Wind power substation main transformer state monitoring method and system based on digital twinning

Abstract

The application discloses a wind power substation main transformer state monitoring method and system based on digital twinning, which first acquires characteristic gas concentration and top layer oil temperature time sequence, then carries out multilayer wavelet packet transformation on the gas concentration sequence, and decomposes the same to different frequency bands; then, based on the main transformer top layer oil temperature time sequence, correlation analysis is used to identify and group the frequency bands, so as to accurately decouple the components related to temperature effect and the components truly representing fault characteristics, finally realize signal reconstruction and extract pure fault characteristics. In this way, the accuracy and reliability of the wind power substation main transformer state monitoring are significantly improved, so that fine perception and early warning of the equipment operating condition can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent monitoring, and more specifically, to a wind power substation main transformer state monitoring method and system based on digital twinning. BACKGROUND

[0002] Wind power, as an important part of clean energy, plays a key role in the transformation of global energy structure. Wind power substation is the core hub of wind power grid connection, and the safe and stable operation of the main transformer, as its key equipment, is directly related to the power generation efficiency of the entire wind farm and the reliability of the power grid. However, the main transformer is long-term operated in the variable environment of the wind farm, and is subjected to frequent load fluctuations and complex stresses, which is prone to insulation aging and even internal failure. Once the main transformer fails, not only will it cause huge economic losses, but also will have a serious impact on the stable operation of the power grid. Therefore, it is of great significance to build an efficient and accurate wind power substation main transformer state monitoring scheme to ensure the safety of wind power assets, prolong the service life of equipment, and improve the reliability of the power grid.

[0003] However, the current wind power substation main transformer state monitoring faces many challenges. The aging rate of transformer solid insulation is extremely sensitive to temperature, and the severe fluctuations of wind farm load cause frequent and large fluctuations of internal winding hot spot temperature, forming a "thermal cycle" that accelerates insulation aging. Traditional life assessment models, such as the method based on IEEE C57.91-2011 standard, usually assume slow temperature changes and mainly focus on equivalent thermal aging effects, but ignore the nonlinear damage caused by "thermal cycle", such as the mechanical fatigue of insulation paper in repeated thermal expansion and cold contraction, and its coupling effect with chemical aging, which cannot be effectively captured by these traditional models. In addition, existing monitoring systems mainly focus on the amplitude of temperature, and cannot effectively capture key features such as temperature change rate and fluctuation frequency, resulting in inaccurate assessment of the actual aging speed of the transformer under dynamic load when the average load and maximum temperature are similar. More importantly, in the state monitoring based on dissolved gas analysis (DGA), the thermal inertia effect inside the transformer causes a significant time delay between gas concentration changes and temperature fluctuations that drive them. Conventional instantaneous correlation analysis methods often underestimate the true correlation strength between gas concentration and oil temperature due to the failure to consider this time lag, which can easily misjudge the frequency band dominated by temperature effects as a fault feature, or drown the real fault signal in the temperature background noise, thereby introducing uncertainty and misjudgment risk for subsequent fault diagnosis, seriously affecting the sensitivity and reliability of diagnosis.

[0004] Therefore, an optimized wind power substation main transformer state monitoring method based on digital twinning is expected. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. The present application provides a wind power substation main transformer state monitoring method and system based on digital twinning, which first acquires characteristic gas concentration and top layer oil temperature time series, then performs multi-layer wavelet packet transformation on the gas concentration series to decompose it into different frequency bands; then, based on the main transformer top layer oil temperature time series, correlation analysis is used to identify and group these frequency bands, thereby accurately decoupling the components related to temperature effect and the components truly representing fault characteristics, and finally realizing signal reconstruction and extracting pure fault characteristics. In this way, the accuracy and reliability of the wind power substation main transformer state monitoring are significantly improved, thereby enabling fine perception and early warning of the equipment operating condition.

[0006] According to one aspect of the present application, a wind power substation main transformer state monitoring method based on digital twinning is provided, which comprises:

[0007] acquiring an original characteristic gas concentration time series and a main transformer top layer oil temperature time series;

[0008] performing multi-layer wavelet packet transformation on the original characteristic gas concentration time series to obtain wavelet packet coefficient series of all frequency bands;

[0009] based on the main transformer top layer oil temperature time series, performing correlation-based frequency band identification and grouping on the wavelet packet coefficient series of all frequency bands to obtain a temperature effect related frequency band index set and a fault characteristic related frequency band index set;

[0010] based on the temperature effect related frequency band index set and the fault characteristic related frequency band index set, performing signal reconstruction and decoupling on the wavelet packet coefficient series of all frequency bands to obtain a reconstructed pure temperature effect component and a reconstructed pure fault characteristic component;

[0011] performing fault diagnosis based on the reconstructed pure fault characteristic component to obtain a fault type diagnosis result.

