Power transformer fault identification method

Abstract

The invention relates to the technical field of electric power, and discloses a power transformer fault identification method, which comprises the following steps that: a least square support vector machine model of a preset grey wolf optimization algorithm driven by a mechanical energy field is adopted as a transformer mechanical fault identification model, so that better performance is realized in a small sample scene, and the fault identification accuracy is improved. The problem that deep learning is prone to overfitting in a small sample scene is avoided, the accuracy and reliability of fault recognition are improved, meanwhile, data of multiple sensors are comprehensively considered, target feature vectors and signal feature vectors are obtained through multi-feature extraction, splicing and dimension reduction processing, the state of the transformer can be represented more comprehensively and accurately, and the fault recognition accuracy and reliability are improved. The defect of incomplete representation of a single feature is overcome, and a powerful guarantee is provided for safe and stable operation of the power transformer.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a method for identifying faults of a power transformer. Background Art

[0002] Power transformers are critical equipment in power systems, and their status is directly related to the safety and stability of the entire system. Traditional power transformer operation and maintenance models rely primarily on conventional threshold judgments and fixed-period maintenance strategies. However, this model has significant inherent flaws. Conventional threshold judgments struggle to accurately and accurately perceive the actual status of power transformers in real time, making it difficult to detect subtle abnormalities during equipment operation. Furthermore, fixed-period maintenance strategies lack specificity, potentially leading to over- or under-maintenance, wasted resources, and untimely equipment failures.

[0003] Power transformers have complex internal mechanical structures and exhibit diverse failure modes, making traditional mechanical fault identification methods face numerous challenges in practical application. Traditional methods struggle to establish effective evaluation systems. This is primarily due to the following reasons: While deep learning technology possesses powerful self-learning capabilities for feature self-learning, it is prone to overfitting when processing small sample sizes, resulting in poor model generalization and inability to accurately identify faults. Furthermore, the non-stationary nature of power transformer vibration signals significantly limits feature extraction, making it difficult to fully and accurately represent the transformer's state with a single feature alone. This, in turn, impacts the accuracy and reliability of fault identification. Summary of the Invention

[0004] Based on this, it is necessary to propose a power transformer fault identification method to address the above problems. By adopting a preset least squares support vector machine model based on the gray wolf optimization algorithm driven by mechanical energy field as the transformer mechanical fault identification model, it has better performance in small sample scenarios, avoiding the problem of deep learning being prone to overfitting in small sample scenarios, and improving the accuracy and reliability of fault identification. At the same time, by comprehensively considering the data of multiple sensors, through multi-feature extraction, splicing and dimensionality reduction processing, the target feature vector and signal feature vector are obtained, which can more comprehensively and accurately characterize the status of the transformer, overcome the defect of incomplete single feature representation, and provide a strong guarantee for the safe and stable operation of the power transformer.

[0005] To achieve the above object, the present invention provides a method for identifying a power transformer fault in a first aspect, the method comprising:

[0006] Acquire vibration signals from various sensors, wherein the various sensors are installed at different locations on the power transformer;

[0007] Performing multi-feature extraction on the vibration signal of each sensor to obtain a signal feature vector, and sequentially splicing the vibration signals of each sensor to obtain a high-dimensional feature vector, and performing dimensionality reduction processing on the high-dimensional feature vector to obtain a target feature vector;

[0008] The target feature vector and the signal feature vector are input into a transformer mechanical fault identification model to obtain a mechanical fault identification result of the power transformer; wherein, the transformer mechanical fault identification model is a preset least squares support vector machine model based on a gray wolf optimization algorithm driven by a mechanical energy field.

[0009] Optionally, the method further includes:

[0010] Obtain historical vibration signals from each sensor;

[0011] Performing multi-feature extraction on the historical vibration signals of each sensor to obtain a historical signal feature vector, sequentially concatenating the historical vibration signals of each sensor to obtain a historical high-dimensional feature vector, and performing dimensionality reduction processing on the historical high-dimensional feature vector to obtain a historical target feature vector;

[0012] Determining a label corresponding to the historical target feature vector and a label corresponding to each signal feature in the historical signal feature vector;

[0013] Using the Grey Wolf optimization algorithm driven by the mechanical energy field, the model parameters in the least squares support vector machine model are adjusted to obtain an initial model, wherein the initial model is an initial least squares support vector machine model based on the Grey Wolf optimization algorithm driven by the mechanical energy field;

[0014] The historical target feature vector, the historical signal feature vector, the label corresponding to the historical target feature vector, and the label corresponding to each signal feature in the historical signal feature vector are input into the initial model for training to obtain the transformer mechanical fault recognition model.

[0015] Optionally, the use of the Grey Wolf optimization algorithm driven by the mechanical energy field to adjust the model parameters in the least squares support vector machine model to obtain the initial model includes:

[0016] Using the mechanical energy field-driven gray wolf optimization algorithm, the position of each gray wolf in the gray wolf population is updated, and the population position of the gray wolf population is updated according to the positions of all gray wolves in the gray wolf population, until the number of iterations corresponding to the current iteration is equal to the preset maximum number of iterations, and the optimal model parameters corresponding to the global optimal fitness value in all iterations are obtained, wherein the position of each gray wolf includes the model parameters;

[0017] The model parameters in the least squares support vector machine model are adjusted according to the optimal model parameters corresponding to all iterations to obtain the initial model.

[0018] Optionally, the position of each gray wolf in the gray wolf population is updated using the gray wolf optimization algorithm based on mechanical energy field drive, and the population position of the gray wolf population is updated according to the positions of all gray wolves in the gray wolf population, until the number of iterations corresponding to the current iteration is equal to the preset maximum number of iterations, and the optimal model parameters corresponding to the global optimal fitness value in all iterations are obtained, including:

[0019] Dividing the historical vibration signal of each sensor into time windows to obtain the vibration signal of each sensor in each time window, wherein the total number of time window divisions is greater than or equal to the preset maximum number of iterations;

[0020] According to the vibration signals of each sensor in the i-th time window, the mechanical energy field function in the i-th time window is constructed, and the initial value of i is 1;

[0021] Determining the position of an alpha gray wolf in the gray wolf population at the i+1th iteration according to the mechanical energy field function in the i-th time window;

[0022] Determining the gradient of the β-gray wolf in the gray wolf population at the i-th iteration based on the mechanical energy field function in the i-th time window and the position of the β-gray wolf in the gray wolf population at the i-th iteration, and determining the gradient of the δ-gray wolf in the gray wolf population at the i-th iteration based on the mechanical energy field function in the i-th time window and the position of the δ-gray wolf in the gray wolf population at the i-th iteration;

[0023] Determine the position of the β-gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the β-gray wolf in the gray wolf population at the i-th iteration, and determine the position of the δ-gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the δ-gray wolf in the gray wolf population at the i-th iteration;

[0024] Determine the position of the ω gray wolf in the gray wolf population at the i+1th iteration based on the vibration signals of each sensor in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration;

[0025] Determine the population position of the gray wolf population at the i+1th iteration based on the position of the α gray wolf in the i+1th iteration, the position of the β gray wolf in the i+1th iteration, the position of the δ gray wolf in the i+1th iteration, and the position of the ω gray wolf in the i+1th iteration;

[0026] The population position of the gray wolf population at the (i+1)th iteration is used as the position of the gray wolf newly added to the gray wolf population at the (i+1)th iteration;

[0027] Let i=i+1 until i is equal to the preset maximum number of iterations, and use the position of the gray wolf corresponding to the global optimal fitness value in all iterations as the optimal model parameter corresponding to the global optimal fitness value in all iterations.

[0028] Optionally, constructing a mechanical energy field function in the i-th time window according to the vibration signals of each sensor in the i-th time window includes:

[0029] Using EWT, the vibration signal of each sensor in the i-th time window is decomposed to obtain the modal components of each sensor in the i-th time window;

[0030] Determine the energy proportion and maximum singular value of each modal component of each sensor in the i-th time window;

[0031] Determine the average energy proportion of each sensor in the i-th time window based on the energy proportion of each modal component of each sensor in the i-th time window, and determine the global maximum singular value of each sensor in the i-th time window based on the maximum singular value of each modal component of each sensor in the i-th time window;

[0032] Determine the time domain correlation coefficient between the data point at the tth moment in the vibration signal of the nth sensor in the i-th time window and the data point at the tth moment in the vibration signal of each sensor except the nth sensor in the i-th time window; where n successively takes integers greater than 0 until n equals the total number of sensors, and obtain the respective time domain correlation coefficients of each sensor at the tth moment in the i-th time window, where the initial value of t is 1;

[0033] Determine the weighted comprehensive index at the t-th moment in the i-th time window according to the time domain correlation coefficients of the sensors at the t-th moment in the i-th time window;

[0034] Let t = t + 1, until t is equal to the total number of moments in the i-th time window, and obtain the weighted comprehensive index of each moment in the i-th time window;

[0035] Determine the mechanical energy of the nth sensor at each moment in the i-th time window based on the weighted comprehensive index at each moment in the i-th time window, as well as the average energy proportion and global maximum singular value of the n-th sensor in the i-th time window; where n is an integer greater than 0 in sequence until n equals the total number of sensors, and obtain the mechanical energy of each sensor at each moment in the i-th time window;

[0036] According to the mechanical energy of each sensor at each moment in the i-th time window and the position of each sensor, a mechanical energy field function in the i-th time window is constructed.

[0037] Optionally, determining the mechanical energy of the nth sensor at each moment in the i-th time window based on the weighted comprehensive index at each moment in the i-th time window, and the average energy proportion and global maximum singular value of the nth sensor in the i-th time window includes:

[0038] Using the formula Determine the mechanical energy of the nth sensor at each moment in the i-th time window;

[0039] Among them, E n,i,t is the mechanical energy of the nth sensor at the tth moment in the i-th time window, k1 is the first preset empirical constant, P n,i is the weighted average energy of the nth sensor in the i-th time window, λ max,n,i is the global maximum singular value of the nth sensor in the i-th time window, e is a natural constant, k2 is the second preset empirical constant, d i,t is the weighted comprehensive index at the tth moment in the i-th time window.

[0040] Optionally, determining the position of the β gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the β gray wolf in the gray wolf population at the i-th iteration, and determining the position of the δ gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the δ gray wolf in the gray wolf population at the i-th iteration, includes:

[0041] Using the formula Determine the position of the β-grey wolf in the gray wolf population at the i+1th iteration, and the position of the δ-grey wolf at the i+1th iteration;

[0042] Among them, when r = β, X β (i+1) is the position of the β gray wolf in the gray wolf population at the i+1th iteration, X β (i) is the position of β gray wolf in the gray wolf population at the i-th iteration, is the gradient of the β gray wolf in the gray wolf population at the i-th iteration. When r = δ, X δ (i+1) is the position of the gray wolf delta in the gray wolf population at the i+1th iteration, X δ (i) is the position of δ gray wolves in the gray wolf population at the i-th iteration, is the gradient of δ gray wolves in the gray wolf population at the i-th iteration, A i is the random attenuation value corresponding to the β-gray wolf or δ-gray wolf in the gray wolf population at the i-th iteration, and Δi is the preset time step.

