Hydro-generator partial discharge pulse extraction method based on Taylor-S transformation

Abstract

The invention relates to a hydro-generator partial discharge pulse extraction method based on Taylor-S transformation. The method comprises the following steps: collecting a partial discharge signal of a stator winding of a hydro-generator; preprocessing the partial discharge signal to obtain a preprocessed signal; performing time-frequency transformation on the preprocessed signal to obtain a time-frequency matrix; estimating an instantaneous frequency based on phase information of the time-frequency matrix; synchronously extracting the time-frequency matrix according to the instantaneous frequency to obtain a time-frequency energy matrix; extracting a time-frequency ridge line based on the time-frequency energy matrix, and performing feature optimization on the time-frequency ridge line to obtain an optimized ridge line; and determining starting and ending positions of the partial discharge pulse according to the time-frequency energy matrix and the optimized ridge line. According to the method, Taylor-S transformation is combined with synchronous extraction operation, so that the problems of energy dispersion and serious cross term interference in a strong noise environment of a traditional time-frequency analysis method are effectively solved, and the definition and focusing property of partial discharge pulse time-frequency representation are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of generator partial discharge detection, specifically relating to a method for extracting partial discharge pulses from hydro-generators based on Taylor-S transform. Background Technology

[0002] As a key piece of equipment in the power system, reliable monitoring of the stator winding insulation condition of hydro-generators is crucial for ensuring the safe and stable operation of the unit and preventing sudden insulation faults. Partial discharge (PD) is closely related to the degradation process of insulation materials; it is both a major cause of insulation aging and an important characteristic signal reflecting its condition. Therefore, online monitoring of partial discharge pulses can promptly obtain information on insulation defects, making it an effective means of early fault diagnosis and predictive maintenance.

[0003] Despite extensive research and numerous application achievements in generator partial discharge monitoring both domestically and internationally, this technology still faces significant challenges in practical engineering applications. On the one hand, existing monitoring systems often suffer from incomplete monitoring; on the other hand, accurate extraction of partial discharge pulses remains difficult. This is primarily due to the complex on-site monitoring environment, where the acquired signals inevitably contain strong background noise. Noise masks the rising and falling edges of the partial discharge pulse, leading to blurred pulse waveform characteristics and significantly increasing the difficulty of subsequent signal identification and characteristic analysis. Therefore, accurately locating and extracting the start and end points of the partial discharge pulse from a strong noise background has become a key bottleneck problem in hydroelectric generator partial discharge detection technology.

[0004] At the signal processing level, researchers have attempted various methods to denoise and extract partial discharge signals, such as wavelet thresholding denoising and empirical mode decomposition denoising. However, these methods each have limitations. For example, the performance of denoising methods based on complex wavelet transform largely depends on the selection of the complex wavelet, the determination of the wavelet decomposition order, and the selection of the threshold algorithm; these parameters are difficult to optimize simultaneously. Another study proposed first using spectral kurtosis to design an adaptive bandpass Wiener filter for filtering, followed by wavelet smoothing denoising. However, this method has a drawback: the adaptive bandpass Wiener filter itself has fixed parameters, and its optimal filtering effect depends on prior knowledge of the statistical characteristics of the signal and noise, which is usually unknowable in advance in complex real-world environments.

[0005] In summary, existing technologies for processing partial discharge signals from hydro-generators generally suffer from limited noise immunity, poor adaptability to non-stationary signals, dispersed energy in time-frequency analysis, and reliance on difficult-to-obtain prior knowledge. These issues lead to inaccurate localization of the start and end points of partial discharge pulses, hindering further improvements in condition monitoring and diagnosis. Therefore, there is an urgent need to develop a new method that can robustly and adaptively achieve accurate extraction of partial discharge pulses in high-noise environments. Summary of the Invention

[0006] This invention provides a method for extracting partial discharge pulses from hydro-generators based on Taylor-S transform, aiming to solve the technical problem of accurately locating the start and end points of partial discharge pulses from hydro-generators under strong noise conditions.

