Transformer vibration state monitoring method, system and equipment based on CPO-VMD and CNN-BiLSTM and medium

Abstract

The invention discloses a transformer vibration state monitoring method, system and device based on CPO-VMD and CNN-BiLSTM and a medium, and the method comprises the steps: calculating the total energy of each vibration signal in 24-hour vibration measurement of a transformer, and dividing the load state according to the vibration energy change trend; optimizing the penalty factor and the decomposition modal number of the VMD by using a crown porcupine optimization algorithm to obtain optimal parameter indexes under different loads; decomposing the vibration signals under different loads by using a VMD algorithm optimized by a crown porcupine optimization algorithm, and performing signal reconstruction according to the weight to obtain a signal capable of representing a transformer vibration core excitation source; and constructing a CNN-BiLSTM neural network model, and training the signal subjected to CPO-VMD preprocessing to realize the identification and classification task of the load state of the transformer. And a load supervision system capable of performing long-term supervision is constructed.

Description

Technical Field

[0001] This invention relates to the field of mechanical vibration monitoring technology for converter transformers, and in particular to a method, system, equipment, and medium for monitoring transformer vibration based on CPO-VMD and CNN-BiLSTM. Background Technology

[0002] In ultra-high voltage direct current (UHVDC) transmission systems, transformers, as core equipment for energy conversion and electrical isolation in AC / DC systems, directly affect the reliability of the DC transmission system. Therefore, online monitoring and condition assessment of various transformers in the power grid have become crucial aspects of modern power system operation and maintenance. In recent years, vibration signal-based condition monitoring technology has gradually become a mainstream trend due to its ability to monitor the internal mechanical condition of transformers under live-line conditions.

[0003] Current research on transformer vibration largely focuses on analyzing the signal characteristics of transient vibrations at a specific moment, lacking methods for long-term monitoring of the mechanical state of transformers in operation. This makes the vibration situation of transformers during operation unclear, and prevents timely warnings when slight abnormal vibrations appear in the early stages of a fault. Furthermore, because the load intensity of transformers changes continuously during operation, their vibration signals exhibit certain non-stationary characteristics, making feature extraction difficult and resulting in poor recognition accuracy of related deep learning methods. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] This invention constructs a long-term monitoring system for the mechanical state of transformers. Using a CNN-BiLSTM model based on CPO-VMD, the current state of the transformer can be quickly identified, providing a new technical means for multi-dimensional state assessment of transformers. Experimental verification on a transformer vibration dataset demonstrates that this method has significant advantages in identifying transformer load conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for monitoring the vibration state of a transformer based on CPO-VMD and CNN-BiLSTM, comprising:

[0008] Calculate the total energy of each vibration signal during a 24-hour vibration measurement of the transformer.

[0009] The penalty factor and the number of decomposed modes of the variational mode decomposition algorithm are optimized using a metaheuristic algorithm to obtain the optimal parameter indices under different loads.

[0010] The variational mode decomposition algorithm optimized by the meta-heuristic algorithm is used to decompose the vibration signal under different loads, and the signal is reconstructed according to the weight to obtain the signal that can characterize the core excitation source of transformer vibration.

[0011] A neural network model is constructed to identify and classify the transformer load status.

[0012] As a preferred embodiment of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM described in this invention, wherein: the calculation of the total energy of each vibration signal during 24-hour vibration measurement of the transformer includes,

[0013] Collect real-time data from the transformer and calculate the normalized energy value based on the real-time data;

[0014] Load conditions are classified based on the trend of vibration energy.

[0015] As a preferred embodiment of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM described in this invention, the method employs a metaheuristic algorithm to optimize the penalty factor and the number of decomposed modes of the variational mode decomposition algorithm, obtaining optimal parameter indices under different loads, including:

[0016] Maintaining population diversity and accelerating convergence speed through cyclic population reduction techniques;

[0017] When using metaheuristic algorithms to optimize parameters, a bi-objective decomposition quality fitness function for variational mode decomposition is constructed.

[0018] As a preferred embodiment of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM described in this invention, the method involves: decomposing vibration signals under different loads using a variational mode decomposition algorithm optimized by a metaheuristic algorithm, and reconstructing the signals according to weights to obtain signals that characterize the core excitation source of transformer vibration, including...

