Transformer vibration fault detection methods, devices, equipment, media, and procedures.
By constructing a physical model of transformer vibration and combining it with a data-driven method for error compensation, the accuracy and robustness of transformer vibration fault detection under complex operating conditions are solved, achieving more accurate fault diagnosis and reducing the false diagnosis rate.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TBEA HENGYANG TRANSFORMERS
- Filing Date
- 2026-02-24
- Publication Date
- 2026-06-02
Smart Images
Figure CN122131197A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power fault detection technology, and in particular to a method, apparatus, equipment, medium, and program product for detecting transformer vibration faults. Background Technology
[0002] As power systems develop towards higher voltage and larger capacity, the operating status of power transformers, as core equipment of the power grid, directly determines the stability of the grid and the reliability of power supply. According to statistics from the National Energy Administration, power outages caused by transformer failures account for 23% of all power grid accidents. Among these, transformer failures caused by abnormal vibrations are particularly challenging to diagnose because their early characteristics are often concealed and easily masked by noise.
[0003] Traditional techniques involve collecting vibration signals from the surface of the transformer tank using vibration sensors, and then automatically extracting features and identifying fault modes using algorithms such as deep learning. This approach yields good fault detection results under ideal conditions of high data quality and simple operating conditions.
[0004] However, in actual industrial scenarios, the operating environment of transformers is complex and variable. When vibration signals are contaminated by factors such as strong electromagnetic interference, the features extracted by the model are distorted, and the accuracy of fault detection is significantly reduced. Summary of the Invention
[0005] Therefore, it is necessary to provide a transformer vibration fault detection method, device, equipment, medium, and program product that can improve the accuracy of transformer vibration fault detection in response to the above-mentioned technical problems.
[0006] Firstly, this application provides a method for detecting transformer vibration faults, including:
[0007] Acquire the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within a preset fault detection cycle;
[0008] Using the transformer vibration physical model, the initial vibration acceleration value of the transformer at each fault detection time is determined based on the core vibration signal value and winding vibration signal value corresponding to each fault detection time.
[0009] Physical model detection error compensation is performed on each initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at each fault detection time.
[0010] Vibration fault detection of the transformer is performed based on the vibration acceleration values of each target.
[0011] In one embodiment, vibration fault detection of the transformer is performed based on each target vibration acceleration value, including:
[0012] Obtain the measured values of the transformer's vibration acceleration at each fault detection moment;
[0013] The transformer is subjected to vibration fault detection based on the target residual value between the target vibration acceleration value and the measured vibration acceleration value at each fault detection time.
[0014] In one embodiment, vibration fault detection of the transformer is performed based on the residual value between the target vibration acceleration value and the measured vibration acceleration value at each fault detection time, including:
[0015] Based on the target vibration acceleration value and the measured vibration acceleration value at each fault detection time, detect the target residual value of the preset frequency band;
[0016] Based on the deviation between the target residual value and the residual threshold corresponding to the preset frequency band, the confidence level of the transformer having the target vibration fault corresponding to the preset frequency band is determined;
[0017] If the confidence level exceeds the preset confidence threshold, the transformer is determined to have a target vibration fault.
[0018] If the confidence level does not exceed the preset confidence threshold, it is determined that the transformer does not have the target vibration fault.
[0019] In one embodiment, the target residual value of a preset frequency band is detected based on the target vibration acceleration value and the measured vibration acceleration value at each fault detection time, including:
[0020] The target frequency band acceleration value of the preset frequency band is extracted from each target vibration acceleration value, and the frequency band acceleration value of the preset frequency band is extracted from each vibration acceleration measured value.
[0021] The difference between the target frequency band acceleration value and the measured frequency band acceleration value at each fault detection time is determined as the initial residual value of the preset frequency band at each fault detection time.
[0022] The average value of each initial residual is determined as the target residual value for the preset frequency band.
[0023] In one embodiment, for any fault detection moment, the initial vibration acceleration value is compensated for by physical model detection error to obtain the target vibration acceleration value of the transformer at the fault detection moment, including:
[0024] Acquire the transformer status data at the time of fault detection. The status data should include at least the load current value and the oil temperature value.
[0025] Based on the state data, the initial vibration acceleration value is compensated for by physical model detection error to obtain the target vibration acceleration value of the transformer at the fault detection time.
[0026] In one embodiment, based on the state data, physical model detection error compensation is performed on the initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at the fault detection moment, including:
[0027] The state data and the initial vibration acceleration value at the fault detection time are input into the error compensation model to obtain the error compensation value at the fault detection time. The error compensation model is obtained by training the recurrent fuzzy neural network to be trained based on the transformer's historical state data and the historical initial vibration acceleration value.
[0028] By combining the error compensation value and the initial vibration acceleration value at the time of fault detection, the target vibration acceleration value of the transformer at the time of fault detection is obtained.
[0029] Secondly, this application also provides a transformer vibration fault detection device, comprising:
[0030] The acquisition module is used to acquire the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within a preset fault detection cycle.
[0031] The determination module is used to determine the initial vibration acceleration value of the transformer at each fault detection time based on the core vibration signal value and winding vibration signal value corresponding to each fault detection time using the transformer vibration physical model.
[0032] The compensation module is used to compensate for the physical model detection error of each initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at each fault detection time.
[0033] The detection module is used to detect vibration faults in the transformer based on the vibration acceleration values of each target.
[0034] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:
[0035] Acquire the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within a preset fault detection cycle;
[0036] Using the transformer vibration physical model, the initial vibration acceleration value of the transformer at each fault detection time is determined based on the core vibration signal value and winding vibration signal value corresponding to each fault detection time.
[0037] Physical model detection error compensation is performed on each initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at each fault detection time.
[0038] Vibration fault detection of the transformer is performed based on the vibration acceleration values of each target.