[0012] According to another aspect of the present application, a wind power substation main transformer state monitoring system based on digital twinning is provided, which comprises:

[0013] a data acquisition module for acquiring an original characteristic gas concentration time series and a main transformer top layer oil temperature time series;

[0014] a multi-layer wavelet packet transformation module for performing multi-layer wavelet packet transformation on the original characteristic gas concentration time series to obtain wavelet packet coefficient series of all frequency bands;

[0015] a frequency band identification and grouping module configured to perform correlation-based frequency band identification and grouping on wavelet packet coefficient sequences of all frequency bands based on the time sequence of the top oil temperature of the main transformer to obtain a temperature effect related frequency band index set and a fault feature related frequency band index set;

[0016] a signal reconstruction and decoupling module configured to perform signal reconstruction and decoupling on the wavelet packet coefficient sequences of all frequency bands based on the temperature effect related frequency band index set and the fault feature related frequency band index set to obtain a reconstructed pure temperature effect component and a reconstructed pure fault feature component;

[0017] a fault diagnosis module configured to perform fault diagnosis based on the reconstructed pure fault feature component to obtain a fault type diagnosis result.

[0018] Compared with the prior art, the wind power substation main transformer state monitoring method and system based on digital twinning provided by the present application first acquires a characteristic gas concentration and a time sequence of a top oil temperature, then performs multi-layer wavelet packet transform on the gas concentration sequence to decompose it into different frequency bands, then performs identification and grouping on these frequency bands based on the time sequence of the top oil temperature of the main transformer using correlation analysis, thereby accurately decoupling components related to temperature effects and components that truly represent fault features, finally reconstructing the signals and extracting pure fault features. In this way, the accuracy and reliability of the wind power substation main transformer state monitoring are significantly improved, thereby enabling fine perception and early warning of the equipment operating condition. BRIEF DESCRIPTION OF DRAWINGS

[0019] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application, when taken in conjunction with the accompanying drawings. The drawings provided in the present application are used to provide further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 a flowchart of the wind power substation main transformer state monitoring method based on digital twinning according to the embodiments of the present application;

[0021] Figure 2 a data flow schematic diagram of the wind power substation main transformer state monitoring method based on digital twinning according to the embodiments of the present application;

[0022] Figure 3 a block diagram of the wind power substation main transformer state monitoring system based on digital twinning according to the embodiments of the present application. DETAILED DESCRIPTION

[0023] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part but not all of the embodiments of the present application, and the present application can be implemented in many different ways. Therefore, the attached drawings should not be used to limit and define the present application and the present application should cover all changes falling within the scope of the present application.

[0024] As shown in the present application and claims, unless the context clearly indicates otherwise, the words "one", "an", "a", and / or "the" do not mean to specify only one, but also include the plural. Generally, the terms "comprise" and "include" only indicate the inclusion of the steps and elements explicitly identified, and these steps and elements do not constitute an exclusive list, and the method or device can also include other steps or elements.

[0025] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0026] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously, as needed. Meanwhile, other operations can be added to these processes, or one or more steps of operations can be removed from these processes.

[0027] In the technical solutions of the present application, a wind power substation main transformer state monitoring method based on digital twinning is proposed. Figure 1 A flowchart of the wind power substation main transformer state monitoring method based on digital twinning according to the embodiments of the present application. Figure 2 A data flow schematic diagram of the wind power substation main transformer state monitoring method based on digital twinning according to the embodiments of the present application. As Figure 1 and Figure 2As shown, the wind power substation main transformer state monitoring method based on digital twinning according to the embodiments of the present application comprises the steps: S1, obtaining the original characteristic gas concentration time sequence and the main transformer top layer oil temperature time sequence; S2, performing multi-layer wavelet packet transform on the original characteristic gas concentration time sequence to obtain the wavelet packet coefficient sequence of all frequency bands; S3, based on the main transformer top layer oil temperature time sequence, performing correlation-based frequency band identification and grouping on the wavelet packet coefficient sequence of all frequency bands to obtain a temperature effect related frequency band index set and a fault feature related frequency band index set; S4, based on the temperature effect related frequency band index set and the fault feature related frequency band index set, performing signal reconstruction and decoupling on the wavelet packet coefficient sequence of all frequency bands to obtain a reconstructed pure temperature effect component and a reconstructed pure fault feature component; S5, performing fault diagnosis based on the reconstructed pure fault feature component to obtain a fault type diagnosis result.

[0028] In particular, the S1, obtaining the original characteristic gas concentration time sequence and the main transformer top layer oil temperature time sequence. It should be understood that the wind power substation main transformer, as a key node of the power grid, its operating state directly affects the power generation efficiency of the entire wind farm and the stability of the power grid. The aging and potential failure of the internal insulation material of the transformer often accompanies the change of the concentration of specific gases in the insulating oil, that is, dissolved gas analysis (DGA) is the core means for diagnosing transformer faults. At the same time, the temperature inside the transformer, especially the top layer oil temperature, is a key physical parameter that affects gas generation, dissolution and migration, and under the variable operating environment of the wind farm, the "thermal cycle" effect caused by load fluctuation is significant, so that there is a complex dynamic correlation and time lag effect between oil temperature and gas concentration. Therefore, accurate acquisition of these two time series data not only directly reflects the health status of the transformer, but also lays a data foundation for subsequent decoupling of temperature influence and extraction of pure fault features, thereby ensuring the accuracy and reliability of the diagnosis.

