A new energy automobile charging station transformer inter-turn short circuit detection system

Abstract

The application belongs to the technical field of transformer fault diagnosis, and particularly relates to a new energy automobile charging station transformer inter-turn short circuit detection system, wherein a magnetic field signal acquisition assembly is arranged at a preset detection point outside the transformer to non-contact collect external space magnetic field distribution data. A detection point optimization module selects an optimal feature extraction point according to the root mean square value of the magnetic field signal of each detection point. A time sequence feature coding module codes the time sequence signal of the Z-axis direction magnetic field at the optimal point in the time domain to extract deep time sequence features. A fault degree regression module establishes a mapping relationship between the deep time sequence features and the short circuit resistance value by using a WaveNet network to output a short circuit resistance prediction value. The application determines the optimal measurement point through three-dimensional electromagnetic simulation, realizes the mapping of the magnetic field features and the short circuit resistance value by using deep learning, can output the short circuit resistance prediction value in the form of continuous numerical value, and realizes the quantitative evaluation of the inter-turn short circuit fault severity.

Description

Technical Field

[0001] This invention belongs to the field of transformer fault diagnosis technology, specifically relating to an inter-turn short circuit detection system for transformers in new energy vehicle charging stations. Background Technology

[0002] Transformers play a crucial role in the transmission, transformation, and distribution of electrical energy, and their operational reliability significantly impacts the safety of the power supply system. In new energy vehicle charging stations, transformers, as the core power conversion equipment connecting the grid and charging terminals, directly affect the continuity and reliability of charging services. Charging station transformers operate under high load rates for extended periods, with the load fluctuating frequently depending on the number of charging terminals and the charging power. The windings are subjected to continuous thermal and electrical stress, making insulation performance more prone to degradation and significantly increasing the risk of inter-turn short circuits. When the inter-turn insulation is damaged, a localized short circuit can lead to increased circulating current, altered leakage flux distribution, and accumulated thermal stress, further inducing insulation aging and even equipment failure. This can cause charging station shutdowns and charging service interruptions, adversely affecting user travel and grid dispatch. Therefore, establishing a detection method capable of promptly identifying inter-turn short circuits and quantifying their severity is a clear engineering requirement and has significant application value for ensuring the operational safety of new energy vehicle charging stations.

[0003] In existing technologies, transformer inter-turn short circuit conditions are mostly identified through information such as electrical quantities, mechanical vibration quantities, and thermal characteristics. While these methods have a certain foundation in fault identification, they may still suffer from insufficient sensitivity, limited robustness, and weak quantitative characterization capabilities under conditions of minor faults, complex operating disturbances, or weakened characteristics. Especially in charging station scenarios, electrical disturbances caused by frequent transformer load fluctuations can easily overlap with fault characteristics, further increasing the difficulty of fault detection and severity assessment. Furthermore, in terms of fault severity assessment, existing research mostly focuses on the presence or type of fault, while schemes for continuous regression description of fault severity are relatively few, making it difficult to support the needs of refined diagnosis.

[0004] Compared to detection methods that require connection to the original measurement circuit or direct contact with conductive structures, external magnetic field measurement causes less disturbance to the original system and has the advantage of non-contact acquisition of external response information of the internal electromagnetic state. Therefore, external magnetic field signals can serve as an effective information source for transformer operating status sensing and anomaly detection. Meanwhile, deep learning has demonstrated strong capabilities in modeling complex nonlinear mappings; however, how to establish a stable fault severity characterization model based on external magnetic field time-series signals and form an effective correspondence between magnetic field characteristics and the degree of inter-turn short circuits still requires further research.

[0005] Based on the above background, this invention proposes a transformer inter-turn short circuit detection system based on deep fusion of magnetic array signal features, aiming to achieve non-contact continuous quantitative assessment of the severity of inter-turn short circuit faults under complex load conditions in charging stations. Summary of the Invention

[0006] The purpose of this invention is to provide a transformer inter-turn short circuit detection system for new energy vehicle charging stations, so as to solve the technical problem that existing detection methods are difficult to conduct non-contact continuous quantitative assessment of the severity of inter-turn short circuit faults.

[0007] The present invention achieves the above objectives through the following technical solutions: This invention proposes a transformer inter-turn short circuit detection system for new energy vehicle charging stations, comprising: a magnetic field signal acquisition component, a detection point optimization module, a timing feature encoding module, and a fault degree regression module; The magnetic field signal acquisition component is deployed at preset detection points in the external space of the transformer under test, and is used to collect the external space magnetic field distribution data of the transformer in operation in a non-contact manner. The preset detection points are distributed in different spatial orientations outside the transformer. The detection point selection module is used to obtain the response intensity distribution of the magnetic field signal at each preset detection point, and to sort the detection points according to the response intensity evaluation index, thereby selecting the optimal feature extraction point. The timing feature encoding module is used to receive the magnetic field timing signal corresponding to the optimal feature extraction point and perform time-domain deep feature encoding on it to obtain deep timing features for characterizing the electromagnetic state evolution process of the transformer internal windings. The fault severity regression module has a pre-trained regression decision model built in. The regression decision model takes the deep temporal features as input and the inter-turn short circuit fault severity parameter as the regression target. It is used to establish a nonlinear mapping relationship between magnetic field temporal features and short circuit fault severity, and to give the fault assessment result describing the current inter-turn short circuit severity in continuous numerical form.

