Lightning arrester state diagnosis method, device and equipment

Abstract

The invention discloses a lightning arrester state diagnosis method, device and equipment, and relates to the technical field of lightning arrester state diagnosis, and the method comprises the following steps: obtaining the geometric structure and material information of a lightning arrester, building an electric heating multi-physics field coupling model based on the impact voltage effect, and extracting electric heating response key data; obtaining discharge current and temperature time sequence data corresponding to different degradation stages to construct an electric heating time sequence response feature database, and extracting feature indexes consistent with the electric heating response key data to establish a sample data set; based on a dual-channel Bi-LSTM and an attention mechanism, according to the sample data set, constructing a lightning arrester classification model and carrying out classification identification on the operation state of the lightning arrester; predicting the remaining use time of the lightning arrester under different operation conditions according to classification identification in combination with reinforcement learning; according to the method, the multi-physical field simulation data and the actually measured time sequence characteristics are fused, so that accurate identification of the degradation state of the lightning arrester and dynamic prediction of the residual life of the lightning arrester are realized.

Description

Technical Field

[0001] This invention relates to the field of surge arrester condition diagnosis technology, and more specifically, to surge arrester condition diagnosis methods, apparatus and equipment. Background Technology

[0002] Surge arresters are critical devices in power systems used to limit overvoltages and protect electrical equipment; their operational reliability directly impacts power grid safety. With increasing grid voltage levels and more complex operating environments, surge arresters, subjected to long-term lightning strikes and damp-heat aging, experience gradual deterioration of their internal materials, leading to decreased electrical performance and even failure. Therefore, accurate diagnosis of surge arrester conditions and prediction of remaining life are of paramount importance.

[0003] Currently, surge arrester condition monitoring mainly relies on traditional electrical parameters (such as leakage current and resistive current) and single physical quantity detection methods such as infrared thermography. However, these methods have the following drawbacks.

[0004] Existing technologies are mostly based on the characteristics of a single electric field or temperature field, failing to comprehensively consider the coupling effect of multiple physical fields such as electrothermal fields, resulting in an incomplete characterization of the internal degradation state of surge arresters. For example, leakage current alone is insufficient to reflect local hot spots or non-uniform material degradation, while relying solely on temperature data cannot distinguish between electrical performance degradation and external environmental interference.

[0005] Traditional methods typically rely on static thresholds or empirical formulas to determine the state of surge arresters, lacking in-depth analysis of their time-series dynamic characteristics. Especially under impulse voltage, the electrothermal response of surge arresters exhibits significant time-varying and nonlinear characteristics, making it difficult for existing methods to effectively extract key features and establish accurate mappings between degradation stages. To address these issues, this invention proposes a solution. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a surge arrester condition diagnosis method, apparatus, and device. The surge arrester intelligent condition diagnosis method is based on electrothermal multiphysics coupling modeling, dual-channel Bi-LSTM-attention mechanism classification, and reinforcement learning prediction. By fusing multiphysics simulation data with measured time series characteristics, it achieves accurate identification of surge arrester degradation state and dynamic prediction of remaining life.

[0007] To achieve the above objectives, the present invention provides the following technical solution: Firstly, this application provides a surge arrester condition diagnosis method, which includes: acquiring surge arrester geometric structure and material information, establishing an electrothermal multiphysics coupling model based on impulse voltage and extracting key electrothermal response data; acquiring discharge current and temperature time series data corresponding to different degradation stages to construct an electrothermal time series response feature database, and extracting feature indicators consistent with the key electrothermal response data to establish a sample dataset; constructing a surge arrester classification model based on dual-channel Bi-LSTM and attention mechanism according to the sample dataset and classifying and identifying the surge arrester operating status; and predicting the remaining service time of the surge arrester under different operating conditions based on the classification and identification combined with reinforcement learning.

