Online monitoring and fault self-diagnosis method for excitation system of hydroelectric generating set

Abstract

The invention relates to the technical field of power system automation and control, and discloses a hydroelectric generating set excitation system on-line monitoring and fault self-diagnosis method comprising the following steps: S1, collecting real-time operation data of an excitation system; s2, filtering the acquired data, and segmenting the acquired data into time slices with fixed lengths; s3, constructing a multi-modal tensor model; s4, nonlinear modeling and bifurcation point detection are carried out based on the power system; s5, performing fault classification on the tensor core features by using a classification model; s6, dynamically adjusting a classification threshold value, a bifurcation point sensitivity parameter and a tensor decomposition path of the fault classification model based on a real-time diagnosis result; and S7, outputting a fault diagnosis result to a visual interface. According to the method, the technical scheme of multi-modal tensor modeling and dynamic bifurcation point detection is adopted, the defect that a complex fault mode cannot be accurately recognized due to data feature loss is overcome, and the comprehensiveness and reliability of fault detection are remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of power system automation and control, and in particular to an online monitoring and fault self-diagnosis method for an excitation system of a hydroelectric generator set. Background Art

[0002] As the core component of the generator operation, the excitation system of the hydroelectric generator set is mainly responsible for adjusting the magnetic field strength of the generator, maintaining the stability of the terminal voltage and the dynamic balance of the power grid. However, due to the complex operating environment of the excitation system, it is often affected by load changes, system failures or external environment interference, which can easily cause problems such as low excitation limitation, strong excitation limitation or volt-hertz protection, and even cause system instability or failure. Therefore, the operation monitoring and fault diagnosis technology for the excitation system is particularly important.

[0003] The patent with announcement number CN105823985B provides an online monitoring and evaluation system for the generator excitation system based on WAMS dynamic data. This technology collects dynamic data of the excitation system operating parameters and combines model evaluation to perform online analysis of the generator's operating status. The main feature of this method is to use WAMS dynamic data combined with excitation regulator protection logic and stability analysis to achieve real-time operation monitoring of the generator excitation system.

[0004] However, although this patented technical solution can realize dynamic monitoring and evaluation of a single operating state, it has the following major shortcomings:

[0005] This technology focuses more on the performance analysis of single variable parameters of the generator excitation system and lacks the ability to diagnose the coupling characteristics between multiple variables. Especially when multiple fault modes are superimposed or complex operating status changes occur, its diagnostic accuracy is low.

[0006] This technology uses fixed parameters to evaluate the diagnostic logic and does not fully consider the changes in the operating characteristics of the excitation system in a dynamic environment. This results in poor adaptability of the diagnostic model under complex working conditions and is prone to misdiagnosis or missed diagnosis. Summary of the invention

[0007] In order to make up for the above shortcomings, the present invention provides an online monitoring and fault self-diagnosis method for the excitation system of a hydroelectric generator set, aiming to improve the problems of insufficient multi-fault coupling mode diagnosis capability, poor real-time and adaptability, and single fault classification result output in the prior art.

[0008] In a first aspect, the present invention provides the following technical solution: a method for online monitoring and fault self-diagnosis of an excitation system of a hydroelectric generator set, comprising the following steps:

[0009] S1. Collect real-time operation data of the excitation system, including excitation current, excitation voltage, terminal voltage, frequency, active power, reactive power and environmental parameters;

[0010] S2, filtering the collected data to remove high-frequency noise from the data, and dividing the data into time segments of fixed length based on a time sliding window;

[0011] S3, construct a multi-modal tensor model, convert the collected multi-dimensional data into tensor form, and extract multi-fault coupling characteristics;

[0012] S4. Based on the nonlinear modeling and bifurcation point detection of the power system, the stability of the system operation state is judged, and a fault warning signal is issued when the characteristic value change trend shows that the system is approaching the bifurcation point;

[0013] S5. Use the classification model to classify the tensor core features into fault categories and output the diagnosis results of single fault or multiple fault coupling modes;

[0014] S6. Dynamically adjust the classification threshold, bifurcation point sensitivity parameter and tensor decomposition path of the fault classification model based on real-time diagnosis results to optimize fault feature extraction performance and diagnosis accuracy;

[0015] S7. Output the fault diagnosis results to the visualization interface to display the system operation status, multi-fault coupling characteristics and health assessment results.

[0016] Preferably, the multimodal tensor quantization model step in step S3 includes the following process:

[0017] S31, constructing a multidimensional data tensor according to the real-time collected excitation current, excitation voltage, terminal voltage, frequency, active power, reactive power and environmental parameters;

[0018] S32. Decompose the tensor into a core tensor and a low-dimensional projection matrix, wherein the core tensor is used to extract the global characteristics of the multi-fault coupling mode, and the low-dimensional projection matrix reduces the data complexity by mapping the high-dimensional data to the low-dimensional space.

[0019] Preferably, the nonlinear modeling of the power system in step S4 includes:

[0020] S41, constructing a nonlinear dynamic equation of the excitation system and establishing a state vector model of the system, wherein the state vector includes the excitation current, the excitation voltage and the frequency;

[0021] S42, calculating eigenvalues ​​based on the Jacobian matrix of the power system model to determine the stability of the system operation state;

[0022] S42. When the real part of the system's eigenvalue changes from a negative value to a positive value, it is determined that the system is close to a bifurcation point, and a fault warning signal is triggered.

[0023] Preferably, the fault classification step in step S5 includes:

[0024] S51, extracting the core features after tensor decomposition as input features;

[0025] S52. Fault mode identification is performed using a nonlinear classification model, and the classification results include a single fault mode and a multi-fault coupling mode.

[0026] Preferably, the step S6 adjusts the fault diagnosis model parameters in the following manner:

[0027] a. Dynamically adjust the classification threshold of the fault classification model;

[0028] b. Dynamically adjust the bifurcation point sensitivity parameters;

[0029] c. Dynamically adjust the tensor decomposition path;

[0030] d. During the parameter adjustment process, the optimization strategy is updated in real time to meet different operating environments and load conditions.