[0039] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:
[0040] Acquire the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within a preset fault detection cycle;
[0041] Using the transformer vibration physical model, the initial vibration acceleration value of the transformer at each fault detection time is determined based on the core vibration signal value and winding vibration signal value corresponding to each fault detection time.
[0042] Physical model detection error compensation is performed on each initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at each fault detection time.
[0043] Vibration fault detection of the transformer is performed based on the vibration acceleration values of each target.
[0044] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:
[0045] Acquire the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within a preset fault detection cycle;
[0046] Using the transformer vibration physical model, the initial vibration acceleration value of the transformer at each fault detection time is determined based on the core vibration signal value and winding vibration signal value corresponding to each fault detection time.
[0047] Physical model detection error compensation is performed on each initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at each fault detection time.
[0048] Vibration fault detection of the transformer is performed based on the vibration acceleration values of each target.
[0049] The aforementioned transformer vibration fault detection method, device, equipment, medium, and program products firstly acquire vibration signals from the transformer core and windings. Using a pre-constructed transformer vibration physical model, they calculate the initial vibration acceleration value, reflecting the transformer's inherent electromagnetic-mechanical coupling mechanism, based on the core and winding vibration signal values. This achieves accurate extraction of vibration characteristics with clear physical meaning. Subsequently, addressing the shortcomings of the transformer vibration physical model in representing complex nonlinear faults, data-driven physical model detection error compensation is applied to the initial vibration acceleration value to obtain a more accurate target vibration acceleration value reflecting the transformer's actual state. Finally, more precise fault diagnosis can be achieved based on the target vibration acceleration value. Thus, this application constrains the basic evolution law of vibration signals through a physical model and then corrects the limitations of the physical model in nonlinear fault feature identification through a data-driven approach, thereby significantly improving the robustness and accuracy of fault detection under complex operating conditions such as strong electromagnetic interference. Attached Figure Description
[0050] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0051] Figure 1 This is a diagram illustrating the application environment of a transformer vibration fault detection method in one embodiment of this application.
[0052] Figure 2 This is a flowchart illustrating a transformer vibration fault detection method in one embodiment of this application;
[0053] Figure 3 This is a flowchart illustrating a transformer vibration fault detection method in another embodiment of this application;
[0054] Figure 4 This is a structural block diagram of a transformer vibration fault detection device in one embodiment of this application;
[0055] Figure 5 This is an internal structural diagram of a computer device in one embodiment of this application. Detailed Implementation
[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0057] As power systems develop towards higher voltage and larger capacity, the operating status of power transformers, as core equipment of the power grid, directly determines the stability of the grid and the reliability of power supply. According to statistics, power outages caused by transformer failures account for 23% of all power grid accidents. Among them, transformer failures caused by abnormal vibrations are the core challenge in fault diagnosis because their early characteristics are hidden and easily masked by noise.
[0058] Currently, transformer vibration fault diagnosis technology mainly includes two types: physical modeling and data-driven methods.
[0059] Physical modeling, based on the electromagnetic-mechanical coupling principle of transformers, establishes vibration transmission equations and locates faults by calculating the relationship between the magnetostrictive force of the core, the electrodynamic force of the windings, and the vibration displacement. However, this type of method relies on precise structural parameters such as the lamination coefficient of the transformer core and the stiffness of the windings. In actual operation, these parameters change over time due to temperature rise and aging, resulting in modeling errors often exceeding 15%, which is insufficient to meet the diagnostic accuracy requirements.
[0060] Data-driven approaches collect vibration signals from the surface of the oil tank using vibration sensors and extract fault features using deep learning algorithms such as neural networks and support vector machines. For example, CNNs (Convolutional Neural Networks) are used to extract the temporal and frequency domain features of the vibration signals, or LSTMs (Long Short-Term Memory) networks are used to capture the temporal correlation of vibration signal sequences. However, these methods ignore the physical nature of transformer vibration, and their diagnostic accuracy drops sharply and their generalization ability is poor when faced with strong electromagnetic interference or insufficient sample size.
[0061] To combine physical laws and data characteristics, an end-to-end large model can be trained using vibration signals and a large amount of data related to physical laws. This allows the large model to autonomously learn the intrinsic correlation between physical laws, data characteristics, and transformer vibration faults. However, the diagnostic delay of such methods exceeds 500ms, which is difficult to meet the requirements of real-time monitoring.
[0062] To deeply integrate physical laws and data characteristics, a large-scale end-to-end model can be jointly trained using massive vibration signals and associated physical parameters (such as current, temperature, and magnetic flux density). This model autonomously learns the complex mapping relationship between physical constraints, signal characteristics, and fault states. However, such methods rely on a large number of parameters and complex model structures, resulting in high inference computation and significant diagnostic delays, typically exceeding 500 milliseconds, which cannot meet the practical requirements of real-time response in transformer fault detection.
[0063] Furthermore, transformer vibration signals exhibit complex characteristics such as strong nonlinearity, time-varying noise, and multi-source coupling. Under normal operating conditions, the vibration energy is mainly concentrated in the 100Hz fundamental frequency and its harmonics; when a core loosening fault occurs, the vibration amplitude in the 200-300Hz frequency band can increase to 3-5 times the normal level, accompanied by a significant increase in random noise; while winding deformation manifests as an abnormal amplitude ratio between characteristic frequency bands such as 100Hz and 200Hz. Traditional techniques struggle to effectively distinguish between background noise, electromagnetic interference, and early, subtle fault characteristics, leading to a misdiagnosis rate often exceeding 20% in practical engineering.
[0064] Based on this, this application proposes a transformer vibration fault detection method that integrates physical constraints and data-driven approaches. By constraining the basic evolution law of vibration signals through a physical model, and then correcting the limitations of the physical model in nonlinear fault feature identification through a data-driven approach, the robustness and accuracy of fault detection are significantly improved under complex operating conditions such as strong electromagnetic interference. This is of great significance for improving transformer operation and maintenance efficiency and reducing unplanned power outage losses.
