Generator partial discharge online monitoring system

Abstract

The invention relates to the field of monitoring, and discloses a generator partial discharge on-line monitoring system, which is used for realizing multiplication effect of early warning reliability and improvement of prediction accuracy. According to the generator partial discharge online monitoring system, a degradation track model is established by continuously accumulating historical data, and a comparison reference is provided for real-time monitoring; the fault identification precision is improved by using mode matching and anomaly detection, and the result reliability is ensured by combining a life prediction correction model; and finally generating a comprehensive monitoring report fused with multi-dimensional analysis. According to the invention, the adaptive capability and long-term monitoring accuracy of the system are improved, and a continuously optimized scientific support is provided for generator state evaluation and maintenance decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of monitoring, in particular to a generator partial discharge online monitoring system. BACKGROUND

[0002] As the core equipment of power system, the operation reliability of generator is directly related to the safety and stability of the entire power grid. Partial discharge (PD) is an important indicator to reflect the insulation state of generator, and effective PD monitoring can early warn insulation faults and avoid catastrophic accidents. With the development of power equipment towards high voltage and large capacity, higher requirements are put forward for generator insulation state monitoring technology. The existing technology has the following shortcomings: 1. Limitations of traditional threshold alarm technology; The existing PD monitoring system can only detect when the insulation degradation reaches a certain level, missing the opportunity for early warning. It is unable to predict the development trend of insulation state, making it difficult to support preventive maintenance decisions. The fixed threshold cannot adapt to the state changes under different operating conditions, and is prone to false alarms or missed alarms; 2. Monitoring method based on statistical analysis; A large amount of historical fault data is needed to establish a statistical model, and the recognition ability for new types of faults is limited. The traditional statistical method is based on linear assumption, which cannot effectively handle the nonlinear characteristics of PD signals. It mainly relies on time domain and frequency domain features, and fails to fully exploit the deep dynamic information in the signals; 3. Limitations of machine learning methods; The model decision-making process is not transparent, and lacks physical meaning explanation, which has trust barriers in key power equipment applications. A large amount of labeled data is needed for training, while the fault samples of generator are scarce, which restricts the model generalization ability. In the case of limited samples, it is easy to overfit, affecting the reliability in actual working conditions.

[0003] Therefore, we propose a generator partial discharge online monitoring system to solve the above problems. SUMMARY

[0004] The present application provides a generator partial discharge online monitoring system for realizing the multiplication effect of early warning reliability and the improvement of prediction accuracy.

[0005] The first aspect of the present application provides a kind of generator partial discharge online monitoring method, the kind of generator partial discharge online monitoring method includes: processing module obtains multiple-point partial discharge signal by acquisition device, generates pure partial discharge signal sequence using adaptive noise suppression technology to signal pre-processing;Extraction module, the pure partial discharge signal sequence is analyzed and phase space reconstruction is implemented, generates multi-scale fractal feature vector and high-dimensional phase space trajectory point cloud;Analysis module, using the high-dimensional phase space trajectory point cloud, executes nonlinear dynamics stability analysis and algebraic topological invariant calculation, produces dynamics stability index set, and topological feature set;Fusion module, the multi-scale fractal feature vector, dynamics stability index set and topological feature set are carried out feature fusion, constructs time series prediction system, generates quantitative prediction result;Output module, according to the quantitative prediction result, combine generator parameters, generate multi-level early warning signal and customized maintenance scheme.

[0006] Optionally, in the first implementation manner of the first aspect of the present application, a high-frequency wide-band sensor array is used to collect multiple original partial discharge signals to generate an original multi-channel signal data set; the original multi-channel signal data set is analyzed to generate a noise feature matrix; based on the noise feature matrix, optimal wavelet packet decomposition parameters are determined, the original partial discharge signals are decomposed into wavelet packet nodes to generate a wavelet packet decomposition coefficient matrix; according to the energy distribution characteristics of each wavelet packet node and the signal-to-noise ratio evaluation results, a coefficient processing mechanism is established to process the wavelet packet decomposition coefficient matrix to generate a wavelet packet coefficient reconstruction matrix; the wavelet packet coefficient reconstruction matrix is used for reconstruction to generate a pure partial discharge signal sequence.