[0007] To achieve the above objectives, the first aspect of the present invention provides a method for extracting partial discharge pulses from a hydro-generator based on Taylor-S transform, comprising the following steps: Acquire partial discharge signals from the stator windings of the hydro-generator; The partial discharge signal is preprocessed to obtain a preprocessed signal; The preprocessed signal is subjected to time-frequency transformation to obtain the time-frequency matrix; The instantaneous frequency is estimated based on the phase information of the time-frequency matrix; The time-frequency matrix is ​​synchronously extracted based on the instantaneous frequency to obtain the time-frequency energy matrix; Based on the time-frequency energy matrix, time-frequency ridges are extracted, and the time-frequency ridges are feature-optimized to obtain optimized ridges; The start and end positions of the partial discharge pulse are determined based on the time-frequency energy matrix and the optimized ridge line.

[0008] Furthermore, the preprocessing includes using a high-pass filter to remove background noise and normalizing the denoised signal.

[0009] Furthermore, methods for performing time-frequency transformation on the preprocessed signal to obtain the time-frequency matrix include: The S-transform of the preprocessed signal is calculated to obtain a two-dimensional time-frequency matrix, where the formula for calculating the S-transform is:

[0010] in, Represents a two-dimensional time-frequency matrix. This represents the time center parameter, which is changed by... To achieve localized analysis of signals at different points in time, The frequency represents the frequency of the current analysis. This indicates a preprocessed partial discharge signal. ∈(-∞, ∞) represents the global time variable. This represents the Gaussian window function width adjustment factor. Represents the imaginary unit. It represents the base of the natural logarithm.

[0011] Furthermore, the method for estimating the instantaneous frequency based on the phase information of the time-frequency matrix includes: The preprocessed signal is represented in analytical form:

[0012] in, This represents the preprocessed partial discharge signal. Indicates amplitude. Represents the signal phase function; Performing a second-order Taylor expansion on the signal phase function yields:

[0013] in, Indicates time Phase value at that point, Indicates the reference time point for Taylor expansion. Indicates time Phase value at that point, Indicates time The first derivative of the phase at that point, Indicates time Phase second derivative at that point, This represents the time offset relative to a reference time point; Calculate the phase function of the time-frequency matrix; Taking the first derivative of the phase function of the time-frequency matrix, we obtain the first-order instantaneous frequency:

[0014] in, The first-order instantaneous frequency, Indicates time The partial derivatives, Representing the time-frequency matrix The phase function; Taking the second derivative of the phase function of the time-frequency matrix, we obtain the second derivative: Based on the first-order instantaneous frequency and the second-order derivative, calculate the second-order corrected instantaneous frequency:

[0015] in, For the second-order corrected instantaneous frequency, Indicates time The second-order partial derivative, This indicates a time delay estimate.

[0016] Furthermore, the method for synchronously extracting the time-frequency matrix based on the instantaneous frequency to obtain the time-frequency energy matrix includes: The second-order corrected instantaneous frequency is used as the estimate of the true instantaneous frequency of the signal; The energy of the time-frequency matrix is ​​redistributed to the actual instantaneous frequency by a synchronous extraction operator, resulting in the time-frequency energy matrix, which is calculated using the following formula:

[0017] in, This represents the time-frequency energy matrix after synchronous extraction. This represents the Dirichlet function, also known as the kernel function. This represents the estimated angular frequency.

[0018] Furthermore, the method for extracting time-frequency ridges based on the time-frequency energy matrix and optimizing the features of the time-frequency ridges to obtain optimized ridges includes: Extract from the time-frequency energy matrix A time-frequency ridge line; Calculate the multi-scale scattering entropy corresponding to each time-frequency ridge; The multi-scale dispersion entropy is arranged in ascending order according to the center frequency of the corresponding time-frequency ridge; The maximum correlation and minimum redundancy criterion is used to optimize the eigenvalues ​​of the multi-scale scatter entropy after arrangement, resulting in the optimized time-frequency ridge.

[0019] Furthermore, the method for determining the start and end positions of the partial discharge pulse based on the time-frequency energy matrix and the optimized ridge line includes: Calculate the energy value of the time-frequency energy matrix:

[0020] in, Indicates energy value; Based on the energy value, the noise floor is calculated:

[0021] in, Indicates the noise floor. This represents the average energy value within one power frequency cycle. Based on the optimized time-frequency ridge, calculate the frequency jump threshold:

[0022] in, Indicates the frequency jump threshold. This represents the highest frequency derived from the optimized time-frequency ridge. The pulse start and end positions are determined based on the energy value and the noise floor.