[0008] In one embodiment, the geometric structure and material information of the surge arrester are acquired, an electrothermal multiphysics coupling model based on impulse voltage is established, and key electrothermal response data are extracted; discharge current and temperature time series data corresponding to different degradation stages are acquired to construct an electrothermal time series response feature database, and feature indicators consistent with the key electrothermal response data are extracted to establish a sample dataset; based on dual-channel Bi-LSTM and attention mechanism, a surge arrester classification model is constructed according to the sample dataset, and the operating state of the surge arrester is classified and identified; based on the classification and identification combined with reinforcement learning, the remaining service time of the surge arrester under different operating conditions is predicted.

[0009] In one embodiment, discharge current and temperature time-series data corresponding to different degradation stages are acquired to construct an electrothermal time-series response feature database, and feature indicators consistent with key electrothermal response data are extracted to establish a labeled dataset. Specifically, the following steps are taken: A surge arrester sample with a structure consistent with the three-dimensional simulation model is acquired, and the sample undergoes damp heat and impulse current aging treatment; a standard lightning impulse voltage waveform is applied to the sample after each treatment stage to acquire discharge current and temperature time-series data; based on the moisture time and current carrying number in the damp heat and impulse current aging treatment, degradation stage labels are defined, including initial state, mild degradation, moderate degradation, and severe degradation; wavelet denoising processing is performed on the discharge current and temperature time-series data, and the electrothermal time-series response feature database is constructed in conjunction with the degradation stage labels; key response features consistent with electrothermal physical property parameters are extracted from the electrothermal time-series response feature database, cross-referenced, and judged by the Pearson correlation coefficient and a preset threshold; samples with correlation coefficients greater than the preset threshold are selected to construct a sample dataset.

[0010] In one embodiment, the discharge current and temperature time series data are subjected to wavelet denoising processing, and an electrothermal time series response feature database is constructed by combining the degradation stage labels. Specifically, the discharge current and temperature time series data are time-synchronized, using the rising edge of the zero point of the impulse voltage as the time reference, and the current and temperature time series data are mapped to continuous time signals under the standard time axis to form a dual-channel synchronous signal pair, which includes a current signal and a temperature sequence; wavelet transform is applied to the current signal and temperature sequence respectively for multi-scale decomposition, and the db4 wavelet basis function is used to perform three-level wavelet decomposition of the signal; soft thresholding is used to filter and denoise the detail coefficients of each level. An inverse transform is performed on the denoised multi-scale wavelet coefficients to reconstruct denoised current and temperature time-series data. Feature extraction is performed on the current and temperature time-series data to obtain electrical and thermal response features. The electrical response features include peak current density, maximum electric field strength, rising slope, and half-peak width. The thermal response features include maximum temperature rise, maximum temperature rise rate, uneven heat distribution, and local hotspot propagation time. The extracted electrical and thermal response features are organized into a vector structure according to a unified format and associated with the current sample degradation stage label to form feature sample entries. All sample entries are fused to generate an electrothermal time-series response feature database.

[0011] In one embodiment, based on a dual-channel Bi-LSTM and attention mechanism, a surge arrester classification model is constructed according to the sample dataset to classify and identify the operating status of the surge arrester. Specifically: Electrical and thermal response features are extracted from the sample dataset to generate electrical and thermal response channels. Independent Bi-LSTM network branches are constructed for each channel, outputting the bidirectional hidden state at each time step. Each Bi-LSTM branch consists of a forward LSTM and a backward LSTM. The outputs of the two branches are connected to the in-channel attention mechanism to calculate the attention weights at each time step and obtain weighted representation vectors, resulting in electrical response weighted vectors and thermal response weighted vectors. These vectors are concatenated to form a fused feature vector. The fused feature vector is input into a three-layer perceptron, where each layer has a preset number of neurons and classifications, and ReLU activation is used for feature mapping. The output layer obtains the arrester state classification probability vector using the softmax function. Based on the degradation stage label corresponding to the maximum probability vector, the arrester's operating state is identified.

