Hydroelectric generating set deflector working condition abnormity monitoring method based on MFCC and hypersphere model

Abstract

The invention relates to a hydroelectric generating set guide plate working condition abnormity monitoring method based on an MFCC and a hypersphere model, and the method comprises the steps: collecting a voiceprint signal of a volute guide plate operation state, and obtaining a top cover vibration signal, a set bearing throw signal and a pressure pulsation signal as feature data of guide plate health and abnormal state recognition; constructing an optimal hyper-sphere model for quickly diagnosing the fault of the guide plate; lSTM is utilized to calculate and obtain unit key parameters related to the health of the guide plate, namely, guide plate voiceprint and unit vibration, throw and pressure pulsation characteristic data, the difference between the unit key parameters and actual unit characteristic data is calculated to obtain a predicted value residual error of the unit key parameters, and the health degree of the guide plate is calculated and obtained according to the predicted value residual error, the allowable residual error and the maximum residual error of the unit key parameters. And judging whether the flow guide plate has a deterioration trend or a fault symptom according to the change trend of the health degree of the unit flow guide plate. The fault diagnosis of the flow guide plate is realized, and the problem that the water turbine volute flow guide plate is located in the flow channel and cannot be directly monitored is solved.

Description

Technical Field

[0001] The invention belongs to the field of hydropower generation monitoring, and in particular relates to a method for monitoring abnormal working conditions of a guide plate of a hydropower unit based on MFCC and a hypersphere model. Background Art

[0002] Large vertical hydro-turbine generator sets are welded with circular guide plates on the outer circle of the upper and lower rings of the seat ring. The guide plates are connected to the seat ring by welding, and the adjacent guide plates are welded and connected by arc ribs to form a transition component at the connection between the upper and lower rings of the seat ring and the volute. Its main function is to form a stable circulation of the water flow entering the fixed guide vane, and then flow smoothly through the fixed guide vane and enter the runner. Restricted by factors such as the strength of the guide plate itself, under the action of strong water flow, the tearing and falling off of the guide plate become a common defect. The falling off of the guide plate causes the water flow in the volute to be turbulent, aggravates the hydraulic vibration in the volute, and then causes abnormal vibration of the flow-passing components and water guide mechanisms. In extreme cases, the vibration of hydraulic machinery can cause cracks in the turbine components, and even cause fatigue and fracture of the components. Strong vibration will affect the normal operation of the turbine unit, reduce the service life of the unit and some parts; it will also affect the load distribution and power supply quality. When it causes resonance of the plant pressure pipeline and the power system, the unit will not be able to operate normally. If it is not controlled, it will lead to serious power accidents and cause significant economic losses. Therefore, it is of great significance to strengthen the abnormal monitoring of the guide plate working conditions, discover abnormalities as early as possible, and eliminate hidden dangers in time.

[0003] Currently, there is no directional monitoring method or means for the volute guide vane and other flow-passing parts of the hydro-turbine generator set. The most common online monitoring method is the monitoring of the unit vibration, swing and pressure pulsation, including the horizontal and vertical vibration of the top cover in all directions, the swing of the bearings of each part of the hydro-turbine generator set, the volute pressure pulsation, the pressure pulsation of the bladeless area, the pressure pulsation under the top cover, the pressure pulsation of the tail water, etc., combined with the manual inspection method, a comprehensive analysis is carried out.

[0004] The existing monitoring methods for hydropower units have the following shortcomings: (1) The monitoring is not targeted. The monitoring of water pressure pulsation, bearing swing, and flow component vibration in the existing technology can only comprehensively reflect the operating conditions of the unit under the current working conditions. There is a lack of targeted monitoring of the volute guide plate, and it cannot directly reflect the abnormal working conditions of the volute guide plate. For example, among dozens of cases of volute guide plate tearing in a giant hydropower station, there was only one sudden change in the swing of the bearings of various parts of the unit. It can be seen that the monitoring of the swing of the bearings of various parts of the unit cannot effectively identify the abnormality of the guide plate.

[0005] (2) The causes of vibration are complex. The causes of vibration in hydropower station turbines are often difficult to determine. Vibrations of various frequency characteristics are mixed together, which makes it difficult to trace the source of abnormal vibration. In addition, some vibration faults are related to operating parameters and occur more occasionally. It is difficult to capture fault characteristics and accurately identify fault characteristics and make early diagnosis of faults.

[0006] Vibration has limitations relative to acoustic emission. Vibration is a periodic reciprocating motion state of an object under the action of external force, including changes in parameters such as displacement, velocity and acceleration, while acoustic emission is a transient elastic wave field generated by the rapid release of local energy such as defect formation, crack extension, and plastic deformation inside the material. The signal frequency range is wider and the sensitivity is high. Compared with vibration analysis, it can detect fault information in advance. Therefore, the present invention studies the guide plate abnormality diagnosis method based on the guide plate acoustic print signal characteristic data. Summary of the invention