[0065] The request processing method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located on the cloud or other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.
[0066] In one exemplary embodiment, such as Figure 2 As shown, a request processing method is provided, which is applied to Figure 1 Taking terminal 102 as an example, the explanation includes the following steps 202 to 208. Wherein:
[0067] Step 202: Obtain the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within the preset fault detection cycle.
[0068] The core vibration signal value can refer to the original physical quantity characterizing the intensity of the core vibration, which is collected by a vibration sensor installed at the core. In some embodiments, the core vibration signal value can be the core vibration acceleration value.
[0069] The winding vibration signal value can refer to the raw physical quantity characterizing the intensity of winding vibration, which is collected by a vibration sensor installed at the winding. In some embodiments, the winding vibration signal value can be the winding vibration acceleration value.
[0070] In some embodiments, the vibration sensor is an acceleration sensor, the core vibration signal value is the core vibration acceleration value, and the winding vibration signal value is the winding vibration acceleration value.
[0071] As a core electromagnetic energy conversion device in a power system, a transformer's physical structure comprises two main components: the iron core and the windings. The iron core is made of stacked high-permeability silicon steel sheets, forming a closed magnetic circuit that provides an efficient channel for alternating magnetic fields. The high-voltage and low-voltage windings are coaxially nested on the iron core columns, forming a circuit system. When alternating current passes through the windings, a periodically changing magnetic flux is generated in the iron core. Through the principle of electromagnetic induction, a voltage is induced in the winding on the other side, realizing the transfer of electrical energy. This "electric-magnetic-electric" energy conversion mechanism determines that the transformer operates accompanied by mechanical vibrations excited by electromagnetic forces.
[0072] Transformer vibration is closely related to the strong coupling between its electromagnetic field and mechanical structure. Core vibration is primarily driven by the magnetostrictive effect and Maxwellian electromagnetic stress: silicon steel sheets undergo periodic microscopic expansion and contraction under an alternating magnetic field, while electromagnetic attraction pulsates between the core laminations, proportional to the square of the magnetic flux density. Both factors jointly excite the core to generate a vibration spectrum. Winding vibration originates from the Lorentz force: the alternating electrodynamic force experienced by the load current in the winding leakage magnetic field causes forced vibration of the winding conductors. When faults such as core loosening or winding deformation occur, changes in the stiffness, damping, or coupling state of the mechanical structure lead to a redistribution of vibration energy in specific frequency bands.
[0073] Therefore, in order to construct vibration characteristics that can directly reflect the coupling effect between electromagnetic excitation and mechanical structure inside the transformer, we can first obtain the core vibration signal value that characterizes the magnetostrictive vibration of the core and the winding vibration signal value that characterizes the electrodynamic vibration of the winding.
[0074] The fault detection period refers to a fixed-length data window extracted from continuously acquired vibration signals and used to perform a single transformer vibration fault detection. The length of the fault detection period can be determined based on actual needs or test results, such as 50ms or 100ms. The fault detection moment refers to the specific time point within the fault detection period where data is sampled; each fault detection moment corresponds to a discrete vibration signal sample value. For example, if the sampling frequency of the core vibration signal is 1kHz, then a 50ms fault detection period contains 50 sampling points, thus acquiring 50 core vibration signal values.
[0075] In a real-time detection scenario, the end time of each fault detection cycle is the current time. For example, assuming the length of the fault detection cycle is 50ms, a transformer vibration fault detection can be triggered at 50ms, 100ms, ..., 50Nms. At 50ms, the vibration signal values collected from 0 to 50ms are used to perform a transformer vibration fault detection. At 100ms, the vibration signal values collected from 50 to 100ms are used to perform a transformer vibration fault detection. ..., at 50Nms, the vibration signal values collected from 50N-50 to 50Nms are used to perform a transformer vibration fault detection.
[0076] For example, at least one core vibration signal value within a preset fault detection period can be extracted from the vibration signal collected by the vibration sensor installed at the core through a sliding window; simultaneously, at least one core vibration signal value within a preset fault detection period can also be extracted from the vibration signal collected by the vibration sensor installed at the winding through a sliding window.
[0077] Step 204: Using the transformer vibration physical model, determine the initial vibration acceleration value of the transformer at each fault detection time based on the core vibration signal value and winding vibration signal value corresponding to each fault detection time.
[0078] Among them, the transformer vibration physical model can refer to the mathematical model used to describe the theoretical relationship between transformer vibration acceleration and transformer operating parameters.
[0079] In some embodiments, a physical model of transformer vibration can be established based on the principles of electromagnetics and structural mechanics.
[0080] We can first establish independent physical models for the two main vibration sources: the iron core and the winding.
[0081] Core vibration primarily originates from the magnetostrictive effect and the periodic interaction of electromagnetic attraction between laminations (Maxwell stress). When lamination loosening occurs, random impacts and friction caused by changes in mechanical clearance are also introduced. Therefore, the core vibration source model can be represented as:
[0082]
[0083] in, The vibration acceleration value at the iron core; the first term The first term is the steady-state periodic term, used to describe periodic oscillations under healthy conditions; the second term... is the fault random term, used to describe the abnormal vibration introduced by the loosening fault; k1 is the magnetostriction coefficient, which can comprehensively reflect the material characteristics of silicon steel sheet and the structural amplification effect; B0 is the peak value of the iron core magnetic flux density; f is the power frequency; k2 is the loosening influence coefficient, which characterizes the amplification factor of the vibration response generated by random excitation per unit gap; The time-varying function of the lamination gap; This is random vibration noise, used to simulate broadband noise generated by random collisions and friction caused by gaps.