[0023] Furthermore, the method for determining the pulse start and end positions based on the energy value and the noise floor includes: Scan the energy value Find satisfaction: and time point ,Will Determined as the pulse start point location; Scan the energy value Find satisfaction: and time point ,Will The location was determined to be the end point of the pulse.

[0024] Furthermore, the method also includes verifying the pulse interval, the verification method comprising: The discharge pulse pre-selection interval is determined based on the pulse start position and the pulse end position; Calculate the absolute value of the frequency difference at each time point within the pre-selected interval of the discharge pulse:

[0025] in, The absolute value of the frequency difference. This refers to the optimized ridge line; If the preselected interval of the discharge pulse satisfies If the number of time points exceeds 10% of the total number of time points in the interval, the pre-selected discharge pulse interval is determined to be a noise interval and is removed; otherwise, it is retained as a valid pulse interval.

[0026] To achieve the above objectives, a second aspect of the present invention provides a partial discharge pulse extraction system for hydro-generators based on Taylor-S transform, comprising the following modules: The signal acquisition module is used to acquire partial discharge signals from the stator windings of the hydro-generator. The signal preprocessing module is used to preprocess the partial discharge signal to obtain the preprocessed signal; The time-frequency transformation module is used to perform time-frequency transformation on the preprocessed signal to obtain a time-frequency matrix. An instantaneous frequency estimation module is used to estimate the instantaneous frequency based on the phase information of the time-frequency matrix; The synchronous extraction module is used to synchronously extract the time-frequency matrix according to the instantaneous frequency to obtain the time-frequency energy matrix; The ridge processing module is used to extract time-frequency ridges based on the time-frequency energy matrix; The feature optimization module is used to perform feature optimization on the time-frequency ridge to obtain an optimized ridge; The pulse positioning module is used to determine the start and end positions of the partial discharge pulse based on the time-frequency energy matrix and the optimized ridge line.

[0027] The beneficial effects of this invention are: Compared with existing technologies, this invention provides a method for extracting partial discharge pulses from hydro-generators based on Taylor-S-transform. It obtains a time-frequency matrix reflecting the non-stationary characteristics of the signal by performing an S-transform on the preprocessed signal. Subsequently, an instantaneous frequency estimation method based on the second-order Taylor expansion of the phase function is introduced. By differentiating the phase of the time-frequency matrix and performing a second-order correction, the accuracy and robustness of frequency estimation under noise interference are significantly improved. Based on this, a synchronous extraction operator is used to redistribute and focus the dispersed time-frequency energy according to the accurately estimated instantaneous frequency, generating a highly concentrated time-frequency energy matrix. Furthermore, this method extracts time-frequency ridges representing the core components of the signal from the focused time-frequency energy matrix and introduces multi-scale scatter entropy for quantitative analysis, effectively distinguishing deterministic discharge pulses from random noise at the "regularity" level. Finally, the ridge features are optimized using the maximum correlation and minimum redundancy criterion, eliminating redundant information to obtain the optimized ridges that best represent the essential characteristics of the discharge pulse. Finally, by combining the energy information provided by the time-frequency energy matrix and the frequency information provided by the optimized ridge, and by setting an adaptive frequency jump threshold and noise floor, the start and end boundaries of the pulse energy are accurately determined. Furthermore, the effectiveness and accuracy of the extracted pulse interval are ensured through posterior verification of frequency stability. The entire methodology forms a complete technical closed loop from signal representation and feature extraction to decision-making, achieving adaptive and high-precision localization of the start and end points of partial discharge pulses under strong noise backgrounds without relying on prior statistical knowledge of the signal and noise. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0029] Figure 1 This is a flowchart of a method for extracting partial discharge pulses from a hydro-generator based on Taylor-S transform, as disclosed in an embodiment of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] According to embodiments of the present invention, it should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the following manufacturing method, in some cases the steps shown or described may be performed in a different order than that shown here.

