic system based on this, which can not only evaluate whether the power transformer has functional defects or damage, but also its degree of wear.

[0045] The system in this embodiment trains the machine learning model by using acoustic signals and corresponding fault types. The trained machine learning model can automatically extract features from the acoustic signals obtained during the operation of the equipment, and accurately determine the type of fault in the equipment based on these features, thereby effectively improving the accuracy and efficiency of fault detection, reducing equipment downtime and maintenance costs, and realizing intelligent management and maintenance of mechanical equipment.

[0046] In a specific embodiment, the third embodiment of the present application provides a transformer fault detection device based on acoustic emission, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method described in the first embodiment of the present application. The device in this embodiment trains a machine learning model by using acoustic signals and corresponding fault types. The trained machine learning model can automatically extract features from the acoustic signals obtained during the operation of the equipment, and accurately determine the type of fault in the equipment based on these features, thereby effectively improving the accuracy and efficiency of fault detection, reducing equipment downtime and maintenance costs, and realizing intelligent management and maintenance of mechanical equipment.

[0047] In a specific embodiment, the fourth embodiment of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the processor executes the steps of the method described in the first embodiment of the present application. The storage medium in this embodiment trains the machine learning model by using acoustic signals and corresponding fault types. The trained machine learning model can automatically extract features from the acoustic signals obtained during the operation of the equipment, and accurately determine the type of fault in the equipment based on these features, thereby effectively improving the accuracy and efficiency of fault detection, reducing equipment downtime and maintenance costs, and realizing intelligent management and maintenance of mechanical equipment.

[0048] Figure 3 The internal structure of a computer device in one embodiment is shown. The computer device can be a terminal or a server. Figure 3 The computer device includes a processor, a memory, etc. connected via a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement the method in this embodiment. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement the method in this embodiment. Those skilled in the art can understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0049] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

[0050] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A transformer fault detection method based on acoustic emission, characterized in that: The method comprises: Construct a 3D model of the transformer; The three-dimensional model is used to continuously simulate the operating state of the transformer under different working conditions to obtain the acoustic signal data of the transformer under different operating states; the operating state includes a normal operating state and a fault state corresponding to different fault types; Extracting the sound features of the sound signal data, and using the sound features as input variables, and using the operating conditions corresponding to the sound features as target variables to output a preset machine learning model, and obtaining the training accuracy of the preset machine learning model, until the training accuracy is greater than a preset threshold, completing the training of the preset machine learning model, and obtaining a trained machine learning model; Acquiring the acoustic signal data to be identified of the faulty transformer; The acoustic signal data to be identified is input into the trained machine learning model to obtain the fault state corresponding to the acoustic signal to be identified.

2. The transformer fault detection method based on acoustic emission according to claim 1, characterized in that: The three-dimensional model of the transformer is constructed, including: according to the structural size and physical characteristics of the transformer, setting physical field parameters and material properties in finite element software to establish the three-dimensional model of the transformer.

3. The transformer fault detection method based on acoustic emission according to claim 1, characterized in that: The extracting the sound features of the sound signal data comprises: Framing and windowing the acoustic signal data to obtain a plurality of short time frames; Performing Fourier transform on each short time frame to obtain frequency spectrum information of the short time frame; Construct a filter bank according to the Mel scale, and filter the spectrum signal through the Mel filter bank; Calculate the logarithmic energy of each filter output for the spectral signal processed by the Mel filter bank; The logarithmic energy is subjected to discrete cosine transformation to obtain the cepstrum coefficients; The cepstral coefficients of preset values ​​are extracted to form the sound features.

4. The transformer fault detection method based on acoustic emission according to claim 1, characterized in that: The training method for training the preset machine learning model includes: Evaluate the correlation between the sound features and the transformer fault status, and select the sound features with the highest correlation to train the preset machine learning model; The relationship between the sound features and the target variable output was evaluated using the distance correlation method.

5. The transformer fault detection method based on acoustic emission according to claim 1, characterized in that: During the training process of the preset machine learning model, a population crossover mutation algorithm is used to train the preset machine learning model.

6. The transformer fault detection method based on acoustic emission according to claim 5, characterized in that: The population crossover mutation algorithm also includes: A decision space is constructed for multiple dynamic decision variables caused by environmental changes during the cross-mutation process. By analyzing the historical data of sound features, the correlation of decision variables at different time points is identified. The decision variables are selected in new environments through the relevance selection optimization algorithm.

7. The transformer fault detection method based on acoustic emission according to claim 5, characterized in that: The population crossover mutation algorithm also includes: Use the tournament selection method to select the individuals with the highest fitness from the current population for reproduction; The selected individuals are subjected to crossover and mutation operations to generate new individuals; the crossover is a single-point crossover, and the mutation is a Gaussian mutation.

8. A transformer fault detection system based on acoustic emission, characterized in that: The system comprises: a model building module, an acoustic signal acquisition module, a training module, a to-be-identified signal acquisition module and a fault analysis module; The model building module is used to build a three-dimensional model of the transformer; The acoustic signal acquisition module is used to continuously simulate the operating state of the transformer under different working conditions using the three-dimensional model to obtain the acoustic signal data of the transformer under different operating conditions; the operating conditions include normal operating conditions and fault conditions corresponding to different fault types; The training module is used to extract the sound features of the acoustic signal data, and use the sound features as input variables, and use the operating conditions corresponding to the sound features as target variables to output and train the preset machine learning model, and obtain the training accuracy of the preset machine learning model, until the training accuracy is greater than a preset threshold, the training of the preset machine learning model is completed, and the trained machine learning model is obtained; The to-be-identified signal acquisition module is used to acquire to-be-identified acoustic signal data of a faulty transformer; The fault analysis module is used to input the acoustic signal data to be identified into the trained machine learning model to obtain the fault state corresponding to the acoustic signal to be identified.

9. A transformer fault detection device based on acoustic emission, comprising a memory and a processor, characterized in that: The memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the steps of the method as claimed in claims 1 to 7.

Images