Dynamic fault analysis method and system for pumped storage hydropower station

Abstract

The invention discloses a dynamic fault analysis method and system for a pumped storage hydropower station, and the method comprises the steps: receiving multi-dimensional state parameters of the hydropower station, including the rotating speed of a water turbine, the opening degree of a guide vane, the water pressure and vibration data, and building a multi-dimensional state space model; determining segmentation points according to the change rate of the state parameters, performing initial segmentation on a state space by adopting an adaptive grid division algorithm, and establishing a linear subsystem model in each segmentation region to obtain a segmented affine (PWA) approximation model; and constructing a hierarchical learning framework, and carrying out online updating on the PWA approximation model and the hierarchical learning framework through an incremental learning algorithm, a sliding time window and a knowledge distillation technology.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis of hydropower stations, and in particular to a dynamic fault analysis method and system for a pumped-storage hydropower station. Background Art

[0002] Pumped-storage hydropower stations are important peak-shaving and frequency-regulating equipment in power systems, and their safe and stable operation has a significant impact on the safety and reliability of power grids. At present, fault diagnosis of hydropower stations mainly relies on expert experience and static threshold judgment, which has the following technical problems: First, the traditional single threshold judgment method cannot effectively capture the dynamic characteristics of the hydropower station system, resulting in insufficient accuracy of fault diagnosis; second, the existing fault analysis method lacks a systematic modeling framework, making it difficult to accurately identify the fault type and trace the root cause of the fault; third, the traditional method has a low utilization rate of historical data, and the model cannot be adaptively updated as the dynamic characteristics of the system change, which affects the real-time and reliability of fault diagnosis. Summary of the invention

[0003] In response to the above technical problems, the present application provides a method and system for dynamic fault analysis of a pumped storage hydropower station. The method combines PWA approximation and hierarchical confidence learning technology, establishes an accurate model by adaptively segmenting the state space, and uses confidence upper bound constraints to ensure the reliability of the learning process, thereby achieving efficient and accurate fault diagnosis.

[0004] In view of this, the present application provides a method and system for dynamic fault analysis of a pumped-storage hydropower station, which solves the problems of insufficient accuracy in fault diagnosis of hydropower stations, lack of a systematic modeling framework, and poor model adaptability in the prior art.

[0005] The embodiment of the present application provides a method for dynamic fault analysis of a pumped storage hydropower station, including: collecting multi-dimensional state parameters of the hydropower station, including turbine speed, guide vane opening, water pressure and vibration data; establishing a piecewise affine PWA approximate model, including determining segmentation points according to the rate of change of the state parameters, using an adaptive grid partitioning algorithm to initially segment the state space, and establishing a linear subsystem model in each segmented area; constructing a hierarchical learning framework for fault analysis, the hierarchical learning framework including a feature extraction layer, a fault identification layer and a cause reasoning layer; wherein: in the feature extraction layer, the state deviation is calculated based on the PWA approximate model, and time domain and frequency domain features are extracted; in the fault identification layer, the fault confidence is calculated based on the time domain and frequency domain features, effective features are screened according to the set confidence upper limit constraints, the screened features are pattern matched with the fault feature library, and the fault type probability distribution is calculated based on the confidence weighting; in the cause reasoning layer, historical fault cases are screened and weighted according to the fault confidence, a fault causal network of the hydropower station system is constructed based on the weighted historical cases, and the root cause and propagation path of the fault are derived in combination with the fault type probability distribution.

[0006] Optionally, the step of establishing a piecewise affine PWA approximate model includes: preprocessing the operating parameters of the hydropower station, including normalizing the turbine speed, performing median filtering and denoising on the guide vane opening signal, interpolating and repairing the water pressure data, and performing sliding window statistical feature calculations on the vibration data; calculating the dynamic characteristics of the system based on the preprocessed parameters, including calculating the first-order derivatives of the turbine speed and the guide vane opening, analyzing the correlation of water pressure changes, and extracting the periodic characteristics of the vibration signal; determining spatial segmentation based on the dynamic characteristics, using smaller grids in areas where parameters change dramatically, using larger grids in areas where changes are gentle, and merging adjacent grids with similar dynamic characteristics.

[0007] Optionally, the step of calculating the state deviation and extracting time domain and frequency domain features based on the PWA approximate model in the feature extraction layer includes: calculating the statistical characteristics of the state parameters of the hydropower station, including calculating the mean and variance of the turbine speed, analyzing the skewness and kurtosis of the guide vane opening, and calculating the correlation coefficient matrix of the water pressure parameters; extracting the trend characteristics of the operating parameters of the hydropower station, including calculating the turbine speed change rate, detecting the inflection point of the guide vane opening change, and analyzing the change trend of the water pressure parameters; identifying the mutation characteristics of the parameters of the hydropower station, including detecting abnormal points of the vibration signal, calculating the amplitude of the water pressure mutation, and recording the duration of the parameter mutation; performing frequency domain analysis, including calculating the power spectral density of the vibration signal, analyzing the main frequency components of the turbine speed, and statistically analyzing the energy distribution of each frequency band.