[0031] Preferably, the bifurcation point detection in step S6 includes the following process:

[0032] S61, calculating the Jacobian matrix of the power system according to the real-time operation data;

[0033] S62, by calculating the eigenvalue of the Jacobian matrix, determining whether the real part of the eigenvalue changes from a negative value to a positive value;

[0034] S63. When it is detected that the system is approaching a bifurcation point, a fault warning signal is triggered, and a bifurcation characteristic value curve of real-time operation data is output.

[0035] Preferably, the visual output of the diagnosis result in step S7 includes the following contents:

[0036] Output multi-fault coupling characteristic diagram to show the correlation between faults;

[0037] Dynamically update the system operation health score and quantify the score results;

[0038] Output operation status optimization suggestions, including excitation system parameter adjustment suggestions and maintenance strategies.

[0039] In a second aspect, the present invention provides the following technical solution, an online monitoring and fault self-diagnosis system for an excitation system of a hydroelectric generator set, the system comprising:

[0040] A data acquisition module is used to collect the operating parameters and environmental parameters of the excitation system and transmit the collected data to the data processing module;

[0041] The data processing module is used for filtering and denoising, time series sliding window segmentation and tensor modeling, and transmits the generated multi-dimensional tensor to the dynamic modeling and bifurcation point detection module;

[0042] Dynamic modeling and bifurcation point detection module, used to establish a nonlinear dynamic model of the system, detect bifurcation points in real time and transmit warning signals and bifurcation characteristic values ​​to the fault diagnosis module;

[0043] Fault diagnosis module, which is used to extract tensor core features and classify single faults and multiple faults, and transmit the diagnosis results to the visualization module;

[0044] Adaptive optimization module, used to optimize diagnostic logic by dynamically adjusting classification thresholds, bifurcation point detection sensitivity, and tensor decomposition paths;

[0045] The visualization module is used to display fault classification results, multi-fault coupling relationships and operation status evaluation results.

[0046] In the third aspect, the invention provides the following technical solution: a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set when executing the computer program.

[0047] In a fourth aspect, the present invention provides the following technical solution: a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the above-mentioned method for online monitoring and fault self-diagnosis of an excitation system of a hydroelectric generator set is implemented.

[0048] The present invention has the following beneficial effects:

[0049] 1. In the present invention, multi-modal tensor modeling and dynamic bifurcation point detection technical solutions are adopted to achieve the technical effect of accurate extraction of multi-fault coupling characteristics of the excitation system and real-time early warning. Compared with the existing solutions that only rely on single variable analysis or low-dimensional fault modeling, the present invention solves the problem of being unable to accurately identify complex fault modes due to loss of data features, and significantly improves the comprehensiveness and reliability of fault detection.

[0050] 2. In the present invention, by introducing an adaptive optimization module and dynamically adjusting the classification threshold and bifurcation point sensitivity parameters, efficient adaptation to different operating environments and complex working conditions is achieved. Compared with the situation in the prior art where fixed parameter settings lead to limited diagnostic accuracy, the present invention effectively solves the problem of insufficient flexibility of traditional diagnostic systems when facing changes in working conditions, ensuring the continued high efficiency and accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a general method flow chart of the on-line monitoring and fault self-diagnosis method of the excitation system of a hydroelectric generator set proposed by the present invention;

[0052] Figure 2 It is a detailed flow chart of S3 of the method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set proposed by the present invention;

[0053] Figure 3 It is a detailed flow chart of S4 of the method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set proposed by the present invention;

[0054] Figure 4 It is a detailed flow chart of S5 of the method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set proposed by the present invention;

[0055] Figure 5 It is a detailed flow chart of S6 of the method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set proposed by the present invention;

[0056] Figure 6 This is a schematic diagram of the online monitoring and fault self-diagnosis system for the excitation system of a hydroelectric generator set proposed in the present invention. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] Embodiment 1

[0059] Reference Figure 1-Figure 5 In a first embodiment of the present invention, the present invention provides an online monitoring and fault self-diagnosis method for an excitation system of a hydroelectric generator set, comprising the following steps:

[0060] S1. Collect real-time operation data of the excitation system, including excitation current, excitation voltage, terminal voltage, frequency, active power, reactive power and environmental parameters;

[0061] Specifically, step S1 is mainly responsible for real-time acquisition of the operating data of the excitation system, which is an important basis for subsequent processing, modeling and diagnosis. By acquiring key parameters in real time, it can effectively support the extraction of multi-fault coupling characteristics and the optimization of diagnostic accuracy, while providing sufficient data support for the nonlinear analysis of the power system. In general, the data to be collected include electrical parameters and environmental parameters that reflect the dynamic characteristics of the excitation system to ensure comprehensive monitoring and analysis of system operation.

[0062] In this embodiment, the collected parameters mainly include the following:

[0063] Excitation current: refers to the DC current provided by the excitation system to the generator rotor, which is used to adjust the magnetic field strength of the generator.

[0064] Excitation voltage: reflects the output voltage of the excitation device and is an important basis for adjusting the working state of the excitation system.

[0065] Generator terminal voltage: The AC voltage output at the generator terminal is directly related to the regulation performance of the excitation system.

[0066] Frequency: The frequency of electricity output by the generator, reflecting the stability of system operation.

[0067] Active power and reactive power: characterize the power output status of the generator and are closely related to the stable operation of the power system.

[0068] Environmental parameters: including the operating temperature of the excitation equipment, external ambient temperature and vibration intensity, etc., providing additional information on the health status of the equipment.

[0069] In some embodiments, in order to achieve accurate parameter acquisition, high-precision sensors are installed in the excitation system, such as:

[0070] The current sensor is used to monitor the excitation current in real time and transmit the signal to the data acquisition unit; the voltage sensor is used to measure the excitation voltage and the terminal voltage; the power metering device monitors the active power and reactive power; the environmental sensor is used to collect the temperature, humidity and vibration signals of the external environment.

