Cavitation state analysis method, system and equipment for mixed-flow water turbine and medium

The pressure monitoring data of mixed flow turbines was analyzed through fractional Fourier transform and parallel CNN-BiLSTM model, and its cavitation state was extracted, which solved the problems of energy loss and safe vibration in the cavitation state of mixed flow turbines, and achieved early detection and The effect of avoiding accidents.

CN119066565APending Publication Date: 2024-12-03CHINA THREE GORGES CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411058768.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

When a hybrid flow turbine operates in a cavitation state, it will lead to increased energy loss, decreased efficiency and dangerous vibrations, and it will be difficult to detect cavitation state early, which may cause irreversible accidents.

Method used

Using a method based on fractional Fourier transform and parallel CNN-BiLSTM, the pressure monitoring data between the rotor and the guide vane are obtained through sensors, time slice and fractional Fourier transform are performed, and a single-channel convolutional neural network CNN and The single-channel BiLSTM network analyzes the real and imaginary data and extracts the cavitation state of the mixed flow turbine.

Benefits of technology

It can detect the cavitation phenomenon of the turbine in the early stage, extract information on abnormal pressure fluctuations, avoid the occurrence of major safety accidents, and provide important information for the maintenance of the turbine status.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066565A_ABST
    Figure CN119066565A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of state maintenance of hydroelectric power generation equipment, and provides a mixed-flow water turbine cavitation state analysis method, system and equipment and a medium. The method comprises the steps that pressure monitoring data between a rotating wheel and a guide vane is obtained through a sensor; processing the pressure monitoring data; performing fractional Fourier transform on the processed pressure monitoring data to obtain real part data and imaginary part data; the real part data and the imaginary part data are input into a trained parallel CNN-Bi LSTM model for analysis, and the cavitation erosion state of the mixed-flow water turbine is obtained; wherein the parallel CNN-Bi LSTM model comprises a parallel single-channel convolutional neural network CNN and a single-channel Bi LSTM network, real part data and imaginary part data can be obtained by performing fractional Fourier transform on pressure monitoring data, and feature extraction is performed on the data from a time domain and a frequency domain through the parallel CNN-Bi LSTM model, so that the real part data and imaginary part data can be obtained. The cavitation erosion phenomenon of the water turbine can be found as early as possible.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of condition maintenance of hydropower equipment, and in particular relates to a method, system, equipment and medium for analyzing the cavitation state of a Francis turbine. Background Art

[0002] Francis turbines are key equipment in hydropower generation systems. During turbine operation, they often face severe erosion challenges caused by the collapse of cavitation bubbles in the water flow. As time accumulates, part of the surface of the turbine runner will experience a process of fading from gloss, turning gray, and then forming pitting, and finally evolving into sponge-like or honeycomb-like damage, which is called turbine cavitation.

[0003] When a Francis turbine is running in a cavitation state, the turbine in the cavitation state will destroy the normal flow of water, increase energy loss, and significantly reduce the efficiency and output of the turbine. In addition, the turbine in the cavitation state is prone to low-frequency and large-amplitude pressure pulsation, which in severe cases will cause dangerous vibrations of the entire unit and the hydropower plant building. Summary of the invention

[0004] In response to the problems in the background technology, the present invention proposes a cavitation state analysis method for a mixed flow turbine based on fractional-order Fourier transform and parallel CNN-BiLSTM, and uses fractional-order Fourier transform and parallel CNN-BiLSTM to analyze abnormal pressure pulsation to detect turbine cavitation as early as possible and avoid the system from generating larger pressure pulses and causing irreversible accidents.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention proposes a method for analyzing the cavitation state of a Francis turbine based on fractional-order Fourier transform and parallel CNN-BiLSTM, comprising:

[0007] Obtain pressure monitoring data between the runner and the guide vane through sensors;

[0008] processing the pressure monitoring data;

[0009] Performing a fractional Fourier transform on the processed pressure monitoring data to obtain real part data and imaginary part data;

[0010] The real part data and the imaginary part data are input into a trained parallel CNN-BiLSTM model for analysis to obtain the cavitation state of the mixed flow turbine; wherein the parallel CNN-BiLSTM model includes a parallel single-channel convolutional neural network CNN and a single-channel BiLSTM network.