[0043] Optionally, determining the position of the ω gray wolf in the gray wolf population at the (i+1)th iteration based on the vibration signals of each sensor in the (i)th time window and the position of the ω gray wolf in the gray wolf population at the (i)th iteration includes:

[0044] Reconstruct the component time domain signal of each sensor in the i-th time window according to the second modal component and the third modal component of each sensor in the i-th time window;

[0045] The component time domain signals of each sensor in the i-th time window are subjected to feature extraction to obtain the kurtosis of each sensor in the i-th time window;

[0046] According to the kurtosis of each sensor in the i-th time window, the weighted average kurtosis in the i-th time window is determined;

[0047] The position of the ω gray wolf in the gray wolf population at the i+1th iteration is determined according to the weighted average kurtosis in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration.

[0048] Optionally, determining the position of the ω gray wolf in the gray wolf population at the i+1th iteration based on the weighted average kurtosis in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration includes:

[0049] Using formula X ω (i+1)=X ω (i)+C i ·S i rand() determines the position of the ω gray wolf in the gray wolf population at the i+1th iteration;

[0050] Among them, X ω (i+1) is the position of the ω gray wolf in the gray wolf population at the i+1th iteration, X ω (i) is the position of the ω gray wolf in the gray wolf population at the i-th iteration, C i is the random disturbance value corresponding to the ω gray wolf in the gray wolf population at the i-th iteration, S i is the weighted average kurtosis in the i-th time window, and rand() is the random number generation function.

[0051] Optionally, determining the population position of the gray wolf population at the i+1th iteration based on the position of the α gray wolf in the gray wolf population at the i+1th iteration, the position of the β gray wolf in the i+1th iteration, the position of the δ gray wolf in the i+1th iteration, and the position of the ω gray wolf in the i+1th iteration includes:

[0052] Determining the mechanical energy of the α gray wolf at the i+1th iteration, the mechanical energy of the β gray wolf at the i+1th iteration, the mechanical energy of the δ gray wolf at the i+1th iteration, and the mechanical energy of the ω gray wolf at the i+1th iteration in the gray wolf population according to the position of the α gray wolf at the i+1th iteration, the position of the β gray wolf at the i+1th iteration, the mechanical energy of the δ gray wolf at the i+1th iteration, and the mechanical energy of the ω gray wolf at the i+1th iteration in the gray wolf population;

[0053] Determine a weight value of the α gray wolf at the i+1th iteration, a weight value of the β gray wolf at the i+1th iteration, a weight value of the δ gray wolf at the i+1th iteration, and a weight value of the ω gray wolf at the i+1th iteration in the gray wolf population according to the mechanical energy of the α gray wolf at the i+1th iteration, the mechanical energy of the β gray wolf at the i+1th iteration, the mechanical energy of the δ gray wolf at the i+1th iteration, and the mechanical energy of the ω gray wolf at the i+1th iteration;

[0054] The population position of the gray wolf population at the i+1 iteration is determined based on the weight value and position of the α gray wolf in the gray wolf population at the i+1 iteration, the weight value and position of the β gray wolf in the i+1 iteration, the weight value and position of the δ gray wolf in the i+1 iteration, and the weight value and position of the ω gray wolf in the i+1 iteration.

[0055] To achieve the above object, the present invention provides, in a second aspect, a power transformer fault identification device, the device comprising:

[0056] an acquisition module, configured to acquire vibration signals from various sensors, wherein the various sensors are installed at different positions of the power transformer;

[0057] A feature module is used to extract multiple features from the vibration signal of each sensor to obtain a signal feature vector, and to sequentially concatenate the vibration signals of each sensor to obtain a high-dimensional feature vector, and to perform dimensionality reduction processing on the high-dimensional feature vector to obtain a target feature vector;

[0058] A model prediction module is used to input the target feature vector and the signal feature vector into a transformer mechanical fault identification model to obtain a mechanical fault identification result of the power transformer; wherein the transformer mechanical fault identification model is a preset least squares support vector machine model based on a gray wolf optimization algorithm driven by a mechanical energy field.

[0059] To achieve the above-mentioned object, the present invention provides, in a third aspect, a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the method as described in any one of the first aspects.

[0060] To achieve the above-mentioned objectives, the present invention provides a computer device in a fourth aspect, comprising 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 method as described in any one of the first aspects.

[0061] The embodiment of the present invention has the following beneficial effects: the above method obtains the vibration signals of each sensor, wherein each sensor has been installed at a different position of the power transformer, and then performs multi-feature extraction on the vibration signal of each sensor to obtain a signal feature vector, and sequentially splices the vibration signals of each sensor to obtain a high-dimensional feature vector, and performs dimensionality reduction processing on the high-dimensional feature vector to obtain a target feature vector, and finally inputs the target feature vector and the signal feature vector into the transformer mechanical fault recognition model to obtain the mechanical fault recognition result of the power transformer, wherein the transformer mechanical fault recognition model is a preset gray wolf based on mechanical energy field drive The least squares support vector machine model of the optimization algorithm; that is, by adopting the preset least squares support vector machine model of the gray wolf optimization algorithm driven by the mechanical energy field as the transformer mechanical fault identification model, it has better performance in small sample scenarios, avoids the problem of overfitting of deep learning in small sample scenarios, and improves the accuracy and reliability of fault identification. At the same time, by comprehensively considering the data of multiple sensors, through multi-feature extraction, splicing and dimensionality reduction processing, the target feature vector and signal feature vector are obtained, which can more comprehensively and accurately characterize the status of the transformer, overcome the defect of incomplete single feature representation, and provide a strong guarantee for the safe and stable operation of power transformers. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] in:

[0064] Figure 1 A schematic diagram of a power transformer fault identification method according to an embodiment of the present application;

[0065] Figure 2 This is a schematic diagram of a power transformer fault identification device according to an embodiment of the present application;

[0066] Figure 3 1 is a diagram of the internal structure of a computer device in some embodiments. DETAILED DESCRIPTION

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0068] Power transformers are critical equipment in power systems, and their status is directly related to the safety and stability of the entire system. Traditional power transformer operation and maintenance models rely primarily on conventional threshold judgments and fixed-period maintenance strategies. However, this model has significant inherent flaws. Conventional threshold judgments struggle to accurately and accurately perceive the actual status of power transformers in real time, making it difficult to detect subtle abnormalities during equipment operation. Furthermore, fixed-period maintenance strategies lack specificity, potentially leading to over- or under-maintenance, wasted resources, and untimely equipment failures.

[0069] Power transformers have complex internal mechanical structures and exhibit diverse failure modes, making traditional mechanical fault identification methods face numerous challenges in practical application. Traditional methods struggle to establish effective evaluation systems. This is primarily due to the following reasons: While deep learning technology possesses powerful self-learning capabilities for feature self-learning, it is prone to overfitting when processing small sample sizes, resulting in poor model generalization and inability to accurately identify faults. Furthermore, the non-stationary nature of power transformer vibration signals significantly limits feature extraction, making it difficult to fully and accurately represent the transformer's state with a single feature alone. This, in turn, impacts the accuracy and reliability of fault identification.

[0070] In response to the above problems, the present application proposes a method for identifying power transformer faults. By adopting a preset least squares support vector machine model based on the gray wolf optimization algorithm driven by a mechanical energy field as a transformer mechanical fault identification model, it has better performance in small sample scenarios, avoids the problem of deep learning being prone to overfitting in small sample scenarios, and improves the accuracy and reliability of fault identification. At the same time, by comprehensively considering the data of multiple sensors, the target feature vector and signal feature vector are obtained through multi-feature extraction, splicing and dimensionality reduction processing, which can more comprehensively and accurately characterize the state of the transformer, overcome the defect of incomplete single feature representation, and provide a strong guarantee for the safe and stable operation of the power transformer. The specific implementation principle will be described in detail in the following embodiments.

[0071] In a first aspect, the present application provides a method for identifying power transformer faults.

[0072] See also Figure 1, is a schematic diagram of a power transformer fault identification method according to an embodiment of the present application, the method comprising:

[0073] Step 110: Acquire vibration signals of various sensors, wherein the various sensors are installed at different positions of the power transformer.

[0074] The number of sensors can be pre-set by the operator according to actual needs.

[0075] In some embodiments, the present application preferably sets the number of sensors to 4, that is, the 4 sensors are installed at different positions of the power transformer in a diagonal distribution of two by two.

[0076] Step 120: Perform multi-feature extraction on the vibration signal of each sensor to obtain a signal feature vector, and sequentially concatenate the vibration signals of each sensor to obtain a high-dimensional feature vector, and perform dimensionality reduction processing on the high-dimensional feature vector to obtain a target feature vector.

[0077] Among them, the signal eigenvector includes but is not limited to the maximum singular value, energy proportion, center frequency, kurtosis, waveform factor, approximate entropy, center of gravity frequency, frequency standard deviation, margin factor, root mean square value, etc.

[0078] Regarding the method of determining the signal eigenvector, in some embodiments, EWT can be used to decompose the vibration signal of each sensor to obtain each modal component of each sensor, and then determine the maximum singular value, energy proportion and center frequency of each modal component of each sensor, and reconstruct the component time domain signal of each sensor based on the second modal component and the third modal component of each sensor, and perform multi-feature extraction on the component time domain signal of each sensor to obtain the kurtosis and waveform factor of each sensor, and finally use the maximum singular value, energy proportion and center frequency of each modal component of each sensor, as well as the kurtosis and waveform factor of each sensor as the signal eigenvector.

[0079] Among them, EWT stands for Empirical Wavelet Transform, that is, empirical wavelet transform.

[0080] Regarding the number of modal components into which the vibration signal of each sensor is decomposed, in some embodiments, the present application preferably decomposes the vibration signal of each sensor into three modal components.

[0081] Regarding the method for determining the maximum singular value of each modal component of each sensor, in some embodiments, a continuous wavelet transform can be used to transform each modal component of each sensor to obtain a time-frequency diagram of each modal component of each sensor, and then the time-frequency diagram of each modal component of each sensor is transformed to obtain a time-frequency matrix of each modal component of each sensor. Finally, singular value decomposition is performed on the time-frequency matrix of each modal component of each sensor to obtain the maximum singular value of each modal component of each sensor.

[0082] Regarding the method for determining the target feature vector, in some embodiments, PCA may be used to perform dimensionality reduction processing on the high-dimensional feature vector to obtain the target feature vector.

[0083] Among them, PCA stands for Principal Component Analysis.

[0084] Step 130: Input the target feature vector and the signal feature vector into the transformer mechanical fault identification model to obtain the mechanical fault identification result of the power transformer; wherein the transformer mechanical fault identification model is a preset least squares support vector machine model based on the gray wolf optimization algorithm driven by the mechanical energy field.