[0071] In the specific implementation, the sensors are reasonably deployed with the main components of the excitation system. For example: the excitation current sensor can be installed at the output end of the excitation rectifier to directly collect the excitation current data; the excitation voltage sensor is set at both ends of the excitation winding to accurately measure the excitation voltage; the machine-end voltage sensor is usually connected to the output terminal of the generator to collect the AC voltage signal; the vibration sensor is installed at the bracket of the excitation equipment to obtain the vibration signal. In order to ensure the real-time and accuracy of the data, a high sampling frequency data acquisition device is used in this embodiment. Generally, the sampling frequency is set between 1kHz and 10kHz to capture rapidly changing electrical signals and dynamically fluctuating operating conditions.

[0072] As an option, the data acquisition device is connected to the central processing unit via an industrial communication protocol, such as a CAN bus or Modbus protocol. Through these protocols, the collected real-time data can be transmitted to the data processing module within milliseconds. In addition, in a distributed system, some embodiments also use wireless communication technology (such as LoRa or Wi-Fi) to adapt to complex field wiring requirements.

[0073] Specifically, in order to eliminate noise and interference in the acquisition process, the present invention also adds hardware filtering circuits and software filtering algorithms in the data acquisition process:

[0074] The hardware filtering circuit uses a low-pass filter to suppress high-frequency noise;

[0075] Software filtering algorithms use sliding average or Kalman filtering to smooth the signal.

[0076] In another possible implementation, the collected parameters will be stored in a local storage unit as historical data for subsequent analysis. At the same time, the real-time data is transmitted to the edge computing device through the data transmission module for subsequent data segmentation, modeling and analysis.

[0077] In some embodiments, in order to improve the reliability and redundancy of data, the acquisition system also introduces multi-sensor fusion technology. By comprehensively calculating the acquisition results of multiple sensors, the impact of single sensor failure or error on data accuracy can be effectively reduced. For example: multiple acquisition results of the same variable (such as excitation current) are fused using the weighted average method; if a sensor fails, it is compensated by the redundant data of other sensors.

[0078] As a possible extension, the real-time collected data can not only be used for subsequent modeling and diagnosis, but also as input for system operation optimization. For example, the excitation current can be dynamically adjusted according to the frequency fluctuation data to improve the stability of system operation.

[0079] In general, step S1 provides basic data support for the entire fault diagnosis method. These collected data not only cover the electrical characteristics of the excitation system, but also include supplementary information on the equipment operating environment.

[0080] S2, filtering the collected data to remove high-frequency noise from the data, and dividing the data into time segments of fixed length based on a time sliding window;

[0081] Specifically, in this embodiment, the filtering process mainly adopts a combination of hardware filtering and software filtering to accurately retain the effective components of the data. The hardware filter preferentially performs low-pass filtering in the data acquisition module to suppress local frequency noise in the signal. Software filtering is used as a subsequent supplement to data processing, and commonly used methods include sliding average filtering and adaptive Kalman filtering. Among them, the sliding average filter performs multiple iterative calculations on the data, which can smooth out small fluctuations generated during the acquisition process; the Kalman filter is based on a dynamic model and can adaptively optimize the filtering parameters in combination with noise characteristics, which is particularly suitable for excitation systems with complex dynamic characteristics.

[0082] Specifically, the filtering operation can be expressed as follows:

[0083]

[0084] Among them, y i is the filtered data value, x j is the raw data value collected for the jth time, and n is the size of the sliding window, which is usually set between 5 and 50 according to the sampling frequency and the dynamic characteristics of the system. When using Kalman filtering, the state equation and measurement equation are defined as:

[0085] x k =Ax k-1 +w k

[0086] z k =Hx k +v k

[0087] Among them, x k is the system state, z k is the measured value, A and H are the state transfer matrix and observation matrix respectively, w k and v k is the system noise and the measurement noise, both of which are assumed to be Gaussian distribution. In some embodiments, in order to improve the real-time performance of filtering, the spectrum distribution of the signal can also be analyzed in combination with the fast Fourier transform (FFT). By accurately identifying the noise frequency band, the filtering parameters can be further optimized, thereby better retaining the main components of the signal.

[0088] After the filtering process is completed, the time sliding window segmentation step is entered. Generally, the real-time acquired data is segmented at fixed time intervals to form uniform time segments for subsequent tensor modeling and feature extraction. As an option, the length and step size of the sliding window can be adjusted according to the dynamic characteristics of the system.

[0089] Specifically, the window length is defined as T w , the step size is defined as T s In a possible implementation, the window length is set to 200 ms and the step length is set to 50 ms to ensure the temporal continuity and data overlap rate of the segments.

[0090] In order to more clearly describe the principle of time window segmentation, let the real-time sampling frequency be f s , the number of data points in a time segment can be expressed as:

[0091] N w =T w ·f s

[0092] The number of data points corresponding to the sliding step is:

[0093] N s =T s ·f s

[0094] For example, when the sampling frequency is 1kHz, the window length is 200ms, and the step length is 50ms, each time segment contains 200 data points, and there are 150 data points overlapping between adjacent time segments. This overlapping design can effectively enhance the temporal continuity of features. In some extended embodiments, the sliding window segmentation can also be adaptively adjusted in combination with data features. For example, when the system operating state changes significantly, the length of the sliding window is dynamically shortened to capture abnormal signals more promptly; when the system is running stably, the window length can be appropriately increased to improve computing efficiency.

[0095] As an improvement, time segments can also be stored as a hierarchical structure to adapt to subsequent tensor modeling. For example, the data of each time segment is divided into a basic electrical parameter layer and an extended environmental parameter layer, which represent the main operating status and the influencing factors of the external environment respectively. This hierarchical design can more intuitively reflect the multimodal characteristics of the data while reducing data redundancy in the modeling process.

[0096] S3, construct a multi-modal tensor model, convert the collected multi-dimensional data into tensor form, and extract multi-fault coupling characteristics;

[0097] S31, constructing a multidimensional data tensor according to the real-time collected excitation current, excitation voltage, terminal voltage, frequency, active power, reactive power and environmental parameters;

[0098] S32. Decompose the tensor into a core tensor and a low-dimensional projection matrix, wherein the core tensor is used to extract the global characteristics of the multi-fault coupling mode, and the low-dimensional projection matrix reduces the data complexity by mapping the high-dimensional data to the low-dimensional space.

