Transformer Fault Location Method and System Based on Palladium Thin Film Hydrogen Sensor Array

Abstract

This invention relates to the field of power equipment condition monitoring and intelligent diagnosis technology, specifically to a transformer fault location method and system based on a palladium thin-film hydrogen sensor array. The system deploys a palladium thin-film resistive hydrogen sensor array at multiple key locations in the transformer tank to collect real-time dynamic changes in hydrogen concentration in the oil. A virtual transformer fault dataset generated from a digital twin model of the transformer's operating state is used for training using artificial intelligence algorithms. The trained model can intelligently analyze and match features from sensor signals collected during actual operation, thereby achieving rapid identification and spatial location of internal transformer faults. This invention, by integrating the high sensitivity of palladium thin-film resistive sensors, the simulation capabilities of a digital twin system, and the data-driven characteristics of artificial intelligence algorithms, achieves early warning, fuzzy location, and intelligent diagnosis of transformer faults, offering advantages such as fast response, high accuracy, and strong scalability.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and intelligent diagnosis technology, specifically to a transformer fault location method and system based on a palladium thin-film hydrogen sensor array. Background Technology

[0002] As a critical piece of equipment in the power system, the operating status of transformers directly affects the safety and reliability of the power grid. During long-term operation, transformers may experience internal faults due to insulation aging, partial discharge, winding overheating, oil arcing, and other reasons. These faults typically cause the decomposition of insulating oil or insulating materials, generating gases such as hydrogen, carbon monoxide, and methane. Among these, hydrogen is considered the most representative early fault indicator gas. Therefore, transformer fault monitoring methods based on hydrogen detection have been widely researched and applied.

[0003] In existing technologies, mainstream hydrogen detection methods include chromatographic gas analysis, optical sensing, and metal oxide semiconductor gas sensors. Among these, chromatographic methods offer high detection accuracy but require sampling and laboratory analysis, making online real-time monitoring difficult. Optical methods involve complex equipment and high costs, and suffer from insufficient stability in oil-based measurement environments. Metal oxide sensors respond well at high temperatures, but exhibit problems such as large drift, poor selectivity, and excessively high operating temperatures in transformer oil environments.

[0004] In recent years, palladium thin-film resistive hydrogen sensors have gradually become a research hotspot for dissolved gas detection in oil due to their high sensitivity to hydrogen, fast response, and room temperature operation. However, existing research mostly focuses on single-point concentration monitoring or trend analysis, lacking systematic analysis of the multi-point hydrogen diffusion characteristics and fault spatial distribution inside transformers, making it difficult to achieve spatial fault location.

[0005] On the other hand, traditional fault diagnosis methods based on single sensors or fixed models often rely on empirical thresholds or statistical regularities, failing to fully consider complex factors such as transformer structure, oil flow field distribution, and temperature gradients. The emerging digital twin technology can reproduce equipment operating states through virtual models, providing high-quality data support for fault simulation and algorithm training. Meanwhile, artificial intelligence algorithms, especially deep learning and reinforcement learning models, have significant advantages in pattern recognition and time-series data analysis, and can be used to learn hydrogen diffusion patterns and fault location characteristics from multi-point response signals.

[0006] Therefore, how to combine palladium thin-film resistive hydrogen sensor array technology with digital twin modeling and artificial intelligence algorithms to establish an intelligent detection system that can realize early identification and spatial fuzzy positioning of internal transformer faults has become an urgent technical problem to be solved in the field of intelligent transformer monitoring. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing transformer hydrogen detection technologies, such as low monitoring accuracy, inability to achieve spatial positioning, and reliance on experience for diagnostic results. This invention proposes a transformer fault location method and system based on a palladium thin-film hydrogen sensor array. By deploying a high-sensitivity palladium thin-film resistive hydrogen sensor array at multiple locations inside the transformer, and combining this with a virtual fault dataset generated from a digital twin model of the transformer's operating state, artificial intelligence algorithms are used to model and analyze the hydrogen diffusion time-series characteristics. This enables early identification, type judgment, and spatial fuzzy positioning of internal transformer faults, thereby improving the intelligence and reliability of fault detection.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A transformer fault location method based on a palladium thin-film hydrogen sensor array includes the following:

[0010] Palladium thin-film resistive hydrogen sensor arrays were deployed at multiple key locations on the transformer to collect hydrogen concentration response signals in real time.

[0011] The collected signals were time-calibrated and feature-extracted to obtain the arrival time series data of hydrogen response signals from multiple measurement points;

[0012] A digital twin model of the transformer's operating state was established to simulate the hydrogen diffusion process under different fault types and locations, in order to generate a virtual dataset for algorithm training.

[0013] The digital twin model of transformer operating status includes structural parameter model, oil flow characteristic model, thermal model, electric field model, fault model, and hydrogen diffusion and migration model.

[0014] The artificial intelligence algorithm is trained using the virtual dataset to form a fault location model;

[0015] The artificial intelligence algorithm includes at least one machine learning model or deep learning model, including support vector machine, convolutional neural network, temporal neural network, physical constraint reinforcement learning, and physical information neural network;

[0016] In actual transformer operation, the time series data of the palladium thin-film resistive hydrogen sensor signal collected in real time is input into the fault location model, and the spatial location result of the fault is output.