[0032] like Figure 1 As shown, this invention provides a method for extracting partial discharge pulses from a hydro-generator based on Taylor-S transform, comprising the following steps: Step S100: Acquire the partial discharge signal of the stator winding of the hydro-generator; This step uses a capacitively coupled sensor installed at the neutral point of the hydro generator to sense and capture the pulse current signal generated by the partial discharge activity inside the stator winding caused by insulation defects.

[0033] Step S200: Preprocess the partial discharge signal to obtain a preprocessed signal; The acquired raw partial discharge signals are preprocessed to improve signal quality; the main purpose of preprocessing is to suppress background noise interference and standardize signal amplitude.

[0034] The preprocessing operation includes using a high-pass filter to remove low-frequency background noise from the signal; this process can effectively retain high-frequency partial discharge pulse components while filtering out low-frequency noise such as power frequency interference and its harmonics.

[0035] After denoising, the signal is normalized. This operation adjusts the signal amplitude to a uniform scaling range to eliminate amplitude differences caused by factors such as sensor gain and transmission attenuation, ensuring the stability of subsequent processing. The preprocessed partial discharge signal is denoted as... .

[0036] Step S300: Perform time-frequency transformation on the preprocessed signal to obtain a time-frequency matrix; This step involves preprocessing the partial discharge signal. The time-frequency transformation aims to convert a one-dimensional time signal into a two-dimensional representation that can simultaneously reflect the time and frequency domain characteristics of the signal. This transformation is used to reveal the energy distribution characteristics of partial discharge pulses in the time-frequency domain.

[0037] The time-frequency transformation method used is the S-transform. The formula for calculating the S-transform is:

[0038] in, Represents a two-dimensional time-frequency matrix. This represents the time center parameter, which is changed by... To achieve localized analysis of signals at different points in time, The frequency represents the frequency of the current analysis. This indicates a preprocessed partial discharge signal. ∈(-∞, ∞) represents the global time variable. This represents the Gaussian window function width adjustment factor. Represents the imaginary unit. It represents the base of the natural logarithm.

[0039] Through this calculation, the signal Mapped to a two-dimensional complex matrix This time-frequency matrix contains both the amplitude and phase information of the signal.

[0040] Step S400: Estimate the instantaneous frequency based on the phase information of the time-frequency matrix; This step improves estimation accuracy by utilizing phase information and applying Taylor expansion for correction. The specific method is as follows: First, the preprocessed signal This can be represented in analytical form:

[0041] in, This represents the preprocessed partial discharge signal. Indicates amplitude. Represents the signal phase function; To approximate the phase locally, the phase function is... At reference time point Perform a second-order Taylor expansion at this point:

[0042] in, Indicates time Phase value at that point, Indicates the reference time point for Taylor expansion. Indicates time Phase value at that point, Indicates time The first derivative of the phase at that point, Indicates time Phase second derivative at that point, This represents the time offset relative to a reference time point.

[0043] Two-dimensional time-frequency matrix obtained based on S-transform The first-order instantaneous frequency estimate is calculated. This estimate is obtained by applying the phase function of the time-frequency matrix. Regarding time The first-order partial derivative is obtained, and the calculation formula is:

[0044] in, The first-order instantaneous frequency, Indicates time The partial derivatives, Representing the time-frequency matrix The phase function.

[0045] To further improve the accuracy of frequency estimation under non-stationary signal conditions, a second-order phase derivative term is introduced for correction, and the instantaneous frequency after second-order correction is calculated. The calculation formula is as follows:

[0046] in, For the second-order corrected instantaneous frequency, Indicates time The second-order partial derivative, This represents the time delay estimate. This correction term effectively compensates for the deviation introduced by the change in signal frequency over time. The final result is... Will be used as a signal at time and frequency points A more accurate estimate of the true instantaneous frequency.

[0047] Step S500: Synchronously extract the time-frequency matrix according to the instantaneous frequency to obtain the time-frequency energy matrix; In this step, the energy in the time-frequency matrix is ​​redistributed and focused using the precise instantaneous frequency estimated in the previous step, in order to generate a time-frequency representation with more concentrated energy and higher time-frequency resolution.

[0048] The synchronous extraction transform is implemented through a synchronous extraction operator. The basic principle of this operator is: in the traditional time-frequency distribution, each time-frequency point... The energy may be dispersed in the surrounding area. This is determined by judging whether the estimated instantaneous frequency at that time-frequency point matches the current analysis frequency. Consistency allows the dispersed energy to be refocused onto the true instantaneous frequency trajectory of the signal.

