Transformer oil acetylene abnormity early warning system and early warning method based on multi-physics field data fusion and computer readable storage medium

Abstract

The invention relates to a transformer oil acetylene abnormity early warning system and early warning method based on multi-physics field data fusion and a computer readable storage medium, and solves the problem that a transformer fault cannot be accurately judged based on a single parameter in the prior art. According to the technical scheme and technical effects, the system integrates gas, thermal field, electric field and magnetic field sensors, collects acetylene concentration, temperature, electric field intensity, magnetic field and other multi-source data in real time, inputs the data into a deep neural network DNN fusion model after wavelet transform denoising, normalization and machine learning preprocessing, extracts features in combination with PCA, analyzes multi-physics field association, and obtains the multi-source data of acetylene concentration, temperature, electric field intensity, magnetic field and the like. Graded early warning is formed based on DNN output and a dynamic threshold value, the accuracy and predictive capacity of fault diagnosis are effectively improved, accurate monitoring and intelligent early warning are achieved, safe operation of the transformer is guaranteed, meanwhile, due to the modular design, the system can adapt to transformers of different models and complex working condition scenes, and high practicability and expansibility are achieved.

Description

Technical Field

[0001] This invention relates to the field of transformer anomaly detection technology, and in particular to a transformer oil acetylene anomaly early warning system, early warning method, and computer-readable storage medium based on multi-physics field data fusion. Background Technology

[0002] In power systems, transformers are core equipment for voltage transformation, power transmission, and distribution, and their operating status directly affects the safety and reliability of the power grid. With the continuous growth of power load and the aging of transformers, they face the challenges of high-load operation and complex working conditions, significantly increasing the risk of internal faults. Transformer oil, as an insulating and cooling medium, decomposes when faults such as overheating or partial discharge occur, producing characteristic gases including acetylene. Acetylene is a marker gas for high-energy arc discharge or high-temperature overheating, and its abnormal concentration is a key indicator for judging the nature and severity of transformer faults.

[0003] Traditional acetylene monitoring methods mainly rely on the detection of a single parameter (such as the concentration of acetylene in oil), which has the following limitations: a single parameter is difficult to fully reflect the complex fault mechanisms inside the transformer, which can easily lead to misjudgment or missed detection; it cannot integrate other physical field parameters (such as temperature, vibration, partial discharge), making it difficult to distinguish fault types (such as overheating or discharge); traditional methods are mostly passive monitoring, making it difficult to predict fault development trends and affecting operation and maintenance decisions; traditional sensors are susceptible to electromagnetic interference or cross-sensitivity, reducing detection reliability.

[0004] Therefore, there is an urgent need for an acetylene anomaly early warning system based on multi-physics data fusion, which can improve the accuracy and predictive ability of fault diagnosis by integrating multi-source data and ensure the safe operation of transformers. Summary of the Invention

[0005] The purpose of this invention is to solve the above-mentioned problems existing in the prior art by providing a transformer oil acetylene anomaly early warning system based on multi-physics field data fusion. Through multi-source data acquisition and fusion analysis, it achieves accurate monitoring and intelligent early warning, effectively improves the accuracy and predictive ability of fault diagnosis, ensures the safe operation of transformers, and effectively reduces unplanned transformer downtime losses. At the same time, its modular design makes it adaptable to different transformer models and complex operating conditions, and has strong practicality and scalability.

[0006] The above-mentioned technical objectives of this invention are mainly achieved through the following technical solutions: The technical solution of the first technical subject of this invention: A transformer oil acetylene anomaly early warning system based on multi-physics data fusion includes: A transformer oil tank, which contains transformer windings and is filled with transformer oil. The multi-physics sensing module includes a thermal field sensor installed inside the transformer tank, an electric field sensor and a magnetic field sensor located near the transformer windings; The oil and gas analysis module includes a degassing device connected to the transformer oil tank, an acetylene gas sensor connected to the degassing device, and an acetylene signal extraction unit connected to the acetylene gas sensor. The data processing module includes a multi-source data acquisition unit connected to the multi-physics sensing module and the acetylene signal extraction unit, a data preprocessing unit connected to the multi-source data acquisition unit, and a data analysis unit connected to the data preprocessing unit. The control module includes an early warning unit connected to the data analysis unit, and a control unit connected to the early warning unit and the transformer winding, respectively.

[0007] In this technical solution, a transformer oil acetylene anomaly early warning system based on multi-physics field data fusion integrates gas, thermal, electric, and magnetic field detection, and collects multi-source data such as acetylene concentration, temperature, electric field strength, and magnetic field data in real time. It employs microelectromechanical systems (MEMS) technology and wireless ad hoc network transmission. After wavelet transform denoising, normalization, and machine learning preprocessing, the data is input into a deep neural network (DNN) fusion model. Principal component analysis (PCA) is used to extract features and analyze multi-physics field correlations. Based on the DNN output and dynamic thresholds, the system provides graded early warnings (mild alert, moderate alarm, severe tripping) and simultaneously transmits the analysis results to the monitoring center. This system significantly improves the accuracy and real-time performance of transformer fault early warning, enhances anti-interference capabilities, supports fault source tracing and trend prediction based on multi-physics field data, and effectively reduces unplanned equipment downtime losses. Furthermore, its modular design allows it to adapt to different transformer models and complex operating conditions, possessing strong practicality and scalability.