[0017] Furthermore, the method for simulating the hydrogen diffusion process under different fault types and locations to generate a virtual dataset for algorithm training is as follows:

[0018] Using the structural parameter model as the core, the actual physical parameters of the transformer are input to construct the geometric and physical foundation of the digital twin; at the same time, the material and dielectric parameter field functions are set in combination with the actual operation scenario to provide an initial state reference for subsequent simulation.

[0019] Fault models are used to characterize the equivalent hydrogen release source term under different fault types, locations, and durations. As the source of hydrogen diffusion;

[0020] The thermal model is used to solve for the internal temperature field T(x,t) of the output transformer, and to determine the temperature-dependent diffusion coefficient D(T,x) in the hydrogen diffusion model and / or the temperature-dependent intensity function g in the fault release model. s (E,T) is used to achieve coupled simulation of heat-fluid-diffusion;

[0021] The electric field model is used to solve the internal electric potential field of a transformer. The output and electric field strength E(x,t) are used in the fault model for the electrical stress-related gas release intensity function g. s (E,T) and / or fault type discrimination and localization training data generation;

[0022] The oil flow characteristic model simulates the flow velocity and direction of insulating oil and outputs oil flow data;

[0023] Based on data from other models, the hydrogen diffusion and migration model simulates the diffusion process of hydrogen from the fault point within the transformer, outputting a hydrogen concentration field. At each sensor location x i Samples are taken to form a virtual sensor response sequence to generate a virtual dataset for training artificial intelligence algorithms.

[0024] Furthermore, in the aforementioned structural parameter model, the structural parameter vector is:

[0025]

[0026] Where L, W, and H are the external dimensions of the fuel tank or cavity, {d j} represents characteristic dimensions such as oil passages / clearances, {A j} represents the equivalent cross-sectional area, V g x represents the gas space volume (or air gap size parameter). i Install coordinates for the i-th sensor;

[0027] Material and medium parameter field functions:

[0028] ρ(x,T),μ(x,T),k(x,T),x(x,T),D(x,T),ε(x,T),σ(x,T), where each of the above parameters corresponds to density, viscosity, thermal conductivity, specific heat, diffusion coefficient, dielectric constant, and electrical conductivity, respectively.

[0029] Furthermore, the oil flow characteristic model includes:

[0030] (1) Three-dimensional incompressible fluid (CFD) model

[0031] Oil flow velocity field u(x,t) and pressure p(x,t):

[0032]

[0033] In the formula, ρ(T) is the density of the oil, μ(T) is the viscosity of the oil, and is allowed to vary with temperature; g is the gravitational acceleration vector; F pump This is the equivalent pump source term for forced oil circulation;

[0034] (2) The Boussinesq approximation can be used for natural cycles:

[0035] ρ(T)≈ρ0[1-β(T-T0)]

[0036] If turbulence is considered, a closed turbulence model can be added.

[0037] Alternatively, a simplified oil passage network pressure drop-flow rate model can be used, which discretizes the oil passage into several segments, with a flow rate Q. j satisfy:

[0038] Δp j =R j Q j ,R j = f(μ, channel geometry)

[0039] And use thermal buoyancy or pump pressure as the driving source.

[0040] Furthermore, the thermal model is used to solve for the internal temperature field T(x,t) of the transformer, where x is the spatial coordinate and t is time; in the oil domain Ω oil The internal energy equation containing convection is satisfied:

[0041]

[0042] Where ρ(x,T) is the density of the oil medium, c(x,T) is the specific heat capacity of the oil medium, k(x,T) is the thermal conductivity of the oil medium, u(x,t) is the oil flow velocity field output by the oil flow characteristic model, and Q loss (x,t) represents the volume heat source term corresponding to normal operating losses, Q f (x,t) represents the fault-related heat source term; and is located at the outer boundary of the transformer. Convection heat transfer boundary conditions are applied at this location:

[0043]

[0044] Where n is the outward normal unit vector of the outer boundary, h is the convective heat transfer coefficient, and T ∞ The ambient temperature is used; and the temperature field T(x,t) output by the thermal model is used to determine the temperature-dependent diffusion coefficient D(T,x) in the hydrogen diffusion model and / or the temperature-dependent intensity function g in the fault release model. s (E,T) is used to achieve coupled simulation of heat-flow-diffusion.

[0045] Furthermore, the aforementioned electric field model is used to solve for the internal electric potential field of the transformer. The electric field intensity E(x,t), where x is the spatial coordinate and t is time; the electric field model satisfies the quasi-static potential control equation:

[0046]

[0047] Where ε(x,T) is the dielectric constant parameter field and can vary with space and temperature T(x,t); at the high-voltage electrode boundary Γ HV Apply potential boundary conditions at the location At the grounding electrode boundary Γ G Apply potential boundary conditions at the location Where V(t) is the applied voltage; and the conditions of continuous potential and continuous normal electric displacement are satisfied at the multi-dielectric interface. and Where n is the unit vector normal to the interface; E(x,t) output by the electric field model is used as the electric stress-related gas release intensity function g in the fault model. s (E,T) and / or fault type discrimination and location training data generation.

