Multi-dimensional state sensing and diagnosing system for transformer

Abstract

The invention discloses a transformer multi-dimensional state sensing and diagnosing system, which comprises a multi-dimensional state sensing module for acquiring transformer multi-dimensional data in real time through various sensors; the data acquisition and processing module performs processing such as synchronous sampling and filtering on the data and extracts features; the state evaluation and fault diagnosis module evaluates a state and diagnoses a fault by using a three-dimensional evaluation model and a CNN-LSTM fusion model; the self-adaptive early warning module dynamically adjusts a threshold value according to an evaluation result and sends out multi-level early warning; the communication and interaction module realizes data encryption transmission, has a visual interface, and supports interaction diagnosis and remote configuration updating. The system can comprehensively collect data, accurately diagnose faults and efficiently early warn, is convenient to operate and maintain, and improves the operation reliability and the operation and maintenance efficiency of the transformer.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer diagnosis systems, and in particular to a transformer multi-dimensional state perception and diagnosis system. Background Art

[0002] In the field of power systems, the stable operation of transformers plays a decisive role in the reliability of the entire power grid. Figure 1 As shown, current transformer condition monitoring and diagnosis technologies face numerous challenges. Existing technologies, such as the oil chromatography-based transformer fault diagnosis system described in Chinese patent CN112393856A, exhibit significant limitations. Furthermore, research such as "Intelligent Perception of Power Transformer Health Status Assessment Based on Multi-Source Information Fusion" also highlights the shortcomings of current technologies.

[0003] CN112393856A, a transformer fault diagnosis system based on oil chromatography, focuses only on oil chromatography data and lacks monitoring of other key operating parameters of the transformer, such as mechanical vibration, partial discharge, and other information. This makes it impossible to fully assess the operating status of the transformer, and there are monitoring blind spots. The technology involved in "Intelligent Perception Power Transformer Health Status Assessment Based on Multi-Source Information Fusion" attempts multi-source information fusion and adopts ultra-high frequency, high frequency, ultrasonic and other partial discharge channels and vibration signal monitoring methods, but there is still room for improvement in data processing depth, diagnostic accuracy, and interactive functions. Its fault diagnosis mainly relies on traditional algorithms, and the accuracy of identifying complex fault modes is difficult to meet actual needs; in terms of interaction, it lacks intuitive and efficient visualization and remote interactive diagnosis functions, and cannot provide convenient and comprehensive equipment status information to operation and maintenance personnel. Summary of the Invention

[0004] The purpose of the present invention is to provide a transformer multi-dimensional state perception and diagnosis system to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a transformer multi-dimensional state perception and diagnosis system, comprising:

[0006] S1. Multi-dimensional state perception module: It consists of a variety of sensors deployed on the transformer body and its surrounding environment, including a high-precision fiber Bragg grating temperature sensor for real-time monitoring of the temperature of the transformer winding, core, and oil body, with a measurement accuracy of ±0.1°C and a measurement range of -40°C to 120°C; a piezoelectric vibration sensor, installed on the transformer casing, which can detect vibrations in the frequency range of 0-20kHz with a sensitivity of 100mV / g; an ultra-high frequency partial discharge sensor with a detection frequency range of 300MHz-3GHz, which is used to detect partial discharge phenomena inside the transformer; and an oil spectrum sensor that can detect multiple gases including hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide with a detection accuracy of ppm level. These sensors collect data on the transformer's temperature distribution, mechanical vibration, partial discharge, dissolved gas in the oil, and other data in real time to form a multi-dimensional state data set.

[0007] S2, Data Acquisition and Processing Module: Connected to the multi-dimensional state perception module, it includes a data acquisition unit and a data processing unit. The data acquisition unit synchronously samples, filters, amplifies, and digitally converts the multi-dimensional data collected by the sensors. It samples the sensor data synchronously with a sampling frequency of ≥1kHz and supports Modbus protocol transmission. The data processing unit performs wavelet denoising and normalization on the data and extracts characteristic parameters using an improved variational mode decomposition algorithm.

[0008] S3, state assessment and fault diagnosis module: This module includes a state assessment unit and a fault diagnosis unit. Based on the analytic hierarchy process, a three-dimensional assessment model is constructed that includes temperature field gradient, vibration energy entropy, discharge characteristics, and oil spectrum ratio. The fault diagnosis unit uses a CNN-LSTM fusion model, which inputs a standardized feature vector and outputs the fault type and confidence level.