[0008] Optionally, the step of calculating the fault confidence and making a fault type judgment based on the time domain and frequency domain features at the fault identification layer includes: screening based on the confidence upper limit constraint of the hydropower station parameter characteristics, including calculating the confidence interval of the turbine speed characteristics, evaluating the reliability of the guide vane opening characteristics, and analyzing the confidence of the water pressure and vibration characteristics; constructing a hydropower station fault feature vector, including combining time domain and frequency domain features, normalizing the features, and establishing a multi-scale feature representation; performing fault pattern matching, including calculating the similarity with historical fault cases, calculating the probability based on confidence weighting, and outputting the fault type judgment result.

[0009] Optionally, the PWA approximation model and hierarchical learning framework are updated online through an incremental learning algorithm, specifically including using recursive least squares method to update linear subsystem parameters, adaptively adjusting the time window size based on the dynamic characteristics of the system, migrating historical model knowledge to the current model through soft label transfer, and dynamically adjusting the confidence upper limit constraint parameters of each layer according to the model prediction performance.

[0010] The dynamic fault analysis method for a pumped-storage hydropower station according to claim 5 is characterized in that the step of updating the linear subsystem parameters using the recursive least squares method includes: initializing the system parameters of the hydropower station, including setting the initial estimated values ​​of the unit speed, guide vane opening, water pressure and vibration parameters; calculating the prediction error based on new data collected in real time, including comparing the deviation between the measured value and the model predicted value; updating the model parameters, including calculating the gain matrix, updating the parameter estimation value, and introducing a forgetting factor to reduce the weight of historical data.

[0011] Optionally, the step of migrating historical model knowledge to the current model by soft label transfer includes: constructing a teacher model, including selecting the model with the best performance in the historical operation data of the hydropower station as the basic teacher model; training the student model, including training with the soft labels of the hydropower station parameters generated by the teacher model, setting the knowledge distillation temperature parameters, and adding regularization constraints; realizing model fusion, including designing a progressive knowledge transfer mechanism based on the operation performance indicators of the hydropower station, updating the model parameters by exponential sliding average, and dynamically adjusting the weights of new and old knowledge.

[0012] Optionally, the step of dynamically adjusting the confidence upper limit constraint parameters of each layer according to the model prediction performance includes: establishing a performance evaluation system, including evaluating the accuracy and recall rate of hydropower station fault diagnosis, calculating the confidence interval of the model prediction, and monitoring the false alarm rate and the missed alarm rate; constructing a multi-level constraint transmission mechanism, including setting the basic confidence threshold of each layer, establishing the constraint mapping relationship between the parameter feature layers of the hydropower station, and maintaining the consistency of the constraint strength; performing adaptive adjustment, including calculating the performance index based on the sliding time window, adjusting the constraint strength according to the change of the index, and appropriately relaxing the constraints when the system is stable.

[0013] Optionally, the step of adaptively adjusting the time window size based on the dynamic characteristics of the system includes: initializing window parameters, including setting a basic window length according to the maximum response time of the hydropower station unit, setting a sliding step of the sampling data, and configuring the data cache size; performing adaptive adjustment, including counting the frequency of hydropower station faults, using a larger window under normal conditions, and reducing the window to increase the response speed under fault conditions; performing data quality control, including detecting and eliminating abnormal data points, smoothing and filtering the unit operation data, and calculating data reliability indicators.

[0014] An embodiment of the present application also provides a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned method for dynamic fault analysis of a pumped-storage hydropower station.

[0015] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the above-mentioned method for dynamic fault analysis of a pumped-storage hydropower station.

[0016] An embodiment of the present application also provides a computer program product, including computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the method for dynamic fault analysis of the pumped-storage hydropower station are implemented.

[0017] This application has the following technical effects:

[0018] By establishing a piecewise affine PWA approximate model, the dynamic characteristics of the hydropower station system can be accurately described, and the accuracy of fault diagnosis is improved;

[0019] A hierarchical learning framework consisting of feature extraction layer, fault identification layer and cause inference layer was constructed, which achieved systematic modeling from fault feature extraction to root cause analysis and improved the reliability of fault diagnosis.

[0020] An incremental learning algorithm is used to update the model online, and knowledge migration is achieved through soft label transfer, so that the model can adaptively adapt to changes in system characteristics and improve the real-time performance of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for describing the embodiments are briefly introduced below.

[0023] Figure 1 It is an overall flow chart of the dynamic fault analysis method of a pumped storage hydropower station provided in an embodiment of the present application;

[0024] Figure 2 is a flowchart of building a PWA approximate model provided in an embodiment of the present application;

[0025] Figure 3 It is a schematic diagram of the hierarchical learning framework structure provided in the embodiment of the present application;

[0026] Figure 4 It is a schematic diagram of the online update mechanism provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The embodiments of the present application are described below in conjunction with the drawings in the present application. It should be understood that the implementation methods described below in conjunction with the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0028] It will be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application refer to that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude the implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the technical field. It should be understood that when an embodiment of the present invention refers to an element being "connected" or "coupled" to another element, the one element may be directly connected or coupled to the other element, or may refer to the one element and the other element establishing a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein indicates at least one of the items defined by the term, for example, "A and / or B" may be implemented as "A", or as "B", or as "A and B".