[0029] Specifically, the original characteristic gas concentration time sequence refers to the ordered data set of the concentration of key characteristic gases such as hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide extracted and measured from the transformer insulating oil over time under continuous or periodic monitoring. These gases will be generated in different proportions and concentrations under different fault types of the transformer (such as partial discharge, overheating, arc), and their time series data can reveal the evolution trend of the fault. The main transformer top layer oil temperature time sequence refers to the continuous measurement value of the temperature of the transformer top insulating oil over time. The top layer oil temperature is an important indicator for measuring the overall thermal load and heat dissipation condition of the transformer, and its fluctuation directly affects the aging rate of the insulating material and the behavior of the dissolved gas in the oil. The synchronous acquisition of these two time series data is particularly important for understanding the internal physicochemical mechanism between fault gas generation and temperature, especially considering the time lag effect caused by the thermal inertia inside the transformer.

[0030] In implementation, the original feature gas concentration time series and the main transformer top layer oil temperature time series are usually obtained by relying on advanced online monitoring systems deployed on the main transformer of the wind power substation. For gas concentration data, an online dissolved gas analysis (DGA) monitoring device is used, which can extract samples from the transformer oil in real time or quasi-real time, and separate and quantitatively analyze the dissolved gas by techniques such as gas chromatography, photoacoustic spectroscopy or infrared absorption, record the obtained various feature gas concentration values together with the corresponding sampling time stamps to form continuous time series data. At the same time, in order to obtain the top layer oil temperature time series, the transformer is usually equipped with high-precision temperature sensors such as platinum resistance (RTD) or thermocouple, which are installed at the top of the transformer oil tank to continuously monitor and record oil temperature data. These temperature data also have accurate time stamps, which are synchronized in time with the gas concentration data. All the collected original data, including gas concentration and oil temperature, are transmitted to the data acquisition system or digital twin platform for unified storage and management, providing reliable and comprehensive data support for subsequent signal processing and fault diagnosis.

[0031] In particular, S2, the original feature gas concentration time series is subjected to multi-layer wavelet packet transform to obtain wavelet packet coefficient sequences of all frequency bands. It should be understood that the original feature gas concentration time series is often complex and non-stationary, and may mix signals generated by multiple physical and chemical processes inside the transformer, including normal temperature effects and potential fault features. In order to effectively identify and distinguish these signals from different sources, the original signal needs to be carefully decomposed in time and frequency. As an advanced time-frequency analysis tool, multi-layer wavelet packet transform can provide more detailed frequency band division than traditional Fourier transform or single wavelet transform, thereby revealing hidden information of the signal at different frequency scales, and providing more distinguishable basic data for subsequent signal reconstruction and fault diagnosis. Therefore, in the technical solution of the present application, the original feature gas concentration time series is subjected to multi-layer wavelet packet transform to obtain wavelet packet coefficient sequences of all frequency bands. Specifically, multi-layer wavelet packet transform is a mathematical tool that decomposes signals in different frequency subbands. It not only decomposes the low-frequency (approximation) part like wavelet transform, but also further decomposes the high-frequency (detail) part, thereby achieving comprehensive and fine division of the entire frequency domain. The obtained wavelet packet coefficient sequence refers to the sequence of wavelet packet decomposition coefficients corresponding to the original signal in each specific frequency subband after multi-layer wavelet packet transform. These coefficients represent the energy distribution and local feature information of the signal in that frequency band.

[0032] In implementation, first, the wavelet base function and the target decomposition level are determined. The selection of the wavelet base function is crucial, and its shape and symmetry (e.g. Daubechies wavelet, Symlets wavelet, etc.) should match the characteristics of the signal to be analyzed to maximize the capture of the local characteristics of the signal; the target decomposition level defines the depth of signal decomposition and the degree of precision of frequency band division; the higher the decomposition level, the more frequency bands obtained, the stronger the ability to analyze the frequency details of the signal, but at the same time, the calculation complexity will also increase;

[0033] Further, based on the wavelet base function and the target decomposition level, the original characteristic gas concentration time series is decomposed and iterated to obtain the wavelet packet coefficient sequence of all frequency bands. That is, after determining the appropriate base function and decomposition level, the wavelet packet transform realizes the decomposition of the signal by iteratively applying a pair of orthogonal filters (low-pass filter and high-pass filter). In each iteration, a signal or sub-band is decomposed into two new sub-bands: one representing its low-frequency component (approximation coefficient), and the other representing its high-frequency component (detail coefficient). Unlike traditional wavelet transform, which only iteratively decomposes the low-frequency component, wavelet packet transform further decomposes each branch (including low-frequency and high-frequency) until the target decomposition level is reached. For example, for an L-level decomposition, the original signal will finally be decomposed into non-overlapping frequency bands. Each frequency band is represented by its corresponding set of wavelet packet coefficient sequences, which accurately capture the dynamic changes of the original gas concentration signal in a specific frequency range.

