Dry-type bushing fault diagnosis method, system and device and storage medium

Abstract

The invention relates to the technical field of electrical equipment fault detection, in particular to a dry-type bushing fault diagnosis method, system and device and a storage medium, and the method comprises the steps: obtaining the feature data of a dry-type bushing, inputting the feature data into a pre-trained fault diagnosis model, and outputting a corresponding fault type and confidence thereof; according to the method, the pulse current waveform, the temperature distribution matrix, the stress value, the concentration of gas generated by decomposition of epoxy resin and other multi-dimensional operation data of the dry-type bushing are taken, the improved particle swarm optimization algorithm is adopted to carry out optimization selection on the data, and the improved particle swarm optimization algorithm combines the advantages of the particle swarm optimization algorithm and the genetic algorithm. The method has higher global search capability and convergence speed, can quickly find a feature subset most sensitive to the fault in a complex feature space, and reduces the risk of falling into a local optimal solution, thereby enhancing the robustness of a fault diagnosis model, and improving the fault diagnosis accuracy. The problem that existing dry-type bushing fault detection cannot accurately and comprehensively reflect the operation state of equipment is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical equipment fault detection, specifically to a dry-type bushing fault diagnosis method, system, device and storage medium, and in particular to a dry-type bushing fault diagnosis method, system, device and storage medium based on an improved particle swarm optimization algorithm. Background Art

[0002] Dry-type bushings are an insulating component widely used in high-voltage electrical equipment, such as transformers, reactors, and mutual inductors. They primarily connect the high-voltage conductors within the equipment to the external power system, providing both electrical insulation and mechanical support. Compared to traditional oil-immersed bushings, dry-type bushings offer advantages such as oil-free operation, fire and explosion resistance, and simplified maintenance. They are gradually replacing traditional oil-immersed bushings and becoming the mainstream choice for insulation in high-voltage electrical equipment. With continued technological advancements, dry-type bushings will play an even more important role in power systems, providing reliable assurance for the safe and stable operation of power equipment.

[0003] Dry-type bushings are subject to long-term influences such as electric fields, thermal stress, and mechanical vibration, making them prone to failures such as aging, cracking, and partial discharge. Therefore, fault detection of dry-type bushings is crucial to ensuring safe equipment operation. Currently, equipment fault diagnosis primarily involves partial discharge (PD) detection, oil chromatography analysis, and vibration monitoring. PD detection uses ultrasonic, ultra-high frequency, or high-frequency current sensors to capture discharge signals and then determine the discharge type using phase spectrum analysis. However, this method has low sensitivity to early, weak fault signals, making it difficult to detect potential faults in a timely manner. Furthermore, it is susceptible to external interference, resulting in reduced detection accuracy (for example, in environments with strong electromagnetic interference, PD detection has high false alarm and missed alarm rates). Oil chromatography analysis determines the fault type by detecting characteristic gases produced by the decomposition of insulating oil due to overheating or discharge within the equipment. This requires regular sampling, resulting in long cycles and poor test timeliness. It also lacks real-time monitoring and is difficult to detect sudden faults. Vibration monitoring determines the fault type by extracting the characteristic frequencies of mechanical components. This method only detects mechanical faults in mechanisms such as machinery and switches, is insensitive to electrical faults, is susceptible to environmental interference, and cannot fundamentally and accurately determine the fault type. Consequently, these methods rely on a single detection method, have certain limitations, and cannot fully reflect the operating status of the equipment. Summary of the Invention

[0004] In view of the problem that dry-type bushing fault detection in the prior art is greatly limited and cannot accurately and comprehensively reflect the operating status of the equipment, the present invention provides a dry-type bushing fault diagnosis method, system, equipment and storage medium.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The present invention provides a dry-type bushing fault diagnosis method, comprising:

[0007] Obtain characteristic data of dry casing;

[0008] Input the characteristic data of dry-type bushing into the pre-trained fault diagnosis model and output the corresponding fault type and its confidence level;

[0009] The characteristic data is obtained in the following manner:

[0010] Acquiring operating data of the dry bushing; wherein the operating data includes a pulse current waveform, a temperature distribution matrix, a stress value, and a gas concentration generated by decomposition of the epoxy resin of the dry bushing;

[0011] The improved particle swarm optimization algorithm is used to optimize the operation data and obtain the characteristic data;

[0012] The pre-trained fault diagnosis model is obtained by training a neural network model using feature data;

[0013] The improved particle swarm optimization algorithm is obtained by introducing a genetic algorithm into the particle swarm optimization algorithm, and the goal of the optimization selection is to screen out the optimal subset that is sensitive to faults.