[0007] The purpose of the present invention is to provide a method for monitoring the abnormal working condition of the guide plate of a hydropower unit based on MFCC and a hypersphere model in order to solve the above problems. A full circle of optical fiber soundprint sensors are pre-buried at the connection between the upper and lower rings of the turbine seat ring and the volute, and a contact soundprint sensor is arranged at the volute door to collect the soundprint signal of the running state of the volute guide plate, and obtain the horizontal and vertical vibration signals of the top cover in all directions, the bearing swing of the upper guide, lower guide and water guide of the hydropower generator set, the volute pressure pulsation, and the pressure pulsation signal of the bladeless area as the feature data for identifying the healthy and abnormal state of the guide plate, and to construct an optimal hypersphere model for rapid diagnosis of guide plate faults. ; Use the AI ​​model to calculate the key parameters of the unit related to the health of the guide plate, namely the guide plate soundprint and the unit vibration, swing, and pressure pulsation characteristic data, and subtract them from the actual unit characteristic data to obtain the predicted value residuals of the key parameters of the unit. According to the predicted value residuals, allowable residuals, and maximum residuals of the key parameters of the unit, the health of the guide plate is calculated. According to the changing trend of the health of the unit's guide plate, it is determined whether the guide plate shows a degradation trend or signs of failure. Real-time online monitoring of the unit's guide plate is achieved, guide plate abnormalities are diagnosed and identified, the risk of unexpected shutdown due to guide plate failure is reduced, and the safety and stability of power production are ensured.

[0008] To achieve the above object, the technical solution of the present invention is: The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model includes the following steps: Step 1: Collect the operating status data of the hydropower unit, including the guide plate sound print signal and the unit vibration, swing and pressure pulsation signal data; Step 2: Use the Mel frequency cepstral coefficient method to extract the voiceprint feature vector from the guide plate voiceprint signal; Step 3: Using the historical operating status data of the hydropower unit and the corresponding health status of the guide plate, the optimal hypersphere model of the guide plate is constructed; Step 4: Using the optimal hypersphere model of the guide plate obtained in step 3, according to the real-time operating status data of the unit, preliminarily diagnose and judge whether the current operating status of the unit is faulty, that is, the current health status of the guide plate; Step 5: Construct a prediction model for key parameters of the unit, use historical operating status data and corresponding soundprint feature vectors to train the prediction model for key parameters of the unit, and use the trained prediction model for key parameters of the unit to obtain the predicted values ​​of guide plate soundprint, top cover vibration, unit swing, and pressure pulsation feature data according to real-time operating status data; Step 6: Calculate the residual between the predicted value obtained in step 5 and the corresponding actual unit characteristic data, that is, the predicted value residual; and determine the allowable residual and maximum residual of the key parameters of each unit based on the guide plate historical fault data; Step 7: Using the unit health model, the health of each unit's key parameters is obtained according to the predicted value residual, allowable residual, and maximum residual of the unit's key parameters obtained in step 6; Step 8: Determine the weight of the health of each key parameter to the health of the guide plate; Step 9: According to the health of the key parameters in step 7 and the weight values ​​obtained in step 8, the health of the guide plate under the current working condition is calculated, and compared with the health of the guide plate at the historical moment to determine whether the guide plate shows a deterioration trend.

[0009] Furthermore, in step 1, the operating status data includes water head, guide vane opening, unit active power, unit guide plate sound print signal, and unit vibration, swing and pressure pulsation signal data.

[0010] Preferably, step 2 comprises the following sub-steps: Step 2.1: Preprocessing the voiceprint signal: pre-emphasize, frame and window the guide plate voiceprint signal, filter out the non-audio components in the voiceprint signal and perform truncation processing; Step 2.2: Perform fast Fourier transform on the pre-processed voiceprint signal data to obtain frequency domain feature data in the voiceprint signal; Step 2.3: Take the square of the modulus of the frequency domain feature data of the voiceprint signal; Step 2.4: Perform Mel filtering to convert the frequency in the voiceprint signal into Mel scale; Step 2.5: Take the logarithm of the filtering result obtained in step 2.4 to obtain the feature vector of the voiceprint signal.

[0011] Preferably, in step 3, the construction of the optimal hypersphere model of the guide plate specifically includes: 1) Data preprocessing, Normalize the operating status data of the hydropower unit to eliminate the dimensional differences between different dimensions; 2) Generate the smallest hypersphere, Using the support vector data description method, the preprocessed operating state data of the hydropower unit is learned to generate a minimum hypersphere containing all healthy states, namely, an optimal hypersphere, which surrounds all healthy samples, namely, healthy operating state data samples of the guide plate, and has a radius as small as possible; 3) Initial node selection, In the generated optimal hypersphere, 70% of the nodes are deleted and 30% of the nodes are retained as initial nodes. During the initial node selection process, it is ensured that the deleted nodes can be represented by the nearest nodes. 4) Delete point backfill and distance calculation, Calculate the distance from each deleted point to its nearest initial node; For each deleted point, use the Gaussian membership function to calculate its health status membership HSMD relative to the nearest node; If the relative health status membership of the deleted point is less than the membership threshold, the deleted point is backfilled as a node of the optimal hypersphere; 5) Repeat step 4) until the membership of all deleted points is greater than or equal to the membership threshold, or the upper limit of the number of nodes is reached.

[0012] Furthermore, in step 4, the real-time operating status data of the unit is preprocessed and mapped to the hypersphere space. If the sample point corresponding to the real-time operating status data of the unit is located within the optimal hypersphere, it is judged that the current operating status of the unit is healthy; otherwise, it is judged that the current operating status of the unit is unhealthy or in a faulty state.

