A method and system for predicting the insulation condition of a generator motor

Abstract

The present application relates to the technical field of power generation motor, and discloses a power generation motor insulation state prediction method and system. The method collects three-phase stator current signals and operation mode identification signals of the power generation motor, and constructs an insulation degradation characteristic vector. The power generation mode and the motor mode are distinguished based on window average active power, and an operation mode variable and a mode change indicator are constructed. The basic coupling coefficient between the degradation characteristics is calculated, the coupling weight is adjusted by introducing a mode sensitivity function, an adaptive adjacency matrix driven by the operation mode is constructed, and the event-driven update of the graph structure is realized. Based on the adaptive graph structure, spatial feature extraction and time dimension fusion are carried out, and a normalized insulation health index is output for state evaluation and trend analysis. The method can weaken the influence of the difference between the two working conditions on the coupling relationship of the characteristics, improve the stability and reliability of the insulation degradation evaluation, and is suitable for online monitoring and early warning of the power generation motor of the pumped storage unit.

Description

Technical Field

[0001] This invention relates to the field of generator motor technology, and in particular to a method and system for predicting the insulation status of generator motors. Background Technology

[0002] With the continuous increase in the proportion of new energy power generation and the sustained strengthening of power system regulation needs, pumped storage power stations are playing an increasingly prominent role in grid peak shaving, frequency regulation, and power balance. As the core equipment of pumped storage units, the generator motor needs to frequently switch between power generation and motoring modes, and its operating status directly affects the safety and stability of the power grid and the reliability of the equipment. Under high load regulation and frequent start-stop conditions, the electromagnetic force, thermal stress, and mechanical vibration within the motor are superimposed, causing the insulation system to be in a complex multi-stress coupling environment for a long time, making it highly susceptible to insulation aging and degradation.

[0003] The insulation system of a generator motor includes multiple parts such as stator main insulation, stator inter-turn insulation, and rotor winding insulation. During long-term operation, the insulation materials are affected by electrical stress, temperature rise changes, and mechanical vibration, which gradually lead to problems such as decreased dielectric properties, enhanced partial discharge, and reduced insulation resistance. When insulation degradation develops to a certain extent, it may cause inter-turn short circuits or grounding faults, and in severe cases, lead to unit shutdown accidents.

[0004] Currently, research on motor insulation condition prediction mainly focuses on lifetime prediction methods based on degradation models, trend analysis methods based on electrical characteristic signals, and condition assessment methods based on data-driven models. For example, by extracting electrical characteristics such as negative sequence current, even harmonics, or third harmonics, and combining them with statistical models or neural network models for insulation condition prediction, the prediction accuracy has been improved to some extent. However, most methods are mainly based on fixed feature structures or static correlations for modeling, such as:

[0005] Chinese invention patent application number CN202310507059.9, entitled "Two-Stage Motor Insulation Life Prediction Method," establishes a two-stage insulation degradation model through accelerated aging tests under different temperature and stress conditions, and combines it with a Kalman filter algorithm to predict the remaining life of motor insulation. This method has certain reference value in material degradation mechanism modeling, and its prediction is mainly based on accelerated test data and a fixed degradation model. Chinese invention patent application number CN202510423979.1, entitled "A Winding Insulation Remaining Life Prediction Method and System," constructs a piecewise mathematical model to describe the degradation process of insulation materials. The prediction result is generated by superimposing the theoretical aging component of the degradation mechanism model with the residual compensation component learned by the neural network. The number of segments is custom-defined based on the dataset. The Chinese invention patent with application number CN202511824139.2 and patent title "Method and Device for Motor Insulation Diagnosis Based on Multi-Physical Field Data Fusion" collects multi-dimensional physical quantity signals related to insulation degradation; extracts multiple feature parameters characterizing the insulation state based on the multi-dimensional physical quantity signals; and inputs the multiple feature parameters into a pre-constructed insulation health assessment model.

[0006] Therefore, in summary, existing patents essentially still rely on fixed feature results for modeling. However, the insulation degradation of generator motors is an extremely complex process. Modeling with fixed state feature data alone requires improvement in accuracy. At the very least, existing patents do not fully consider the changing coupling relationships of degradation features under different operating modes of generator motors. Specifically, they do not consider the impact of changes in electromagnetic and thermal stress distribution on the sensitivity and correlation of degradation features during the switching between generator and motor modes. Furthermore, the lack of a dynamic update mechanism for spatial correlation between degradation features, coupled with the continued use of a fixed structure for prediction when operating modes switch, may affect the stability and accuracy of the prediction results. Summary of the Invention

[0007] Technical Problem: In order to solve the technical problems existing in the prior art, the present invention provides a method for predicting the insulation state of a generator motor. This method solves the problem that the insulation state prediction results are not stable and accurate when the changes in feature sensitivity and inter-feature coupling relationship are not taken into account during the insulation degradation process of the generator motor under frequent switching under dual operating conditions. In addition, this application also provides a generator motor insulation state prediction system.

[0008] Technical Solution: According to the technical solution provided by this invention, on one hand, this invention provides a method for predicting the insulation state of a generator motor, the method comprising:

[0009] During normal operation of the generator motor, three-phase stator current signals are collected, and the negative sequence current amplitude and the normalized amplitudes of the second, third, and fourth harmonics are spliced ​​together to construct the insulation degradation feature vector at each moment. Based on the average active power within the set sliding time window and the set discrimination dead zone threshold, operation mode variables and mode confidence are constructed, and a comprehensive feature vector is obtained based on the insulation degradation feature vector and the operation mode variables.

[0010] Within a fixed statistical time window, the time series of each feature dimension in the comprehensive feature vector is constructed to obtain the basic coupling coefficient between two different features. The normalization function is used to construct the adaptive adjacency matrix of the operation mode based on the basic coupling coefficient and the operation mode sensitivity function. The operation mode sensitivity function is used to determine the difference in sensitivity of the degradation feature to the insulation state between the power generation mode and the electric mode according to the operation mode variable.

[0011] Based on the running mode variable, the mode change indicator of two adjacent windows is obtained. The mode change indicator is used to use the running mode change as a trigger condition for dynamic updating of the graph structure. The nodes in the graph structure correspond one-to-one with the degenerate feature dimension in the comprehensive feature vector.

[0012] The adaptive adjacency matrix of the mode is adaptively updated according to the value of the mode change indicator, and the nodes in the graph structure are used as graph signal input. Graph convolution is used to aggregate the spatial association information between nodes to obtain spatial feature representation. The spatial feature representation is subjected to time dimension fusion processing to characterize the evolution trend of insulation degradation state and output insulation health index. By continuously monitoring the insulation health index, the degree of insulation degradation can be quantitatively assessed and trend analyzed.