[0009] S4, Adaptive Early Warning Module: Dynamically adjusts the alarm threshold based on the health status score output by the status assessment model. When the score falls below the preset safety threshold or the failure probability exceeds the critical value, a multi-level early warning signal containing fault location information is generated. When the score is less than 70 or the fault confidence level is greater than 85%, a third-level early warning is triggered. The warning signal includes GIS positioning coordinates, with a transmission delay of ≤2 seconds. Warning methods include audible and visual alarms, SMS notifications, and email reminders.

[0010] S5, Communication and Interaction Module: Adopts a multi-protocol compatible industrial-grade gateway to realize encrypted transmission of sensor data, diagnostic results and warning signals, and supports 4G / 5G / fiber hybrid networking; it also has a visual interactive interface for displaying the spatiotemporal distribution of multi-dimensional status parameters, health status evolution trends and maintenance decision recommendations, supports users to interactively diagnose equipment status through a virtual reality interface or other interactive methods, and supports remote parameter configuration and diagnostic strategy updates.

[0011] Furthermore, the high-precision fiber Bragg grating temperature sensor arranges 8-12 temperature measuring points at a spacing of 20cm±5% in the axial direction of the winding, and 4-6 temperature measuring points are arranged at a spacing of 30cm±10% between the core columns. The optical fiber adopts an electromagnetic interference-resistant double-sheath structure, the outer sheath temperature resistance level is ≥150°C, and the temperature measurement nodes are calibrated online using platinum resistance, with a calibration error of ≤±0.05°C.

[0012] Furthermore, the piezoelectric vibration sensor has a built-in temperature-frequency composite compensation module, and the compensated vibration acceleration α acorrected satisfy:

[0013] α acorrected =α raw ·[1-0.015%·(T-25)]

[0014] Where T is the internal temperature of the sensor (detection accuracy ±0.5°C), the roughness of the sensor installation surface Ra≤1.6μm, and the resonant frequency ≥25kHz.

[0015] Furthermore, the ring antenna array of the ultra-high frequency partial discharge sensor is composed of 8 ceramic substrate patch antennas, the antenna impedance matching network has a standing wave ratio of ≤1.5:1 in the 300MHz-3GHz frequency band, and the pulse capture unit can identify discharge pulses with a rise time ≥2ns, with a time resolution of 100ps.

[0016] Furthermore, the oil spectrum sensor uses an internal standard method to correct the gas concentration, and the correction formula is:

[0017]

[0018] Among them, P cal is the real-time measurement value of CO2 internal standard, p ref =5.0 PPM As a reference benchmark value, the temperature control fluctuation of the gas separation column is ≤±0.1℃.

[0019] Furthermore, the wavelet transform uses the db6 wavelet basis to perform a 5-layer decomposition, and the high-frequency coefficient denoising threshold λ is calculated as:

[0020]

[0021] Where σ is the estimated value of the noise variance, N is the signal length, and nonlinear compression is applied to the temperature data during normalization. The compression ratio gradually changes from 1:1 to 1:3 with the increase of the temperature rise rate.

[0022] Furthermore, the dynamic adjustment algorithm of the weight coefficient of the state assessment model is:

[0023] w i (t) =0.7wi (t-1) +0.3Δw i

[0024] where Δw i Calculated based on the coefficient of variation of characteristic parameters in the last 24 hours, when the weight of a dimension exceeds 0.5, an expert system review is triggered.

[0025] Furthermore, the convolutional neural network contains 4 convolutional layers, the convolution kernel sizes are 5×5, 3×3, 3×3, and 1×1 respectively, the number of channels increases in steps of 32-64-128-256, the number of hidden layer units of the LSTM network is set to 128, and the time step matches 60 sampling periods.

[0026] Furthermore, the dynamic alarm threshold update satisfies:

[0027] Threshold t =μ t -25σ t +0.2I pd

[0028] Among them I pd It is the partial discharge activity index. When the threshold is exceeded three times in a row, a multi-level alarm is activated, and the alarm delay time is ≤ 2 seconds.