[0084] Considering that temperature changes cause thermal expansion and contraction of metal components, thus affecting mechanical clearances, a time-varying function of the lamination clearance can be used. Introducing temperature compensation, and the time-varying function of the lamination gap after temperature compensation. It can be represented as:
[0085]
[0086] in, The initial gap at rated temperature; T ref γ is the rated reference temperature; γ is the equivalent thermal expansion coefficient; T(t) is the real-time oil temperature at time t.
[0087] By dynamically correcting fault parameters, the accuracy of the state characterization of the core vibration source model under different temperature conditions is improved, and time-varying deviations are reduced.
[0088] Winding vibration is mainly considered in terms of electrodynamic forces and changes in winding stiffness. Therefore, the winding vibration source model can be expressed as:
[0089]
[0090] in, This represents the vibration acceleration value at the winding. As an incentive item; For system response items; m w k is the equivalent mass of the winding; k3 is the electrodynamic coefficient; I(t) is the load current at time t; c w k is the winding damping coefficient. w For winding stiffness; This represents the vibration displacement of the winding. This represents the vibration velocity of the winding.
[0091] The vibrations of the core and windings are transmitted to the tank surface through the mechanical structure. The measured vibrations are a linear superposition of the two vibrations after attenuation or amplification through different transmission paths, and are also mixed with environmental interference. Therefore, by coupling and superimposing the core vibration source model and the winding vibration source model, a comprehensive vibration model transmitted to the tank surface is established, which is the transformer vibration physical model. The transformer vibration physical model can be expressed as:
[0092]
[0093] in, α is the theoretical vibration acceleration value of the oil tank surface, which is also the transformer vibration acceleration value; α is the vibration contribution coefficient. This is environmental noise interference.
[0094] To enhance the model's practicality, the original signal can be preprocessed with power frequency filtering before using the model to reduce the impact of strong electromagnetic interference from the power grid's power frequency and its adjacent frequency bands on the vibration signal.
[0095] After the transformer vibration physical model is constructed, historical data or test data can be used for fitting to determine the various model parameters in the transformer vibration physical model.
[0096] Winding mass in the physical model of transformer vibration Rated stiffness Magnetic flux density peak These parameters can be extracted from the transformer's manufacturing data.
[0097] The magnetostriction coefficient k1 can be calculated through an unloaded test with a test duration of 30 minutes and a sampling interval of 1 second. The average value is then taken as the magnetostriction coefficient k1.
[0098] The loosening influence coefficient k2 and electrodynamic coefficient k3 can be corrected hourly using the load current I(t) and real-time oil temperature T(t). The correction formula is as follows:
[0099]
[0100]
[0101] Among them, I n t is the rated current; k2(t) is the loosening influence coefficient at time t, k2(0) is the initial value of the loosening influence coefficient; k3(t) is the electrodynamic coefficient at time t, k3(0) is the initial value of the electrodynamic coefficient.
[0102] The initial vibration acceleration value refers to the theoretical vibration level estimate obtained by inputting the extracted core vibration signal value and winding vibration signal value into the transformer vibration physical model, which conforms to the electromagnetic-mechanical coupling mechanism. The initial vibration acceleration value is derived based on the first principles of physics and reflects the vibration acceleration level that the transformer should have when it is in a theoretically healthy state under the current operating conditions.
[0103] For example, using a pre-built transformer vibration physical model, the core vibration signal value and winding vibration signal value collected at each fault detection moment are used as inputs to calculate the initial vibration acceleration value of the transformer at each fault detection moment. The calculation process of the transformer vibration physical model can be performed point-by-point in the time domain to obtain the initial vibration acceleration value corresponding to each fault detection moment, thereby preserving the time-varying characteristics of the vibration signal and laying the foundation for subsequent accurate fault diagnosis.
[0104] Step 206: Perform physical model detection error compensation on each initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at each fault detection time.
[0105] The target vibration acceleration value can refer to a corrected estimate that is closer to the actual vibration level of the transformer tank surface.
[0106] Simplified assumptions and inaccurate parameters in the physical model of transformer vibration may cause deviations between the initial vibration acceleration values and the actual vibration acceleration values on the transformer tank surface. Physical model detection error compensation refers to the process of correcting these deviations using data-driven methods. Methods for physical model detection error compensation can include neural networks, regression models, and adaptive algorithms.
[0107] For example, a data-driven approach is used to correct the initial vibration acceleration value at each fault detection moment to compensate for systematic biases introduced by the simplified assumptions and parameter mismatches of the transformer vibration physical model. Simultaneously, it learns and integrates nonlinear dynamics and complex coupling effects that the model fails to fully represent. This combines the initial vibration acceleration value derived from the mechanism with the nonlinear characteristics implicit in the data, ultimately generating a more accurate target vibration acceleration value that reflects the actual vibration state of the transformer tank surface.
[0108] Step 208: Vibration fault detection of the transformer is performed based on the vibration acceleration values of each target.
[0109] For example, by using the target vibration acceleration value obtained through dual optimization of physics and data, mechanical faults such as transformer core loosening and winding deformation can be identified and their conditions assessed through threshold discrimination, pattern recognition or machine learning classification algorithms.
[0110] In the aforementioned transformer vibration fault detection method, firstly, vibration signals from the transformer core and windings are collected. Using a pre-constructed transformer vibration physical model, the initial vibration acceleration value, reflecting the transformer's inherent electromagnetic-mechanical coupling mechanism, is calculated based on the core and winding vibration signal values, thus achieving accurate extraction of vibration features with clear physical meaning. Subsequently, addressing the shortcomings of the transformer vibration physical model in representing complex nonlinear faults, data-driven physical model detection error compensation is applied to the initial vibration acceleration value to obtain a more accurate target vibration acceleration value reflecting the actual state of the transformer. Finally, more precise fault diagnosis can be achieved based on the target vibration acceleration value. In this way, this application constrains the basic evolution law of vibration signals through a physical model and then corrects the limitations of the physical model in nonlinear fault feature identification through a data-driven approach, thereby significantly improving the robustness and accuracy of fault detection under complex operating conditions such as strong electromagnetic interference.