[0034] In particular, the S3, based on the main transformer top oil temperature time series, performs correlation-based frequency band identification and grouping on the wavelet packet coefficient sequence of all frequency bands to obtain a temperature effect related frequency band index set and a fault feature related frequency band index set. It should be understood that in the transformer, a complex physical system with significant thermal inertia, there is a non-negligible time delay between the gas concentration signal and the temperature fluctuation driving the change. The conventional frequency band identification method often judges by calculating the instantaneous Pearson correlation coefficient, which inherently assumes that the change in gas concentration and temperature fluctuation occur synchronously. However, the time delay exists in the physical and chemical processes such as heat conduction from the fault point to the oil body, gas precipitation and migration to the sensor, which leads to the original instantaneous correlation analysis to underestimate the correlation strength of the two when there is a phase difference, and may misjudge the frequency band essentially dominated by temperature effect as irrelevant fault feature, thereby introducing uncertainty and potential misjudgment risk for subsequent fault diagnosis. Therefore, in order to accurately distinguish the source of the signal and accurately decouple the temperature effect component and the feature component that truly reflects the potential fault, in the technical solution of the present application, the dynamic time lag relationship is explored to ensure the accuracy and reliability of subsequent fault diagnosis. Specifically, the cross-correlation function is used to quantify the similarity of two time series at different time shifts to capture the delay effect of temperature change on gas concentration. The temperature effect related frequency band index set finally obtained refers to the set of indexes (identifiers) corresponding to the frequency bands whose wavelet packet coefficient sequences are significantly correlated with the main transformer top oil temperature time series after considering the time lag. These frequency bands mainly reflect the influence of temperature change caused by normal operation or load fluctuation of the transformer on gas concentration; while the fault feature related frequency band index set refers to the set of indexes corresponding to those frequency bands which are not significantly correlated with the top oil temperature time series (or significantly correlated but unreasonable in time lag). The fluctuations of these frequency bands are more likely to come from potential faults or abnormalities inside the transformer.

[0035] In specific implementation, first, the wavelet packet coefficient sequence of the first frequency band of the first layer is extracted from the wavelet packet coefficient sequence of all frequency bands. The wavelet packet coefficient sequence of the first frequency band of the first layer is extracted from the wavelet packet coefficient sequence of all frequency bands. This means that for each specific frequency sub-band obtained by wavelet packet transform decomposition, its corresponding wavelet packet coefficient sequence will be individually investigated as a gas concentration signal component to be analyzed.

[0036] Next, the cross-correlation function of the wavelet packet coefficient sequence of the first frequency band of the first layer and the main transformer top oil temperature time series is calculated. The cross-correlation function of the wavelet packet coefficient sequence of the first frequency band of the first layer and the main transformer top oil temperature time series is calculated. The maximum cross-correlation coefficient and optimal time delay between the wavelet packet coefficient sequence of a frequency band and the time series of the top oil temperature of the main transformer are determined. It is understandable that conventional frequency band identification methods exhibit a fundamental limitation in distinguishing signal sources. This method determines the source by calculating the instantaneous Pearson correlation coefficient between the wavelet packet coefficient sequence and the oil temperature sequence, based on the assumption that changes in gas concentration and the temperature fluctuations driving those changes occur synchronously. However, in the complex physical system of a wind power substation main transformer with significant thermal inertia, the conduction of heat from the local fault point to the oil, and the precipitation and migration of gas due to decreased solubility to the online monitoring sensors all involve non-negligible time delays. Therefore, the original instantaneous correlation analysis, failing to consider this time delay characteristic, may incorrectly underestimate the correlation strength between the gas concentration signal and the oil temperature signal when there is a phase difference. This could lead to misjudging a frequency band essentially dominated by temperature effects as an irrelevant fault feature, thus introducing uncertainty and potential misjudgment risks into subsequent fault diagnosis.

[0037] To overcome the aforementioned shortcomings, a frequency band adaptive identification mechanism based on time-delay-aware cross-correlation maximum entropy and physical constraints is proposed. This mechanism abandons instantaneous, static correlation measurements and instead delves into the dynamic time-delay relationship between signals, achieving precise decoupling of the temperature effect frequency band and the fault characteristic frequency band through the following steps.

[0038] In this process, firstly, calculate the... Layer The cross-correlation function between the wavelet packet coefficient sequence of each frequency band and the time series of the top oil temperature of the main transformer is used. In other words, by calculating the cross-correlation function, a tool is constructed to quantify the similarity between the two time series at different time shifts, in order to capture the delayed effect of temperature changes on gas concentration. Specifically, for each gas concentration frequency band coefficient sequence, a cross-correlation operation is performed with the synchronized oil temperature time series to generate a cross-correlation function regarding the time delay; this process is expressed by the formula:

[0039] ,

[0040] in, It is a cross-correlation function. For the first Layer Wavelet packet coefficient sequence of each frequency band The time series of top oil temperature of the main transformer. Due to time lag, The sequence length is given; this function fully depicts the correlation between a specific gas concentration band and oil temperature fluctuations throughout the entire time-delay scan range, providing a comprehensive data foundation for revealing the true intrinsic relationship between the two that may be obscured by time delay.