[0048] Furthermore, the aforementioned fault model is used to characterize the equivalent hydrogen release source term under different fault types, locations, and durations. Where x is the spatial coordinate and t is time; the hydrogen release source term satisfies:

[0049]

[0050] Where s is the fault type identifier and s∈{thermal fault, partial discharge, electric arc…}, A s g is the intensity coefficient related to the fault type. s (E,T) represents the hydrogen release intensity function related to the fault type, E(x,t) is the electric field intensity output by the electric field model, and T(x,t) is the temperature field output by the thermal model. f The location is the fault center, and δ is the fault impact scale parameter; the location distribution function ω(x; x) f ,δ) is used to characterize the source term space weights, and can take the following values:

[0051]

[0052] The time-gated function u(t; t0, τ) is used to characterize the source term start-stop control, and can take the following values:

[0053] u(t;t0,τ)=H(t-t0)-H(t-t0-τ)

[0054] Where t0 is the fault initiation time, τ is the fault duration, and H(·) is the Heaviside step function.

[0055] Furthermore, the aforementioned hydrogen diffusion and migration model is used to solve for the Ω value in the oil region. oil Internal hydrogen concentration field Where x is the spatial coordinate and t is time; the hydrogen diffusion and migration model satisfies the convection-diffusion-source-sink control equations:

[0056]

[0057] Where u(x,t) is the oil flow velocity field output by the oil flow characteristic model, and D(x,T) is the diffusion coefficient of hydrogen in the oil, which can vary with space and temperature T(x,t). The hydrogen release source term is given by the fault model, where λ is the equivalent decay coefficient used to characterize the consumption effects of hydrogen being adsorbed, dissolved, or recombinated; and at the sealed boundary Γ seal Zero flux boundary conditions are applied at this point:

[0058]

[0059] Where n is the unit normal vector outside the boundary; and at the oil-gas interface Γ og Mass transfer boundary conditions are applied at this point:

[0060]

[0061] Where k m C is the mass transfer coefficient. eq (T,C g The relationship between temperature T and the concentration of hydrogen in the interfacial gas phase C is... g The determined equilibrium concentration; the output of the hydrogen diffusion and migration model. At each sensor location x i Samples are taken to form a virtual sensor response sequence to generate a virtual dataset for training artificial intelligence algorithms.

[0062] A transformer fault location system based on a palladium thin-film hydrogen sensor array includes:

[0063] Sensor array module: Installed at multiple preset monitoring locations in the transformer oil tank or gas space, the sensor is a palladium thin-film resistive hydrogen sensor, used to detect changes in hydrogen concentration in transformer oil in real time and output a resistance signal;

[0064] Signal acquisition and synchronization module: used to synchronously acquire and time-calibrate the output signals of each sensor to obtain the arrival time sequence of hydrogen response;

[0065] Digital twin simulation module: Based on transformer structure, oil flow characteristics and fault model, establish a digital twin model of transformer working state, including but not limited to structural parameter model, oil flow characteristic model, thermal model, electric field model, fault model and hydrogen diffusion and migration model, to simulate the hydrogen diffusion process under different fault types and locations, so as to generate a virtual dataset for algorithm training.

[0066] Artificial intelligence algorithm module: Uses the virtual dataset for supervised learning training to form a fault location prediction model;

[0067] Fault location module: During the actual operation of the transformer, it receives real-time response signals from various sensors, inputs them into the trained artificial intelligence model, and outputs fuzzy location results of the fault location.

[0068] The palladium thin-film resistive hydrogen sensor includes a palladium or palladium alloy thin-film hydrogen resistance wire assembly and a temperature control or compensation assembly, used to achieve high sensitivity and fast response hydrogen detection.

[0069] The digital twin model of transformer operating status can simulate the diffusion rate and spatiotemporal distribution characteristics of hydrogen in oil under different operating conditions; the artificial intelligence algorithm module can learn the mapping relationship between hydrogen diffusion path and fault source location based on the temporal characteristics and concentration gradient information of sensor signals; the fault location result is output in the form of spatial probability distribution, which can be combined with historical data to realize fault trend prediction and health assessment.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] Achieving spatial fuzzy localization of internal transformer faults: By analyzing the hydrogen response time difference and concentration characteristics at multiple measurement points, the fault location can be quickly determined;

[0072] Combining digital twins with artificial intelligence algorithms improves detection accuracy: the virtual model provides rich training samples, enabling the artificial intelligence algorithm to have strong generalization ability and fault identification ability;

[0073] Highly sensitive and fast response: Employing a palladium thin-film resistive sensor, it can achieve high-sensitivity hydrogen detection at relatively low temperatures, making it suitable for complex environments in oil.

[0074] The system is highly scalable: the sensor layout and model parameters can be flexibly adjusted according to different transformer structures and capacities;

[0075] It has early warning and predictive maintenance functions: Through continuous learning of historical trends by AI models, potential failure risks can be identified in advance.

[0076] This invention integrates hydrogen sensing and detection technology, digital twin modeling technology, machine learning and reinforcement learning algorithms, and other interdisciplinary approaches to achieve early identification, fault type determination and spatial fuzzy location of internal transformer faults, belonging to the field of intelligent monitoring and predictive maintenance of power systems.

[0077] This invention achieves intelligent detection and spatial positioning of internal transformer faults through the organic integration of sensor arrays, digital twins, and artificial intelligence algorithms, and has broad engineering application value and promotion prospects. Attached Figure Description

[0078] Figure 1 A schematic diagram of a transformer fault location system based on a palladium thin-film hydrogen sensor array is shown.