[0029] Furthermore, the three-dimensional thermal map rendering engine supports 0.1°C temperature gradient visualization and discharge hotspot positioning accuracy ≤ 10cm 3 When the acetylene concentration is greater than 5ppm and the vibration energy rises by more than 5dB per hour in the 100-300kHz frequency band, a red warning signal is automatically generated.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] By deploying a variety of high-precision sensors, including high-precision fiber Bragg grating (FBG) temperature sensors, piezoelectric vibration sensors, ultra-high frequency (UHF) partial discharge (PD) sensors, and oil spectrum sensors, real-time data collection is achieved for multi-dimensional data, including transformer temperature, mechanical vibration, partial discharge (PD), and dissolved gas in oil. The high-precision measurement and specialized placement of the FBG temperature sensors enable precise temperature monitoring across the transformer. The wide frequency detection range of the UHF PD sensors effectively captures minute PD signals. The data acquisition and processing module efficiently processes multi-dimensional data, ensuring data quality through simultaneous sampling, filtering, amplification, and digitization. Preprocessing includes noise filtering and normalization, as well as the use of wavelet transforms and machine learning algorithms to extract key characteristic parameters and generate standardized state feature vectors, enhancing data usability. In the condition assessment and fault diagnosis module, a transformer condition assessment model based on key characteristic parameters quantifies health status using multiple dimensional weighting coefficients that can be dynamically adjusted. By integrating historical operating data, environmental parameters, and equipment inventory information, precise diagnosis of transformer fault type, location, severity, and probability distribution is achieved, significantly improving diagnostic accuracy and reliability. The adaptive early warning module dynamically adjusts alarm thresholds based on the health status score output by the status assessment model. When the score falls below a preset safety threshold or the failure probability exceeds a critical value, it quickly generates a multi-level early warning signal containing fault location information. Users can interactively diagnose device status through a virtual reality interface or other interactive methods, providing operators with a deeper understanding of device status. Furthermore, remote parameter configuration and diagnostic strategy updates are supported, enabling the system to quickly adjust to actual needs, improving both operational efficiency and system practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic diagram of a conventional transformer according to the present invention;

[0033] Figure 2 Schematic diagram of the system flow of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0035] See Figure 1-2 The present invention provides a technical solution: a transformer multi-dimensional state perception and diagnosis system, comprising:

[0036] S1. Multi-dimensional state perception module: It consists of a variety of sensors deployed on the transformer body and its surrounding environment, including a high-precision fiber Bragg grating temperature sensor for real-time monitoring of the temperature of the transformer winding, core, and oil body, with a measurement accuracy of ±0.1°C and a measurement range of -40°C to 120°C; a piezoelectric vibration sensor, installed on the transformer casing, which can detect vibrations in the frequency range of 0-20kHz with a sensitivity of 100mV / g; an ultra-high frequency partial discharge sensor with a detection frequency range of 300MHz-3GHz, which is used to detect partial discharge phenomena inside the transformer; and an oil spectrum sensor that can detect multiple gases including hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide with a detection accuracy of ppm level. These sensors collect data on the transformer's temperature distribution, mechanical vibration, partial discharge, dissolved gas in the oil, and other data in real time to form a multi-dimensional state data set.

[0037] Choose an oil level and temperature sensor with high-precision measurement and remote transmission capabilities, such as the YW-300, which has an oil level measurement accuracy of ±5 mm and an oil temperature measurement accuracy of ±0.5°C. Install it at the bottom side of the transformer's oil conservator, connecting it to the transformer oil via a stainless steel pipe to ensure accurate oil level and temperature measurement.

[0038] Using a highly sensitive spectral sensor, it can detect a variety of characteristic gases, including hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), and carbon dioxide (CO2), with a detection accuracy of 1ppm. The characteristic gas sensor is installed near the gas sampling port of the transformer and extracts gases from the transformer oil through a polytetrafluoroethylene sampling pipe for detection.

[0039] At the same time, use highly sensitive, broadband acoustic sensors, such as the SMA-500 piezoelectric acoustic sensor, which has a frequency response range of 20Hz-20kHz and can effectively capture the various characteristic sound signals emitted by the transformer during operation. During installation, evenly distribute the acoustic sensors at different locations on the transformer casing, such as the top center and the centers of the four sides. Secure them with dedicated rubber shock-absorbing mounting brackets and silicone shock-absorbing pads to ensure close contact between the sensors and the transformer casing, reducing signal transmission loss and external interference.