[0007] Optionally, in the second implementation manner of the first aspect of the present application, the pure partial discharge signal sequence is processed to generate a signal analysis segment, each signal analysis segment is eliminated to generate a signal segment set; the signal segment set is analyzed to generate a scale behavior feature set by calculating a fluctuation function value and a statistical moment; based on the scale behavior feature set, a multi-scale fractal feature vector is generated by calculation and analysis; the pure partial discharge signal sequence is calculated to determine optimal reconstruction parameters to generate a phase space reconstruction parameter optimization combination; the one-dimensional pure partial discharge signal sequence is mapped to a high-dimensional phase space using the phase space reconstruction parameter optimization combination to generate a high-dimensional phase space trajectory point cloud; the multi-scale fractal feature vector and the high-dimensional phase space trajectory point cloud are processed to generate a multi-fractal-phase space joint feature data set.

[0008] Optionally, in the third implementation form of the first aspect of the present application, the high-dimensional phase space trajectory point cloud is reconstructed and analyzed, the exponential divergence rate is calculated, and the Lyapunov exponent spectrum feature is generated; based on the Lyapunov exponent spectrum feature, the Lyapunov exponent and the exponential spectrum distribution feature are extracted, the correlation dimension and the time variation rate thereof are calculated, and the dynamic stability index set is generated; the high-dimensional phase space trajectory point cloud is analyzed, the Vietoris-Rips complex sequence is constructed, and the persistent homology feature is generated; the evolution sequence is extracted from the persistent homology feature, the lifetime distribution of the topological feature and the topological entropy value are calculated, and the topological feature set is generated; and the dynamic stability index set and the topological feature set are analyzed, and the dynamic-topological joint feature matrix is generated.

[0009] Optionally, in the fourth implementation form of the first aspect of the present application, the multi-scale fractal feature vector, the dynamic stability index set and the topological feature set are processed, the feature fusion vector is constructed, and the system state feature matrix is generated; based on the system state feature matrix, the model is constructed, the time evolution law is described, and the state space prediction model is generated; the system state feature matrix is predicted by using the state space prediction model, and the system state prediction sequence is generated; based on the system state prediction sequence, the trajectory model is established in combination with the data of the generator system, the quantitative life assessment result is generated through the mapping relationship; and the system state prediction sequence and the quantitative life assessment result are analyzed, and the quantitative prediction result is generated through the analysis.

[0010] Optionally, in the fifth implementation form of the first aspect of the present application, the quantitative prediction result is analyzed with the generator parameters, and the comprehensive state evaluation matrix is generated; based on the comprehensive state evaluation matrix, different early warning threshold values are set for the topological invariant feature, the dynamic stability index and the multi-fractal feature by using the decision algorithm, and the dynamic early warning threshold matrix is generated; the multi-level early warning trigger judgment is performed according to the dynamic early warning threshold matrix, and the three-level early warning signal set is generated through calculation and mapping; based on the three-level early warning signal set and the quantitative prediction result, the maintenance strategy set is generated in combination with the maintenance data and the equipment operation specification; and the three-level early warning signal set and the maintenance strategy set are optimized, and the customized maintenance scheme is generated.

[0011] Optionally, in the sixth implementation form of the first aspect of the present application, a generator evolution database is established, the quantitative prediction result, the multi-scale fractal feature vector, the dynamic stability index set and the topological feature set are stored in time sequence to generate a state feature database; based on the state feature database, a trajectory model is constructed, a fault mode recognition system is generated through a key feature evolution mode; the fault mode recognition system is used to match and compare data to generate a mode recognition report; based on the mode recognition report, a life prediction correction model is established in combination with generator parameters and history, and a prediction quality report is generated through data and analysis; the mode recognition report and the prediction quality report are analyzed to generate a comprehensive monitoring report.