[0008] As a further optimization of the present invention, the new energy vehicle charging station transformer is a step-down transformer connecting the power grid side and the charging terminal. The magnetic field signal acquisition component is set on the surface of the outer wall of the transformer. The preset detection points are distributed on the front and side of the transformer wall, and the preset detection points are arranged in an array from top to bottom on the front and side.

[0009] As a further optimization of the present invention, the magnetic field signal acquisition component includes a TMR magnetic sensor array and its signal conditioning circuit. The TMR magnetic sensor array is deployed at the preset detection point and is used to convert the magnetic field signal of the external space of the transformer into an electrical signal, and to generate a corresponding electrical output according to the change of the magnetic field around the transformer. The signal conditioning circuit is used to perform amplitude enhancement and interference suppression processing on the electrical output.

[0010] As a further optimization of the present invention, each sensor in the TMR magnetic sensor array is a biaxial TMR magnetic sensor, used to synchronously acquire magnetic field components in two orthogonal directions, the X-axis and the Z-axis; the time-series feature encoding module extracts the optimal feature extraction point's Z-axis direction magnetic field time series data from the biaxial magnetic field components acquired at each detection point as the input signal for its time-domain depth feature encoding, wherein the Z-axis direction is a direction perpendicular to the transformer box wall surface outward and parallel to the horizontal plane.

[0011] As a further optimization of the present invention, the response intensity evaluation index is the root mean square value of the magnetic field signal at each detection point, and the detection point selection module calculates the root mean square value of the magnetic field signal at each detection point according to the following formula: ; Where T is the sampling time period of the magnetic field signal, and B(t) is the magnetic field strength at time t; The detection point optimization module sorts the root mean square values ​​calculated for each detection point in descending order of numerical value, and selects the detection point with the highest root mean square value as the optimal feature extraction point.

[0012] As a further optimization of the present invention, the detection point selection module is also used to obtain the magnetic field response data of each preset detection point under normal working conditions based on the three-dimensional electromagnetic simulation model, and to calculate the root mean square value of each detection point based on the magnetic field response data.

[0013] As a further optimization of the present invention, the regression decision model of the fault severity regression module is pre-trained in the following manner: Step 1: Set up a short-circuit branch at the winding of the transformer under test, and adjust the resistance value of the short-circuit branch within a preset resistance range to construct multiple sets of inter-turn fault operating conditions with different short-circuit degrees. Step 2: Under each fault condition, the magnetic field signal at each preset detection point is collected by the magnetic field signal acquisition component, and the magnetic field time sequence signal corresponding to the optimal feature extraction point under each fault condition is extracted as a training sample, and the short-circuit resistance value corresponding to each fault condition is used as the sample label. Step 3: Input the training samples into the regression decision model, use the magnetic field time series signal as input and the corresponding short-circuit resistance value as output label to perform supervised regression training until the model converges, and establish a nonlinear mapping relationship between the magnetic field time series features and the short-circuit resistance value.

[0014] As a further optimization of the present invention, the short-circuit branch is a two-turn short-circuit branch set between the transformer winding layers; the training samples of the regression decision model are a portion of the fault condition samples selected from all fault conditions according to the short-circuit resistance value at equal intervals, and the remaining fault condition samples are used to verify or test the trained regression decision model.

[0015] As a further optimization of the present invention, the regression decision model is a WaveNet network, and the continuous quantitative evaluation result output by the fault severity regression module is the short-circuit resistance prediction value, which serves as a continuous indicator for measuring the severity of inter-turn short circuits in the transformer.

[0016] The beneficial effects of the present invention are as follows: (1) The present invention uses magnetic field data as the basis for diagnosing short circuit faults between transformer turns. Compared with electrical parameters such as current, voltage, and inductance, magnetic field data can be obtained without directly contacting the internal conductive parts of the transformer or changing the original electrical connection structure, thus having better non-invasiveness.

[0017] (2) The present invention uses a magnetic sensor to collect the external magnetic field response information of the transformer, which can reflect the changes in the external magnetic field caused by the changes in the internal electromagnetic state of the transformer in real time, thus having good real-time performance.

[0018] (3) This invention first analyzes the magnetic field response of multiple candidate detection locations under normal operating conditions using a three-dimensional electromagnetic simulation model, and determines the optimal data selection point based on the RMS value of the magnetic field signal in the Z-axis direction. Since the signal corresponding to the measurement point with a stronger magnetic field response is less affected by noise interference in actual detection, it is beneficial for subsequent further analysis.

[0019] (4) The present invention uses a magnetic sensor array to collect magnetic field data at multiple candidate locations outside the transformer, and selects the magnetic field data corresponding to the best data selection point to input into a deep learning neural network to establish a nonlinear mapping relationship between magnetic field data and the severity of inter-turn short circuit faults, thereby realizing continuous quantitative assessment of the severity of inter-turn short circuit faults of transformers and improving diagnostic accuracy.