[0012] In one embodiment, the remaining service time of the surge arrester under different operating conditions is predicted based on classification and reinforcement learning. Specifically, the following steps are taken: the degradation stage label output by the surge arrester classification model is obtained; the time-series monitoring data associated with the corresponding degradation stage is obtained and the current operating condition parameters are extracted; based on the reinforcement learning algorithm, the state space is defined by the degradation stage label and the operating condition parameters; the action space is defined as the remaining service time decay rate adjustment action, and a reward function is set, which includes premature maintenance penalty, prediction error penalty, and late maintenance reward; based on the PPO policy network output action, the preset remaining service time process model is updated to obtain the remaining service time of the surge arrester.

[0013] Secondly, this application provides a surge arrester condition diagnostic device, which includes: The simulation module is used to acquire the geometric structure and material information of the surge arrester, establish an electrothermal multiphysics coupling model based on impulse voltage, and extract key data of electrothermal response. The sample dataset acquisition module is used to acquire discharge current and temperature time series data corresponding to different degradation stages to construct an electrothermal time series response feature database, and extract feature indicators consistent with key electrothermal response data to establish a sample dataset. The surge arrester status classification and recognition module is used to construct a surge arrester classification model based on the sample dataset and classify and recognize the operating status of the surge arrester based on a dual-channel Bi-LSTM and attention mechanism. The surge arrester remaining time prediction module is used to predict the remaining service time of the surge arrester under different operating conditions based on classification and reinforcement learning.

[0014] Thirdly, this application provides an electronic device, characterized in that it includes: Memory, used to store computer programs; A processor, configured to implement the steps of the surge arrester condition diagnosis method as described in any one of claims 1 to 8 when executing the computer program.

[0015] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages: 1. By establishing a multi-physics coupled electrothermal model, comprehensively considering the dynamic responses of electric field, current density, and temperature field, the limitations of traditional single-physical-quantity detection are overcome, enabling a more comprehensive and accurate characterization of the arrester's degradation state. Employing wavelet denoising and temporal feature mining techniques, combined with dual-channel Bi-LSTM and attention mechanisms, the spatiotemporal correlation of the electrothermal response is effectively captured, significantly improving the accuracy and robustness of state classification, particularly suitable for complex nonlinear and high-noise operating conditions. A dynamic prediction model based on reinforcement learning (PPO algorithm) can adaptively adjust the remaining life assessment strategy according to real-time classification results and operating conditions (such as humidity and voltage fluctuations), avoiding the errors of traditional fixed-parameter models. Simultaneously, a reward and punishment mechanism optimizes maintenance decisions, balancing safety and economy. From simulation modeling to measured data verification, a closed-loop technology chain is formed, which can directly connect to existing monitoring systems, realizing an intelligent upgrade from periodic inspections to predictive maintenance, providing a scientific basis for power equipment operation and maintenance, and improving the reliability of power grid operation. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of the surge arrester condition diagnosis method provided in the embodiments of this application.

[0017] Figure 2 A schematic diagram of the structure of the surge arrester condition diagnosis device provided in the embodiments of this application.

[0018] Figure 3 Schematic flowchart of the surge arrester condition diagnosis method provided in the embodiments of this application Figure 2 . Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Reference Figure 1 As shown in the schematic diagram of the surge arrester condition diagnosis method provided in this application embodiment, it includes the following steps: S1. Obtain the geometric structure and material information of the surge arrester, establish an electrothermal multiphysics coupling model based on impulse voltage, and extract key data of electrothermal response.

[0021] In one exemplary embodiment, Figure 3 Schematic flowchart of the surge arrester condition diagnosis method provided in the embodiments of this application Figure 2 ,like Figure 3 As shown, in Figure 1Based on this, the steps of the multimodal image fusion method are illustrated by example. In step S1, an electrothermal multiphysics coupling model based on the action of impulse voltage is established, specifically as follows: A three-dimensional simulation model is constructed based on the geometric structure and material information of the surge arrester, and corresponding electrothermal physical property parameters are defined for each structural component. The electrothermal physical property parameters include electrical conductivity, thermal conductivity and specific heat capacity. A standard lightning impulse voltage waveform is applied to the model, and the upper and lower electrodes are set as excitation and grounding boundaries to construct a complete electric field simulation region including the air domain. It should be noted that applying a standard lightning impulse voltage waveform (such as a 1.2 / 50μs waveform) to the simulation model can effectively reproduce the high-voltage excitation environment that the surge arrester experiences in an actual lightning strike. Setting the upper and lower electrodes as the excitation end and ground boundary, respectively, ensures that the electric field distribution conforms to the actual current-carrying path. Simultaneously, introducing a complete electric field simulation region including the air domain avoids electric field boundary truncation errors, ensuring the natural expansion and dissipation of electric field lines in the simulation, thereby improving simulation accuracy. This step provides high-fidelity boundary conditions and initial excitation for subsequent current density, Joule heat source, and thermal coupling analysis, and is fundamental to the entire multiphysics coupling simulation.