[0013] Furthermore, including:

[0014] The method of concatenating the negative sequence current amplitude and the normalized amplitudes of the second, third, and fourth harmonics as the insulation degradation feature vector at each time point includes:

[0015] Within each time window, the fundamental components of the three-phase current are first extracted to obtain the corresponding fundamental amplitude. Based on the fundamental amplitude, symmetrical component decomposition is performed to obtain zero-sequence, positive-sequence, and negative-sequence components. The negative-sequence current amplitude is selected as one of the characteristics of stator inter-turn insulation degradation.

[0016] The discrete Fourier transform of the current discrete sequence within the same time window is performed at several sampling points to obtain the spectral coefficients. The harmonics in the spectral coefficients are normalized by the fundamental wave. Finally, the normalized amplitudes of the second, third, and fourth harmonics are selected as other characteristics of stator inter-turn insulation degradation.

[0017] Furthermore, including:

[0018] The construction of operating mode variables and mode confidence based on the average active power within the set sliding time window and the set dead zone threshold includes:

[0019] Within each time window, the instantaneous three-phase active power of the current generator unit is calculated, and the average value of the window is obtained based on the number of sampling points within the window, thereby obtaining the average active power within that time window.

[0020] The dead zone threshold is determined by multiplying the rated active power of the current unit by the proportional coefficient. If the average active power at the current moment is not less than the dead zone threshold, the unit is considered to be in generating mode; otherwise,

[0021] If the average active power at the current moment is not greater than the negative of the dead zone threshold, the unit is considered to be in electric mode; otherwise,

[0022] If the absolute value of the average active power at the current moment is less than the dead zone threshold, it is considered to be in the power switching or near-zero power range.

[0023] The pattern confidence is defined by the absolute value of the average active power at the current moment and the discrimination dead zone threshold, and is expressed as: ,in, For tiny positive numbers that approach zero, To determine the dead zone threshold, for Average active power at any given time.

[0024] Furthermore, including:

[0025] The step of constructing an adaptive adjacency matrix for operating modes based on the basic coupling coefficient and the operating mode sensitivity function using a normalization function includes:

[0026] To characterize the fundamental spatial coupling strength between degradation features, within a fixed statistical time window Construct time series for each feature dimension, let the first feature be... The time series of each degradation feature within the statistical window is as follows: ,in, Indicates time The extracted first Each time step has a degenerate feature value, and each time step corresponds to a sliding time window. Indicates the length of the time series;

[0027] Define features With features The basic coupling coefficient is: ,in, For covariance; Features and characteristics The standard deviation of the sequence within the statistical window; Used to characterize the fundamental coupling strength between features;

[0028] Considering the different sensitivities of degradation characteristics to insulation status between power generation and electric operation modes, a sensitivity function for the aforementioned operating mode is introduced: ,in, For the first The basic sensitivity coefficient of each degradation feature under the power generation mode, i.e. =0, This represents the increase in sensitivity of the electric mode relative to the power generation mode, i.e. =1, and set ;

[0029] Based on the aforementioned fundamental coupling coefficient With mode sensitivity function Define the elements of the adaptive adjacency matrix for the operating mode as follows: ,in, The normalized mapping function maps the edge weights to... The interval is used to obtain the candidate adjacency matrix. .

[0030] Furthermore, including:

[0031] The basic sensitivity coefficient under the power generation mode And the increase in sensitivity of electric mode relative to power generation mode The methods for determining this include:

[0032] Data grouping and windowing: collecting historical operating data of the unit and corresponding insulation health reference values. ,in, The historical operating data is divided into power generation mode subsets according to the operating mode, using either the time-aligned values ​​of offline detection test indicators or the degradation level labels obtained by mapping maintenance records. and electric mode subset Each degradation characteristic was calculated using a sliding time window. ;

[0033] Calculate the association strength between the feature and the health reference value: calculate the feature in both modes. Insulation health reference For the correlation strength, the absolute Pearson correlation coefficient is preferred: ,in, This is the function for calculating the correlation coefficient.

[0034] Normalization yields the sensitivity parameter: To eliminate the influence of dimensions and ensure comparability between different features, the correlation strength under the two modes is normalized to obtain the first sensitivity parameter. The fundamental sensitivity coefficient of each feature under power generation mode: ;in, The number of dimensions of the degenerate features. To prevent extremely small positive numbers with a denominator of zero, the sensitivity enhancement of the electric mode relative to the power generation mode is defined as: Thus ensuring .

[0035] Furthermore, including:

[0036] The adaptive update of the mode-adaptive adjacency matrix based on the value of the mode change indicator includes:

[0037] If the mode change indicator This indicates that the operating mode remains unchanged, the adjacency matrix remains unchanged, or only for... Perform a smooth update, that is: ;otherwise,

[0038] If the mode change indicator When this occurs, it indicates a change in operating mode, and a smooth transition update is performed on the adjacency matrix, represented as: ,in, The larger the value, the more reliable the pattern discrimination, and the closer the degree of adjacency matrix update. .

[0039] Furthermore, including:

[0040] The nodes in the combined graph structure are used as graph signal inputs, and graph convolution is used to aggregate the spatial association information between nodes to obtain spatial feature representations, including:

[0041] Based on the candidate adjacency matrix Construct the degree matrix The diagonal element is Based on this, construct the normalized graph Laplace matrix: ,in, It is the identity matrix;

[0042] To achieve local smoothing and multi-order neighborhood information aggregation of graph signals in the spectral domain, the graph convolution is approximated using Chebyshev polynomials. Let the convolution order be... The graph convolution output is defined as: ;in, For a moment The image signal, Representation of spatial features; For the first Convolution weights of order; For Chebyshev polynomials, To The matrix after spectral scaling.

[0043] Furthermore, including:

[0044] The process of performing time-dimensional fusion processing on the spatial feature representation to characterize the evolution trend of insulation degradation and output an insulation health index includes:

[0045] Let the length of the time series be... , define continuity The spatial feature sequence of each time window is: ;in, The length of the time series;

[0046] Perform a time-dimensional fusion operation on the spatial feature sequence to obtain a time-comprehensive representation vector: ;in, For time fusion functions, gated recurrent units, long short-term memory networks, or attention structures can be used;

[0047] Based on time-integrated representation vector Construct an insulation health index: ,in, For mapping weights, For bias terms, The normalization function makes the insulation health index... The insulation health index This is used to characterize the insulation condition of generator motors; the lower the value, the higher the degree of degradation. The insulation health index is used to characterize the insulation condition of generator motors. Continuous monitoring enables quantitative assessment and trend analysis of insulation degradation.