[0014] Optionally, the operating data includes a pulse current waveform, a temperature distribution matrix, a stress value, and a gas concentration generated by decomposition of the dry bushing.

[0015] Optionally, the neural network model adopts a 1D-CNN and Transformer hybrid architecture as the model framework.

[0016] Optionally, the method of optimizing and selecting the operating data using the improved particle swarm optimization algorithm to obtain the characteristic data is:

[0017] De-noising the running data;

[0018] Normalize the denoised running data and build a feature pool;

[0019] The particle swarm optimization algorithm is used to optimize the features in the feature pool to obtain feature data.

[0020] Optionally, the operating data is denoised using a Daubechies-4 wavelet basis.

[0021] Optionally, Z-score standardization is used to normalize the denoised running data into a standard normal distribution.

[0022] Optionally, the method of using the particle swarm optimization algorithm to optimize and select features in the feature pool to obtain feature data is:

[0023] Introducing elite particles x generated by genetic algorithm GA ; Among them, elite particles are generated by simulating binary crossover and polynomial mutation;

[0024] Elite particle x GA Encode and initialize, and update and iterate;

[0025] After a preset number of iterations, the individuals with the highest fitness are selected from the elite particles generated by GA and injected into the PSO population;

[0026] Add a weighted fitness function and map the continuous speed value to a binary feature selection result:

[0027] When the preset termination condition is reached, the iteration is terminated and the feature data is obtained.

[0028] The present invention also provides a dry-type casing fault diagnosis system, comprising:

[0029] Characteristic data acquisition module: used to obtain characteristic data of dry casing;

[0030] Fault type and confidence output module: used to input the characteristic data of dry bushing into the pre-trained fault diagnosis model and output the corresponding fault type and confidence;

[0031] The characteristic data is obtained in the following manner:

[0032] Obtain operating data of dry casing;

[0033] The improved particle swarm optimization algorithm is used to optimize the operation data and obtain the characteristic data;

[0034] The pre-trained fault diagnosis model is obtained by training a neural network model using feature data;

[0035] The improved particle swarm optimization algorithm is obtained by introducing a genetic algorithm into the particle swarm optimization algorithm, and the goal of the optimization selection is to screen out the optimal subset that is sensitive to faults.

[0036] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0037] A computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the above method when executed by a processor.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention provides a dry-type bushing fault diagnosis method. The method obtains characteristic data of the dry-type bushing, inputs the characteristic data of the dry-type bushing into a pre-trained fault diagnosis model, and outputs a corresponding fault type and its confidence level. The method obtains multi-dimensional operating data of the dry-type bushing, such as a pulse current waveform, a temperature distribution matrix, a stress value, and a gas concentration generated by epoxy resin decomposition, and optimizes and selects these data using an improved particle swarm optimization algorithm. The improved particle swarm optimization algorithm combines the advantages of a particle swarm optimization algorithm (PSO) and a genetic algorithm (GA), has stronger global search capabilities and convergence speed, can quickly find the feature subset that is most sensitive to faults in a complex feature space, reduces the risk of falling into a local optimal solution, and thus enhances the robustness of the fault diagnosis model. This optimization selection process helps to remove redundant and noisy data, so that the fault diagnosis model is more focused on key features, thereby improving the accuracy of fault diagnosis. Using optimized feature data to train a neural network model enables the model to learn the complex mapping relationship between dry bushing faults and feature data. Furthermore, the neural network model possesses powerful nonlinear fitting capabilities, enabling it to capture fault modes difficult to identify using traditional methods. This enables real-time fault diagnosis of dry bushings, which is crucial for timely detection and resolution of dry bushing faults, helping to prevent fault escalation and more serious consequences, and further improving the accuracy of fault diagnosis. Furthermore, by comprehensively utilizing various dry bushing operational data, such as pulse current waveforms and temperature distribution matrices, a more comprehensive picture of the dry bushing's operational status can be generated, reducing the uncertainty associated with a single data source and improving the stability and reliability of fault diagnosis. By monitoring the dry bushing's operational status in real time and outputting fault diagnosis results and confidence levels, users can develop reasonable maintenance plans based on this information, achieving intelligent dry bushing maintenance.