[0079] Numbering on the map:

[0080] 1. Palladium thin-film resistive hydrogen sensor array module; 2. Signal acquisition and synchronization module; 3. Digital twin simulation module; 4. Artificial intelligence algorithm module; 5. Fault location block.

[0081] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Detailed Implementation

[0082] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. Although the drawings and embodiments show specific embodiments of the invention, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0083] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this invention is determined by the appended claims.

[0084] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front," and "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship in the working state of this invention, and are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. Furthermore, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0085] Example 1

[0086] like Figure 1 As shown.

[0087] A transformer fault location system based on a palladium thin-film hydrogen sensor array, characterized in that it includes:

[0088] Palladium thin-film resistive hydrogen sensor array module 1: Installed at multiple preset monitoring locations in transformer oil tank or gas space, the sensor is a palladium thin-film resistive hydrogen sensor, used to detect changes in hydrogen concentration in transformer oil in real time and output resistance signal;

[0089] Signal acquisition and synchronization module 2: used to synchronously acquire and time-calibrate the output signals of each sensor to obtain the arrival time sequence of hydrogen response;

[0090] Digital Twin Simulation Module 3: Based on the transformer structure, oil flow characteristics and fault model, establish a digital twin model of the transformer's working state, including but not limited to structural parameter model, oil flow characteristic model, thermal model, electric field model, fault model and hydrogen diffusion and migration model, to simulate the hydrogen diffusion process under different fault types and locations, so as to generate a virtual dataset for algorithm training.

[0091] Artificial intelligence algorithm module 4: Uses the virtual dataset for supervised learning training to form a fault location prediction model;

[0092] Fault location module 5: During the actual operation of the transformer, it receives the real-time response signals from each sensor, inputs them into the trained artificial intelligence model, and outputs the fuzzy location result of the fault location.

[0093] The palladium thin-film resistive hydrogen sensor array module 1 is arranged in multiple typical locations within the transformer tank or gas space, such as near the windings, core, oil surface, and radiator. Each sensor includes a palladium or palladium alloy thin-film electrode layer, a substrate layer, and a temperature-compensated resistor layer. The palladium thin film undergoes adsorption and diffusion reactions in the presence of hydrogen, resulting in a change in resistance; the resistance signal output by the sensor corresponds to the hydrogen concentration. To reduce the influence of temperature on resistance changes, this embodiment incorporates a temperature compensation circuit in each sensor, obtaining a high-precision hydrogen response signal through differential measurement. Through the multi-point layout of the sensor array, dynamic change curves of hydrogen concentration at multiple locations within the transformer can be acquired simultaneously, providing timing data support for subsequent positioning algorithms.

[0094] The signal acquisition and synchronization module 2 acquires the resistance output signals of all sensors in parallel and achieves millisecond-level time alignment through a time synchronization mechanism (such as a timestamp or a high-precision clock). The acquired signals are filtered and standardized to form a multi-channel hydrogen concentration response matrix. By analyzing the arrival time of the response signals, the propagation delay of hydrogen diffusion at different locations can be calculated, thereby inferring the spatial relationship between the fault source and each sensor.

[0095] The digital twin simulation module 3 is primarily used to obtain high-quality training samples. In this embodiment, a digital twin model is established based on the transformer's structural parameters, oil flow distribution, and heat conduction characteristics. This model utilizes computational fluid dynamics and finite element analysis to simulate the diffusion process of hydrogen inside the transformer under different fault types (such as partial discharge, winding overheating, and arc discharge) and locations. The output includes data on the evolution of the hydrogen concentration field over time, corresponding to the sensor array placement points. By changing the fault location and energy intensity, multiple sets of virtual samples can be generated for subsequent training of artificial intelligence algorithms.

[0096] The artificial intelligence algorithm module 4 establishes a fault location prediction model based on digital twin simulation data. In this embodiment, the module's data preprocessing part normalizes and interpolates the simulated and measured signals to form a unified feature format; the module's feature extraction part extracts features such as arrival time difference, peak concentration change rate, and diffusion gradient from multiple measurement points; the module's model training part uses machine learning or deep learning algorithms for training, which may include: convolutional neural networks for learning spatiotemporal features; long short-term memory networks for handling time series dependencies; and reinforcement learning-based policy networks for optimizing parameters by minimizing the location error reward function. The model validation and optimization part calibrates and adaptively optimizes the model using test samples generated by the digital twin and some actual operating data to reduce errors caused by environmental differences.

[0097] During actual transformer operation, the fault location module 5 receives hydrogen response signals from multiple measurement points in real time and inputs them into a trained fault location model. The model outputs a fault spatial distribution probability map based on the input features to determine the most likely fault area. The location results can be projected onto a two-dimensional or three-dimensional coordinate system in conjunction with transformer structural parameters for intuitive viewing by maintenance personnel. Simultaneously, the system can determine the degree of fault development based on the hydrogen response rate and its changing trend.

[0098] Example 1:

[0099] This embodiment provides a hydrogen monitoring system for oil-immersed transformers, which enables real-time detection of internal transformer faults and concentration trend analysis.

[0100] Five key locations were selected inside the oil tank of a 110kV oil-immersed transformer to deploy an array of palladium thin-film resistive hydrogen sensors, including the upper part of the winding, the middle of the core, the bottom oil layer, near the oil surface, and the oil outlet. Each sensor uses a palladium-silver alloy thin-film electrode with a film thickness of approximately 100nm, an operating temperature range of -20℃ to 80℃, and a response time of less than 10s.