[0111] In one exemplary embodiment, such as Figure 3 As shown, vibration fault detection of the transformer is performed based on the vibration acceleration values of each target, including steps 302 to 304. Wherein:
[0112] Step 302: Obtain the measured values of vibration acceleration of the transformer at each fault detection time.
[0113] It should be noted that comparing the measured vibration acceleration values with a fixed threshold or directly inputting them into the classification model can lead to a high false alarm rate and difficulty in identifying early, minor faults because the measured values are easily affected by complex working conditions such as load fluctuations, temperature changes, and strong electromagnetic interference.
[0114] Among them, the measured value of vibration acceleration can refer to the physical quantity of acceleration that characterizes the actual vibration intensity of the transformer tank surface at each fault detection moment, determined by the original electrical signal directly collected from the vibration sensor on the surface of the transformer tank, after calibration conversion and standardization processing.
[0115] For example, in conjunction with vibration sensors at the core and windings, vibration signals from the transformer can also be acquired via vibration sensors mounted on the surface of the transformer tank. The acquired transformer vibration signals can be directly transmitted to a terminal or stored in a designated data storage path. When performing transformer vibration fault detection, the terminal can read the transformer vibration signals acquired within a preset fault detection period from the vibration sensors mounted on the transformer tank surface or from the designated data storage path. By analyzing the transformer vibration signals, the measured vibration acceleration values of the transformer at each fault detection moment within the preset fault detection period can be determined.
[0116] Step 304: Perform vibration fault detection on the transformer based on the target residual value between the target vibration acceleration value and the measured vibration acceleration value at each fault detection time.
[0117] The target residual value refers to the difference between the measured vibration acceleration value and the target vibration acceleration value at the same fault detection time. The target residual value is used to quantify the deviation between the target vibration acceleration value after dual optimization by physical model and data-driven approach and the actual observed value of vibration acceleration on the transformer tank surface.
[0118] For example, for each fault detection moment, the difference between the measured vibration acceleration value and the target vibration acceleration value at that fault detection moment is calculated to obtain the target residual value at that fault detection moment. After calculating the target residual values for each fault detection moment within the preset fault detection period, mechanical faults such as transformer core loosening and winding deformation can be identified and their conditions assessed based on the target residual values for each fault detection moment within the preset fault detection period, using threshold discrimination, pattern recognition, or machine learning classification algorithms.
[0119] In some embodiments, a static or dynamic threshold can be set for the absolute value of the target residual or its energy in a specific frequency domain. When the target residual continuously exceeds the set threshold, a transformer vibration fault is determined to exist.
[0120] In some embodiments, the statistical characteristics (e.g., mean, variance, etc.), time series patterns (e.g., whether periodic fluctuations occur), or spectral characteristics of the target residual value sequence can be analyzed and matched with predefined fault modes to determine the fault type based on the matching results.
[0121] In some embodiments, the target residual value sequence or the feature values extracted from it (such as time-domain statistics, frequency-domain components, etc.) can be used as input feature vectors and input into a pre-trained fault classification model (such as support vector machine, random forest, etc.) to output fault diagnosis results through the fault classification model.
[0122] In this embodiment, on the one hand, the target vibration acceleration value itself has already undergone preliminary suppression and state optimization of interference in the original signal through physical models and data-driven methods. Calculating the residual based on this allows the residual signal to better highlight real anomalies that do not conform to model expectations, while filtering out a large amount of common-mode interference caused by operating condition fluctuations. On the other hand, the residual directly reflects the deviation between the actual transformer system and the optimized theoretical model. This deviation is directly related to changes in the internal mechanical state, thus significantly improving the signal-to-noise ratio and separability of fault characteristics. Therefore, fault detection based on the target residual value can achieve more accurate and earlier fault identification, effectively reduce false alarms caused by environmental noise and changes in operating conditions, and improve the reliability and practicality of fault detection.
[0123] In an exemplary embodiment, vibration fault detection of the transformer based on the residual value between the target vibration acceleration value and the measured vibration acceleration value at each fault detection time includes:
[0124] Based on the target vibration acceleration value and the measured vibration acceleration value at each fault detection time, the target residual value of the preset frequency band is detected; based on the deviation between the target residual value and the residual threshold corresponding to the preset frequency band, the confidence level of the transformer having a target vibration fault corresponding to the preset frequency band is determined; if the confidence level exceeds the preset confidence threshold, it is determined that the transformer has a target vibration fault; if the confidence level does not exceed the preset confidence threshold, it is determined that the transformer does not have a target vibration fault.
[0125] It should be noted that different fault types have different characteristic frequency distributions. Full-band evaluation will dilute the sensitive characteristics of specific faults, resulting in insufficient detection sensitivity for early and weak faults and an inability to effectively distinguish fault types.
[0126] The preset frequency band refers to a frequency range that is pre-defined based on the physical mechanism of transformer faults and is strongly correlated with a specific type of fault. There can be one or more preset frequency bands, and for each preset channel, the same method can be used to calculate the target residual value and determine the fault sequentially.
[0127] The residual threshold can refer to a limit value set for each preset frequency band to determine whether the residual of that frequency band is significant. The residual threshold can be a fixed value or dynamically adjusted according to operating conditions.
[0128] Confidence level is used to characterize the probability or degree of certainty that a transformer has a certain vibration fault.