[0013] Preferably, in step 5, LSTM is used to construct a key parameter prediction model of the unit, and the input data of the key parameter prediction model of the unit includes the soundprint feature vector, the horizontal vibration and vertical vibration of the top cover in the X direction, the horizontal vibration and vertical vibration of the Y direction, the upper guide X-direction swing and the upper guide Y-direction swing, the lower guide X-direction swing and the lower guide Y-direction swing, the water guide X-direction swing and the water guide Y-direction swing, as well as the volute pressure pulsation, the bladeless area pressure pulsation and the pressure pulsation signal characteristic data under the top cover.

[0014] Furthermore, in step 6, the residual matrix of the key parameters of the unit is calculated based on the prediction results of the key parameter prediction model of the unit, ; Among them, Yres(t) represents the residual matrix of the key parameters of the unit at time t, Yres1 is the residual of the soundprint feature, Yres2 is the residual vector of the top cover vibration, Yres3 is the residual vector of the swing, and Yres4 is the residual vector of the pressure pulsation; The allowable residual matrix of the key parameters of the unit is: ; in, Indicates the allowable residual of the voiceprint feature. represents the allowable residual error of the top cover vibration, represents the allowable residual error of the swing, Indicates the allowable residual of pressure pulsation; The difference between the alarm value of the key parameter of the unit and the predicted value of the corresponding key parameter prediction model of the unit is taken as the maximum residual of the key parameter of the unit. The maximum residual matrix of the key parameter of the unit is: ; in, represents the maximum residual of the voiceprint feature, represents the maximum residual of the top cover vibration, represents the maximum residual error of the swing, Indicates the maximum residual of pressure pulsation.

[0015] Preferably, in step 7, the unit health model uses a quadratic parabolic membership function to calculate the health of key parameters of the unit.

[0016] Preferably, in step 8, the independence weight method is used to determine the weight value corresponding to the health of the key parameters of each unit.

[0017] Preferably, in step 9, the guide plate health degree under the current working condition is calculated, and the calculation formula of the guide plate health degree is: ; In the formula represents the overall health of the guide plate at time t; Indicates the health of key parameters of the unit The corresponding weight value.

[0018] The guide plate operating abnormality monitoring system of the above-mentioned guide plate operating abnormality monitoring method for a hydropower unit comprises: The unit operating status data acquisition module is used to collect the operating status data of the hydropower unit, including water head, guide vane opening, unit active power, unit guide plate sound print signal, and unit vibration, swing and pressure pulsation signal data, and store them in the unit operating status database; A voiceprint feature extraction module is used to extract a voiceprint feature vector from the guide plate voiceprint signal; The optimal hypersphere module constructs an optimal hypersphere containing healthy sample data based on the historical operating status data of the unit and the corresponding health status of the guide plate. The optimal hypersphere is used to quickly diagnose and judge whether the guide plate under the current operating status of the unit is healthy; when the diagnosis identifies a guide plate fault, the guide plate sound pattern, top cover vibration, unit swing, and pressure pulsation characteristic data are stored in the guide plate fault database; The unit key parameter prediction module calculates the predicted values ​​of the unit key parameters related to the health of the guide plate; The residual calculation module calculates the residual, allowable residual and maximum residual of each key parameter of each unit according to the predicted value output by the key parameter prediction module of the unit; The unit key parameter health degree calculation module calculates the health degree of the unit key parameters according to the residual, allowable residual and maximum residual of the unit key parameters output by the residual calculation module; The weight calculation module is used to calculate the health weight of each unit's key parameters; A guide plate health calculation module calculates the overall health of the guide plate according to the health of the key parameters of the unit output by the key parameter health calculation module and the weight of the health of the key parameters of the unit output by the weight calculation module, and stores it in the guide plate health database; Fault symptom judgment module: Based on the overall health output by the guide plate health calculation module, the overall health of the guide plate at the historical time and the time when the guide plate showed signs of fault in the guide plate health database is compared to diagnose and judge whether the guide plate shows deterioration trend or fault signs under the current working condition; when the guide plate fault signs are identified, the time when the fault signs appear, the overall health of the guide plate, and the guide plate sound pattern, top cover vibration, unit swing, and pressure pulsation characteristic data are stored in the fault symptom database; The guide plate abnormality alarm module, based on the diagnosis result of the optimal hypersphere module or the fault symptom judgment module, issues a guide plate fault alarm signal when the diagnosis result is that the guide plate is unhealthy, that is, the guide plate is faulty; when the guide plate shows a deterioration trend or the guide plate shows signs of failure, it issues a guide plate abnormality alarm signal.

[0019] Preferably, the unit operating status database, the guide plate fault database and the fault symptom database are all incremental databases. When a new guide plate fault is diagnosed, the guide plate sound print, top cover vibration, unit swing, and pressure pulsation characteristic data corresponding to the new guide plate fault are stored in the guide plate fault database, and the sample data of the unit operating status database and the fault symptom database are regularly used to retrain the optimal hypersphere model, so that the optimal hypersphere model can adapt to the latest changes of the unit guide plate, thereby ensuring the accuracy and adaptability of the fault diagnosis of the optimal hypersphere model.