[0020] (5) In some embodiments, the present invention can use a small number of typical fault condition samples to train a deep learning neural network and achieve prediction of the severity of other fault conditions. This shows that the solution can avoid the inefficiency caused by repeatedly trying different test points on a physical platform and reduce the testing cost and cycle in the early development stage. Attached Figure Description

[0021] Figure 1 This is a flowchart of a transformer inter-turn short circuit detection system for new energy vehicle charging stations according to the present invention. Figure 2 This is a schematic diagram of the overall technical roadmap for a transformer inter-turn short-circuit detection system for a new energy vehicle charging station according to the present invention. Figure 3 This is a schematic diagram of a three-dimensional electromagnetic field simulation model of a single-phase transformer obtained using Ansys Maxwell. Figure 4 This is a schematic diagram showing the arrangement of detection points of the magnetic array signal detection device on the front and side of the transformer's external tank wall in this invention; Figure 5 This is a schematic diagram of a fault circuit used to simulate the inter-turn short circuit state of a transformer in this invention; Figure 6 This is a schematic diagram showing the comparison results of the root mean square values ​​of the magnetic field signals in the Z-axis direction of the nine candidate detection points under normal operating conditions in this invention. Figure 7 This is a schematic diagram of the simulation results of the magnetic field waveform in the Z-axis direction at the optimal feature extraction point under the inter-turn short-circuit fault condition in the three-dimensional electromagnetic simulation model of this invention. Figure 8 This is a schematic diagram of the measured results of the magnetic field waveform in the Z-axis direction at the optimal feature extraction point under the condition of inter-turn short-circuit fault in the physical testing platform of this invention. Figure 9 This is a scatter plot comparing the predicted and actual short-circuit resistance values ​​under all fault conditions after the WaveNet network training is completed in this invention. Detailed Implementation

[0022] The following description provides specific application scenarios and requirements for this specification, intended to enable those skilled in the art to make and use the contents of this specification. Various partial modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but rather to the widest scope consistent with the claims.

[0023] The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not restrictive. For example, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. When used in this specification, the terms “comprising,” “including,” and / or “containing” mean that the associated integers, steps, operations, elements, and / or components are present, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components, and / or groups, or that other features, integers, steps, operations, elements, components, and / or groups may be added to the system / method.

[0024] Considering the following description, these and other features of this specification, as well as the operation and function of the related components of the structure, and the economy of assembly and manufacture of the parts, can be significantly improved. All of these form part of this specification with reference to the accompanying drawings. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0025] The flowcharts used in this specification illustrate operations implemented according to some embodiments of this specification. It should be clearly understood that the operations in the flowcharts may not be implemented in a sequential order. Instead, the operations may be implemented in reverse order or simultaneously. Furthermore, one or more additional operations may be added to the flowcharts. One or more operations may be removed from the flowcharts.

[0026] First Embodiment A preferred embodiment of the present invention provides a transformer inter-turn short-circuit detection system for new energy vehicle charging stations, such as... Figure 1 As shown, the system includes: a magnetic field signal acquisition component, a detection point optimization module, a timing feature encoding module, and a fault degree regression module.

[0027] In this embodiment, the magnetic field signal acquisition component belongs to the hardware layer and is deployed at preset detection points in the external space of the transformer under test. It is used to collect external space magnetic field distribution data of the transformer under operating conditions in a non-contact manner. The detection point selection module, the timing feature encoding module, and the fault severity regression module belong to the algorithm layer and run in a host computer or embedded processing platform. Among them, the detection point selection module is used to select the optimal feature extraction point based on the response intensity evaluation index of the magnetic field signal at each detection point; the timing feature encoding module is used to perform time-domain deep feature encoding on the magnetic field timing signal at the optimal point to extract deep timing features; the fault severity regression module has a built-in pre-trained regression decision model to map the deep timing features to the short-circuit resistance prediction value and output a continuous quantized value corresponding to the severity of the inter-turn short circuit, enabling the system to complete the online assessment of the degree of inter-turn short circuit of the transformer under non-contact detection conditions.

[0028] In the signal processing chain of the above four components: the magnetic field signal acquisition component is responsible for the extraction and digitization of the original magnetic field signal; the detection point selection module is responsible for selecting the channel with the best signal quality from the multi-channel signal; the time sequence feature encoding module is responsible for extracting deep features from the selected time sequence signal; and the fault degree regression module is responsible for mapping the features to the final fault degree quantification value.

[0029] (I) Specific Implementation Method of Magnetic Field Signal Acquisition Component The magnetic field signal acquisition component is deployed at preset detection points in the external space of the transformer under test, and is used to collect the external space magnetic field distribution data of the transformer in operation in a non-contact manner. The preset detection points are distributed in different spatial orientations outside the transformer.