[0099] Specifically, in this embodiment, a multimodal tensor is constructed based on time-segmented data. Specifically, the tensor model construction process includes multivariate data integration, tensor structure design, and high-order data representation. The dimension of the constructed tensor T is determined by the number of variables in the system and the length of the time segment. For example, when the number of variables is 6 (including excitation current, excitation voltage, machine-end voltage, frequency, active power, and reactive power), and each time segment contains 200 data points, the dimension of the tensor is 6×200. In some embodiments, in order to further enrich the expressive power of the model, environmental parameters (such as equipment vibration and operating temperature) can also be used as additional tensor dimensions.

[0100] In general, a tensor is defined as follows:

[0101]

[0102] Among them, t ijk is the element value of the tensor, N v Indicates the number of variables, N t Indicates the length of the time segment, N e Represents the dimension of environmental parameters (if environmental data exists). For example, for a time segment, the sampling value of the excitation current at a certain moment can be represented as an element of the tensor. In order to extract the multi-fault coupling characteristics, the constructed high-dimensional tensor needs to be reduced in dimensionality.

[0103] In this embodiment, the Zhangli decomposition technique is used to decompose the original tensor T into a core tensor C and several projection matrices:

[0104]

[0105] in:

[0106] is the core tensor, representing the global features of multiple faults after dimensionality reduction;

[0107] U (n) It is the n-th dimension projection matrix, used to map the tensor to a lower-dimensional space.

[0108] Specifically, the core tensor The global information and high-order coupling characteristics of the tensor are preserved. For example, when the fault coupling of excitation current overshoot and frequency fluctuation occurs in a certain time segment, its characteristics will be concentrated on certain dimensions of the core tensor.

[0109] In some embodiments, the projection matrix U (1) Mainly used for variable dimension reduction, projection matrix U (2) Used for time segment dimensionality reduction. Variable dimensionality reduction can reduce the impact of redundant variables on the model, and time segment dimensionality reduction can focus on the key time window. In a possible implementation, the dimension of the projection matrix is ​​determined by singular value decomposition (SVD) to ensure the dimensionality reduction effect of the model.

[0110] In the process of implementing tensor decomposition, orthogonal constraints can also be used to ensure the orthogonality of the projection matrix, thereby avoiding repeated expression of information. The optimization objective of the projection matrix can be expressed as:

[0111]

[0112] Among them, ||·|| is the Frobenius norm of the tensor, which is used to measure the similarity between the original tensor and the decomposed tensor.

[0113] As an option, if the dimension of the core tensor is too large, the core tensor C can be further sparsely processed to enhance the expressiveness of the features. For example, by adding a sparse regularization term, only a few feature dimensions in the core tensor can be restricted to have significant values, thereby improving the resolution of multi-fault features.

[0114] In a possible extension, the present invention also supports feature visualization of tensor decomposition results. For example, through the two-dimensional projection of the core tensor, the coupling relationship between different faults can be intuitively displayed, thereby providing a reference for subsequent fault classification. The core tensor is used as an output to provide support for subsequent bifurcation point detection and fault classification. In some embodiments, the core tensor can also be directly used as an input to monitor the changing trend of fault characteristics in real time.

[0115] S4. Based on the nonlinear modeling and bifurcation point detection of the power system, the stability of the system operation state is judged, and a fault warning signal is issued when the characteristic value change trend shows that the system is approaching the bifurcation point;

[0116] S41, constructing a nonlinear dynamic equation of the excitation system and establishing a state vector model of the system, wherein the state vector includes the excitation current, the excitation voltage and the frequency;

[0117] S42, calculating eigenvalues ​​based on the Jacobian matrix of the power system model to determine the stability of the system operation state;

[0118] S42. When the real part of the system's eigenvalue changes from a negative value to a positive value, it is determined that the system is close to a bifurcation point, and a fault warning signal is triggered.

[0119] Specifically, this step inherits the multi-fault coupling features extracted in step S3, takes the core features of the tensor decomposition output as input, builds a nonlinear dynamic model of the excitation system, calculates the dynamic changes of key parameters, and analyzes its stability trend. In general, the stability of the excitation system is affected by many factors, including electrical parameters, load disturbances, and external environmental changes. Through bifurcation point detection technology, the critical points of the system operation state can be identified in advance, providing a basis for fault classification and operation optimization.

[0120] In this embodiment, the nonlinear dynamic modeling of the excitation system is based on the state space description. Specifically, the state equation of the system can be expressed as:

[0121]

[0122] Among them, x(t) is the system state vector, which represents the main dynamic variables of the system (such as excitation current, excitation voltage, frequency, etc.); u(t) is the input vector, which represents the external control or disturbance input; A and B are linear coefficient matrices, which describe the basic structure of the system; N(x) is a nonlinear term, which reflects the nonlinear characteristics of the system. In order to analyze the stability of the system, the Jacobian matrix of the system is calculated in this embodiment:

[0123]

[0124] Where J represents the linearization matrix of the system in a specific state, and its eigenvalue can reflect the local stability of the system. When the real part of an eigenvalue of the Jacobian matrix changes from negative to positive, the system may undergo Hopf bifurcation and enter a periodic oscillation state. This state is usually associated with unstable operation of the excitation system.

[0125] In general, the detection of bifurcation points includes the following key steps:

[0126] 1. Calculate the Jacobian matrix of the system in its current state,

[0127] 2. Solve for the eigenvalues ​​of the Jacobian matrix and determine the sign of its real part.

[0128] 3. When it is detected that the real part of the eigenvalue changes from a negative value to a positive value, a bifurcation point warning signal is triggered.

[0129] As an option, in order to improve the accuracy of bifurcation point detection, the calculation process of the Jacobian matrix is ​​optimized in this embodiment. For example, when the system state changes rapidly, the calculation frequency of the Jacobian matrix is ​​increased; when the system state is relatively stable, the calculation frequency is appropriately reduced to save computing resources. In addition, the interpolation method can be used to estimate the eigenvalue change trend between system states to further improve the detection sensitivity of the bifurcation point.