[0048] According to the technical solution provided by the present invention, in another aspect, the present invention also provides a generator motor insulation state prediction system, the system comprising:

[0049] The comprehensive feature vector construction module is used to collect three-phase stator current signals during the normal operation of the generator motor and splice the negative sequence current amplitude and the normalized amplitudes of the second, third and fourth harmonics as the insulation degradation feature vectors at each time point; construct operating mode variables and mode confidence based on the average active power within the set sliding time window and the set discrimination dead zone threshold, and obtain the comprehensive feature vector based on the insulation degradation feature vectors and operating mode variables.

[0050] The sensitivity difference determination module is used to construct the time series of each feature dimension in the comprehensive feature vector within a fixed statistical time window, obtain the basic coupling coefficient between two different features, and construct the operating mode adaptive adjacency matrix based on the basic coupling coefficient and the operating mode sensitivity function using a normalization function. The operating mode sensitivity function is used to determine the sensitivity difference of the degradation feature to the insulation state between the power generation mode and the electric mode according to the operating mode variable.

[0051] The indicator construction module is used to obtain the mode change indicator of two adjacent windows based on the running mode variable. The mode change indicator is used to use the running mode change as the trigger condition for dynamic updating of the graph structure. The nodes in the graph structure correspond one-to-one with the degenerate feature dimension in the comprehensive feature vector.

[0052] The state prediction module is used to adaptively update the mode adaptive adjacency matrix according to the value of the mode change indicator, and combine the nodes in the graph structure as graph signal input. It uses graph convolution to aggregate the spatial association information between nodes to obtain spatial feature representation. The spatial feature representation is then subjected to time-dimensional fusion processing to characterize the evolution trend of insulation degradation state and outputs insulation health index. Through continuous monitoring of the insulation health index, the quantitative assessment and trend analysis of the degree of insulation degradation can be realized.

[0053] According to the technical solution provided by the present invention, in a third aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method described in any of the above-mentioned embodiments.

[0054] Beneficial Effects: This invention addresses the changing sensitivity of degradation features and the coupling relationships between features under conditions of frequent switching between generator and motor operation modes. First, it constructs a spatial correlation weight generation mechanism driven by the operating mode, enabling adaptive adjustment of the spatial correlation structure according to changes in the operating mode. This achieves dynamic updates of the correlation relationships between insulation degradation features, thereby improving the accuracy and stability of insulation state prediction. Second, considering the potential differences in the sensitivity of degradation features to insulation state under generator and motor modes, this invention introduces an operating mode sensitivity function. Through the mode sensitivity function and mode change indicators, it achieves adaptive updates and smooth reconstruction of the adjacency matrix, allowing the spatial coupling relationships between degradation features to be dynamically adjusted with switching between generator and motor operating conditions. Third, this invention's mode sensitivity... The parameters of the sensitivity factor in the degree function no longer use a fixed threshold, but instead consider the correlation strength between features and health reference quantities according to different operating modes. In order to eliminate the influence of dimensions and ensure the comparability between different features, the correlation strength in the two modes is normalized to obtain the basic sensitivity coefficient in the power generation mode and the sensitivity enhancement of the electric mode relative to the power generation mode. Finally, after determining the constructed adaptive adjacency matrix of the operating mode, the present invention performs graph convolution operation on the graph signal to extract the spatial correlation representation between degradation features. The spatial correlation features are extracted by graph convolution, and the obtained spatial feature representation is combined with time dimension fusion modeling to characterize the degradation evolution trend. Thus, the operating mode information is explicitly fused in the state assessment process, enhancing the stability and robustness of the degradation feature correlation expression under complex operating conditions. Attached Figure Description

[0055] Figure 1 This is a flowchart of the method for predicting the insulation state of a generator motor according to an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the adjacency matrix weight generation and event-triggered update mechanism driven by the operating mode as described in an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of the system structure for constructing insulation degradation characteristics of a generator motor according to an embodiment of the present invention;

[0058] Figure 4 This is a schematic diagram illustrating the process of constructing the insulation health index of a generator motor according to an embodiment of the present invention. Detailed Implementation

[0059] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0060] This invention provides a method for predicting the insulation state of a generator-motor based on an adaptive spatiotemporal graph structure constructed and dynamically updated according to operating modes. By identifying the current operating mode of the generator-motor, a mode sensitivity adjustment function is constructed to adaptively calculate the spatial correlation weights between insulation degradation features. When the operating mode changes, graph structure reconstruction is triggered, achieving dynamic updates of the correlation relationships between degradation features. Based on this, the constructed adaptive spatiotemporal graph structure is used in an insulation state prediction model to obtain the predicted insulation state of the generator-motor. (See also...) Figure 1 As shown, it includes the following steps:

[0061] S1: During the normal operation of the generator motor, the three-phase stator current signal is collected and the negative sequence current amplitude and the normalized amplitudes of the second, third and fourth harmonics are spliced ​​together to construct the insulation degradation feature vector at each moment; the operating mode variables and mode confidence are constructed according to the average active power within the set sliding time window and the set discrimination dead zone threshold, and the comprehensive feature vector is obtained according to the insulation degradation feature vector and the operating mode variables.

[0062] S2: Construct the time series of each feature dimension in the comprehensive feature vector within a fixed statistical time window to obtain the basic coupling coefficient between two different features. Use the normalization function to construct the adaptive adjacency matrix of the operating mode based on the basic coupling coefficient and the operating mode sensitivity function. The operating mode sensitivity function is used to determine the difference in sensitivity of the degradation feature to the insulation state between the power generation mode and the electric mode according to the operating mode variable.

[0063] S3: Obtain the mode change indicator of two adjacent windows based on the running mode variable. The mode change indicator is used to use the running mode change as a trigger condition for dynamic updating of the graph structure. The nodes in the graph structure correspond one-to-one with the degenerate feature dimension in the comprehensive feature vector.

[0064] S4: Adaptively update the mode adaptive adjacency matrix according to the value of the mode change indicator, and combine the nodes in the graph structure as graph signal input. Use graph convolution to aggregate the spatial association information between nodes to obtain spatial feature representation. Perform time-dimensional fusion processing on the spatial feature representation to characterize the evolution trend of insulation degradation state and output insulation health index. Through continuous monitoring of the insulation health index, realize the quantitative assessment and trend analysis of the degree of insulation degradation.

