A transformer diagnosis analysis and early warning method based on a multi-source heterogeneous data model

Abstract

The application discloses the technical field of transformer diagnosis and analysis and early warning, and particularly relates to a transformer diagnosis, analysis and early warning method, which comprises the following specific steps: detecting the partial discharge of a transformer by using a UHF partial discharge detection method, monitoring and analyzing the oil chromatogram by using a transformer chromatogram online monitoring system, installing a high-precision zero-flux current sensor in series on an iron core / filing grounding point to detect the current, then amplifying and transmitting the weak current, and finally performing calculation and analysis by using a digital signal method. The application is beneficial to quantitatively evaluating and dynamically sensing the real-time operation health state of the main transformer equipment, that is, timely capturing the clues of the initial information of the hidden danger, correctly diagnosing and timely warning the "disease" mechanism and "lesion" position of the potential fault, and accurately evaluating the possibility of the development of the hidden danger into a dominant fault.

Description

Technical Field

[0001] This invention relates to the field of transformer diagnosis, analysis and early warning technology, specifically a transformer diagnosis, analysis and early warning method based on a multi-source heterogeneous data model. Background Technology

[0002] In power system operation, the stability, safety, and reliability of electrical equipment directly affect the system's operation. Electrical equipment maintenance and testing serve as the basis for ensuring the normal operation of electrical equipment. They not only promptly identify equipment problems but also effectively address related issues, thereby minimizing equipment failure rates and reducing power system losses. Therefore, in practical work, it is essential to correctly understand online monitoring and condition-based maintenance of electrical equipment to ensure the normal operation of the power system.

[0003] However, current power plant transformer status monitoring methods are relatively simple and cannot effectively assess the actual operating status of transformer equipment. Therefore, a transformer diagnostic analysis and early warning method based on a multi-source heterogeneous data model is invented. Summary of the Invention

[0004] In view of the problems existing in the above and / or the existing transformer diagnosis analysis and early warning method based on multi-source heterogeneous data model, the present invention is proposed.

[0005] Therefore, the purpose of this invention is to provide a transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model, which can solve the aforementioned existing problems.

[0006] To address the aforementioned technical problems, according to one aspect of the present invention, the present invention provides the following technical solution: A transformer diagnostic analysis and early warning method based on a multi-source heterogeneous data model includes the following specific steps: Step 1, Partial Discharge: The partial discharge of the transformer is detected using the ultra-high frequency partial discharge detection method; Step 2, Oil Chromatography Monitoring and Analysis: The oil chromatogram is monitored and analyzed using a transformer online chromatographic monitoring system; Step 3, core grounding monitoring: A high-precision zero-flux current sensor is connected in series at the core grounding point to detect the current. Then, the weak current is amplified and transmitted, and finally, digital signal methods are used for calculation and analysis. Step 4, Vibration Monitoring and Analysis: Monitor transformer vibration using an online electromagnetic oscillation monitoring device; Step 5, Transformer temperature monitoring: Two sets of temperature sensors are configured on each transformer, with two temperature sensors in each set. They are arranged at the upper and lower oil inlets of the transformer radiator to monitor the heat dissipation effect of the transformer. The two sets of temperature sensors are arranged diagonally. Step six involves combining the linear correlation index and the two nonlinear indices to establish a comprehensive correlation index for analyzing the correlation between transformer state parameters.

[0007] As a preferred embodiment of the transformer diagnosis, analysis and early warning method based on a multi-source heterogeneous data model described in this invention, the process of the ultra-high frequency partial discharge detection method is as follows: the UHF electromagnetic waves radiated by the partial discharge are received by a UHF sensor to realize the detection of partial discharge.

[0008] As a preferred embodiment of the transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model described in this invention, the transformer chromatographic online monitoring system consists of a chromatographic analysis system, a circuit control data storage and processing system, a communication module, a carrier gas, an oil sampling valve, and an ambient temperature regulator.

[0009] As a preferred embodiment of the transformer diagnosis, analysis and early warning method based on a multi-source heterogeneous data model described in this invention, the chromatographic analysis system includes an oil sample acquisition module, an oil-gas separation module, an automatic sample injection module, a sample component separation module and a component detection module.

[0010] As a preferred embodiment of the transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model described in this invention, the transformer electromagnetic oscillation online monitoring device consists of an acquisition module and an analysis and diagnosis module. The acquisition module includes 16 electromagnetic oscillation sensors respectively set around the transformer to realize synchronous data acquisition of transformer vibration signals. The analysis and diagnosis module realizes the diagnosis and early warning of latent faults in transformer operation by monitoring and predicting the data of the transformer core and windings and reasonably modeling the logic devices. The data types include frequency complexity, vibration stability, energy similarity, and vibration correlation.