[0130] Specifically, in one possible implementation, the optimization goal of bifurcation point detection is to minimize the error and calculation time, and the formula is:

[0131] min(||J calculated -J true ||+λ·T compute )

[0132] Among them, J calculated is the calculated Jacobian matrix, J t rue is the theoretical value, ||·|| represents the Frobenius norm of the matrix, T compute is the calculation time, and λ is the time weight factor.

[0133] In some embodiments, in order to achieve long-term stability evaluation of the excitation system, the present invention further combines Lyapunov stability theory.

[0134] By constructing the Lyapunov function:

[0135] V(x)=x T Px

[0136] Where P is a symmetric positive definite matrix, which can verify the global stability of the system. When the Lyapunov function satisfies When , the system is in a stable state; when As an improvement, in order to improve the adaptability of bifurcation point detection to different operating environments, the present invention introduces a sensitivity analysis method.

[0137] By calculating the sensitivity coefficients of key parameters (such as excitation current and excitation voltage), the impact of these parameters on system stability can be quantified. For example:

[0138]

[0139] Among them, S i is the sensitivity coefficient, λ is the characteristic value, p i is the parameter value. The sensitivity analysis results can be used to dynamically adjust the threshold of bifurcation point detection to avoid false alarms caused by parameter fluctuations.

[0140] In another possible implementation, the results of bifurcation point detection can also be displayed in real time through a visual interface. For example, by drawing an eigenvalue trajectory graph to show the change of eigenvalues ​​over time; by using a health index curve to quantify the stability level of the system.

[0141] S5. Use the classification model to classify the tensor core features into fault categories and output the diagnosis results of single fault or multiple fault coupling modes;

[0142] S51, extracting the core features after tensor decomposition as input features;

[0143] S52. Fault mode identification is performed using a nonlinear classification model, and the classification results include a single fault mode and a multi-fault coupling mode.

[0144] Specifically, this step inherits the tensor core features extracted in step S3 and the warning signal of bifurcation point detection in step S4, and further analyzes and classifies the system status by combining the data-driven fault classification model. In general, the excitation system may have various types of faults, including low excitation limit, strong excitation limit, volt-hertz protection triggering, etc. These faults may occur alone or overlap each other during operation. Therefore, accurate classification of these complex patterns is the key to improving the system's diagnostic capabilities.

[0145] In this embodiment, the input of fault classification is the core feature matrix after tensor decomposition The core feature matrix retains the global information of the system operation status, while significantly reducing the data dimension, which is convenient for the training and reasoning of the classification model. In general, the fault classification process can adopt a multi-classification model based on machine learning, such as support vector machine (SVM) or decision tree model.

[0146] Specifically, the support vector machine model achieves classification of different fault categories by constructing a hyperplane in a high-dimensional feature space. The core optimization goal of the classification model is to maximize the interval of the classification boundary, and its mathematical expression is as follows:

[0147]

[0148] Under the constraints:

[0149] y i (w·x i +b)≥1,i=1,2,…,N

[0150] Among them, w is the normal vector of the classification hyperplane, b is the bias, and x i is the feature vector of the i-th sample, y iis the corresponding classification label, and N is the number of samples. In some embodiments, in order to improve the adaptability of fault classification, a kernel function can be introduced to map features from the original space to a high-dimensional feature space, thereby processing nonlinear fault modes. Common kernel functions include radial basis kernel function (RBF):

[0151] K(x i ,x j )=exp(-γ||x i -x j || 2 )

[0152] Among them, γ is the parameter of the kernel function, which is used to control the width of the kernel function.

[0153] As an option, in order to further improve the classification accuracy of the multi-fault coupling mode, this embodiment may also adopt an ensemble learning method, such as a random forest model. The random forest constructs multiple decision tree classifiers and votes on the classification results to obtain a more robust fault classification result.

[0154] Specifically, the construction process of the random forest model includes:

[0155] Randomly extract subsamples from the training sample set to generate multiple decision tree classifiers;

[0156] In each decision tree, some features are randomly selected as the basis for splitting to avoid overfitting of a single feature;

[0157] The classification results of all decision trees are weighted and voted on, and the fault category is finally output.

[0158] In one possible implementation, the classification results are not limited to a single fault category, but can also output a probability distribution of fault categories to assess the uncertainty of the classification. For example, for a set of Tensor Core features, the classification model may output the following probability results:

[0159] Low reward limit: 70%

[0160] Enforcement limit: 20%

[0161] Volts Hertz protection trigger: 10%

[0162] In some embodiments, the training data of the classification model can be derived from historical operating data and fault simulation data. By comprehensively considering actual operating conditions and extreme operating conditions, the generalization ability and stability of the classification model can be improved.

[0163] As an improvement, in order to enhance the adaptability of the classification model to real-time data, this embodiment supports online updating of the classification model. For example, after a new fault sample is introduced during system operation, the classification model can be dynamically adjusted through incremental training to ensure the continuous reliability of the classification results.

[0164] In another possible implementation, the results of the classification model can also be combined with the early warning signal of the bifurcation point detection to improve the accuracy of the classification. For example, when the bifurcation point detection indicates that the system is close to an unstable state, the fault mode related to the frequency fluctuation is given priority; when the bifurcation point detection indicates that the system is in the low excitation area, the low excitation limit fault is detected first.

[0165] S6. Dynamically adjust the classification threshold, bifurcation point sensitivity parameter and tensor decomposition path of the fault classification model based on real-time diagnosis results to optimize fault feature extraction performance and diagnosis accuracy;

[0166] Step S6 adjusts the fault diagnosis model parameters in the following ways:

[0167] a. Dynamically adjust the classification threshold of the fault classification model;

[0168] b. Dynamically adjust the bifurcation point sensitivity parameters;

[0169] c. Dynamically adjust the tensor decomposition path;

[0170] d. During the parameter adjustment process, the optimization strategy is updated in real time to meet different operating environments and load conditions.