[0085] The transformer mechanical fault identification model herein refers to a pre-trained model used to predict the mechanical fault identification result of the output power transformer based on the input target feature vector and signal feature vector.

[0086] It should be noted that the transformer mechanical fault identification model is a preset least squares support vector machine model based on the Grey Wolf optimization algorithm driven by the mechanical energy field, that is, the Grey Wolf optimization algorithm is optimized by the mechanical energy field drive, and then the model parameters of the least squares support vector machine model are optimized by the optimized Grey Wolf optimization algorithm. The optimized least squares support vector machine model is the least squares support vector machine model based on the Grey Wolf optimization algorithm driven by the mechanical energy field.

[0087] Among them, the mechanically-driven Grey Wolf optimization algorithm is Mechanically-Driven Grey Wolf Optimizer (MD-GWO), and the least squares support vector machine model is Least Squares Support Vector Machines (LSSVM).

[0088] Regarding the types included in the mechanical fault identification results of the power transformer, in some embodiments, the mechanical fault identification results of the power transformer include but are not limited to normal state, loose insulation plate state, loose moving contact state, loose upper static contact state, loose lower static contact state, contact ablation state and contact wear state, etc.

[0089] In an embodiment of the present application, by adopting a preset least squares support vector machine model based on the gray wolf optimization algorithm driven by mechanical energy field as a transformer mechanical fault identification model, it has better performance in small sample scenarios, avoids the problem of deep learning being prone to overfitting in small sample scenarios, and improves the accuracy and reliability of fault identification. At the same time, by comprehensively considering the data of multiple sensors, through multi-feature extraction, splicing and dimensionality reduction processing, the target feature vector and signal feature vector are obtained, which can more comprehensively and accurately characterize the state of the transformer, overcome the defect of incomplete single feature representation, and provide a strong guarantee for the safe and stable operation of the power transformer.

[0090] In addition, in addition to the above-mentioned advantages of avoiding the overfitting of deep learning in small sample scenarios, improving the accuracy and reliability of fault identification, and overcoming the defects of incomplete single feature representation, the power transformer fault identification method also has the following advantages: Multi-dimensional information integration: Vibration signals are obtained by multiple sensors installed at different positions of the power transformer, and these signals are comprehensively considered. It is possible to capture the operating status information of the transformer from multiple angles and dimensions. Sensors at different positions may be more sensitive to different types of faults or abnormal changes. The fusion of multi-sensor data can make up for the incompleteness and limitations of single sensor information, thereby more comprehensively reflecting the actual status of the transformer; Improve fault detection sensitivity: Multi-sensor data fusion helps Discover some subtle anomalies that are difficult to detect with a single sensor. For example, some faults may cause weak vibration changes in local areas, which may not be accurately captured by a single sensor. The collaborative work of multiple sensors can enhance the perception of these subtle changes and improve the sensitivity of fault detection; comprehensively characterize fault characteristics: adopt a variety of feature extraction methods, such as empirical wavelet transform (EWT) to extract the maximum singular value, energy proportion, center frequency, kurtosis and waveform factor of the modal component, which can describe the characteristics of the vibration signal from different angles. These features cover multiple aspects such as the time domain and frequency domain of the signal, and can more comprehensively characterize the fault characteristics of the transformer, which helps to improve the accuracy of fault identification; enhance feature robustness: by splicing multiple The vibration signal of the sensor is processed and dimension reduction is performed to obtain the target feature vector, and the signal feature vector of each sensor is combined to form a comprehensive feature set. This multi-feature fusion method can enhance the robustness of the feature to noise and interference, and reduce the misjudgment caused by the influence of noise on a single feature; Optimize model parameters: Use the mechanical energy field driven gray wolf optimization algorithm (MD-GWO) to optimize the model parameters of the least squares support vector machine model (LSSVM). The gray wolf optimization algorithm is an optimization algorithm based on swarm intelligence with good global search capability and convergence speed. The mechanical energy field drive further enhances the optimization performance of the algorithm, and can more effectively find the optimal combination of model parameters, thereby improving the model. Improved generalization ability and recognition accuracy of the model; Adaptation to complex fault modes: The internal mechanical structure of the power transformer is complex and the fault modes are diverse. The least squares support vector machine model based on the gray wolf optimization algorithm driven by the mechanical energy field can better adapt to such complex fault modes. By optimizing the model parameters, the model can more accurately identify different types of faults, including normal state, loose insulation plate state, loose moving contact state, and other states; Prevention of equipment failure: Accurate mechanical fault identification can timely discover potential fault hazards of the transformer, take maintenance and repair measures in advance, avoid further development and expansion of the fault, thereby preventing the occurrence of equipment failure, reducing the risk of power outages and equipment damage, and improving the operational safety of the power system;Optimize maintenance strategies: Traditional fixed-cycle maintenance strategies lack specificity and may lead to over- or under-maintenance. This mechanical fault identification method can accurately perceive the actual status of the transformer in real time, providing a basis for formulating reasonable maintenance strategies. Based on the fault identification results, targeted maintenance can be carried out, improving maintenance efficiency and reducing maintenance costs, while ensuring reliable equipment operation and enhancing power system stability. Reduce resource waste: This avoids the excessive maintenance that may be caused by traditional fixed-cycle maintenance strategies, reduces unnecessary manpower, material resources, and time investment, and reduces operation and maintenance costs. Extend equipment life: By promptly detecting and handling faults, it can reduce damage to equipment caused by faults, extend equipment service life, and further reduce equipment replacement costs.

[0091] In a feasible implementation, the method in the above embodiment also includes: obtaining historical vibration signals of each sensor; performing multi-feature extraction on the historical vibration signals of each sensor to obtain a historical signal feature vector, and splicing the historical vibration signals of each sensor in sequence to obtain a historical high-dimensional feature vector, and performing dimensionality reduction processing on the historical high-dimensional feature vector to obtain a historical target feature vector; determining the label corresponding to the historical target feature vector and the label corresponding to each signal feature in the historical signal feature vector; using the gray wolf optimization algorithm driven by a mechanical energy field to adjust the model parameters in the least squares support vector machine model to obtain an initial model, which is an initial least squares support vector machine model based on the gray wolf optimization algorithm driven by a mechanical energy field; inputting the historical target feature vector, the historical signal feature vector, the label corresponding to the historical target feature vector, and the label corresponding to each signal feature in the historical signal feature vector into the initial model for training to obtain a transformer mechanical fault recognition model.

[0092] The initial model here refers to the original model that has not been trained.

[0093] The method for determining the historical signal feature vector and the historical target feature vector is similar to the method for determining the signal feature vector and the target feature vector. You can refer to the above-mentioned relevant embodiments for determining the signal feature vector and the target feature vector, and will not be repeated here.

[0094] Regarding the objective function used in the training process of the transformer mechanical fault identification model, in some embodiments, the expression of the objective function is: Among them, f(x) is the objective function value, P is the total number of labels, α p is the pth Lagrange multiplier, y p is the pth label, K(x p ,x q ) is the kernel function, x pis the feature corresponding to the p-th label (such as the various signal features contained in the signal feature vector, and the target feature vector, etc.), x q is the feature corresponding to the qth label (such as the various signal features contained in the signal feature vector, and the target feature vector, etc.), and b is the bias term.

[0095] In the embodiment of the present application, by using historical vibration signals for model training, a more accurate and reliable transformer mechanical fault identification model can be obtained, further improving the accuracy and reliability of fault identification, and providing a more solid guarantee for the safe and stable operation of power transformers.

[0096] It is understandable that to improve the accuracy of the model: obtain the historical vibration signals of each sensor, and perform multi-feature extraction, splicing and dimensionality reduction processing on them to obtain historical target feature vectors and historical signal feature vectors. At the same time, determine the labels corresponding to these feature vectors. These labels contain the real information of the transformer in different states. Inputting these labeled historical feature vectors into the initial model for training can enable the model to learn a more accurate mapping relationship between fault characteristics and states, thereby improving the accuracy of the model in fault identification; enhance model reliability: use the gray wolf optimization algorithm driven by mechanical energy field to adjust the model parameters in the least squares support vector machine model to obtain the initial model. This optimization algorithm has good global search capability and convergence speed, and can more effectively find the optimal model parameter combination. During the training process, the model continuously adjusts the parameters To adapt to historical data, the trained transformer mechanical fault identification model has better generalization ability and robustness, and can more reliably respond to various situations in actual operation; achieve accurate fault identification: the transformer mechanical fault identification model obtained through historical data training can more accurately predict the mechanical fault identification results of the output power transformer based on the input target feature vector and signal feature vector, which helps to timely discover potential fault hazards of the transformer, take maintenance and repair measures in advance, and avoid further development and expansion of the fault, thereby ensuring the safe and stable operation of the power system; optimize the model training process: using historical vibration signals for model training can make full use of existing data resources and avoid data waste. At the same time, through reasonable feature extraction and label determination, the model training process is more scientific and effective, and the efficiency and quality of model training are improved.

[0097] In a feasible implementation method, the gray wolf optimization algorithm based on mechanical energy field driving is used in the above embodiment to adjust the model parameters in the least squares support vector machine model to obtain an initial model, including: using the gray wolf optimization algorithm based on mechanical energy field driving to update the position of each gray wolf in the gray wolf population, and updating the population position of the gray wolf population according to the positions of all gray wolves in the gray wolf population, until the number of iterations corresponding to the current iteration is equal to the preset maximum number of iterations, and obtaining the optimal model parameters corresponding to the global optimal fitness value in all iterations, wherein the position of each gray wolf includes the model parameters; according to the optimal model parameters corresponding to all iterations, the model parameters in the least squares support vector machine model are adjusted to obtain the initial model.

[0098] The preset maximum number of iterations may be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics. Of course, it may also be pre-set by the operator based on actual needs.

[0099] It should be noted that, in the process of updating the position of each wolf in the wolf population using the gray wolf optimization algorithm driven by the mechanical energy field, a local fitness value is calculated for each wolf in each iteration, and then the global fitness value of each iteration is determined based on the local fitness value of each wolf in each iteration (that is, in one iteration, among the local fitness values ​​of each wolf, the minimum local fitness value is used as the global fitness value), and then the global fitness value of each iteration is compared with the global optimal fitness value in all iterations, and the global optimal fitness value in all iterations can be updated (if If the global fitness value of the i-th iteration is less than the global optimal fitness value in all iterations from 1 to i-1, the global fitness value of the i-th iteration can be used as the global optimal fitness value in all iterations from 1 to i-1; if the global fitness value of the i-th iteration is greater than or equal to the global optimal fitness value in all iterations from 1 to i-1, it will not be replaced), and when the number of iterations corresponding to the current iteration is equal to the preset maximum number of iterations, the position of the gray wolf corresponding to the global optimal fitness value in all iterations will be used as the optimal model parameter corresponding to the global optimal fitness value in all iterations.