[0065] Specifically, this embodiment focuses on a method for assessing the degradation status of the stator winding insulation system of a pumped-storage generator motor. The aim is to construct insulation degradation-sensitive features using three-phase stator current signals under conditions of frequent switching between generator and motor modes, and to achieve online assessment of the insulation health index based on an adaptive graph structure of the operating mode. Specifically, it provides a method for predicting the insulation status of a generator motor based on the construction and dynamic updating of an adaptive graph structure of the operating mode. Figure 4 As shown, the steps are as follows:

[0066] Step a: During the normal operation of the generator motor, collect the three-phase stator current signal. And a mode signal used to determine the operating status of the unit. The current signal is sampled at a fixed frequency. Discretize the data below, and use a length of... Feature computation is performed using a sliding time window, with each time window corresponding to a feature computation time. .

[0067] Specifically, the three-phase current at the sampling frequency Discretize to And within each time window (sampling point number within the window) The fundamental and harmonic characteristics are calculated. To ensure the effectiveness of the three-phase symmetrical component calculation, the three-phase currents are sampled synchronously using the same sampling clock and with a unified timestamp. Before entering the frequency domain calculation, the signal within the window is de-DC processed to eliminate the influence of zero drift and bias on the spectral amplitude.

[0068] Within each time window, the fundamental component of the three-phase current is first extracted to obtain the corresponding fundamental amplitude. Among them, the fundamental amplitude The amplitude is obtained using a method based on discrete Fourier transform: the current of each phase is calculated within a window. Point-based discrete Fourier transform, taking the fundamental frequency (Rated power frequency, 50Hz or 60Hz) Corresponding frequency spectrum points The amplitude is taken as the fundamental wave amplitude of the phase. The method for determining it is as follows:

[0069] Formula 1

[0070] In the above formula, This represents the integer operation. Sampling frequency, This represents the number of sampling points within the window.

[0071] Symmetrical component decomposition based on the three-phase fundamental amplitude yields zero-sequence, positive-sequence, and negative-sequence components:

[0072] Formula 2

[0073] In the above formula, This represents the fundamental amplitude of the three-phase current. It is a zero-order component; For positive order components; It is a negative-order component; is the phase rotation constant.

[0074] Negative order components Used to characterize the degree of three-phase current imbalance, when there is inter-turn insulation degradation in the stator winding, impedance asymmetry leads to an increase in the negative sequence component. Therefore, the negative sequence current amplitude is selected. As one of the sensitive characteristics of stator turn-to-turn insulation degradation.

[0075] Subsequently, the discrete current sequence within the same time window was processed. Point-based discrete Fourier transform is used to obtain the spectral coefficients. To avoid ambiguity, the following is used: Taking phase current as an example, let The spectral coefficients are then calculated as follows:

[0076] Formula 3

[0077] In the above formula, This represents a discrete sequence of currents within a window. The spectrum index; This represents the number of sampling points within the window.

[0078] No. The spectral index corresponding to the subharmonic From fundamental frequency Confirmed, the calculation method is as follows:

[0079] Formula 4

[0080] in, These correspond to the second, third, and fourth harmonics, respectively, and their harmonic amplitudes are... To eliminate the impact of load changes on the overall proportion of current amplitude, the first... The normalized amplitude of the subharmonic is:

[0081] Formula 5

[0082] In the above formula, For the first Second harmonic amplitude; The amplitude is the fundamental wave.

[0083] By normalizing the fundamental frequency, the harmonic characteristics primarily reflect changes in insulation condition rather than load variations, thus mitigating the overall scaling effect of current amplitude caused by load changes. Since load variations typically cause each order of spectral components to scale synchronously with the fundamental frequency amplitude, directly using... Load amplitude factors can easily be incorporated; after fundamental frequency normalization, the eigenvalues ​​mainly reflect changes in the relative energy distribution of harmonics. Nonlinearity and asymmetry caused by insulation degradation alter the relative energy distribution; therefore, It can be used to more stably characterize changes in insulation state.

[0084] According to the insulation degradation mechanism, stator turn imbalance usually manifests as an increase in negative sequence current, enhanced nonlinearity of main insulation usually manifests as third harmonic variation, and magnetic field asymmetry or rotor abnormality usually manifests as enhanced even harmonics. Therefore, the negative sequence current amplitude is selected. and the normalized amplitudes of the second, third, and fourth harmonics. Constructing Time Insulation degradation feature vector:

[0085] Formula 6

[0086] In the above formula, , The negative sequence component is obtained by symmetrical component decomposition of the three-phase fundamental components within the time window. , To obtain the first... Second harmonic amplitude The amplitude is the fundamental frequency, and For the first The spectral index corresponding to the subharmonic. All the above features are calculated from data within the same sliding time window to ensure temporal consistency and comparability of the features.

[0087] Step b: Determine the current operating mode of the generator motor based on the unit's operating data to obtain the operating mode variable. And generate pattern change indicators. This serves as the triggering condition for the adaptive update of the graph structure in subsequent step c; simultaneously, to quantify the reliability of pattern discrimination, a pattern confidence score is constructed. This is used to adaptively adjust the intensity of graph structure updates in subsequent steps.

[0088] Since the generator motor operates in two basic modes during pumped storage hydroelectric power generation: generator mode and motor mode, and the difference between these two modes can be clearly determined by the direction of the unit's active power, this embodiment uses the three-phase voltage and three-phase current on the stator side of the unit to calculate the active power, and uses the sign of the active power as the criterion for the operating mode. Within each time window, the instantaneous three-phase active power of the unit is calculated and the average value of the window is taken to obtain the window active power. Its expression is:

[0089] Formula 7

[0090] In the above formula, The first in the window The three-phase stator phase voltages of phases A, B, and C at each sampling point; The first in the window The three-phase stator currents of phases A, B, and C at each sampling point; This represents the number of sampling points within the window. This represents the average active power within that time window. Using the window average can reduce the impact of instantaneous disturbances on mode discrimination, thus stabilizing the discrimination results.

[0091] Optionally, in this embodiment, to avoid the unit from experiencing issues during start-up, shutdown, switching, or light load phases... Approaching zero can lead to misjudgments or pattern jitter; a dead zone threshold can be set. It should be noted that a dead-zone threshold is used and... Maintaining the previous judgment result when the device is within the dead zone is a well-known steady-state vibration suppression measure. This embodiment further utilizes the change in operating mode to trigger subsequent event updates of the graph structure. As an implementation method that facilitates engineering configuration, the dead zone threshold can be set to the rated active power. Fixed ratio:

[0092] Formula 8

[0093] In the above formula, The rated active power of the unit; This is the dead zone scaling factor, and its value range is... By binding the dead zone threshold to the rated power, the threshold has a clear physical meaning and can be determined by the equipment nameplate parameters, avoiding the uncertainty brought about by "experience thresholds".