[0171] S61, calculating the Jacobian matrix of the power system according to the real-time operation data;

[0172] S62, by calculating the eigenvalue of the Jacobian matrix, determining whether the real part of the eigenvalue changes from a negative value to a positive value;

[0173] S63. When it is detected that the system is approaching a bifurcation point, a fault warning signal is triggered, and a bifurcation characteristic value curve of real-time operation data is output.

[0174] Specifically, this step inherits the fault classification result of step S5, and adjusts the parameters of the diagnosis model to adapt to different working conditions and operating environments in combination with the real-time operation data of the excitation system. In general, the operating state of the excitation system may change dynamically due to load changes, external disturbances or equipment aging, so adaptive optimization of the diagnosis model is a necessary means to ensure the real-time and accuracy of fault diagnosis.

[0175] In this embodiment, the dynamic adjustment of the fault classification model is achieved through the reinforcement learning algorithm. Reinforcement learning is an interactive machine learning method that continuously optimizes model parameters through interaction with the operating environment so that the classification model can adapt to the dynamic changes of the system in real time. In this process, the basic elements of reinforcement learning include state space, action space and reward function.

[0176] Specifically, the design of the reinforcement learning model in this embodiment is as follows:

[0177] The state space is defined as the current operating state and real-time operating data characteristics of the fault classification model, expressed as:

[0178]

[0179] Among them, C is the core feature after tensor decomposition, λ current is the current bifurcation point detection sensitivity parameter, E current is the real-time classification error.

[0180] The action space is defined as the dynamic adjustment operations of the classification model, which includes the following three categories:

[0181] Adjust the classification threshold to optimize the classification model's ability to distinguish single faults and multiple fault coupling modes;

[0182] Adjust the bifurcation point detection sensitivity parameters to improve the accuracy of bifurcation point detection and the timeliness of early warning;

[0183] Optimize the tensor decomposition path to improve the extraction accuracy of tensor core features.

[0184] The reward function is used to evaluate the contribution of an action to the model optimization effect, and its mathematical expression is:

[0185] R=w1·(1-E classification )-w2·T update

[0186] Among them, E classification is the classification error, T u pdate is the time required for parameter update, w1 and w2 are weight coefficients.

[0187] In general, the weight coefficient of the reward function can be adjusted according to the real-time requirements of the system. For example, when the system is running under high load or high risk, the weight of w1 can be increased to prioritize the classification accuracy; while when the system is running stably, the weight of w2 can be appropriately increased to reduce the computational burden.

[0188] In the implementation of reinforcement learning, the Deep Deterministic Policy Gradient (DDPG) algorithm is used to support the optimization of continuous action space. The DDPG algorithm combines the policy network and the value network to achieve efficient adjustment of high-dimensional parameters.

[0189] In this embodiment, in order to further improve the optimization effect, a sliding window mechanism is introduced to update the state space in real time. The size of the sliding window is determined by the dynamic characteristics of the operating data. For example, when the system changes rapidly, the size of the sliding window can be reduced to increase the optimization frequency; when the system is running stably, the sliding window can be appropriately enlarged to reduce unnecessary adjustments. As an option, this embodiment also supports dynamic optimization of tensor decomposition paths. By analyzing the impact of different tensor decomposition paths on core features, the optimal path can be selected for tensor dimensionality reduction, thereby improving the representation ability of features. For example, for a multidimensional tensor It can be decomposed through the following paths:

[0190]

[0191] During the optimization process, the projection matrix U can be dynamically adjusted (n) The dimension of is used to balance the feature extraction accuracy and computational complexity. Specifically, in a possible implementation, the classification threshold adjustment process is as follows:

[0192] 1. According to the current output of the classification model, calculate the classification probability distribution of single faults and multiple faults;

[0193] 2. Dynamically adjust the classification threshold to minimize the overlapping area between different categories;

[0194] 3. The updated classification threshold is used for subsequent fault diagnosis tasks.

[0195] In the optimization process of the bifurcation point detection sensitivity parameters, the change trend of the key parameters can be calculated based on the sensitivity analysis method. For example, the bifurcation point sensitivity coefficient is calculated by the following formula:

[0196]

[0197] Among them, S i is the sensitivity coefficient, λ is the characteristic value, p i is a key parameter. When the sensitivity coefficient exceeds the set threshold, the system automatically adjusts the sensitivity parameter of bifurcation point detection to adapt to the current operating state. In the implementation of tensor decomposition path optimization, the optimal projection path can be selected by analyzing the impact of projection matrices of different dimensions on the core tensor. For example, for the dimensionality reduction task of high-dimensional tensors, singular value decomposition (SVD) can be used to select the projection matrix corresponding to the first k eigenvalues ​​to retain the main features.

[0198] S7. Output the fault diagnosis results to the visualization interface to display the system operation status, multi-fault coupling characteristics and health assessment results.

[0199] Specifically, this step inherits the output results of the aforementioned fault classification model and the dynamically optimized diagnostic parameters, and provides comprehensive fault information support for operation and maintenance personnel through intuitive and systematic visualization methods to ensure the efficiency of fault location and decision-making. In general, the operating status of the excitation system is complex and the fault modes are diverse. Therefore, the use of multi-dimensional visualization methods can make complex diagnostic information easy to understand and operate.

[0200] In this embodiment, the visualization interface is divided into multiple levels, which are used to display fault diagnosis information of different dimensions, including multi-fault coupling characteristics, system operation status trends, and health score results. In general, the interface design needs to take into account real-time, intuitive, and comprehensive data, as follows.

[0201] Specifically, the display of multi-fault coupling features is based on the visualization of tensor core features. High-dimensional tensors are projected into two-dimensional or three-dimensional space through dimensionality reduction technology, and the correlation between different fault features is displayed in the form of heat maps or point cloud maps. For example, in the heat map, the depth of different colors indicates the coupling strength between different fault features. The point cloud map shows the aggregation of fault features with the distribution position and density of three-dimensional scattered points, thereby intuitively reflecting the complexity of the fault mode.