[0020] Compared with the prior art, the beneficial effects of the present invention include: 1) The present invention collects the soundprint signal of the running state of the turbine volute guide vane, uses the Mel frequency cepstral coefficient method to extract feature data from the soundprint signal, and combines the top cover vibration, unit swing, and pressure pulsation feature data as feature data for identifying the healthy and abnormal state of the guide vane; constructs an optimal hypersphere model, and uses the optimal hypersphere model to achieve rapid diagnosis of guide vane faults, solving the problem that the turbine volute guide vane is in the flow channel and cannot be directly monitored. The multimodal data of the guide vane soundprint signal feature data combined with the top cover vibration, unit swing, and pressure pulsation feature data is used for guide vane fault diagnosis, which improves the accuracy of fault diagnosis; the soundprint signal has high sensitivity and can identify guide vane faults or fault signs in advance compared to a single vibration signal analysis method.

[0021] 2) The present invention uses the LSTM model to calculate the key parameters of the unit related to the health of the guide plate, namely the guide plate sound print and top cover vibration, unit swing, and pressure pulsation characteristic data, and subtracts them from the actual unit characteristic data to obtain the predicted value residuals of the key parameters of the unit. According to the predicted value residuals, allowable residuals and maximum residuals of the key parameters of the unit, the health of the guide plate is calculated using the unit health model. According to the health of the guide plate of the unit, effective identification and judgment of the deterioration trend or fault signs of the guide plate are achieved, which facilitates timely taking measures against the deterioration trend or fault signs of the guide plate, and can prevent the guide plate from failing in advance and affecting the normal operation of the unit.

[0022] 3) The abnormal operating condition monitoring system of the unit guide plate of the present invention establishes a unit operating status database, a guide plate fault database, a guide plate health database and a fault symptom database, and uses the optimal hypersphere model to quickly diagnose and identify turbine volute guide plate faults; the guide plate health is calculated based on the predicted value residual, allowable residual and maximum residual of the unit key parameters related to the guide plate health, and the guide plate degradation trend and fault symptoms are identified and judged based on the health, thereby realizing real-time online monitoring of the turbine volute guide plate.

[0023] 4) The method and system of the present invention, when diagnosing a guide plate fault, sends an alarm signal to the unit operation and maintenance personnel; when the guide plate shows a deterioration trend or the guide plate shows signs of failure, sends a warning signal to the unit operation and maintenance personnel, so as to facilitate timely handling measures, reduce the risk of hydropower unit shutdown, and improve the safety and reliability of the power system.

[0024] 5) The present invention adopts an incremental database and regularly uses sample data from the unit operating status database and the fault symptom database to retrain the optimal hypersphere model for guide plate fault diagnosis, thereby improving the adaptability of the optimal hypersphere model to the changes of the unit guide plates over time, being able to adapt to the latest changes of the unit guide plates, and ensuring the accuracy and precision of the fault diagnosis of the optimal hypersphere model. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0026] Figure 1 The present invention is a structural diagram of a volute guide plate of a hydropower unit according to an embodiment of the present invention.

[0027] Figure 2 Schematic diagram of a data acquisition system according to an embodiment of the present invention.

[0028] Figure 3 Schematic diagram of the optimal hypersphere model of the guide plate according to an embodiment of the present invention.

[0029] Figure 4 Schematic diagram of an LSTM model according to an embodiment of the present invention.

[0030] Explanation of the reference numerals: upper guide plate 1, lower guide plate 2, volute 3, seat ring 4, optical fiber voiceprint sensor 5. DETAILED DESCRIPTION

[0031] like Figure 1 As shown, a whole circle of optical fiber soundprint sensors 5 are embedded in the connection between the turbine seat ring 4 and the volute 3, and a contact soundprint sensor is arranged at the volute door to collect soundprint signals of the volute guide plate under normal and abnormal conditions, and collect horizontal and vertical vibration signals of the top cover in all directions, bearing swing signals of various parts of the turbine generator set, volute pressure pulsation, bladeless area pressure pulsation signals and other multi-dimensional parameters as feature data for identifying abnormal states of the guide plate. The data acquisition system of the embodiment is as shown in FIG. Figure 2 shown.

[0032] The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model includes the following steps: Step 1: Collect the operating status data of the hydropower unit, water head, guide vane opening, unit active power, unit guide plate sound print signal, and unit vibration, swing and pressure pulsation signal data; Step 2: Use the Mel frequency cepstral coefficient method to extract the voiceprint feature vector from the guide plate voiceprint signal; Step 2.1: Preprocessing the voiceprint signal: pre-emphasize, frame and window the guide plate voiceprint signal, filter out the non-audio components in the voiceprint signal and perform truncation processing; The windowing adopts a Hamming window, and the calculation formula of the Hamming window is:

[0033] In the formula represents the Hamming window function, I Indicates the number of frames, is the coefficient, Indicates the frame number; Step 2.2: Perform fast Fourier transform on the preprocessed voiceprint signal data to obtain the frequency domain feature data in the voiceprint signal. The calculation formula of fast Fourier transform is: ; In the formula is the fast Fourier transform function; Represents the input voiceprint signal, represents the number of points of the fast Fourier transform, n F Indicates the sequence number of the sampling point in the time domain; N Indicates the total number of sampling points; Step 2.3: Take the square of the modulus of the frequency domain feature data of the voiceprint signal; ; In the formula P Represents the spectral line energy of the sound signal, Indicates the first frame, FFT ( ) represents fast Fourier transform; Step 2.4: Perform Mel filtering to convert the frequency in the voiceprint signal into Mel scale; ; In the formula, Mel ( ) represents the Mel function, f represents the frequency in the voiceprint signal, and ln represents the natural logarithmic function; Step 2.5: Take the logarithm of the filtering result obtained in step 2.4 to obtain the feature vector of the voiceprint signal; Among them, the calculation formula of the logarithmic energy of the filter output is: ; In the formula S ( m ) indicates the m The logarithmic energy of the filter output; M is the number of triangular filters; m Indicates the filter number; Represents the frequency response function of a triangular filter.