[0101] The sensor output resistance signal is converted into a hydrogen concentration value by the signal acquisition module, with a sampling frequency of 1Hz. The system achieves multi-channel synchronous acquisition through a high-precision clock module.

[0102] When partial discharge or overheating occurs inside the transformer, the hydrogen concentration in the oil rises, which is first detected by the sensor at the nearest fault point, and its response time is earlier than that of other measuring points.

[0103] The signal acquisition module uploads the response time and concentration change rate of each sensor to the data processing unit. By analyzing the temporal differences in the hydrogen concentration change curves, it is possible to determine whether a fault has occurred and to preliminarily estimate the approximate location of the fault.

[0104] This embodiment achieves multi-point synchronous monitoring of hydrogen concentration in transformer oil through a simple sensor array arrangement. Compared with traditional single-point detection, it can effectively improve fault identification sensitivity and response speed, and provide basic data for subsequent intelligent positioning algorithms.

[0105] Example 2:

[0106] A transformer fault location method based on a palladium thin-film hydrogen sensor array includes the following:

[0107] Palladium thin-film resistive hydrogen sensor arrays were deployed at multiple key locations on the transformer to collect hydrogen concentration response signals in real time.

[0108] The collected signals were time-calibrated and feature-extracted to obtain the arrival time series data of hydrogen response signals from multiple measurement points;

[0109] A digital twin model of the transformer's operating state was established to simulate the hydrogen diffusion process under different fault types and locations, in order to generate a virtual dataset for algorithm training.

[0110] The digital twin model of transformer operating status includes structural parameter model, oil flow characteristic model, thermal model, electric field model, fault model, and hydrogen diffusion and migration model.

[0111] The artificial intelligence algorithm is trained using the virtual dataset to form a fault location model;

[0112] The artificial intelligence algorithm includes at least one machine learning model or deep learning model, including support vector machine, convolutional neural network, temporal neural network, physical constraint reinforcement learning, and physical information neural network;

[0113] In actual transformer operation, the real-time acquired palladium thin-film resistive hydrogen sensor signal is input into the fault location model, and the spatial location result of the fault is output.

[0114] Based on a digital twin model of transformer operating conditions, the hydrogen diffusion process under different fault types and locations is simulated to generate a virtual dataset for algorithm training. The method is as follows:

[0115] Using the structural parameter model as the core, the actual physical parameters of the transformer are input to construct the geometric and physical foundation of the digital twin; at the same time, the material and dielectric parameter field functions are set in combination with the actual operation scenario to provide an initial state reference for subsequent simulation.

[0116] Fault models are used to characterize the equivalent hydrogen release source term under different fault types, locations, and durations. As the source of hydrogen diffusion;

[0117] The thermal model is used to solve for the internal temperature field T(x,t) of the output transformer, and to determine the temperature-dependent diffusion coefficient D(T,x) in the hydrogen diffusion model and / or the temperature-dependent intensity function g in the fault release model. s (E,T) is used to achieve coupled simulation of heat-fluid-diffusion;

[0118] The electric field model is used to solve the internal electric potential field of a transformer. The output and electric field strength E(x,t) are used in the fault model for the electrical stress-related gas release intensity function g. s (E,T) and / or fault type discrimination and localization training data generation;

[0119] The oil flow characteristic model simulates the flow velocity and direction of insulating oil and outputs oil flow data;

[0120] Based on data from other models, the hydrogen diffusion and migration model simulates the diffusion process of hydrogen from the fault point within the transformer, outputting a hydrogen concentration field. At each sensor location x i Samples are taken to form a virtual sensor response sequence to generate a virtual dataset for training artificial intelligence algorithms.

[0121] In the aforementioned structural parameter model, the structural parameter vector is:

[0122]

[0123] Where L, W, and H are the external dimensions of the fuel tank or cavity, {d j} represents characteristic dimensions such as oil passages / clearances, {A j} represents the equivalent cross-sectional area, V g x represents the gas space volume (or air gap size parameter). i Install coordinates for the i-th sensor;

[0124] Material and medium parameter field functions:

[0125] ρ(x,T),μ(x,T),k(x,T),c(x,T),D(x,T),ε(x,T),σ(x,T), where each of the above parameters corresponds to density, viscosity, thermal conductivity, specific heat, diffusion coefficient, dielectric constant, and electrical conductivity, respectively.

[0126] The oil flow characteristic model described above is implemented in two steps;

[0127] (1) Three-dimensional incompressible fluid (CFD) model

[0128] Oil flow velocity field u(x,t) and pressure p(x,t):

[0129]

[0130] In the formula, ρ(T) is the density of the oil, μ(T) is the viscosity of the oil, and is allowed to vary with temperature; g is the gravitational acceleration vector; F pump This is the equivalent pump source term for forced oil circulation;

[0131] (2) The Boussinesq approximation can be used for natural cycles:

[0132] ρ(T)≈ρ0[1-β(T-T0)]

[0133] If turbulence is considered, a closed turbulence model can be added.