[0094] Under the above optional implementation methods, the operating mode variable It can be defined according to the following rules:

[0095] Formula 9

[0096] In the above formula, For a moment The running mode variable; when Time indicates the power generation mode; when When indicates electric mode; when This indicates that the system is in a power switching or near-zero power range. To avoid mode jitter, the mode from the previous moment is maintained, i.e., the value is taken as... .

[0097] Furthermore, to use changes in operating mode as a trigger condition for dynamic updates of the graph structure, a mode change indicator is defined.

[0098] Formula 10

[0099] In the above formula, Indicates whether the operating mode of two adjacent windows has switched; when The time indicates that the mode has switched. The time indication mode remains unchanged. This will be used to trigger the reconstruction or update of the spatial association structure in subsequent steps.

[0100] Furthermore, to quantify the reliability of operational mode discrimination, a mode confidence level is defined:

[0101] Formula 11

[0102] in, It is a very small positive number. Characterizes the "distance" of the current window power amplitude from the dead zone threshold. The larger the value, the more reliable the pattern discrimination; the stated This is used to adaptively adjust the adjacency matrix update strength or transition smoothing coefficient in subsequent steps.

[0103] Step c: After obtaining the insulation degradation characteristics constructed in step a and the operating mode variables determined in step b. Mode change indicator and pattern confidence Subsequently, a spatiotemporal graph structure is constructed to characterize the spatial relationships between degenerate features, and the adjacency matrix is ​​adaptively updated and smoothly transitioned based on the change events of the operating mode.

[0104] First, at the feature calculation time corresponding to each sliding time window The comprehensive feature vector is defined as follows:

[0105] Formula 12

[0106] In the above formula: This represents the amplitude of the negative sequence current. These are the normalized amplitudes of the second, third, and fourth harmonics; This is the zero-sequence current component; The operating mode variable obtained in step b. This feature vector serves as the attribute input for nodes in the graph structure, describing the insulation degradation state and operating conditions at the current moment.

[0107] Furthermore, the insulation degradation system is modeled as a graph structure, defining time intervals. The image is as follows:

[0108] Formula 13

[0109] in, For example, a set of nodes, where each node corresponds one-to-one with a dimension of the degradation feature. Each corresponds to a different node; It is the set of edges connecting nodes; Let be an adjacency matrix, with matrix elements denoted as . Used to characterize time node With nodes Spatial correlation strength.

[0110] To characterize the fundamental spatial coupling strength between degradation features, within a fixed statistical time window The time series of each feature dimension are constructed internally. Let the first feature be... The time series of each degradation feature within the statistical window is as follows:

[0111] Formula 14

[0112] in, Indicates time (Corresponding to a sliding time window) the extracted first A degenerate eigenvalue, for example, when When corresponding to negative order features, ;when When corresponding to the third harmonic characteristics, .therefore, and These are feature value sequences formed at multiple adjacent time points for two different feature dimensions.

[0113] Based on this, such as Figure 2 As shown, the features are defined. With features The basic coupling coefficient (standardized correlation coefficient) is:

[0114] Formula 15

[0115] In the above formula, For covariance; These represent the standard deviations of the corresponding feature sequences within the statistical window; The coupling matrix obtained by Equation 15 is used to characterize the basic coupling strength between features and to represent the basic correlation between different degenerate features.

[0116] Considering the difference in degradation sensitivity between power generation mode and electric mode, an operation mode variable is introduced. ,in Indicates the power generation mode. This represents the electric mode. To characterize the moderating effect of different operating modes on the sensitivity to degradation features, a mode sensitivity function is constructed: Equation 16, in the above equation, For the first A degradation characteristic in power generation mode ( The basic sensitivity coefficient under ) Electric mode relative to power generation mode ( The sensitivity enhancement amount is preferably set to... It should be noted that, and These are fixed parameters obtained from offline calibration; changes in operating mode are only achieved by modifying... The value of makes exist and Switching between them does not change and The value.

[0117] In this embodiment and The data statistical calibration method using different operating modes is used to determine the following steps:

[0118] (1) Data grouping and windowing: Collect historical operating data of the unit and corresponding insulation health reference values. .in, This can be a time-aligned value of offline test indicators (such as dielectric loss, partial discharge, insulation resistance, etc.) or a degradation level label mapped from maintenance records. Historical data is divided into power generation mode subsets according to operating modes. and electric mode subset Each degradation characteristic was calculated using a sliding time window. .

[0119] (2) Calculate the correlation strength between the feature and the health reference value: Calculate the feature under both modes. With health reference level For the correlation strength, the absolute Pearson correlation coefficient is preferred:

[0120] Formula 17

[0121] In the above formula, This is the function for calculating the correlation coefficient. If... For rank labels, Spearman's rank correlation coefficient can also be used instead.

[0122] (3) Normalization to obtain sensitivity parameters: In order to eliminate the influence of dimensions and ensure the comparability between different features, the correlation strength under the two modes is normalized to obtain the sensitivity parameter. The fundamental sensitivity coefficient of each feature under power generation mode:

[0123] Formula 18

[0124] In the above formula The number of dimensions of the degenerate features. To prevent extremely small positive numbers with a denominator of zero.

[0125] Therefore, the sensitivity enhancement of the electric mode relative to the power generation mode is defined as:

[0126] Formula 19

[0127] Thus guarantee .

[0128] To improve parameter stability, in a preferred embodiment, the correlation coefficient can be repeatedly calculated using multiple statistical windows, and the mean / median can be taken. When the sample size is insufficient, a minimum sample threshold can be set. ,like or If so, data from adjacent operating conditions or historical periods will be used to supplement the data.

[0129] Based on the fundamental coupling coefficient With mode sensitivity function Define the elements of the adaptive adjacency matrix for the operating mode as follows:

[0130] Formula 20

[0131] In the above formula, The normalized mapping function maps the edge weights to... Intervals, such as linear truncation normalization or sigmoid mapping, are used to obtain the candidate adjacency matrix. .

[0132] To avoid spatial feature jitter caused by sudden changes in the adjacency matrix during mode switching, an adjacency matrix update strategy of "event triggering + confidence control" is adopted: when the mode changes, the indicator... At this time, the operating mode remains unchanged. To reduce computational load and maintain structural stability, the adjacency matrix remains unchanged or is only modified. Perform a smooth update, preferably:

[0133] Formula 21

[0134] When the mode changes the indicator When this occurs, it indicates a change in the operating mode. First, the candidate adjacency matrix is ​​calculated according to Equation 20. Furthermore, the pattern confidence obtained in step b is introduced. As a transition coefficient, the adjacency matrix is ​​updated smoothly:

[0135] Formula 22

[0136] In the above formula, The larger the value, the more reliable the pattern discrimination, and the closer the degree of adjacency matrix update. ;when When the size is small, the adjacency matrix tends to maintain the structure of the previous time step in order to suppress erroneous updates and oscillations caused by switching transients.