[0011] As a preferred embodiment of the transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model described in this invention, wherein: the frequency complexity The operation process is as follows: Step 1, Data Selection: The complexity of the frequency components in the reaction signal is considered. Lower complexity means the tank wall vibration energy is concentrated in a few frequency components; conversely, higher complexity means the energy is more dispersed. During normal transformer operation, the frequency components are typically concentrated in only a few finite frequencies, meaning the frequency complexity is relatively low. The transformer vibration components are mainly concentrated within 2000Hz, and all harmonic frequencies are integer multiples of 50Hz. Therefore, only harmonic components within the range of 50-2000Hz that are integer multiples of 50Hz are analyzed. The formula for calculating the harmonic proportion Pf at frequency f is defined as follows: , ; : The magnitude of the vibration harmonic amplitude at frequency f; The weighting coefficient with frequency f; Step 2, Frequency Complexity Calculation: Complexity of the Vibration Frequency Components of the Transformer Tank Wall : .

[0012] As a preferred embodiment of the transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model described in this invention, wherein: the vibration stability... The operation process is as follows: Step 1, Data Selection: Reflects the mechanical stability of each vibration monitoring point of the transformer. When structural abnormalities occur, the system becomes uncertain and random vibrations will occur. A transformer under normal operation is a deterministic system, and will produce the same vibrations for the same input parameters and initial conditions. Step 2, Calculation: A recursion graph is represented by the following formula: , For a time series Uk (k=1,2,...n), with a sampling time interval of Δt, the appropriate embedding dimension m and delay time t are determined through relevant theoretical calculations, and then the time series is reconstructed. The reconstructed dynamical system is as follows: , In the formula, ; Calculate the distance between point xi (i) and point xj (j) in the reconstructed phase space: , In the formula, Represents the norm; Calculate the recursive value: .

[0013] As a preferred embodiment of the transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model described in this invention, wherein: the energy similarity The operation process is as follows: Step 1, Data Selection: The acquired vibration time-domain signal is subjected to Fourier transform, and the resulting frequency-domain vibration is divided into different energy bands. The transformer vibration components are mainly concentrated within 2000Hz, and all harmonic frequencies are integer multiples of 50Hz. According to the vibration characteristics, only the vibration components within 400-2000Hz are analyzed. These energies are divided into 4 groups, each with a bandwidth of 400Hz. The energy of the nth energy band is defined as follows: , Where Af is the vibration amplitude corresponding to frequency f, f is an integer, = (400, 450, ..., 2000); Step 2, Data Normalization: Convert the energy x(n) of the nth energy band into a relative percentage value: , Where u(1) is the relative percentage of energy converted from the first energy band (400-800Hz), and so on; Step 3, Average Energy Calculation: The target value is extracted from a large amount of historical data. Assuming there are N feature vector samples ui (i=1, 2, ..., N) (e.g., taking data from the past week, 7*24*6=1008 times), the average energy can be calculated as follows: ; Step 4, Energy Similarity Calculation: Energy similarity analysis determines whether transformer vibration is abnormal by comparing the energy analysis of the measured signal with the target energy distribution. When the energy similarity value of a certain measuring point suddenly increases, it means that the mechanical structure of the transformer near that measuring point may be abnormal. Energy similarity is expressed by energy difference rate, which is the average distance between multiple measured values ​​and the target value.

[0014] As a preferred embodiment of the transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model described in this invention, wherein: the vibration correlation... The operation process is as follows: Step 1, Data Selection: Based on the transformer vibration principle, the vibration of both the winding and the core is based on 100Hz as the fundamental frequency. The diagnostic model for correlation analysis only needs to consider the 100Hz component of the vibration. α1 and α2 are the vibration values ​​under different loads, respectively, and β is the phase difference between the two vibrations (i.e., the power factor angle). The COSβ of the high-voltage side vibration sensor is taken from the COSβ of the main transformer's high-voltage side, and the COSβ of the low-voltage side vibration sensor is taken from the COSβ of the main transformer's low-voltage side. , Its Δωij is the vibration change obtained from the j-th vibration sensor and from the i-th and i+1-th measurements, and the characteristic matrix Xorigin is composed of Δωij, note that n>m; ; Step 2, Data Dimensionality Reduction: Principal component analysis is used to reduce the dimensionality of the data. First, the above feature matrices are normalized: , Next, the corresponding covariance matrix is ​​calculated for the above data, and then the matrix is ​​decomposed by eigenvalue decomposition or singular value decomposition, where each column in matrix U represents an eigenvector. ,

[0015] in: ; Step 3, Calculate vibration characteristic parameters: Vibration characteristic parameters: , which represents the proportion of the principal component's energy to the total energy;