[0019] The vibration signals under four load conditions were decomposed and reconstructed. The time-domain waveforms and spectra of each mode after decomposition were plotted, and the proportion of energy of each mode in the original signal energy was calculated.

[0020] The top five modes of each load condition are selected for reconstruction to obtain the filtered reconstructed signal.

[0021] As a preferred embodiment of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM described in this invention, wherein: the method of maintaining population diversity and accelerating convergence speed through cyclic population reduction technology is expressed as follows:

[0022]

[0023] Where N represents the population size; parameter T controls the number of iterations; T max N represents the maximum number of iterations. min This represents the minimum population size; t is the current function value.

[0024] As a preferred embodiment of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM described in this invention, the step of constructing a neural network model and training the signal preprocessed by CPO-VMD to achieve the task of identifying and classifying transformer load states includes:

[0025] Construct convolutional neural network layers to extract local spatial features of vibration signals;

[0026] A bidirectional long short-term memory network is connected in series after the CNN layer;

[0027] The output of the BiLSTM layer is flattened into a one-dimensional vector and connected to a fully connected layer. Finally, it is connected to the output layer and the Softmax activation function is used to classify the vibration signal.

[0028] Validated using vibration datasets from transformers in operation;

[0029] Dimensionality reduction algorithms are used to achieve dimensionality reduction mapping of data, enabling the visualization of data changes.

[0030] As a preferred embodiment of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM described in this invention, the step of achieving dimensionality reduction mapping of data through a dimensionality reduction algorithm to complete the visualization of data changes includes:

[0031] The similarity between data points in a high-dimensional space is calculated using the Gaussian kernel function. Let the dataset be X, x i and x j Let x represent the i-th and j-th data points in X, respectively. i and x j The similarity between them is represented as,

[0032]

[0033] in, Let i be the variance parameter related to i. The similarity is converted into a probability distribution, expressed as:

[0034]

[0035] Where n represents the total number of data points;

[0036] In a low-dimensional space, we define a probability distribution Q for a data point y in that low-dimensional space. i y j Calculate the similarity value Q i|j , is represented as ,

[0037]

[0038] Q i|j Converted to a probability distribution, it is expressed as follows:

[0039]

[0040] Minimizing the KL divergence between the P and Q distributions is expressed as:

[0041]

[0042] Secondly, this invention provides a transformer vibration state monitoring system based on CPO-VMD and CNN-BiLSTM, comprising:

[0043] The system is divided into modules, and the total energy of each vibration signal during 24-hour vibration monitoring of the transformer is calculated. The load state is then classified based on the trend of vibration energy change.

[0044] The calculation module uses the Crowned Porcupine optimization algorithm to optimize the penalty factor and decomposition mode number of VMD, and obtains the optimal parameter index under different loads;

[0045] The reconstruction module uses the VMD algorithm optimized by the porcupine optimization algorithm to decompose the vibration signal under different loads, and reconstructs the signal according to the weight to obtain the signal that can characterize the core excitation source of transformer vibration.

[0046] The classification module constructs a CNN-BiLSTM neural network model, trains it on signals preprocessed by CPO-VMD, and realizes the task of identifying and classifying transformer load status.

[0047] Thirdly, the present invention provides an electronic device, comprising:

[0048] Memory and processor;

[0049] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM.

[0050] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows: First, the present invention establishes a normal operating transformer vibration dataset by measuring vibration, and further subdivides the load status of the transformer in operation by the trend of vibration energy change, thus constructing a load monitoring system that can be monitored for a long time.

[0052] Secondly, the VMD algorithm is used to preprocess the vibration signals of the transformer at various load stages. An improved CPO optimization algorithm is then used to globally optimize the penalty factor α and the number of modes K during the decomposition process. The CPO-optimized VMD algorithm can simultaneously consider modal complexity and reconstruction fidelity, avoiding over- or under-decomposition caused by manually set parameters. Experiments show excellent decomposition results, effectively separating clutter from the transformer vibration signal and making the characteristic frequencies of the stable excitation sources corresponding to the windings and core more prominent.