[0041] Secondly, the maximum cross-correlation coefficient and optimal time delay are extracted from the cross-correlation function. That is, from the generated cross-correlation function, the key parameters that best represent the essence of the correlation between signals are extracted, simplifying the complex functional relationship into a scalar indicator that can be used for decision-making. Specifically, the normalized cross-correlation function... In all possible time delays A search is performed on the upper bound of its absolute value to determine the time delay corresponding to reaching that upper bound. In this process, firstly, based on the... Layer The standard deviations of the wavelet packet coefficient sequences of each frequency band and the time series of the top oil temperature of the main transformer are used to transform the cross-correlation function to obtain the cross-correlation coefficient function; this process is expressed by the formula:

[0042] ,

[0043] in, It is a cross-correlation function. and The first Layer The standard deviations of the wavelet packet coefficient sequences and the top oil temperature time series of the main transformer are analyzed. Then, the maximum correlation coefficient and its corresponding time delay are searched from the cross-correlation function to obtain the maximum cross-correlation coefficient and the optimal time delay. That is, the supremum of the absolute value of the normalized cross-correlation function is searched over all possible time delays, and the time delay corresponding to reaching this supremum is recorded. This process is expressed by the formula:

[0044] ,

[0045] ,

[0046] in, For cross-correlation function, This indicates a search for the maximum value of a function over its entire domain. This indicates that the function returns the value of the independent variable when it reaches its maximum value. To maximize the cross-correlation coefficient, This is the optimal time delay.

[0047] In this way, the maximum linear correlation strength between the two signals, unaffected by the phase difference, can be precisely quantified. And the optimal time delay required to achieve this maximum correlation strength. . Its execution effect is, output two parameters with clear physical meaning: degree represent the real strength of the association, while reveals the dynamic response time of the system to temperature disturbance, together constitute a solid basis for subsequent intelligent grouping;

[0048] Further, based on the maximum cross-correlation coefficient and the optimal time delay, the wavelet packet coefficient sequence of the first layer and the first frequency band is adaptively grouped to determine whether it belongs to the temperature effect related frequency band index set or the fault feature related frequency band index set. That is, a set of rigorous decision logic needs to be established to avoid misjudging the occasional higher correlation or the correlation at the physically unreasonable time delay as effective association. Specifically, a double-verification discriminant function is constructed, which performs two tests in parallel: first, the maximum cross-correlation coefficient is converted into a variable that approximately obeys normal distribution using Fisher's z-transformation, and it is tested whether it is significant at a given statistical significance level ; second, it is tested whether the extracted optimal time delay falls within a physically reasonable time delay window preset according to the knowledge of transformer heat transfer and fluid dynamics.

[0049] In this process, first, the maximum cross-correlation coefficient is transformed into a variable that approximately obeys normal distribution; second, the variable that approximately obeys normal distribution and the optimal time delay are input into the discriminant function to adaptively group the wavelet packet coefficient sequence of the first layer and the first frequency band. The discriminant function constructs a set of rigorous double-verification logic to avoid misjudging the occasional higher correlation or the correlation at the physically unreasonable time delay as effective association. Specifically, the discriminant function is expressed as:

[0050] ,

[0051] wherein is the variable that approximately obeys normal distribution, is the value of the standard normal distribution at the quantile, is the statistical significance level, is the optimal time delay, a physically reasonable time delay window. It is worth mentioning that when the discriminant function outputs 1, it means that the frequency band has a significant and physically reasonable correlation with the temperature effect, and thus it is classified into the temperature effect related frequency band index set; on the contrary, when the output is 0, it is considered that the frequency band is more likely to represent a fault signal or random noise, and it is classified into the fault feature related frequency band index set. Through this double constraint, the accuracy and robustness of the frequency band grouping are ensured.

[0052] In this way, the basis for classifying a frequency band as a temperature effect must meet both statistical significance and physical process rationality. Thus, through this double constraint, the accuracy and robustness of the frequency band grouping are greatly improved, effectively identifying and classifying those frequency bands with real physical correlation with temperature fluctuations into the temperature effect related frequency band index set, and classifying the remaining frequency bands representing fault signals or random noise into the fault feature related frequency band index set.

[0053] It should be understood that the above preferred mechanism realizes high-fidelity decoupling of online monitoring DGA signals. By introducing time delay perception ability and double verification logic, it can penetrate the time delay fog caused by transformer system inertia, accurately separate the temperature effect component mixed in the original measurement data caused by load fluctuation from the feature signal component that truly reflects the evolution trend of latent faults. In this way, an unprecedented "clean" input is provided for subsequent fault diagnosis models. By using the decoupled pure fault feature component for analysis, the problem of DGA diagnosis misjudgment caused by the dramatic changes in wind farm working conditions can be fundamentally solved, effectively avoiding the situation of gas concentration fluctuation caused by temperature effect being misjudged as a fault, and preventing the real fault signal from being missed due to being masked by temperature background noise, thereby significantly improving the sensitivity of wind power substation main transformer state monitoring and the reliability of diagnosis results, providing a solid technical support for realizing precise predictive maintenance and ensuring the safe and stable operation of power grid assets.