[0040] The present invention also provides a dry bushing fault diagnosis system. This system, through the high integration of a feature data acquisition module and a fault type and confidence output module, acquires dry bushing feature data, inputs the dry bushing feature data into a pre-trained fault diagnosis model, and outputs the corresponding fault type and confidence. By improving the fault diagnosis of a particle swarm optimization algorithm and integrating it with a neural network model, the system leverages the global search capabilities of the particle swarm optimization algorithm and the local search advantages of the genetic algorithm to select an optimal subset of fault-sensitive features as feature data, providing a higher-quality data foundation for subsequent fault diagnosis. This significantly reduces the amount of data input to the fault diagnosis model, reduces model computational complexity, improves diagnostic speed, and avoids interference from redundant data on diagnostic results, making the diagnostic process more efficient. Furthermore, the system utilizes the powerful nonlinear mapping capabilities of the neural network to accurately capture the complex relationship between dry bushing feature data and fault type, enabling dry bushing fault diagnosis. The system can output not only the fault type but also the corresponding confidence level, providing more comprehensive diagnostic results for maintenance personnel. This allows them to make more reasonable maintenance decisions based on the confidence level, avoid blind repairs, and improve maintenance efficiency. The system has a simple structure, good reliability, high fault diagnosis accuracy and efficiency, and has certain versatility and scalability. It is expected to promote the progress and development of fault diagnosis technology in the entire industry.

[0041] The present invention also provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method when executing the computer program; the processor is capable of quickly executing the above-mentioned process of obtaining characteristic data of the dry bushing, inputting the characteristic data of the dry bushing into a pre-trained fault diagnosis model, and outputting the corresponding fault type and its confidence level, thereby ensuring the accuracy and efficiency of fault diagnosis; the computer program in the memory can be modified and optimized according to actual needs to adapt to the diagnostic requirements of different devices.

[0042] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned method; the computer-readable storage medium (such as a solid-state drive (SSD) and a Flash memory) has high-speed reading capabilities and can quickly load the computer program into the processor for execution, ensuring that the above-mentioned method is executed quickly and accurately. It has the characteristics of flexibility and portability, high reliability and stability, support for large-scale data storage, easy integration and expansion, reduced development and maintenance costs, high security, energy saving and environmental protection, support for multiple application scenarios, and promotion of standardization and normalization. It provides strong support for power equipment fault diagnosis and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The figure is a flow chart of a dry-type bushing fault diagnosis method of the present invention.

[0044] Figure 2 A framework diagram is established for a model of a dry-type bushing fault diagnosis method of the present invention.

[0045] Figure 3 The present invention is a flow chart of a method for obtaining characteristic data by optimizing and selecting operating data using an improved particle swarm optimization algorithm in a dry-type casing fault diagnosis method.

[0046] Figure 4 This is a structural diagram of the fault diagnosis model of the present invention.

[0047] Figure 5 This is a structural diagram of a dry-type casing fault diagnosis system of the present invention. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0049] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0050] The present invention will be further described in detail below with reference to specific embodiments, which are intended to explain the present invention rather than to limit it.

[0051] See also Figure 1 and Figure 2 The present invention discloses a dry-type bushing fault diagnosis method, comprising:

[0052] S1: Obtain the characteristic data of the dry casing, specifically:

[0053] S1.1: Acquire operating data of the dry bushing; wherein the operating data includes the pulse current waveform, temperature distribution matrix, stress value, and gas concentration generated by epoxy resin decomposition of the dry bushing, specifically:

[0054] A UHF sensor is embedded in the flange connection of the dry bushing and the conductor lead-out section (5 mm from the conductor surface) to collect the pulse current waveform with a sampling rate of 1 GHz and extract the pulse amplitude A. p , rise time t r (refers to the time required for the signal's stable value to rise from 10% to 90%) and the phase distribution entropy H Φ A temperature measuring point is arranged every 30 cm on the axial surface of the dry casing. The temperature measuring point covers the heat conduction path of the casing. The temperature distribution matrix is ​​collected using an infrared thermal imager to capture the longitudinal temperature gradient and calculate the axial temperature gradient. in, The system uses a threshold value >5°C / m to trigger an early warning. Fiber Bragg Grating (FBG) sensors are placed at the support root of the dry bushing to monitor the stress value of the dry bushing. Micro gas sensors are installed inside the bushing and near the operating location to monitor the concentration of gases (such as H2S and SO2) produced by the thermal decomposition of epoxy resin.