[0053] Third, a CNN-BiLSTM neural network model was established. This model first extracts the signal features reconstructed by VMD decomposition through CNN layers, and then utilizes the powerful understanding of the context of time-series signals by BiLSTM layers to fully learn the non-steady-state characteristics of transformer vibration signals, ultimately achieving the classification of transformer vibration signals under different loads. The classification process was visualized using t-SNE, verifying the feasibility and superiority of the above method in transformer load state identification and classification.

[0054] Fourth, comparative experiments have verified that the CNN-BiLSTM model based on CPO-VMD signal preprocessing in this invention achieves an average classification accuracy of 99.95% across multiple trials when identifying and classifying various loads of the transformer, with the loss value approaching 0. Compared to CNN, LSTM, and BiLSTM models, this demonstrates superior performance. Figure 14 The model of this invention improves the average accuracy by more than 5% in the classification test of four load states of transformers, which is a significant improvement. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0056] Figure 1This invention relates to a method for monitoring the vibration state of a transformer based on CPO-VMD and CNN-BiLSTM, which is described in one embodiment of the present invention. The method describes the signal vibration intensity trend and load division.

[0057] Figure 2 This is a convergence graph of the CPO algorithm in each load stage according to an embodiment of the present invention.

[0058] Figure 3 This is a schematic diagram of VMD decomposition according to an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram illustrating the ratio of modal energy to original signal energy according to an embodiment of the present invention.

[0060] Figure 5 This is a schematic diagram of the original vibration signal and reconstructed signal under various loads according to an embodiment of the present invention.

[0061] Figure 6 This is a schematic diagram of a convolutional neural network structure according to an embodiment of the present invention.

[0062] Figure 7 This is a schematic diagram of a bidirectional long short-term memory network according to an embodiment of the present invention.

[0063] Figure 8 This is a CNN-BiLSTM model structure based on CPO-VMD as described in one embodiment of the present invention.

[0064] Figure 9 This is a schematic diagram illustrating the model training and testing accuracy according to an embodiment of the present invention.

[0065] Figure 10 This is a schematic diagram illustrating the model training and testing loss according to an embodiment of the present invention.

[0066] Figure 11 This is a schematic diagram of a confusion matrix according to an embodiment of the present invention.

[0067] Figure 12 This is a schematic diagram illustrating the dimensionality reduction visualization of raw data according to an embodiment of the present invention.

[0068] Figure 13 This is a schematic diagram illustrating the final dimensionality reduction visualization according to one embodiment of the present invention.

[0069] Figure 14 This is a schematic diagram comparing the classification effects of different models according to an embodiment of the present invention. Detailed Implementation

[0070] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail 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 them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0071] Example 1, an embodiment of the present invention, provides a method for monitoring transformer vibration status based on CPO-VMD and CNN-BiLSTM, including:

[0072] S1: Calculate the total energy of each vibration signal during the 24-hour vibration measurement of the transformer;

[0073] S2: The hog optimization algorithm is used to optimize the penalty factor and the number of decomposition modes of the variational mode decomposition algorithm to obtain the optimal parameter index under different loads;

[0074] S3: The variational mode decomposition algorithm optimized by the porcupine optimization algorithm is used to decompose the vibration signal under different loads, and the signal is reconstructed according to the weight to obtain the signal that can characterize the core excitation source of transformer vibration.

[0075] S4: Construct a neural network model to train the signal preprocessed by CPO-VMD and realize the task of identifying and classifying transformer load status.

[0076] It should be noted that, addressing the issues of insufficient long-term mechanical condition monitoring in transformer condition monitoring, difficulties in feature extraction due to the non-stationarity of vibration signals, and low accuracy of deep learning recognition, this paper firstly achieves comprehensive monitoring of the long-term operating status of transformers by calculating the total energy of 24-hour continuous vibration signals (S1). Next, a metaheuristic algorithm is used to optimize signal processing parameters (S2), and the optimized algorithm is applied to decompose and reconstruct vibration signals under different loads (S3), effectively extracting signals that characterize the core excitation source and solving the feature extraction problem caused by non-stationarity. Finally, a CNN-BiLSTM neural network model is constructed and trained on the preprocessed signals (S4), significantly improving the accuracy of transformer load condition identification. This provides a more effective online monitoring and condition assessment method for the reliable operation of transformers in ultra-high voltage direct current transmission systems.