[0011] Furthermore, performing time series processing on the pressure monitoring data includes:

[0012] The pressure monitoring data is time sliced.

[0013] Further, time slicing the pressure monitoring data includes:

[0014] Setting the starting time of the pressure monitoring data and the time window length of the slice;

[0015] Slicing the pressure monitoring data based on the start time of the pressure monitoring data and the time window length of the slice to obtain a data segment;

[0016] A total number of slices is determined based on the data segment.

[0017] Furthermore, the processed pressure monitoring data is subjected to a fractional Fourier transform to obtain real part data and imaginary part data, including:

[0018] The processed pressure monitoring data is subjected to a p-order fractional Fourier transform, and the expression is:

[0019]

[0020] Among them, F p is the fractional Fourier operator, K p (t,u) is the transformation kernel, p is the fractional order, x i (t) is the i-th slice signal, u is the variable after the fractional Fourier transform of the time independent variable t;

[0021] Transformation kernel K p The expression of (t,u) is:

[0022]

[0023] in, α=pπ / 2 represents the rotation angle of the time-frequency plane.

[0024] Further, the real part data and the imaginary part data are input into the parallel CNN-BiLSTM model, and the cavitation state of the Francis turbine is obtained, including:

[0025] Using a single-channel convolutional neural network (CNN) to process the real part data to obtain a first output result;

[0026] Using a single-channel BiLSTM network to process the imaginary part data to obtain a second output result;

[0027] The first output result and the second output result are integrated to obtain the cavitation state of the Francis turbine.

[0028] Further, training the parallel CNN-BiLSTM model includes:

[0029] Acquire monitoring data samples; the monitoring data samples include the pressure between the runner and the guide vane;

[0030] Time slicing the monitoring data samples;

[0031] Performing a fractional Fourier transform on the pressure monitoring data after the time slice to obtain real part data and imaginary part data;

[0032] Dividing the real part data and the imaginary part data into a training set and a test set respectively;

[0033] A single-channel convolutional neural network CNN is used to train the signal samples of the real part data training set, and a single-channel BiLSTM network is used to train the imaginary part data training set to obtain a trained CNN-BiLSTM model;

[0034] The trained CNN-BiLSTM model is tested using the real part data test set and the imaginary part data test set.

[0035] Furthermore, the ratio of the training set to the test set is 8:2.

[0036] In a second aspect, the present invention proposes a Francis turbine cavitation state analysis system based on fractional Fourier transform and parallel CNN-BiLSTM, comprising:

[0037] An acquisition module, used for acquiring pressure monitoring data between the runner and the guide vane through a sensor;

[0038] A processing module, used for processing the pressure monitoring data;

[0039] A fractional Fourier transform module, used for performing a fractional Fourier transform on the processed pressure monitoring data to obtain real part data and imaginary part data;

[0040] The analysis module is used to input the real part data and the imaginary part data into the trained CNN-BiLSTM model for analysis to obtain the cavitation state of the Francis turbine.

[0041] In a third aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon.

[0042] When the computer instructions are executed, the above-mentioned Francis turbine cavitation state analysis method based on fractional-order Fourier transform and parallel CNN-BiLSTM can be realized.

[0043] In a fourth aspect, the present invention proposes a device comprising a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, and wherein the processor executes the computer instructions using the above-mentioned mixed flow turbine cavitation state analysis method based on fractional-order Fourier transform and parallel CNN-BiLSTM.

[0044] Beneficial effects of the present invention:

[0045] 1. The method of the present invention can obtain real part data and imaginary part data by performing fractional Fourier transform on pressure monitoring data, and extract features of the data from the time domain and frequency domain through a parallel CNN-BiLSTM model, which is conducive to early detection of cavitation phenomenon of turbines.

[0046] 2. The method of the present invention adopts two different deep learning models to perform feature extraction and state recognition on the real part data and the imaginary part data respectively, and combines the results of the two deep learning models to finally output the diagnosis result, which is conducive to extracting abnormal pressure fluctuation information and timely discovering abnormal pressure pulsation, which can effectively avoid the occurrence of major safety accidents and provide important information for the status maintenance of the turbine.