[0055] Through the acquisition of the aforementioned ultra-high frequency sensors, infrared thermal imagers, fiber Bragg grating sensors, and gas sensors, high-frequency discharge pulses caused by insulation defects are captured, abnormal temperature gradients caused by thermal stress are monitored, indicating poor heat dissipation or local overheating, quantifying mechanical stress distribution, identifying structural deformation, and detecting characteristic gases (such as SO2) produced by the decomposition of insulating materials to assist in determining the degree of aging. When training the fault diagnosis model, typical faults (such as partial discharge defects) can be artificially created to obtain fault data. This data, combined with normal data, can be divided into training, validation, and test sets.

[0056] Among them, the training set: 70% data (including normal and fault samples, mixed in the ratio of normal:fault=7:3;

[0057] Validation set: 15% data (independent of the training set, used for parameter adjustment and early stopping);

[0058] Test set: 15% of the data (for final evaluation of model performance).

[0059] S1.2: Use the improved particle swarm optimization algorithm to optimize the operating data and obtain feature data. The method is as follows:

[0060] Denoising the running data:

[0061] Since the original sensor signals often contain noise (power frequency interference, environmental vibration, electromagnetic noise), and different dimensional features need to be normalized to the same scale, the collected data needs to be preprocessed to extract effective features through denoising and normalization;

[0062] Because the tight support of the Daubechies-4 wavelet is suitable for capturing the transient characteristics of pulse current waveforms (partial discharge (PD) pulses) and avoiding signal distortion, the Daubechies-4 wavelet basis was chosen for denoising. This smoothes the waveform, reduces boundary effects during decomposition, and improves the signal-to-noise ratio (SNR). The number of decomposition layers, L, was set to 5. A 5-layer decomposition divides the 1 GHz signal into a frequency band of 31.25 MHz to 1 GHz, effectively separating the PD pulses (100 MHz–1 GHz) from the power-frequency noise (50 / 60 Hz).

[0063] By first performing a 5-layer Daubechies-4 wavelet decomposition on the PD pulse signal, the corresponding detail coefficients d1~d5 and the approximate coefficient a5 are obtained. The highest layer detail coefficient d1 (noise-dominated) is selected and the noise standard deviation σ is calculated:

[0064]

[0065] Wherein, MAD is the median absolute deviation of noise;

[0066] Based on the Donoho-Johnstone threshold theory, the global threshold is set:

[0067]

[0068] Wherein, N is the signal length, N=1024;

[0069] Then apply soft threshold processing to the detail coefficients d1 to d5 layer by layer:

[0070]

[0071] in, The denoising result of the detail coefficient of the kth layer after soft threshold processing, k = 1, 2, 3, 4, 5; the PD pulse signal with an amplitude exceeding the soft threshold is retained, and finally the PD pulse signal is reconstructed using the denoising coefficient, retaining the valid PD pulse signal and suppressing the noise to complete the denoising process.

[0072] Normalize the denoised running data and build a feature pool:

[0073] Z-score normalization is used to unify multi-dimensional features (such as PD pulse signal amplitude mV and temperature gradient °C / m) into a standard normal distribution:

[0074]

[0075] Among them, X is the original eigenvalue of the multi-dimensional feature; X norm is the standardized eigenvalue; μ is the mean, σ is the standard deviation;

[0076] In addition, the PD pulse signal and temperature signal must be aligned based on the power frequency phase (50 Hz) to ensure causality, and cubic spline interpolation is used to fit and fill in short-term missing data segments.

[0077] Since the original signal has high dimensions and contains redundant information, it is necessary to extract key features and optimize their combination to improve model efficiency;

[0078] Build a feature pool, including:

[0079] PD pulse characteristics: pulse amplitude A p and pulse density ρ PD , used to reflect the discharge intensity, which increases significantly when the insulation ages; the phase distribution entropy H Φ It is used to measure the randomness of discharge within the power frequency cycle. Partial discharge shows specific phase aggregation.

[0080] Temperature characteristics: axial temperature gradient Abnormal heat dissipation or internal failure causes the gradient to increase, and the hotspot offset rate v hotspot .

[0081] Gas characteristics: concentration ratio of H2S and SO2.