[0077] Example 2, refer to Figures 1-13 As an embodiment of the present invention, a transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM is provided, comprising:

[0078] In this embodiment, step S1, calculating the total energy of each vibration signal during the 24-hour vibration measurement of the transformer, includes steps S11-S12:

[0079] S11: Collect real-time data from the transformer and calculate the normalized energy value based on the real-time data.

[0080] S12: Classify the load state based on the trend of vibration energy.

[0081] In this embodiment of the application, the specific calculation method of step S11 is as follows:

[0082] The total energy of each vibration signal during 24-hour vibration measurement is calculated, and the formula for calculating the signal energy is shown in Equation 1.

[0083] The formula for calculating its signal energy is as follows:

[0084]

[0085] Where: f s = 8192Hz, which is the sampling frequency; x(n) is a set of acceleration signals, in m / s². 2 E is the normalized energy value, in meters. 2 / s 4 .

[0086] The specific division method of step S12 is as follows:

[0087] The daily load variation of the transformer is subdivided based on the trend of vibration energy change. Since the vibration energy change exhibits a symmetrical bimodal shape, it is divided into four sub-states: ① low load ② load increase ③ high load fluctuation ④ load decrease. See details below. Figure 1 .

[0088] In one optional embodiment, real-time data of the transformer is collected, and a normalized energy value is calculated based on the real-time data. The calculation method can be based on root mean square (RMS) energy calculation.

[0089] In another optional embodiment, real-time data of the transformer is collected, and a normalized energy value is calculated based on the real-time data. The calculation method can be frequency domain-based energy calculation.

[0090] Step S2 uses a metaheuristic algorithm to optimize the penalty factor and the number of decomposed modes of the variational mode decomposition algorithm to obtain the optimal parameter indices under different loads, including steps S21-S22:

[0091] In this embodiment, the metaheuristic algorithm used is the Crested Porcupine Optimizer (CPO).

[0092] S21: Maintaining population diversity and accelerating convergence speed through cyclic population reduction techniques; The execution process of the Crowned Porcupine optimization algorithm includes a global exploration phase and a local development phase;

[0093] S22: When optimizing parameters using the Crowned Porcupine optimization algorithm, construct a bi-objective decomposition quality fitness function for the variational mode decomposition algorithm. The function integrates the mean sample entropy and the signal reconstruction error.

[0094] The specific calculation method for step S21 is as follows:

[0095] The Crested Porcupine Optimizer (CPO) algorithm is employed for multi-objective parameter optimization to achieve synergistic optimization of the VMD algorithm's penalty factor α and the number of decomposition modes K. The CPO optimization algorithm maintains population diversity and accelerates convergence through cyclic population reduction techniques. Its mathematical model is as follows:

[0096]

[0097] Where N represents the population size; parameter T controls the number of iterations; T max N represents the maximum number of iterations. min This represents the minimum population size; t is the current function value.

[0098] It should be noted that the CPO execution process includes a global exploration phase and a local development phase:

[0099] Global exploration phase: Expand the search scope by simulating the visual and auditory defense behaviors of the hooded porcupine to avoid local optima.

[0100] Visual intimidation:

[0101]

[0102] Sound intimidation:

[0103]

[0104] in, This represents the position of individual i after iteration t, where τ1, τ2, and τ3 are random numbers. U1 represents the position of the predator in the t-th iteration, and U1 is a binary vector.

[0105] It should be noted that the partial development phase involves simulating the scent and physical attack behaviors of hogs to refine the potential solution areas.

[0106] Odor attack:

[0107]

[0108] Physical attack:

[0109]

[0110] Wherein, γ1 is the defense factor. δ is the odor diffusion factor, α is the direction control parameter, and α is the convergence rate factor; τ4 and τ5 are random numbers. It is an inelastic collision force.

[0111] In an alternative embodiment, the metaheuristic algorithm employs the particle swarm optimization (PSO) algorithm, which is computationally efficient and suitable for rapid optimization, but may lead to local optima.

[0112] In another alternative embodiment, the metaheuristic algorithm uses a genetic algorithm (GA), which has strong global search capabilities and is suitable for complex multi-objective optimization, but is computationally time-consuming.