[0030] like Figure 4 As shown, the transformer in a new energy vehicle charging station is located between the power grid supply side and the charging terminal load side, and is used to perform voltage reduction conversion. Its function is to convert the 10kV or 35kV medium-voltage power from the grid side into 400V or 690V low-voltage power for use by the charging terminal. This type of transformer is usually arranged in a box-type substation in the charging station. Its outer box wall is made of steel plate, which has no shielding effect on the power frequency magnetic field. Therefore, the magnetic field generated by the winding current inside can penetrate the box wall and propagate to the external space, providing a physical basis for the detection of external magnetic fields. The magnetic field signal acquisition component is set on the surface of the outer box wall of the transformer, rather than in the free space away from the transformer. The purpose is to reduce the distance between the magnetic sensor and the leakage magnetic source inside the transformer, so as to improve the signal-to-noise ratio of the acquired signal.

[0031] like Figure 4 As shown, the preset detection points are distributed on the front and sides of the transformer tank wall, but not on the top surface. The top surface is excluded because the top of the charging station transformer typically houses the power input line (i.e., the high-voltage side incoming cable) and its terminals. This area is surrounded by strong power frequency electric and magnetic field interference, and the cable routing and arrangement vary significantly depending on the site construction. If detection points were placed on the top surface, the collected magnetic field signal would include not only the leakage magnetic field component from the transformer's internal windings but also the current magnetic field component from the incoming cable, introducing uncertain interference components into the collected results and affecting the accuracy of subsequent fault identification and severity assessment.

[0032] Based on the above considerations, the present invention limits the candidate detection points to the front and sides of the transformer. These two areas are far away from the high-voltage incoming cable, and the magnetic field environment is relatively pure, which is conducive to obtaining external magnetic field signals that can truly reflect the internal electromagnetic state of the transformer.

[0033] The preset detection points are arranged in an array from top to bottom on the front and sides.

[0034] In one specific implementation, the preset detection points are nine, such as... Figure 4 As shown, the nine detection points are located on the front and sides of the transformer tank wall, arranged in three vertical layers, with three detection points in each layer. This three-row x three-point array arrangement covers the vertical height range of the transformer windings and different locations along the horizontal winding width. Since a partial short circuit in the windings can cause significant changes in the leakage magnetic response in the vicinity of the fault, this array-style multi-point arrangement allows the sensors to be placed in locations that can fully detect changes in the leakage magnetic field, avoiding missed fault signals due to improper sensor placement. Simultaneously, the multi-point arrangement provides sufficient candidate locations for subsequent optimal detection point selection, enabling the system to select the channel with the best signal quality from multiple measurement points.

[0035] like Figure 4 As shown, the magnetic field signal acquisition component mainly consists of a TMR (Tunnel Magnetoresistance) sensor array and a signal conditioning unit connected to it. A TMR sensor is a magnetically sensitive element based on the electron tunneling effect in a magnetic multilayer film structure. Its working principle is as follows: the TMR device utilizes the change in tunneling resistance caused by the change in the relative magnetization directions of the free layer and the pinned layer to sense the magnetic field. Therefore, the external magnetic field strength can be reflected through the resistance output response. Compared to conventional Hall effect devices, TMR devices have higher sensitivity in weak magnetic field detection, providing a more sufficient detection basis for early inter-turn short-circuit magnetic signal identification.

[0036] The TMR magnetic sensor array is deployed at the preset detection points, meaning each of the nine detection points is equipped with a TMR magnetic sensor. This sensor converts the magnetic field changes in the outer periphery of the transformer under test into an electrical signal that can be processed by subsequent circuitry. The signal conditioning circuit amplifies and filters the electrical signal to eliminate common-mode noise and power frequency harmonic interference, thereby improving signal quality.

[0037] In this embodiment, the signal conditioning circuit consists of a differential amplifier unit and a low-pass filter unit. The differential amplifier unit is mainly used to reduce common-mode interference components. The cutoff frequency of the low-pass filter circuit is set to 500Hz to retain the fundamental frequency (50Hz) and main harmonic components (up to the 10th harmonic, i.e., 500Hz) of the magnetic field signal, while filtering out noise interference in higher frequency bands to ensure that the signal entering the digital processing stage has a sufficient signal-to-noise ratio.

[0038] (II) Selection of the magnetic field signal in the Z-axis direction of the biaxial TMR sensor In this embodiment, each TMR detection unit possesses biaxial magnetic field sensing capability, enabling parallel acquisition of magnetic field components in two mutually perpendicular directions, X and Z. Since the biaxial TMR device integrates orthogonally arranged magnetic sensing units, it can obtain relatively complete spatial magnetic field component information for subsequent algorithm analysis. The time-series feature encoding module extracts the Z-axis magnetic field time-series data from the biaxial magnetic field components acquired at each detection point as the input signal for its time-domain depth feature encoding. The Z-axis direction is perpendicular to the transformer tank wall surface outwards and parallel to the horizontal plane. Figure 6 As shown, Figure 6 The distribution of RMS values ​​of the magnetic field signal in the Z-axis direction at each detection point is given.