[0016] As a preferred embodiment of the transformer diagnosis analysis and early warning method based on a multi-source heterogeneous data model described in this invention, the linear correlation index is the Pearson correlation coefficient, and the nonlinear index is the Spearman correlation coefficient and the Kendall correlation coefficient. A comprehensive correlation index is established by taking the average of the three. The Pearson correlation coefficient, also known as the Pearson correlation coefficient, is often used to describe the relationship between two continuous variables, reflecting the degree of linear correlation between the two variables. The Pearson correlation coefficients between the k-th state and the l-th state parameter of the data sample are shown below; , in, , The first k The state parameter and the first l The average value of each state parameter data; Let be the Pearson correlation coefficient between the k-th state parameter and the l-th state parameter; The Pearson correlation coefficient ranges from [-1, 1]. Indicates wireless connectivity. Indicates perfect linear dependence. This indicates different degrees of linear correlation; The Spearman correlation coefficient, also known as the Spearman rank correlation, is independent of the specific values ​​of the two sets of state parameters. It is calculated based on the differences between each pair of ranks of the two sets of state parameters. Let the data sample be the first k The state parameter and the first l The state parameter data are as follows: Sort the two sets of state parameter data in ascending or descending order to obtain the corresponding vectors. , In the middle element The position in the middle is denoted as The Spearman correlation coefficients of the two state parameters are shown below.

[0017] in, For the k-th state parameter and the th Spearman correlation coefficients of the state parameters; for The average value;

[0018]

[0019] Substituting the above two equations, we obtain the Spearman correlation coefficient;

[0020] The Spearman correlation coefficient ranges from [-1, 1]. Indicates nonlinear correlation. Indicates perfect linear dependence. This indicates different degrees of linear correlation; The Kendall correlation coefficient, also known as the Kendall rank correlation coefficient, is a statistical value used to detect the correlation between two random variables. set up It is a two-dimensional detection value of two state parameters; when Then it is called and Satisfy synergy; conversely, when Then it is called and Incoordination; Kendall correlation coefficients are shown below;

[0021] in, Let be the Kendall correlation coefficient between the k-th state parameter and the l-th state parameter; The number of cooperating pairs is the logarithm; It is a non-cooperative logarithm; The number of samples; In summary, the Pearson correlation index is suitable for two linearly related states; the Spearman correlation index is independent of the specific values ​​of the state parameters and has low requirements for the data of the two states, so it is suitable for nonlinear data; the Kendall correlation coefficient has the same data requirements as Spearman, and is also suitable for situations where there is interference in the data. Due to the fluctuations in power and temperature, the data collected by the transformer data system has complex correlations. A comprehensive correlation index (CCI) is proposed as shown below.

[0022] in, For the first The state parameter and the first A comprehensive correlation index of several state parameters; , , These are the Pearson correlation coefficient, Spearman correlation coefficient, and Kendall correlation coefficient indices for the k-th state parameter and the l-th state parameter, respectively.