[0113] The specific implementation method of step S22 is as follows:

[0114] When optimizing parameters using the CPO algorithm, a VMD bi-objective decomposition quality fitness function was constructed. This function integrates two metrics: sample entropy and reconstruction error, as shown in the following formula:

[0115]

[0116] In the formula, E is the mean of the sample entropy, and ε recon The signal reconstruction error is used to obtain the bi-objective decomposition quality fitness function through weighted summation. The calculation methods for the mean sample entropy and signal reconstruction error are shown in Equations (8) and (9).

[0117]

[0118] K represents the number of modes in the decomposition, u k E represents the k-th modal component after VMD decomposition. s Let B be the entropy of a single modality sample, m and r represent the embedding dimension and tolerance coefficient, respectively. m (r): The number of vector pairs satisfying distance ≤ r at embedding dimension m, A m (r): The number of vector pairs that satisfy distance ≤ r at embedding dimension m+1.

[0119]

[0120] f is the original signal. This is the reconstructed signal.

[0121] In this embodiment, parameter optimization is performed on four different load signals of the transformer, and the convergence visualization curves during the optimization process are obtained as follows: Figure 2 As shown, where:

[0122] Low load: α=1932, K=12, optimal comprehensive index is: 0.453233;

[0123] Load increase: α=3996, K=11, optimal comprehensive index is: 0.384723;

[0124] High load fluctuation: α=1344, K=12, the optimal comprehensive index is: 0.354415;

[0125] Load reduction: α=3023, K=11, the optimal comprehensive index is: 0.348883.

[0126] In this embodiment of the application, step S3 uses the variational mode decomposition algorithm optimized by the porcupine optimization algorithm to decompose the vibration signal under different loads, and reconstructs the signal according to the weights to obtain the signal that can characterize the core excitation source of transformer vibration, including steps S31-S32:

[0127] S31: The vibration signals under four load conditions are decomposed and reconstructed. Taking the signal under high load fluctuation as an example, the time-domain waveforms and spectrum diagrams of each mode after decomposition are shown in the figure. Figure 3 And calculate the proportion of each mode energy in the original signal energy. Figure 4 .

[0128] S32: The top 5 modes with weights for each load condition are selected for reconstruction to obtain the filtered reconstructed signal. A comparison of the original waveforms and reconstructed waveforms for the four load conditions is shown below. Figure 5 .

[0129] In this embodiment, step S4, which involves constructing a neural network model and training the signal preprocessed by CPO-VMD to achieve the task of identifying and classifying transformer load conditions, includes steps S41-S46:

[0130] It should be noted that, in the embodiments of this application, the neural network used is a CNN-BiLSTM neural network model, and the specific implementation method is as follows:

[0131] S41: Construct a Convolutional Neural Network (CNN) layer to extract local spatial features of the vibration signal. A schematic diagram of the CNN is shown below. Figure 6 As shown.

[0132] S42: A Bidirectional Long Short-Term Memory (BiLSTM) network is connected in series after the CNN layer to capture the dependencies between non-stationary signals of the transformer and establish long-range dependencies. See the BiLSTM schematic diagram. Figure 7 .

[0133] S43: Flatten the output of the BiLSTM layer into a one-dimensional vector, connect it to a fully connected layer, and finally connect it to the output layer. Use the Softmax activation function to classify the vibration signal.

[0134] S44: See the overall model structure setup. Figure 8 .

[0135] In an optional embodiment, the neural network uses a CNN-LSTM neural network model, which is suitable for situations where there are complex nonlinear relationships in the vibration signal or where it is necessary to focus on the characteristics of specific time steps, such as scenarios where the load state changes rapidly and it is necessary to distinguish subtle differences.

[0136] In another alternative embodiment, the neural network uses a CNN-BiLSTM-Attention neural network model, which is suitable for situations where there are complex nonlinear relationships in the vibration signal or where it is necessary to focus on the features of a specific time step, such as scenarios where the load state changes rapidly and it is necessary to distinguish subtle differences.

[0137] S45: Validated using a vibration dataset of in-operation transformers. Model parameter settings are shown in Table 1, and training set partitioning is shown in Table 2. Training and testing accuracy are shown in [Table 1]. Figure 9 Training and testing losses are shown in [link to relevant documentation]. Figure 10 See confusion matrix Figure 11 .