[0049] The specific calculations for the synchronous extraction transformation are performed using the following formula:

[0050] in, This represents the time-frequency energy matrix after synchronous extraction, which mainly contains the redistributed signal energy information. This represents the Dirichlet function, also known as the kernel function. This represents the estimated angular frequency.

[0051] This operation effectively removes the energy from regions on the time-frequency plane where the instantaneous frequency does not match the analysis frequency, while the energy of the main signal components is concentrated on their time-frequency ridges, resulting in a clean and well-defined time-frequency energy matrix. .

[0052] Step S600: Extract time-frequency ridges based on the time-frequency energy matrix, and perform feature optimization on the time-frequency ridges to obtain optimized ridges; The purpose of this step is to extract time-frequency ridges representing the main components of the signal from the synchronously extracted time-frequency energy matrix, and to select the optimized ridges that best represent the partial discharge pulse using feature optimization methods. The specific methods include: Step S601: From the time-frequency energy matrix Extract Time-frequency ridge line ,in .

[0053] Step S602: Calculate the multi-scale scattering entropy corresponding to each time-frequency ridge. ,in Multiscale scatter entropy is an indicator used to quantify the complexity and regularity of time series, effectively distinguishing deterministic signals from random noise. By calculating the multiscale scatter entropy of time-frequency ridges, the regularity characteristics of the signal components represented by each ridge can be quantified.

[0054] Step S603: Arrange the calculated multi-scale scattering entropy in ascending order according to the center frequency of the corresponding time-frequency ridge.

[0055] Step S604: The multi-scale scatter entropy features after arrangement are optimized using the maximum correlation and minimum redundancy criterion. This criterion selects the most representative feature subset with the lowest information redundancy by evaluating the correlation between features and the target, as well as the redundancy between features. Through this optimization process, the final optimized time-frequency ridge is obtained. .

[0056] Step S700: Determine the start and end positions of the partial discharge pulse based on the time-frequency energy matrix and the optimized ridge line.

[0057] Based on the time-frequency energy matrix and optimized ridge lines obtained in the preceding steps, the start and end positions of the partial discharge pulse are accurately determined by setting adaptive thresholds and logical judgments. The specific method includes: First, calculate the energy value corresponding to the time-frequency energy matrix. The calculation formula is as follows:

[0058] in, Indicates energy value; Meanwhile, based on the final optimized time-frequency ridge... Determine the highest frequency of the signal .

[0059] Set two thresholds: Frequency jump threshold: ,in, Indicates the frequency jump threshold. This represents the highest frequency derived from the optimized time-frequency ridge. Noise substrate: ,in, Indicates the noise floor. This represents the average energy value within one power frequency cycle. The setting of the noise floor allows for effective differentiation between signal energy and background noise levels.

[0060] The specific methods for determining the pulse start and end positions based on the energy value and noise floor mentioned above include: Scan the energy value Find satisfaction: and time point ,Will Determined as the pulse start point location; Scan the energy value Find satisfaction: and time point ,Will The location was determined to be the end point of the pulse.

[0061] The pre-selected time interval of the discharge pulse can be determined by the pulse start position and pulse end position. .

[0062] Finally, the pre-selected pulse interval was verified by calculating the absolute value of the frequency difference at each time point within the interval:

[0063] in, The absolute value of the frequency difference. This refers to the optimized ridge line.

[0064] If the interval satisfies If the number of time points exceeds 10% of the total number of points in the interval, the interval is determined to be a noise interference interval and is removed; otherwise, it is retained as a valid partial discharge pulse interval.

[0065] This invention effectively overcomes the problems of energy dispersion and severe cross-term interference in strong noise environments by combining Taylor-S transform with synchronous extraction operations, significantly improving the clarity and focus of the time-frequency characterization of partial discharge pulses. By introducing multi-scale scattering entropy to quantify the complexity of time-frequency ridges and combining the maximum correlation and minimum redundancy criterion for feature optimization, it achieves an essential distinction between discharge pulses and background noise, enhancing the method's anti-interference capability. Through an adaptive pulse localization mechanism based on energy threshold and frequency stability verification, it greatly improves the accuracy and reliability of partial discharge pulse start and end point localization, providing a more accurate data foundation for the insulation status assessment of hydro-generators.