[0100] It should be further explained that the method for determining the initial position of each gray wolf in the gray wolf population can be set to a default initialization position, and of course, it can also be determined by a random initialization position; the method for determining the local fitness value can be calculated according to the fitness formula in the gray wolf optimization algorithm.

[0101] Of course, in other embodiments, the accuracy calculation method of the initial model after cross-validation can also be used as the fitness formula to calculate the local fitness value.

[0102] Regarding the specific value of the preset maximum number of iterations, in some embodiments, the present application preferably sets the preset maximum number of iterations to 30.

[0103] Regarding the method of adjusting the model parameters in the least squares support vector machine model, in some embodiments, the model parameters in the least squares support vector machine model can be directly adjusted to the optimal model parameters corresponding to all iterations to obtain an initial model; of course, in other embodiments, the model parameters in the least squares support vector machine model can also be adjusted to the sum or difference between the optimal model parameters corresponding to all iterations and the preset model error parameters to obtain an initial model; wherein, the preset model error parameters can be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics, and of course, can also be pre-set by the operator based on actual needs.

[0104] Regarding the parameters included in the optimal model parameters, in some embodiments, the optimal model parameters include but are not limited to the penalty factor and kernel function parameters in the least squares support vector machine model; wherein the kernel function parameters include but are not limited to kernel width, order, scaling coefficient, bias term, etc.

[0105] In an embodiment of the present application, the least squares support vector machine model parameters are adjusted by using a gray wolf optimization algorithm driven by a mechanical energy field, which effectively improves the effect of model parameter optimization and thereby enhances the performance and fault identification capability of the model.

[0106] It can be understood that the model parameter optimization effect is improved: the position of each gray wolf in the gray wolf population is updated using a gray wolf optimization algorithm driven by a mechanical energy field, and the population position is updated according to the positions of all gray wolves in the gray wolf population until the preset maximum number of iterations is reached, and the optimal model parameters corresponding to the global optimal fitness value are obtained. This method can more effectively find the optimal model parameter combination, avoid the problem that traditional optimization methods may fall into local optimality, and improve the effect of model parameter optimization; enhance model performance: according to the optimal model parameters corresponding to all iterations, the model parameters in the least squares support vector machine model are adjusted to obtain the initial model. Since the optimized model parameters are more in line with the actual data characteristics, the generalization ability and robustness of the model can be enhanced, so that the model has better performance in practical applications; improve fault identification ability: the optimized model can more accurately identify mechanical faults of power transformers, because after the model parameters are optimized, they can better capture the characteristic information in the data, thereby improving the accuracy and reliability of fault identification, which is of great significance for ensuring the safe and stable operation of the power system.

[0107] In a feasible implementation, the gray wolf optimization algorithm based on mechanical energy field drive in the above embodiment is used to update the position of each gray wolf in the gray wolf population, and update the population position of the gray wolf population according to the positions of all gray wolves in the gray wolf population, until the number of iterations corresponding to the current iteration is equal to the preset maximum number of iterations, and the optimal model parameters corresponding to the global optimal fitness value in all iterations are obtained, including: dividing the historical vibration signal of each sensor into time windows, obtaining the vibration signal of each sensor in each time window, wherein the total number of time window divisions is The number is greater than or equal to the preset maximum number of iterations; according to the vibration signals of each sensor in the i-th time window, a mechanical energy field function is constructed in the i-th time window, and the initial value of i is 1; according to the mechanical energy field function in the i-th time window, the position of the α gray wolf in the gray wolf population in the i+1 iteration is determined; according to the mechanical energy field function in the i-th time window and the position of the β gray wolf in the gray wolf population in the i-th iteration, the gradient of the β gray wolf in the gray wolf population in the i-th iteration is determined, and according to the mechanical energy field function in the i-th time window and the position of the δ gray wolf in the gray wolf population in the i-th iteration, the gradient of the β gray wolf in the gray wolf population in the i-th iteration is determined. According to the position of the δ gray wolf in the gray wolf population at the i-th iteration, the gradient of the δ gray wolf in the gray wolf population at the i-th iteration is determined; according to the gradient and position of the β gray wolf in the gray wolf population at the i-th iteration, the position of the β gray wolf in the gray wolf population at the i+1 iteration is determined, and according to the gradient and position of the δ gray wolf in the gray wolf population at the i-th iteration, the position of the δ gray wolf in the gray wolf population at the i+1 iteration is determined; according to the vibration signals of each sensor in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration, the position of the ω gray wolf in the gray wolf population at the i+1 iteration is determined; according to the α gray wolf in the gray wolf population at the i-th iteration, the position of the ω gray wolf in the gray wolf population at the i+1 iteration is determined. The position of the gray wolf population at the i+1th iteration, the position of the β gray wolf at the i+1th iteration, the position of the δ gray wolf at the i+1th iteration and the position of the ω gray wolf at the i+1th iteration are used to determine the population position of the gray wolf population at the i+1th iteration; the population position of the gray wolf population at the i+1th iteration is used as the position of the gray wolf newly added to the gray wolf population at the i+1th iteration; let i=i+1 until i is equal to the preset maximum number of iterations, and the position of the gray wolf corresponding to the global optimal fitness value in all iterations is used as the optimal model parameter corresponding to the global optimal fitness value in all iterations.

[0108] The total number of time window divisions can be pre-set by the operator according to actual needs, as long as the total number of time window divisions is greater than or equal to the preset maximum number of iterations, which is not limited here.

[0109] Regarding the method for determining the position of the α gray wolf in the gray wolf population at the i+1th iteration, in some embodiments, the formula Determine the position of α gray wolf in the gray wolf population at the i+1th iteration; where X a(i+1) is the position of the α gray wolf in the gray wolf population at the i+1th iteration, To obtain the position of the independent variable (x, y, z) when the mechanical energy field function in the i-th time window is at its maximum value, E i (x, y, z, t) is the mechanical energy field function in the i-th time window.

[0110] Regarding the determination of the gradient of the β gray wolf in the gray wolf population at the i-th iteration and the gradient of the δ gray wolf at the i-th iteration, in some embodiments, the formula Determine the gradient of the gray wolf population at iteration β and the gradient of the gray wolf population at iteration δ; where, when r = β, is the gradient of the gray wolf population β in the i-th iteration, (x β,i ,y β,i ,z β,i ) is the position of β gray wolf in the gray wolf population at the i-th iteration, (x β,i ,y β,i ,z β,i )=X β (i), when r = δ, is the gradient of δ gray wolf in the gray wolf population at the i-th iteration, (x β,i ,y β,i ,z β,i ) is the position of δ gray wolf in the gray wolf population at the i-th iteration, (x β,i ,y β,i ,z β,i )=X δ (i) For E i The partial derivative of (x,y,z,t) with respect to x, For E i The partial derivative of (x,y,z,t) with respect to y, For E i The partial derivative of (x,y,z,t) with respect to z, E i (x, y, z, t) is the mechanical energy field function in the i-th time window.

[0111] Regarding the determination of the population position of the gray wolf population at the i+1th iteration, in some embodiments, the average position of the position of the α gray wolf in the gray wolf population at the i+1th iteration, the position of the β gray wolf at the i+1th iteration, the position of the δ gray wolf at the i+1th iteration, and the position of the ω gray wolf at the i+1th iteration can be used as the population position of the gray wolf population at the i+1th iteration; of course, in other embodiments, the weighted average position of the position of the α gray wolf in the gray wolf population at the i+1th iteration, the position of the β gray wolf at the i+1th iteration, the position of the δ gray wolf at the i+1th iteration, and the position of the ω gray wolf at the i+1th iteration can also be used as the population position of the gray wolf population at the i+1th iteration.

[0112] In an embodiment of the present application, the position of the gray wolf population is dynamically updated by using a gray wolf optimization algorithm driven by a mechanical energy field, combined with time window division and mechanical energy field function, which can more accurately find the optimal model parameters and improve model performance and fault identification capabilities.

[0113] It can be understood that the position of the gray wolf population is dynamically updated: by dividing the historical vibration signal of each sensor into time windows, and constructing a mechanical energy field function based on the vibration signal of each time window, the position of different gray wolves (α gray wolf, β gray wolf, δ gray wolf, ω gray wolf) in the gray wolf population in each iteration is dynamically determined. This method can make full use of the timing information of the historical vibration signal, so that the position update of the gray wolf population is more in line with the actual data characteristics, which helps to find better model parameters; accurately determine the optimal model parameters: during the iteration process, according to the mechanical energy field function and the position of each gray wolf in the gray wolf population, gradually update the position of the gray wolf until the preset maximum number of iterations is reached. At this time, the position of the gray wolf corresponding to the global optimal fitness value in all iterations is used as the optimal model parameter. This method can more effectively avoid trapping. Enter the local optimum, accurately find the optimal combination of model parameters, and improve the effect of model parameter optimization; Improve model performance: Since the optimized model parameters are more in line with the actual data characteristics, the generalization ability and robustness of the model can be enhanced. In practical applications, the optimized model can better adapt to different data distributions and noise interference, improve the accuracy and reliability of fault identification, and provide stronger guarantees for the safe and stable operation of power transformers; Enhance fault identification ability: By more accurately finding the optimal model parameters, the optimized model can better capture the characteristic information in the data, thereby improving the ability to identify mechanical faults of power transformers, which helps to timely discover potential fault hazards of transformers, take maintenance and repair measures in advance, avoid further development and expansion of faults, and ensure the safe and stable operation of the power system.

[0114] In a feasible implementation, the mechanical energy field function in the i-th time window is constructed according to the vibration signal of each sensor in the i-th time window in the above embodiment, including: using EWT to decompose the vibration signal of each sensor in the i-th time window to obtain each modal component of each sensor in the i-th time window; determining the energy proportion and maximum singular value of each modal component of each sensor in the i-th time window; determining the average energy proportion of each sensor in the i-th time window according to the energy proportion of each modal component of each sensor in the i-th time window, and determining the global maximum singular value of each sensor in the i-th time window according to the maximum singular value of each modal component of each sensor in the i-th time window; determining each time domain correlation coefficient between the data point at the t-th time in the vibration signal of the n-th sensor in the i-th time window and the data point at the t-th time in the vibration signal of each sensor except the n-th sensor in the i-th time window; wherein n is successively greater than 0 integers until n is equal to the total number of sensors, and the time domain correlation coefficients of each sensor at the t-th moment in the i-th time window are obtained, where the initial value of t is 1; based on the time domain correlation coefficients of each sensor at the t-th moment in the i-th time window, the weighted comprehensive index at the t-th moment in the i-th time window is determined; let t = t + 1, until t is equal to the total number of moments in the i-th time window, and the weighted comprehensive index at each moment in the i-th time window is obtained; based on the weighted comprehensive index at each moment in the i-th time window, and the average energy proportion and global maximum singular value of the n-th sensor in the i-th time window, the mechanical energy of the n-th sensor at each moment in the i-th time window is determined; where n successively takes integers greater than 0 until n is equal to the total number of sensors, and the mechanical energy of each sensor at each moment in the i-th time window is obtained; based on the mechanical energy of each sensor at each moment in the i-th time window and the position of each sensor, a mechanical energy field function in the i-th time window is constructed.