[0008] As a further improvement and supplement to the above technical solution, the present invention adopts the following technical measures: Preferably, the thermal field sensor is a distributed fiber optic temperature sensor; The electric field sensor is a photoelectric electric field sensor; The magnetic field sensor is an optical fiber magnetic field sensor; The thermal field sensor, the electric field sensor, and the magnetic field sensor all transmit signals to the multi-source data acquisition unit via optical fiber; The acetylene gas sensor is a photoacoustic spectroscopy acetylene gas sensor with a detection accuracy of ppb level and selective anti-interference capability for hydrogen and methane.

[0009] Preferably, the data analysis unit incorporates a deep neural network model, which includes a feature fusion layer, a fault classification layer, and a trend prediction layer, for realizing the fusion analysis of multi-physics data, fault mode identification, and trend prediction.

[0010] Preferably, the early warning unit is configured to output three levels of early warning signals, including a mild alert, a moderate alarm, and a severe trip command; Based on the aforementioned three-level warning signals, the control unit performs the following actions: A maintenance report is pushed out when there is a minor alert; the transformer load is reduced when there is a moderate alarm; and the power supply is automatically cut off if there is no manual intervention within a preset time when there is a severe trip.

[0011] The technical solution of the second technical subject matter involved in this invention: A method for early warning of acetylene anomalies in transformer oil based on multi-physics data fusion, which is based on the aforementioned early warning system for acetylene anomalies in transformer oil based on multi-physics data fusion, includes the following early warning steps: Multi-physics field data acquisition: Temperature field data, electric field strength data, and magnetic field strength data of the transformer are acquired through the multi-physics field sensing module, and acetylene gas concentration data in the degassed transformer oil is acquired through the oil and gas analysis module. Data preprocessing: The collected multiphysics data are denoised, normalized, and cleaned. Multiphysics data fusion analysis: A deep neural network model is used to fuse and calculate the preprocessed multiphysics data to identify fault characteristics and predict the trend of acetylene concentration changes; Tiered early warning: Based on the results of fusion analysis, early warning signals of corresponding levels are generated; Early warning control operation: Execute corresponding control operations based on the early warning signal.

[0012] Preferably, the data preprocessing steps include: Noise reduction processing: The signal is processed in layers using Discrete Wavelet Transform (DWT). Noise components are removed using wavelet basis functions Daubechies and soft thresholding functions; The denoised signal is reconstructed using inverse wavelet transform IDWT; Normalization process: The minimum-maximum normalization method was used to perform minimum-maximum normalization on temperature data and acetylene concentration data with clear physical boundaries; Z-Score standardization is performed to standardize electric and magnetic field data that exhibit fluctuation characteristics. Cleaning process: The σ standard is used to identify outliers, where σ is the standard deviation; Fill in transiently missing data using linear interpolation or spline interpolation; High-frequency redundant data is processed by downsampling.

[0013] As a preferred method, feature extraction is performed based on normalization processing, and the steps include: Constructing a multidimensional time series matrix: Divide the temperature data, electric field data, and magnetic field data into time windows, and construct the matrix based on the divided data; Principal component analysis dimensionality reduction: Calculate the covariance matrix of the multidimensional time series matrix and extract the eigenvalues ​​corresponding to the principal components; perform eigenvalue decomposition on the covariance matrix; extract the first k principal components, project the original data onto the principal component space, and obtain the dimensionality-reduced eigenvectors, where the selection of k satisfies the cumulative variance contribution rate; Spectral feature extraction: Perform Fast Fourier Transform (FFT) on electric field and magnetic field data to extract characteristic frequencies, peak amplitudes, and total harmonic distortion; extract energy features of specific frequency bands through wavelet packet decomposition.

[0014] Preferably, the deep neural network model includes: Feature fusion layer: Receives the preprocessed temperature feature vector, electric field-magnetic field spectrum feature matrix, and acetylene concentration time-series feature, and calculates the weight coefficients of each feature through a multi-head self-attention mechanism. Among them, the weight coefficient of the temperature feature is positively correlated with the rate of change of acetylene concentration. Fault classification layer: A convolutional neural network (CNN) is used to perform convolution operations on the fused features, and the fault probability distribution is output through the softmax activation function; Trend prediction layer: A time series prediction model is built based on the Long Short-Term Memory (LSTM) network. The input is the fault classification results and acetylene concentration data, and the output is the acetylene concentration prediction curve and confidence interval.