[0012] The mechanism of the present application is as follows: the self-similarity evolution law of partial discharge signals under multi-scale is revealed through multi-fractal detrended fluctuation analysis, a one-dimensional time series is mapped to a high-dimensional dynamic trajectory by using phase space reconstruction theory, the sensitivity of the system to initial conditions and the instability precursor are quantitatively characterized based on Lyapunov exponent spectrum, and the continuity change of the internal structure of the system is identified from the geometric topological point of view by introducing persistent homology theory in algebraic topology, thereby realizing the paradigm shift from "phenomenon monitoring" to "essence dynamics analysis".

[0013] Beneficial effects: the present application realizes the deep integration of multi-fractal analysis, nonlinear dynamics and algebraic topology theory, and constructs a complete cognitive system from signal features to system essence. The multi-fractal analysis reveals the micro features of insulation deterioration, the phase space reconstruction exhibits the system dynamics evolution law, the Lyapunov exponent warns the instability trend, and the topological invariant identifies the structural change, thereby forming a multi-dimensional collaborative monitoring mechanism. The system realizes the synergistic effect of early warning, accurate prediction and intelligent decision-making. The present application improves the fault warning sensitivity through the mutual verification of the micro changes of topological invariants, the sensitivity of Lyapunov exponent and the evolution of fractal spectrum; the time series prediction of multi-feature fusion improves the accuracy; the dynamic adaptive threshold mechanism combined with the multi-level warning strategy ensures the timeliness and accuracy of the warning. The system forms a technical chain of "data acquisition-feature extraction-state analysis-prediction evaluation-decision output", and each module closely cooperates, the previous output optimizes the subsequent input, the subsequent result feedback guides the previous adjustment, and the maximization of monitoring data is realized. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 FIG. 1 is an embodiment schematic diagram of a generator partial discharge online monitoring system according to the present application; Figure 2 FIG. 2 is an embodiment schematic diagram of a generator partial discharge online monitoring device according to the present application. DETAILED DESCRIPTION

[0015] This invention provides an online monitoring system for generator partial discharge, which enhances the reliability of early warning and improves prediction accuracy. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0016] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the generator partial discharge online monitoring method of the present invention includes: 101. Processing module: Simultaneously acquires multi-point partial discharge signals from the generator stator winding through a multi-channel high-frequency signal acquisition device, and preprocesses the original signals using adaptive noise suppression technology based on wavelet packet decomposition to generate a clean partial discharge signal sequence with optimized signal-to-noise ratio.

[0017] It is understood that the executing entity of this invention can be a generator partial discharge online monitoring device, a terminal, or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0018] It should be noted that by using a high-frequency broadband sensor array arranged at different phases and positions of the generator stator winding, multiple raw partial discharge signals are synchronously acquired at a sampling rate of not less than 100MHz, generating a raw multi-channel signal dataset with time-stamped synchronization information. Multi-dimensional noise characteristic analysis is performed on the original multi-channel signal dataset, including background electromagnetic interference feature extraction, periodic noise identification, and random noise statistical distribution analysis, generating a noise feature matrix containing noise type identification results; Based on the noise feature matrix, the optimal wavelet packet decomposition parameters are determined by an adaptive wavelet packet basis function selection algorithm. Each original partial discharge signal is decomposed into wavelet packet nodes at different resolution scales to generate a wavelet packet decomposition coefficient matrix with time-frequency localization characteristics. Based on the energy distribution characteristics and signal-to-noise ratio evaluation results of each wavelet packet node, a coefficient processing mechanism based on a nonlinear threshold function is established to perform multi-scale adaptive threshold processing on the wavelet packet decomposition coefficient matrix and generate a denoised wavelet packet coefficient reconstruction matrix. Multi-resolution signal reconstruction is performed using the denoised wavelet packet coefficient reconstruction matrix, and a clean partial discharge signal sequence with optimized signal-to-noise ratio and waveform fidelity is generated through signal integrity verification and time-domain waveform optimization.

[0019] 102. Extraction module: Simultaneously performs multifractal detrended fluctuation analysis and phase space reconstruction based on Takens' theorem on the pure partial discharge signal sequence, generating multi-scale fractal feature vectors containing fractal spectrum width, fractal spectrum asymmetry and generalized Hurst exponential distribution, as well as high-dimensional phase space trajectory point clouds reflecting the intrinsic dynamic characteristics of the system.

