A high-voltage switchgear fault tracing positioning method, device, medium and product

Abstract

The application discloses a high-voltage switchgear fault tracing positioning method, device, medium and product, relates to the technical field of power equipment online monitoring and fault diagnosis, and comprises the following steps: acquiring actual structure data of a target high-voltage switchgear; establishing a geometric simulation model; combining a gas diffusion mechanism equation, carrying out CFD numerical simulation of fault characteristic gas under multiple fault points, obtaining time-space concentration data sets of multiple monitoring points corresponding to different fault positions; screening fault differentiation characteristic parameters, constructing a fault identification feature vector, dividing a fault area, and generating a labeled sample set; training a double-branch intelligent classification model; collecting real-time gas concentration data at the monitoring points of the target high-voltage switchgear, extracting a fault identification feature vector and inputting the trained double-branch intelligent classification model, and outputting corresponding fault area prediction results. The application can realize real-time online accurate tracing of high-voltage switchgear faults and improve fault tracing positioning accuracy.

Description

Technical Field

[0001] This application relates to the field of online monitoring and fault diagnosis technology for power equipment, and in particular to a method, equipment, medium and product for tracing and locating faults in high-voltage switchgear. Background Technology

[0002] High-voltage switchgear refers to the core equipment of substations and power plant distribution units in power transmission and distribution systems. Air is used as the insulating medium inside high-voltage switchgear. When partial discharge or contact overheating faults occur, the insulating air decomposes to generate characteristic fault gases such as O3 (ozone), NO (nitric oxide), CO (carbon monoxide), and CO2 (carbon dioxide). Existing fault location methods mainly include ultrasonic location, ultra-high frequency location, combined acoustic and electrical location, and traditional gas analysis. Among these, ultrasonic and ultra-high frequency location methods are easily affected by the internal structure of the equipment, resulting in problems such as signal refraction, reflection, and environmental noise interference. While the combined acoustic and electrical location method can complement each other to some extent, it still relies on the deployment of multiple sensors and has limited location accuracy. Traditional gas analysis methods are mostly static offline detection; although they can determine whether there is a fault in the equipment, they cannot achieve real-time online fault tracing and ignore the diffusion effect of characteristic component gases inside the equipment, leading to inaccurate fault location determination.

[0003] Therefore, how to achieve real-time online accurate source tracing of faults in high-voltage switchgear and improve the accuracy of fault source tracing and location has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] The purpose of this application is to provide a method, equipment, medium, and product for tracing and locating faults in high-voltage switchgear, which can realize real-time online accurate tracing of faults in high-voltage switchgear and improve the accuracy of fault tracing and location.

[0005] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for tracing and locating faults in high-voltage switchgear, comprising the following steps: Obtain the actual structural data of the target high-voltage switchgear.

[0006] Based on the actual structural data, a geometric simulation model of the target high-voltage switchgear is established.

[0007] Based on the geometric simulation model and combined with the gas diffusion mechanism equation, CFD numerical simulation of the fault characteristic gas under multiple fault locations is carried out to obtain spatiotemporal concentration datasets of multiple monitoring points corresponding to different fault locations; the gas diffusion mechanism equation includes the molecular diffusion equation and the convection-diffusion coupling equation.

[0008] Fault discrimination feature parameters are selected from the spatiotemporal concentration dataset, fault identification feature vectors are constructed and fault regions are divided to generate a labeled sample set.

[0009] A dual-branch intelligent classification model is trained based on the labeled sample set to obtain the trained dual-branch intelligent classification model. The dual-branch intelligent classification model is an intelligent classification model that takes the fault identification feature vector as input and the corresponding fault area prediction result as output for fault source tracing and localization prediction. The dual-branch intelligent classification model includes a lightweight embedded branch and a cloud-based high-precision branch. The lightweight embedded branch adopts an SVM model, and the cloud-based high-precision branch adopts a CNN-LSTM hybrid model.

[0010] Real-time gas concentration data at the monitoring point of the target high-voltage switchgear is collected, fault identification feature vectors are extracted and input into the trained dual-branch intelligent classification model, and the corresponding fault area prediction results are output.

[0011] Optionally, the fault characteristic gas is at least one of O3, NO, CO, and CO2 generated by the decomposition of insulating air under fault conditions of the target high-voltage switchgear; the monitoring point is a pre-set gas intake port of the target high-voltage switchgear.

[0012] Optionally, based on the geometric simulation model and combined with the gas diffusion mechanism equation, CFD numerical simulation of the fault characteristic gas at multiple fault locations is carried out to obtain a spatiotemporal concentration dataset of multiple monitoring points corresponding to different fault locations. This specifically includes the following steps: The geometric simulation model is meshed and its mesh independence is verified to obtain effective mesh parameters.

[0013] Based on the effective mesh parameters and preset operating condition parameters, the simulation boundary conditions of the geometric simulation model are configured; the simulation boundary conditions include the working pressure of the equipment cavity, the ambient reference temperature, the fault point temperature, and the initial release amount of characteristic gas.