[0138] Table 1 Model Parameter Settings

[0139] CNN-BiLSTM Parameter settings Number of parameters Convolutional layer C1 16×16×16 4112 Batch normalization layer B1 / 32 Max pooling layer M1 Pooling window 3×3, stride 2 0 Convolutional layer C2 32×7×7 25120 Batch normalization layer B2 / 64 Max pooling layer M2 Pooling window 3×3, stride 2 0 Convolutional layer C3 64×3×3 18496 Batch normalization layer B3 / 128 Maximum pooling layer M3 Pooling window 3×3, stride 2 0 BiLSTM layer Number of neurons 32×2 16608 Fully connected layer 64 2112 Output layer 4 260

[0140] Table 2. Training Set Partition Table

[0141]

[0142]

[0143] In this embodiment, the dimensionality reduction algorithm used is the t-SNE algorithm, and the specific implementation is as follows:

[0144] S46: The t-SNE algorithm is used to achieve dimensionality reduction mapping of data, thereby enabling a visual representation of data changes. Its calculation principle is as follows:

[0145] It should be noted that the Gaussian kernel function is used to calculate the similarity between data points in high-dimensional space. Let the dataset be X, x... i and x j Let represent the i-th and j-th data points in X, respectively. Then x i and x j The similarity between them can be expressed as:

[0146]

[0147] in, Let p be the variance parameter related to i. i|j The larger the value, the better for data point x. i and x j The more similar the similarity, the better. The similarity can be converted into a probability distribution using the following formula:

[0148]

[0149] In the formula, n represents the total number of data points.

[0150] It should be noted that in low-dimensional space, a probability distribution Q is defined to represent the similarity between different data points in the low-dimensional space. For data points y in the low-dimensional space... i y j Calculate its similarity value Q i|j The calculation formula is as follows:

[0151]

[0152] Q i|j Convert to probability distribution:

[0153]

[0154] The formula for minimizing the KL divergence between the P and Q distributions is:

[0155]

[0156] See raw data visualization and categorized data visualization. Figure 12 , Figure 13 .

[0157] In an optional embodiment, the dimensionality reduction algorithm uses principal component analysis. In transformer vibration signal analysis, if the data itself does not have a clear category label, PCA can help remove redundant information and improve the efficiency of subsequent processing.

[0158] In another alternative embodiment, the dimensionality reduction algorithm uses linear discriminant analysis. In transformer vibration signal analysis, if it is necessary to classify vibration signals under different load conditions, LDA can improve the classification effect by retaining category information.

[0159] Example 2 illustrates the schematic scheme of a transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM. It should be noted that the technical solution of this transformer vibration state monitoring system based on CPO-VMD and CNN-BiLSTM belongs to the same concept as the aforementioned transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM. Details not described in detail in this example of the transformer vibration state monitoring system based on CPO-VMD and CNN-BiLSTM can be found in the description of the aforementioned transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM.

[0160] This embodiment also provides a transformer vibration condition monitoring system based on CPO-VMD and CNN-BiLSTM, including:

[0161] The module is divided, the total energy of each vibration signal in the 24-hour vibration measurement of the transformer is calculated, and the load state is divided according to the trend of vibration energy change;

[0162] The calculation module uses the Crowned Porcupine optimization algorithm to optimize the penalty factor and decomposition mode number of VMD, and obtains the optimal parameter index under different loads;

[0163] The reconstruction module uses the VMD algorithm optimized by the porcupine optimization algorithm to decompose the vibration signal under different loads, and reconstructs the signal according to the weight to obtain the signal that can characterize the core excitation source of transformer vibration.

[0164] The classification module constructs a CNN-BiLSTM neural network model, trains it on signals preprocessed by CPO-VMD, and realizes the task of identifying and classifying transformer load status.

[0165] This embodiment also provides an electronic device suitable for transformer vibration state monitoring based on CPO-VMD and CNN-BiLSTM, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as proposed in the above embodiment.

[0166] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as proposed in the above embodiments.