[0020] It should be noted that an adaptive sliding window segmentation process is performed on the pure partial discharge signal sequence to generate multiple signal analysis segments with overlapping characteristics. Then, a polynomial fitting technique is used to eliminate the trend of each signal analysis segment to generate a set of detrended signal segments. Multi-scale fluctuation analysis is performed on the detrended signal segment set. By calculating the fluctuation function values ​​and their statistical moments at different scales, a scaling behavior feature set describing the self-similarity characteristics of the signal is generated. Based on the scaling behavior feature set, a multi-scale fractal feature vector containing fractal spectrum width, fractal spectrum asymmetry and generalized Hurst exponent distribution is generated through generalized Hurst exponent calculation and multifractal spectrum analysis. The time delay parameter and embedding dimension of the clean partial discharge signal sequence are optimized simultaneously. The optimal reconstruction parameters are determined by the mutual information function method and the pseudo nearest neighbor method, and the optimal combination of phase space reconstruction parameters is generated. By optimizing the combination of phase space reconstruction parameters, a one-dimensional pure partial discharge signal sequence is mapped to a high-dimensional phase space. A high-dimensional phase space trajectory point cloud reflecting the intrinsic dynamic characteristics of the system is generated through a trajectory point reconstruction algorithm. By performing feature alignment and time-stamping on multi-scale fractal feature vectors and high-dimensional phase space trajectory point clouds, a multi-fractal-phase space joint feature dataset with temporal correspondence is generated.

[0021] 103. The analysis module utilizes high-dimensional phase space trajectory point clouds to perform nonlinear dynamic stability analysis based on Lyapunov exponent spectrum and algebraic topological invariant calculation based on continuous homology theory in parallel. This generates a set of dynamic stability indices including the maximum Lyapunov exponent and the rate of change of the correlation dimension, as well as a set of topological features based on the Betti number evolution sequence and topological entropy.

[0022] It should be noted that trajectory matrix reconstruction and neighbor point tracking analysis are performed on high-dimensional phase space trajectory point clouds. By calculating the exponential divergence rate of adjacent trajectory points, Lyapunov exponential spectrum features describing the system's sensitivity to initial conditions are generated. Based on the Lyapunov exponent spectrum characteristics, the maximum Lyapunov exponent and the exponent spectrum distribution characteristics are extracted. At the same time, the correlation dimension of the phase space trajectory and its time change rate are calculated to generate a set of dynamic stability indices containing quantitative indicators of system stability. Multi-scale topological structure analysis is performed on high-dimensional phase space trajectory point clouds. By constructing Vietoris-Rips complex sequences under different scale parameters, continuous cohomological features describing the evolution of the system's topological structure are generated. The evolution sequence of Betti numbers in each dimension is extracted from the persistent homology features, the persistent lifetime distribution and topological entropy value of the topological features are calculated, and a set of topological features reflecting the structural complexity of the system is generated. By performing time alignment and feature correlation analysis on the dynamic stability index set and the topological feature set, a joint dynamic-topological feature matrix with spatiotemporal correspondence is generated.

[0023] 104. The fusion module integrates multi-scale fractal feature vectors, dynamic stability index sets, and topological feature sets based on nonlinear dynamic systems to construct a time series prediction system based on a state-space evolution model, generating quantitative prediction results that include insulation state development trends and remaining lifetime assessments.

[0024] It should be noted that the multi-scale fractal feature vector, dynamic stability index set and topological feature set are subjected to feature alignment and normalization processing. A feature fusion vector reflecting the multi-dimensional state of the generator insulation system is constructed through a nonlinear coupling algorithm, generating a system state feature matrix with spatiotemporal consistency. Based on the system state characteristic matrix, a state-space evolution model is constructed. The time evolution law of the system state is described by a set of nonlinear differential equations, and a state-space prediction model containing state transition characteristics and system dynamic response is generated. Using a state-space prediction model, the system state feature matrix is ​​predicted in multiple steps, and the state evolution trajectory is solved by numerical integration to generate a system state prediction sequence containing future time series. Based on the system state prediction sequence and combined with the historical degradation data of the generator insulation system, an insulation state degradation trajectory model is established. Through the mapping relationship from state space to lifetime space, a quantitative lifetime assessment result containing the remaining lifetime probability distribution and confidence interval is generated. By comprehensively analyzing the system state prediction sequence and quantitative lifetime assessment results, and through trend extrapolation and state evolution analysis, quantitative prediction results containing insulation state development trend curves and key time nodes are generated.