[0040] High-precision three-phase voltage and current sensors, such as the HL-500, offer voltage and current measurement accuracies of ±0.2% and ±0.2%, respectively, enabling accurate, real-time measurement of the transformer's three-phase voltage and current values. The energy sensor is installed on the transformer's incoming line, connected to the transformer's three-phase circuits via well-insulated copper conductors. Ensure secure connections and good contact to prevent measurement errors caused by excessive contact resistance.

[0041] S2, Data Acquisition and Processing Module: Connected to the multi-dimensional state perception module, it includes a data acquisition unit and a data processing unit. The data acquisition unit synchronously samples, filters, amplifies, and digitally converts the multi-dimensional data collected by the sensors. It samples the sensor data synchronously with a sampling frequency of ≥1kHz and supports Modbus protocol transmission. The data processing unit performs wavelet denoising and normalization on the data and extracts characteristic parameters using an improved variational mode decomposition algorithm.

[0042] The transformer's characteristic acoustic signal is collected at a first preset interval, such as every 10 minutes. During the acquisition process, the acoustic signal is digitized, converting the analog signal into a digital signal. The sampling frequency is set to 44.1kHz to facilitate subsequent transmission and processing. Furthermore, to prevent data loss and ensure data integrity, the acoustic sensor is equipped with a 1GB cache memory. In the event of a network transmission failure, the collected data can be temporarily stored and transmitted again after the network is restored.

[0043] When no activation signal is received, the power sensor and oil-gas combination sensor remain dormant to reduce energy consumption. When the edge computing terminal sends a first activation signal based on the acoustic sensor's first diagnostic result, the power sensor and oil-gas combination sensor are activated. Upon activation, the power sensor collects three-phase voltage and current values in real time and transmits them 10 times per second. The oil level and temperature sensors in the oil-gas combination sensor collect oil level and temperature data every five minutes, and the characteristic gas sensor continuously collects characteristic gas signals from the transformer oil and transmits this data in real time to the edge computing terminal.

[0044] S3, state assessment and fault diagnosis module: This module includes a state assessment unit and a fault diagnosis unit. Based on the analytic hierarchy process, a three-dimensional assessment model is constructed that includes temperature field gradient, vibration energy entropy, discharge characteristics, and oil spectrum ratio. The fault diagnosis unit uses a CNN-LSTM fusion model, which inputs a standardized feature vector and outputs the fault type and confidence level.

[0045] The first built-in diagnostic model takes multiple characteristic sound signals as input. The model structure utilizes a convolutional neural network, a deep learning technique. The network consists of three convolutional layers with kernel sizes of 3x3, 5x5, and 3x3, respectively, all with a stride of 1 and the same padding; two pooling layers with a kernel size of 2x2 and a stride of 2; and one fully connected layer. During model training, 1,000 sets of historical sound signal data, including those from normal operation and various fault states, were used to annotate the data categories. The Adam optimizer was used to adjust the model's weight parameters, with a learning rate set to 0.001. Training was performed for 50 epochs, enabling the model to accurately identify fault features in sound signals. When the currently collected characteristic sound signal is input, the model outputs the first diagnostic result. If the diagnostic result is abnormal, the first predetermined condition is met, and the edge computing terminal sends the first activation signal.

[0046] The second diagnostic model uses a probabilistic neural network (PNN) model, taking multiple characteristic gas signals as input. When using the PNN model, the preprocessed characteristic gas signals are input into the model's input layer. The PNN model performs pattern classification based on the Bayesian classification rule and the Parzen probability density function. The Gaussian kernel width of the pattern layer nodes is set to 0.1. By adjusting the model parameters, the model can accurately determine changes in the characteristic gas in the transformer oil and thus diagnose whether the transformer is faulty. When the oil-gas combination sensor is activated and collects the characteristic gas signals, the edge computing terminal inputs the preprocessed characteristic gas signals into the second transformer diagnostic model to obtain the second diagnostic result.

[0047] S4, Adaptive Early Warning Module: Dynamically adjusts the alarm threshold based on the health status score output by the status assessment model. When the score falls below the preset safety threshold or the failure probability exceeds the critical value, a multi-level early warning signal containing fault location information is generated. When the score is less than 70 or the fault confidence level is greater than 85%, a third-level early warning is triggered. The warning signal includes GIS positioning coordinates, with a transmission delay of ≤2 seconds. Warning methods include audible and visual alarms, SMS notifications, and email reminders.