[0015] As a preferred option, tiered early warning systems include: A mild warning will be issued when the acetylene concentration exceeds the normal threshold by 10%-30%. When the acetylene concentration exceeds the normal threshold by 30%-50%, a moderate alarm is issued and operating parameters are adjusted. A severe trip command is issued when the acetylene concentration exceeds 50% of the normal threshold or is predicted to reach a dangerous level.

[0016] The technical solution of the third technical subject matter involved in this invention: A computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the aforementioned method for early warning of transformer oil acetylene anomalies based on multiphysics data fusion.

[0017] The beneficial effects of this invention are as follows: This invention proposes a transformer oil acetylene anomaly early warning system based on multi-physics field data fusion. By fusing multi-physics field data (acetylene concentration, temperature, electric field, and magnetic field), it overcomes the limitations of single-parameter monitoring, comprehensively reflects the complex fault mechanisms inside the transformer, significantly improves the accuracy of fault diagnosis, and reduces misjudgments and omissions. Through early fault detection, precise early warning, and coordinated control, it reduces the economic losses caused by unplanned transformer outages and ensures the safety and reliability of power system operation. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structure of the transformer oil acetylene anomaly early warning system based on multi-physics field data fusion, which is involved in this invention.

[0019] In the diagram: 1. Transformer oil tank; 2. Transformer winding; 3. Thermal field sensor; 4. Electric field sensor; 5. Magnetic field sensor; 6. Degassing device; 7. Acetylene gas sensor; 8. Acetylene signal extraction unit; 9. Multi-source data acquisition unit; 10. Data preprocessing unit; 11. Data analysis unit; 12. Early warning unit; 13. Control unit. Detailed Implementation

[0020] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0021] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] Example 1: The technical solution of the first technical subject matter involved in this invention: like Figure 1 As shown, a transformer oil acetylene anomaly early warning system based on multi-physics data fusion includes: Transformer oil tank 1, which contains transformer windings 2 and is filled with transformer oil; The multi-physics sensing module includes a thermal field sensor 3 installed inside the transformer tank 1, an electric field sensor 4 and a magnetic field sensor 5 located near the transformer winding 2; The oil and gas analysis module includes a degassing device 6 connected to the transformer oil tank 1, an acetylene gas sensor 7 connected to the degassing device 6, and an acetylene signal extraction unit 8 connected to the acetylene gas sensor 7. The data processing module includes a multi-source data acquisition unit 9 connected to the multi-physics sensing module and the acetylene signal extraction unit 8, a data preprocessing unit 10 connected to the multi-source data acquisition unit 9, and a data analysis unit 11 connected to the data preprocessing unit 10. The control module includes an early warning unit 12 connected to the data analysis unit 11, and a control unit 13 connected to the early warning unit 12 and the transformer winding 2 respectively.

[0023] The main difference between this technical solution and existing technologies lies in addressing the shortcomings of relying solely on acetylene concentration to determine transformer operating conditions. Instead, it utilizes multi-source data from multiple physical fields to assess transformer operation, thereby improving the accuracy and predictive ability of transformer fault diagnosis and ensuring safe transformer operation. Specifically: In this technical solution, the transformer oil acetylene anomaly early warning system based on multi-physics field data fusion integrates gas, thermal, electric, and magnetic field detection, and collects multi-source data such as acetylene concentration, temperature, electric field strength, and magnetic field data in real time. It employs microelectromechanical systems (MEMS) technology and wireless ad hoc network transmission. After wavelet transform denoising, normalization, and machine learning preprocessing, the data is input into a deep neural network (DNN) fusion model. Principal component analysis (PCA) is used to extract features and analyze the multi-physics field correlation. Based on the DNN output and dynamic thresholds, the system provides graded early warnings (mild alert, moderate alarm, severe tripping) and simultaneously transmits the analysis results to the monitoring center.

[0024] Therefore, the transformer oil acetylene anomaly early warning system based on multi-physics data fusion can significantly improve the accuracy and real-time performance of transformer fault early warning, enhance anti-interference capabilities, support fault source tracing and trend prediction based on multi-physics data, and effectively reduce unplanned equipment downtime losses. Furthermore, its modular design allows it to adapt to different transformer models and complex operating conditions, possessing strong practicality and scalability.

[0025] In practical applications, the thermal field sensor 3 is a distributed fiber optic temperature sensor.

[0026] In practical applications, the electric field sensor 4 is a photoelectric electric field sensor.

[0027] In practical applications, the magnetic field sensor 5 is an optical fiber magnetic field sensor.

[0028] In practical applications, the thermal field sensor 3, the electric field sensor 4, and the magnetic field sensor 5 all transmit signals to the multi-source data acquisition unit 9 via optical fiber.

[0029] In practical applications, the acetylene gas sensor 7 is a photoacoustic spectroscopy acetylene gas sensor with a detection accuracy of ppb level and selective anti-interference capability for hydrogen and methane.