[0014] Based on the simulation boundary conditions, the molecular diffusion coefficients of each fault characteristic gas are calculated using the FSG empirical formula.

[0015] Based on the molecular diffusion coefficients of the characteristic gases of each fault, a laminar flow model or a standard k-ε turbulent flow model is selected according to the Reynolds number of the gas flow. A high-precision convection discretization scheme and a pressure-velocity coupled iterative algorithm are used to solve the diffusion control equations to obtain the full-time-series concentration raw data of a single fault location.

[0016] The fault location coordinates are transformed using the controlled variable method, and all fault locations are simulated. The full-time-series raw concentration data of each monitoring point are continuously collected to form the spatiotemporal concentration dataset.

[0017] Optionally, the high-precision convection discretization scheme adopts the QUICK discretization scheme, and the pressure-velocity coupled iterative algorithm adopts the SIMPLE algorithm.

[0018] Optionally, fault discrimination feature parameters are selected from the spatiotemporal concentration dataset to construct fault identification feature vectors and divide fault regions, generating a labeled sample set. This specifically includes the following steps: Based on the spatiotemporal concentration dataset, the concentration-distance decay model combined with the dual monitoring point location formula is used to screen fault discrimination feature parameters to obtain the screened feature parameters; the fault discrimination feature parameters include at least one of the following: concentration at a specified time of monitoring point, concentration ratio of multiple monitoring points, concentration rise rate, and gas diffusion uniformity time.

[0019] Based on the filtered feature parameters, a fault identification feature vector is constructed.

[0020] Based on the fault identification feature vector, the fault area is divided according to the fault location and labeled to generate the labeled sample set.

[0021] Optionally, the SVM model uses a radial basis function kernel and optimizes the penalty parameter and kernel width parameter through cross-validation; the CNN-LSTM hybrid model uses a weighted hybrid loss function including classification cross-entropy loss, mechanism consistency loss and region smoothing loss for iterative optimization.

[0022] Optionally, the lightweight embedded branch is deployed on the field measurement and control terminal corresponding to the target high-voltage switchgear, and the cloud-based high-precision branch is deployed on the remote server corresponding to the target high-voltage switchgear.

[0023] The on-site monitoring and control terminal is used to collect the real-time gas concentration data, perform preliminary fault tracing and location based on the real-time gas concentration data, and output the local fault area prediction result; at the same time, it also uploads the real-time gas concentration data to the remote server.

[0024] The remote server is used to perform fault source tracing and location again based on the real-time gas concentration data, and outputs cloud-based fault area prediction results, which can realize secondary verification of the fault area prediction results.

[0025] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the high-voltage switchgear fault tracing and location method described in any one of the above descriptions.

[0026] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the high-voltage switchgear fault tracing and location method described in any one of the above descriptions.

[0027] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the high-voltage switchgear fault tracing and location method described in any one of the above descriptions.

[0028] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, equipment, medium, and product for fault tracing and localization in high-voltage switchgear. First, it utilizes equipment structure modeling combined with gas diffusion mechanism equations to conduct CFD simulations and generate spatiotemporal concentration datasets. Samples are constructed based on simulation data, eliminating the need for extensive on-site destructive testing to collect fault samples, enabling efficient batch acquisition of effective data across multiple operating conditions and fault locations. Second, by selecting fault discrimination feature parameters to construct feature vectors and dividing fault regions, a labeled sample set is obtained. This eliminates invalid data, strengthens the correlation between features and fault locations, and improves the effectiveness of sample data. Third, it combines SVM and CNN-LSTM to form a dual-branch intelligent classification model. The lightweight embedded branch based on SVM and the high-precision cloud branch based on CNN-LSTM work collaboratively, respectively meeting the needs for rapid on-site assessment and refined cloud-based verification. Furthermore, features are extracted from real-time gas data from on-site monitoring points and input into the trained model to automatically predict fault areas. Fault tracing and localization can be completed based on gas monitoring, enabling real-time online accurate fault tracing in high-voltage switchgear and improving fault tracing and localization accuracy. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 An application environment diagram for a fault tracing and location method for high-voltage switchgear provided in an embodiment of this application; Figure 2 A flowchart illustrating a fault tracing and location method for high-voltage switchgear provided in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the principle of fault tracing and location for high-voltage switchgear, provided in one embodiment of this application. Figure 4This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0032] The purpose of this application is to provide a method, equipment, medium, and product for tracing and locating faults in high-voltage switchgear. This method can be applied to intelligent fault assessment scenarios in high-voltage switchgear at substations of multiple voltage levels, such as 10kV and 35kV. It can also be extended to fault detection in enclosed high-voltage electrical equipment such as GIS (Gas Insulated Metal Enclosed Switchgear) gas-filled switchgear. The method primarily involves gas tracing and locating faults such as partial discharge and overheating within high-voltage switchgear. It captures the diffusion patterns of characteristic gases under different fault conditions by conducting CFD (Computational Fluid Dynamics) numerical simulations of fault characteristic gases at multiple fault locations based on gas diffusion mechanisms. By extracting concentration characteristic parameters during the diffusion process and combining them with machine learning algorithms, it achieves multi-classification tracing of fault locations. This method eliminates the need for complex sensor deployments, enabling real-time online monitoring and accurate tracing, improving fault tracing and locating accuracy, and solving the problems of low location accuracy and inability to perform online tracing in existing technologies.