[0048] When a normal diagnostic signal is received, the user control layer can appropriately increase the first preset period, such as adjusting the acoustic signal collection from 10 minutes to 15 minutes, to reduce the amount of data collected and system energy consumption. If a normal diagnostic signal is not received for three consecutive times, the user control layer can maintain or shorten the first preset period to 5-minute collection intervals to monitor the transformer status more frequently. Users can also manually control the start and stop of the sensor layer through the user control layer. When the transformer is undergoing special operations or maintenance, some sensors can be temporarily stopped to avoid unnecessary data collection and interference.

[0049] The user control layer identifies the transformer's status based on the first and second diagnostic results, as well as the transformer's electrical energy parameters obtained by the power sensor. Rule-based reasoning or artificial intelligence algorithms can be used for status identification. Taking rule-based reasoning as an example, a series of rules are set. For example, when the first diagnostic result indicates the presence of an abnormal sound, and the second diagnostic result indicates that the acetylene (C2H2) concentration exceeds 5ppm, and the three-phase current imbalance exceeds 10%, the transformer is judged to be faulty. Based on the identification results, the user control layer provides decision support to users, such as generating alarm information to inform operation and maintenance personnel of the transformer's abnormal condition through audio and visual alarms, text message notifications, and email reminders. It also provides fault handling suggestions, recommending appropriate maintenance measures and repair times based on the fault type and severity, helping operation and maintenance personnel to promptly handle the fault and ensure the safe and stable operation of the transformer.

[0050] S5, Communication and Interaction Module: Adopts a multi-protocol compatible industrial-grade gateway to realize encrypted transmission of sensor data, diagnostic results and warning signals, and supports 4G / 5G / fiber hybrid networking; it also has a visual interactive interface for displaying the spatiotemporal distribution of multi-dimensional status parameters, health status evolution trends and maintenance decision recommendations, supports users to interactively diagnose equipment status through a virtual reality interface or other interactive methods, and supports remote parameter configuration and diagnostic strategy updates.

[0051] High-precision fiber Bragg grating temperature sensors are arranged with 8-12 temperature measurement points at a spacing of 20cm±5% in the winding axis, and 4-6 temperature measurement points are arranged at a spacing of 30cm±10% between the core columns. The optical fiber adopts an anti-electromagnetic interference double-sheath structure, and the outer sheath temperature resistance level is ≥150℃. The temperature measurement nodes are calibrated online using platinum resistance, and the calibration error is ≤±0.05℃.

[0052] The piezoelectric vibration sensor has a built-in temperature-frequency composite compensation module, and the compensated vibration acceleration α acorrected satisfy:

[0053] α acorrected =α raw ·[1-0.015%·(T-25)]

[0054] Where T is the internal temperature of the sensor (detection accuracy ±0.5°C), the roughness of the sensor installation surface Ra≤1.6μm, and the resonant frequency ≥25kHz.

[0055] The ring antenna array of the UHF partial discharge sensor consists of 8 ceramic substrate patch antennas. The antenna impedance matching network has a standing wave ratio of ≤1.5:1 in the 300MHz-3GHz frequency band. The pulse capture unit can identify discharge pulses with a rise time of ≥2ns, with a time resolution of 100ps.

[0056] The oil spectrum sensor uses the internal standard method to correct the gas concentration. The correction formula is:

[0057]

[0058] Among them, P cal is the real-time measurement value of CO2 internal standard, P ref =5.0 PPM As a reference benchmark value, the temperature control fluctuation of the gas separation column is ≤±0.1℃.

[0059] The wavelet transform uses the db6 wavelet basis to perform a 5-layer decomposition, and the high-frequency coefficient denoising threshold λ is calculated as:

[0060]

[0061] Where σ is the estimated value of the noise variance, N is the signal length, and nonlinear compression is applied to the temperature data during normalization. The compression ratio gradually changes from 1:1 to 1:3 with the increase of the temperature rise rate.

[0062] The dynamic adjustment algorithm of the weight coefficient of the state assessment model is:

[0063] w i (t) =0.7w i (t-1) +0.3Δw i

[0064] where Δw i Calculated based on the coefficient of variation of characteristic parameters in the last 24 hours, when the weight of a dimension exceeds 0.5, an expert system review is triggered.