[0039] The Z-axis magnetic field component was chosen as the input signal for subsequent analysis based on the spatial distribution characteristics of the transformer's leakage magnetic field. When alternating current is applied to the transformer windings, a time-varying magnetic field is formed around them. The magnetic field component distributed along the normal to the transformer tank wall (i.e., the Z-axis component) is most sensitive to changes in the current distribution within the windings. When an inter-turn short circuit occurs, the short-circuit current formed in the faulty winding changes the original current distribution and further causes significant disturbances to the external spatial Z-axis magnetic field component. The X-axis component, parallel to the tank wall surface, mainly reflects the symmetry of the overall transformer magnetic circuit and has relatively low sensitivity to local inter-turn short circuits. Therefore, this embodiment selects the Z-axis magnetic field time series data as the input signal for subsequent feature extraction to maximize the efficiency of fault information acquisition while reducing the processing burden of redundant data. In addition, the Z-axis magnetic field signal attenuates relatively slowly with distance in space, which is beneficial for obtaining sufficient signal strength even when there is an installation gap between the sensor and the transformer tank wall.

[0040] (III) Determination of the three-dimensional electromagnetic simulation model and the optimal detection point like Figure 3 As shown, the detection point selection module is also used to obtain the magnetic field response data of each preset detection point under normal operating conditions based on the three-dimensional electromagnetic simulation model, and to calculate the RMS root mean square parameters at different detection points using the magnetic field response results. The three-dimensional electromagnetic simulation model is built using Ansys Maxwell software, and the model's geometry, winding arrangement, and relative positional relationships correspond to the actual transformer.

[0041] Specifically, the three-dimensional electromagnetic simulation model corresponds to the physical transformer in the following aspects: (1) the number of winding turns corresponds; (2) the core structure corresponds, the transformer adopts a core structure, and the winding adopts a concentric winding; (3) the material properties correspond, including the core material and the winding conductor material; (4) the main structural dimensions correspond, including the core size, the winding size and the relative positional dimensions between each component.

[0042] By controlling the consistency of the above parameters, it is ensured that the magnetic field distribution characteristics of the simulation model can truly reflect the magnetic field distribution characteristics of the actual transformer, so that the optimal detection point obtained based on simulation analysis also has the best signal quality in actual detection.

[0043] After establishing the three-dimensional electromagnetic simulation model, transient simulation analysis was performed on the model to obtain the magnetic field data of each candidate detection location changing over time under normal operating conditions. Transient simulation analysis involves solving Maxwell's equations in the time domain to obtain the complete waveform of the magnetic field changing over time, rather than just obtaining the steady-state amplitude. This time-domain waveform data can fully reflect the harmonic components and waveform distortion information in the magnetic field signal, providing a complete statistical basis for subsequent RMS value calculation.

[0044] Subsequently, the detection point selection module calculates the root mean square value of the magnetic field signal at each detection point according to the following formula: Where T is the sampling time period of the magnetic field signal, and B(t) is the magnetic field strength at time t. The formula means that within the sampling time period T, the square of the magnetic field strength B(t) is integrated, then divided by T to obtain the average value, and finally the square root is taken to obtain the root mean square (RMS) value. The RMS value characterizes the average energy level of the magnetic field signal within the time period, comprehensively reflecting the strength of the magnetic field response at the detection point. Compared to the peak value, the RMS value is less affected by instantaneous spike noise and has better stability and repeatability.

[0045] The detection point optimization module sorts the root mean square values ​​calculated for each detection point in descending order of numerical value, and selects the detection point with the highest root mean square value as the optimal feature extraction point. For example... Figure 6 As shown, Figure 6 The RMS comparison results for nine candidate detection points are presented. The horizontal axis P1 to P9 represent the numbers of the nine detection points, and the vertical axis represents the RMS calculation result of the magnetic field response in the Z-axis direction for each detection point. Figure 6 As can be seen intuitively, the RMS value of point P5 (located in the middle of the side of the transformer) is the highest among all candidate points, and therefore it is determined as the optimal feature extraction point in this embodiment.

[0046] The advantage of using simulation to predetermine the optimal detection point is that: On the one hand, 3D electromagnetic simulation can pre-evaluate the signal quality of different detection points without building a physical testing platform, reducing the need for blind point placement and repeated experiments during physical testing, thereby reducing testing costs and time consumption in the R&D phase. On the other hand, simulation analysis can evaluate the intrinsic response intensity of each point under ideal conditions (no environmental noise, no measurement error), thus obtaining purer and more objective evaluation results, avoiding deviations introduced by factors such as installation errors and individual sensor differences in physical measurements. After the optimal points are determined by simulation, sensors only need to be deployed at the corresponding physical locations in the actual detection system, without the need to configure sensors at all candidate locations, further reducing system hardware costs.

[0047] (iv) Specific implementation of the time-series feature coding module The time-series feature encoding module is used to receive the magnetic field time-series signal corresponding to the optimal feature extraction point and perform time-domain deep feature encoding on it to obtain a deep sequence representation for describing the time evolution process of the electromagnetic state inside the transformer winding.