[0129] For example, for each fault detection time, a first spectrum diagram is obtained by performing spectral analysis on the measured vibration acceleration value at that fault detection time, and a second spectrum diagram is obtained by performing spectral analysis on the target vibration acceleration value at that fault detection time. Then, for each preset frequency band related to a specific fault, the amplitude of the first frequency point in that frequency band and the amplitude of the second frequency point in that frequency band are calculated respectively. The absolute value of the difference between the amplitude of the first frequency point and the amplitude of the second frequency point are further calculated as the residual value of that frequency band at the fault detection time. After determining the residual value of that frequency band at each fault detection time within the preset fault detection period, the residual values are averaged, and the average value is determined as the target residual value of that frequency band. For each preset frequency band and its corresponding target vibration fault, the target residual value of that frequency band is compared with a residual threshold set in advance for that frequency band, and the deviation between the two is calculated. This deviation can be a simple difference, for example, deviation = residual value - threshold, or it can be a normalized relative value, for example, deviation = (residual value - threshold) / threshold. Then, the calculated bias value is mapped to a specific confidence level using a predefined confidence function. For example, the larger the bias, the higher the confidence level; if the bias is negative, that is, the residual value is below the threshold, the confidence level may be 0 or a pre-set lower confidence limit.
[0130] After calculating the confidence level for each target vibration fault, these confidence levels are compared one by one with a pre-set confidence threshold, either globally or specifically for that fault type. For any fault type, if the confidence level exceeds the pre-set confidence threshold, the transformer is determined to have the target vibration fault; otherwise, if the confidence level does not exceed the pre-set confidence threshold, the transformer is determined not to have the target vibration fault.
[0131] In this embodiment, by calculating the target residual value corresponding to the preset frequency band, the analysis focus is on the frequency components directly related to a specific fault type, which effectively enhances the signal-to-noise ratio of the fault characteristics and makes the early and weak fault characteristics stand out.
[0132] In an exemplary embodiment, the target residual value of a preset frequency band is detected based on the target vibration acceleration value and the measured vibration acceleration value at each fault detection time, including:
[0133] The target frequency band acceleration value of the preset frequency band is extracted from each target vibration acceleration value, and the measured frequency band acceleration value of the preset frequency band is extracted from each vibration acceleration measured value. The difference between the target frequency band acceleration value and the measured frequency band acceleration value of the preset frequency band at each fault detection time is determined as the initial residual value of the preset frequency band at each fault detection time. The average value of each initial residual value is determined as the target residual value of the preset frequency band.
[0134] Among them, the target frequency band acceleration value can refer to the vibration acceleration component belonging to that frequency band separated from the target vibration acceleration value for a certain preset frequency band.
[0135] The measured value of frequency band acceleration refers to the vibration acceleration component belonging to that frequency band, which is separated from the measured value of vibration acceleration for a certain preset frequency band.
[0136] For example, a digital bandpass filter can be designed, with its passband range set to a preset frequency band. This filter is then used to filter a time series of target vibration acceleration values to obtain a first time-domain sub-signal, and the same filter is used to filter a time series of measured vibration acceleration values to obtain a second time-domain sub-signal. The effective value or average amplitude of the first time-domain sub-signal is then calculated, and this result is used as the target frequency band acceleration value for that frequency band. Similarly, the effective value or average amplitude of the second time-domain sub-signal is calculated, and this result is used as the measured frequency band acceleration value for that frequency band. For each fault detection time, the difference between the measured frequency band acceleration value and the target frequency band acceleration value at that fault detection time is calculated, and this difference is determined as the initial residual value for that frequency band at that fault detection time. The arithmetic mean of the initial residual values corresponding to each fault detection time within a preset fault detection period is then calculated, and the resulting average value is determined as the target residual value for that preset frequency band. The target residual value can reflect the deviation level between the actual vibration intensity of the transformer in the preset frequency band within the preset fault detection period and the ideal vibration intensity of the transformer in the ideal state where there is no fault.
[0137] In this embodiment, by averaging the initial residual values at each fault detection time, abnormal fluctuations caused by random noise and instantaneous impacts can be effectively smoothed out, highlighting the systematic deviations caused by persistent faults, thereby reducing false alarms caused by instantaneous interference.
[0138] In an exemplary embodiment, for any fault detection moment, physical model detection error compensation is performed on the initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at the fault detection moment, including:
[0139] Obtain the transformer's state data at the time of fault detection. The state data includes at least the load current value and the oil temperature value. Based on the state data, perform physical model detection error compensation on the initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at the time of fault detection.
[0140] It should be noted that relying solely on the initial vibration acceleration value or its historical sequence for fault detection makes it difficult for the fault detection results to adapt to changes in key operating conditions such as load and temperature. Especially when operating conditions fluctuate drastically, it is difficult to accurately correct the systematic deviations in the physical model caused by time-varying parameters (such as changes in winding electrodynamics with load current and changes in material properties with temperature), thus significantly reducing the accuracy of fault detection.
[0141] Status data refers to the set of physical quantities that reflect the operating conditions of the transformer at the moment of target fault detection. Status data includes at least load current and oil temperature values, and may also include voltage, tap changer position, etc.
[0142] The load current value reflects the electrical energy transmission load of the transformer at the time of target fault detection. The first oil temperature value reflects the thermal state of the transformer at the time of target fault detection.
[0143] For example, the transformer's state data at each fault detection moment can be read from monitoring modules such as current transformers and temperature sensors, or obtained from a specified data storage path. A data-driven approach is then used to correct the initial vibration acceleration value at each fault detection moment using the state data, thus obtaining the target vibration acceleration value of the transformer at each fault detection moment.
[0144] In this embodiment, considering that the vibration characteristics of a transformer are strongly correlated with load and temperature (e.g., the winding electrodynamic force is proportional to the square of the current, and material properties and mechanical clearances are affected by temperature), compensating for the initial vibration acceleration value based on the transformer's state data can dynamically and accurately correct the systematic deviations of the transformer's vibration physical model under different operating conditions. This makes the compensation process no longer static or blind, but rather capable of adapting to operating conditions. Therefore, the obtained target vibration acceleration value can maintain higher accuracy under different loads and operating temperatures, more realistically reflecting the transformer's vibration state. This provides more stable, reliable, and high-fidelity input data for subsequent fault detection based on the target vibration acceleration value, thereby improving the accuracy and robustness of transformer fault detection.