[0066] Based on the aforementioned method steps, this invention also provides a partial discharge pulse extraction system for hydro-generators based on Taylor-S transform, which achieves accurate extraction of partial discharge pulses through the coordinated operation of corresponding functional modules. The specific implementation of the system is described in detail below in conjunction with the functions of each module. The system includes the following modules: The signal acquisition module is used to acquire partial discharge signals from the stator windings of the hydro-generator. The signal preprocessing module is used to preprocess the partial discharge signal to obtain the preprocessed signal; The time-frequency transformation module is used to perform time-frequency transformation on the preprocessed signal to obtain a time-frequency matrix. An instantaneous frequency estimation module is used to estimate the instantaneous frequency based on the phase information of the time-frequency matrix; The synchronous extraction module is used to synchronously extract the time-frequency matrix according to the instantaneous frequency to obtain the time-frequency energy matrix; The ridge processing module is used to extract time-frequency ridges based on the time-frequency energy matrix; The feature optimization module is used to perform feature optimization on the time-frequency ridge to obtain an optimized ridge; The pulse positioning module is used to determine the start and end positions of the partial discharge pulse based on the time-frequency energy matrix and the optimized ridge line.

[0067] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0068] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0069] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0070] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0071] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for extracting partial discharge pulses of a hydro-generator based on Taylor-S transform, characterized in that, The method comprises the following steps: Collecting partial discharge signals of a stator winding of a hydroelectric generator; Pretreating the partial discharge signals to obtain pretreated signals; Performing time-frequency transformation on the pretreated signals to obtain a time-frequency matrix; Estimating an instantaneous frequency based on phase information of the time-frequency matrix; Synchronously extracting the time-frequency matrix according to the instantaneous frequency to obtain a time-frequency energy matrix; Extracting time-frequency ridges based on the time-frequency energy matrix, and optimizing features of the time-frequency ridges to obtain optimized ridges; Determining start and end positions of a partial discharge pulse according to the time-frequency energy matrix and the optimized ridges.

2. The Taylor-S transform based partial discharge pulse extraction method for hydrogenerators as claimed in claim 1, wherein, The pretreatment comprises removing background noise by using a high-pass filter, and performing normalization processing on the noise-removed signals.

3. The Taylor-S transform based partial discharge pulse extraction method for hydrogenerators as claimed in claim 1, wherein, The method for performing time-frequency transformation on the pretreated signals to obtain a time-frequency matrix comprises: Calculating an S transform of the pretreated signals to obtain a two-dimensional time-frequency matrix, wherein a calculation formula of the S transform is: wherein denotes a two-dimensional time-frequency matrix, denotes a time center parameter, by varying a localized analysis of the signal at different time points is achieved, is the frequency, denoting the frequency of the current analysis, denotes a pre-processed partial discharge signal, ∈(-∞,∞) denotes a global time variable, denotes a Gaussian window function width adjustment factor, denotes the imaginary unit, denotes the base of the natural logarithm.

4. The Taylor-S transform based partial discharge pulse extraction method for hydrogenerators as claimed in claim 3, wherein, The method for estimating an instantaneous frequency based on phase information of the time-frequency matrix comprises: Expressing the pretreated signals in an analytical form: wherein, represents the pre-processed partial discharge signal, represents the amplitude, represents the signal phase function; Performing a second-order Taylor expansion on the signal phase function to obtain: wherein denotes a phase value at time denotes a phase value at time denotes a reference time point of the Taylor expansion, denotes a phase value at time denotes a phase value at time denotes a first derivative of the phase at time denotes a first derivative of the phase at time denotes a second derivative of the phase at time denotes a second derivative of the phase at time denotes a time offset with respect to the reference time point; Calculating a phase function of the time-frequency matrix; Performing a first-order derivation on the phase function of the time-frequency matrix to obtain a first-order instantaneous frequency: wherein is the first order instantaneous frequency, denotes the partial derivative with respect to time , and denotes the phase function of the time-frequency matrix ; Performing a second-order derivation on the phase function of the time-frequency matrix to obtain a second-order derivative: Based on the first-order instantaneous frequency and the second-order derivative, calculating a second-order modified instantaneous frequency: wherein is the second order modified instantaneous frequency, denotes the second order partial derivative with respect to time t, denotes the time delay estimate.