[0202] As an option, visualization of core features can also be combined with the classification results of fault categories to mark different fault modes in the point cloud with colors or shapes. For example, low excitation limit faults are marked in red, volt-hertz limit faults are marked in blue, and multiple fault coupling modes are marked with mixed points of different colors. This marking method makes it easier for operators to quickly locate key fault modes.

[0203] In one possible implementation, the results of bifurcation point detection can be superimposed on the time series graph. For example, when the system approaches a bifurcation point, the key position of the time series curve will be displayed with a marker, and the characteristic value information of the bifurcation point will be attached to remind maintenance personnel to pay attention to potential instability risks.

[0204] Embodiment 2:

[0205] Reference Figure 6 In a second embodiment of the present invention, the present invention provides an online monitoring and fault self-diagnosis system for an excitation system of a hydroelectric generator set, the system comprising:

[0206] A data acquisition module is used to collect the operating parameters and environmental parameters of the excitation system and transmit the collected data to the data processing module;

[0207] Specifically, the data acquisition module is the basic module of the system, which is responsible for real-time acquisition of key operating parameters and external environmental parameters of the excitation system. The collected operating parameters include excitation current, excitation voltage, machine terminal voltage, frequency, active power, reactive power, etc., and external environmental parameters include temperature, humidity and vibration intensity around the equipment. Through high-precision sensor arrays and industrial communication protocols, the module can ensure the accuracy and real-time nature of data acquisition. The collected data is transmitted to the data processing module via wired or wireless communication, supporting high-frequency sampling and large-scale data transmission, and providing data support for subsequent signal processing and modeling analysis.

[0208] Through the implementation of the data acquisition module, the dynamic change characteristics and fault signs of the excitation system can be accurately captured, and environmental factors can be included in the analysis scope to provide a comprehensive data basis for system status evaluation. The high reliability and anti-interference of the module can ensure stability in complex operating environments.

[0209] The data processing module is used for filtering and denoising, time series sliding window segmentation and tensor modeling, and transmits the generated multi-dimensional tensor to the dynamic modeling and bifurcation point detection module;

[0210] Specifically, the data processing module is used to preprocess and format the raw data transmitted by the acquisition module. The module first removes noise from the data, suppresses high-frequency noise through hardware filtering, and smoothes data fluctuations with software filtering. The processed data is divided into multiple time segments according to the time sliding window, and each time segment has a fixed length and overlap ratio to ensure the time continuity and feature integrity of the data.

[0211] Subsequently, the data processing module organizes the segmented time segments into multimodal tensors, and retains the high-dimensional relationship between variables through data tensor modeling. The output of the multidimensional tensor is not only convenient for subsequent feature extraction and diagnostic analysis, but also can intuitively reflect the multivariate dynamic characteristics of the system. The final processed data is transmitted to the dynamic modeling and bifurcation point detection module through a standardized interface for modeling and state analysis.

[0212] The role of the data processing module is to improve the validity and structure of the original data, making it more suitable for subsequent analysis and modeling, while reducing the impact of noise and data redundancy on diagnostic accuracy.

[0213] Dynamic modeling and bifurcation point detection module, used to establish a nonlinear dynamic model of the system, detect bifurcation points in real time and transmit warning signals and bifurcation characteristic values ​​to the fault diagnosis module;

[0214] Specifically, the dynamic modeling and bifurcation point detection module receives the tensorized data output by the data processing module and builds a nonlinear dynamic model based on the operating characteristics of the excitation system. By analyzing the dynamic changes of the operating parameters, the module can detect the possible bifurcation points of the system in real time, that is, the critical transition points of the operating state. When the bifurcation point is approaching, the module generates an early warning signal to indicate the possible instability risk faced by the system.

[0215] At the same time, the bifurcation point detection module generates key eigenvalue information about the operating status through bifurcation feature extraction, and transmits this information to the fault diagnosis module to support higher-precision fault classification and pattern recognition. The module also achieves effective characterization of the complex nonlinear characteristics of the system through integration with system dynamic modeling, providing a theoretical basis for system stability assessment.

[0216] The implementation of dynamic modeling and bifurcation point detection modules effectively solves the problem that traditional monitoring systems are difficult to predict nonlinear faults, and enhances the safety of system operation through real-time early warning functions.

[0217] Fault diagnosis module, which is used to extract tensor core features and classify single faults and multiple faults, and transmit the diagnosis results to the visualization module;

[0218] Specifically, the fault diagnosis module completes the classification analysis of the fault mode of the excitation system based on the input results of the dynamic modeling and bifurcation point detection modules, combined with the core features in the tensor data. Through the multi-category fault model, the module can identify single fault modes (such as low excitation limit, volt-hertz limit) and multi-fault coupling modes (such as the superposition effect between multiple faults). The classification results are output in the form of probability to describe the possibility of different fault modes.

[0219] The module design fully considers the complexity of the excitation system and the coupling of multiple fault characteristics, and improves the ability to identify fault modes through classification model optimization. The diagnosis results are transmitted to the visualization module, providing a clear basis for fault analysis for operation and maintenance personnel.

[0220] Adaptive optimization module, used to optimize diagnostic logic by dynamically adjusting classification thresholds, bifurcation point detection sensitivity, and tensor decomposition paths;

[0221] Specifically, the adaptive optimization module is the core adjustment module of the system, which realizes dynamic adjustment of model parameters by real-time monitoring of the classification accuracy and parameter changes of the diagnosis model. The optimization content includes the classification threshold adjustment of the fault classification model, the optimization of the bifurcation point detection sensitivity parameters, and the optimization selection of the tensor decomposition path.

[0222] By combining real-time operation data with optimization algorithms, the module can dynamically adjust the parameters of the classification model according to the current working conditions, ensuring the adaptability and stability of the diagnostic logic under different operating environments. The optimized parameter settings are further passed to the diagnosis module and bifurcation point detection module to continuously optimize the diagnostic accuracy and computational efficiency.

[0223] The visualization module is used to display fault classification results, multi-fault coupling relationships and operation status evaluation results.