[0054] In particular, the S4 reconstructs and decouples the wavelet packet coefficient sequences of all frequency bands based on the temperature effect related frequency band index set and the fault feature related frequency band index set to obtain a reconstructed pure temperature effect component and a reconstructed pure fault feature component. That is, through signal reconstruction and decoupling, the temperature effect component mixed in the original measurement data caused by load fluctuation is accurately separated from the feature signal component that truly reflects the evolution trend of the latent fault. It is worth mentioning that signal reconstruction and decoupling refers to the process of respectively synthesizing the corresponding wavelet packet coefficient sequences into the original time domain signal by using these grouped frequency band indexes, thereby separating the original mixed signal into independent and pure components. The reconstructed pure temperature effect component obtained finally refers to the time domain signal obtained by inversely wavelet packet transforming all signal components belonging to the temperature effect related frequency band, which represents the change in dissolved gas concentration caused by temperature fluctuation; the reconstructed pure fault feature component refers to the time domain signal obtained by inversely wavelet packet transforming all signal components belonging to the fault feature related frequency band, which removes the interference of temperature effect and more directly reflects the real evolution of the internal potential fault of the transformer.

[0055] In specific implementation, first, the wavelet packet coefficient sequences of all frequency bands identified as being related to temperature effect are screened from the wavelet packet coefficient sequences of all frequency bands. The wavelet packet coefficient sequences corresponding to the frequency bands not belonging to the set are set to zero or directly excluded; then, the constructed pure temperature effect wavelet packet coefficient set is taken as input to perform inverse wavelet packet transformation. The inverse wavelet packet transformation is the inverse process of wavelet packet transformation, which gradually combines the decomposed frequency band coefficient sequences by iteratively applying a pair of synthesis filters (low-pass synthesis filter and high-pass synthesis filter), and finally reconstructs the original time domain signal. Here, the reconstructed pure temperature effect component can be obtained by inversely transforming the pure temperature effect wavelet packet coefficient set. The component is a time domain signal that accurately represents the feature gas concentration fluctuation caused by temperature change of the transformer;

[0056] Similarly, the wavelet packet coefficient sequences of all frequency bands identified as being related to fault feature are screened from the wavelet packet coefficient sequences of all frequency bands. The wavelet packet coefficient sequences corresponding to the frequency bands not belonging to the set are also set to zero or excluded;

[0057] Finally, the constructed pure fault feature wavelet packet coefficient set is taken as input to perform inverse wavelet packet transformation. The reconstructed pure fault feature component can be obtained by inversely transforming the pure fault feature wavelet packet coefficient set. The component is also a time domain signal that maximally eliminates the interference of temperature effect, so that the feature gas concentration change caused by fault is clearly presented.

[0058] Through the above fine signal reconstruction and decoupling process, the temperature effect component caused by load fluctuation mixed in the original measurement data can be accurately separated from the characteristic signal component that truly reflects the evolution trend of the latent fault, thereby providing a clean input for the subsequent fault diagnosis model, and significantly improving the sensitivity of the main transformer state monitoring of the wind power substation and the reliability of the diagnosis result.

[0059] In particular, the S5 performs fault diagnosis based on the reconstructed pure fault characteristic component to obtain a fault type diagnosis result. It should be understood that the reconstructed pure fault characteristic component has excluded the non-fault gas concentration fluctuation caused by the transformer load fluctuation and the environmental temperature change to the greatest extent, more directly and clearly reflects the gas change pattern generated by the abnormal condition inside the transformer, thereby fundamentally solving the problem of misjudgment of non-fault temperature effect fluctuation as a fault and the problem of missing report of the real fault signal being masked by noise, significantly improving the sensitivity of the main transformer state monitoring of the wind power substation and the reliability of the diagnosis result, and providing solid technical support for realizing precise predictive maintenance and ensuring the safe and stable operation of the power grid assets. Therefore, in the technical solution of the present application, the fault diagnosis based on the reconstructed pure fault characteristic component ensures the authenticity and effectiveness of the diagnosis result, avoids misjudgment and missing judgment, and has decisive significance for ensuring the safety of the power grid assets and optimizing the operation and maintenance strategy.

[0060] Among them, the fault diagnosis refers to analyzing and discriminating the reconstructed pure fault characteristic component to determine whether the transformer has a fault and the specific type of the fault (such as partial discharge, overheat fault, arc discharge, etc.). The final fault type diagnosis result refers to the specific category judgment about the current or potential fault of the transformer after diagnosis.

[0061] In specific implementation, the time sequence characteristics of the reconstructed pure fault characteristic component or the extracted key parameters (such as gas production rate, ratio, etc.) can be taken as inputs and sent into a rule-based expert system, a fuzzy logic system, an artificial neural network (ANN), a support vector machine (SVM), or a deep learning model (such as a long short-term memory network LSTM, a convolutional neural network CNN, etc.). These models can establish a complex nonlinear mapping relationship by learning a large amount of historical fault data and corresponding characteristic gas patterns, thereby intelligently identifying the current input component and outputting an accurate fault type diagnosis result.

[0062] Specifically, first, a series of key features that can represent the fault type are extracted from the reconstructed pure fault feature components for effective utilization of these information for fault classification. These features can include but not limited to: the average concentration, maximum concentration, gas generation rate (i.e. the rate of change of concentration over time), and feature ratios between different gases, etc. These ratios are widely used in fault discrimination in Duval triangle method or Rogers ratio method.