[0022] In the complete electric field simulation region, a multiphysics coupling model containing electric field and heat conduction equations is constructed. By solving the multiphysics coupling model, the voltage intensity, current density, and temperature field distribution are obtained. Based on the Joule heat source term driven by the current density, the electric field and the temperature field are dynamically coupled to obtain the electrothermal simulation system. Transient simulation is performed on the electrothermal simulation system, and the key physical quantity distribution data in the simulation results are extracted. The key physical quantity distribution data includes response data of internal electric field intensity, current density and temperature distribution evolving over time. Based on the distribution data of key physical quantities, key electrothermal response data are extracted according to a preset time window and spatial location. The key electrothermal response data includes the maximum electric field strength, peak current density, local temperature rise rate, and heat distribution non-uniformity, which are used to characterize the response state of the surge arrester under impulse voltage.

[0023] The electric field equations are calculated using the following formulas:

[0024] In the formula, For electric field strength, For the potential gradient, Electric potential, also known as voltage.

[0025] The specific formula for calculating current density is as follows:

[0026] In the formula, For current density, This represents temperature-dependent conductivity.

[0027] The specific calculation formula for the Joule heat source term is as follows:

[0028] In the formula, For the Joule heat source term; It should be noted that the Joule heat source term formula is the core connecting term of electrothermal coupling, converting the electric field calculation result E into the heat source term. The temperature field is updated by inputting it into the heat conduction equation.

[0029] The heat conduction equation, and the specific calculation formula are as follows:

[0030] In the formula, For the density of the medium, Where is the specific heat capacity, and T is the temperature field. The local temperature rise rate describes the rate of temperature change per unit time, and k is the thermal conductivity, reflecting the material's ability to conduct heat. For temperature gradient, The divergence of heat flux.

[0031] The specific formula for calculating uneven heat distribution is as follows:

[0032] In the formula, The variance of the temperature field reflects the non-uniformity of the overall heat distribution. The effective volume inside the surge arrester. Let r be the temperature field at spatial location r. The average temperature is the temperature over the entire volume.

[0033] S2. Obtain discharge current and temperature time series data corresponding to different degradation stages to construct an electrothermal time series response feature database, and extract feature indicators consistent with key electrothermal response data to establish a sample dataset.

[0034] In this embodiment, discharge current and temperature time-series data corresponding to different degradation stages are obtained to construct an electrothermal time-series response feature database, and feature indicators consistent with key electrothermal response data are extracted to establish a labeled dataset, specifically: Obtain a surge arrester sample with a structure consistent with the three-dimensional simulation model, and perform damp heat and impulse current aging treatments on the sample; A standard lightning impulse voltage waveform was applied to the sample after each processing stage to obtain discharge current and temperature time-series data. Based on the moisture absorption time and current flow frequency during the damp heat and impact current aging treatment, the degradation stage labels are divided into initial state (no moisture, no current flow), mild degradation (current peak decrease of more than 10%, or temperature rise of more than 5°C), moderate degradation (current decrease of more than 25%, or temperature rise of more than 10°C), and severe degradation (current decrease of more than 40%, or temperature rise of more than 15°C). Wavelet denoising was performed on the discharge current and temperature time series data, and an electrothermal time series response feature database was constructed by combining the degradation stage labels. Key response features consistent with electrothermal physical property parameters are extracted from the electrothermal time-series response feature database, cross-referenced, and judged by Pearson correlation coefficient and preset threshold. Sample datasets are constructed by selecting samples with correlation coefficients greater than a preset threshold.