[0033] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] The fault tracing and location method for high-voltage switchgear provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up separately, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send the actual structural data of the target high-voltage switchgear and the real-time gas concentration data at the monitoring points to server 104. After receiving the actual structural data of the target high-voltage switchgear and the real-time gas concentration data at the monitoring points, server 104, based on the actual structural data, establishes a geometric simulation model; combines the gas diffusion mechanism equation to conduct CFD numerical simulation of fault characteristic gases at multiple fault locations, obtaining spatiotemporal concentration datasets for multiple monitoring points corresponding to different fault locations; filters fault discrimination feature parameters, constructs fault identification feature vectors and divides fault regions, generating a labeled sample set; trains a dual-branch intelligent classification model; and extracts fault identification feature vectors based on real-time gas concentration data and inputs them into the trained dual-branch intelligent classification model, outputting the corresponding fault region prediction results. Server 104 can feed back the obtained fault area prediction results to terminal 102. Furthermore, in some embodiments, the fault tracing and location method for high-voltage switchgear can also be implemented independently by server 104 or terminal 102. For example, terminal 102 can directly perform fault tracing and location processing based on the actual structural data of the target high-voltage switchgear and the real-time gas concentration data at the monitoring points; alternatively, server 104 can obtain the actual structural data of the target high-voltage switchgear and the real-time gas concentration data at the monitoring points from the data storage system and perform fault tracing and location processing based on these data.

[0035] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, and IoT devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0036] In one exemplary embodiment, such as Figure 2 As shown, a method for tracing and locating faults in high-voltage switchgear is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the following steps are included: S1: Obtain the actual structural data of the target high-voltage switchgear.

[0037] S2: Based on the actual structural data, establish a geometric simulation model of the target high-voltage switchgear.

[0038] S3: Based on the geometric simulation model and combined with the gas diffusion mechanism equation, CFD numerical simulation of the fault characteristic gas under multiple fault locations is carried out to obtain the spatiotemporal concentration dataset of multiple monitoring points corresponding to different fault locations; the gas diffusion mechanism equation includes the molecular diffusion equation and the convection-diffusion coupling equation.

[0039] S4: Select fault discrimination feature parameters from the spatiotemporal concentration dataset, construct fault identification feature vectors and divide fault regions to generate a labeled sample set.

[0040] S5: Train a dual-branch intelligent classification model based on the labeled sample set to obtain the trained dual-branch intelligent classification model. The dual-branch intelligent classification model is an intelligent classification model used for fault source tracing and localization prediction, taking the fault identification feature vector as input and the corresponding fault area prediction result as output. The dual-branch intelligent classification model includes a lightweight embedded branch and a cloud-based high-precision branch. The lightweight embedded branch uses an SVM (Support Vector Machine) model, and the cloud-based high-precision branch uses a CNN-LSTM (Convolutional-Long Short-Term Memory Neural Network) hybrid model.

[0041] S6: Collect real-time gas concentration data at the monitoring point of the target high-voltage switchgear, extract fault identification feature vectors and input them into the trained dual-branch intelligent classification model, and output the corresponding fault area prediction results.

[0042] By implementing steps S1 to S6 above, CFD numerical simulations of fault characteristic gases at multiple fault locations are conducted based on the gas diffusion mechanism to capture the diffusion patterns of characteristic component gases under different fault conditions. By extracting concentration characteristic parameters during the diffusion process and combining them with machine learning algorithms, a dual-branch intelligent classification model architecture of "lightweight embedded branch (SVM) + cloud high-precision branch (CNN-LSTM)" is used to predict fault source location. This enables real-time online monitoring and accurate source tracing, improving the accuracy of fault source location.

[0043] As an optional implementation, in step S3, the fault characteristic gas is at least one of O3, NO, CO, and CO2 generated by the decomposition of insulating air under fault conditions of the target high-voltage switchgear; the monitoring point is a pre-set gas intake port of the target high-voltage switchgear.

[0044] As an optional implementation, in step S3, based on the geometric simulation model and combined with the gas diffusion mechanism equation, CFD numerical simulation of the fault characteristic gas under multiple fault locations is carried out to obtain a spatiotemporal concentration dataset of multiple monitoring points corresponding to different fault locations. This specifically includes the following steps: S31: Perform mesh generation and mesh independence verification on the geometric simulation model to obtain effective mesh parameters.

[0045] S32: Based on the effective mesh parameters and preset operating condition parameters, configure the simulation boundary conditions of the geometric simulation model; the simulation boundary conditions include the working pressure of the equipment cavity, the ambient reference temperature, the fault point temperature, and the initial release amount of characteristic gas.