[0025] 105. Output module: Based on the quantitative prediction results and combined with the real-time operating parameters of the generator, the module uses a dynamic adaptive threshold decision algorithm to generate multi-level early warning signals and customized maintenance plans. The early warning signals include three levels: early warning based on changes in topological invariants, mid-term early warning based on dynamic stability indicators, and emergency early warning based on multifractal feature mutations. The maintenance plan includes preventive maintenance time windows and maintenance strategy suggestions based on the prediction results.

[0026] It should be noted that the quantitative prediction results are correlated with the real-time operating parameters of the generator in multiple dimensions. Through the state feature weighted fusion algorithm, a comprehensive state assessment matrix that takes into account the development trend of insulation state and the influence of the operating environment is generated. Based on the comprehensive state evaluation matrix, a dynamic adaptive threshold decision algorithm is adopted to set early warning thresholds with different sensitivities for topological invariant features, dynamic stability indicators and multifractal features, respectively, and generate a dynamic early warning threshold matrix containing three levels of early warning thresholds. Based on the dynamic early warning threshold matrix, multi-level early warning trigger judgment is performed on real-time monitoring features. Through feature anomaly calculation and early warning level mapping, a three-level early warning signal set containing early warning, mid-term early warning and emergency early warning signals is generated. Based on the three-level early warning signal set and quantitative prediction results, combined with the generator's historical maintenance data and equipment operation specifications, a set of maintenance strategies including preventive maintenance time windows and specific maintenance measures suggestions is generated through a maintenance strategy optimization algorithm. The three-level early warning signal set and maintenance strategy set are integrated and optimized, and a customized maintenance plan is generated through the decision support system, which includes an early warning level description, maintenance time suggestion, maintenance measure recommendation and risk assessment report.

[0027] 106. It also includes: establishing a generator insulation state evolution database, storing quantitative prediction results, multi-scale fractal feature vectors, dynamic stability index sets and topological feature sets in time series, and generating a state feature database with complete historical evolution records; Based on the state feature database, a generator insulation state degradation trajectory model is constructed. The key feature evolution mode of insulation degradation is identified through the nonlinear dynamic system inversion algorithm, and a fault mode recognition system containing a typical fault mode feature library and degradation path map is generated. Using a fault mode recognition system, pattern matching and trend comparison are performed on real-time monitoring data. Through similarity analysis and abnormal pattern detection, a pattern recognition report containing fault type identification results and evolution stage judgment is generated. Based on the pattern recognition report, combined with generator design parameters and operating history, a life prediction correction model based on nonlinear dynamics is established. Through deviation analysis between real-time data and historical patterns, a prediction quality report containing the confidence level of the prediction results and reliability assessment is generated. By comprehensively analyzing pattern recognition reports and prediction quality reports, and optimizing early warning signals and maintenance plans through decision fusion algorithms, a comprehensive monitoring report is generated that includes fault warnings, lifespan predictions, and maintenance recommendations.

[0028] In this embodiment of the invention, the beneficial effects are as follows: By establishing an insulation state evolution database and a fault mode recognition system, continuous accumulation and self-optimization of monitoring data are achieved. The degradation trajectory model constructed using historical data provides a comparison benchmark for real-time monitoring. Fault identification accuracy is improved through pattern matching and anomaly detection. The reliability of prediction results is assessed by combining a life prediction correction model. The final comprehensive monitoring report integrates multi-dimensional analysis results, forming a complete closed loop of "data accumulation - pattern learning - real-time diagnosis - prediction correction - decision optimization". This effectively improves the system's adaptive capability and the accuracy of long-term monitoring, providing a scientific basis for continuous optimization of generator condition assessment and maintenance decisions.