[0030] In this technical solution, the connection sequence of each component, device, or unit in the transformer oil acetylene anomaly early warning system based on multi-physics field data fusion is as follows: the transformer oil tank 1 is filled with transformer oil and the transformer winding 2 is installed; the thermal field sensor 3 is installed inside the transformer oil tank 1; the electric field sensor 4 and the magnetic field sensor 5 are installed near the transformer winding 2; the degassing device 6 is installed outside the transformer oil tank 1 and connected to the acetylene gas sensor 7; the acetylene gas sensor 7 is connected to the acetylene signal extraction unit 8; the acetylene signal extraction unit 8 is connected to the multi-source data acquisition unit 9; the multi-source data acquisition unit 9 is connected to the thermal field sensor 3, the electric field sensor 4, the magnetic field sensor 5, and the data preprocessing unit 10; the data preprocessing unit 10 is connected to the data analysis unit 11; the data analysis unit 11 is connected to the early warning unit 12; the early warning unit 12 is connected to the control unit 13; and the control unit 13 is connected to the transformer winding 2.

[0031] Multi-physical field data is collected by thermal field sensor 3, electric field sensor 4, magnetic field sensor 5 and acetylene gas sensor 7. The multi-physical field data is analyzed and processed by data preprocessing unit 10 and data analysis unit 11 to form corresponding early warning signals (for mild prompts, moderate alarms and severe trips).

[0032] Under fault conditions such as overheating or high-energy discharge (e.g., electric arc), transformer winding 2 can cause transformer oil decomposition, generating characteristic gases such as acetylene. A transformer oil acetylene anomaly early warning system based on multi-physics data fusion provides accurate and reliable acetylene anomaly early warning by precisely monitoring the transformer's multi-physics data. Specifically: In practical applications, the transformer winding 2 adopts a layered or partitioned design to reduce the local electric field strength and the probability of electric arcing. This, combined with electric field / magnetic field sensors, allows for more accurate detection of field disturbances caused by faults.

[0033] In practical applications, the thermal field sensor 3 collects temperature data around the transformer oil and transformer winding 2 in real time, quickly responds to minute temperature changes in the oil tank, captures thermal anomalies caused by early faults, and provides timely and reliable data support for acetylene anomaly early warning.

[0034] In practical applications, the thermal field sensor 3 is preferably an optical fiber temperature sensor, whose distributed optical fiber temperature measurement (DTS) can cover the entire oil tank, thereby providing comprehensive temperature field data, which is highly compatible with the fusion analysis requirements of the DNN model.

[0035] In practical applications, the electric field sensor 4 collects electric field strength data inside the transformer tank 1 and around the transformer winding 2 in real time, quickly captures electric field disturbances caused by partial discharge or electric arc, and promptly reflects early fault signals inside the transformer, providing key data support for acetylene anomaly early warning.

[0036] In practical applications, the electric field sensor 4 is preferably a photoelectric electric field sensor that uses optical fiber to transmit signals. It is immune to strong electromagnetic field interference inside the transformer and can detect electric field changes at the ppb level, making it suitable for early partial discharge detection.

[0037] In practical applications, the magnetic field sensor 5 collects magnetic field data inside the transformer tank 1 and around the transformer winding 2 in real time, and detects magnetic field disturbances caused by faults such as abnormal current, partial discharge or electric arc.

[0038] In practical applications, the magnetic field sensor 5 is preferably an optical fiber magnetic field sensor. It uses optical fiber to transmit signals and is immune to strong electric field interference inside the transformer. It can detect magnetic field changes at the µT level, making it suitable for early fault signal capture. The optical fiber material can be encapsulated as oil-resistant and high-temperature resistant, making it suitable for long-term immersion in transformer oil. It can achieve multi-point magnetic field monitoring along the length of the optical fiber, covering the entire oil tank or winding.

[0039] In practical applications, the degassing device 6 is installed outside the transformer oil tank 1 and connected to the tank. It separates dissolved gases (such as acetylene, hydrogen, methane, etc.) from the transformer oil to provide a gas sample to be detected for the acetylene gas sensor.

[0040] In practical applications, the acetylene gas sensor 7 utilizes photoacoustic spectroscopy to detect minute concentration changes (down to the ppb level) and exhibits good selectivity for other gases (such as hydrogen and methane), reducing cross-sensitivity interference.

[0041] In practical applications, the acetylene signal extraction unit 8 converts the raw acetylene concentration signal detected by the acetylene gas sensor 7 into a processable standardized signal and amplifies it to improve the signal-to-noise ratio, ensuring the accuracy and reliability of the signal.

[0042] In practical applications, the multi-source data acquisition unit 9 collects data from different sensors in real time, including acetylene gas sensor 7 (acetylene gas data determined by the acetylene signal extraction unit), thermal field sensor 3 (temperature data), electric field sensor 4 (electric field strength data), and magnetic field sensor 5 (magnetic field strength data), realizing the unified collection of multi-physical field data.