[0023] Compared with existing technologies: This invention facilitates the quantitative assessment and dynamic perception of the real-time operational health status of main transformer equipment. Specifically, it enables the timely capture of early signs of potential hazards, accurate diagnosis and real-time warning of the pathogenesis and location of potential faults, and accurate assessment of the likelihood of potential hazards developing into manifest faults. Through fault analysis and graded early warning strategies, information supports condition-based maintenance and proactive operation and maintenance, effectively curbing the further deterioration of defects and minimizing the impact, scope, and economic losses of accidents, thereby ensuring the safe and reliable operation of the main transformer. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the ultra-high frequency partial discharge detection method of the present invention; Figure 2 This is a schematic diagram of the online chromatographic monitoring system for transformers according to the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0026] This invention provides a transformer diagnostic analysis and early warning method based on a multi-source heterogeneous data model. Please refer to [link / reference]. Figures 1-2 The specific steps are as follows: Step 1, Partial Discharge: The partial discharge of the transformer is detected using the ultra-high frequency partial discharge detection method; The process of the UHF partial discharge detection method is as follows: the UHF sensor receives the UHF electromagnetic waves radiated by the partial discharge to realize the detection of the partial discharge; Specifically, to address the shortcomings of traditional detection methods, a new detection method has emerged in recent years – the ultra-high frequency (UHF) detection method. UHF partial discharge detection detects and locates partial discharges by detecting the UHF electrical signals generated by partial discharges inside the transformer, thus achieving anti-interference purposes. Each partial discharge is accompanied by a steep current pulse and radiates electromagnetic waves to the surrounding area. Studies have shown that the rise time of the partial discharge pulse in the transformer is approximately 12 ns, and the emitted electromagnetic waves contain a rich amount of UHF components. These UHF components can be received using capacitive sensors or UHF antennas. The UHF method differs from the pulse current method; the frequency measurement range of the pulse current method is generally no more than 1 MHz, while the frequency range of the UHF method is 300–3000 MHz. Studies suggest that each partial discharge process is accompanied by the neutralization of positive and negative charges and the appearance of a steep current pulse, while simultaneously radiating electromagnetic waves. The spectral characteristics of the electromagnetic waves radiated by partial discharge are related to the geometry of the partial discharge source and the insulation strength of the discharge gap. When the discharge gap is relatively small and the insulation strength of the discharge gap is relatively high, the discharge process is shorter, the current pulse is steeper, and the radiated electromagnetic wave signal has richer UHF components. Experiments have shown that partial discharges in power equipment such as GIS and transformers can excite very high-frequency electromagnetic waves, reaching up to several GHz. By using antenna sensors to receive the UHF electromagnetic waves radiated during the partial discharge process, the partial discharge situation can be detected. The ultra-high frequency method has a higher frequency band, which can effectively avoid electrical interference such as corona and switching operation in conventional testing, and has higher detection sensitivity. At present, because the measurement mechanism of the UHF method is different from that of the pulse current method, the sensitivity of external sensors is generally significantly lower than that of built-in sensors. Therefore, some modifications to the structure of the transformer at the site are generally required. This usually involves pre-embedding the sensor in the transformer or using an oil drain valve to extend the ultra-high frequency sensor into the transformer tank. Installation method: Install a UHF sensor on the transformer oil drain valve. The analog signal of the sensor is connected to the monitoring unit in the terminal cabinet next to the transformer. After signal conditioning and high-frequency acquisition, it is digitally processed and the final result is sent to the back-end server. Step 2, Oil Chromatography Monitoring and Analysis: The oil chromatogram is monitored and analyzed using a transformer online chromatographic monitoring system; The transformer online chromatographic monitoring system employs chromatographic analysis principles, utilizing dynamic headspace (purge-trap) degassing technology and a high-sensitivity microbridge detector to detect seven components (H2, CO, CO2, CH4, C2H4, C2H6, C2H2) in transformer oil. Optional additional features include the detection of O2, N2, and trace amounts of water. This system boasts high detection sensitivity, short analysis cycles, and comparability with laboratory data, making it a truly comprehensive online chromatographic monitoring unit. The entire system integrates chromatographic analysis, expert diagnostics, automatic control, and communication technology. By measuring and analyzing dissolved gases in insulating oil, it enables online monitoring of the internal operating status of large transformers, allowing for timely detection and diagnosis of internal faults and real-time monitoring of equipment operation. This overcomes the limitations of long monitoring cycles in laboratory chromatographic analysis, providing technical support for ensuring the safe and economical operation and condition-based maintenance of transformers. In addition, based on the principles of the three-ratio method, the David triangle method, and the cube diagram method, an online oil quality analysis database and analysis model for transformers are established within the online transformer chromatography monitoring system. The software system automatically detects and analyzes in real time, and automatically pushes the cause of the fault when an anomaly occurs, making the oil quality data analysis intelligent and real-time. The system also integrates offline transformer test data, and the offline data is correlated with the real-time data to form a micro-cloud database, which can generate correlation trend curves, providing convenience for technicians to further confirm and analyze in the background. The online chromatographic monitoring system for transformers consists of a chromatographic analysis system, a circuit control data storage and processing system, a communication module, a carrier gas, an oil sampling valve, and an ambient temperature regulator. The chromatographic analysis system includes an oil sample acquisition module, an oil-gas separation module, an automatic sample injection module, a sample component separation module, and a component detection module. 1) Chromatographic Analysis System: All chromatographic analysis processes are completed here. The function of the chromatographic analysis system is to complete the entire series of chromatographic analysis processes, including oil sample collection, oil-gas separation, automatic sample injection, component separation, and component detection, under the control of the circuit control system program instructions. Internally, it mainly includes oil-gas separation section, component detection section, gas path control section, etc. 2) Circuit control data storage and processing system: including circuit motherboard, various power supply modules, industrial control computer modules and other circuit components, its function is to control the circuit and gas path of the whole machine and to process, calculate and transmit the component signals detected by the chromatographic analysis system; 3) Communication module: Under the control of the main circuit, it completes the communication work with the client, including transmitting analysis data, transmitting control commands, and transmitting instrument status data. 4) Carrier gas: The system has a built-in gas source module that uses ambient air as the carrier gas source; 5) Power: 2.5kW; 6) Power supply voltage: 220V ±10%, 50Hz ±1%; 7) External dimensions (unit: mm): 627 (length) × 550 (width) × 1100 (height); 8) Weight: 75 kg; 9) Sampling valve: The sampling valve is a special valve machined according to the size of the oil drain port thread of the transformer. The valve is mainly composed of a valve body and a shut-off valve. After installation, it can provide oil samples to the online chromatograph through the shut-off valve, and can also perform fully sealed sampling through the oil drain port at the front to supply the laboratory for analysis of other items. Sampling valves are generally divided into two types: threaded and flanged, depending on the design of the transformer. Step 3, core grounding monitoring: A high-precision zero-flux current sensor is connected in series at the core grounding point to detect the current. Then, the weak current is amplified and transmitted, and finally, digital signal methods are used for calculation and analysis. Specifically, the online monitoring system for iron core grounding current employs a high-precision zero-flux current sensor, connected in series at the iron core grounding point. Taking into full account the amplification and transmission of weak currents, and utilizing digital signal methods for calculation and analysis, a PC is used as the hardware platform. The signal is digitized locally and uploaded to the next-level software monitoring system via a communication cable. The software employs modern digital signal processing theory and features functions such as live measurement of iron core grounding current, analysis and judgment, alarm, and remote data transmission. It can continuously, in real-time, and online monitor the iron core, enabling real-time monitoring of the equipment's insulation status. Simultaneously, by monitoring the iron core grounding current, it can promptly detect faults such as internal insulation dampness or damage, multi-point grounding of the iron core, foreign objects inside the enclosure, and sludge deposition in the oil tank, allowing for early prevention and handling of accidents. This provides a reliable technical basis for condition-based maintenance of such equipment. Step 4, Vibration Monitoring and Analysis: Monitor transformer vibration using an online electromagnetic oscillation monitoring device; The transformer electromagnetic oscillation online monitoring device consists of an acquisition module and an analysis and diagnosis module. The acquisition module includes 16 electromagnetic oscillation sensors that are set around the transformer to realize synchronous data acquisition of transformer vibration signals. The analysis and diagnosis module monitors and predicts data from the transformer core and windings and performs reasonable modeling of logic devices to diagnose and warn of latent faults in transformer operation. Data categories include frequency complexity, vibrational stability, energy similarity, and vibrational correlation. Frequency complexity The operation process is as follows: Step 1, Data Selection: The complexity of the frequency components in the reaction signal is considered. Lower complexity means the tank wall vibration energy is concentrated in a few frequency components; conversely, higher complexity means the energy is more dispersed. During normal transformer operation, the frequency components are typically concentrated in only a few finite frequencies, meaning the frequency complexity is relatively low. The transformer vibration components are mainly concentrated within 2000Hz, and all harmonic frequencies are integer multiples of 50Hz. Therefore, only harmonic components within the range of 50-2000Hz that are integer multiples of 50Hz are analyzed. The formula for calculating the harmonic proportion Pf at frequency f is defined as follows:

[0027] ; : The magnitude of the vibration harmonic amplitude at frequency f; The weighting coefficient with frequency f; Step 2, Frequency Complexity Calculation: Complexity of the Vibration Frequency Components of the Transformer Tank Wall : ; Vibration stability The operation process is as follows: Step 1, Data Selection: Reflects the mechanical stability of each vibration monitoring point of the transformer. When structural abnormalities occur, the system becomes uncertain and random vibrations will occur. A transformer under normal operation is a deterministic system, and will produce the same vibrations for the same input parameters and initial conditions. Step 2, Calculation: A recursion graph is represented by the following formula:

[0028] For a time series Uk (k=1,2,...n), with a sampling time interval of Δt, the appropriate embedding dimension m and delay time t are determined through relevant theoretical calculations, and then the time series is reconstructed. The reconstructed dynamical system is as follows:

[0029] In the formula, ; Calculate the distance between point xi (i) and point xj (j) in the reconstructed phase space:

[0030] In the formula, Represents the norm; Calculate the recursive value: ; Energy similarity The operation process is as follows: Step 1, Data Selection: The acquired vibration time-domain signal is subjected to Fourier transform, and the resulting frequency-domain vibration is divided into different energy bands. The transformer vibration components are mainly concentrated within 2000Hz, and all harmonic frequencies are integer multiples of 50Hz. According to the vibration characteristics, only the vibration components within 400-2000Hz are analyzed. These energies are divided into 4 groups, each with a bandwidth of 400Hz. The energy of the nth energy band is defined as follows:

[0031] Where Af is the vibration amplitude corresponding to frequency f, f is an integer, = (400, 450, ..., 2000); Step 2, Data Normalization: Convert the energy x(n) of the nth energy band into a relative percentage value:

[0032] Where u(1) is the relative percentage of energy converted from the first energy band (400-800Hz), and so on; Step 3, Average Energy Calculation: The target value is extracted from a large amount of historical data. Assuming there are N feature vector samples ui (i=1, 2, ..., N) (e.g., taking data from the past week, 7*24*6=1008 times), the average energy can be calculated as follows:

[0033] Step 4, Energy Similarity Calculation: Energy similarity analysis determines whether transformer vibration is abnormal by comparing the energy analysis of the measured signal with the target energy distribution. When the energy similarity value of a certain measuring point suddenly increases, it means that the mechanical structure of the transformer near that measuring point may be abnormal. Energy similarity is expressed by energy difference rate, which is the average distance between multiple measured values ​​and the target value.