[0035] Furthermore, wavelet denoising processing is performed on the discharge current and temperature time series data, and an electrothermal time series response feature database is constructed by combining the degradation stage labels, specifically: Time synchronization processing is performed on the discharge current and temperature timing data. The rising edge of the zero point of the impulse voltage is used as the time reference. The current and temperature timing data are mapped into a continuous time signal under the standard time axis to form a dual-channel synchronization signal pair. The signal pair includes the current signal and the temperature sequence. Wavelet transform was applied to the current signal and temperature sequence for multi-scale decomposition, and the db4 wavelet basis function was used to perform three-level wavelet decomposition of the signal. The specific calculation formula for the db4 wavelet basis function is as follows:

[0036] In the formula, The original signal refers to the current signal and temperature sequence. For the approximate (low-frequency) component of the Jth layer, Let J be the detail (high-frequency) component of the j-th layer, and J be the decomposition layer number.

[0037] A soft thresholding function is used to filter and denoise each layer of detail coefficients, setting coefficients below the threshold to zero and shrinking those above the threshold. The specific calculation formula for the soft thresholding function is as follows:

[0038] In the formula, For the threshold, Let n be the noise standard deviation and n be the signal length.

[0039] Perform an inverse transform on the denoised multiscale wavelet coefficients to reconstruct the denoised current time series data and temperature time series data. Feature extraction is performed on current time series data and temperature time series data to obtain electrical response features and thermal response features. The electrical response features include peak current density, maximum electric field strength, rise slope, and half-peak width. The thermal response features include maximum temperature rise, maximum temperature rise rate, uneven heat distribution, and local hot spot propagation time. The extracted electrical and thermal response features are organized into a vector structure in a unified format and associated with the current sample degradation stage label to form feature sample entries. All sample entries are fused to generate an electrothermal time-series response feature database.

[0040] It should be noted that by using time synchronization processing based on the rising edge of the lightning impulse zero point, the error problem of time domain alignment between the current signal and the temperature sequence is solved, ensuring the accuracy of signal fusion. By using the db4 wavelet basis function for multi-scale decomposition and soft threshold filtering, high-frequency noise is effectively removed while retaining key signal features, improving the signal-to-noise ratio and feature stability. Through the fine extraction of key physical features (such as peak current density and maximum temperature rise rate) in the electrical and thermal time series signals, the multidimensional response state of the surge arrester under impulse voltage is comprehensively characterized. The final structured electrothermal feature database has the advantages of strong labeling and high time resolution, and can be directly used for various intelligent analysis tasks such as supervised learning, cluster analysis, or degradation trend prediction, significantly enhancing the accuracy, reliability, and engineering practicality of surge arrester condition diagnosis.

[0041] S3, based on dual-channel Bi-LSTM and attention mechanism, constructs a surge arrester classification model according to the sample dataset and classifies and identifies the operating status of surge arresters.

[0042] In this embodiment, based on a dual-channel Bi-LSTM and attention mechanism, a surge arrester classification model is constructed according to the sample dataset to classify and identify the operating status of surge arresters. Specifically: The electrical response features and thermal response features in the sample dataset are extracted and electrical response channels and thermal response channels are generated. Independent Bi-LSTM network branches are constructed for the electrical response channel and the thermal response channel respectively, and the bidirectional hidden state at each time step is output. Each Bi-LSTM network branch is composed of a forward LSTM and a backward LSTM, which are used to capture the forward and backward dependencies in the time series simultaneously. The outputs of the two branches are connected to the in-channel attention mechanism, and the attention weights at each time step are calculated. And obtain the weighted representation vector F= * The weighted vectors of the electrical response and the thermal response are obtained respectively. The attention weight The specific calculation formula is as follows:

[0043] In the formula, For a trainable parameter matrix, It is in a two-way hidden state.