[0043] In practical applications, the data preprocessing unit 10 transforms the raw data provided by the multi-source data acquisition unit into high-quality, structured input through denoising, normalization, feature extraction, and cleaning, providing a reliable foundation for the data analysis unit 11, wherein the data analysis unit 11 is preferably a DNN model.

[0044] In practical applications, the data analysis unit incorporates a deep neural network model, which includes a feature fusion layer, a fault classification layer, and a trend prediction layer, to achieve fusion analysis of multi-physics data, fault mode identification, and trend prediction.

[0045] In practical applications, the early warning unit is configured to output three levels of early warning signals, including mild prompts, moderate alarms, and severe tripping commands; Based on the aforementioned three-level warning signals, the control unit performs the following actions: A maintenance report is pushed out when there is a minor alert; the transformer load is reduced when there is a moderate alarm; and the power supply is automatically cut off if there is no manual intervention within a preset time when there is a severe trip.

[0046] In practical applications, when the transformer oil acetylene anomaly early warning system based on multi-physics field data fusion is activated, all sensors and devices operate synchronously. If the transformer winding 2 in the transformer oil tank 1 experiences overheating or high-energy discharge during operation, it will cause the transformer oil in the tank to decompose, producing characteristic gases including acetylene. The thermal field sensor 3 collects temperature data of the transformer oil and the area around the winding in real time, while the electric field sensor 4 and magnetic field sensor 5 capture the electric field strength and magnetic field data inside the tank and around the winding, respectively. These data directly reflect whether there are signs of thermal anomalies, partial discharges, or arcing in the equipment. Simultaneously, the degassing device 6 separates dissolved gases (including acetylene) from the transformer oil and delivers them to the acetylene gas sensor 7. The acetylene gas sensor 7 uses photoacoustic spectroscopy to detect the acetylene concentration, achieving a detection accuracy of ppb and effectively reducing cross-interference from other gases. Subsequently, the acetylene signal extraction unit 8 converts the raw acetylene concentration signal detected by the acetylene gas sensor 7 into a standardized signal and amplifies it, improving the signal-to-noise ratio. The multi-source data acquisition unit 9 collects data from the thermal field sensor 3, electric field sensor 4, magnetic field sensor 5, and acetylene concentration data processed by the acetylene signal extraction unit 8, achieving the convergence of multi-physics field data. The data preprocessing unit 10 processes the collected raw data: it removes noise such as electromagnetic interference through wavelet transform; it eliminates dimensional differences between different physical field data using minimum-maximum normalization and Z-score standardization; it extracts key features using methods such as principal component analysis (PCA) to reduce data dimensionality; and it identifies and processes outliers, missing values, and redundant data to ensure high-quality and structured data input to the data analysis unit 11. The preprocessed multi-physics field data is input to the data analysis unit 11, which performs fusion analysis using a deep neural network (DNN) model. The DNN model combines temperature, electric field, magnetic field, and acetylene concentration data to identify fault modes, determine fault types (such as overheating or discharge), predict fault development trends, and generate dynamic thresholds. Based on the output of the data analysis unit 11, the early warning unit 12 implements tiered early warnings using dynamic thresholds: when the acetylene concentration is slightly high or a single physical field exhibits a minor anomaly, a mild alert is issued to notify maintenance personnel; if the acetylene concentration continues to rise or multiple physical fields exhibit coordinated anomalies, a moderate alarm is triggered, suggesting adjustments to operating parameters or a shutdown for inspection; if the acetylene concentration is severely exceeded and multiple physical fields display an emergency fault, a severe trip warning is initiated. The control unit 13 responds to the tiered early warning signals from the early warning unit 12, performing corresponding operations for different warning levels: for a mild alert, only the anomaly is recorded; for a moderate alarm, transformer operating parameters are adjusted (e.g., load reduction) to mitigate the fault; for a severe trip, the transformer power supply is automatically disconnected, triggering a protection mechanism to prevent equipment damage or power grid accidents.The entire process forms a complete closed loop from data acquisition (transformer tank 1, transformer winding 2, thermal field sensor 3, electric field sensor 4, magnetic field sensor 5, degassing device 6, acetylene gas sensor 7, acetylene signal extraction unit 8, multi-source data acquisition unit 9), data processing (data preprocessing unit 10), data analysis (data analysis unit 11) to early warning control (early warning unit 12, control unit 13), ensuring the safe and reliable operation of the transformer.