[0034] Vibration correlation The operation process is as follows: Step 1, Data Selection: Based on the transformer vibration principle, the vibration of both the winding and the core is based on 100Hz as the fundamental frequency. The diagnostic model for correlation analysis only needs to consider the 100Hz component of the vibration. α1 and α2 are the vibration values ​​under different loads, respectively, and β is the phase difference between the two vibrations (i.e., the power factor angle). The COSβ of the high-voltage side vibration sensor is taken from the COSβ of the main transformer's high-voltage side, and the COSβ of the low-voltage side vibration sensor is taken from the COSβ of the main transformer's low-voltage side.

[0035] Its Δωij is the vibration change obtained from the j-th vibration sensor and from the i-th and i+1-th measurements, and the characteristic matrix Xorigin is composed of Δωij, note that n>m; ; Step 2, Data Dimensionality Reduction: Principal component analysis is used to reduce the dimensionality of the data. First, the above feature matrices are normalized:

[0036] Next, the corresponding covariance matrix is ​​calculated for the above data, and then the matrix is ​​decomposed by eigenvalue decomposition or singular value decomposition, where each column in matrix U represents an eigenvector.

[0037]

[0038] in: ; Step 3, Calculate vibration characteristic parameters: Vibration characteristic parameters: , which represents the proportion of the principal component's energy to the total energy;

[0039] Step 5, Transformer temperature monitoring: Two sets of temperature sensors are configured on each transformer, with two temperature sensors in each set. They are arranged at the upper and lower oil inlets of the transformer radiator to monitor the heat dissipation effect of the transformer. The two sets of temperature sensors are arranged diagonally. Step 6: Combine the linear correlation index and the two nonlinear indices to establish a comprehensive correlation index to analyze the correlation between transformer state parameters. The linear correlation index is the Pearson correlation coefficient, and the non-linear indexes are the Spearman correlation coefficient and the Kendall correlation coefficient. The comprehensive correlation index is established by taking the average of the three. The Pearson correlation coefficient, also known as the Pearson correlation coefficient, is often used to describe the relationship between two continuous variables, reflecting the degree of linear correlation between the two variables. The Pearson correlation coefficients between the k-th state and the l-th state parameter of the data sample are shown below;

[0040] in, , The first k The state parameter and the first l The average value of each state parameter data; Let be the Pearson correlation coefficient between the k-th state parameter and the l-th state parameter; The Pearson correlation coefficient ranges from [-1, 1]. Indicates wireless connectivity. Indicates perfect linear dependence. This indicates different degrees of linear correlation; The Spearman correlation coefficient, also known as the Spearman rank correlation, is independent of the specific values ​​of the two sets of state parameters. It is calculated based on the differences between each pair of ranks of the two sets of state parameters. Let the data sample be the first k The state parameter and the first l The state parameter data are as follows: Sort the two sets of state parameter data in ascending or descending order to obtain the corresponding vectors. , In the middle element The position in the middle is denoted as The Spearman correlation coefficients of the two state parameters are shown below.

[0041] in, For the k-th state parameter and the th Spearman correlation coefficients of the state parameters; for The average value;

[0042]

[0043] Substituting the above two equations, we obtain the Spearman correlation coefficient;

[0044] The Spearman correlation coefficient ranges from [-1, 1]. Indicates nonlinear correlation. Indicates perfect linear dependence. This indicates different degrees of linear correlation; The Kendall correlation coefficient, also known as the Kendall rank correlation coefficient, is a statistical value used to detect the correlation between two random variables. set up It is a two-dimensional detection value of two state parameters; when Then it is called and Satisfy synergy; conversely, when Then it is called and Incoordination; Kendall correlation coefficients are shown below;

[0045] in, Let be the Kendall correlation coefficient between the k-th state parameter and the l-th state parameter; The number of cooperating pairs is the logarithm; It is a non-cooperative logarithm; The number of samples; In summary, the Pearson correlation index is suitable for two linearly related states; the Spearman correlation index is independent of the specific values ​​of the state parameters and has low requirements for the data of the two states, so it is suitable for nonlinear data; the Kendall correlation coefficient has the same data requirements as Spearman, and is also suitable for situations where there is interference in the data. Due to the fluctuations in power and temperature, the data collected by the transformer data system has complex correlations. A comprehensive correlation index (CCI) is proposed as shown below.

[0046] in, For the first The state parameter and the first A comprehensive correlation index of several state parameters; , , These are the Pearson correlation coefficient, Spearman correlation coefficient, and Kendall correlation coefficient indices for the k-th state parameter and the l-th state parameter, respectively.