[0046] S33: Based on the simulation boundary conditions, the molecular diffusion coefficients of each fault characteristic gas are calculated using the FSG (Fuller-Schettler-Giddings) empirical formula.

[0047] S34: Based on the molecular diffusion coefficients of the characteristic gases of each fault, a laminar flow model or a standard k-ε (turbulent kinetic energy-dissipation rate) turbulent flow model is selected according to the Reynolds number of the gas flow. A high-precision convection discretization scheme and a pressure-velocity coupled iterative algorithm are used to solve the diffusion control equations to obtain the full-time-series concentration raw data of a single fault location.

[0048] S35: The fault location coordinates are transformed using the controlled variable method, and the simulation of all fault locations is traversed. The full-time-series raw concentration data of each monitoring point is continuously collected to form the spatiotemporal concentration dataset.

[0049] As an optional implementation, in step S34, the high-precision convection discretization scheme adopts the QUICK (third-order upwind quadratic interpolation) discretization scheme, and the pressure-velocity coupled iterative algorithm adopts the SIMPLE (semi-implicit pressure coupling equations) algorithm.

[0050] As an optional implementation, step S4 involves selecting fault discrimination feature parameters from the spatiotemporal concentration dataset, constructing a fault identification feature vector, dividing the fault region, and generating a labeled sample set. This specifically includes the following steps: S41: Based on the spatiotemporal concentration dataset, the concentration-distance decay model combined with the dual monitoring point location formula is used to screen the fault discrimination feature parameters to obtain the screened feature parameters; the fault discrimination feature parameters include at least one of the following: concentration at a specified time of monitoring point, concentration ratio of multiple monitoring points, concentration rise rate, and gas diffusion uniformity time.

[0051] S42: Based on the filtered feature parameters, construct a fault identification feature vector.

[0052] S43: Based on the fault identification feature vector, the fault area is divided according to the fault location and a label is added to generate the labeled sample set.

[0053] As an optional implementation, in step S5, the SVM model uses a radial basis function (RBF) and optimizes the penalty parameter and kernel width parameter through cross-validation; the CNN-LSTM hybrid model uses a weighted hybrid loss function including classification cross-entropy loss, mechanism consistency loss and region smoothing loss for iterative optimization.

[0054] As an optional implementation, in step S5, the lightweight embedded branch is deployed on the field measurement and control terminal corresponding to the target high-voltage switchgear, and the cloud high-precision branch is deployed on the remote server corresponding to the target high-voltage switchgear. Furthermore, the field measurement and control terminal and the remote server are wirelessly connected.

[0055] The on-site monitoring and control terminal is used to collect the real-time gas concentration data, perform preliminary fault tracing and location based on the real-time gas concentration data, and output the local fault area prediction result; at the same time, it also uploads the real-time gas concentration data to the remote server.

[0056] The remote server is used to perform further fault tracing and location based on the real-time gas concentration data, outputting a cloud-based fault area prediction result. This cloud-based prediction result is compared and verified with the local fault area prediction result, enabling a secondary check of the fault area prediction result. Alternatively, the final fault area prediction result can be determined by combining the cloud-based and local prediction results, or the more accurate cloud-based prediction result can be directly used as the final fault area prediction result.

[0057] In an exemplary embodiment, to illustrate the technical solutions provided in the embodiments of this application in detail, the following are provided: Figure 3 The method for tracing and locating faults in high-voltage switchgear, as shown, specifically includes the following implementation steps: Step 1: Geometric modeling and boundary condition setting of high-voltage switchgear.

[0058] In this embodiment, based on the actual structural parameters of the high-voltage switchgear, a three-dimensional or two-dimensional geometric simulation model including components such as gas chamber, gas intake port, and internal guide rod is constructed. A mesh generation tool is used to perform mesh generation, and the reliability of the geometric simulation model is ensured through mesh independence verification.

[0059] In this embodiment, the boundary conditions are set as follows: the pressure in the gas chamber is 0.4 MPa (which meets the operating pressure of mainstream equipment), the ambient temperature is 300 K (close to the operating temperature of the equipment), the temperature at the fault point is set according to the fault type (e.g., the partial discharge fault is set to 700 K), the gas inlet is set to a closed state, external external force interference is ignored, and only the gas diffusion driven by molecular thermal motion and concentration difference is considered.

[0060] In this embodiment of the application, taking a 10kV high-voltage switchgear as an example, the gas chamber structure parameters are as follows: 1) Main body of the air chamber: cuboid, 800mm in diameter, 1000mm in length, with a volume of 0.5024m³. 3 .

[0061] 2) Air intake ports: 2, symmetrically located at the center of the bottom surface on both sides of the air chamber, with a diameter of 10mm and a length of 30mm.

[0062] 3) Internal components: Includes guide rods (100mm in diameter, made of copper) and insulating supports (made of epoxy resin). The model mesh was generated using the FLUENT Meshing tool, with a final mesh count of 51,271 and a node count of 287,623. The mesh quality is ≥0.844, which meets the simulation accuracy requirements.