[0047] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0048] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and system can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0049] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0050] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0051] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0052] Example 2: The technical solution of the second technical subject of this invention: a method for early warning of acetylene anomalies in transformer oil based on multi-physics data fusion, which is based on the early warning system for early warning of acetylene anomalies in transformer oil based on multi-physics data fusion described in Example 1, and its early warning steps include: Multi-physics field data acquisition: Temperature field data, electric field strength data, and magnetic field strength data of the transformer are acquired through the multi-physics field sensing module, and acetylene gas concentration data in the degassed transformer oil is acquired through the oil and gas analysis module. Data preprocessing: The collected multiphysics data are denoised, normalized, and cleaned. Multiphysics data fusion analysis: A deep neural network model is used to fuse and calculate the preprocessed multiphysics data to identify fault characteristics and predict the trend of acetylene concentration changes; Tiered early warning: Based on the results of fusion analysis, early warning signals of corresponding levels are generated; Early warning control operation: Execute corresponding control operations based on the early warning signal.

[0053] In practical applications, data preprocessing steps include: Noise reduction processing: The signal is processed in layers using Discrete Wavelet Transform (DWT). Noise components are removed using wavelet basis functions Daubechies and soft thresholding functions; The denoised signal is reconstructed using inverse wavelet transform IDWT; Normalization process: The minimum-maximum normalization method was used to perform minimum-maximum normalization on temperature data and acetylene concentration data with clear physical boundaries; Z-Score standardization is performed to standardize electric and magnetic field data that exhibit fluctuation characteristics. Cleaning process: The σ standard is used to identify outliers, where σ is the standard deviation; Fill in transiently missing data using linear interpolation or spline interpolation; High-frequency redundant data is processed by downsampling.

[0054] In this technical solution, noise reduction processing is used to remove noise from sensor data, improve the signal-to-noise ratio, and ensure that the data reflects the true changes in the physical field.

[0055] Next, the steps of the noise reduction process will be explained in more detail: S11 performs wavelet decomposition on the original signal (such as acetylene concentration, temperature, electric field, and magnetic field data), using Discrete Wavelet Transform (DWT) to decompose the signal into components of different frequencies.

[0056] S12 employs wavelet basis functions Daubechies to improve denoising performance and ensure data quality and reliability.

[0057] S13. Apply a soft thresholding method to the decomposed wavelet coefficients to remove high-frequency noise components while retaining low-frequency signals that represent real physical changes.

[0058] S14 reconstructs the denoised signal using inverse wavelet transform IDWT.

[0059] In this technical solution, the steps for normalization processing are further improved: S21, Calculate the historical minimum and maximum values ​​of each sensor's data, perform minimum-maximum normalization on the data, and map the original data X (i.e., the denoised data of each sensor) to the [0,1] interval: , In the formula: This represents the normalized data value, which is usually in the range [0,1], and represents the position of the original data X after normalization; This represents the maximum value in the dataset, used to determine the upper bound of the normalization range; This represents the minimum value in the dataset, used to determine the lower bound of the normalization range; X represents the original data value, i.e. the data point that needs to be normalized; S22, calculate the mean and standard deviation of the time window for each sensor data, and convert the data into standardized values ​​with a mean of 0 and a standard deviation of 1; , In the formula: Represents the Z-Score standardized value; This represents the data mean; It represents the standard deviation.

[0060] In practical applications, feature extraction based on normalization processing includes the following steps: Constructing a multidimensional time series matrix: Divide the temperature data, electric field data, and magnetic field data into time windows, and construct the matrix based on the divided data; Principal component analysis dimensionality reduction: Calculate the covariance matrix of the multidimensional time series matrix and extract the eigenvalues ​​corresponding to the principal components; perform eigenvalue decomposition on the covariance matrix; extract the first k principal components, project the original data onto the principal component space, and obtain the dimensionality-reduced eigenvectors, where the selection of k satisfies the cumulative variance contribution rate; Spectral feature extraction: Perform Fast Fourier Transform (FFT) on electric field and magnetic field data to extract characteristic frequencies, peak amplitudes, and total harmonic distortion; extract energy features of specific frequency bands through wavelet packet decomposition.

[0061] Next, the steps for feature extraction based on normalization processing will be explained in further detail: S31, Constructing a multidimensional time series matrix based on multiphysics field data: X=[X C2H2 ,X Temp ,X E-field ,X B-field ] , X C2H2 This represents acetylene concentration data; X Temp Represents temperature data; X E-field Represents electric field strength data; X B-field This represents magnetic field strength data. S32, Principal Component Analysis Dimensionality Reduction: Calculate the covariance matrix of the data matrix, extract the eigenvalues ​​corresponding to the principal components (eigenvectors), and perform eigenvalue decomposition on the covariance matrix; Wherein, the covariance matrix C is: , In the formula: X represents the centered data matrix (mean is 0); This represents the transpose of the centered data matrix.

[0062] Perform eigenvalue decomposition on the covariance matrix C: , In the formula: Represents a diagonal matrix containing eigenvalues. This indicates the importance of each principal component; V represents an orthogonal matrix, with columns representing eigenvectors v1, v2, ..., v m , indicating the direction of the principal component.

[0063] S33, select the first k principal components (usually retaining 90-95% of the variance), project the original data onto the principal component space, and obtain the dimensionality-reduced feature vectors.

[0064] Calculate the cumulative contribution rate: .