[0034] Step 3: Use the historical operating status data of the hydropower unit and the corresponding health status of the guide plate to construct the optimal hypersphere model of the guide plate.

[0035] like Figure 3 As shown in Figure 2, the specific process of constructing the optimal hypersphere model of the guide plate includes: 1) Data preprocessing, Normalize the operating status data of the hydropower unit to eliminate the dimensional differences between different dimensions; 2) Generate the smallest hypersphere, Using the support vector data description method, the preprocessed operating state data of the hydropower unit is learned to generate a minimum hypersphere containing all healthy states, namely, an optimal hypersphere, which surrounds all healthy samples, namely, healthy operating state data samples of the guide plate, and has a radius as small as possible; 3) Initial node selection, In the generated optimal hypersphere, 70% of the nodes are deleted and 30% of the nodes are retained as initial nodes. During the initial node selection process, it is ensured that the deleted nodes can be represented by the nearest nodes. 4) Delete point backfill and distance calculation, Calculate the distance from each deleted point to its nearest initial node; For each deleted point, use the Gaussian membership function to calculate its health status membership relative to the nearest node; If the relative health status membership of the deleted point is less than the membership threshold, the deleted point is backfilled as a node of the optimal hypersphere; 5) Repeat step 4) until the membership of all deleted points is greater than or equal to the membership threshold, or the upper limit of the number of nodes is reached.

[0036] Step 4: Using the optimal hypersphere model of the guide plate obtained in step 3, according to the real-time operating status data of the unit, preliminarily diagnose and judge whether the current operating status of the unit is faulty, that is, the current health status of the guide plate; In an embodiment, the real-time operating status data of the unit is preprocessed and mapped to the hypersphere space. If the sample point corresponding to the real-time operating status data of the unit is located within the optimal hypersphere, it is judged that the current operating status of the unit is healthy; otherwise, it is judged that the current operating status of the unit is unhealthy or in a fault state.

[0037] Step 5: Use the long short-term memory network (LSTM) to build a prediction model for key parameters of the unit. Use the historical operating status data and the corresponding soundprint feature vectors to train the prediction model for key parameters of the unit. Use the trained prediction model for key parameters of the unit to obtain the predicted values ​​of the guide vane soundprint, top cover vibration, unit swing, and pressure pulsation feature data based on the real-time operating status data.

[0038] In the embodiment, the input data of the key parameter prediction model of the unit include the soundprint feature vector, the horizontal vibration and vertical vibration of the top cover in the X direction and the horizontal vibration and vertical vibration in the Y direction, the upper guide X-direction swing and the upper guide Y-direction swing, the lower guide X-direction swing and the lower guide Y-direction swing, the water guide X-direction swing and the water guide Y-direction swing, as well as the volute pressure pulsation, the bladeless area pressure pulsation and the pressure pulsation signal characteristic data under the top cover.

[0039] The LSTM model of the embodiment is as follows Figure 4 As shown, each LSTM unit contains input gating, forget gating, selection gating and output gating. Indicates the current input. represents the output of the previous hidden layer, To forget the gate, To select the gate, For input gating, For output gating, Indicates the previous state. Indicates the current state. represents the current hidden layer output, Represents output.

[0040] Forget about gating

[0041] in, is the weight of the forget gate, To forget the bias of the gate, Represents the activation function.

[0042] Select Gating

[0043] in, is the weight of the selected gate, To select the bias of the gate.

[0044] Input Gating

[0045] in, is the input weight, is the input bias, tanh represents the hyperbolic tangent function; Output Gating

[0046] in, is the weight of the output gate, is the bias of the output gate.

[0047] Current Status

[0048] Output of hidden layer nodes and the final output The calculation formula is:

[0049]

[0050] in, is the final output weight, is the bias of the final output.

[0051] Step 6: Calculate the residual between the predicted value obtained in step 5 and the corresponding actual unit characteristic data, that is, the predicted value residual; and determine the allowable residual and maximum residual of the key parameters of each unit based on the guide plate historical fault data; Residual matrix of key parameters of the unit ; Among them, Yres(t) represents the residual matrix of the key parameters of the unit at time t, Yres1 is the residual of the soundprint feature, Yres2 is the residual vector of the top cover vibration, Yres3 is the residual vector of the swing, and Yres4 is the residual vector of the pressure pulsation; The allowable residual matrix of the key parameters of the unit is: ; in, Indicates the allowable residual of the voiceprint feature. represents the allowable residual error of the top cover vibration, represents the allowable residual error of the swing, Indicates the allowable residual of pressure pulsation; The difference between the alarm value of the key parameter of the unit and the predicted value of the corresponding key parameter prediction model of the unit is taken as the maximum residual of the key parameter of the unit. The maximum residual matrix of the key parameter of the unit is: ; in, represents the maximum residual of the voiceprint feature, represents the maximum residual of the top cover vibration, represents the maximum residual error of the swing, Indicates the maximum residual of pressure pulsation.