[0063] In this embodiment, the basic simulation parameters are set as follows: 1) Air chamber pressure: 0.4 MPa (normal atmospheric pressure).

[0064] 2) Ambient temperature: 300K (27℃, normal operating conditions in substations).

[0065] 3) Fault point temperature: 700K (typical temperature for partial discharge faults).

[0066] 4) Characteristic component gases: Four relatively stable characteristic component gases, O3, NO, CO, and CO2, were selected, with an initial release amount of 8 × 10⁻⁶. -5 mol (typical gas production in a single partial discharge).

[0067] Step 2: Based on the geometric simulation model established in Step 1, conduct CFD numerical simulation of the fault characteristic gas at multiple fault locations to obtain spatiotemporal concentration data of multiple monitoring points corresponding to different fault locations.

[0068] In this embodiment, stable and easily detectable characteristic component gases (such as O3, NO, CO, and CO2) generated by air decomposition are selected as the research object. Based on computational fluid dynamics theory, the diffusion process is simulated using Fluent software. Fick's law and the FSG empirical formula are used as the theoretical basis, employing a laminar flow model (when Reynolds number < 2300) or a standard k-ε turbulence model (when Reynolds number ≥ 2300), combined with the QUICK discretization scheme and SIMPLE algorithm to ensure simulation accuracy. Using the controlled variable method, the effects of component type (difference in diffusion coefficient), initial gas quantity, and fault location (horizontal and vertical axes) on the diffusion effect are studied, and the concentration change curves at the gas intake at different times are recorded.

[0069] In this embodiment, the diffusion coefficients of four characteristic component gases in air (under the conditions of 300K and 0.4MPa) were calculated according to the FSG empirical formula. The calculation results of the diffusion parameters of each characteristic component gas are shown in Table 1 below.

[0070] Table 1 Calculation results of gas diffusion parameters for each characteristic component

[0071] Conclusion: The diffusion coefficient is negatively correlated with the diffusion uniformity time. O3 diffuses the fastest and CO2 diffuses the slowest, which is consistent with the simulation results.

[0072] For diffusion simulation at different fault locations, this application embodiment sets 5 fault points (coordinates based on a Cartesian coordinate system with the center of the gas chamber as the origin), and the location distribution parameters are shown in Table 2.

[0073] Table 2 Location distribution parameters of each fault point

[0074] Step 3: During the numerical simulation in Step 2, collect spatiotemporal concentration simulation data of fault characteristic gases at different fault locations, extract and screen fault characteristic parameters, and construct fault identification feature vectors.

[0075] In this embodiment, the feature parameters extracted from the simulation data include: the concentration values ​​of each feature component at the gas intake at different times, the rate of concentration increase, the peak concentration, and the diffusion uniformity time. For different fault locations, key feature parameters that can distinguish the fault area (such as the concentration combination at the gas intake at 230s or 270s) are selected to construct a feature parameter matrix, providing data support for fault tracing.

[0076] In this embodiment of the application, CO2 is used as an example. The concentration changes at each fault point at the gas intake ports A and B are recorded (the critical moment is 270s, when the concentration difference is the largest and the fault location is the easiest to distinguish). The concentration and concentration ratio results at each fault point are shown in Table 3.

[0077] Table 3. Concentration and concentration ratio results at the gas intake ports of each fault point.

[0078] In this embodiment of the application, “concentration at gas outlet A, concentration at gas outlet B, and concentration ratio (A / B) at 270s” are selected as the core feature parameters for fault tracing during feature extraction.

[0079] Step 4: Based on the multi-dimensional fault identification feature vectors extracted and constructed in Step 3, introduce machine learning algorithms to carry out research on equipment fault location and source tracing.

[0080] In this embodiment, when dividing the fault area, the gas chamber of the high-voltage switchgear is divided into several fault areas along the line connecting the gas inlets, each area corresponding to different distance characteristics. Specifically, the gas chamber is divided into 5 areas along the line connecting the gas inlets AB (corresponding to the areas where the 5 fault points are located), and 200 sets of simulation data (including interference data with different initial gas volumes and slight temperature fluctuations) are collected for each area, for a total of 1000 sets of samples. When dividing the data, 80% (800 sets) is used as the training set and 20% (200 sets) is used as the test set.

[0081] In this embodiment of the application, during model training, the SVM model and the CNN-LSTM hybrid model are used as intelligent classification models to be trained, the fault identification feature vector is used as the model input, and the corresponding fault area label is used as the output for model training.

[0082] In this embodiment, during model optimization, the SVM model uses the radial basis function kernel and optimizes the penalty parameter c and kernel variance g through cross-validation; the CNN-LSTM hybrid model uses the ReLU activation function and Nadam optimization algorithm, adjusts the number of iterations and the training set size to improve classification accuracy.