[0115] Regarding the number of modal components decomposed by the vibration signal of each sensor in the i-th time window, in some embodiments, the present application preferably decomposes the vibration signal of each sensor in the i-th time window into three modal components.

[0116] The method for determining the maximum singular value of each modal component of each sensor in the i-th time window is similar to the method for determining the maximum singular value of each modal component of each sensor described above. Reference may be made to the above-mentioned embodiment for determining the maximum singular value of each modal component of each sensor, and details thereof will not be repeated here.

[0117] In some embodiments, the average energy proportion of each sensor in the i-th time window can be determined using the formula Determine the average energy proportion of each sensor in the i-th time window; where P n,iis the average energy proportion of the nth sensor in the i-th time window, M is the total number of modal components, P n,i,m is the energy proportion of the mth modal component of the nth sensor in the i-th time window.

[0118] Regarding the method for determining the global maximum singular value of each sensor in the i-th time window, in some embodiments, for the global maximum singular value of the n-th sensor in the i-th time window, the largest maximum singular value among the maximum singular values ​​of each modal component of the n-th sensor in the i-th time window can be used as the global maximum singular value of the n-th sensor in the i-th time window.

[0119] Regarding the method for determining each time domain correlation coefficient of the nth sensor at the tth moment in the i-th time window, in some embodiments, for the kth time domain correlation coefficient of the nth sensor at the tth moment in the i-th time window, the product of the data point at the tth moment in the vibration signal of the nth sensor in the i-th time window and the data point at the tth moment in the vibration signal of the kth sensor in the i-th time window other than the nth sensor can be used as the kth time domain correlation coefficient of the nth sensor in the i-th time window; of course, in other embodiments, the formula can be used. Determine the time domain correlation coefficients of the nth sensor at the tth moment in the i-th time window; where ρ n,i,k,t is the kth time domain correlation coefficient of the nth sensor at the tth moment in the i-th time window, x n,i,t is the data point at time t in the vibration signal of the nth sensor in the i-th time window, is the average value of the vibration signal of the nth sensor in the i-th time window between the data point at time t-1, the data point at time t, and the data point at time t+1, x k,i,t is the data point at time t in the vibration signal of the kth sensor except the nth sensor in the i-th time window, is the average value of the data point at time t-1, the data point at time t, and the data point at time t+1 in the vibration signal of the kth sensor excluding the nth sensor in the i-th time window.

[0120] Regarding the method for determining the weighted comprehensive index at the t-th moment in the i-th time window, in some embodiments, the formula Determine the weighted comprehensive index at the tth moment in the i-th time window; where d i,t is the weighted comprehensive index at the tth moment in the i-th time window, N is the total number of sensors, K is N-1, ρ n,i,k,t is the kth time domain correlation coefficient of the nth sensor at the tth moment in the i-th time window.

[0121] Regarding the method of constructing the mechanical energy field function in the i-th time window, in some embodiments, a function with independent variables including x, y, z and t can be fitted according to the mechanical energy of each sensor at each moment in the i-th time window and the position of each sensor to construct the mechanical energy field function in the i-th time window; it can be understood that the position of each sensor includes x, y and z, and the mechanical energy of each sensor at each moment in the i-th time window includes the function value and t. Therefore, a function with independent variables including x, y, z and t can be fitted, that is, the fitting coefficient is first solved, and then the mechanical energy field function in the i-th time window is constructed.

[0122] In an embodiment of the present application, by comprehensively considering the modal component energy, singular values ​​and time domain correlation of the multi-sensor vibration signals, a mechanical energy field function is constructed, which can more accurately reflect the operating status of the power transformer and provide a more effective optimization basis for the Grey Wolf optimization algorithm, thereby improving the model parameter optimization effect and fault identification capability.

[0123] It can be understood that the vibration signal information is fully utilized: the empirical wavelet transform (EWT) is used to decompose the vibration signal of each sensor in the i-th time window to obtain various modal components, and the energy proportion and maximum singular value of each modal component are determined. This method can fully utilize the information of the vibration signal in different frequency bands, more accurately characterize the signal characteristics, and provide a rich data basis for the subsequent construction of the mechanical energy field function; consider the sensor position and time domain correlation: by determining the average energy proportion and global maximum singular value of each sensor in the i-th time window, and calculating the time domain correlation coefficient of different sensors at the same time, the sensor position information and the time domain change of the vibration signal are comprehensively considered. This method can more comprehensively reflect the operating status of the power transformer at different positions and different times, and improve the accuracy of the mechanical energy field function; accurately determine the mechanical energy: according to the weighted comprehensive index at each moment in the i-th time window, and the average energy proportion and global maximum singular value of each sensor in the i-th time window, determine the energy proportion of each sensor in the i-th time window. The mechanical energy at each moment in the i-th time window can be more accurately quantified by this method, which provides key data support for constructing the mechanical energy field function; constructing a more effective mechanical energy field function: based on the mechanical energy of each sensor at each moment in the i-th time window and the position of each sensor, a mechanical energy field function in the i-th time window is constructed. This method can more accurately reflect the mechanical energy distribution of the power transformer at different positions and at different times, provide a more effective optimization basis for the Grey Wolf optimization algorithm, help to find a better model parameter combination, and improve the effect of model parameter optimization; improve model performance and fault identification ability: since the mechanical energy field function more accurately reflects the operating status of the power transformer, the model parameter optimization of the Grey Wolf optimization algorithm based on this function can obtain model parameters that are more in line with the actual data characteristics. Therefore, it can enhance the generalization ability and robustness of the model, improve the accuracy and reliability of fault identification, and provide a stronger guarantee for the safe and stable operation of the power transformer.

[0124] In a feasible implementation, the above embodiment determines the mechanical energy of the nth sensor at each moment in the i-th time window based on the weighted comprehensive index at each moment in the i-th time window, and the average energy proportion and global maximum singular value of the nth sensor in the i-th time window, including:

[0125] Using the formula Determine the mechanical energy of the nth sensor at each moment in the i-th time window;

[0126] Among them, E n,i,t is the mechanical energy of the nth sensor at the tth moment in the i-th time window, k1 is the first preset empirical constant, P n,iis the weighted average energy of the nth sensor in the i-th time window, λ max,n,i is the global maximum singular value of the nth sensor in the i-th time window, e is a natural constant, k2 is the second preset empirical constant, d i,t is the weighted comprehensive index at the tth moment in the i-th time window.

[0127] The first preset empirical constant and the second preset empirical constant can both be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics. Of course, they can also be pre-set by the operator based on actual needs.

[0128] In some embodiments, the present application preferably sets the first preset empirical constant to 0.87 and the second preset empirical constant to 0.95.

[0129] In the embodiment of the present application, by introducing the average energy ratio, the global maximum singular value and the weighted comprehensive index, and using the preset empirical constants to calculate the calculation formula of mechanical energy, the mechanical energy of each sensor at different times can be quantified more accurately, providing key data support for constructing a more effective mechanical energy field function, thereby improving the model parameter optimization effect and fault identification ability.

[0130] It can be understood that multi-dimensional information is comprehensively considered: the formula comprehensively considers the average energy proportion of the nth sensor in the ith time window, the global maximum singular value and the weighted comprehensive index at the tth moment in the ith time window. These indicators respectively reflect the energy distribution, singular value characteristics and time domain correlation of the sensor vibration signal, so that the calculated mechanical energy more comprehensively and accurately characterizes the operating status of the power transformer; introducing preset empirical constants: by introducing the first preset empirical constant and the second preset empirical constant, the calculation formula is optimized. These empirical constants can be adjusted according to actual operating conditions and experimental data, so that the calculation of mechanical energy is more in line with actual conditions, and the accuracy and reliability of the calculation are improved; accurately quantifying mechanical energy: using the above formula, the mechanical energy of each sensor at different times can be accurately quantified. Mechanical energy. This precise quantification helps to more accurately reflect the mechanical energy distribution of the power transformer at different positions and at different times, and provides key data support for the subsequent construction of the mechanical energy field function; improves the optimization effect of model parameters: Since the calculation of mechanical energy is more accurate, the mechanical energy field function constructed based on this energy is also more effective, which provides a more reliable optimization basis for the Grey Wolf optimization algorithm, helps to find a better combination of model parameters, thereby improving the effect of model parameter optimization; enhances fault identification capabilities: The optimized model parameters can better capture the characteristic information in the data, improve the accuracy and reliability of fault identification, and by more accurately quantifying mechanical energy, the model can more sensitively perceive changes in the operating status of the power transformer, promptly discover potential fault hazards, and provide stronger guarantees for the safe and stable operation of the power transformer.

[0131] In a feasible implementation, the above embodiment determines the position of the β gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the β gray wolf in the gray wolf population at the i-th iteration, and determines the position of the δ gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the δ gray wolf in the gray wolf population at the i-th iteration, including:

[0132] Using the formula Determine the position of the gray wolf β in the gray wolf population at the i+1th iteration, and the position of the gray wolf δ at the i+1th iteration;

[0133] Among them, when r = β, X β (i+1) is the position of β gray wolf in the gray wolf population at the i+1th iteration, X β (i) is the position of β gray wolf in the gray wolf population at the i-th iteration, is the gradient of the gray wolf population β in the i-th iteration. When r = δ, X δ (i+1) is the position of the gray wolf δ in the gray wolf population at the i+1th iteration, X δ(i) is the position of δ gray wolves in the gray wolf population at the i-th iteration, is the gradient of δ gray wolf in the gray wolf population at the i-th iteration, A i is the random attenuation value corresponding to the β-gray wolf or δ-gray wolf in the gray wolf population at the i-th iteration, and Δi is the preset time step.

[0134] Regarding the method for determining the random attenuation value corresponding to the β-gray wolf or δ-gray wolf in the gray wolf population at the i-th iteration, in some embodiments, the formula Determine the random attenuation value corresponding to the β-gray wolf or δ-gray wolf in the gray wolf population at the i-th iteration; where A i is the random attenuation value corresponding to the β-gray wolf or δ-gray wolf in the gray wolf population at the i-th iteration, r1 is the first random number in [0,1], and I is the preset maximum number of iterations.

[0135] Regarding the specific value of the preset time step, in some embodiments, the present application preferably sets the preset time step to 0.1.

[0136] In the embodiment of the present application, by introducing a random attenuation value and a preset time step, the iterative positions of the β-gray wolf and the δ-gray wolf in the gray wolf population can be adjusted more flexibly, thereby enhancing the global search capability and convergence speed of the algorithm, thereby improving the effect of model parameter optimization and fault identification capability.