[0052] Step 7: Using the unit health model, the health of each unit's key parameters is obtained according to the predicted value residual, allowable residual, and maximum residual of the unit's key parameters obtained in step 6; In the embodiment, the unit health model uses a quadratic parabolic membership function to calculate the health of the key parameters of the unit. The quadratic parabolic membership function is: ; In the formula Indicates Among the key parameters of this type of unit The healthiness of each portion, Corresponding to sound pattern, vibration, swing, and pressure pulsation respectively. j is the component number, Indicates Among the key parameters of this type of unit The residual of the components, Indicates Among the key parameters of this type of unit The allowed residual of the components, Indicates Among the key parameters of this type of unit The maximum residual of the components.

[0053] Step 8: Use the independence weight method to determine the health of key parameters of the unit The corresponding weight value .

[0054] The independence weight method is an objective weighting method that uses the strength of collinearity between indicators to determine the weight. If an indicator is highly correlated with other indicators, it means that there is a large overlap in information, and the weight of the indicator will be relatively low; conversely, if an indicator is weakly correlated with other indicators, it means that the indicator carries a large amount of information, and a higher weight should be given to the indicator.

[0055] In one embodiment, the complex correlation coefficient R is used to measure the health of the key parameters of the unit. The strength of collinearity between indicators is determined by using the inverse of the multiple correlation coefficient 1 / R as an intermediate variable to determine the weight. The closer the multiple correlation coefficient R is to 1, the stronger the collinearity between indicators and the higher the degree of information overlap. Therefore, the smaller its inverse 1 / R is, the lower the weight is.

[0056] Step 9: According to the health of the key parameters in step 7 and the weight values ​​obtained in step 8, the health of the guide plate under the current working condition is calculated. The calculation formula of the health of the guide plate is: ; In the formula Represents the overall health of the guide plate at time t.

[0057] By comparing the health of the guide plate at the previous moment, it is determined whether the guide plate has a deterioration trend, or by comparing the health of the guide plate at the time when the historical fault symptom appeared, it is determined whether the guide plate has a fault symptom.

[0058] The guide plate abnormal condition monitoring system based on the above method includes the following modules: The unit operating status data acquisition module is used to collect the operating status data of the hydropower unit, including water head, guide vane opening, unit active power, unit guide plate sound print signal, and unit vibration, swing and pressure pulsation signal data, and store them in the unit operating status database; A voiceprint feature extraction module is used to extract a voiceprint feature vector from the guide plate voiceprint signal; The optimal hypersphere module constructs an optimal hypersphere containing healthy sample data based on the historical operating status data of the unit and the corresponding health status of the guide plate. The optimal hypersphere is used to quickly diagnose and judge whether the guide plate under the current operating status of the unit is healthy; when the diagnosis identifies a guide plate fault, the guide plate sound pattern, top cover vibration, unit swing, and pressure pulsation characteristic data are stored in the guide plate fault database; The unit key parameter prediction module calculates the predicted values ​​of the unit key parameters related to the health of the guide plate; The residual calculation module calculates the residual, allowable residual and maximum residual of each key parameter of each unit according to the predicted value output by the key parameter prediction module of the unit; The unit key parameter health degree calculation module calculates the health degree of the unit key parameters according to the residual, allowable residual and maximum residual of the unit key parameters output by the residual calculation module; The weight calculation module is used to calculate the health weight of each unit's key parameters; A guide plate health calculation module calculates the overall health of the guide plate according to the health of the key parameters of the unit output by the key parameter health calculation module and the weight of the health of the key parameters of the unit output by the weight calculation module, and stores it in the guide plate health database; Fault symptom judgment module: Based on the overall health output by the guide plate health calculation module, the overall health of the guide plate at the historical time and the time when the guide plate showed signs of fault in the guide plate health database is compared to diagnose and judge whether the guide plate shows deterioration trend or fault signs under the current working condition; when the guide plate fault signs are identified, the time when the fault signs appear, the overall health of the guide plate, and the guide plate sound pattern, top cover vibration, unit swing, and pressure pulsation characteristic data are stored in the fault symptom database; The guide plate abnormality alarm module, based on the diagnosis result of the optimal hypersphere module or the fault symptom judgment module, issues a guide plate fault alarm signal when the diagnosis result is that the guide plate is unhealthy, that is, the guide plate is faulty; when the guide plate shows a deterioration trend or the guide plate shows signs of failure, it issues a guide plate abnormality alarm signal.

[0059] The present invention introduces a soundprint sensor into the traditional online monitoring equipment of the hydropower unit, which greatly expands the monitoring range of the signal frequency of the unit equipment, helps to more comprehensively monitor the operating conditions of the unit equipment and perceive abnormal conditions of the unit equipment; the acoustic emission signal has high sensitivity, and compared with the single vibration signal analysis method, it can discover fault information in advance, as well as the comprehensive detection of multi-modal data such as pressure pulsation, vibration, swing, etc., which can improve the accuracy and timeliness of online monitoring and fault warning.