[0137] The above method enables dynamic updating of the graph structure based on operational mode change events, allowing the spatial relationships between degradation features to adaptively adjust with changing operating conditions under frequent switching between power generation and electric modes. Step d: The operational mode adaptive adjacency matrix constructed in step c. Once determined, the time... The image signal Graph convolution operations are performed to extract spatial relationships between degenerate features. It should be noted that... This refers to the candidate adjacency matrix calculated in step c based on the basic coupling coefficient and the mode sensitivity function. Based on the pattern change indication quantity and pattern confidence The final adjacency matrix is ​​updated (including smooth transition); the construction of the normalized Laplacian matrix and the graph convolution operation in this step are both based on the final adjacency matrix. Finish.

[0138] The time is obtained in step c The operating mode of the adaptive adjacency matrix Then, the composite node attribute vector is used as the graph signal input, and graph convolution is used to aggregate the spatial correlation information between degenerate features to obtain spatial feature representation. It should be noted that graph Laplacian matrix construction and spectral domain graph convolution are conventional graph signal processing methods. The improvement in this embodiment lies in the adjacency matrix upon which graph convolution depends. The system is designed to adapt to different operating modes and is updated by mode change events, thereby dynamically adjusting spatial aggregation relationships according to operating conditions.

[0139] Based on adjacency matrix Construct the degree matrix The diagonal element is Based on this, construct the normalized graphical Laplace matrix:

[0140] Formula 23

[0141] In the above formula, It is the identity matrix. Represents the inverse square root of the degree matrix; The normalized graph Laplacian matrix is ​​used to describe the topological properties of graph structures and serves as a fundamental operator for spectral domain convolution.

[0142] To avoid To mitigate the high computational complexity of eigenvalue decomposition, and to achieve local smoothing and multi-order neighborhood information aggregation of graph signals in the spectral domain, Chebyshev polynomials are used to approximate the graph convolution. Let the convolution order be... Define the scaling Laplacian matrix as:

[0143] Formula 24

[0144] In the above formula, for The largest eigenvalue is used to... Spectral range scaled to This is to meet the numerical stability requirements of the Chebyshev polynomial; This is the scaled Laplace matrix.

[0145] The graph convolution output is defined as:

[0146] Formula 25

[0147] in, For a moment The graph signal, which is the corresponding node attribute input, can be the comprehensive feature vector from step c in this embodiment. Arrangement along the node dimension Representation of spatial features; For the first Convolution weights of order; For Chebyshev polynomials; To The matrix after spectral scaling. The order of the polynomial is used to control the neighborhood range of spatial aggregation. The larger the value, the more distant graph neighborhood information can be aggregated.

[0148] The underlying Chebyshev polynomials satisfy the recurrence relation:

[0149] Formula 26

[0150] The above calculations yielded... This can be reflected in the adaptive adjacency matrix of the operating mode. The spatial coupling relationship between the degenerate features under constraints serves as the input for subsequent temporal fusion modeling. That is, through the aforementioned recursive operations, the modeling of the graph structure is achieved. The aggregation of neighborhood information yields a representation vector reflecting the spatial coupling relationship of degenerate features. The aforementioned As a result of spatial modeling, it will be input into the subsequent time-dimensional modeling module to characterize the dynamic evolution of the insulation degradation process.

[0151] Step e: Spatial feature representation obtained in step d Based on this, a time-dimensional fusion process is performed to characterize the evolution trend of insulation degradation, thereby outputting an insulation health index. .

[0152] Specifically, during the online evaluation process, as the sliding time window scrolls forward, a series of spatial feature representations can be obtained continuously. To achieve time modeling of the degradation and evolution process, a length of [length value missing] is taken. The time series window, which is updated in step a with respect to the sliding window, is arranged according to the feature calculation time and constructed using time... Spatial feature sequence of the ends:

[0153] Formula 27

[0154] In the above formula, This represents the input sequence used for time modeling; The sequence length is used to control the range of time dependencies; This represents the spatial characteristics of a corresponding historical moment.

[0155] Perform a time-dimensional fusion operation on the spatial feature sequence to obtain a time-comprehensive representation vector:

[0156] Formula 28

[0157] in, This is a time fusion function used to dynamically model the evolution of spatial feature sequences and extract degradation trend information. In this embodiment, it can be implemented using a gated recurrent unit, a long short-term memory network, or an attention structure. Preferably, the time fusion function can be implemented using any engineering method capable of processing time series inputs and outputting fixed-length representation vectors, such as sliding weighted convergence, gated recursive structures, or other time series modeling modules.

[0158] Optionally, to reduce the disturbance to timing modeling caused by the instantaneous switching of operating modes, when the mode change indicator... At that time, the hidden state of the time fusion module is decayed or reset to enhance robustness when switching between different operating modes.

[0159] Based on time-integrated representation vector Construct an insulation health index To ensure the output has a uniform dimension and is easy to use in engineering applications, the following will be implemented: Linear mapping and normalization to An interval is defined as:

[0160] Formula 29

[0161] in, For mapping weights, For bias terms, The normalization function makes the insulation health index... Insulation health index For a moment The insulation health index is a quantitative indicator; the closer the value is to 1, the healthier the insulation condition, and the closer the value is to 0, the more severe the insulation degradation. By analyzing the insulation health index... Continuous monitoring enables quantitative assessment and trend analysis of insulation degradation.

[0162] Another aspect of this embodiment provides a generator motor insulation state prediction system, such as... Figure 3 As shown, the system includes:

[0163] The comprehensive feature vector construction module is used to collect three-phase stator current signals during the normal operation of the generator motor and splice the negative sequence current amplitude and the normalized amplitudes of the second, third and fourth harmonics as the insulation degradation feature vectors at each time point; construct operating mode variables and mode confidence based on the average active power within the set sliding time window and the set discrimination dead zone threshold, and obtain the comprehensive feature vector based on the insulation degradation feature vectors and operating mode variables.

[0164] The sensitivity difference determination module is used to construct the time series of each feature dimension in the comprehensive feature vector within a fixed statistical time window, obtain the basic coupling coefficient between two different features, and construct the operating mode adaptive adjacency matrix based on the basic coupling coefficient and the operating mode sensitivity function using a normalization function. The operating mode sensitivity function is used to determine the sensitivity difference of the degradation feature to the insulation state between the power generation mode and the electric mode according to the operating mode variable.

[0165] The indicator construction module is used to obtain the mode change indicator of two adjacent windows based on the running mode variable. The mode change indicator is used to use the running mode change as the trigger condition for dynamic updating of the graph structure. The nodes in the graph structure correspond one-to-one with the degenerate feature dimension in the comprehensive feature vector.