[0029] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0030] The core of the present invention is to provide a method and system for generating intelligent software test data based on Internet public information. The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0031] It should be understood that the terms "including" and "having" and any variations thereof used herein are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0032] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0033] like Figure 1 , the embodiment of the present application includes the following steps:

[0034] S1: Collect multi-dimensional state parameters of the hydropower station, including turbine speed, guide vane opening, water pressure and vibration data;

[0035] In this embodiment, collecting multi-dimensional state parameters of the hydropower station is the basis and prerequisite for fault analysis. Specifically, the collection system obtains the operating data of each key monitoring point of the hydropower station in real time through a distributed data collection network. These data include:

[0036] Turbine speed data: collected through the speed sensor installed on the main shaft of the turbine, with a sampling frequency of 100Hz. The speed data reflects the rotation state of the unit and can be used to monitor abnormal conditions such as unit speed fluctuation, overspeed, underspeed, etc. The collected speed data includes the instantaneous speed value and its change trend, which is of great significance for identifying faults such as turbine vibration, stall and cavitation.

[0037] Guide vane opening data: collected through the guide vane opening sensor, with a sampling frequency of 10Hz. The guide vane opening data reflects the flow regulation status of the unit and is an important parameter for evaluating the operating conditions of the turbine. The system collects the actual opening value of the guide vane and its rate of change. These data can be used to analyze guide vane regulation system failures, such as guide vane jamming and governor failure.

[0038] Water pressure data: collected through pressure sensors installed in key locations such as the water inlet pipe, volute and tailwater pipe, with a sampling frequency of 200Hz. Water pressure data includes static pressure and dynamic pressure, reflecting the hydraulic working conditions of the turbine. The system collects pressure values ​​and fluctuation characteristics at different measuring points. These data can be used to analyze hydraulic system anomalies, such as cavitation, water hammer and pressure pulsation.

[0039] Vibration data: collected through acceleration sensors installed in the bearing seat, frame and foundation, with a sampling frequency of 1kHz. Vibration data reflects the mechanical state of the unit and is a key parameter for diagnosing mechanical failures. The system collects vibration acceleration in three directions (radial, axial and tangential). These data can be used to analyze mechanical failures such as shaft misalignment, bearing wear, and blade cracks.

[0040] During the collection process, the system uses the following key technologies to ensure the quality and reliability of data:

[0041] 1. Adopt high-precision synchronous sampling technology to ensure the time synchronization of data at multiple measuring points, and the sampling clock error is controlled within 1μs;

[0042] 2. Realize real-time data transmission of distributed data acquisition nodes, adopt industrial Ethernet protocol, and control network delay within 10ms;

[0043] 3. Set up anti-interference measures for data collection, including signal shielding, isolation and filtering, to ensure that the signal-to-noise ratio of the collected data is not less than 40dB;

[0044] 4. Establish a redundant backup mechanism for data collection, configure dual sensors at key measuring points, and ensure data reliability through data comparison.

[0045] In addition, the system is also equipped with a data warning function. When the collected parameters exceed the preset range, the alarm signal is triggered in time to provide warning information for subsequent fault analysis.

[0046] S2: Building a piecewise affine PWA approximation model

[0047] like Figure 2 The purpose of establishing the piecewise affine PWA (Piecewise Affine) approximation model is to accurately describe the dynamic characteristics of the hydropower station under different working conditions. This model decomposes the complex nonlinear system into multiple simple linear subsystems, which not only ensures the accuracy of the model but also reduces the computational complexity. The specific implementation process is as follows:

[0048] S2.1 Preprocessing of hydropower station operation parameters

[0049] First, preprocess all types of collected data to improve data quality:

[0050] Normalize the turbine speed: Use the maximum and minimum value normalization method to map the speed value to the [0,1] interval. The specific calculation formula is:

[0051] normalized_speed=(current_speedmin_speed) / (max_speedmin_speed)

[0052] Among them, min_speed and max_speed are respectively ±10% of the rated speed of the turbine. This treatment can eliminate the dimension effect and facilitate the subsequent model calculation.

[0053] Median filtering and denoising of the guide vane opening signal: a 5-point median filtering algorithm is used with a sliding window size of 0.5 seconds. The specific steps are:

[0054] Sort the data within the time window

[0055] Select the middle value after sorting as the filtering result

[0056] Sliding window point by point to complete the full sequence filtering

[0057] This method can effectively remove the sudden noise in the guide vane opening signal while maintaining the rapid change characteristics of the signal.

[0058] Interpolation repair of water pressure data: cubic spline interpolation method is used to process missing values ​​in water pressure data. The specific implementation is:

[0059] Detecting missing points in a data series

[0060] Take 4 valid data points before and after the missing point

[0061] Constructing a cubic spline function for interpolation calculation

[0062] Perform physical constraint checks on interpolation results to ensure that the interpolation results are within a reasonable range

[0063] This method can maintain the continuity and smoothness of water pressure data.

[0064] Calculate the sliding window statistical features of the vibration data: Use a 2-second sliding time window with a window overlap rate of 50% to calculate the following statistical features:

[0065] Root mean square value (RMS): reflects the vibration intensity

[0066] Kurtosis factor: reflects the impact characteristics

[0067] Waveform factor: reflects the regularity of the waveform

[0068] Margin factor: reflects the waveform concentration

[0069] These features can effectively characterize the time domain characteristics of vibration signals.