[0145] In an exemplary embodiment, physical model detection error compensation is performed on the initial vibration acceleration value based on the state data to obtain the target vibration acceleration value of the transformer at the time of target fault detection, including:
[0146] The state data and the initial vibration acceleration value at the time of target fault detection are input into the error compensation model to obtain the error compensation value at the time of target fault detection. The error compensation model is obtained by training the recurrent fuzzy neural network to be trained based on the historical state data of the transformer and the historical value of the initial vibration acceleration. The error compensation value and the initial vibration acceleration value at the time of target fault detection are aggregated to obtain the target vibration acceleration value of the transformer at the time of target fault detection.
[0147] It should be noted that if an error compensation method based on a fixed formula or a simple linear relationship is used, it is difficult to accurately characterize the model residuals in the transformer vibration system caused by complex nonlinear and time-varying characteristics (such as aging of core materials, softening of insulation materials, etc.) and dynamic coupling between states. It is also difficult to fully learn and adapt to the complex dynamic mapping relationship between the residuals and multi-dimensional state data, resulting in limited improvement in accuracy after compensation.
[0148] Among them, the error compensation model can refer to a mathematical model used to predict the error of the transformer vibration physical model at the fault detection time based on the input state data and initial vibration acceleration value, and output the corresponding error compensation value.
[0149] The error compensation model is obtained by training the recurrent fuzzy neural network to be trained based on the historical state data of the transformer and the historical value of the initial vibration acceleration.
[0150] Recurrent fuzzy neural networks (RNNs) are a hybrid model that combines the interpretability of fuzzy logic systems with the dynamic temporal memory capabilities of recurrent neural networks. RNNs can handle complex system modeling problems that exhibit fuzziness, nonlinearity, and temporal dependence.
[0151] In one exemplary embodiment, the input to the error compensation model may include the initial vibration acceleration value a at any fault detection time t. phy The load current value I(t) and oil temperature value T(t) are given. The output of the error compensation model is the error compensation value Δa(t) at the fault detection moment.
[0152] The error compensation model can include a fuzzification layer, a rule layer, a normalization layer, a recursive layer, and a defuzzification layer. The fuzzification layer transforms the input data into fuzzy semantic concepts (e.g., "high current," "medium temperature," etc.); the rule layer calculates the activation strength of each fuzzy rule; the normalization layer normalizes the activation strength of the rules; the recursive layer updates and transmits the internal memory state of each rule, which is crucial for the model's temporal memory capability; and the defuzzification layer aggregates the fuzzy inference results, outputting a clear error compensation value.
[0153] For example, before detecting transformer vibration faults, historical state data, historical values of initial vibration acceleration, and historical measured values of vibration acceleration can be collected for the transformer during periods when its health or condition is known. Using the historical state data and historical values of initial vibration acceleration as training samples, and with the goal of minimizing the difference between the compensated target historical vibration acceleration value and the historical measured vibration acceleration value, a pre-defined recurrent fuzzy neural network is trained under supervision. After training, the trained recurrent fuzzy neural network is used as the error compensation model.
[0154] In online applications, the currently obtained state data and initial vibration acceleration values are input into the error compensation model. The error compensation model performs forward processing through fuzzy inference and recursive memory mechanisms, and outputs an error compensation value. This error compensation value is then aggregated with the initial vibration acceleration value at the same fault detection time, and the aggregation result is used as the target vibration acceleration value of the transformer at the fault detection time. In some embodiments, the aggregation operation method includes at least one of summation, weighted summation, etc.
[0155] In this embodiment, the recurrent fuzzy neural network integrates the semantic rule expression capability of fuzzy logic with the dynamic temporal memory capability of recurrent neural networks. It can automatically learn and internalize the complex, nonlinear mapping relationship between errors, multidimensional state data, and historical vibration trends from historical data. The recurrent layers of the recurrent fuzzy neural network enhance the ability to capture time-varying features through historical information, maintaining stable diagnostic performance even in scenarios with strong electromagnetic interference and scarce samples. Through the memory function of the recurrent fuzzy neural network, time-varying noise can be effectively suppressed. As noise intensity increases, the diagnostic misjudgment rate is significantly reduced, thus effectively distinguishing noise interference from early fault characteristics.
[0156] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0157] Based on the same inventive concept, this application also provides a transformer vibration fault detection device for implementing the transformer vibration fault detection method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the transformer vibration fault detection device provided below can be found in the limitations of the transformer vibration fault detection method described above, and will not be repeated here.
[0158] In one exemplary embodiment, such as Figure 4 As shown, a transformer vibration fault detection device is provided, comprising: an acquisition module 402, a determination module 404, a compensation module 406, and a detection module 408, wherein:
[0159] The acquisition module 402 is used to acquire the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within a preset fault detection cycle.
[0160] The determination module 404 is used to determine the initial vibration acceleration value of the transformer at each fault detection time based on the core vibration signal value and winding vibration signal value corresponding to each fault detection time using the transformer vibration physical model.
[0161] The compensation module 406 is used to perform physical model detection error compensation on each initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at each fault detection time.
[0162] The detection module 408 is used to detect vibration faults in the transformer based on the vibration acceleration values of each target.
[0163] In one exemplary embodiment, the detection module 408 is further configured to:
[0164] Obtain the measured values of the transformer's vibration acceleration at each fault detection moment;
[0165] The transformer is subjected to vibration fault detection based on the target residual value between the target vibration acceleration value and the measured vibration acceleration value at each fault detection time.