[0137] It can be understood that the introduction of random attenuation value: the formula introduces the random attenuation value corresponding to the β gray wolf or δ gray wolf in the gray wolf population in the i-th iteration, which is calculated by random numbers and the preset maximum number of iterations. The introduction of random attenuation value makes the position update of the gray wolf in the iteration process more flexible, which helps the algorithm to jump out of the local optimum and enhance the global search ability; preset time step: by setting the preset time step, the moving step of the gray wolf in each iteration can be controlled, making the algorithm more stable during the search process. The reasonable setting of the preset time step helps to balance the exploration and development capabilities of the algorithm and improve the convergence speed; flexibly adjust the iteration position: using the above formula, the gradient and position of the β gray wolf and δ gray wolf in the gray wolf population in the i-th iteration can be adjusted. , and flexibly determine their positions in the i+1th iteration. This flexible position adjustment method helps the algorithm better adapt to different optimization problems and improve the optimization effect; improve the model parameter optimization effect: because the iterative positions of β gray wolves and δ gray wolves in the gray wolf population are more flexible and accurate, when optimizing the model parameters based on these gray wolf positions, it is possible to more effectively find the optimal model parameter combination and improve the effect of model parameter optimization; enhance fault identification ability: the optimized model parameters can better capture the characteristic information in the data and improve the accuracy and reliability of fault identification. By more flexibly adjusting the iterative position of the gray wolf, the algorithm can more comprehensively search the solution space and find better model parameters, thereby enhancing the ability to identify mechanical faults of power transformers.

[0138] In a feasible implementation, the above embodiment determines the position of the ω gray wolf in the gray wolf population at the i+1 iteration based on the vibration signal of each sensor in the i-time window and the position of the ω gray wolf in the gray wolf population at the i-iteration, including: reconstructing the component time domain signal of each sensor in the i-time window based on the second modal component and the third modal component of each sensor in the i-time window; performing feature extraction on the component time domain signal of each sensor in the i-time window to obtain the kurtosis of each sensor in the i-time window; determining the weighted average kurtosis in the i-time window based on the kurtosis of each sensor in the i-time window; and determining the position of the ω gray wolf in the gray wolf population at the i+1 iteration based on the weighted average kurtosis in the i-time window and the position of the ω gray wolf in the gray wolf population at the i-iteration.

[0139] Regarding the determination of the weighted average kurtosis in the i-th time window, in some embodiments, the formula Determine the weighted average kurtosis in the i-th time window; where S i is the weighted average kurtosis of the nth sensor in the i-th time window, N is the total number of sensors, w n is the weight value of the nth sensor (related to the position of the nth sensor, ranging from 0.65 to 0.85), S n,iis the kurtosis of the nth sensor in the i-th time window.

[0140] In the embodiment of the present application, by comprehensively considering the kurtosis characteristics of the multi-sensor vibration signals and calculating the weighted average kurtosis, it is possible to more accurately guide the iterative position update of the ω gray wolf in the gray wolf population, enhance the global search capability and convergence of the algorithm, and thus improve the effect of model parameter optimization and fault identification capability.

[0141] It can be understood that the integrated multi-sensor information: by reconstructing the component time domain signal according to the second modal component and the third modal component of each sensor in the i-th time window, and extracting the kurtosis feature, the vibration signal information provided by the multi-sensor is fully utilized. This method of integrating multi-sensor information can more comprehensively reflect the operating status of the power transformer and provide richer data support for the subsequent determination of the iterative position of the ω gray wolf; calculate the weighted average kurtosis: use the formula to calculate the weighted average kurtosis in the i-th time window, in which the weight value of each sensor (related to the sensor position) is taken into account. This weighted average method can more reasonably reflect the contribution of different sensors to the kurtosis feature, making the calculated weighted average kurtosis more accurate and reliable; accurately guide the iteration of the ω gray wolf: according to the weighted average kurtosis in the i-th time window and the ω gray wolf in the gray wolf population in the i-th iteration Position, determine the position of the ω gray wolf in the i+1th iteration. This method can more accurately guide the iterative process of the ω gray wolf, make it better adapt to the actual data characteristics, help the algorithm to jump out of the local optimum, and enhance the global search ability; improve the model parameter optimization effect: because the iterative position of the ω gray wolf is more accurate and flexible, when optimizing the model parameters based on these gray wolf positions, it is possible to more effectively find the optimal model parameter combination, which helps to improve the effect of model parameter optimization and make the optimized model have better generalization ability and robustness; enhance fault identification ability: the optimized model parameters can better capture the characteristic information in the data, improve the accuracy and reliability of fault identification, and by more accurately guiding the iterative position of the ω gray wolf, the algorithm can more comprehensively search the solution space and find better model parameters, thereby enhancing the ability to identify mechanical faults of power transformers.

[0142] In a feasible implementation, the above embodiment determines the position of the ω gray wolf in the gray wolf population at the i+1th iteration based on the weighted average kurtosis in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration, including:

[0143] Using formula X ω (i+1)=X ω (i)+C i ·S i rand() determines the position of the ω gray wolf in the gray wolf population at the i+1th iteration;

[0144] Among them, X ω (i+1) is the position of the ω gray wolf in the gray wolf population at the i+1th iteration, X ω (i) is the position of the ω gray wolf in the gray wolf population at the i-th iteration, C i is the random disturbance value corresponding to the ω gray wolf in the gray wolf population at the i-th iteration, S i is the weighted average kurtosis in the i-th time window, and rand() is the random number generation function.

[0145] Regarding the method for determining the random disturbance value corresponding to the ω gray wolf in the gray wolf population in the i-th iteration, in some embodiments, the random disturbance value corresponding to the ω gray wolf in the gray wolf population in the i-th iteration is equal to 2r2, where r2 is a second random number in [0,1].

[0146] In the embodiment of the present application, by introducing random disturbance values ​​and weighted average kurtosis, the iterative position of ω gray wolf in the gray wolf population can be adjusted more flexibly, the global search ability and convergence of the algorithm can be enhanced, and falling into local optimality can be avoided, thereby improving the effect of model parameter optimization and fault identification ability.

[0147] It can be understood that the introduction of random disturbance value: the formula introduces the random disturbance value corresponding to the ω gray wolf in the gray wolf population in the i-th iteration, which is generated by random numbers. The introduction of random disturbance value makes the position update of ω gray wolf in the iteration process more flexible, which helps the algorithm to jump out of the local optimal solution, enhance the global search ability, and explore a broader solution space; consider weighted average kurtosis: the formula takes the weighted average kurtosis of the i-th time window as an important factor affecting the iterative position of ω gray wolf. The weighted average kurtosis integrates the kurtosis characteristics of the multi-sensor vibration signal and takes into account the weight value of each sensor. It can more accurately reflect the operating status of the power transformer. By incorporating it into the iterative position update formula of ω gray wolf, the iterative process of ω gray wolf is more in line with the actual data characteristics; flexibly adjust the iterative position: using the above formula, the position of ω gray wolf in the gray wolf population in the i-th iteration, the random disturbance value and the weighted Average kurtosis, flexibly determine the position of the ω gray wolf in the i+1th iteration. This flexible position adjustment method helps the algorithm better adapt to different optimization problems and improve the optimization effect; improve the model parameter optimization effect: because the iterative position of the ω gray wolf is more flexible and accurate, when optimizing the model parameters based on these gray wolf positions, it is possible to more effectively find the optimal model parameter combination, avoid falling into local optimality, improve the effect of model parameter optimization, and make the optimized model have better generalization ability and robustness; enhance fault identification ability: the optimized model parameters can better capture the characteristic information in the data, improve the accuracy and reliability of fault identification, and by more flexibly adjusting the iterative position of the ω gray wolf, the algorithm can more comprehensively search the solution space and find better model parameters, thereby enhancing the ability to identify mechanical faults of power transformers, timely discover potential fault hazards, and ensure the safe and stable operation of the power system.

[0148] In a feasible implementation, the above embodiment determines the population position of the gray wolf population at the i+1 iteration based on the position of the α gray wolf in the gray wolf population at the i+1 iteration, the position of the β gray wolf in the i+1 iteration, the position of the δ gray wolf in the i+1 iteration, and the position of the ω gray wolf in the i+1 iteration, including: determining the mechanical energy of the α gray wolf in the gray wolf population at the i+1 iteration, the mechanical energy of the β gray wolf in the i+1 iteration, the mechanical energy of the δ gray wolf in the i+1 iteration, and the mechanical energy field function in the i-th time window. the mechanical energy of the position of the α gray wolf in the gray wolf population at the i+1 iteration, the mechanical energy of the β gray wolf at the i+1 iteration, the mechanical energy of the δ gray wolf at the i+1 iteration and the mechanical energy of the ω gray wolf at the i+1 iteration, determine the weight value of the α gray wolf in the gray wolf population at the i+1 iteration, the weight value of the β gray wolf at the i+1 iteration, the weight value of the δ gray wolf at the i+1 iteration and the weight value of the ω gray wolf at the i+1 iteration; determine the population position of the gray wolf population at the i+1 iteration according to the weight value and position of the α gray wolf in the gray wolf population at the i+1 iteration, the weight value and position of the β gray wolf at the i+1 iteration, the weight value and position of the δ gray wolf at the i+1 iteration, and the weight value and position of the ω gray wolf at the i+1 iteration.

[0149] Regarding the method for determining the mechanical energy of the α gray wolf at the position of the i+1th iteration, the mechanical energy of the β gray wolf at the position of the i+1th iteration, the mechanical energy of the δ gray wolf at the position of the i+1th iteration, and the mechanical energy of the ω gray wolf at the position of the i+1th iteration in the gray wolf population, in some embodiments, the position of the α gray wolf at the i+1th iteration, the position of the β gray wolf at the i+1th iteration, the position of the δ gray wolf at the i+1th iteration, and the position of the ω gray wolf at the i+1th iteration in the gray wolf population are respectively substituted into the mechanical energy field function of the i-th time window to obtain the gray wolf respectively. The mechanical energy of the α gray wolf in the population at the position of the i+1th iteration, the mechanical energy of the β gray wolf at the position of the i+1th iteration, the mechanical energy of the δ gray wolf at the position of the i+1th iteration, and the mechanical energy of the ω gray wolf at the position of the i+1th iteration; among them, the independent variable time t substituted into the mechanical energy field function in the i-th time window can be any time in the i-th time window, or each time in the i-th time window can be substituted into the calculation, and then the mechanical energy at each time is obtained. Among the mechanical energies at each time, the largest mechanical energy is selected as the final mechanical energy.