[0070] S2.2 Calculate the system dynamic characteristics based on the preprocessed parameters

[0071] Calculate the first-order derivative of turbine speed and guide vane opening: Use the central difference method with a time step of 0.1 seconds to calculate the rate of change of these two parameters. The calculation formula is:

[0072] derivative=(value(t+Δt)value(t-Δt)) / (2Δt)

[0073] Where Δt is the time step. These first-order derivatives reflect the dynamic response characteristics of the system.

[0074] Analyze the correlation of water pressure changes: Calculate the correlation coefficient matrix of water pressure data at the three measuring points of the water inlet pipe, volute and tailwater pipe, and use the Pearson correlation coefficient method to reflect the correlation of pressure changes at each measuring point. The significance level of the correlation coefficient is set to 0.05 to determine the transmission relationship of pressure fluctuations.

[0075] Extract the periodic characteristics of the vibration signal: Use the autocorrelation analysis method to detect the periodic components of the vibration signal. The specific steps include:

[0076] Calculate the autocorrelation function of the vibration signal

[0077] Find the local maximum point of the autocorrelation function

[0078] Analyze the maximum point interval to determine the main periodic components of the signal

[0079] These features can be used to identify periodic faults such as shaft rotation and blade passing.

[0080] S2.3 Determine spatial segmentation based on dynamic features

[0081] In the area where the parameters change dramatically, a smaller grid (grid size is 1 / 4 of the standard deviation) is used, and in the area where the parameters change slowly, a larger grid (grid size is 1 / 2 of the standard deviation) is used. The specific implementation is:

[0082] Calculate the variance of the rate of change of each parameter

[0083] Adaptively adjust the grid size according to the variance

[0084] This adaptive meshing method, which merges adjacent grids with similar dynamic characteristics (similarity threshold is 0.85), can optimize computational efficiency while ensuring model accuracy.

[0085] In each segmented area, the least squares method is used to establish a linear subsystem model. The general form of the model is:

[0086] x(k+1)=Ax(k)+Bu(k)+f

[0087] Among them, x is the state variable, u is the input variable, A and B are system matrices, and f is the bias term.

[0088] S3: Building a hierarchical learning framework for fault analysis

[0089] Hierarchical learning framework Figure 3 shown, specifically,

[0090] S3.1 Feature Extraction Layer

[0091] First, the system calculates the statistical characteristics of the state parameters of the hydropower station. In terms of turbine speed, the mean, standard deviation, skewness and kurtosis coefficients are calculated through data within a 10-minute time window. These indicators can fully reflect the stability level of the unit operation. For the guide vane opening parameters, not only its basic statistics are analyzed, but also the frequency and response time of the opening adjustment are focused on to evaluate the dynamic characteristics of the speed control system. At the same time, by calculating the correlation coefficient matrix of water pressure parameters at different measuring points, the transmission relationship and synchronization of pressure fluctuations are deeply analyzed, and abnormal pressure fluctuations are discovered in time.

[0092] Secondly, the system focuses on extracting the trend characteristics of each operating parameter. For the turbine speed, the speed change rate is obtained by calculating the first-order difference sequence, and its acceleration characteristics and periodic characteristics are further analyzed. In the guide vane opening analysis, the three-point method is used to detect the inflection point of the opening change, and the time distribution characteristics of the inflection point are combined to evaluate the stability of the regulation process. For the water pressure parameters, the sliding regression method is used to extract the trend term, and the dynamic characteristics of the water pressure change are accurately grasped by analyzing the slope and curvature of the trend.

[0093] Thirdly, the system focuses on identifying the mutation characteristics of each parameter. In the vibration signal processing, the 3σ criterion and wavelet analysis method are combined to achieve accurate positioning of abnormal points, and the amplitude, duration and frequency of abnormal points are recorded in detail. For the detection of water pressure mutations, the dynamic threshold method is used to not only calculate the amplitude and rate of the mutation, but also deeply analyze its propagation characteristics and evaluate the possible harm caused by the mutation. At the same time, the system will record the duration of various mutations in detail, analyze the recovery process of the mutation, and provide a basis for evaluating the degree of impact on the system.

[0094] Finally, the system conducts in-depth frequency domain analysis on various types of signals. In vibration signal processing, the Welch method is used to calculate the power spectrum density, and the key characteristic frequency components are identified by analyzing the energy distribution characteristics of the spectrum. For the turbine speed signal, FFT technology is used for spectrum analysis, focusing on the frequency modulation characteristics, and the complexity of the spectrum is quantified by the spectrum entropy value. In terms of frequency band energy analysis, the system divides the spectrum into multiple characteristic frequency bands, and comprehensively evaluates the spectrum structure characteristics by calculating the energy distribution ratio of each frequency band.

[0095] S3.2 Fault Identification Layer

[0096] First, the system performs feature screening based on the confidence upper limit constraints of the hydropower station parameter characteristics. For the turbine speed characteristics, the characteristic indicators with statistical significance are screened out by calculating the 95% confidence interval. In the guide vane opening feature analysis, the Bootstrap method is used to evaluate the reliability of the features, and the features with confidence levels below the threshold are eliminated. For water pressure and vibration characteristics, the system comprehensively considers the measurement accuracy and physical constraints, and establishes a multi-dimensional feature reliability evaluation system.