[0029] Figure 2 This is a schematic diagram of the structure of an online generator partial discharge monitoring device 200 provided in an embodiment of the present invention. This online generator partial discharge monitoring device 200 can vary considerably due to differences in configuration or performance. The device 200 includes a transmitter 201, a receiver 202, and a processor 203. The processor 203 can also be a controller. Figure 2 The device is referred to as "controller / processor 203". Optionally, the device 200 may also include a modem processor 205, wherein the modem processor 205 may include an encoder 206, a modulator 207, a decoder 208, and a demodulator 209.

[0030] In one example, transmitter 201 modulates (e.g., analog-to-analog conversion, filtering, amplification, and up-conversion, etc.) the output sample and generates an uplink signal, which is transmitted via an antenna to an access network device. On the downlink, the antenna receives the downlink signal transmitted by the access network device. Receiver 202 modulates (e.g., filtering, amplification, down-conversion, and digitization, etc.) the signal received from the antenna and provides an input sample. In modem processor 205, encoder 206 receives traffic data and signaling messages to be transmitted on the uplink and processes (e.g., formatting, encoding, and interleaving) the traffic data and signaling messages. Modulator 207 further processes (e.g., symbol mapping and modulation) the encoded traffic data and signaling messages and provides an output sample. Demodulator 209 processes (e.g., demodulates) the input sample and provides a symbol estimate. Decoder 208 processes (e.g., deinterleaving and decoding) the symbol estimate and provides decoded data and signaling messages to device 200. Encoder 206, modulator 207, demodulator 209, and decoder 208 can be implemented by a combined modem processor 205. These units process data according to the radio access technology used by the radio access network (e.g., LTE and other evolved systems access technologies). It should be noted that when device 200 does not include modem processor 205, the aforementioned functions of modem processor 205 can also be performed by processor 203.

[0031] The processor 203 controls and manages the operation of the device 200, and is used to execute the processing procedures performed by the device 200 in the above embodiments of this disclosure. For example, the processor 203 is also used to execute various steps of the transmitting or receiving device in the above method embodiments, and / or other steps of the technical solutions described in the embodiments of this disclosure.

[0032] Furthermore, the device 200 may also include a memory 204 for storing program code and data for the device 200.

[0033] Understandable, Figure 2 Only a simplified design of device 200 is shown. In practical applications, device 200 can include any number of transmitters, receivers, processors, modem processors, memory, etc., and all devices that can implement the embodiments of this disclosure are within the protection scope of the embodiments of this disclosure.

[0034] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A generator partial discharge on-line monitoring system characterized by, The generator partial discharge on-line monitoring system comprises: a processing module, which acquires multi-point partial discharge signals through an acquisition device, pre-processes the signals by using an adaptive noise suppression technology, and generates a pure partial discharge signal sequence; an extraction module, which analyzes and phase space reconstructs the pure partial discharge signal sequence, generates a multi-scale fractal feature vector and a high-dimensional phase space trajectory point cloud; an analysis module, which utilizes the high-dimensional phase space trajectory point cloud to perform nonlinear dynamics stability analysis and algebraic topological invariant calculation, generates a dynamics stability index set and a topological feature set; a fusion module, which fuses the multi-scale fractal feature vector, the dynamics stability index set and the topological feature set, constructs a time series prediction system, and generates a quantitative prediction result; an output module, which generates a multi-level early warning signal and a customized maintenance scheme according to the quantitative prediction result and in combination with generator parameters.

2. A generator partial discharge on-line monitoring system according to claim 1, characterised in that, It comprises: a high-frequency broadband sensor array is used to acquire multiple original partial discharge signals and generate an original multi-channel signal data set; the original multi-channel signal data set is analyzed to generate a noise feature matrix; based on the noise feature matrix, optimal wavelet packet decomposition parameters are determined, the original partial discharge signal is decomposed into wavelet packet nodes, and a wavelet packet decomposition coefficient matrix is generated; a coefficient processing mechanism is established according to the energy distribution characteristics and the signal-to-noise ratio evaluation results of each wavelet packet node, the wavelet packet decomposition coefficient matrix is processed, and a wavelet packet coefficient reconstruction matrix is generated; the wavelet packet coefficient reconstruction matrix is used for reconstruction to generate a pure partial discharge signal sequence.