[0047] 3. The method of the present invention ensures the integrity of signal characteristics by first slicing the signal and then performing fractional Fourier transform on the signal.

[0048] Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A flow chart of a Francis turbine cavitation state analysis method based on fractional-order Fourier transform and parallel CNN-BiLSTM of the present invention is shown;

[0051] Figure 2 The CNN-BiLSTM model structure diagram of the present invention is shown;

[0052] Figure 3The framework diagram of the Francis turbine cavitation state analysis system based on fractional-order Fourier transform and parallel CNN-BiLSTM of the present invention is shown;

[0053] Figure 4 A schematic diagram of the device structure of the present invention is shown. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Refer to the instruction manual Figure 1 As shown, the present invention discloses a cavitation state analysis method for a Francis turbine based on fractional Fourier transform and parallel CNN-BiLSTM, which uses fractional Fourier transform and a parallel CNN-BiLSTM deep learning model to fuse and process the transformed data features and extract abnormal pressure fluctuation information.

[0056] The cavitation state analysis method of Francis turbine based on fractional Fourier transform and parallel CNN-BiLSTM specifically includes the following steps:

[0057] Step S1, obtaining pressure monitoring data s(t) between the runner and the guide vane through a sensor;

[0058] In step S1 , the pressure monitoring data obtained is the basis for subsequent cavitation analysis.

[0059] Step S2, processing the acquired pressure monitoring data s(t) between the runner and the guide vane;

[0060] In step S2, if the obtained operating data of the pressure monitoring data s(t) between the runner and the guide vane is long, it will be unfavorable for the processing of the model. Therefore, the pressure monitoring data needs to be processed first.

[0061] In a specific embodiment, the method for processing the pressure monitoring data s(t) is time slicing, and the time slicing method specifically includes the following steps:

[0062] Step S21, determining the starting time of the pressure monitoring data s(t), and determining the time window length L of the pressure monitoring data s(t) slice;

[0063] Step S22: Slice the pressure monitoring data s(t), the data segment is: s 1(t)—s 1 (t+L), s 2 (t+1)—s 2 (t+1+L), s 3 (t+2)—s 3 (t+2+L),…s i (t+i-1)—s i (t+i-1+L);

[0064] Step S23, determining that the total number of signal slices is M, where M is an even number, i∈M;

[0065] Step S3: Perform a fractional Fourier transform on the processed pressure monitoring data s(t) between the runner and the guide vane to obtain the real part data and the imaginary part data

[0066] Firstly, the signal is sliced ​​and then fractional Fourier transform is performed on the signal to ensure the integrity of the pressure monitoring data characteristics.

[0067] Fractional Fourier transform (FRFT) is a unified time-frequency transform that reflects the information of the signal in both the time domain and the frequency domain. Compared with the traditional Fourier transform, it is more suitable for processing non-stationary data. After performing fractional Fourier transform on the data, two sets of data, the real part and the imaginary part, will be obtained.

[0068] In a specific embodiment, the processed pressure monitoring data s(t) between the runner and the guide vane is subjected to a fractional Fourier transform, which specifically includes the following steps:

[0069] Step S31, perform a p-order fractional Fourier transform on the processed pressure monitoring data to obtain:

[0070]

[0071] In formula (1), F p is the fractional Fourier operator, K p (t,u) is the transformation kernel, p is the fractional order, x i (t) is the i-th slice signal, and u is the variable after the fractional Fourier transform of the time independent variable t.

[0072] Transformation kernel K p The expression of (t,u) is:

[0073]

[0074] In formula (2), α=pπ / 2 represents the rotation angle of the time-frequency plane.

[0075] After the above fractional Fourier transform, the pressure monitoring data will be divided into two parts, the real part and the imaginary part, namely the real part data and the imaginary part data, which are recorded as: and

[0076] Step S4, input the real part data and the imaginary part data into the trained CNN-BiLSTM model for analysis to obtain the cavitation state of the Francis turbine. Both the real part data and the imaginary part data contain important turbine state information, so it is necessary to use two different deep learning models for feature extraction and state recognition from the time domain and frequency domain for the real part data and the imaginary part data, namely, a single-channel convolutional neural network CNN and a single-channel BiLSTM network.