5. The Taylor-S transform based partial discharge pulse extraction method for hydrogenerators as claimed in claim 4, wherein, The method for synchronously extracting the time-frequency matrix according to the instantaneous frequency to obtain a time-frequency energy matrix comprises: Taking the second-order modified instantaneous frequency as a real instantaneous frequency estimation of the signal; Re-distributing energy of the time-frequency matrix to the real instantaneous frequency by using a synchronous extraction operator to obtain a time-frequency energy matrix, and a calculation formula thereof is: wherein, denotes the time-frequency energy matrix after synchronization extraction, denotes the Dirichlet function, i.e. the kernel function, denotes the angular frequency estimate.

6. The Taylor-S transform based partial discharge pulse extraction method for hydrogenerators as claimed in claim 1, wherein, The method for extracting time-frequency ridges based on the time-frequency energy matrix, and optimizing features of the time-frequency ridges to obtain optimized ridges comprises: extracting from the time-frequency energy matrix a bar time-frequency ridge; Calculating a multi-scale scatter entropy corresponding to each time-frequency ridge; Arranging the multi-scale scatter entropies in an order from small to large according to center frequencies of the corresponding time-frequency ridges; Optimizing feature values of the arranged multi-scale scatter entropies by using a maximum correlation minimum redundancy criterion to obtain optimized time-frequency ridges.

7. The Taylor-S transform based partial discharge pulse extraction method for hydrogenerators as claimed in claim 5, wherein, The method for determining start and end positions of a partial discharge pulse according to the time-frequency energy matrix and the optimized ridges comprises: Calculating energy values of the time-frequency energy matrix: wherein, represents an energy value; Based on the energy values, calculating a noise base: wherein represents a noise floor, represents an average of energy values within a power frequency cycle; Based on the optimized time-frequency ridges, calculating a frequency jump threshold value: wherein, denotes a frequency hopping threshold, denotes the highest frequency derived from the optimized time-frequency ridge; Determining a pulse start position and an end position according to the energy values and the noise base.

8. The Taylor-S transform based partial discharge pulse extraction method for hydrogenerators as claimed in claim 7, wherein, The method for determining a pulse start position and an end position according to the energy values and the noise base comprises: scanning the energy values , finding a time point that satisfies: and determining as the pulse start position scanning the energy values , finding a time point that satisfies: and , determining as the pulse end position.

9. The Taylor-S transform based partial discharge pulse extraction method for hydrogenerators as claimed in claim 8, wherein, The method further comprises verifying a pulse interval, and the verification method comprises: Determining a discharge pulse preselected interval according to the pulse start position and the pulse end position; Calculating an absolute value of a frequency difference at each time point in the discharge pulse preselected interval: wherein is the absolute value of the frequency difference, denotes the optimized spine; If the proportion of the number of time points satisfying exceeds 10% of the total number of time points in the preselected interval, the preselected interval of the discharge pulse is determined as a noise interval and is rejected, otherwise it is retained as a valid pulse interval.

10. A Taylor-S transformation based partial discharge pulse extraction system for a hydro-generator characterized by, The method comprises the following modules: A signal collection module for collecting partial discharge signals of a stator winding of a hydroelectric generator; A signal pretreatment module for pretreating the partial discharge signals to obtain pretreated signals; a time-frequency transform module configured to perform time-frequency transform on the preprocessed signal to obtain a time-frequency matrix; an instantaneous frequency estimation module configured to estimate an instantaneous frequency based on phase information of the time-frequency matrix; a synchronization extraction module configured to perform synchronization extraction on the time-frequency matrix according to the instantaneous frequency to obtain a time-frequency energy matrix; a ridge processing module configured to extract a time-frequency ridge based on the time-frequency energy matrix; a feature optimization module configured to perform feature optimization on the time-frequency ridge to obtain an optimized ridge; a pulse positioning module configured to determine start and end positions of a partial discharge pulse based on the time-frequency energy matrix and the optimized ridge.

Images