3. A generator partial discharge on-line monitoring system according to claim 1, wherein, It comprises: the pure partial discharge signal sequence is processed to generate a signal analysis segment, and each signal analysis segment is eliminated to generate a signal segment set; the signal segment set is analyzed by calculating the fluctuation function value and its statistical moments to generate a scale behavior feature set; based on the scale behavior feature set, multi-scale fractal feature vectors are generated by calculation and analysis; the pure partial discharge signal sequence is calculated to determine optimal reconstruction parameters and generate an optimized combination of phase space reconstruction parameters; the one-dimensional pure partial discharge signal sequence is mapped to a high-dimensional phase space by using the optimized combination of phase space reconstruction parameters to generate a high-dimensional phase space trajectory point cloud; the multi-scale fractal feature vector and the high-dimensional phase space trajectory point cloud are processed to generate a multi-fractal-phase space joint feature data set.

4. A generator partial discharge on-line monitoring system according to claim 3, wherein, It comprises: the high-dimensional phase space trajectory point cloud is reconstructed and analyzed to calculate the exponential divergence rate and generate Lyapunov exponent spectrum features; based on the Lyapunov exponent spectrum features, Lyapunov exponents and exponent spectrum distribution features are extracted, the correlation dimension and its time variation rate are calculated, and a dynamics stability index set is generated; the high-dimensional phase space trajectory point cloud is analyzed to construct a Vietoris-Rips complex sequence and generate persistent homology features; evolution sequences are extracted from the persistent homology features, the lifetime distribution and topological entropy value of topological features are calculated, and a topological feature set is generated; the dynamics stability index set and the topological feature set are analyzed to generate a dynamics-topology joint feature matrix.

5. A generator partial discharge on-line monitoring system according to claim 1, wherein, It comprises: The multi-scale fractal feature vector, the dynamic stability index set and the topological feature set are processed to construct a feature fusion vector and generate a system state feature matrix; Based on the system state feature matrix, a model is constructed to describe the time evolution law and generate a state space prediction model; The system state feature matrix is predicted using the state space prediction model to generate a system state prediction sequence; Based on the system state prediction sequence, a trajectory model is established by combining the data of the generator system to generate a quantitative life assessment result through a mapping relationship; The system state prediction sequence and the quantitative life assessment result are analyzed to generate a quantitative prediction result.

6. A generator partial discharge on-line monitoring system according to claim 5, wherein, It includes: The quantitative prediction result is analyzed with the generator parameters to generate a comprehensive state assessment matrix; Based on the comprehensive state assessment matrix, different early warning threshold values are set for the topological invariant features, the dynamic stability indexes and the multi-fractal features using a decision algorithm to generate a dynamic early warning threshold matrix; According to the dynamic early warning threshold matrix, multi-level early warning trigger judgments are made to generate a three-level early warning signal set through calculation and mapping; Based on the three-level early warning signal set and the quantitative prediction result, a maintenance strategy set is generated by combining the maintenance data and the equipment operation specifications; The three-level early warning signal set and the maintenance strategy set are optimized to generate a customized maintenance plan.

7. A generator partial discharge on-line monitoring system according to claim 1, wherein, It also includes: A generator evolution database is established to store the quantitative prediction result, the multi-scale fractal feature vector, the dynamic stability index set and the topological feature set in time sequence to generate a state feature database; Based on the state feature database, a trajectory model is constructed to generate a fault mode recognition system through a key feature evolution mode; The data is matched and compared using the fault mode recognition system to generate a mode recognition report; According to the mode recognition report, a life prediction correction model is established by combining the generator parameters and the history to generate a prediction quality report through data and analysis; The mode recognition report and the prediction quality report are analyzed to generate a comprehensive monitoring report.

Images