[0097] Secondly, the system constructs a comprehensive fault feature vector of the hydropower station. By combining the filtered time domain and frequency domain features, a multi-dimensional feature space is formed. In the feature processing process, the Z-score standardization method is used to normalize the features of different dimensions to ensure the comparability between features. At the same time, the system also establishes a multi-scale feature representation, realizes multi-resolution analysis of features through wavelet packet decomposition, and enhances the expressiveness of features.

[0098] Finally, the system performs the fault pattern matching process. By calculating the cosine similarity between the current feature vector and the historical fault case, the possible fault type is preliminarily identified. In the matching process, a feature weight mechanism is introduced to assign different weight coefficients according to the diagnostic capabilities of different features. Based on the confidence-weighted voting mechanism, the system finally outputs the probability distribution of the fault type, providing a reliable basis for subsequent fault diagnosis.

[0099] S3.3 Cause Reasoning Layer

[0100] First, the system screens and weights historical fault cases according to the fault confidence. By setting a dynamic threshold, historical cases with high similarity to the current fault characteristics are screened out. In the weighting process, an exponential decay function is used so that historical cases with higher similarity have greater weights, thereby ensuring the accuracy of the reasoning results. At the same time, considering the timeliness of the fault, the system also introduces a time decay factor to reasonably balance the time correlation of historical cases.

[0101] Secondly, the system constructs the fault causal network of the hydropower station system based on weighted historical cases. The topological structure of the fault components is established by analyzing the physical connection relationship and fault propagation law between the fault components. In the process of network construction, the Bayesian network model is used to describe the conditional dependency relationship between components, and the probability parameters in the network are calculated using historical case data. The system also optimizes and verifies the network structure by introducing expert knowledge.

[0102] Finally, the system combines the probability distribution of fault types to deduce the root cause and propagation path of the fault. By performing probabilistic reasoning in the fault causal network, the posterior probability of each possible root cause is calculated. Based on the maximum a posteriori probability criterion, the most likely root cause of the fault is identified. At the same time, the system also analyzes the propagation link of the fault and predicts other components that may be affected, providing support for fault isolation and maintenance decisions.

[0103] For example, suppose that Unit 2 of a pumped storage power station has an abnormal situation. This example is used to illustrate the analysis process of the hierarchical learning framework:

[0104] S3.1 Feature Extraction Layer

[0105] First, the system calculates the statistical characteristics of the state parameters of Unit 2. For example, within a 10-minute time window, the mean of the turbine speed is 300 rpm, the standard deviation is 2.3 rpm, the skewness coefficient is 0.15, and the kurtosis coefficient is 3.8, indicating that the speed fluctuates greatly and the distribution is slightly asymmetric. The average guide vane opening is 85%, the adjustment frequency is 12 times / hour, and the response time is 1.2 seconds, which is significantly longer than the normal value of 1.0 seconds. At the same time, the pressure correlation coefficient between the water inlet pipe and the volute dropped from 0.92 under normal conditions to 0.78, indicating that the hydraulic system characteristics are abnormal.

[0106] Secondly, the system extracts the trend characteristics of each parameter. Through analysis, it was found that the turbine speed fluctuated periodically with a fluctuation period of about 1.5 seconds, and the fluctuation amplitude showed an increasing trend, reaching 15% within 10 minutes. Multiple irregular inflection points appeared during the adjustment of the guide vane opening, and the smoothness of the adjustment curve was significantly reduced. The trend analysis of the water pressure parameters showed that the root mean square value of the volute pressure fluctuation increased by 25% compared with the normal operating conditions.

[0107] Again, the system identified several significant mutation features. The vibration signal showed abnormal peaks in the radial direction, with the maximum instantaneous acceleration reaching 9.8m / s 2 , which is 2.3 times the normal value, and the abnormality lasted for more than 3 seconds. The water pressure mutation analysis showed that the water inlet pipe pressure had 5 instantaneous fluctuations exceeding 20% ​​of the rated value, and the fluctuations spread to the entire water system within 0.5 seconds.

[0108] Finally, through frequency domain analysis, it was found that there were obvious characteristic frequency components of 20Hz and 40Hz in the spectrum of the vibration signal, and their amplitudes were 35% and 28% of the fundamental frequency component respectively. The spectrum entropy value of the turbine speed signal increased from 0.65 in normal working conditions to 0.89, indicating that the dynamic characteristics of the system have become more complex.

[0109] S3.2 Fault Identification Layer

[0110] First, the system screens features and retains feature indicators with a confidence level of more than 90%, including: periodicity of speed fluctuation (confidence level 95.2%), guide vane response delay (confidence level 93.8%), radial vibration peak (confidence level 97.5%), etc. These features significantly deviate from the normal operating range and have high diagnostic value.

[0111] Secondly, the fault feature vector constructed by the system contains 25 dimensions, including 12 time domain features and 13 frequency domain features. After standardization, the feature vector is matched with the historical case library to obtain the three most similar fault types and their probabilities:

[0112] Guide vane cylinder leakage (matching degree 85.6%)

[0113] Bearing wear (matching degree 72.3%)

[0114] Cavitation vibration (matching degree 65.8%)

[0115] S3.3 Cause Reasoning Layer

[0116] First, the system screened out 32 relevant cases from the historical case library, including 15 cases of guide vane cylinder leakage, 12 cases of bearing wear, and 5 cases of cavitation vibration. Through similarity weighting, the system focused on analyzing 12 cases with a matching degree of more than 80%.