[0082] Strain characteristics: strain gradient, location of maximum strain concentration area.

[0083] See also Figure 3 , the particle swarm optimization algorithm is used to optimize the features in the feature pool and obtain the feature data:

[0084] Traditional PSO is prone to fall into local optimality, and the elite particles x generated by genetic algorithm are introduced GA ,PSO is responsible for rapid local development, GA provides global exploration, and the two work together to improve search efficiency;

[0085] Among them, first, for the elite particle x GA Encoding and initialization are performed, and the update iteration is performed. Each elite particle is a 42-dimensional binary vector, with 1 representing the selected feature and 0 representing the unselected feature; the population size is 50 particles, and the number of iterations is 100; initialization: 50 particles are randomly generated, and the initial probability p of each feature being selected isinit =0.3. The speed update formula is:

[0086]

[0087] in, is the updated velocity vector of particle i at iteration number t+1; is the current velocity vector of particle i at iteration number t; r1, r2 and r3 are random numbers uniformly distributed in the range [0,1], p best is the individual optimal solution, is the current position vector of particle i at iteration number t, g best is the global optimal solution; i is the i-th particle; t is the current number of iterations, w(t) is the inertia weight; T max is the maximum number of iterations of the algorithm;

[0088] Among them, the inertia weight w(t) decreases linearly from 0.9 to 0.4, that is, w max =0.9, w min =0.4, balance global search and local development, elite particle x GA , learning factors c1=1.5, c2=1.5, c3=0.5, r1, r2, r3~U(0,1).

[0089] Then, after a preset number of iterations, the individuals with the highest fitness are selected from the elite particles generated by the GA and injected into the PSO population. For example, every 10 iterations, the individuals with the highest fitness are selected from the elite particles generated by the GA and injected into the PSO population. The elite particles are generated through simulated binary crossover (SBX) and polynomial mutation to ensure diversity.

[0090] The simulated binary crossover is:

[0091]

[0092] The polynomial mutates to:

[0093] x mut =x+δ(x max -x min )

[0094] Where, δ=(2r)1 / (η+1)-1, r~U(0,1), mutation rate η=20; x new is the position of the newly generated particle, x i and x j Are both parent particle position vectors, x mut is the position of the particle after mutation, x is the original particle position, x maxand x min are the maximum and minimum positions of the two parent particles participating in the crossover operation.

[0095] A weighted fitness function is added to achieve a multi-objective balance among accuracy, efficiency, and resource usage. The multi-objective fitness function is:

[0096]

[0097] Among them, FLOPs max The upper limit of the model calculation is usually 22.8×10 6 ;Mem max Mem is the upper limit of memory usage. max =16.0MB; Mem is the model memory usage; the accuracy weight is 70%, the computational effort 20%, and the memory 10%; Acc is the model diagnosis accuracy, and FLOPs is the model computational effort (number of floating-point operations).

[0098] Add a weighted fitness function and map the continuous speed value to a binary feature selection result:

[0099]

[0100] in, is the position vector of particle i at iteration t+1, To represent the Sigmoid function acting on the back ; sigmoid(v) is the Sigmoid function, which maps the continuous velocity to the probability interval, and v is the velocity vector of the particle during the optimization process.

[0101] When a preset termination condition is reached, iterations are terminated and feature data is obtained; for example, when the maximum number of iterations (100) is reached, or when the fitness function changes by less than 1e-4 after 20 consecutive iterations. An optimization algorithm is used to select the optimal subset of 15 fault-sensitive features (such as PD pulse density and temperature gradient) from the 42-dimensional feature set (25 for partial discharge, 17 for temperature, 15 for strain, and 5 for gas), reducing the model input dimensionality and improving computational efficiency.

[0102] In order to achieve high accuracy with limited computing resources and avoid overfitting, it is necessary to build a lightweight fault diagnosis model. Figure 4 , the model framework is:

[0103] Input layer: optimized 20-dimensional features. 1D-CNN layer (extracting local spatiotemporal features): 3 convolution layers (kernel size = 5, number of channels = 32 → 16 → 8), stride = 2, ReLU activation. Transformer encoder (capturing global feature dependencies): 2-head attention, hidden layer dimension = 64. Output layer: Softmax classification, output failure probability:

[0104]

[0105] Among them, z k is the model output value of the kth class (k=1~5), z is the natural constant e k Power.