[0048] In this embodiment, the magnetic field time series signal collected at the optimal feature extraction point is magnetic field time series data in the Z-axis direction. Its sampling frequency is set to 1kHz, and each sample is truncated to cover one 50Hz power frequency cycle, corresponding to a duration of 20ms, forming a magnetic field sequence containing 20 sampling points. The time series feature encoding module receives this one-dimensional time series signal as input, performs nonlinear transformation on it through a multi-layer neural network, and extracts deep time series features.

[0049] It is understandable that the deep features referred to here are relative to the shallow features of the original signal (such as amplitude, RMS value and other statistics). They refer to the high-order abstract feature representation obtained after multiple nonlinear transformations, which can capture complex pattern information in the original signal that is not easy to obtain through manually designed feature extraction methods.

[0050] Specifically, the time-series feature encoding module employs a one-dimensional convolutional neural network structure to perform temporal convolution operations on the input time-series signal. By stacking multiple convolutional layers, the cascading effect of these layers increases the model's coverage of the correlation between preceding and subsequent sampling points, thereby extracting multi-scale magnetic field temporal patterns. When an inter-turn short-circuit fault occurs, the change in the external magnetic field signal is not a single transient abrupt change, but rather exhibits different characteristics at multiple time scales: at the microscopic time scale, the current waveform distortion caused by the fault manifests as waveform detail distortion in the magnetic field signal; at the macroscopic time scale, the overall change in the magnetic field amplitude caused by the fault manifests as the rise and fall of the signal envelope. Through the cascading of multiple temporal convolutions, the time-series feature encoding module can simultaneously capture the fault features at these different time scales, encoding them into a fixed-dimensional deep feature vector for subsequent fault severity regression module to map fault severity.

[0051] (v) Specific implementation of the fault severity regression module The fault severity regression module has a pre-trained regression decision model built in. The regression decision model takes the deep temporal features as input and the inter-turn short circuit fault severity parameter as the regression target. It is used to establish a nonlinear mapping relationship between magnetic field temporal features and short circuit fault severity, and to give the fault assessment result describing the current inter-turn short circuit severity in continuous numerical form.

[0052] In this embodiment, the regression decision model is a WaveNet network, and the continuous quantitative evaluation result output by the fault severity regression module is the short-circuit resistance prediction value, which serves as a continuous indicator for measuring the severity of inter-turn short circuits in the transformer.

[0053] In this invention, the WaveNet network is migrated to the regression modeling task of the time-series signal of the external magnetic field of the transformer, and its powerful time-series modeling capability is used to establish a nonlinear mapping relationship between the magnetic field waveform characteristics and the short-circuit resistance value.

[0054] The short-circuit resistance value refers to the equivalent resistance value in an inter-turn short-circuit branch. The smaller the value, the more severe the short circuit (close to a metallic short circuit); the larger the value, the milder the short circuit (close to a high-resistance short circuit). Therefore, by predicting the short-circuit resistance value, the system can achieve a continuous quantitative assessment of the severity of inter-turn short-circuit faults, rather than simply outputting discrete category labels such as "faulty / no fault" or "minor / severe," thereby providing maintenance personnel with more refined fault status information and supporting condition-based maintenance decisions.

[0055] The training process of the regression decision model is as follows: Step 1: Set up a short-circuit branch at the winding of the transformer under test, and adjust the resistance value of the short-circuit branch within a preset resistance range to obtain multiple sets of inter-turn short-circuit simulation conditions with different short-circuit degrees.

[0056] Specifically, such as Figure 5 As shown, the equivalent circuit of the transformer under test includes a primary input circuit and a secondary output circuit. Among them, U ac R represents the AC power supply applied to the primary side of the transformer. p I represents the equivalent series resistance of the primary winding. p L represents the primary current. p This represents the equivalent inductance of the primary winding; M Used to characterize the magnetic coupling mutual inductance between the primary and secondary windings. In the secondary output branch, R... s The equivalent resistance of the secondary winding in series, I s Corresponding to the secondary side output current, U out The corresponding output voltage at the secondary side port.

[0057] In this embodiment, to simulate an inter-turn short-circuit fault in the transformer winding, a short-circuit branch is drawn at the inter-layer location of the secondary winding. Specifically, Figure 5 L in s1 and L s2 These represent the equivalent inductance of local winding segments located in different winding layers of the secondary winding, where L s1 Corresponding to the first winding layer, i.e., Layer 1, L s2 This corresponds to the second winding layer, or Layer 2. A connecting branch is provided between the first and second winding layers, and a short-circuit resistor R is connected in series in this connecting branch. sc This forms a two-turn short-circuit fault simulation loop between layers. Wherein, R sc This is an adjustable short-circuit resistor, used to adjust the equivalent fault resistance value of the short-circuit branch; sc This indicates the short-circuit circulation formed in the short-circuit branch.