[0044] The weighted vectors of electrical response and thermal response are concatenated to form a fused feature vector; The fused feature vector is input into a three-layer perceptron, where each layer has a preset number of neurons and a preset number of categories, and the ReLU activation function is used for feature mapping. The output layer obtains the arrester state classification probability vector based on the softmax function; Based on the degradation stage label corresponding to the maximum probability vector, the operating status of the surge arrester is identified.

[0045] It should be noted that this surge arrester classification model, based on dual-channel Bi-LSTM and an attention mechanism, significantly improves the accuracy and robustness of state recognition by independently processing the temporal features of electrical and thermal responses and fusing bidirectional dependencies, combined with the attention mechanism to dynamically focus on key fault signals. Its advantages lie in the adaptive fusion of multimodal features, deep capture of temporal dynamic characteristics, and classification optimization achieved through attention weighting and ReLU multi-layer mapping, ultimately outputting probabilistic results that support interpretable decision-making. Compared to traditional methods, this approach is more suitable for early fault warning and refined state assessment in complex electromagnetic environments, combining computational efficiency with engineering practicality, providing reliable technical support for intelligent operation and maintenance of surge arresters.

[0046] S4, based on classification and reinforcement learning, predicts the remaining service time of the surge arrester under different operating conditions.

[0047] In this embodiment, the remaining service time of the surge arrester under different operating conditions is predicted based on classification and reinforcement learning, specifically as follows: Obtain the degradation stage label output by the surge arrester classification model, obtain the time-series monitoring data associated with the corresponding degradation stage, and extract the current operating condition parameters. The time-series monitoring data includes the average discharge current, temperature gradient, and partial discharge amplitude. The current operating condition parameters include ambient humidity, system voltage fluctuation coefficient, and cumulative current carrying times. Based on the reinforcement learning algorithm, the state space is defined by the degradation stage label and the operating condition parameters; Define the action space as an action space for adjusting the remaining usage time decay rate, and set a reward function, which includes a premature maintenance penalty. Prediction error penalty and late maintenance rewards ; Based on the PPO policy network output action, the preset remaining usage time process model is updated to obtain the remaining usage time of the surge arrester.

[0048] The reward function is calculated using the following formula:

[0049] In the formula, For the reward function, , , These are the weighting coefficients.

[0050] The specific calculation formula for the remaining usage time process model is as follows:

[0051] In the formula, Remaining usage time As the reference attenuation rate, Where is the diffusion coefficient. For standard Brownian motion, For action.

[0052] It should be noted that by integrating the arrester degradation classification results with reinforcement learning (PPO algorithm), dynamic and accurate prediction of remaining service time is achieved. Its innovation lies in: First, fusing the degradation stage labels output by the classification model with real-time monitoring data (discharge current, temperature, etc.) and operating parameters (humidity, voltage fluctuations, etc.) to construct a state space significantly improves the comprehensiveness of the prediction; second, using the PPO algorithm to dynamically adjust the decay rate of remaining service time overcomes the limitations of traditional fixed-parameter models; finally, through a composite reward function that includes a reward and penalty mechanism for premature / late maintenance, maintenance costs are optimized while ensuring safety. This solution can be directly integrated with existing monitoring systems, enabling an intelligent upgrade from periodic inspections to predictive maintenance, providing a scientific basis for arrester health management, effectively extending equipment life and ensuring stable grid operation.

[0053] Reference Figure 2 As shown in the schematic diagram, the surge arrester condition diagnosis device provided by the present invention includes a simulation module, a sample dataset acquisition module, a surge arrester condition classification and recognition module, and a surge arrester remaining time prediction module. The simulation module is used to acquire the geometric structure and material information of the surge arrester, establish an electrothermal multiphysics coupling model based on impulse voltage, extract key data of the electrothermal response, and predict the remaining time of the surge arrester. These modules are interconnected. The sample dataset acquisition module is used to acquire discharge current and temperature time series data corresponding to different degradation stages to construct an electrothermal time series response feature database, and extract feature indicators consistent with key electrothermal response data to establish a sample dataset. The surge arrester status classification and recognition module is used to construct a surge arrester classification model based on the sample dataset and classify and recognize the operating status of the surge arrester based on a dual-channel Bi-LSTM and attention mechanism. The surge arrester remaining time prediction module is used to predict the remaining service time of the surge arrester under different operating conditions based on classification and reinforcement learning.