[0117] Second, the fault causal network constructed based on these cases shows that:

[0118] The probability of delayed regulation response due to guide vane cylinder leakage is 92%

[0119] The probability of speed fluctuation caused by delayed regulation response is 87%

[0120] The probability of speed fluctuation exacerbating vibration abnormality is 83%

[0121] Finally, the system deduced the most likely cause chain of the fault: guide vane cylinder seal deterioration (root cause) → leakage in the cylinder → guide vane adjustment hysteresis → load adjustment instability → speed fluctuation → increased vibration. Based on this diagnostic result, the system recommends focusing on checking the seal status of the guide vane cylinder during the next maintenance.

[0122] This example clearly shows how the hierarchical learning framework can gradually extract features from raw data, identify fault types, and ultimately deduce the root cause of the fault, providing clear repair suggestions for maintenance personnel. The advantage of this framework is that it can systematically analyze complex faults and provide reliable diagnostic results.

[0123] S4 is updated online through incremental learning algorithm

[0124] Online update mechanism such as Figure 4 . First, the system uses recursive least squares to update the parameters of the linear subsystem. In the initialization stage, the initial estimates of the unit speed, guide vane opening, water pressure and vibration parameters are set according to the historical operating data. When new data arrives, the prediction error is calculated by comparing the deviation between the measured value and the model prediction value. Subsequently, the system updates the gain matrix based on the prediction error, and combines the forgetting factor mechanism to gradually update the model parameters to ensure that the model can adapt to the dynamic changes of system characteristics.

[0125] Secondly, the system adaptively adjusts the time window size based on dynamic characteristics. When the window parameters are initialized, the basic window length is set according to the maximum response time of the hydropower station unit, and the appropriate data cache size is configured. During operation, the system dynamically adjusts the window size by counting the frequency of faults. When the unit is in normal operating conditions, a larger time window is used to improve the stability of the model; when signs of faults appear, the window size is automatically reduced to increase the response speed of the system. At the same time, the system also implements strict data quality control, including outlier detection, data smoothing, and reliability assessment.

[0126] Finally, the system transfers historical model knowledge through soft label transfer. First, the historical model with the best performance is selected as the teacher model, and soft labels are generated through this model to guide the training of the new model. When training the student model, the system sets reasonable knowledge distillation temperature parameters and introduces regularization constraints to prevent overfitting. In the model fusion stage, the system designs a progressive knowledge transfer mechanism based on operating performance indicators, uses the exponential sliding average method to update model parameters, and dynamically adjusts the weight ratio of new and old knowledge to ensure the stability and effectiveness of model updates.

[0127] In S4, in a pumped storage hydropower station, taking a linear subsystem in a certain operating condition interval as an example, its state space model can be expressed as:

[0128] x(k+1)=Ax(k)+Bu(k)+f

[0129] Among them, the state variable x includes:

[0130] x1: turbine speed (rated value 300rpm)

[0131] x2: Guide vane opening (0-100%)

[0132] x3: Water inlet pipe pressure (rated value 2.5MPa)

[0133] x4: Spindle radial vibration (alarm value 4.5mm / s)

[0134] First, during the initialization phase, the system sets the initial parameter values ​​based on historical data:

[0135] By analyzing the normal operating data of the last 30 days

[0136] Calculate the steady-state values ​​of each parameter as initial estimates

[0137] For example:

[0138] Initial speed value: x1(0) = 300 rpm

[0139] Initial value of guide vane opening: x2(0)=85%

[0140] Initial value of water inlet pipe pressure: x3(0)=2.5MPa

[0141] Initial vibration value: x4(0)=2.0mm / s

[0142] Secondly, when new data arrives, the system performs the following calculation process: 1. Construct the observation equation:

[0143] y(k)=H(k)θ(k)+v(k)

[0144] in:

[0145] y(k) is the measured output vector

[0146] H(k) is the measurement matrix

[0147] θ(k) is the parameter vector to be estimated

[0148] v(k) is the observation noise

[0149] 2. Calculate the prediction error:

[0150]

[0151] For example, the prediction error at a certain moment:

[0152] Speed ​​prediction error: 1.5rpm

[0153] Guide vane opening prediction error: 2.1%

[0154] Pressure prediction error: 0.15MPa

[0155] Vibration prediction error: 0.3mm / s

[0156] 3. Update the gain matrix K(k):

[0157] K(k)=P(k-1)H(k)T[H(k)P(k-1)H(k)T+R(k)]-1 where:

[0158] P(k) is the parameter estimation error covariance matrix

[0159] R(k) is the measurement noise covariance matrix

[0160] 4. Introduce the forgetting factor mechanism:

[0161] λ(k)=λ0*β(k)

[0162] in:

[0163] λ0 is the basic forgetting factor, with a value of 0.98

[0164] β(k) is the adaptive adjustment coefficient

[0165] When the system characteristics change greatly, β(k) takes a smaller value (such as 0.95) to speed up the parameter update;

[0166] When the system characteristics are stable, β(k) takes a larger value (such as 0.99) to maintain the stability of parameter estimation.