[0047] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for transformer diagnosis, analysis, and early warning based on a multi-source heterogeneous data model, characterized in that, The specific steps are as follows: Step 1, Partial Discharge: The partial discharge of the transformer is detected using the ultra-high frequency partial discharge detection method; Step 2, Oil Chromatography Monitoring and Analysis: The oil chromatogram is monitored and analyzed using a transformer online chromatographic monitoring system; Step 3, core grounding monitoring: A high-precision zero-flux current sensor is connected in series at the core grounding point to detect the current. Then, the weak current is amplified and transmitted, and finally, digital signal methods are used for calculation and analysis. Step 4, Vibration Monitoring and Analysis: Monitor transformer vibration using an online electromagnetic oscillation monitoring device; Step 5, Transformer temperature monitoring: Two sets of temperature sensors are configured on each transformer, with two temperature sensors in each set. They are arranged at the upper and lower oil inlets of the transformer radiator to monitor the heat dissipation effect of the transformer. The two sets of temperature sensors are arranged diagonally. Step 6: Combine the linear correlation index and the two nonlinear indices to establish a comprehensive correlation index to analyze the correlation between transformer state parameters. The linear correlation index is the Pearson correlation coefficient, and the nonlinear index is the Spearman correlation coefficient and the Kendall correlation coefficient. A comprehensive correlation index is established by taking the average of the three. The Pearson correlation coefficient, also known as the Pearson correlation coefficient, is used to measure the relationship between two continuous variables and reflects the degree of linear correlation between the two variables. The Pearson correlation coefficients between the k-th state and the l-th state parameter of the data sample are shown below; ， in, , The first k The state parameter and the first l The average value of each state parameter data; Let be the Pearson correlation coefficient between the k-th state parameter and the l-th state parameter; The Pearson correlation coefficient ranges from [-1, 1]. Indicates wireless connectivity. Indicates perfect linear dependence. This indicates different degrees of linear correlation; The Spearman correlation coefficient, also known as the Spearman rank correlation, is independent of the specific values ​​of the two sets of state parameters. It is calculated based on the differences between each pair of ranks of the two sets of state parameters. Let the data sample be the first k The state parameter and the first l The state parameter data are as follows: Sort the two sets of state parameter data in ascending or descending order to obtain the corresponding vectors. , In the middle element The position in the middle is denoted as The Spearman correlation coefficients of the two state parameters are shown below. ， in, For the k-th state parameter and the th Spearman correlation coefficients of the state parameters; for The average value; ， ， Substituting the above two equations, we obtain the Spearman correlation coefficient; ， The Spearman correlation coefficient ranges from [-1, 1]. Indicates nonlinear correlation. Indicates perfect linear dependence. This indicates different degrees of linear correlation; The Kendall correlation coefficient, also known as the Kendall rank correlation coefficient, is a statistical value used to detect the correlation between two random variables. set up It is a two-dimensional detection value of two state parameters; when Then it is called and Satisfy synergy; conversely, when Then it is called and Incoordination; Kendall correlation coefficients are shown below; ， in, Let be the Kendall correlation coefficient between the k-th state parameter and the l-th state parameter; The number of cooperating pairs is the logarithm; It is a non-cooperative logarithm; The number of samples; In summary, the Pearson correlation index is suitable for two linearly related states; the Spearman correlation index is independent of the specific values ​​of the state parameters and has low requirements for the data of the two states, so it is suitable for nonlinear data; the Kendall correlation coefficient has the same data requirements as Spearman, and is also suitable for situations where there is interference in the data. Due to the fluctuations in power and temperature, the data collected by the transformer data system has complex correlations. A comprehensive correlation index (CCI) is proposed as shown below. ， in, For the first The state parameter and the first A comprehensive correlation index of several state parameters; , , These are the Pearson correlation coefficient, Spearman correlation coefficient, and Kendall correlation coefficient indices for the k-th state parameter and the l-th state parameter, respectively.

2. The transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model according to claim 1, characterized in that, The process of the UHF partial discharge detection method is as follows: the UHF electromagnetic waves radiated by the partial discharge are received by a UHF sensor to realize the detection of partial discharge.

3. The transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model according to claim 1, characterized in that, The transformer online chromatographic monitoring system consists of a chromatographic analysis system, a circuit control data storage and processing system, a communication module, a carrier gas, an oil sampling valve, and an ambient temperature regulator.

4. The transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model according to claim 3, characterized in that, The chromatographic analysis system includes an oil sample acquisition module, an oil-gas separation module, an automatic sample injection module, a sample component separation module, and a component detection module.

5. The transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model according to claim 1, characterized in that, The transformer electromagnetic oscillation online monitoring device consists of an acquisition module and an analysis and diagnosis module. The acquisition module includes 16 electromagnetic oscillation sensors that are respectively set around the transformer to realize synchronous data acquisition of transformer vibration signals. The analysis and diagnosis module realizes the diagnosis and early warning of latent faults in transformer operation by monitoring and predicting the data of transformer core and windings and reasonably modeling the logic devices. The data types include frequency complexity, vibration stability, energy similarity, and vibration correlation.