[0134] Alternatively, a simplified oil passage network pressure drop-flow rate model can be used, which discretizes the oil passage into several segments, with a flow rate Q. j satisfy:

[0135] Δp j =R j Q j ,R j = f(μ, channel geometry)

[0136] And use thermal buoyancy or pump pressure as the driving source.

[0137] The thermal model is used to solve for the internal temperature field T(x,t) of the transformer, where x is the spatial coordinate and t is time; in the oil domain Ω oil The internal energy equation containing convection is satisfied:

[0138]

[0139] Where ρ(x,T) is the density of the oil medium, c(x,T) is the specific heat capacity of the oil medium, k(x,T) is the thermal conductivity of the oil medium, u(x,t) is the oil flow velocity field output by the oil flow characteristic model, and Q loss (x,t) represents the volume heat source term corresponding to normal operating losses, Q f (x,t) represents the fault-related heat source term; and is located at the outer boundary of the transformer. Convection heat transfer boundary conditions are applied at this location:

[0140]

[0141] Where n is the outward normal unit vector of the outer boundary, h is the convective heat transfer coefficient, and T ∞ The ambient temperature is used; and the temperature field T(x,t) output by the thermal model is used to determine the temperature-dependent diffusion coefficient D(T,x) in the hydrogen diffusion model and / or the temperature-dependent intensity function g in the fault release model. s (E,T) is used to achieve coupled simulation of heat-flow-diffusion.

[0142] The electric field model described above is used to solve for the internal electric potential field of a transformer. The electric field intensity E(x,t), where x is the spatial coordinate and t is time; the electric field model satisfies the quasi-static potential control equation:

[0143]

[0144] Where ε(x,T) is the dielectric constant parameter field and can vary with space and temperature T(x,t); at the high-voltage electrode boundary Γ HV Apply potential boundary conditions at the location At the grounding electrode boundary Γ G Apply potential boundary conditions at the location Where V(t) is the applied voltage; and the conditions of continuous potential and continuous normal electric displacement are satisfied at the multi-dielectric interface. and Where n is the unit vector normal to the interface; E(x,t) output by the electric field model is used as the electric stress-related gas release intensity function g in the fault model. s (E,T) and / or fault type discrimination and location training data generation.

[0145] The aforementioned fault model is used to characterize the equivalent hydrogen release source term under different fault types, locations, and durations. Where x is the spatial coordinate and t is time; the hydrogen release source term satisfies:

[0146]

[0147] Where s is the fault type identifier and s∈{thermal fault, partial discharge, electric arc…}, A s g is the intensity coefficient related to the fault type. s (E,T) represents the hydrogen release intensity function related to the fault type, E(x,t) is the electric field intensity output by the electric field model, and T(x,t) is the temperature field output by the thermal model. f The location is the fault center, and δ is the fault impact scale parameter; the location distribution function ω(x; x) f ,δ) is used to characterize the source term space weights, and can take the following values:

[0148]

[0149] The time-gated function u(t; t0, τ) is used to characterize the source term start-stop control, and can take the following values:

[0150] u(t;t0,τ)=H(t-t0)-H(t-t0-τ)

[0151] Where t0 is the fault initiation time, τ is the fault duration, and H(·) is the Heaviside step function.

[0152] The hydrogen diffusion and migration model described above is used to solve for the Ω value in the oil region. oil Internal hydrogen concentration field Where x is the spatial coordinate and t is time; the hydrogen diffusion and migration model satisfies the convection-diffusion-source-sink control equations:

[0153]

[0154] Where u(x,t) is the oil flow velocity field output by the oil flow characteristic model, and D(x,T) is the diffusion coefficient of hydrogen in the oil, which can vary with space and temperature T(x,t). The hydrogen release source term is given by the fault model, where λ is the equivalent decay coefficient used to characterize the consumption effects of hydrogen being adsorbed, dissolved, or recombinated; and at the sealed boundary Γ seal Zero flux boundary conditions are applied at this point:

[0155]

[0156] Where n is the unit normal vector outside the boundary; and at the oil-gas interface Γ og Mass transfer boundary conditions are applied at this point:

[0157]

[0158] Where k m C is the mass transfer coefficient. eq (T,C g The relationship between temperature T and the concentration of hydrogen in the interfacial gas phase C is... g The determined equilibrium concentration; the output of the hydrogen diffusion and migration model. At each sensor location x i Samples are taken to form a virtual sensor response sequence to generate a virtual dataset for training artificial intelligence algorithms.

[0159] A digital twin model of the transformer's operating state was established based on its structural parameters, oil flow distribution, and heat conduction characteristics. Computational fluid dynamics (CFD) was used to simulate the diffusion process of hydrogen in the oil under different fault locations and energy conditions, outputting a simulated concentration field. By changing the fault location (e.g., middle of the winding, edge of the core, bottom) and energy level, 1000 sets of virtual samples were generated for algorithm training.

[0160] Next, the artificial intelligence algorithm uses a Long Short-Term Memory (LSTM) network model as its core algorithm. Input features include: the arrival time difference of hydrogen response at multiple measurement points, the concentration change rate at each measurement point, and the time interval between concentration peaks. The model output is the probability distribution of the fault location coordinates or a region label.

[0161] During training, data generated by digital twins is used as the training set, and on-site operational data is used as the validation set. The model parameters are optimized using the backpropagation algorithm, enabling the model to accurately locate the fault area under different operating conditions.