[0065] Using the selected feature vector V k (First k columns) Projecting the original data onto the principal component space yields the dimensionality-reduced feature matrix Z: Z=XV k .

[0066] S34 calculates the mean, standard deviation, peak value, slope, skewness, kurtosis, etc. of each sensor data to reflect the trend and anomalies of the signal.

[0067] S35 performs Fast Fourier Transform (FFT) analysis on electric and magnetic field data to extract high-frequency pulses of partial discharge.

[0068] S36 performs wavelet packet decomposition on the non-stationary signal of magnetic field disturbance caused by electric arc to extract energy features of a specific frequency band.

[0069] In this technical solution, the steps for identifying and processing outliers, missing values, or redundant data to ensure high-quality, structured data input to the DNN model are as follows: S41. Use the 3σ criterion to identify outliers. If the acetylene concentration changes by more than three times the standard deviation of historical data, it is marked as an anomaly. Set a threshold based on the transformer operating parameters. If the acetylene concentration is >1000ppb, it is considered an anomaly and the outlier is removed or replaced.

[0070] S42, use linear interpolation or spline interpolation to fill in the data that is temporarily missing (such as when a sensor goes offline); for short-term missing data, use the data from the previous or next moment to fill in the data; if a sensor has no data for a long time, mark it as a fault, notify the early warning unit and suspend the relevant data input.

[0071] S43 downsamples high-frequency sampling data (such as electric field pulses) to retain data at key time points and reduce computational burden; it uses a sliding window to detect duplicate data (such as identical values ​​caused by sensor lag) and removes redundant records.

[0072] In practical applications, the deep neural network model includes: Feature fusion layer: Receives the preprocessed temperature feature vector, electric field-magnetic field spectrum feature matrix, and acetylene concentration time-series feature, and calculates the weight coefficients of each feature through a multi-head self-attention mechanism. Among them, the weight coefficient of the temperature feature is positively correlated with the rate of change of acetylene concentration. Fault classification layer: A convolutional neural network (CNN) is used to perform convolution operations on the fused features, and the fault probability distribution is output through the softmax activation function; Trend prediction layer: A time series prediction model is built based on the Long Short-Term Memory (LSTM) network. The input is the fault classification results and acetylene concentration data, and the output is the acetylene concentration prediction curve and confidence interval.

[0073] In practical applications, tiered early warning systems include: When the acetylene concentration exceeds the normal threshold by 10%-30%, a mild alert will be issued. The mild alert mainly includes: recording the anomaly and notifying the operation and maintenance personnel, without the need for immediate intervention. When the acetylene concentration exceeds the normal threshold by 30%-50%, a moderate alarm is issued and operating parameters are adjusted. The moderate alarm mainly includes adjusting transformer operating parameters (such as reducing the load) to slow down the development of the fault. When the acetylene concentration exceeds 50% of the normal threshold or is predicted to reach a dangerous level, a severe trip command is issued. Severe tripping mainly includes: automatically disconnecting the transformer power supply and triggering protection mechanisms to avoid equipment damage or power grid accidents caused by serious faults (such as arc discharge or overheating).

[0074] Example 3: The technical solution of the third technical subject of the present invention: a computer-readable storage medium.

[0075] The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the transformer oil acetylene anomaly early warning method based on multi-physics data fusion as described in Example 2.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations can be made to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A transformer oil acetylene anomaly early warning system based on multi-physics data fusion, characterized in that, include: The transformer oil tank (1) contains transformer windings (2) and is filled with transformer oil. The multi-physics sensing module includes a thermal field sensor (3) installed inside the transformer tank (1), an electric field sensor (4) and a magnetic field sensor (5) located near the transformer winding (2). The oil and gas analysis module includes a degassing device (6) connected to the transformer tank (1), an acetylene gas sensor (7) connected to the degassing device (6), and an acetylene signal extraction unit (8) connected to the acetylene gas sensor (7). The data processing module includes a multi-source data acquisition unit (9) connected to the multi-physics sensing module and the acetylene signal extraction unit (8), a data preprocessing unit (10) connected to the multi-source data acquisition unit (9), and a data analysis unit (11) connected to the data preprocessing unit (10). The control module includes an early warning unit (12) connected to the data analysis unit (11), and a control unit (13) connected to the early warning unit (12) and the transformer winding (2) respectively.

2. The transformer oil acetylene anomaly early warning system based on multi-physics data fusion according to claim 1, characterized in that, The thermal field sensor (3) is a distributed fiber optic temperature sensor; The electric field sensor (4) is a photoelectric electric field sensor; The magnetic field sensor (5) is an optical fiber magnetic field sensor; The thermal field sensor (3), the electric field sensor (4), and the magnetic field sensor (5) all transmit signals to the multi-source data acquisition unit (9) via optical fiber. The acetylene gas sensor (7) is a photoacoustic spectroscopy acetylene gas sensor with a detection accuracy of ppb level and selective anti-interference capability for hydrogen and methane.