[0166] The state prediction module is used to adaptively update the mode adaptive adjacency matrix according to the value of the mode change indicator, and combine the nodes in the graph structure as graph signal input. It uses graph convolution to aggregate the spatial association information between nodes to obtain spatial feature representation. The spatial feature representation is then subjected to time-dimensional fusion processing to characterize the evolution trend of insulation degradation state and outputs insulation health index. Through continuous monitoring of the insulation health index, the quantitative assessment and trend analysis of the degree of insulation degradation can be realized.

[0167] Other technical features of the generator motor insulation state prediction system described in this embodiment are similar to the corresponding generator motor insulation state prediction method, and will not be repeated here.

[0168] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this application.

[0169] Furthermore, some embodiments of this application also provide an electronic device. The electronic device can be various forms of digital computer, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, etc. The electronic device can also be various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices.

[0170] The electronic device includes: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform a method for predicting the insulation state of a generator motor as provided in any one or more of the above embodiments. The electronic device includes: one or more central processing units (CPUs), and interfaces for connecting various components, such as displays, infrared sensors, and cameras. That is, the various components are interconnected using different buses and can be mounted on a common motherboard or otherwise installed as needed. The processor can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory sets, if desired. Similarly, multiple electronic devices can be connected, each providing some of the necessary operations. The components, their connections and relationships, and their functions shown in this embodiment are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0171] In a preferred embodiment of this invention, the electronic device may further include an input device and an output device. The processing unit, memory, input device, and output device may be connected via a bus or other means.

[0172] The input device can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the electronic device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. The output device may include a display device, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The display device may include, but is not limited to, liquid crystal displays, light-emitting diode displays, and plasma displays. In some embodiments, the display device may be a touchscreen.

[0173] To provide interaction with the user, the electronic device can be a computer. The computer has: a display device (e.g., a cathode ray tube or LCD monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback); and input from the user can be received in any form (e.g., voice input or tactile input).

[0174] In this embodiment, a computer-readable medium stores a computer program / instructions, which, when executed by a processor, implement a method for predicting the insulation state of a generator motor provided in any one or more of the above embodiments. This computer-readable medium may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into that device. The aforementioned computer-readable medium carries one or more computer-readable instructions.

[0175] Memory can serve as a non-transitory computer-readable storage medium, used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The central processing unit executes various server functions and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the program instructions / modules corresponding to the methods provided in any one or more of the embodiments described above in this application.

[0176] The memory may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device, etc. Furthermore, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, and these remote memories may be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0177] It should be noted that the computer-readable medium described in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. Computer-readable media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections having one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0178] Computer-readable media include permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, read-only optical discs, digital versatile optical discs or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0179] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including local area networks (LANs) or wide area networks (WANs), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0180] In the above embodiments, all or part of the implementation can be achieved through software, hardware, firmware, or any combination thereof. For example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In some embodiments, the software program of this application can be executed by a processor to implement the above steps or functions. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, magnetic or optical drives, floppy disks, and similar devices. In addition, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

[0181] The computer program product provided in this application includes one or more computer programs / instructions. When executed by a processor, these computer programs / instructions generate, in whole or in part, the processes or functions described in this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0182] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0183] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device in software or hardware. Terms such as "first," "second," etc., are used only for distinguishing descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.

[0184] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily made by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.

Claims

1. A method for predicting the insulation state of a generator motor, characterized in that, The method includes: During normal operation of the generator motor, three-phase stator current signals are collected, and the negative sequence current amplitude and the normalized amplitudes of the second, third, and fourth harmonics are spliced ​​together to construct the insulation degradation feature vector at each moment. Based on the average active power within the set sliding time window and the set discrimination dead zone threshold, operation mode variables and mode confidence are constructed, and a comprehensive feature vector is obtained based on the insulation degradation feature vector and the operation mode variables. Within a fixed statistical time window, the time series of each feature dimension in the comprehensive feature vector is constructed to obtain the basic coupling coefficient between two different features. The normalization function is used to construct the adaptive adjacency matrix of the operation mode based on the basic coupling coefficient and the operation mode sensitivity function. The operation mode sensitivity function is used to determine the difference in sensitivity of the degradation feature to the insulation state between the power generation mode and the electric mode according to the operation mode variable. Based on the running mode variable, the mode change indicator of two adjacent windows is obtained. The mode change indicator is used to use the running mode change as a trigger condition for dynamic updating of the graph structure. The nodes in the graph structure correspond one-to-one with the degenerate feature dimension in the comprehensive feature vector. The adaptive adjacency matrix of the mode is adaptively updated according to the value of the mode change indicator, and the nodes in the graph structure are used as graph signal input. Graph convolution is used to aggregate the spatial association information between nodes to obtain spatial feature representation. The spatial feature representation is subjected to time dimension fusion processing to characterize the evolution trend of insulation degradation state and output insulation health index. By continuously monitoring the insulation health index, the degree of insulation degradation can be quantitatively assessed and trend analyzed.

2. The method for predicting the insulation state of a generator motor according to claim 1, characterized in that, The method of concatenating the negative sequence current amplitude and the normalized amplitudes of the second, third, and fourth harmonics as the insulation degradation feature vector at each time point includes: Within each time window, the fundamental components of the three-phase current are first extracted to obtain the corresponding fundamental amplitude. Based on the fundamental amplitude, symmetrical component decomposition is performed to obtain zero-sequence, positive-sequence, and negative-sequence components. The negative-sequence current amplitude is selected as one of the characteristics of stator inter-turn insulation degradation. The discrete Fourier transform of the current discrete sequence within the same time window is performed at several sampling points to obtain the spectral coefficients. The harmonics in the spectral coefficients are normalized by the fundamental wave. Finally, the normalized amplitudes of the second, third, and fourth harmonics are selected as other characteristics of stator inter-turn insulation degradation.

3. The method for predicting the insulation state of a generator motor according to claim 2, characterized in that, The construction of operating mode variables and mode confidence based on the average active power within the set sliding time window and the set dead zone threshold includes: Within each time window, the instantaneous three-phase active power of the current generator unit is calculated, and the average value of the window is obtained based on the number of sampling points within the window, thereby obtaining the average active power within that time window. The dead zone threshold is determined by multiplying the rated active power of the current unit by the proportional coefficient. If the average active power at the current moment is not less than the dead zone threshold, the unit is considered to be in generating mode; otherwise, If the average active power at the current moment is not greater than the negative of the dead zone threshold, the unit is considered to be in electric mode; otherwise, If the absolute value of the average active power at the current moment is less than the dead zone threshold, it is considered to be in the power switching or near-zero power range. The pattern confidence is defined by the absolute value of the average active power at the current moment and the discrimination dead zone threshold, and is expressed as: ,in, For tiny positive numbers that approach zero, To determine the dead zone threshold, for Average active power at any given time.