[0162] In actual testing, when a partial discharge fault is simulated inside the transformer, the AI ​​model, based on the real-time signal input from the sensor array, can provide a prediction of the fault area within approximately 30 seconds.

[0163] The results show that, relative to the overall size of the transformer, the positioning error is less than 5%, which is significantly better than traditional ultrasonic positioning methods.

[0164] This embodiment illustrates that by introducing digital twin-generated training data and combining it with deep learning algorithms, high-precision fault location can be achieved without relying on a large number of real fault samples, significantly improving the system's intelligence level and reliability.

Claims

1. A transformer fault location method based on a palladium thin-film hydrogen sensor array, characterized in that, Includes the following: Palladium thin-film resistive hydrogen sensor arrays were deployed at multiple key locations on the transformer to collect hydrogen concentration response signals in real time. The collected signals were time-calibrated and feature-extracted to obtain the arrival time series data of hydrogen response signals from multiple measurement points; A digital twin model of the transformer's operating state was established to simulate the hydrogen diffusion process under different fault types and locations, in order to generate a virtual dataset for algorithm training. The digital twin model of transformer operating status includes structural parameter model, oil flow characteristic model, thermal model, electric field model, fault model, and hydrogen diffusion and migration model. The artificial intelligence algorithm is trained using the virtual dataset to form a fault location model; The artificial intelligence algorithm includes at least one machine learning model or deep learning model, including support vector machine, convolutional neural network, temporal neural network, physical constraint reinforcement learning, and physical information neural network; In actual transformer operation, the time series data of the palladium thin-film resistive hydrogen sensor signal collected in real time is input into the fault location model, and the spatial location result of the fault is output.

2. The transformer fault location method based on a palladium thin-film hydrogen sensor array according to claim 1, characterized in that, The method for simulating hydrogen generation and diffusion processes under different fault types and locations to generate a virtual dataset for algorithm training is as follows: Using the structural parameter model as the core, the actual physical parameters of the transformer are input to construct the geometric and physical foundation of the digital twin; at the same time, the material and dielectric parameter field functions are set in combination with the actual operation scenario to provide an initial state reference for subsequent simulation. Fault models are used to characterize the equivalent hydrogen release source term under different fault types, locations, and durations. As the source of hydrogen diffusion; The thermal model is used to solve for the internal temperature field T(x,t) of the output transformer, and to determine the temperature-dependent diffusion coefficient D(T,x) in the hydrogen diffusion model and / or the temperature-dependent intensity function g in the fault release model. s (E,T) is used to achieve coupled simulation of heat-fluid-diffusion; The electric field model is used to solve the internal electric potential field of a transformer. The output and electric field strength E(x,t) are used in the fault model for the electrical stress-related gas release intensity function g. s (E,R) and / or fault type discrimination and localization training data generation; The oil flow characteristic model simulates the flow velocity and direction of insulating oil and outputs oil flow data; Based on data from other models, the hydrogen diffusion and migration model simulates the generation of hydrogen and its diffusion from the fault point within the transformer, outputting a hydrogen concentration field. At each sensor location x i Samples are taken to form a virtual sensor response sequence to generate a virtual dataset for training artificial intelligence algorithms.

3. The transformer fault location method based on a palladium thin-film hydrogen sensor array according to claim 1 or 2, characterized in that, The structural parameter model described above has the following structural parameter vector: Where l, W, and H are the external dimensions of the fuel tank or cavity, {d j } represents characteristic dimensions such as oil passages / clearances, {A j } represents the equivalent cross-sectional area, V g x represents the gas space volume (or air gap size parameter). i Install coordinates for the i-th sensor; Material and medium parameter field functions: ρ(x,T),μ(x,T),k(x,T),c(x,T),D(x,T),ε(x,T),σ(x,T), where each of the above parameters corresponds to density, viscosity, thermal conductivity, specific heat, diffusion coefficient, dielectric constant, and electrical conductivity, respectively.

4. The transformer fault location method based on a palladium thin-film hydrogen sensor array according to claim 1 or 2, characterized in that, The oil flow characteristic model includes: (1) Three-dimensional incompressible fluid (CFD) model Oil flow velocity field u(x,t) and pressure p(x,t): In the formula, ρ(T) is the density of the oil, μ(T) is the viscosity of the oil, and is allowed to vary with temperature; g is the gravitational acceleration vector; F pump This is the equivalent pump source term for forced oil circulation; (2) The Boussinesq approximation can be used for natural cycles: ρ(T)≈ρ0[1-β(T-T0)] If turbulence is considered, a closed turbulence model can be added. Alternatively, a simplified oil passage network pressure drop-flow rate model can be used, which discretizes the oil passage into several segments, with a flow rate Q. j satisfy: Δp j =R j Q j ,R j = f(μ, channel geometry) And use thermal buoyancy or pump pressure as the driving source.