[0065] The convolutional neural network contains 4 convolutional layers with convolution kernel sizes of 5×5, 3×3, 3×3, and 1×1 respectively. The number of channels increases from 32 to 64 to 128 to 256. The number of hidden layer units in the LSTM network is set to 128, and the time step matches 60 sampling periods.

[0066] Dynamic alarm threshold updates satisfy:

[0067] Threshold t =μ t -2.5σ t +0.2I pd

[0068] Among them I pd It is the partial discharge activity index. When the threshold is exceeded three times in a row, a multi-level alarm is activated, and the alarm delay time is ≤ 2 seconds.

[0069] The 3D thermal map rendering engine supports 0.1°C temperature gradient visualization, and the discharge hotspot positioning accuracy is ≤10cm. 3When the acetylene concentration is greater than 5ppm and the vibration energy rises by more than 5dB per hour in the 100-300kHz frequency band, a red warning signal is automatically generated.

[0070] During the initial system operation, multiple transformers were monitored in real time. During operation, the acoustic sensor detected an abnormal sound signal on one transformer. The first transformer diagnostic model on the edge computing terminal identified a possible fault and immediately sent a first activation signal, activating the power sensor and the combined oil-gas sensor. The characteristic gas signal collected by the combined oil-gas sensor indicated a gradual increase in the acetylene (C2H2) concentration in the transformer oil, reaching 8 ppm. The second transformer diagnostic model further confirmed an internal discharge fault. The deep learning layer aggregated and analyzed the diagnostic results from the edge computing terminal and found that the difference in diagnostic results between two consecutive cycles exceeded a preset value. The deep learning layer then sent a feedback signal to the cloud feedback layer. The cloud feedback layer updated the edge computing terminal's diagnostic model based on the feedback signal, improving its accuracy. Upon receiving the fault alarm, the user control layer promptly notified the operations and maintenance personnel for action. Based on the fault diagnosis results and recommended actions provided by the system, the operations and maintenance personnel inspected the transformer and discovered a localized discharge point within the transformer. After repairs, the transformer resumed normal operation.

Claims

1. A transformer multi-dimensional state perception and diagnosis system, characterized in that: include: S1. Multi-dimensional state perception module: It consists of a variety of sensors deployed on the transformer body and its surrounding environment, including a high-precision fiber Bragg grating temperature sensor for real-time monitoring of the temperature of the transformer winding, core, and oil body, with a measurement accuracy of ±0.1°C and a measurement range of -40°C to 120°C; a piezoelectric vibration sensor, installed on the transformer casing, which can detect vibrations in the frequency range of 0-20kHz with a sensitivity of 100mV / g; an ultra-high frequency partial discharge sensor with a detection frequency range of 300MHz-3GHz, which is used to detect partial discharge phenomena inside the transformer; and an oil spectrum sensor that can detect multiple gases including hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide with a detection accuracy of ppm level. These sensors collect data on the transformer's temperature distribution, mechanical vibration, partial discharge, dissolved gas in the oil, and other data in real time to form a multi-dimensional state data set. S2, Data Acquisition and Processing Module: Connected to the multi-dimensional state perception module, it includes a data acquisition unit and a data processing unit. The data acquisition unit synchronously samples, filters, amplifies, and digitally converts the multi-dimensional data collected by the sensors. It samples the sensor data synchronously with a sampling frequency of ≥1kHz and supports Modbus protocol transmission. The data processing unit performs wavelet denoising and normalization on the data and extracts characteristic parameters using an improved variational mode decomposition algorithm. S3, state assessment and fault diagnosis module: This module includes a state assessment unit and a fault diagnosis unit. Based on the analytic hierarchy process, a three-dimensional assessment model is constructed that includes temperature field gradient, vibration energy entropy, discharge characteristics, and oil spectrum ratio. The fault diagnosis unit uses a CNN-LSTM fusion model, which inputs a standardized feature vector and outputs the fault type and confidence level. S4, Adaptive Early Warning Module: Dynamically adjusts the alarm threshold based on the health status score output by the status assessment model. When the score falls below the preset safety threshold or the failure probability exceeds the critical value, a multi-level early warning signal containing fault location information is generated. When the score is less than 70 or the fault confidence level is greater than 85%, a third-level early warning is triggered. The warning signal includes GIS positioning coordinates, with a transmission delay of ≤2 seconds. Warning methods include audible and visual alarms, SMS notifications, and email reminders. S5, Communication and Interaction Module: Adopts a multi-protocol compatible industrial-grade gateway to realize encrypted transmission of sensor data, diagnostic results and warning signals, and supports 4G / 5G / fiber hybrid networking; it also has a visual interactive interface for displaying the spatiotemporal distribution of multi-dimensional status parameters, health status evolution trends and maintenance decision recommendations, supports users to interactively diagnose equipment status through a virtual reality interface or other interactive methods, and supports remote parameter configuration and diagnostic strategy updates.