4. The method for predicting the insulation state of a generator motor according to claim 3, characterized in that, The step of constructing an adaptive adjacency matrix for operating modes based on the basic coupling coefficient and the operating mode sensitivity function using a normalization function includes: To characterize the fundamental spatial coupling strength between degradation features, within a fixed statistical time window Construct time series for each feature dimension, let the first feature be... The time series of each degradation feature within the statistical window is as follows: ,in, Indicates time The extracted first A degenerate eigenvalue, Indicates the length of the time series; Define features With features The basic coupling coefficient is: ,in, For covariance; Features and characteristics The standard deviation of the sequence within the statistical window; Used to characterize the fundamental coupling strength between features; Considering the different sensitivities of degradation characteristics to insulation status between power generation and electric operation modes, a sensitivity function for the aforementioned operating mode is introduced: ,in, For the first The basic sensitivity coefficient of each degradation characteristic under the power generation mode This represents the increase in sensitivity of the electric mode relative to the power generation mode. For a moment The running mode variable, and set ; Based on the aforementioned fundamental coupling coefficient With mode sensitivity function Define the elements of the adaptive adjacency matrix for the operating mode as follows: ,in, The normalized mapping function maps the edge weights to... The interval is used to obtain the candidate adjacency matrix. .

5. The method for predicting the insulation state of a generator motor according to claim 4, characterized in that, The basic sensitivity coefficient under the power generation mode And the increase in sensitivity of electric mode relative to power generation mode The methods for determining this include: Data grouping and windowing: collecting historical operating data of the unit and corresponding insulation health reference values. ,in, The historical operating data is divided into power generation mode subsets according to the operating mode, using either the time-aligned values ​​of offline detection test indicators or the degradation level labels obtained by mapping maintenance records. and electric mode subset Each degradation characteristic was calculated using a sliding time window. ; Calculate the association strength between the feature and the health reference value: calculate the feature in both modes. Insulation health reference The correlation strength was determined using the absolute Pearson correlation coefficient: ,in, This is the function for calculating the correlation coefficient. Normalization yields the sensitivity parameter: To eliminate the influence of dimensions and ensure comparability between different features, the correlation strength under the two modes is normalized to obtain the first sensitivity parameter. The fundamental sensitivity coefficient of each feature under power generation mode: ;in, The number of dimensions of the degenerate features. To prevent extremely small positive numbers with a denominator of zero, the sensitivity enhancement of the electric mode relative to the power generation mode is defined as: Thus ensuring .

6. The method for predicting the insulation state of a generator motor according to claim 4, characterized in that, The adaptive update of the mode-adaptive adjacency matrix based on the value of the mode change indicator includes: If the mode change indicator When this occurs, it indicates that the operating mode remains unchanged, the adjacency matrix remains unchanged, or only for... Perform a smooth update, that is: ;otherwise, If the mode change indicator When this occurs, it indicates a change in operating mode, and a smooth transition update is performed on the adjacency matrix, represented as: ,in, The larger the value, the more reliable the pattern discrimination, and the closer the degree of adjacency matrix update. .

7. The method for predicting the insulation state of a generator motor according to claim 6, characterized in that, The nodes in the combined graph structure are used as graph signal inputs, and graph convolution is used to aggregate the spatial association information between nodes to obtain spatial feature representations, including: Based on the candidate adjacency matrix Construct the degree matrix The diagonal element is Based on this, construct the normalized graph Laplace matrix: ,in, It is the identity matrix; To achieve local smoothing and multi-order neighborhood information aggregation of graph signals in the spectral domain, the graph convolution is approximated using Chebyshev polynomials. Let the convolution order be... The graph convolution output is defined as: ;in, For a moment The image signal, Representation of spatial features; For the first Convolution weights of order; For Chebyshev polynomials, To The matrix after spectral scaling.

8. The method for predicting the insulation state of a generator motor according to claim 7, characterized in that, The process of performing time-dimensional fusion processing on the spatial feature representation to characterize the evolution trend of insulation degradation and output an insulation health index includes: Let the length of the time series be... , define continuity The spatial feature sequence of each time window is: ;in, The length of the time series; Perform a time-dimensional fusion operation on the spatial feature sequence to obtain a time-comprehensive representation vector: ;in, For time fusion function; Based on time-integrated representation vector Construct an insulation health index: ,in, For mapping weights, For bias terms, The normalization function makes the insulation health index... The insulation health index This is used to characterize the insulation condition of generator motors; the lower the value, the higher the degree of degradation. The insulation health index is used to characterize the insulation condition of generator motors. Continuous monitoring enables quantitative assessment and trend analysis of insulation degradation.

9. A generator motor insulation condition prediction system, characterized in that, The system includes: The comprehensive feature vector construction module is used to collect three-phase stator current signals during the normal operation of the generator motor and splice the negative sequence current amplitude and the normalized amplitudes of the second, third and fourth harmonics as the insulation degradation feature vectors at each time point; construct operating mode variables and mode confidence based on the average active power within the set sliding time window and the set discrimination dead zone threshold, and obtain the comprehensive feature vector based on the insulation degradation feature vectors and operating mode variables. The sensitivity difference determination module is used to construct the time series of each feature dimension in the comprehensive feature vector within a fixed statistical time window, obtain the basic coupling coefficient between two different features, and construct the operating mode adaptive adjacency matrix based on the basic coupling coefficient and the operating mode sensitivity function using a normalization function. The operating mode sensitivity function is used to determine the sensitivity difference of the degradation feature to the insulation state between the power generation mode and the electric mode according to the operating mode variable. The indicator construction module is used to obtain the mode change indicator of two adjacent windows based on the running mode variable. The mode change indicator is used to use the running mode change as the trigger condition for dynamic updating of the graph structure. The nodes in the graph structure correspond one-to-one with the degenerate feature dimension in the comprehensive feature vector. The state prediction module is used to adaptively update the mode adaptive adjacency matrix according to the value of the mode change indicator, and combine the nodes in the graph structure as graph signal input. It uses graph convolution to aggregate the spatial association information between nodes to obtain spatial feature representation. The spatial feature representation is then subjected to time-dimensional fusion processing to characterize the evolution trend of insulation degradation state and outputs insulation health index. Through continuous monitoring of the insulation health index, the quantitative assessment and trend analysis of the degree of insulation degradation can be realized.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 8.

Images