[0150] In some embodiments, the weight value of the gray wolf α in the gray wolf population at the i+1th iteration, the weight value of the gray wolf β at the i+1th iteration, the weight value of the gray wolf δ at the i+1th iteration, and the weight value of the gray wolf ω at the i+1th iteration can be determined using the formula: Determine the weight value of the α gray wolf in the gray wolf population at the i+1th iteration, the weight value of the β gray wolf at the i+1th iteration, the weight value of the δ gray wolf at the i+1th iteration, and the weight value of the ω gray wolf at the i+1th iteration; wherein, when r = α, w a,i+1 is the weight value of α gray wolf in the gray wolf population at the i+1th iteration, E a,i+1 is the mechanical energy of the position of the α gray wolf in the gray wolf population at the i+1th iteration. When r = β, w β,i+1 is the weight value of the β gray wolf in the gray wolf population at the i+1th iteration, E β,i+1 is the mechanical energy of the position of β gray wolf in the gray wolf population at the i+1th iteration. When r = δ, w δ,i+1 is the weight value of the gray wolf δ in the gray wolf population at the i+1th iteration, E δ,i+1 is the mechanical energy of the position of the gray wolf δ in the gray wolf population at the i+1th iteration. When r = ω, w ω,i+1 is the weight value of the ω gray wolf in the gray wolf population at the i+1th iteration, E ω,i+1 is the mechanical energy of the position of the ω gray wolf in the gray wolf population at the i+1th iteration.

[0151] Regarding the method for determining the population location of the gray wolf population at the i+1th iteration, in some embodiments, the formula X(i+1)=w a,i+1 X a (i+1)+w β,i+1 X aβ (i+1)+w δ,i+1 X δ (i+1)+w ω,i+1 X ω (i+1) determines the population position of the gray wolf population at the i+1th iteration; where X(i+1) is the population position of the gray wolf population at the i+1th iteration, w a,i+1 is the weight value of α gray wolf in the gray wolf population at the i+1th iteration, X a (i+1) is the position of the α gray wolf in the gray wolf population at the i+1th iteration, w β,i+1 is the weight value of the β gray wolf in the gray wolf population at the i+1th iteration, X β (i+1) is the position of β gray wolf in the gray wolf population at the i+1th iteration, w δ,i+1 is the weight value of the gray wolf δ in the gray wolf population at the i+1th iteration, X δ(i+1) is the position of the gray wolf δ in the gray wolf population at the i+1th iteration, w ω,i+1 is the weight value of the ω gray wolf in the gray wolf population at the i+1th iteration, X ω (i+1) is the position of the ω gray wolf in the gray wolf population at the i+1th iteration.

[0152] In an embodiment of the present application, by combining the mechanical energy field function with the position and mechanical energy of each gray wolf in the gray wolf population, the weight value of each gray wolf is dynamically determined, and then the population position of the gray wolf population in the i+1th iteration is determined, which enhances the global search capability and convergence of the algorithm, helps to find better model parameters, and improves model performance and fault identification capabilities.

[0153] It can be understood that, combined with the mechanical energy field function: by substituting the positions of α, β, δ and ω wolves in the gray wolf population at the i+1 iteration into the mechanical energy field function of the i-th time window, the mechanical energy of their positions at the i+1 iteration is obtained respectively. This method makes full use of the mechanical energy distribution of the power transformer at different positions and at different times reflected by the mechanical energy field function, making the mechanical energy calculation of the gray wolf position more accurate and reliable; dynamically determine the weight value: according to the mechanical energy of the position of each gray wolf in the gray wolf population at the i+1 iteration, dynamically determine their weight value at the i+1 iteration. The gray wolf with greater mechanical energy has a greater weight value and a greater influence in determining the population position. This method can more reasonably reflect the importance of each gray wolf in the iteration process, making the algorithm more scientific and effective in the search process; accurately determine the population position: use the weight value and position of each gray wolf at the i+1 iteration to accurately determine the gray wolf population. In the population position of the i+1th iteration, this method can comprehensively consider the information of each gray wolf, making the determination of the population position more accurate and reasonable, helping the algorithm to better adapt to the actual data characteristics and enhance the global search capability; improving the model parameter optimization effect: because the population position of the gray wolf population in the i+1th iteration is more accurate and reasonable, when optimizing the model parameters based on this population position, it is possible to more effectively find the optimal model parameter combination, avoid falling into local optimality, improve the effect of model parameter optimization, and make the optimized model have better generalization ability and robustness; enhance fault identification ability: the optimized model parameters can better capture the characteristic information in the data, improve the accuracy and reliability of fault identification, and by more accurately determining the population position of the gray wolf population, the algorithm can more comprehensively search the solution space and find better model parameters, thereby enhancing the ability to identify mechanical faults of power transformers, timely discover potential fault hazards, and ensure the safe and stable operation of the power system.

[0154] In a second aspect, the present application provides a power transformer fault identification device.

[0155] See also Figure 2 , is a schematic diagram of a power transformer fault identification device according to an embodiment of the present application, wherein the device 210 includes:

[0156] An acquisition module 211 is configured to acquire vibration signals from various sensors, wherein the various sensors are installed at different locations on the power transformer;

[0157] The feature module 212 is used to perform multi-feature extraction on the vibration signal of each sensor to obtain a signal feature vector, and to sequentially splice the vibration signals of each sensor to obtain a high-dimensional feature vector, and to perform dimensionality reduction processing on the high-dimensional feature vector to obtain a target feature vector;

[0158] The model prediction module 213 is used to input the target feature vector and the signal feature vector into the transformer mechanical fault identification model to obtain the mechanical fault identification result of the power transformer; wherein the transformer mechanical fault identification model is a preset least squares support vector machine model based on the gray wolf optimization algorithm driven by the mechanical energy field.

[0159] In the embodiment of the present application, the relevant contents of the acquisition module 211, the feature module 212 and the model prediction module 213 can be found in Figure 1 The contents of the illustrated embodiments are not described in detail here.

[0160] It should be noted that the device 210 of the present application also includes some other modules. It can be understood that the method of the present application and the device 210 have a one-to-one correspondence. Therefore, the other modules of the device 210 of the present application are the contents corresponding to the method of the present application in the above-mentioned embodiment.

[0161] In an embodiment of the present application, by adopting a preset least squares support vector machine model based on the gray wolf optimization algorithm driven by mechanical energy field as a transformer mechanical fault identification model, it has better performance in small sample scenarios, avoids the problem of deep learning being prone to overfitting in small sample scenarios, and improves the accuracy and reliability of fault identification. At the same time, by comprehensively considering the data of multiple sensors, through multi-feature extraction, splicing and dimensionality reduction processing, the target feature vector and signal feature vector are obtained, which can more comprehensively and accurately characterize the state of the transformer, overcome the defect of incomplete single feature representation, and provide a strong guarantee for the safe and stable operation of the power transformer.

[0162] In addition, in addition to the above-mentioned advantages of avoiding the overfitting of deep learning in small sample scenarios, improving the accuracy and reliability of fault identification, and overcoming the defects of incomplete single feature representation, the mechanical fault identification device of the power transformer also has the following advantages: Multi-dimensional information integration: Vibration signals are obtained by multiple sensors installed at different positions of the power transformer, and these signals are comprehensively considered. It is possible to capture the operating status information of the transformer from multiple angles and dimensions. Sensors at different positions may be more sensitive to different types of faults or abnormal changes. The fusion of multi-sensor data can make up for the incompleteness and limitations of single sensor information, thereby more comprehensively reflecting the actual status of the transformer; Improve fault detection sensitivity: Multi-sensor data fusion has It helps to discover some subtle anomalies that are difficult to detect with a single sensor. For example, some faults may cause weak vibration changes in local areas, which may not be accurately captured by a single sensor. The collaborative work of multiple sensors can enhance the perception of these subtle changes and improve the sensitivity of fault detection; comprehensively characterize fault characteristics: adopt a variety of feature extraction devices, such as empirical wavelet transform (EWT) to extract the maximum singular value, energy proportion, center frequency, kurtosis and waveform factor of the modal component, which can describe the characteristics of the vibration signal from different angles. These features cover multiple aspects such as the time domain and frequency domain of the signal, and can more comprehensively characterize the fault characteristics of the transformer, which helps to improve the accuracy of fault identification; enhance feature robustness: by splicing The vibration signals of multiple sensors are processed and dimensionally reduced to obtain the target feature vector, and the signal feature vector of each sensor is combined to form a comprehensive feature set. This multi-feature fusion method can enhance the robustness of the feature to noise and interference, and reduce the misjudgment caused by the influence of noise on a single feature; Optimize model parameters: Use the mechanical energy field driven gray wolf optimization algorithm (MD-GWO) to optimize the model parameters of the least squares support vector machine model (LSSVM). The gray wolf optimization algorithm is an optimization algorithm based on swarm intelligence with good global search capability and convergence speed. The mechanical energy field drive further enhances the optimization performance of the algorithm, and can more effectively find the optimal combination of model parameters, thereby improving The model's generalization ability and recognition accuracy; Adaptability to complex fault modes: The internal mechanical structure of power transformers is complex and the fault modes are diverse. The least squares support vector machine model based on the Gray Wolf optimization algorithm driven by the mechanical energy field can better adapt to such complex fault modes. By optimizing the model parameters, the model can more accurately identify different types of faults, including normal state, loose insulation plate state, loose moving contact state, and other states; Prevention of equipment failure: Accurate mechanical fault identification can promptly discover potential fault hazards of transformers, take maintenance and repair measures in advance, avoid further development and expansion of faults, thereby preventing the occurrence of equipment failures, reducing the risk of power outages and equipment damage, and improving the operational safety of the power system;Optimize maintenance strategies: Traditional fixed-cycle maintenance strategies lack specificity and may lead to over- or under-maintenance. This mechanical fault identification device can accurately sense the actual status of the transformer in real time, providing a basis for formulating reasonable maintenance strategies. Based on the fault identification results, targeted maintenance can be carried out, improving maintenance efficiency and reducing maintenance costs, while ensuring reliable equipment operation and enhancing power system stability. Reduce resource waste: This avoids the over-maintenance that may be caused by traditional fixed-cycle maintenance strategies, reduces unnecessary manpower, material resources, and time investment, and reduces operation and maintenance costs. Extend equipment life: By promptly detecting and handling faults, damage to equipment caused by faults can be reduced, extending equipment life and further reducing equipment replacement costs.

[0163] In a third aspect, the present application further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes a power transformer fault identification method in the above method embodiment.

[0164] In a fourth aspect, the present application further provides a computer device including a memory and a processor, wherein the memory stores a computer program. When the computer program is executed by the processor, the processor executes a power transformer fault identification method in the above method embodiment.

[0165] Figure 3 The internal structure diagram of the computer device in some embodiments is shown. The computer device can be a terminal, a server, or a gateway. Figure 3 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus.

[0166] 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 various steps in the above method embodiment. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement the various steps in the above method embodiment. It will be understood by those skilled in the art 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 shown in the figure, or combine certain components, or have a different component arrangement.

[0167] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0168] Among them, any reference to memory, storage, database or other media used in the various embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchl ink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0169] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0170] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for identifying a power transformer fault, characterized in that: The method comprises: Acquire vibration signals from various sensors, wherein the various sensors are installed at different locations on the power transformer; Performing multi-feature extraction on the vibration signal of each sensor to obtain a signal feature vector, and sequentially splicing the vibration signals of each sensor to obtain a high-dimensional feature vector, and performing dimensionality reduction processing on the high-dimensional feature vector to obtain a target feature vector; The target feature vector and the signal feature vector are input into a transformer mechanical fault identification model to obtain a mechanical fault identification result of the power transformer; wherein, the transformer mechanical fault identification model is a preset least squares support vector machine model based on a gray wolf optimization algorithm driven by a mechanical energy field.