[0083] This application's embodiments are based on the "molecular diffusion theory" and the "convective diffusion theory" to establish a diffusion control equation for the fault characteristic gas, as detailed below: 1) Fick's First Law (steady-state diffusion).

[0084] The steady-state diffusion flux of the fault gas satisfies the following equation: .

[0085] in, This is the diffusion flux, expressed in mol / (m²). 2 ·s); The diffusion coefficient of gas molecules, in meters (m). 2 / s; This represents the gas concentration gradient; the negative sign indicates that the gas diffuses from a high concentration region to a low concentration region.

[0086] 2) Fick's Second Law (Transient Diffusion).

[0087] The transient relationship between gas concentration and time and space satisfies the following equation: .

[0088] in, The rate of change of concentration over time; is the Laplace operator for concentration, characterizing spatial concentration changes.

[0089] 3) Convection-diffusion coupling equation.

[0090] Considering the natural convection effect caused by the temperature field inside the switchgear, the transient diffusion process satisfies the following equation: .

[0091] in, This represents the air convection velocity field. This is the fault gas generation rate source term, used to characterize the gas generation process during a discharge fault.

[0092] 4) Calculation of molecular diffusion coefficient.

[0093] In this embodiment, the diffusion coefficient of the characteristic gas in air is calculated using the FSG empirical formula, which is: .

[0094] in, Components The diffusion coefficient in air, measured in meters (m). 2 / s; Temperature is thermodynamic temperature, in Kelvin (K). This represents the absolute pressure of the gas, measured in Pa. The molar mass of the characteristic gas is expressed in g / mol. The average molar mass of air is used in the embodiments of this application. =29g / mol; The characteristic gas molecule diffusion volume, in cm³. 3 / mol; The volume of air molecule diffusion is shown in the embodiments of this application. =20.1cm 3 / mol.

[0095] In the embodiments of this application, the characteristic gas molar masses of O3, CO, CO2, and NO are... With characteristic gas molecule diffusion volume as follows: O3: =48g / mol =26.9cm 3 / mol.

[0096] NO: =30g / mol =39.84cm 3 / mol.

[0097] CO: =28g / mol =46.67cm 3 / mol.

[0098] CO2: =44g / mol =54.80cm 3 / mol.

[0099] 5) Fault source tracing and location theory formula.

[0100] For the concentration-distance decay model, the gas concentration at the monitoring point and the fault distance satisfy an exponential decay relationship, expressed as follows: .

[0101] in, The concentration at the monitoring point at time t; The initial gas concentration at the fault point; The straight-line distance from the fault point to the monitoring point; This is the gas diffusion attenuation coefficient; The diffusion characteristic time constant is denoted as .

[0102] For the fault location formula with dual monitoring points, let the concentration at monitoring point one be... The distance from the fault point is The concentration at monitoring point two was The distance from the fault point is Then the concentration ratio and the distance difference satisfy the following formula: .

[0103] Taking the logarithm of the formula yields the fault distance difference, calculated as follows: .

[0104] By using real-time concentration data from dual monitoring points, the location of the fault can be determined, enabling precise source tracing.

[0105] In this embodiment, the diffusion control equation and the fault tracing formula provide physical mechanism support for the simulation of fault gas diffusion and theoretical basis and standardized sample source for machine learning models. Together, they constitute a complete technical system of "mechanism-driven simulation - data-trained model - intelligent tracing". First, this embodiment adopts Fick's first law, Fick's second law, convection-diffusion coupling equation and FSG empirical formula to establish a diffusion dynamic mechanism model of fault characteristic gas in the gas chamber of high-voltage switchgear. This model can accurately describe the molecular diffusion, convection transport and spatiotemporal concentration variation of fault gas in the gas chamber. By calculating the diffusion coefficient of different characteristic gases, numerical simulation and quantitative solution of the gas diffusion process are achieved. This provides strict physical mechanism and mathematical constraints for obtaining fault gas concentration data of different fault areas, different times and different gas intakes, ensuring that the simulation data is real, reliable and reproducible. Secondly, the concentration-distance decay model and dual-monitoring-point fault location formula established in this application theoretically reveal the mapping relationship between the fault location, fault distance, and gas intake concentration, clarifying that the fault location can be determined by the inversion of concentration features from the dual gas intakes. This provides a mechanistic criterion and location principle for fault tracing, and is the theoretical basis for fault area division and feature extraction. Furthermore, this application does not directly achieve location by analytically solving the diffusion control equation and tracing formula. Instead, it uses the aforementioned mechanistic formula as the underlying basis for data generation and feature construction. The diffusion control equation drives numerical simulation to generate multiple sets of fault gas concentration samples under different fault areas. Based on the concentration-location mapping law revealed by the tracing formula, key features are extracted and feature vectors and feature parameter matrices are constructed. The feature dataset generated and constrained by the physical mechanistic formula is then input into an SVM or CNN-LSTM hybrid model, enabling the model to automatically learn and fit the fault location-gas concentration-time evolution law contained in the diffusion equation and tracing formula. This achieves higher accuracy and stronger robustness in fault area identification and tracing location than traditional mechanistic formulas under complex structures and multi-interference conditions.