[0106] The model training process is as follows: Optimizer: Adopt AdamW optimizer, combined with adaptive learning rate and weight decay to prevent overfitting (learning rate = 3e-4, weight decay = 0.01). Loss function: In order to alleviate the class imbalance (high proportion of normal data) and prevent the model from being overconfident, label smoothing cross entropy:

[0107]

[0108] Among them, ∈ = 0.1 is the label smoothing parameter, which is used to smooth the true label and improve the generalization ability of the model; C is the number of categories; y i is the probability value of the i-th category in the true label, p i The probability value of the i-th category predicted by the model; the batch size is: 32, the training rounds are: 100; the early stopping strategy is adopted: if the validation set loss does not decrease for 10 consecutive rounds, the training is terminated to avoid overtraining.

[0109] The model validation process is: sliding window KS test:

[0110]

[0111] Among them, D KS is the KS statistic, F new (x) is the empirical distribution function of the new data window; F old (x) is the empirical distribution function of the old data window; the window size is 100 samples, and the model update is triggered when DKS>0.3. 10% of the historical data is retained and mixed with the new data to fine-tune the model parameters.

[0112] S2: Input the characteristic data of dry bushing into the pre-trained fault diagnosis model and output the corresponding fault type and its confidence level.

[0113] The characteristic data is obtained in the following manner:

[0114] Acquire real-time operating data of the dry bushing; wherein the real-time operating data includes the pulse current waveform, temperature distribution matrix, stress value, and gas concentration generated by epoxy resin decomposition of the dry bushing;

[0115] The improved particle swarm optimization algorithm is used to optimize the operation data and obtain the characteristic data;

[0116] The pre-trained fault diagnosis model is obtained by training a neural network model using feature data; the neural network model uses a hybrid architecture of 1D-CNN and Transformer as the model framework; the improved particle swarm optimization algorithm is obtained by introducing a genetic algorithm into the particle swarm optimization algorithm, and the goal of the optimization selection is to screen out the optimal subset of fault-sensitive components, specifically:

[0117] The lightweight model uses real-time characteristic data from bushing operation as input. The model selects the fault type corresponding to the highest probability value output by the Softmax function as the initial judgment result. A confidence threshold (e.g., 90%) is set, and a fault alarm is triggered only when the highest probability exceeds the threshold. A secondary verification is performed using multimodal characteristic values ​​(for example, to determine insulation aging, the following conditions must simultaneously be met: PD pulse density > 150 pulses / s and SO2 concentration > 5 ppm).

[0118] The model determines that a fault has occurred and outputs the fault type (such as "insulation aging") and its confidence level (such as 92%). The edge device (such as Jetson Nano) triggers a local alarm through an LED or buzzer and sends alarm information (including fault type, confidence level, and RUL) to the operation and maintenance platform through the Internet of Things (IoT) module. The edge end directly outputs the results to avoid cloud latency.

[0119] See also Figure 5 The present invention provides a dry-type casing fault diagnosis system, comprising:

[0120] Characteristic data acquisition module: used to obtain characteristic data of dry casing;

[0121] Fault type and confidence output module: used to input the characteristic data of dry bushing into the pre-trained fault diagnosis model and output the corresponding fault type and confidence;

[0122] The characteristic data is obtained in the following manner:

[0123] Obtain operating data of dry casing;

[0124] The improved particle swarm optimization algorithm is used to optimize the operation data and obtain the characteristic data;

[0125] The pre-trained fault diagnosis model is obtained by training a neural network model using feature data;

[0126] The improved particle swarm optimization algorithm is obtained by introducing a genetic algorithm into the particle swarm optimization algorithm, and the goal of the optimization selection is to screen out the optimal subset that is sensitive to faults.

[0127] The system achieves this by integrating a feature data acquisition module with a fault type and confidence output module. This module then inputs the dry bushing feature data into a pre-trained fault diagnosis model, outputting the corresponding fault type and confidence level. By improving the fault diagnosis of a particle swarm optimization algorithm and integrating it with a neural network model, the system leverages the global search capabilities of the particle swarm optimization algorithm and the local search advantages of the genetic algorithm to select an optimal subset of fault-sensitive features as feature data, providing a higher-quality data foundation for subsequent fault diagnosis. This significantly reduces the amount of data input to the fault diagnosis model, lowering the model's computational complexity and improving diagnostic speed while avoiding the interference of redundant data on the diagnostic results, making the diagnostic process more efficient. Furthermore, the system leverages the powerful nonlinear mapping capabilities of the neural network to accurately capture the complex relationship between dry bushing feature data and fault type, enabling dry bushing fault diagnosis. The system not only outputs the fault type but also provides the corresponding confidence level, providing more comprehensive diagnostic results for maintenance personnel. This allows them to make more reasonable maintenance decisions based on the confidence level, avoiding blind repairs and improving maintenance efficiency. The system has a simple structure, good reliability, high fault diagnosis accuracy and efficiency, and has certain versatility and scalability. It is expected to promote the progress and development of fault diagnosis technology in the entire industry.