[0077] Refer to the instruction manual Figure 2 As shown, attached Figure 2 The schematic diagram of the structure of the trained CNN-BiLSTM model is shown in the figure. The parallel CNN-BiLSTM model is composed of a single-channel convolutional neural network CNN and a single-channel BiLSTM network, and the single-channel convolutional neural network CNN and the single-channel BiLSTM network work in parallel. The real part of the data passes through the convolution layer-regularization layer-ReLu activation layer-maximum pooling layer, and then passes through the full connection layer for output. It can be understood that the convolution layer-regularization layer-ReLu activation layer-maximum pooling layer can be repeated multiple times, and different numbers or sizes of convolution kernels may be used each time; after the imaginary part of the data passes through the forward LSTM and the backward LSTM, the results are spliced ​​and finally output through the full connection layer; it should be further explained that the parallel CNN-BiLSTM model in the prior art uses the output of the CMM model as the input of the BiLSTM, that is, the "relay" method is adopted, and in this application, a parallel structure is used to train the real part of the data and the imaginary part of the data in parallel, and finally the output results are fused together.

[0078] In a specific embodiment, the real part data and the imaginary part data are input into a trained CNN-BiLSTM model for analysis, which specifically includes the following steps:

[0079] S41, using a single-channel convolutional neural network CNN to process the real part data to obtain a first output result;

[0080] S42, using a single-channel BiLSTM network to process the imaginary part data to obtain a second output result;

[0081] S43, integrating the first output result and the second output result to obtain the cavitation state of the Francis turbine.

[0082] In a specific embodiment, training the CNN-BiLSTM model specifically includes the following steps:

[0083] S51, obtaining a monitoring data sample; the monitoring data sample includes the pressure between the runner and the guide vane;

[0084] S52, time slicing the monitoring data samples; wherein the method of time slicing is the same as the method of processing the pressure monitoring data s(t) described above, and will not be described in detail here.

[0085] In step S52, the historical operating data in the monitoring data sample is long, which is not conducive to direct processing and deep learning model training. Therefore, it is necessary to perform slicing processing first, and sufficient data samples can be obtained according to actual needs.

[0086] S53, performing a fractional Fourier transform on the pressure monitoring data after the time slice to obtain real part data and imaginary part data;

[0087] S54, dividing the real part data and the imaginary part data into a training set and a test set respectively; wherein the ratio of the training set to the test set is 8:2.

[0088] S55, using a single-channel convolutional neural network CNN to train the signal samples of the real part data training set, and using a single-channel BiLSTM network to train the imaginary part data training set to obtain a trained CNN-BiLSTM model;

[0089] S56. Use the real part data test set and the imaginary part data test set to test the trained CNN-BiLSTM model.

[0090] Based on the same inventive concept, the present invention also proposes a Francis turbine cavitation state analysis system based on fractional-order Fourier transform and parallel CNN-BiLSTM, refer to the attached manual Figure 3 As shown, including,

[0091] An acquisition module, used for acquiring pressure monitoring data between the runner and the guide vane through a sensor;

[0092] A processing module, used for processing pressure monitoring data;

[0093] A fractional Fourier transform module is used to perform a fractional Fourier transform on the processed pressure monitoring data to obtain real part data and imaginary part data;

[0094] The analysis module is used to input the real part data and the imaginary part data into the trained CNN-BiLSTM model for analysis to obtain the cavitation state of the Francis turbine.

[0095] Based on the same inventive concept, the present invention also proposes a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed, the above-mentioned mixed flow turbine cavitation state analysis method based on fractional-order Fourier transform and parallel CNN-BiLSTM can be implemented.

[0096] Based on the same inventive concept, the present invention also proposes a device, referring to the attached specification Figure 4 As shown, it includes a memory and a processor, the memory stores computer instructions that can be run on the processor, and the processor executes the above-mentioned mixed flow turbine cavitation state analysis method based on fractional-order Fourier transform and parallel CNN-BiLSTM when running the computer instructions.