5. The transformer fault location method based on a palladium thin-film hydrogen sensor array according to claim 1 or 2, characterized in that, The aforementioned thermal model is used to solve for the internal temperature field T(x,t) of the transformer, where x is the spatial coordinate and t is time; in the oil domain Ω oil The internal energy equation containing convection is satisfied: Where ρ(x,T) is the density of the oil medium, c(x,T) is the specific heat capacity of the oil medium, k(x,T) is the thermal conductivity of the oil medium, u(x,t) is the oil flow velocity field output by the oil flow characteristic model, and Q loss (x,t) represents the volume heat source term corresponding to normal operating losses, Q f (x,t) represents the fault-related heat source term; and is located at the outer boundary of the transformer. Convection heat transfer boundary conditions are applied at this location: Where n is the outward normal unit vector of the outer boundary, h is the convective heat transfer coefficient, and T ∞ The ambient temperature is used; and the temperature field T(x,t) output by the thermal model is used to determine the temperature-dependent diffusion coefficient D(T,x) in the hydrogen diffusion model and / or the temperature-dependent intensity function g in the fault release model. s (E,T) is used to achieve coupled simulation of heat-flow-diffusion.

6. The transformer fault location method based on a palladium thin-film hydrogen sensor array according to claim 1 or 2, characterized in that, The electric field model described above is used to solve for the internal electric potential field of a transformer. The electric field strength E(x,t) is given by x, where x is the spatial coordinate and t is the time. The electric field model satisfies the quasi-static potential control equation: Where ε(x,T) is the dielectric constant parameter field and can vary with space and temperature T(x,t); at the high-voltage electrode boundary Γ HV Apply potential boundary conditions at the location At the grounding electrode boundary Γ G Apply potential boundary conditions at the location Where V(t) is the applied voltage; and the conditions of continuous potential and continuous normal electric displacement are satisfied at the multi-dielectric interface. and Where n is the unit vector normal to the interface; E(x,t) output by the electric field model is used as the electric stress-related gas release intensity function g in the fault model. s (E,T) and / or fault type discrimination and location training data generation.

7. The transformer fault location method based on a palladium thin-film hydrogen sensor array according to claim 1 or 2, characterized in that, The aforementioned fault model is used to characterize the equivalent hydrogen release source term under different fault types, locations, and durations. Where x is the spatial coordinate and t is time; the hydrogen release source term satisfies: Where s is the fault type identifier and s∈{thermal fault, partial discharge, electric arc…}, A s g is the intensity coefficient related to the fault type. s (E,T) represents the hydrogen release intensity function related to the fault type, E(x,t) is the electric field intensity output by the electric field model, and T(x,t) is the temperature field output by the thermal model. f The location is the fault center, and δ is the fault impact scale parameter; the location distribution function ω(x; x) f ,δ) is used to characterize the source term space weights, and can take the following values: The time-gated function u(t; t0, τ) is used to characterize the source term start-stop control, and can take the following values: u(t;t0,τ)=H(t-t0)-H(t-t0-τ) Where t0 is the fault initiation time, τ is the fault duration, and H(·) is the Heaviside step function.

8. The transformer fault location method based on a palladium thin-film hydrogen sensor array according to claim 1 or 2, characterized in that, The hydrogen diffusion and migration model described above is used to solve for the Ω value in the oil region. oil Internal hydrogen concentration field Where x is the spatial coordinate and t is time; the hydrogen diffusion and migration model satisfies the convection-diffusion-source-sink control equations: Where u(x,t) is the oil flow velocity field output by the oil flow characteristic model, and D(x,T) is the diffusion coefficient of hydrogen in the oil, which can vary with space and temperature T(x,t). The hydrogen release source term is given by the fault model, where λ is the equivalent decay coefficient used to characterize the consumption effects of hydrogen being adsorbed, dissolved, or recombinated; and at the sealed boundary Γ seal Zero flux boundary conditions are applied at this point: Where n is the unit normal vector outside the boundary; and at the oil-gas interface Γ og Mass transfer boundary conditions are applied at this point: Where k m C is the mass transfer coefficient. eq (T,C g The relationship between temperature T and the concentration of hydrogen in the interfacial gas phase C is... g The determined equilibrium concentration; the output of the hydrogen diffusion and migration model. At each sensor location x i Samples are taken to form a virtual sensor response sequence to generate a virtual dataset for training artificial intelligence algorithms.

9. A transformer fault location system based on a palladium thin-film hydrogen sensor array, characterized in that, include: Sensor array module: Installed at multiple preset monitoring locations in the transformer oil tank or gas space, the sensor is a palladium thin-film resistive hydrogen sensor, used to detect changes in hydrogen concentration in transformer oil in real time and output a resistance signal; Signal acquisition and synchronization module: used to synchronously acquire and time-calibrate the output signals of each sensor to obtain the arrival time sequence of hydrogen response; Digital twin simulation module: Based on transformer structure, oil flow characteristics and fault model, establish a digital twin model of transformer working state, including but not limited to structural parameter model, oil flow characteristic model, thermal model, electric field model, fault model and hydrogen diffusion and migration model, to simulate the hydrogen diffusion process under different fault types and locations, so as to generate a virtual dataset for algorithm training. Artificial intelligence algorithm module: Uses the virtual dataset for supervised learning training to form a fault location prediction model; Fault location module: During the actual operation of the transformer, it receives real-time response signals from various sensors, inputs them into the trained artificial intelligence model, and outputs fuzzy location results of the fault location.

10. The transformer fault location system based on a palladium thin-film hydrogen sensor array according to claim 1, characterized in that, The palladium thin-film resistive hydrogen sensor includes a palladium or palladium alloy thin-film hydrogen resistance wire assembly and a temperature control or compensation assembly, used to achieve high sensitivity and fast response hydrogen detection.

Images