3. The transformer oil acetylene anomaly early warning system based on multi-physics data fusion according to claim 1, characterized in that, The data analysis unit (11) has a built-in deep neural network model, which includes a feature fusion layer, a fault classification layer and a trend prediction layer, and is used to realize the fusion analysis of multi-physics field data, fault mode recognition and trend prediction.

4. The transformer oil acetylene anomaly early warning system based on multi-physics data fusion according to claim 1, characterized in that, The early warning unit (12) is configured to output three levels of early warning signals, including mild prompt, moderate alarm and severe trip command; Based on the aforementioned three-level warning signals, the control unit (13) performs the following actions: A maintenance report is pushed out when there is a minor alert; the transformer load is reduced when there is a moderate alarm; and the power supply is automatically cut off if there is no manual intervention within a preset time when there is a severe trip.

5. A method for early warning of acetylene anomalies in transformer oil based on multi-physics data fusion, characterized in that, The transformer oil acetylene anomaly early warning system based on multi-physics data fusion, as described in any one of claims 1-4, includes the following early warning steps: Multi-physics field data acquisition: Temperature field data, electric field strength data, and magnetic field strength data of the transformer are acquired through the multi-physics field sensing module, and acetylene gas concentration data in the degassed transformer oil is acquired through the oil and gas analysis module. Data preprocessing: The collected multiphysics data are denoised, normalized, and cleaned. Multiphysics data fusion analysis: A deep neural network model is used to fuse and calculate the preprocessed multiphysics data to identify fault characteristics and predict the trend of acetylene concentration changes; Tiered early warning: Based on the results of fusion analysis, early warning signals of corresponding levels are generated; Early warning control operation: Execute corresponding control operations based on the early warning signal.

6. The method for early warning of acetylene anomalies in transformer oil based on multi-physics data fusion according to claim 5, characterized in that, Data preprocessing steps include: Noise reduction processing: The signal is processed in layers using Discrete Wavelet Transform (DWT). Noise components are removed using wavelet basis functions Daubechies and soft thresholding functions; The denoised signal is reconstructed using inverse wavelet transform IDWT; Normalization process: The minimum-maximum normalization method was used to perform minimum-maximum normalization on temperature data and acetylene concentration data with clear physical boundaries; Z-Score standardization is performed to standardize electric and magnetic field data that exhibit fluctuation characteristics. Cleaning process: The 3σ standard is used to identify outliers, where σ is the standard deviation; Fill in transiently missing data using linear interpolation or spline interpolation; High-frequency redundant data is processed by downsampling.

7. The method for early warning of acetylene anomalies in transformer oil based on multi-physics data fusion according to claim 6, characterized in that, Feature extraction based on normalization processing includes the following steps: Constructing a multidimensional time series matrix: Divide the temperature data, electric field data, and magnetic field data into time windows, and construct the matrix based on the divided data; Principal component analysis dimensionality reduction: Calculate the covariance matrix of the multidimensional time series matrix and extract the eigenvalues ​​corresponding to the principal components; perform eigenvalue decomposition on the covariance matrix; extract the first k principal components, project the original data onto the principal component space, and obtain the dimensionality-reduced eigenvectors, where the selection of k satisfies the cumulative variance contribution rate; Spectral feature extraction: Perform Fast Fourier Transform (FFT) on electric field and magnetic field data to extract characteristic frequencies, peak amplitudes, and total harmonic distortion; extract energy features of specific frequency bands through wavelet packet decomposition.

8. The method for early warning of acetylene anomalies in transformer oil based on multi-physics data fusion according to claim 5, characterized in that, The deep neural network model includes: Feature fusion layer: Receives the preprocessed temperature feature vector, electric field-magnetic field spectrum feature matrix, and acetylene concentration time-series feature, and calculates the weight coefficients of each feature through a multi-head self-attention mechanism. Among them, the weight coefficient of the temperature feature is positively correlated with the rate of change of acetylene concentration. Fault classification layer: A convolutional neural network (CNN) is used to perform convolution operations on the fused features, and the fault probability distribution is output through the softmax activation function; Trend prediction layer: A time series prediction model is built based on the Long Short-Term Memory (LSTM) network. The input is the fault classification results and acetylene concentration data, and the output is the acetylene concentration prediction curve and confidence interval.

9. The method for early warning of acetylene anomalies in transformer oil based on multi-physics data fusion according to claim 5, characterized in that, Tiered early warning includes: A mild warning will be issued when the acetylene concentration exceeds the normal threshold by 10%-30%. When the acetylene concentration exceeds the normal threshold by 30%-50%, a moderate alarm is issued and operating parameters are adjusted. A severe trip command is issued when the acetylene concentration exceeds 50% of the normal threshold or is predicted to reach a dangerous level.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the transformer oil acetylene anomaly early warning method based on multi-physics data fusion as described in any one of claims 5 to 9.

Images