[0054] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0055] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0056] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0057] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0059] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing the condition of a surge arrester, characterized in that, Includes the following steps: Obtain the geometric structure and material information of the surge arrester, establish an electrothermal multiphysics coupling model based on impulse voltage, and extract key data of electrothermal response. We acquired discharge current and temperature time-series data corresponding to different degradation stages to construct an electrothermal time-series response feature database, and extracted feature indicators consistent with key electrothermal response data to establish a sample dataset. Based on dual-channel Bi-LSTM and attention mechanism, a surge arrester classification model is constructed according to the sample dataset to classify and identify the operating status of surge arresters; Based on classification and reinforcement learning, the remaining service life of the surge arrester under different operating conditions is predicted.

2. The surge arrester condition diagnosis method according to claim 1, characterized in that, The establishment of the electrothermal multiphysics coupling model based on the action of impulse voltage is specifically as follows: A three-dimensional simulation model is constructed based on the geometric structure and material information of the surge arrester, and corresponding electrothermal physical property parameters are defined for each structural component. A standard lightning impulse voltage waveform is applied to the model, and the upper and lower electrodes are set as excitation and grounding boundaries to construct a complete electric field simulation region including the air domain. In the complete electric field simulation region, a multiphysics coupling model containing electric field and heat conduction equations is constructed. By solving the multiphysics coupling model, the voltage intensity, current density, and temperature field distribution are obtained. Based on the Joule heat source term driven by the current density, the electric field and the temperature field are dynamically coupled to obtain the electrothermal simulation system. Transient simulation is performed on the electrothermal simulation system, and the key physical quantity distribution data in the simulation results are extracted. The key physical quantity distribution data includes response data of internal electric field intensity, current density and temperature distribution evolving over time. Based on the distribution data of key physical quantities, key data of electrothermal response are extracted according to a preset time window and spatial location. The key data of electrothermal response include the maximum electric field strength, peak current density, local temperature rise rate, and heat distribution non-uniformity.

3. The surge arrester condition diagnosis method according to claim 2, characterized in that, The process involves acquiring discharge current and temperature time-series data corresponding to different degradation stages to construct an electrothermal time-series response feature database, and extracting feature indicators consistent with key electrothermal response data to establish a labeled dataset. Specifically: Obtain a surge arrester sample with a structure consistent with the three-dimensional simulation model, and perform damp heat and impulse current aging treatments on the sample; A standard lightning impulse voltage waveform was applied to the sample after each processing stage to obtain discharge current and temperature time series data. Based on the moisture exposure time and current flow frequency during damp heat and impact current aging treatment, degradation stage labels are defined, including initial state, mild degradation, moderate degradation, and severe degradation. Wavelet denoising was performed on the discharge current and temperature time series data, and an electrothermal time series response feature database was constructed by combining the degradation stage labels. Key response features consistent with electrothermal physical property parameters are extracted from the electrothermal time-series response feature database, cross-referenced, and judged by Pearson correlation coefficient and preset threshold. Sample datasets are constructed by selecting samples with correlation coefficients greater than a preset threshold.

4. The surge arrester condition diagnosis method according to claim 3, characterized in that, The discharge current and temperature time series data are subjected to wavelet denoising processing, and an electrothermal time series response feature database is constructed by combining the degradation stage labels, specifically: Time synchronization processing is performed on the discharge current and temperature timing data. The rising edge of the zero point of the impulse voltage is used as the time reference. The current and temperature timing data are mapped into a continuous time signal under the standard time axis to form a dual-channel synchronization signal pair. The signal pair includes the current signal and the temperature sequence. Wavelet transform was applied to the current signal and temperature sequence for multi-scale decomposition, and the db4 wavelet basis function was used to perform three-level wavelet decomposition of the signal. A soft thresholding function is used to filter and denoise each level of detail coefficients; Perform an inverse transform on the denoised multiscale wavelet coefficients to reconstruct the denoised current time series data and temperature time series data. Feature extraction is performed on current time series data and temperature time series data to obtain electrical response features and thermal response features. The electrical response features include peak current density, maximum electric field strength, rise slope, and half-peak width. The thermal response features include maximum temperature rise, maximum temperature rise rate, uneven heat distribution, and local hot spot propagation time. The extracted electrical and thermal response features are organized into a vector structure in a unified format and associated with the current sample degradation stage label to form feature sample entries. All sample entries are fused to generate an electrothermal time-series response feature database.