[0060] The volute guide plate is an important flow-passing component. The present invention solves the problem that the turbine volute guide plate is in the flow channel and cannot be directly monitored. The present invention monitors by multiple means such as pre-embedded optical fiber soundprint sensors, which increases the targeted nature of the monitoring and has a significant driving effect on directly and accurately judging the abnormality of the volute guide plate.

Claims

1. A method for monitoring abnormal working condition of guide plates of a unit based on MFCC and optimal hypersphere model, characterized in that: The following steps are involved: Step 1: Collect the operating status data of the hydropower unit, including the guide plate sound print signal and the unit vibration, swing and pressure pulsation signal data; Step 2: Use the Mel frequency cepstral coefficient method to extract the voiceprint feature vector from the guide plate voiceprint signal; Step 3: Using the historical operating status data of the hydropower unit and the corresponding health status of the guide plate, the optimal hypersphere model of the guide plate is constructed; Step 4: Using the optimal hypersphere model of the guide plate obtained in step 3, according to the real-time operating status data of the unit, preliminarily diagnose and judge whether the current operating status of the unit is faulty, that is, the current health status of the guide plate; Step 5: Construct a prediction model for key parameters of the unit, use historical operating status data and corresponding soundprint feature vectors to train the prediction model for key parameters of the unit, and use the trained prediction model for key parameters of the unit to obtain the predicted values ​​of guide plate soundprint, top cover vibration, unit swing, and pressure pulsation feature data according to real-time operating status data; Step 6: Calculate the residual between the predicted value obtained in step 5 and the corresponding actual unit characteristic data, that is, the predicted value residual; And according to the guide plate historical fault data, determine the allowable residual and maximum residual of each unit's key parameters; Step 7: Using the unit health model, the health of each unit's key parameters is obtained according to the predicted value residual, allowable residual, and maximum residual of the unit's key parameters obtained in step 6; Step 8: Determine the weight of the health of each key parameter to the health of the guide plate; Step 9: According to the health of the key parameters in step 7 and the weight values ​​obtained in step 8, the health of the guide plate under the current working condition is calculated, and compared with the health of the guide plate at the historical moment to determine whether the guide plate shows a deterioration trend.

2. The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model according to claim 1 is characterized in that: In step 1, the operating status data includes water head, guide vane opening, unit active power, unit guide plate sound print signal, and unit vibration, swing and pressure pulsation signal data.

3. The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model according to claim 2 is characterized in that: The step 2 includes the following sub-steps: Step 2.1: Preprocessing the voiceprint signal: pre-emphasize, frame and window the guide plate voiceprint signal, filter out the non-audio components in the voiceprint signal and perform truncation processing; The windowing adopts a Hamming window, and the calculation formula of the Hamming window is: In the formula represents the Hamming window function, I Indicates the number of frames, is the coefficient, Indicates the frame number; Step 2.2: Perform fast Fourier transform on the preprocessed voiceprint signal data to obtain the frequency domain feature data in the voiceprint signal. The calculation formula of fast Fourier transform is: ； In the formula is the fast Fourier transform function; Represents the input voiceprint signal, represents the number of points of the fast Fourier transform, n F Indicates the sequence number of the sampling point in the time domain; N Indicates the total number of sampling points; Step 2.3: Take the square of the modulus of the frequency domain feature data of the voiceprint signal; ； In the formula P Represents the spectral line energy of the sound signal, Indicates the first frame, FFT ( ) represents fast Fourier transform; Step 2.4: Perform Mel filtering to convert the frequency in the voiceprint signal into Mel scale; ； In the formula, Mel ( ) represents the Mel function, f represents the frequency in the voiceprint signal, and ln represents the natural logarithmic function; Step 2.5: Take the logarithm of the filtering result obtained in step 2.4 to obtain the feature vector of the voiceprint signal; Among them, the calculation formula of the logarithmic energy of the filter output is: ； In the formula S ( m ) indicates the m The logarithmic energy of the filter output; M is the number of triangular filters; m Indicates the filter number; Represents the frequency response function of a triangular filter.

4. The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model according to claim 2 or 3 is characterized in that: In step 3, the optimal hypersphere model of the guide plate is constructed, and the specific process includes: 1) Data preprocessing, Normalize the operating status data of the hydropower unit to eliminate the dimensional differences between different dimensions; 2) Generate the smallest hypersphere, Using the support vector data description method, the preprocessed operating state data of the hydropower unit is learned to generate a minimum hypersphere containing all healthy states, namely, an optimal hypersphere, which surrounds all healthy samples, namely, healthy operating state data samples of the guide plate, and has a radius as small as possible; 3) Initial node selection, In the generated optimal hypersphere, 70% of the nodes are deleted and 30% of the nodes are retained as initial nodes. During the initial node selection process, it is ensured that the deleted nodes can be represented by the nearest nodes. 4) Delete point backfill and distance calculation, Calculate the distance from each deleted point to its nearest initial node; For each deleted point, use the Gaussian membership function to calculate its health status membership relative to the nearest node; If the relative health status membership of the deleted point is less than the membership threshold, the deleted point is backfilled as a node of the optimal hypersphere; 5) Repeat step 4) until the membership of all deleted points is greater than or equal to the membership threshold, or the upper limit of the number of nodes is reached.