[0167] The storage medium proposed in this embodiment and the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0168] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0169] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM, characterized in that, include: Calculate the total energy of each vibration signal during a 24-hour vibration measurement of the transformer. The penalty factor and the number of decomposed modes of the variational mode decomposition algorithm are optimized using a metaheuristic algorithm to obtain the optimal parameter indices under different loads. The variational mode decomposition algorithm optimized by the meta-heuristic algorithm is used to decompose the vibration signal under different loads, and the signal is reconstructed according to the weight to obtain the signal that can characterize the core excitation source of transformer vibration. A neural network model is constructed to identify and classify the transformer load status.

2. The transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as described in claim 1, characterized in that, The total energy of each vibration signal during the 24-hour vibration measurement of the transformer is expressed as follows: Collect real-time data from the transformer and calculate the normalized energy value based on the real-time data; Load conditions are classified based on the trend of vibration energy.

3. The transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as described in claim 2, characterized in that, The method employs a metaheuristic algorithm to optimize the penalty factor and the number of decomposed modes in the variational mode decomposition algorithm, obtaining optimal parameter indices under different loads, including: Maintaining population diversity and accelerating convergence speed through cyclic population reduction techniques; When using metaheuristic algorithms to optimize parameters, a bi-objective decomposition quality fitness function for variational mode decomposition is constructed.

4. The transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as described in claim 3, characterized in that, The variational mode decomposition algorithm, optimized using a meta-heuristic algorithm, decomposes vibration signals under different loads and reconstructs the signals according to weights to obtain signals that characterize the core excitation source of transformer vibration, including... The vibration signals under four load conditions were decomposed and reconstructed. The time-domain waveforms and spectra of each mode after decomposition were plotted, and the proportion of energy of each mode in the original signal energy was calculated. The top five modes of each load condition are selected for reconstruction to obtain the filtered reconstructed signal.

5. The transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as described in claim 4, characterized in that, The technique of maintaining population diversity and accelerating convergence through cyclic population reduction is expressed as follows: Where N represents the population size; parameter T controls the number of iterations; T max N represents the maximum number of iterations. min This represents the minimum population size; t is the current function value.

6. The transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as described in claim 5, characterized in that, The aforementioned construction of a neural network model, trained on signals preprocessed by CPO-VMD, enables the identification and classification of transformer load conditions, including: Construct convolutional neural network layers to extract local spatial features of vibration signals; A bidirectional long short-term memory network is connected in series after the CNN layer; The output of the BiLSTM layer is flattened into a one-dimensional vector and connected to a fully connected layer. Finally, it is connected to the output layer and the Softmax activation function is used to classify the vibration signal. Validated using vibration datasets from transformers in operation; Dimensionality reduction algorithms are used to achieve dimensionality reduction mapping of data, enabling the visualization of data changes.

7. The transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as described in claim 6, characterized in that, The method of using dimensionality reduction algorithms to achieve dimensionality reduction mapping of data and visualize data changes includes, The similarity between data points in a high-dimensional space is calculated using the Gaussian kernel function. Let the dataset be X, x i and x j Let x represent the i-th and j-th data points in X, respectively. i and x j The similarity between them is represented as, in, Let i be the variance parameter related to i. The similarity is converted into a probability distribution, expressed as: Where n represents the total number of data points; In a low-dimensional space, we define a probability distribution Q for a data point y in that low-dimensional space. i y j Calculate the similarity value Q i|j , is represented as , Q i|j Converted to a probability distribution, it is expressed as follows: Minimizing the KL divergence between the P and Q distributions is expressed as:

8. A transformer vibration state monitoring system based on CPO-VMD and CNN-BiLSTM, using the method described in any one of claims 1-7, characterized in that, include: The module is divided, the total energy of each vibration signal in the 24-hour vibration measurement of the transformer is calculated, and the load state is divided according to the trend of vibration energy change; · The calculation module uses the Crowned Porcupine optimization algorithm to optimize the penalty factor and decomposition mode number of VMD, and obtains the optimal parameter index under different loads; The reconstruction module uses the VMD algorithm optimized by the porcupine optimization algorithm to decompose the vibration signal under different loads, and reconstructs the signal according to the weight to obtain the signal that can characterize the core excitation source of transformer vibration. The classification module constructs a CNN-BiLSTM neural network model, trains it on signals preprocessed by CPO-VMD, and realizes the task of identifying and classifying transformer load status.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the transformer vibration state monitoring method based on CPO-VMD and CNN-BiLSTM as described in any one of claims 1 to 7.

Images