[0106] This application employs a dual-model parallel architecture, including an SVM model suitable for embedded real-time deployment and a CNN-LSTM hybrid model suitable for high-precision analysis in the cloud. Both models share the same feature dataset and constraint system generated by physical mechanisms. Fault source tracing uses the SVM model, and the optimal classification hyperplane is represented by the following equation: .

[0107] in, It is a radial basis kernel function, and ; The width parameter of the radial basis kernel function; The features include concentration, concentration ratio, and rate of rise, and the corresponding output is the fault area label. For symbolic functions, For Lagrange multipliers, The magnitude of the value represents the weight of the corresponding support vector in the classification result: The larger the value, the greater the impact of this support vector on the final fault region determination. For the first The true labels of each training sample This is the bias term, a constant (fixed after training), such as 0.23, -0.15, etc. Its function is to adjust the position of the classification hyperplane to ensure that the feature vectors of the five fault regions are correctly separated. For the first The feature vectors corresponding to each support vector.

[0108] The formula for calculating the accuracy of fault tracing is: .

[0109] in, Indicates the accuracy of fault tracing. It is a true positive. It is a true negative. It was a false positive. This is a false negative. The overall accuracy rate of fault tracing in the embodiments of this application is no less than 94.5%.

[0110] To address the spatiotemporal characteristics of fault gas concentration data, a CNN-LSTM hybrid model is employed. The CNN extracts spatial correlation features from different gas intakes, while the LSTM extracts the temporal evolution features of the concentration. The CNN-LSTM hybrid model sequentially comprises an input layer, two convolutional-pooling units, a flattening layer, an LSTM layer, a fully connected layer, and a Softmax output layer, outputting the probability distribution of each fault region.

[0111] This application's embodiments directly embed physical mechanisms into the training process, proposing a mechanism-constrained weighted hybrid loss function. It can be expressed as the following formula: .

[0112] in, Based on classification cross-entropy loss; To compensate for the loss of consistency in mechanistic data, a concentration-distance decay model is used to penalize predictions that violate physical laws. To smooth the loss at the regional boundary and ensure the continuity of concentration distribution in adjacent regions; , , The weight coefficients for the basic classification cross-entropy loss, mechanism consistency loss, and region boundary smoothing loss are determined through validation set optimization.

[0113] In this embodiment, a radial basis function kernel is used, and parameters are optimized through 5-fold cross-validation. The penalty parameter c has an optimal value of 10 (balancing classification accuracy and generalization ability). The kernel variance g has an optimal value of 0.1 (matching the dimension of the feature parameters). The number of iterations is 500 (model convergence, loss function stability).

[0114] The verification of the traceability results in this application embodiment is shown in Table 4.

[0115] Table 4 Test set validation results

[0116] As can be seen from Table 4, the overall accuracy of the model in tracing the source is 94.5%, which meets the actual needs of the project, and the tracing accuracy is even higher for the fault areas (areas 4 and 5) near the gas intake.

[0117] In the actual fault tracing and location prediction, this application embodiment monitors the concentration changes of each characteristic component gas at the gas intake port online, extracts fault identification feature vectors in real time, inputs them into a trained dual-branch intelligent classification model, and outputs the corresponding fault area prediction results, thereby achieving real-time and accurate fault tracing and providing clear location guidance for equipment maintenance, thus improving the overall efficiency of fault tracing and equipment maintenance.

[0118] This application's embodiments break through the limitations of traditional "static gas detection" and innovatively utilize the dynamic diffusion law of characteristic component gases in the high-voltage switchgear chamber (based on Fick's law + CFD simulation) to establish a correlation mapping of "fault location - diffusion process - concentration characteristics". It proposes a full-process technical solution for 10kV high-voltage switchgear, which includes "equipment geometric modeling → diffusion process simulation → key feature screening → intelligent classification and tracing". It does not require additional deployment of ultrasonic / ultra-high frequency sensors and can complete monitoring through existing gas intake ports, significantly reducing hardware costs and equipment modification difficulty. It enables real-time online fault tracing and shortens maintenance location time from "hours" to "minutes", solving the pain point of existing technologies being unable to accurately locate fault locations through gas data.

[0119] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores the actual structural data of the target high-voltage switchgear and real-time gas concentration data at monitoring points. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for tracing and locating faults in high-voltage switchgear.

[0120] Figure 4 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0121] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0122] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0123] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations and be authorized by the owner of the corresponding device.