2. The transformer multi-dimensional state perception and diagnosis system according to claim 1 is characterized in that: The high-precision fiber Bragg grating temperature sensor has 8-12 temperature measurement points arranged at a spacing of 20cm±5% in the winding axis, and 4-6 temperature measurement points arranged at a spacing of 30cm±10% between the core columns. The optical fiber adopts an electromagnetic interference-resistant double-sheath structure, and the outer sheath has a temperature resistance level of ≥150°C. The temperature measurement nodes are calibrated online using platinum resistance, and the calibration error is ≤±0.05°C.

3. The transformer multi-dimensional state perception and diagnosis system according to claim 1, characterized in that: The piezoelectric vibration sensor has a built-in temperature-frequency composite compensation module, and the compensated vibration acceleration α acorrected satisfy: α acorrected ＝α raw ·[1-0.015％·(T-25)] Where T is the internal temperature of the sensor (detection accuracy ±0.5°C), the roughness of the sensor installation surface Ra≤1.6μm, and the resonant frequency ≥25kHz.

4. The transformer multi-dimensional state perception and diagnosis system according to claim 1, characterized in that: The ring antenna array of the ultra-high frequency partial discharge sensor consists of 8 ceramic substrate patch antennas. The antenna impedance matching network has a standing wave ratio of ≤1.5:1 in the 300MHz-3GHz frequency band. The pulse capture unit can identify discharge pulses with a rise time of ≥2ns, with a time resolution of 100ps.

5. The transformer multi-dimensional state perception and diagnosis system according to claim 1, characterized in that: The oil spectrum sensor uses the internal standard method to correct the gas concentration, and the correction formula is: Among them, P cal is the real-time measurement value of CO2 internal standard, P ref =5.0 PPM As a reference benchmark value, the temperature control fluctuation of the gas separation column is ≤±0.1℃.

6. The transformer multi-dimensional state perception and diagnosis system according to claim 1, characterized in that: The wavelet transform uses the db6 wavelet basis to perform 5-layer decomposition, and the high-frequency coefficient denoising threshold λ is calculated as: Where σ is the estimated value of the noise variance, N is the signal length, and nonlinear compression is applied to the temperature data during normalization. The compression ratio gradually changes from 1:1 to 1:3 with the increase of the temperature rise rate.

7. The transformer multi-dimensional state perception and diagnosis system according to claim 1, characterized in that: The dynamic adjustment algorithm of the weight coefficient of the state assessment model is: w i (t) =0.7w i (t-1) +0.3Δw i where Δw i Calculated based on the coefficient of variation of characteristic parameters in the last 24 hours, when the weight of a dimension exceeds 0.5, an expert system review is triggered.

8. The transformer multi-dimensional state perception and diagnosis system according to claim 1, characterized in that: The convolutional neural network contains 4 convolutional layers with convolution kernel sizes of 5×5, 3×3, 3×3, and 1×1 respectively. The number of channels increases in the order of 32-64-128-256. The number of hidden layer units of the LSTM network is set to 128, and the time step matches 60 sampling periods.

9. The transformer multi-dimensional state perception and diagnosis system according to claim 1, characterized in that: The dynamic alarm threshold update satisfies: Threshold t =μ t -2.5s t +0.21I pd Among them I pd It is the partial discharge activity index. When the threshold is exceeded three times in a row, a multi-level alarm is activated, and the alarm delay time is ≤ 2 seconds.

10. The transformer multi-dimensional state perception and diagnosis system according to claim 1, characterized in that: The three-dimensional thermal map rendering engine supports 0.1°C temperature gradient visualization and discharge hotspot positioning accuracy ≤ 10cm 3 When the acetylene concentration is greater than 5ppm and the vibration energy rises by more than 5dB per hour in the 100-300kHz frequency band, a red warning signal is automatically generated.

Images