[0058] By adjusting R sc The resistance value can be used to construct inter-turn fault states with different short-circuit degrees. When R sc When the value is large, the equivalent fault resistance of the short-circuit branch is large, and the short-circuit circulating current i sc Smaller R corresponds to a less severe high-resistance short-circuit fault; when R sc When the value is small, the equivalent fault resistance of the short-circuit branch is small, and the short-circuit circulating current i sc An increase in resistance corresponds to a more severe low-resistance short-circuit fault. Based on the above structure, Figure 5The fault simulation circuit in the middle can adjust the short-circuit resistance R without changing the mutual inductive coupling relationship between the primary and secondary sides. sc This method constructs inter-turn short-circuit fault samples corresponding to different short-circuit resistance values, providing an experimental basis for subsequent magnetic field signal acquisition, fault sample construction, and short-circuit resistance value regression prediction.

[0059] Step 2: Under each fault condition, the magnetic field signal at each preset detection point is collected by the magnetic field signal acquisition component, and the magnetic field time sequence signal corresponding to the optimal feature extraction point under each fault condition is extracted as a training sample, and the short-circuit resistance value corresponding to each fault condition is used as the sample label.

[0060] In this embodiment, the adjustment range of the short-circuit resistance value covers the entire range from minor to severe faults, and the resistance value intervals between each fault condition ensure uniform coverage of the fault severity space. The training samples for the regression decision model are a subset of fault condition samples selected at equal intervals according to the short-circuit resistance value from all fault conditions. The remaining fault condition samples are used to verify or test the trained regression decision model. This equal-interval selection method ensures that the training samples are evenly distributed in the fault severity space, avoiding the concentration of training samples in a certain fault severity interval, which could lead to insufficient generalization ability of the model in other intervals.

[0061] Step 3: Input the training samples into the regression decision model, use the magnetic field time series signal as input and the corresponding short-circuit resistance value as output label to perform supervised regression training until the model converges, and establish a nonlinear mapping relationship between the magnetic field time series features and the short-circuit resistance value.

[0062] like Figure 9 As shown, Figure 9 A scatter plot comparing the predicted and actual short-circuit resistance values ​​output by the model after training is complete and the magnetic field time-series signals for all fault conditions (including the training and test sets) are input into the model. The scatter plot uses the actual short-circuit resistance value as the horizontal axis and the predicted short-circuit resistance value output by the model as the vertical axis, employing a diagonal reference line to represent the ideal state where the predicted value matches the actual value. Figure 9 It can be seen that the predicted scatter points are close to the ideal diagonal reference line, indicating that the deviation between the model output and the actual short-circuit resistance value is small. This shows that the detection system can establish a stable correlation between the external magnetic field characteristics and the severity of the inter-turn short circuit, and has the ability to regress and predict fault states outside the training set.

[0063] Figure 7 and Figure 8 The simulated fault waveforms and the measured fault waveforms are compared respectively. Figure 7The simulated Z-axis magnetic field waveform obtained at the optimal feature extraction point after setting up an inter-turn short-circuit fault in a three-dimensional electromagnetic simulation model; Figure 8 This refers to the measured Z-axis magnetic field waveform at the same location after an inter-turn short-circuit fault with the same short-circuit resistance value was set on a physical test platform. (Comparison) Figure 7 and Figure 8 It can be seen that the simulated waveform and the measured waveform have good consistency in terms of waveform shape, amplitude variation trend and main harmonic components, thus verifying the effectiveness of the three-dimensional electromagnetic simulation model and proving that the optimal detection point determined based on simulation analysis can effectively guide the sensor deployment of the physical system.

[0064] As a specific application process, Figure 2 The overall execution path of the detection system of this invention, from modeling to diagnosis, is given, including the simulation modeling stage, the optimal data selection point determination stage, the data acquisition stage, and the deep learning training and severity diagnosis stage, as detailed below: In the simulation modeling stage, a corresponding three-dimensional electromagnetic simulation model is established based on the structural parameters of the physical transformer. This model is used to obtain the magnetic field response at each candidate detection location outside the transformer (steps 1 and 2). Then, the optimal data selection point determination stage is entered. The root mean square (RMS) value of the magnetic field signal in the Z-axis direction at each candidate detection location is calculated. By comparing the RMS values ​​at each location, the location with the largest RMS value is determined as the optimal data selection point (steps 3 and 4). After completing the simulation screening, the data acquisition stage begins. An inter-turn short-circuit fault is constructed on the physical transformer. Different short-circuit resistance values ​​are set to characterize different fault severity levels. Magnetic sensor arrays are used to collect magnetic field data under various fault conditions outside the transformer, and the corresponding magnetic field data is extracted based on the optimal data selection point to construct a fault sample set (steps 5 and 6). Finally, the deep learning training and severity diagnosis stage is entered. Some typical fault condition samples are selected as training samples. The corresponding magnetic field data is input into the signal feature deep fusion algorithm unit for training to establish the mapping relationship between magnetic field data and short-circuit resistance value (step 7). The magnetic field data to be diagnosed is input into the trained signal feature deep fusion algorithm unit, and the corresponding short-circuit resistance prediction value is output to realize the fault severity diagnosis (step 8).