[0224] Specifically, the visualization module is used to display diagnostic results and operating status assessment information, providing an intuitive decision support tool for operation and maintenance personnel. The module displays fault classification results and multi-fault coupling relationships in a variety of graphical ways, such as using heat maps to display the correlation strength between different fault modes, and using time series graphs to dynamically display the changing trends of key operating parameters.

[0225] In addition, the visualization module also displays the operating status of the excitation system based on the health scoring system, dividing the operating status into multiple health levels (such as normal, slightly abnormal, seriously abnormal, etc.). The calculation of the health score combines the fault classification results and real-time operating parameters to quantify the health status of the system and provide data support for maintenance strategies.

[0226] Embodiment 3

[0227] The third embodiment of the present invention is based on the same inventive concept and proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the online monitoring and fault self-diagnosis method of the excitation system of the hydropower generator set of the above embodiment are implemented.

[0228] Embodiment 4

[0229] The fourth embodiment of the present invention is based on the same inventive concept. The present invention proposes a computer device, which includes: a processor and a memory; the processor and the memory communicate with each other; the memory is used to store instructions; the processor is used to execute the instructions in the memory, and execute the online monitoring and fault self-diagnosis method of the excitation system of the hydropower generator set of the above embodiment.

[0230] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0231] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for online monitoring and fault self-diagnosis of an excitation system of a hydroelectric generator set, characterized in that: The following steps are involved: S1. Collect real-time operation data of the excitation system, including excitation current, excitation voltage, terminal voltage, frequency, active power, reactive power and environmental parameters; S2, filtering the collected data to remove high-frequency noise from the data, and dividing the data into time segments of fixed length based on a time sliding window; S3, construct a multi-modal tensor model, convert the collected multi-dimensional data into tensor form, and extract multi-fault coupling characteristics; S4. Based on nonlinear modeling of the power system and detection of bifurcation points, the stability of the system operation state is judged, and a fault warning signal is issued when the characteristic value change trend shows that the system is approaching the bifurcation point; S5. Use the classification model to classify the tensor core features into fault categories and output the diagnosis results of a single fault or a multiple fault coupling mode; S6. Dynamically adjust the classification threshold, bifurcation point sensitivity parameter and tensor decomposition path of the fault classification model based on real-time diagnosis results to optimize fault feature extraction performance and diagnosis accuracy; S7. Output the fault diagnosis results to the visualization interface to display the system operation status, multi-fault coupling characteristics and health assessment results.

2. The method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set according to claim 1, characterized in that: The multimodal tensor quantization model step in the S3 step includes the following processes: S31, constructing a multidimensional data tensor according to the real-time collected excitation current, excitation voltage, terminal voltage, frequency, active power, reactive power and environmental parameters; S32. Decompose the tensor into a core tensor and a low-dimensional projection matrix, wherein the core tensor is used to extract the global characteristics of the multi-fault coupling mode, and the low-dimensional projection matrix reduces the data complexity by mapping the high-dimensional data to the low-dimensional space.

3. The method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set according to claim 1, characterized in that: The nonlinear modeling of the power system in step S4 includes: S41, constructing a nonlinear dynamic equation of the excitation system and establishing a state vector model of the system, wherein the state vector includes the excitation current, the excitation voltage and the frequency; S42, calculating eigenvalues ​​based on the Jacobian matrix of the power system model to determine the stability of the system operation state; S42. When the real part of the system's eigenvalue changes from a negative value to a positive value, it is determined that the system is close to a bifurcation point, and a fault warning signal is triggered.

4. The method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set according to claim 1, characterized in that: The fault classification step in step S5 includes: S51, extracting the core features after tensor decomposition as input features; S52. Fault mode identification is performed using a nonlinear classification model, and the classification results include a single fault mode and a multi-fault coupling mode.

5. The method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set according to claim 1, characterized in that: The step S6 adjusts the fault diagnosis model parameters in the following manner: a. Dynamically adjust the classification threshold of the fault classification model; b. Dynamically adjust the bifurcation point sensitivity parameters; c. Dynamically adjust the tensor decomposition path; d. During the parameter adjustment process, the optimization strategy is updated in real time to meet different operating environments and load conditions.

6. The method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set according to claim 1, characterized in that: The bifurcation point detection in step S6 includes the following process: S61, calculating the Jacobian matrix of the power system according to the real-time operation data; S62, by calculating the eigenvalue of the Jacobian matrix, determining whether the real part of the eigenvalue changes from a negative value to a positive value; S63. When it is detected that the system is approaching a bifurcation point, a fault warning signal is triggered, and a bifurcation characteristic value curve of real-time operation data is output.

7. The method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set according to claim 1, characterized in that: The visual output of the diagnostic results in step S7 includes the following: Output multi-fault coupling characteristic diagram to show the correlation between faults; Dynamically update the system operation health score and quantify the score results; Output operation status optimization suggestions, including excitation system parameter adjustment suggestions and maintenance strategies. 8.Online monitoring and fault self-diagnosis system of hydroelectric generator excitation system, characterized by: The method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set according to any one of claims 1 to 7, the system comprising: A data acquisition module is used to collect the operating parameters and environmental parameters of the excitation system and transmit the collected data to the data processing module; The data processing module is used for filtering and denoising, time series sliding window segmentation and tensor modeling, and transmits the generated multi-dimensional tensor to the dynamic modeling and bifurcation point detection module; Dynamic modeling and bifurcation point detection module, used to establish a nonlinear dynamic model of the system, detect bifurcation points in real time and transmit warning signals and bifurcation characteristic values ​​to the fault diagnosis module; Fault diagnosis module, which is used to extract tensor core features and classify single faults and multiple faults, and transmit the diagnosis results to the visualization module; Adaptive optimization module, used to optimize diagnostic logic by dynamically adjusting classification thresholds, bifurcation point detection sensitivity, and tensor decomposition paths; The visualization module is used to display fault classification results, multi-fault coupling relationships and operation status evaluation results.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for online monitoring and fault self-diagnosis of the excitation system of a hydroelectric generator set as described in any one of claims 1 to 7 is implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for online monitoring and fault self-diagnosis of an excitation system of a hydroelectric generator set according to any one of claims 1 to 7 is implemented.

Images