[0124] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0125] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0126] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0127] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only intended to help understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for tracing and locating faults in high-voltage switchgear, characterized in that, include: Obtain the actual structural data of the target high-voltage switchgear; Based on the actual structural data, a geometric simulation model of the target high-voltage switchgear is established; Based on the geometric simulation model and combined with the gas diffusion mechanism equation, CFD numerical simulation of fault characteristic gas under multiple fault locations is carried out to obtain spatiotemporal concentration datasets of multiple monitoring points corresponding to different fault locations. The gas diffusion mechanism equations include molecular diffusion equations and convection-diffusion coupling equations; Fault discrimination feature parameters are selected from the spatiotemporal concentration dataset, fault identification feature vectors are constructed and fault regions are divided to generate a labeled sample set; A dual-branch intelligent classification model is trained based on the labeled sample set to obtain a trained dual-branch intelligent classification model. The dual-branch intelligent classification model is an intelligent classification model that takes the fault identification feature vector as input and the corresponding fault area prediction result as output for fault source tracing and localization prediction. The dual-branch intelligent classification model includes a lightweight embedded branch and a cloud-based high-precision branch. The lightweight embedded branch uses an SVM model, while the cloud-based high-precision branch uses a CNN-LSTM hybrid model. Real-time gas concentration data at the monitoring point of the target high-voltage switchgear is collected, fault identification feature vectors are extracted and input into the trained dual-branch intelligent classification model, and the corresponding fault area prediction results are output.

2. The fault tracing and location method for high-voltage switchgear according to claim 1, characterized in that, The fault characteristic gas is at least one of O3, NO, CO, and CO2 generated by the decomposition of insulating air under fault conditions of the target high-voltage switchgear; the monitoring point is a pre-set gas intake port of the target high-voltage switchgear.

3. The fault tracing and location method for high-voltage switchgear according to claim 1, characterized in that, Based on the aforementioned geometric simulation model and combined with the gas diffusion mechanism equation, CFD numerical simulations of the fault characteristic gas at multiple fault locations were conducted to obtain spatiotemporal concentration datasets for multiple monitoring points corresponding to different fault locations, specifically including: The geometric simulation model is meshed and its mesh independence is verified to obtain effective mesh parameters; Based on the effective mesh parameters and preset operating condition parameters, the simulation boundary conditions of the geometric simulation model are configured; the simulation boundary conditions include the working pressure of the equipment cavity, the ambient reference temperature, the fault point temperature, and the initial release amount of characteristic gas. Based on the simulation boundary conditions, the molecular diffusion coefficients of each fault characteristic gas are calculated using the FSG empirical formula. Based on the molecular diffusion coefficients of the characteristic gases of each fault, a laminar flow model or a standard k-ε turbulent flow model is selected according to the Reynolds number of the gas flow. A high-precision convection discretization scheme and a pressure-velocity coupled iterative algorithm are used to solve the diffusion control equations to obtain the full-time-series concentration raw data of a single fault location. The fault location coordinates are transformed using the controlled variable method, and all fault locations are simulated. The full-time-series raw concentration data of each monitoring point are continuously collected to form the spatiotemporal concentration dataset.

4. The fault tracing and location method for high-voltage switchgear according to claim 3, characterized in that, The high-precision convection discretization scheme adopts the QUICK discretization scheme, and the pressure-velocity coupled iterative algorithm adopts the SIMPLE algorithm.

5. The fault tracing and location method for high-voltage switchgear according to claim 1, characterized in that, The process involves selecting fault discrimination feature parameters from the spatiotemporal concentration dataset, constructing fault identification feature vectors, dividing fault regions, and generating a labeled sample set. Specifically, this includes: Based on the spatiotemporal concentration dataset, the concentration-distance decay model combined with the dual monitoring point location formula is used to screen fault discrimination feature parameters to obtain the screened feature parameters; the fault discrimination feature parameters include at least one of the following: concentration at a specified time of monitoring point, concentration ratio of multiple monitoring points, concentration rise rate, and gas diffusion uniformity time. Based on the filtered feature parameters, a fault identification feature vector is constructed. Based on the fault identification feature vector, the fault area is divided according to the fault location and labeled to generate the labeled sample set.

6. The fault tracing and location method for high-voltage switchgear according to claim 1, characterized in that, The SVM model uses a radial basis function kernel and optimizes the penalty parameter and kernel width parameter through cross-validation; the CNN-LSTM hybrid model uses a weighted hybrid loss function including classification cross-entropy loss, mechanism consistency loss and region smoothing loss for iterative optimization.

7. The fault tracing and location method for high-voltage switchgear according to claim 1, characterized in that, The lightweight embedded branch is deployed on the field measurement and control terminal corresponding to the target high-voltage switchgear, and the cloud high-precision branch is deployed on the remote server corresponding to the target high-voltage switchgear. The on-site monitoring and control terminal is used to collect the real-time gas concentration data, perform preliminary fault tracing and location based on the real-time gas concentration data, and output the local fault area prediction result; at the same time, it also uploads the real-time gas concentration data to the remote server. The remote server is used to perform fault source tracing and location again based on the real-time gas concentration data, and outputs cloud-based fault area prediction results, which can realize secondary verification of the fault area prediction results.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the high-voltage switchgear fault tracing and location method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fault tracing and location method for high-voltage switchgear as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the fault tracing and location method for high-voltage switchgear as described in any one of claims 1-7.

Images