[0065] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0066] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0067] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A short-circuit detection system for transformer turns in a new energy vehicle charging station, characterized in that, include: Magnetic field signal acquisition component, detection point optimization module, timing feature encoding module, and fault degree regression module; The magnetic field signal acquisition component is deployed at preset detection points in the external space of the transformer under test, and is used to collect the external space magnetic field distribution data of the transformer in operation in a non-contact manner. The preset detection points are distributed in different spatial orientations outside the transformer. The detection point selection module is used to obtain the response intensity distribution of the magnetic field signal at each preset detection point, and to sort the detection points according to the response intensity evaluation index, thereby selecting the optimal feature extraction point. The timing feature encoding module is used to receive the magnetic field timing signal corresponding to the optimal feature extraction point and perform time-domain deep feature encoding on it to obtain deep timing features for characterizing the electromagnetic state evolution process of the transformer internal windings. The fault severity regression module has a pre-trained regression decision model built in. The regression decision model takes the deep temporal features as input and the inter-turn short-circuit fault severity parameter as the regression target. It is used to establish a nonlinear mapping relationship between magnetic field temporal features and short-circuit fault severity, and to give a continuous fault severity evaluation value corresponding to the current inter-turn short-circuit state.

2. The inter-turn short-circuit detection system for transformers in new energy vehicle charging stations according to claim 1, characterized in that, The transformer of the new energy vehicle charging station is a step-down transformer that connects the power grid side and the charging terminal. The magnetic field signal acquisition component is set on the surface of the outer wall of the transformer. The preset detection points are distributed on the front and side of the transformer wall, and the preset detection points are arranged in an array from top to bottom on the front and side.

3. The short-circuit detection system for transformer turns in a new energy vehicle charging station according to claim 2, characterized in that, The magnetic field signal acquisition component includes a TMR magnetic sensor array and its signal conditioning circuit. The TMR magnetic sensor array is deployed at the preset detection point and is used to generate corresponding electrical output according to the change of the magnetic field around the transformer. The signal conditioning circuit is used to perform amplitude enhancement and interference suppression processing on the electrical output.

4. The short-circuit detection system for transformer turns in a new energy vehicle charging station according to claim 3, characterized in that, Each sensor in the TMR magnetic sensor array is a biaxial TMR magnetic sensor, used to synchronously acquire magnetic field components in two orthogonal directions, the X-axis and the Z-axis. The time-series feature encoding module extracts the optimal feature extraction point's Z-axis magnetic field time series data from the biaxial magnetic field components acquired at each detection point as the input signal for its time-domain depth feature encoding. The Z-axis direction is a direction perpendicular to the transformer box wall surface outward and parallel to the horizontal plane.

5. The short-circuit detection system for transformer turns in a new energy vehicle charging station according to claim 1, characterized in that, The response intensity evaluation index is the root mean square value of the magnetic field signal at each detection point. The detection point selection module calculates the root mean square value of the magnetic field signal at each detection point according to the following formula: ； Where T is the sampling time period of the magnetic field signal, and B(t) is the magnetic field strength at time t; The detection point optimization module sorts the root mean square values ​​calculated for each detection point in descending order of numerical value, and selects the detection point with the highest root mean square value as the optimal feature extraction point.

6. The inter-turn short-circuit detection system for transformers in new energy vehicle charging stations according to claim 5, characterized in that, The detection point selection module is also used to obtain the magnetic field response data of each preset detection point under normal operating conditions based on the three-dimensional electromagnetic simulation model, and to calculate the root mean square value of each detection point based on the magnetic field response data.

7. The inter-turn short-circuit detection system for transformers in new energy vehicle charging stations according to claim 1, characterized in that, The regression decision model of the fault severity regression module is pre-trained using the following method: Step 1: Set up a short-circuit branch at the winding of the transformer under test, and adjust the resistance value of the short-circuit branch within a preset resistance range to obtain various inter-turn fault conditions corresponding to different short-circuit development levels. Step 2: Under each fault condition, the magnetic field signal at each preset detection point is collected by the magnetic field signal acquisition component, and the magnetic field time sequence signal corresponding to the optimal feature extraction point under each fault condition is extracted as a training sample, and the short-circuit resistance value corresponding to each fault condition is used as the sample label. Step 3: Input the training samples into the regression decision model, use the magnetic field time series signal as input and the corresponding short-circuit resistance value as output label to perform supervised regression training until the model converges, and establish a nonlinear mapping relationship between the magnetic field time series features and the short-circuit resistance value.

8. The inter-turn short-circuit detection system for transformers in new energy vehicle charging stations according to claim 7, characterized in that, The short-circuit branch is a two-turn short-circuit branch set between the transformer winding layers; the training samples of the regression decision model are a portion of the fault condition samples selected from all fault conditions according to the short-circuit resistance value at equal intervals, and the remaining fault condition samples are used to verify or test the trained regression decision model.

9. A short-circuit detection system for transformer turns in a new energy vehicle charging station according to claim 1, characterized in that, The regression decision model is a WaveNet network, and the continuous quantitative evaluation result output by the fault severity regression module is the short-circuit resistance prediction value. The short-circuit resistance prediction value is used as an evaluation index to characterize the severity of the inter-turn short circuit state of the transformer under test.

Images