5. The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model according to claim 4 is characterized in that: In step 4, the real-time operating status data of the unit is pre-processed and mapped to the hypersphere space. If the sample point corresponding to the real-time operating status data of the unit is located in the optimal hypersphere, it is judged that the current operating status of the unit is healthy; otherwise, it is judged that the current operating status of the unit is unhealthy or faulty.

6. The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model according to claim 5 is characterized in that: In the step 5, LSTM is used to construct a key parameter prediction model of the unit. The input data of the key parameter prediction model of the unit includes the soundprint feature vector, the horizontal vibration and vertical vibration of the top cover in the X direction and the horizontal vibration and vertical vibration in the Y direction, the upper guide bearing X-direction swing and the lower guide bearing Y-direction swing, the water guide bearing X-direction swing and the Y-direction swing, and the volute pressure pulsation, the bladeless area pressure pulsation and the pressure pulsation signal feature data under the top cover.

7. The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model according to claim 6 is characterized in that: In step 6, the residual matrix of the key parameters of the unit is calculated based on the prediction results of the key parameter prediction model of the unit. ； Among them, Yres(t) represents the residual matrix of the key parameters of the unit at time t, Yres1 is the residual of the soundprint feature, Yres2 is the residual vector of the top cover vibration, Yres3 is the residual vector of the swing, and Yres4 is the residual vector of the pressure pulsation; The allowable residual matrix of the key parameters of the unit is: ； in, Indicates the allowable residual of the voiceprint feature. represents the allowable residual error of the top cover vibration, represents the allowable residual error of the swing, Indicates the allowable residual of pressure pulsation; The difference between the alarm value of the key parameter of the unit and the predicted value of the corresponding key parameter prediction model of the unit is taken as the maximum residual of the key parameter of the unit. The maximum residual matrix of the key parameter of the unit is: ； in, represents the maximum residual of the voiceprint feature, represents the maximum residual of the top cover vibration, represents the maximum residual error of the swing, Indicates the maximum residual of pressure pulsation.

8. The method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model according to claim 7 is characterized in that: In step 7, the unit health model uses a quadratic parabolic membership function to calculate the health of the key parameters of the unit, and the quadratic parabolic membership function is: ； In the formula Indicates Among the key parameters of this type of unit The healthiness of each portion, Corresponding to sound pattern, vibration, swing, and pressure pulsation respectively. j is the component number, Indicates Among the key parameters of this type of unit The residual of the components, Indicates Among the key parameters of this type of unit The allowed residual of the components, Indicates Among the key parameters of this type of unit The maximum residual of the components.

9. According to the method for monitoring abnormal working condition of guide plate of hydropower unit based on MFCC and hypersphere model in claim 8, in step 9, the health of the guide plate under the current working condition is calculated, and the calculation formula of the health of the guide plate is: ； In the formula represents the overall health of the guide plate at time t, Indicates the health of key parameters of the unit The corresponding weight value.

10. The system of the method for monitoring abnormal working condition of guide plate of a unit according to claim 2 or 3 or 5 or 6 or 7 or 8 or 9, characterized in that: include: The unit operating status data acquisition module is used to collect the operating status data of the hydropower unit, including water head, guide vane opening, unit active power, unit guide plate sound print signal, and unit vibration, swing and pressure pulsation signal data, and store them in the unit operating status database; A voiceprint feature extraction module is used to extract a voiceprint feature vector from the guide plate voiceprint signal; The optimal hypersphere module constructs an optimal hypersphere containing healthy sample data based on the historical operating status data of the unit and the corresponding health status of the guide plate. The optimal hypersphere is used to quickly diagnose and judge whether the guide plate under the current operating status of the unit is healthy; when the diagnosis identifies a guide plate fault, the guide plate sound pattern, top cover vibration, unit swing, and pressure pulsation characteristic data are stored in the guide plate fault database; The unit key parameter prediction module calculates the predicted values ​​of the unit key parameters related to the health of the guide plate; The residual calculation module calculates the residual, allowable residual and maximum residual of each key parameter of each unit according to the predicted value output by the key parameter prediction module of the unit; The unit key parameter health degree calculation module calculates the health degree of the unit key parameters according to the residual, allowable residual and maximum residual of the unit key parameters output by the residual calculation module; The weight calculation module is used to calculate the health weight of each unit's key parameters; A guide plate health calculation module calculates the overall health of the guide plate according to the health of the key parameters of the unit output by the key parameter health calculation module and the weight of the health of the key parameters of the unit output by the weight calculation module, and stores it in the guide plate health database; Fault symptom judgment module: Based on the overall health output by the guide plate health calculation module, the overall health of the guide plate at the historical time and the time when the guide plate showed signs of fault in the guide plate health database is compared to diagnose and judge whether the guide plate shows deterioration trend or fault signs under the current working condition; when the guide plate fault signs are identified, the time when the fault signs appear, the overall health of the guide plate, and the guide plate sound pattern, top cover vibration, unit swing, and pressure pulsation characteristic data are stored in the fault symptom database; The guide plate abnormality alarm module, based on the diagnosis result of the optimal hypersphere module or the fault symptom judgment module, issues a guide plate fault alarm signal when the diagnosis result is that the guide plate is unhealthy, that is, the guide plate is faulty; when the guide plate shows a deterioration trend or the guide plate shows signs of failure, it issues a guide plate abnormality alarm signal.

Images