[0128] The present invention provides a terminal device comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of each of the aforementioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of each module / unit in each of the aforementioned apparatus embodiments are implemented.

[0129] The computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to accomplish the present invention.

[0130] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0131] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0132] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0133] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0134] The above description is merely a preferred embodiment of the present invention and is not intended to impose any limitation on the technical solution of the present invention. Those skilled in the art should understand that, without departing from the spirit and principles of the present invention, the technical solution can also be subjected to several simple modifications and replacements, and these modifications and replacements are also within the scope of protection covered by the claims.

Claims

1. A dry bushing fault diagnosis method, characterized in that: include: Obtain characteristic data of dry casing; Input the characteristic data of dry-type bushing into the pre-trained fault diagnosis model and output the corresponding fault type and its confidence level; The characteristic data is obtained in the following manner: Acquiring operating data of the dry bushing; wherein the operating data includes a pulse current waveform, a temperature distribution matrix, a stress value, and a gas concentration generated by decomposition of the epoxy resin of the dry bushing; The improved particle swarm optimization algorithm is used to optimize the operation data and obtain the characteristic data; The pre-trained fault diagnosis model is obtained by training a neural network model using feature data; The improved particle swarm optimization algorithm is obtained by introducing a genetic algorithm into the particle swarm optimization algorithm, and the goal of the optimization selection is to screen out the optimal subset that is sensitive to faults.

2. The dry bushing fault diagnosis method according to claim 1, characterized in that: The operating data includes a pulse current waveform, a temperature distribution matrix, a stress value, and a gas concentration generated by decomposition of the dry bushing.

3. The dry bushing fault diagnosis method according to claim 1, characterized in that: The neural network model adopts a hybrid architecture of 1D-CNN and Transformer as the model framework.

4. The dry bushing fault diagnosis method according to claim 1, characterized in that: The method of optimizing and selecting the operating data using the improved particle swarm optimization algorithm to obtain characteristic data is as follows: De-noising the running data; Normalize the denoised running data and build a feature pool; The particle swarm optimization algorithm is used to optimize the features in the feature pool to obtain feature data.

5. The dry bushing fault diagnosis method according to claim 4, characterized in that: The Daubechies-4 wavelet basis is used to denoise the running data.

6. The dry bushing fault diagnosis method according to claim 4, characterized in that: Z-score standardization was used to normalize the denoised running data to a standard normal distribution.

7. The dry bushing fault diagnosis method according to claim 4, characterized in that: The method of using the particle swarm optimization algorithm to optimize and select features in the feature pool to obtain feature data is as follows: Introducing elite particles x generated by genetic algorithm GA ; Among them, elite particles are generated by simulating binary crossover and polynomial mutation; Elite particle x GA Encode and initialize, and update and iterate; After a preset number of iterations, the individuals with the highest fitness are selected from the elite particles generated by GA and injected into the PSO population; Add a weighted fitness function and map the continuous speed value to a binary feature selection result: When the preset termination condition is reached, the iteration is terminated and the feature data is obtained.

8. A dry casing fault diagnosis system, characterized in that: include: Characteristic data acquisition module: used to obtain characteristic data of dry casing; Fault type and confidence output module: used to input the characteristic data of dry bushing into the pre-trained fault diagnosis model and output the corresponding fault type and confidence; The characteristic data is obtained in the following manner: Obtain operating data of dry casing; The improved particle swarm optimization algorithm is used to optimize the operation data and obtain the characteristic data; The pre-trained fault diagnosis model is obtained by training a neural network model using feature data; The improved particle swarm optimization algorithm is obtained by introducing a genetic algorithm into the particle swarm optimization algorithm, and the goal of the optimization selection is to screen out the optimal subset that is sensitive to faults.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images