[0167] 5. Update parameter estimates:

[0168] θ(k)=θ(k-1)+K(k)e(k)

[0169] 6. Update the error covariance matrix:

[0170] P(k)=[IK(k)H(k)]P(k-1) / λ(k)

[0171] An example is given to illustrate the effect of parameter update:

[0172] Assuming that at a certain moment it is detected that the guide vane opening suddenly drops from 85% to 75%, the system parameter update process is as follows:

[0173] First iteration:

[0174] Prediction error e1: 2.1%

[0175] Forgetting factor λ1: 0.931 (0.98*0.95, due to large characteristic changes)

[0176] Parameter update amount: 1.8%

[0177] Updated estimate: 76.8%

[0178] Second iteration:

[0179] Prediction error e2: 1.2%

[0180] Forgetting factor λ2: 0.931

[0181] Parameter update amount: 0.9%

[0182] Updated estimate: 75.9%

[0183] Third iteration:

[0184] Prediction error e3: 0.4%

[0185] Forgetting factor λ3: 0.970 (0.98*0.99, due to error reduction)

[0186] Parameter update amount: 0.3%

[0187] Updated estimate: 75.2%

[0188] Through this iterative update mechanism, the system is able to:

[0189] 1. Quickly respond to parameter mutations (by adjusting the forgetting factor)

[0190] 2. Suppress the influence of measurement noise (through covariance matrix update)

[0191] 3. Maintaining stability of parameter estimates (through adaptive forgetting factor)

[0192] 4. Adapt to dynamic changes in system characteristics (through continuous parameter updates)

[0193] The advantages of this parameter update method are: small amount of calculation, good real-time performance, strong adaptability, and can meet the needs of online monitoring under complex working conditions of hydropower stations. Through continuous parameter updates, it is ensured that the model can accurately reflect the dynamic characteristics of the system and provide a reliable basis for subsequent fault diagnosis.

[0194] An embodiment of the present application also provides a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned method for dynamic fault analysis of a pumped-storage hydropower station.

[0195] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the above-mentioned method for dynamic fault analysis of a pumped-storage hydropower station.

[0196] An embodiment of the present application also provides a computer program product, including computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the method for dynamic fault analysis of the pumped-storage hydropower station are implemented.

[0197] In the description of this application, the terms "first", "second", etc. are only used to distinguish different entities or operations, rather than indicating or implying the relative importance or implicit order of these entities or operations. Unless otherwise clearly specified and limited, "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0198] The above is only an optional embodiment of the present application, and is not intended to limit the present application. Those skilled in the art can also make various changes and modifications without departing from the spirit and scope of the present application, and these changes and modifications should fall within the scope of protection of the present application. The above is only a preferred embodiment of the present application, and is not intended to limit the scope of the patent of the present application. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application should be included in the scope of patent protection of the present application. It should be understood that although the flowchart of the embodiment of the present application indicates each operation step by arrows, the implementation order of these steps is not limited to the order indicated by the arrows. Unless clearly stated in this article, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be executed in other orders according to demand. In addition, some or all of the steps in each flowchart can include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage in these sub-steps or stages can also be executed at different times. In scenarios with different execution times, the execution order of these sub-steps or stages can be flexibly configured according to needs, and the embodiments of the present application are not limited to this.

[0199] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of the present application, other similar implementation methods based on the technical ideas of the present application are also within the protection scope of the embodiments of the present application.

Claims

1. A method for dynamic fault analysis of a pumped storage hydropower station, characterized in that: The following steps are involved: Collect multi-dimensional status parameters of the hydropower station, including turbine speed, guide vane opening, water pressure and vibration data; Establishing a piecewise affine PWA approximation model, including determining a segmentation point according to a rate of change of the state parameter, using an adaptive meshing algorithm to initially segment the state space, and establishing a linear subsystem model in each segmented region; A hierarchical learning framework is constructed for fault analysis, wherein the hierarchical learning framework includes a feature extraction layer, a fault identification layer, and a cause reasoning layer; wherein: In the feature extraction layer, the state deviation is calculated based on the PWA approximate model, and the time domain and frequency domain features are extracted; In the fault identification layer, the fault confidence is calculated based on the time domain and frequency domain features, valid features are screened according to the set confidence upper limit constraint, the screened features are pattern matched with the fault feature library, and the probability distribution of the fault type is calculated based on the confidence weighting; At the cause reasoning layer, historical fault cases are screened and weighted according to the fault confidence, a fault causal network of the hydropower station system is constructed based on the weighted historical cases, and the root cause and propagation path of the fault are derived in combination with the fault type probability distribution.

2. The method for dynamic fault analysis of a pumped storage hydropower station according to claim 1, characterized in that: The step of establishing a piecewise affine PWA approximation model comprises: Preprocessing of the hydropower station operating parameters, including normalization of turbine speed, median filtering and denoising of guide vane opening signals, interpolation and repair of water pressure data, and sliding window statistical feature calculation of vibration data; Calculate the system dynamic characteristics based on the preprocessed parameters, including calculating the first-order derivatives of the turbine speed and guide vane opening, analyzing the correlation of water pressure changes, and extracting the periodic characteristics of the vibration signal; The spatial segmentation is determined based on the dynamic characteristics, a smaller grid is used in the area where the parameters change drastically, a larger grid is used in the area where the parameters change gently, and adjacent grids with similar dynamic characteristics are merged.