[0166] In one exemplary embodiment, the detection module 408 is further configured to:
[0167] Based on the target vibration acceleration value and the measured vibration acceleration value at each fault detection time, detect the target residual value of the preset frequency band;
[0168] Based on the deviation between the target residual value and the residual threshold corresponding to the preset frequency band, the confidence level of the transformer having the target vibration fault corresponding to the preset frequency band is determined;
[0169] If the confidence level exceeds the preset confidence threshold, the transformer is determined to have a target vibration fault.
[0170] If the confidence level does not exceed the preset confidence threshold, it is determined that the transformer does not have the target vibration fault.
[0171] In one exemplary embodiment, the detection module 408 is further configured to:
[0172] The target frequency band acceleration value of the preset frequency band is extracted from each target vibration acceleration value, and the frequency band acceleration value of the preset frequency band is extracted from each vibration acceleration measured value.
[0173] The difference between the target frequency band acceleration value and the measured frequency band acceleration value at each fault detection time is determined as the initial residual value of the preset frequency band at each fault detection time.
[0174] The average value of each initial residual is determined as the target residual value for the preset frequency band.
[0175] In an exemplary embodiment, for any fault detection moment, the compensation module 406 is further configured to:
[0176] Acquire the transformer status data at the time of fault detection. The status data should include at least the load current value and the oil temperature value.
[0177] Based on the state data, the initial vibration acceleration value is compensated for by physical model detection error to obtain the target vibration acceleration value of the transformer at the fault detection time.
[0178] In one exemplary embodiment, the compensation module 406 is further configured to:
[0179] The state data and the initial vibration acceleration value at the fault detection time are input into the error compensation model to obtain the error compensation value at the fault detection time. The error compensation model is obtained by training the recurrent fuzzy neural network to be trained based on the transformer's historical state data and the historical initial vibration acceleration value.
[0180] By combining the error compensation value and the initial vibration acceleration value at the time of fault detection, the target vibration acceleration value of the transformer at the time of fault detection is obtained.
[0181] Each module in the aforementioned transformer vibration fault detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.
[0182] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a method for detecting transformer vibration faults. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.
[0183] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0184] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.
[0185] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.
[0186] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.
[0187] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0188] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.
[0189] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0190] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for detecting transformer vibration faults, characterized in that, The method includes: Acquire the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within a preset fault detection cycle; Using a transformer vibration physical model, the initial vibration acceleration value of the transformer at each fault detection time is determined based on the core vibration signal value and winding vibration signal value corresponding to each fault detection time. Physical model detection error compensation is performed on each of the initial vibration acceleration values to obtain the target vibration acceleration values of the transformer at each of the fault detection times. Vibration fault detection is performed on the transformer based on the target vibration acceleration values.
2. The method according to claim 1, characterized in that, The vibration fault detection of the transformer based on the target vibration acceleration values includes: Obtain the measured values of vibration acceleration of the transformer at each of the fault detection times; The transformer is subjected to vibration fault detection based on the target residual value between the target vibration acceleration value and the measured vibration acceleration value at each fault detection time.
3. The method according to claim 2, characterized in that, The step of detecting vibration faults in the transformer based on the residual value between the target vibration acceleration value and the measured vibration acceleration value at each fault detection time includes: Based on the target vibration acceleration value and the measured vibration acceleration value at each fault detection time, the target residual value of the preset frequency band is detected. The confidence level of the transformer having a target vibration fault corresponding to the preset frequency band is determined based on the deviation between the target residual value and the residual threshold corresponding to the preset frequency band. If the confidence level exceeds a preset confidence threshold, it is determined that the transformer has a target vibration fault. If the confidence level does not exceed the preset confidence threshold, it is determined that the transformer does not have a target vibration fault.
4. The method according to claim 3, characterized in that, The step of detecting the target residual value of a preset frequency band based on the target vibration acceleration value and the measured vibration acceleration value at each of the fault detection times includes: The target frequency band acceleration value of the preset frequency band is extracted from each of the target vibration acceleration values, and the frequency band acceleration measured value of the preset frequency band is extracted from each of the measured vibration acceleration values. The difference between the target frequency band acceleration value and the measured frequency band acceleration value at each fault detection time is determined as the initial residual value of the preset frequency band at each fault detection time. The average value of each initial residual value is determined as the target residual value for the preset frequency band.
5. The method according to any one of claims 1 to 4, characterized in that, For any fault detection moment, the initial vibration acceleration value is compensated for by physical model detection error to obtain the target vibration acceleration value of the transformer at the fault detection moment, including: Acquire the status data of the transformer at the time of fault detection, the status data including at least the load current value and the oil temperature value; Based on the state data, physical model detection error compensation is performed on the initial vibration acceleration value to obtain the target vibration acceleration value of the transformer at the fault detection time.
6. The method according to claim 5, characterized in that, The step of performing physical model detection error compensation on the initial vibration acceleration value based on the state data to obtain the target vibration acceleration value of the transformer at the fault detection time includes: The state data and the initial vibration acceleration value at the fault detection time are input into the error compensation model to obtain the error compensation value at the fault detection time. The error compensation model is obtained by training the recurrent fuzzy neural network to be trained based on the historical state data of the transformer and the historical initial vibration acceleration value. By aggregating the error compensation value and the initial vibration acceleration value at the fault detection time, the target vibration acceleration value of the transformer at the fault detection time is obtained.
7. A transformer vibration fault detection device, characterized in that, The device includes: The acquisition module is used to acquire the core vibration signal value and winding vibration signal value collected at at least one fault detection moment within a preset fault detection cycle. The determination module is used to determine the initial vibration acceleration value of the transformer at each of the fault detection times by using the transformer vibration physical model and based on the core vibration signal value and winding vibration signal value corresponding to each of the fault detection times. The compensation module is used to perform physical model detection error compensation on each of the initial vibration acceleration values to obtain the target vibration acceleration value of the transformer at each of the fault detection times. The detection module is used to detect vibration faults in the transformer based on the vibration acceleration values of each target.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.