[0097] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0098] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A Francis turbine cavitation state analysis method based on fractional Fourier transform and parallel CNN-BiLSTM, characterized in that: include: Obtain pressure monitoring data between the runner and the guide vane through sensors; processing the pressure monitoring data; Performing a fractional Fourier transform on the processed pressure monitoring data to obtain real part data and imaginary part data; The real part data and the imaginary part data are input into a trained parallel CNN-BiLSTM model for analysis to obtain the cavitation state of the mixed flow turbine; wherein the parallel CNN-BiLSTM model includes a parallel single-channel convolutional neural network CNN and a single-channel BiLSTM network.

2. The method for analyzing cavitation state of a Francis turbine based on fractional Fourier transform and parallel CNN-BiLSTM according to claim 1, characterized in that: The time series processing of the pressure monitoring data includes: The pressure monitoring data is time sliced.

3. The method for analyzing cavitation state of a Francis turbine based on fractional Fourier transform and parallel CNN-BiLSTM according to claim 2, characterized in that: Time slicing the pressure monitoring data includes: Setting the starting time of the pressure monitoring data and the time window length of the slice; Slicing the pressure monitoring data based on the start time of the pressure monitoring data and the time window length of the slice to obtain a data segment; A total number of slices is determined based on the data segment.

4. The method for analyzing cavitation state of a Francis turbine based on fractional Fourier transform and parallel CNN-BiLSTM according to claim 1, characterized in that: Performing a fractional Fourier transform on the processed pressure monitoring data to obtain real part data and imaginary part data includes: The processed pressure monitoring data is subjected to a p-order fractional Fourier transform, and the expression is: Among them, F p is the fractional Fourier operator, K p (t,u) is the transformation kernel, p is the fractional order, x i (t) is the i-th slice signal, u is the variable after the fractional Fourier transform of the time independent variable t; Transformation kernel K p The expression of (t,u) is: in, α=pπ / 2 represents the rotation angle of the time-frequency plane.

5. The method for analyzing cavitation state of a Francis turbine based on fractional Fourier transform and parallel CNN-BiLSTM according to claim 1, characterized in that: The real part data and the imaginary part data are input into the parallel CNN-BiLSTM model to obtain the cavitation state of the Francis turbine, including: Using a single-channel convolutional neural network (CNN) to process the real part data to obtain a first output result; Using a single-channel BiLSTM network to process the imaginary part data to obtain a second output result; The first output result and the second output result are integrated to obtain the cavitation state of the Francis turbine.

6. The method for analyzing cavitation state of a Francis turbine based on fractional Fourier transform and parallel CNN-BiLSTM according to claim 1, characterized in that: Training the parallel CNN-BiLSTM model includes: Acquire monitoring data samples; the monitoring data samples include the pressure between the runner and the guide vane; Time slicing the monitoring data samples; Performing a fractional Fourier transform on the pressure monitoring data after the time slice to obtain real part data and imaginary part data; Dividing the real part data and the imaginary part data into a training set and a test set respectively; A single-channel convolutional neural network CNN is used to train the signal samples of the real part data training set, and a single-channel BiLSTM network is used to train the imaginary part data training set to obtain a trained CNN-BiLSTM model; The trained CNN-BiLSTM model is tested using the real part data test set and the imaginary part data test set.

7. The method for analyzing cavitation state of a Francis turbine based on fractional Fourier transform and parallel CNN-BiLSTM according to claim 6, characterized in that: The ratio of the training set to the test set is 8:

2.

8. A Francis turbine cavitation state analysis system based on fractional Fourier transform and parallel CNN-BiLSTM, characterized in that: include, An acquisition module, used for acquiring pressure monitoring data between the runner and the guide vane through a sensor; A processing module, used for processing the pressure monitoring data; A fractional Fourier transform module, used for performing a fractional Fourier transform on the processed pressure monitoring data to obtain real part data and imaginary part data; The analysis module is used to input the real part data and the imaginary part data into the trained CNN-BiLSTM model for analysis to obtain the cavitation state of the Francis turbine.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed, the cavitation state analysis method of a Francis turbine based on fractional-order Fourier transform and parallel CNN-BiLSTM as described in any one of claims 1 to 7 can be implemented.

10. A device comprising a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, characterized in that: When the processor runs the computer instructions, it executes the cavitation state analysis method of a Francis turbine based on fractional-order Fourier transform and parallel CNN-BiLSTM as described in any one of claims 1 to 7.