5. The surge arrester condition diagnosis method according to claim 4, characterized in that, The method involves constructing a surge arrester classification model based on a dual-channel Bi-LSTM and attention mechanism, and classifying and identifying the operating states of surge arresters using the sample dataset. Specifically: The electrical response features and thermal response features in the sample dataset are extracted and electrical response channels and thermal response channels are generated. Independent Bi-LSTM network branches are constructed for the electrical response channel and the thermal response channel respectively, and the bidirectional hidden state at each time step is output. Each Bi-LSTM network branch is composed of a forward LSTM and a backward LSTM. The outputs of the two branch structures are connected to the in-channel attention mechanism respectively. The attention weights at each time step are calculated and the weighted representation vectors are obtained, resulting in the electrical response weighted vector and the thermal response weighted vector respectively. The weighted vectors of electrical response and thermal response are concatenated to form a fused feature vector; The fused feature vector is input into a three-layer perceptron, where each layer has a preset number of neurons and a preset number of categories, and the ReLU activation function is used for feature mapping. The output layer obtains the arrester state classification probability vector based on the softmax function; Based on the degradation stage label corresponding to the maximum probability vector, the operating status of the surge arrester is identified.

6. The surge arrester condition diagnosis method according to claim 5, characterized in that, The method of predicting the remaining service time of the surge arrester under different operating conditions based on classification and reinforcement learning is as follows: Obtain the degradation stage labels output by the surge arrester classification model, obtain the time-series monitoring data associated with the corresponding degradation stage, and extract the current operating condition parameters. Based on the reinforcement learning algorithm, the state space is defined by the degradation stage label and the operating condition parameters; Define the action space as the remaining usage time decay rate adjustment action and set a reward function, which includes a premature maintenance penalty, a prediction error penalty, and a late maintenance reward; Based on the PPO strategy network output action, the preset remaining usage time process model is updated to obtain the remaining usage time of the surge arrester.

7. The surge arrester condition diagnosis method according to claim 6, characterized in that, The specific calculation formula for the db4 wavelet basis function is as follows: In the formula, The original signal refers to the current signal and temperature sequence. For the approximate (low-frequency) component of the Jth layer, Let J be the detail (high-frequency) component of the j-th layer, and J be the decomposition layer number.

8. The surge arrester condition diagnosis method according to claim 7, characterized in that, The reward function is calculated using the following formula: In the formula, For the reward function, , , These are the weighting coefficients. To maintain punishment prematurely, Penalty for prediction error Rewards for late maintenance.

9. A surge arrester condition diagnostic device, characterized in that, include: The simulation module is used to acquire the geometric structure and material information of the surge arrester, establish an electrothermal multiphysics coupling model based on impulse voltage, and extract key data of electrothermal response. The sample dataset acquisition module is used to acquire discharge current and temperature time series data corresponding to different degradation stages to construct an electrothermal time series response feature database, and extract feature indicators consistent with key electrothermal response data to establish a sample dataset. The surge arrester status classification and recognition module is used to construct a surge arrester classification model based on the sample dataset and classify and recognize the operating status of the surge arrester based on a dual-channel Bi-LSTM and attention mechanism. The surge arrester remaining time prediction module is used to predict the remaining service time of the surge arrester under different operating conditions based on classification and reinforcement learning.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the surge arrester condition diagnosis method as described in any one of claims 1 to 8 when executing the computer program.

Images