2. The power transformer fault identification method according to claim 1, characterized in that: The method further comprises: Obtain historical vibration signals from each sensor; Performing multi-feature extraction on the historical vibration signals of each sensor to obtain a historical signal feature vector, sequentially concatenating the historical vibration signals of each sensor to obtain a historical high-dimensional feature vector, and performing dimensionality reduction processing on the historical high-dimensional feature vector to obtain a historical target feature vector; Determining a label corresponding to the historical target feature vector and a label corresponding to each signal feature in the historical signal feature vector; Using the Grey Wolf optimization algorithm driven by the mechanical energy field, the model parameters in the least squares support vector machine model are adjusted to obtain an initial model, wherein the initial model is an initial least squares support vector machine model based on the Grey Wolf optimization algorithm driven by the mechanical energy field; The historical target feature vector, the historical signal feature vector, the label corresponding to the historical target feature vector, and the label corresponding to each signal feature in the historical signal feature vector are input into the initial model for training to obtain the transformer mechanical fault recognition model.

3. The power transformer fault identification method according to claim 2, characterized in that: The gray wolf optimization algorithm based on mechanical energy field drive is used to adjust the model parameters in the least squares support vector machine model to obtain an initial model, including: Using the mechanical energy field-driven gray wolf optimization algorithm, the position of each gray wolf in the gray wolf population is updated, and the population position of the gray wolf population is updated according to the positions of all gray wolves in the gray wolf population, until the number of iterations corresponding to the current iteration is equal to the preset maximum number of iterations, and the optimal model parameters corresponding to the global optimal fitness value in all iterations are obtained, wherein the position of each gray wolf includes the model parameters; The model parameters in the least squares support vector machine model are adjusted according to the optimal model parameters corresponding to all iterations to obtain the initial model.

4. The power transformer fault identification method according to claim 3, characterized in that: The mechanical energy field-driven gray wolf optimization algorithm is used to update the position of each gray wolf in the gray wolf population, and the population position of the gray wolf population is updated according to the positions of all gray wolves in the gray wolf population, until the number of iterations corresponding to the current iteration is equal to the preset maximum number of iterations, and the optimal model parameters corresponding to the global optimal fitness value in all iterations are obtained, including: Dividing the historical vibration signal of each sensor into time windows to obtain the vibration signal of each sensor in each time window, wherein the total number of time window divisions is greater than or equal to the preset maximum number of iterations; According to the vibration signals of each sensor in the i-th time window, the mechanical energy field function in the i-th time window is constructed, and the initial value of i is 1; Determining the position of an alpha gray wolf in the gray wolf population at the i+1th iteration according to the mechanical energy field function in the i-th time window; Determining the gradient of the β-gray wolf in the gray wolf population at the i-th iteration based on the mechanical energy field function in the i-th time window and the position of the β-gray wolf in the gray wolf population at the i-th iteration, and determining the gradient of the δ-gray wolf in the gray wolf population at the i-th iteration based on the mechanical energy field function in the i-th time window and the position of the δ-gray wolf in the gray wolf population at the i-th iteration; Determine the position of the β-gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the β-gray wolf in the gray wolf population at the i-th iteration, and determine the position of the δ-gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the δ-gray wolf in the gray wolf population at the i-th iteration; Determine the position of the ω gray wolf in the gray wolf population at the i+1th iteration based on the vibration signals of each sensor in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration; Determine the population position of the gray wolf population at the i+1th iteration based on the position of the α gray wolf in the i+1th iteration, the position of the β gray wolf in the i+1th iteration, the position of the δ gray wolf in the i+1th iteration, and the position of the ω gray wolf in the i+1th iteration; The population position of the gray wolf population at the (i+1)th iteration is used as the position of the gray wolf newly added to the gray wolf population at the (i+1)th iteration; Let i=i+1 until i is equal to the preset maximum number of iterations, and use the position of the gray wolf corresponding to the global optimal fitness value in all iterations as the optimal model parameter corresponding to the global optimal fitness value in all iterations.

5. The power transformer fault identification method according to claim 4, characterized in that: The mechanical energy field function in the i-th time window is constructed according to the vibration signals of each sensor in the i-th time window, including: Using EWT, the vibration signal of each sensor in the i-th time window is decomposed to obtain the modal components of each sensor in the i-th time window; Determine the energy proportion and maximum singular value of each modal component of each sensor in the i-th time window; Determine the average energy proportion of each sensor in the i-th time window based on the energy proportion of each modal component of each sensor in the i-th time window, and determine the global maximum singular value of each sensor in the i-th time window based on the maximum singular value of each modal component of each sensor in the i-th time window; Determine the time domain correlation coefficient between the data point at the tth moment in the vibration signal of the nth sensor in the i-th time window and the data point at the tth moment in the vibration signal of each sensor except the nth sensor in the i-th time window; where n successively takes integers greater than 0 until n equals the total number of sensors, and obtain the respective time domain correlation coefficients of each sensor at the tth moment in the i-th time window, where the initial value of t is 1; Determine the weighted comprehensive index at the t-th moment in the i-th time window according to the time domain correlation coefficients of the sensors at the t-th moment in the i-th time window; Let t = t + 1, until t is equal to the total number of moments in the i-th time window, and obtain the weighted comprehensive index of each moment in the i-th time window; Determine the mechanical energy of the nth sensor at each moment in the i-th time window based on the weighted comprehensive index at each moment in the i-th time window, as well as the average energy proportion and global maximum singular value of the n-th sensor in the i-th time window; where n is an integer greater than 0 in sequence until n equals the total number of sensors, and obtain the mechanical energy of each sensor at each moment in the i-th time window; According to the mechanical energy of each sensor at each moment in the i-th time window and the position of each sensor, a mechanical energy field function in the i-th time window is constructed.

6. The power transformer fault identification method according to claim 5, characterized in that: Determining the mechanical energy of the nth sensor at each moment in the i-th time window based on the weighted comprehensive index at each moment in the i-th time window, and the average energy proportion and the global maximum singular value of the nth sensor in the i-th time window includes: Using the formula Determine the mechanical energy of the nth sensor at each moment in the i-th time window; Among them, E n,i,t is the mechanical energy of the nth sensor at the tth moment in the i-th time window, k1 is the first preset empirical constant, P n,i is the weighted average energy of the nth sensor in the i-th time window, λ max,n,i is the global maximum singular value of the nth sensor in the i-th time window, e is a natural constant, k2 is the second preset empirical constant, d i,t is the weighted comprehensive index at the tth moment in the i-th time window.

7. The power transformer fault identification method according to claim 4, characterized in that: The step of determining the position of the β gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the β gray wolf in the gray wolf population at the i-th iteration, and determining the position of the δ gray wolf in the gray wolf population at the i+1th iteration based on the gradient and position of the δ gray wolf in the gray wolf population at the i-th iteration, comprises: Using the formula Determine the position of the β-grey wolf in the gray wolf population at the i+1th iteration, and the position of the δ-grey wolf at the i+1th iteration; Among them, when r = β, X β (i+1) is the position of the β gray wolf in the gray wolf population at the i+1th iteration, X β (i) is the position of β gray wolf in the gray wolf population at the i-th iteration, is the gradient of the β gray wolf in the gray wolf population at the i-th iteration. When r = δ, X δ (i+1) is the position of the gray wolf delta in the gray wolf population at the i+1th iteration, X δ (i) is the position of δ gray wolves in the gray wolf population at the i-th iteration, is the gradient of δ gray wolves in the gray wolf population at the i-th iteration, A i is the random attenuation value corresponding to the β-gray wolf or δ-gray wolf in the gray wolf population at the i-th iteration, and Δi is the preset time step.

8. The power transformer fault identification method according to claim 4, characterized in that: The step of determining the position of the ω gray wolf in the gray wolf population at the i+1th iteration based on the vibration signals of the sensors in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration includes: Reconstruct the component time domain signal of each sensor in the i-th time window according to the second modal component and the third modal component of each sensor in the i-th time window; The component time domain signals of each sensor in the i-th time window are subjected to feature extraction to obtain the kurtosis of each sensor in the i-th time window; According to the kurtosis of each sensor in the i-th time window, the weighted average kurtosis in the i-th time window is determined; The position of the ω gray wolf in the gray wolf population at the i+1th iteration is determined according to the weighted average kurtosis in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration.

9. The power transformer fault identification method according to claim 8, characterized in that: The method of determining the position of the ω gray wolf in the gray wolf population at the i+1th iteration based on the weighted average kurtosis in the i-th time window and the position of the ω gray wolf in the gray wolf population at the i-th iteration includes: Using formula X ω (i+1)=X ω (i)+C i ·S i rand() determines the position of the ω gray wolf in the gray wolf population at the i+1th iteration; Among them, X ω (i+1) is the position of the ω gray wolf in the gray wolf population at the i+1th iteration, X ω (i) is the position of the ω gray wolf in the gray wolf population at the i-th iteration, C i is the random disturbance value corresponding to the ω gray wolf in the gray wolf population at the i-th iteration, S i is the weighted average kurtosis in the i-th time window, and rand() is the random number generation function.

10. The power transformer fault identification method according to claim 4, characterized in that: Determining the population position of the gray wolf population at the i+1th iteration based on the position of the α gray wolf in the i+1th iteration, the position of the β gray wolf in the i+1th iteration, the position of the δ gray wolf in the i+1th iteration, and the position of the ω gray wolf in the i+1th iteration includes: Determining the mechanical energy of the α gray wolf at the i+1th iteration, the mechanical energy of the β gray wolf at the i+1th iteration, the mechanical energy of the δ gray wolf at the i+1th iteration, and the mechanical energy of the ω gray wolf at the i+1th iteration in the gray wolf population according to the position of the α gray wolf at the i+1th iteration, the position of the β gray wolf at the i+1th iteration, the mechanical energy of the δ gray wolf at the i+1th iteration, and the mechanical energy of the ω gray wolf at the i+1th iteration in the gray wolf population; Determine a weight value of the α gray wolf at the i+1th iteration, a weight value of the β gray wolf at the i+1th iteration, a weight value of the δ gray wolf at the i+1th iteration, and a weight value of the ω gray wolf at the i+1th iteration in the gray wolf population according to the mechanical energy of the α gray wolf at the i+1th iteration, the mechanical energy of the β gray wolf at the i+1th iteration, the mechanical energy of the δ gray wolf at the i+1th iteration, and the mechanical energy of the ω gray wolf at the i+1th iteration; The population position of the gray wolf population at the i+1 iteration is determined based on the weight value and position of the α gray wolf in the gray wolf population at the i+1 iteration, the weight value and position of the β gray wolf in the i+1 iteration, the weight value and position of the δ gray wolf in the i+1 iteration, and the weight value and position of the ω gray wolf in the i+1 iteration.

Images