[0063] Subsequently, these key features extracted from the reconstructed pure fault feature components are grouped into a feature vector as the input of the trained artificial intelligence based fault diagnosis model. For example, a multi-layer feedforward artificial neural network (ANN) can be employed. The ANN has been trained in the offline phase with a large amount of historical fault data (including known fault types and corresponding feature gas patterns). During the training process, the model learns and establishes a complex nonlinear mapping relationship from these gas features to specific fault types (such as partial discharge, low temperature overheating, high temperature overheating, arc discharge, etc.). When a new feature vector (from the current transformer reconstructed pure fault feature components) is input into this trained ANN, the ANN will infer according to its learned knowledge and output a probability distribution or direct classification result about different fault types.

[0064] In summary, the wind power substation main transformer state monitoring method based on digital twinning according to the embodiments of the present application is illustrated, which first acquires the feature gas concentration and top layer oil temperature time series, then performs multi-layer wavelet packet transform on the gas concentration sequence and decomposes it into different frequency bands; then, based on the main transformer top layer oil temperature time series, correlation analysis is used to identify and group these frequency bands, so as to accurately decouple the components related to temperature effect and the components that truly represent fault features, and finally realize signal reconstruction and extract pure fault features. In this way, the accuracy and reliability of the wind power substation main transformer state monitoring are significantly improved, so as to realize fine perception and early warning of the equipment operating condition.

[0065] Further, a wind power substation main transformer state monitoring system based on digital twinning is also provided.

[0066] Figure 3 A block diagram of the wind power substation main transformer state monitoring system based on digital twinning according to the embodiments of the present application is shown. As shown in the figure, Figure 3As shown, the wind power substation main transformer state monitoring system 300 based on digital twinning according to the embodiment of the application comprises: a data acquisition module 310, configured to acquire an original characteristic gas concentration time sequence and a main transformer top layer oil temperature time sequence; a multi-layer wavelet packet transform module 320, configured to perform multi-layer wavelet packet transform on the original characteristic gas concentration time sequence to obtain a wavelet packet coefficient sequence of all frequency bands; a frequency band identification and grouping module 330, configured to perform correlation-based frequency band identification and grouping on the wavelet packet coefficient sequence of all frequency bands based on the main transformer top layer oil temperature time sequence to obtain a temperature effect related frequency band index set and a fault feature related frequency band index set; a signal reconstruction and decoupling module 340, configured to perform signal reconstruction and decoupling on the wavelet packet coefficient sequence of all frequency bands based on the temperature effect related frequency band index set and the fault feature related frequency band index set to obtain a reconstructed pure temperature effect component and a reconstructed pure fault feature component; and a fault diagnosis module 350, configured to perform fault diagnosis based on the reconstructed pure fault feature component to obtain a fault type diagnosis result.

[0067] As described above, the wind power substation main transformer state monitoring system 300 based on digital twinning according to the embodiment of the application can be implemented in various wireless terminals, such as a server with a wind power substation main transformer state monitoring algorithm based on digital twinning, etc. In a possible implementation manner, the wind power substation main transformer state monitoring system 300 based on digital twinning according to the embodiment of the application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the wind power substation main transformer state monitoring system 300 based on digital twinning can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the wind power substation main transformer state monitoring system 300 based on digital twinning can also be one of the many hardware modules of the wireless terminal.

[0068] Alternatively, in another example, the wind power substation main transformer state monitoring system 300 based on digital twinning and the wireless terminal can also be separate devices, and the wind power substation main transformer state monitoring system 300 based on digital twinning can be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in an agreed data format.

[0069] Having described above several embodiments of the disclosure, any modifications and variations that fall within the scope of the described embodiments are also intended to be within the scope of the disclosure. As will be apparent to those skilled in the art, some modifications and variations to the embodiments described above can be practiced while staying within the scope and spirit of the described embodiments. The foregoing description of the described embodiments has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the described embodiments to the precise form disclosed. Many modifications and variations are possible in light of the above teachings. It is intended that the disclosed embodiments be limited only by the claims.

Claims

1. A wind power substation main transformer state monitoring method based on digital twinning, characterized in that, The method comprises the following steps: obtaining an original characteristic gas concentration time sequence and a main transformer top layer oil temperature time sequence; performing multi-layer wavelet packet transformation on the original characteristic gas concentration time sequence to obtain a wavelet packet coefficient sequence of all frequency bands; based on the main transformer top layer oil temperature time sequence, performing correlation-based frequency band identification and grouping on the wavelet packet coefficient sequence of all frequency bands to obtain a temperature effect related frequency band index set and a fault characteristic related frequency band index set, which comprises the following steps: extracting the wavelet packet coefficient sequence of the i-th frequency band from the wavelet packet coefficient sequence of all frequency bands layer i-th frequency band wavelet packet coefficient sequence The computing the first The layer of the The maximum cross-correlation coefficient and the optimal time lag between the wavelet packet coefficient sequence of the first frequency band and the time sequence of the top oil temperature of the main transformer; Based on the maximum cross-correlation coefficient and the optimal time lag, the wavelet packet coefficient sequence of the first layer and the first frequency band is adaptively grouped to determine whether it belongs to the temperature effect related frequency band index set or the fault feature related frequency band index set, comprising: transforming the maximum cross-correlation coefficient into a variable approximately subject to a normal distribution; inputting the variable approximately subject to the normal distribution and the optimal time lag into a discriminant function to adaptively group the wavelet packet coefficient sequence of the first layer and the first frequency band; based on the temperature effect related frequency band index set and the fault characteristic related frequency band index set, performing signal reconstruction and decoupling on the wavelet packet coefficient sequence of all frequency bands to obtain a reconstructed pure temperature effect component and a reconstructed pure fault characteristic component; performing fault diagnosis based on the reconstructed pure fault characteristic component to obtain a fault type diagnosis result.