3. The dynamic fault analysis method of a pumped storage hydropower station according to claim 1, characterized in that: The step of calculating the state deviation based on the PWA approximate model and extracting time domain and frequency domain features at the feature extraction layer includes: Calculate the statistical characteristics of the hydropower station state parameters, including calculating the mean and variance of the turbine speed, analyzing the skewness and kurtosis of the guide vane opening, and calculating the correlation coefficient matrix of the water pressure parameters; Extract trend characteristics of hydropower station operation parameters, including calculating turbine speed change rate, detecting guide vane opening change inflection point, and analyzing water pressure parameter change trend; Identify the sudden change characteristics of hydropower station parameters, including detecting abnormal points of vibration signals, calculating the amplitude of water pressure sudden change, and recording the duration of parameter sudden change; Perform frequency domain analysis, including calculating the power spectrum density of the vibration signal, analyzing the main frequency components of the turbine speed, and statistically analyzing the energy distribution of each frequency band.

4. The method for dynamic fault analysis of a pumped storage hydropower station according to claim 1, characterized in that: The step of calculating the fault confidence and determining the fault type based on the time domain and frequency domain features at the fault identification layer includes: Screening based on the confidence upper limit constraints of the hydropower station parameter characteristics, including calculating the confidence interval of the turbine speed characteristics, evaluating the reliability of the guide vane opening characteristics, and analyzing the confidence of the water pressure and vibration characteristics; Constructing the fault feature vector of the hydropower station, including combining time domain and frequency domain features, normalizing the features, and establishing a multi-scale feature representation; Perform fault pattern matching, including calculating the similarity with historical fault cases, calculating the probability based on confidence weights, and outputting the fault type judgment result.

5. The method for dynamic fault analysis of a pumped storage hydropower station according to claim 1, characterized in that: Also includes: The PWA approximate model and hierarchical learning framework are updated online through an incremental learning algorithm, specifically including using recursive least squares method to update the linear subsystem parameters, adaptively adjusting the time window size based on the dynamic characteristics of the system, migrating historical model knowledge to the current model through soft label transfer, and dynamically adjusting the confidence upper limit constraint parameters of each layer according to the model prediction performance.

6. The method for dynamic fault analysis of a pumped storage hydropower station according to claim 5, characterized in that: The step of updating the linear subsystem parameters by using the recursive least squares method comprises: Initialize the hydropower station system parameters, including setting the initial estimated values ​​of the unit speed, guide vane opening, water pressure and vibration parameters; Calculate the prediction error based on new data collected in real time, including comparing the deviation between the measured value and the model prediction value; Update model parameters, including calculating the gain matrix, updating parameter estimates, and introducing a forgetting factor to reduce the weight of historical data.

7. The method for dynamic fault analysis of a pumped storage hydropower station according to claim 5, characterized in that: The step of migrating the historical model knowledge to the current model by soft label transfer includes: Constructing a teacher model, including selecting the best performing model in the historical operation data of the hydropower station as the basic teacher model; Training the student model, including training with soft labels of hydropower station parameters generated by the teacher model, setting the knowledge distillation temperature parameter, and adding regularization constraints; Realize model fusion, including designing a progressive knowledge transfer mechanism based on the hydropower station operation performance indicators, using exponential sliding average to update model parameters, and dynamically adjusting the weights of new and old knowledge.

8. The method for dynamic fault analysis of a pumped storage hydropower station according to claim 5, characterized in that: The step of dynamically adjusting the confidence upper limit constraint parameters of each layer according to the model prediction performance includes: Establish a performance evaluation system, including evaluating the accuracy and recall of hydropower station fault diagnosis, calculating the confidence interval of model predictions, and monitoring the false alarm rate and missed alarm rate; Construct a multi-level constraint transmission mechanism, including setting the basic confidence threshold of each layer, establishing the constraint mapping relationship between the parameter characteristic layers of the hydropower station, and maintaining the consistency of the constraint strength; Perform adaptive adjustments, including calculating performance indicators based on sliding time windows, adjusting constraint strength based on indicator changes, and relaxing constraints appropriately when the system is stable.

9. A method for dynamic fault analysis of a pumped storage hydropower station according to claim 5, characterized in that: The step of adaptively adjusting the time window size based on the system dynamic characteristics comprises: Initialize window parameters, including setting the basic window length according to the maximum response time of the hydropower station unit, setting the sliding step size of the sampling data, and configuring the data cache size; Perform adaptive adjustments, including counting the frequency of hydropower plant failures, using larger windows under normal conditions, and reducing windows to improve response speed under fault conditions; Perform data quality control, including detecting and eliminating abnormal data points, smoothing and filtering the unit operation data, and calculating data reliability indicators.

10. A dynamic fault analysis system for a pumped storage hydropower station, characterized in that: The system comprises a processor and a memory, wherein the memory stores a computer program, and when the processor executes the program, the method according to any one of claims 1 to 9 is implemented.

Images