6. The transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model according to claim 5, characterized in that, The frequency complexity The operation process is as follows: Step 1, Data Selection: This reflects the complexity of the frequency components in the signal. The lower the complexity, the more concentrated the tank wall vibration energy is on a few frequency components. Conversely, the higher the complexity, the more dispersed the energy. During normal operation, the frequency components of a transformer are typically concentrated on only a few finite frequencies, meaning the frequency complexity is low. The transformer's vibration components are mainly concentrated within 2000Hz, and all harmonic frequencies are integer multiples of 50Hz. Therefore, only harmonic components within the range of 50-2000Hz that are integer multiples of 50Hz are analyzed. The formula for calculating the harmonic proportion Pf at frequency f is defined as follows: ， ； : The magnitude of the vibration harmonic amplitude at frequency f; The weighting coefficient with frequency f; Step 2, Frequency Complexity Calculation: Complexity of the Vibration Frequency Components of the Transformer Tank Wall : 。 7. The transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model according to claim 5, characterized in that, Vibration stability The operation process is as follows: Step 1, Data Selection: Reflects the mechanical stability of each vibration monitoring point of the transformer. When the structure is abnormal, the system becomes uncertain and random vibration will occur. The transformer under normal operation is a deterministic system, and it will produce the same vibration for the same input parameters and initial conditions. Step 2, Calculation: A recursion graph is represented by the following formula: ， For a time series Uk (k=1,2,...n), with a sampling time interval of Δt, the appropriate embedding dimension m and delay time τ are determined through relevant theoretical calculations, and then the time series is reconstructed. The reconstructed dynamical system is as follows: ， In the formula, ; Calculate the distance between point xi (i) and point xj (j) in the reconstructed phase space: ， In the formula, Represents the norm; Calculate the recursive value: 。 8. The transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model according to claim 5, characterized in that, The energy similarity The operation process is as follows: Step 1, Data Selection: The acquired vibration time-domain signal is subjected to Fourier transform, and the resulting frequency-domain vibration is divided into different energy bands. The transformer vibration components are mainly concentrated within 2000Hz, and all harmonic frequencies are integer multiples of 50Hz. According to the vibration characteristics, only the vibration components within 400-2000Hz are analyzed. These energies are divided into 4 groups, each with a bandwidth of 400Hz. The energy of the nth energy band is defined as follows: ， Where Af is the vibration amplitude corresponding to frequency f, and f takes integer values ​​(400, 450, ..., 2000); Step 2, Data Normalization: Convert the energy x(n) of the nth energy band into a relative percentage value: ， Where u(1) is the relative percentage of energy converted from the first energy band 400-800Hz, and so on; Step 3, Average Energy Calculation: The target value is extracted from a large amount of historical data. Assuming there are N data points, taking 7*24*6=1008 data points from the past week, and the feature vector samples ui (i=1, 2, ..., N), the average energy can be calculated as follows: ， Step 4, Energy Similarity Calculation: Energy similarity analysis determines whether transformer vibration is abnormal by comparing the energy analysis of the measured signal with the target energy distribution. When the energy similarity value of a certain measuring point suddenly increases, it means that the mechanical structure of the transformer near that measuring point is abnormal. Energy similarity is expressed by energy difference rate, which is the average distance between multiple measured values ​​and the target value.

9. The transformer diagnosis, analysis, and early warning method based on a multi-source heterogeneous data model according to claim 5, characterized in that, The vibration correlation The operation process is as follows: Step 1, Data Selection: Based on the transformer vibration principle, the vibration of both the winding and the core is based on 100Hz as the fundamental frequency. The diagnostic model for correlation analysis only needs to consider the 100Hz component of the vibration. α1 and α2 are the vibration values ​​under different loads, respectively, and β is the phase difference between the two vibrations, which is the power factor angle. The COSβ of the high-voltage side vibration sensor is taken from the COSβ of the main transformer high-voltage side, and the COSβ of the low-voltage side vibration sensor is taken from the COSβ of the main transformer low-voltage side. ， Its Δωij is the vibration change obtained from the j-th vibration sensor and from the i-th and i+1-th measurements, and the characteristic matrix Xorigin is composed of Δωij, note that n>m; ； Step 2, Data Dimensionality Reduction: Principal component analysis is used to reduce the dimensionality of the data. First, the above feature matrices are normalized: ， Next, the corresponding covariance matrix is ​​calculated for the above data, and then the matrix is ​​decomposed by eigenvalue decomposition or singular value decomposition, where each column in matrix U represents an eigenvector. ， ， in: ; Step 3, Calculate vibration characteristic parameters: Vibration characteristic parameters: , which represents the proportion of the principal component's energy to the total energy;

Images