2. The wind power substation main transformer condition monitoring method based on digital twinning according to claim 1, characterized in that, The method comprises the following steps: performing multi-layer wavelet packet transformation on the original characteristic gas concentration time sequence to obtain a wavelet packet coefficient sequence of all frequency bands, which comprises the following steps: determining a wavelet basis function and a target decomposition layer number; 3. The wind power substation main transformer condition monitoring method based on digital twinning of claim 1, characterized in that, The computing the The layer of the The maximum cross-correlation coefficient and the optimal time lag between the wavelet packet coefficient sequence of the first frequency band and the time sequence of the top oil temperature of the main transformer, comprising: The computing the layer of the wavelet packet coefficient sequence and the main transformer top oil temperature time sequence between the cross-correlation function; based on the wavelet basis function and the target decomposition layer number, performing decomposition iteration on the original characteristic gas concentration time sequence to obtain the wavelet packet coefficient sequence of all frequency bands.

4. The wind power substation main transformer condition monitoring method based on digital twinning of claim 3, characterized in that, extracting a maximum cross-correlation coefficient and an optimal time lag from a cross-correlation function. Based on the first layer of the band wavelet packet coefficient sequence and the standard deviation of the main transformer top oil temperature time sequence, the cross-correlation function is transformed to obtain the cross-correlation coefficient function; The method comprises the following steps:

5. The wind power substation main transformer condition monitoring method based on digital twinning of claim 4, characterized in that, Based on the Layer The standard deviations of the wavelet packet coefficient sequences of each frequency band and the top oil temperature time series of the main transformer are used to transform the cross-correlation function to obtain the cross-correlation coefficient function, including: based on the first frequency band... Layer The standard deviations of the wavelet packet coefficient sequences of each frequency band and the time series of the top oil temperature of the main transformer are used to transform the cross-correlation function to obtain the cross-correlation coefficient function using the following formula: ， wherein, is the cross-correlation function, and are the first and the second wavelet packet coefficient sequence of the first and the standard deviation of the top oil temperature time series of the main transformer, respectively.

6. The wind power substation main transformer condition monitoring method based on digital twinning of claim 1, characterized in that, extracting a maximum cross-correlation coefficient and an optimal time lag from a cross-correlation function, which comprises the following steps: ， wherein, is a variable approximately following a normal distribution, is a value of a standard normal distribution at is a quantile, is a statistical significance level, is an optimal time lag, is a physically reasonable time lag window, is a sequence length.

7. A digital-twin-based wind power substation main transformer condition monitoring system, characterized in that, finding a maximum correlation coefficient and a corresponding time lag from a cross-correlation coefficient function to obtain the maximum cross-correlation coefficient and the optimal time lag. The discriminant function is expressed as: The method comprises the following steps: a data acquisition module for acquiring an original characteristic gas concentration time sequence and a main transformer top layer oil temperature time sequence; extracting the wavelet packet coefficient sequence of the i-th frequency band from the wavelet packet coefficient sequence of all frequency bands layer i-th frequency band wavelet packet coefficient sequence The computing the first The layer of the The maximum cross-correlation coefficient and the optimal time lag between the wavelet packet coefficient sequence of the first frequency band and the time sequence of the top oil temperature of the main transformer; Based on the maximum cross-correlation coefficient and the optimal time lag, the wavelet packet coefficient sequence of the first layer and the first frequency band is adaptively grouped to determine whether it belongs to the temperature effect related frequency band index set or the fault feature related frequency band index set, comprising: transforming the maximum cross-correlation coefficient into a variable approximately subject to a normal distribution; inputting the variable approximately subject to the normal distribution and the optimal time lag into a discriminant function to adaptively group the wavelet packet coefficient sequence of the first layer and the first frequency band; a multi-layer wavelet packet transformation module for performing multi-layer wavelet packet transformation on the original characteristic gas concentration time sequence to obtain a wavelet packet coefficient sequence of all frequency bands; a frequency band identification and grouping module for, based on the main transformer top layer oil temperature time sequence, performing correlation-based frequency band identification and grouping on the wavelet packet coefficient sequence of all frequency bands to obtain a temperature effect related frequency band index set and a fault characteristic related frequency band index set, which comprises the following steps: a signal reconstruction and decoupling module for, based on the temperature effect related frequency band index set and the fault characteristic related frequency band index set, performing signal reconstruction and decoupling on the wavelet packet coefficient sequence of all frequency bands to obtain a reconstructed pure temperature effect component and a reconstructed pure fault characteristic component; a fault diagnosis module for performing fault diagnosis based on the reconstructed pure fault characteristic component to obtain a fault type diagnosis result.

Images