CN-115774841-B - Hydraulic turbine cavitation intelligent recognition method based on combined spectrum feature extraction

CN115774841BCN 115774841 BCN115774841 BCN 115774841BCN-115774841-B

Abstract

The invention discloses an intelligent recognition method for cavitation phenomenon of a water turbine based on combined spectrum feature extraction, and belongs to the field of signal recognition processing. The method comprises the steps of collecting noise data before and after cavitation of a turbine runner model, decomposing firecracker-like sound spectrum and special pressure pulsation sound spectrum in each group of noise data, extracting index parameters representing characteristics of the two sound spectrums, carrying out normalization processing, mapping to different element positions of row vectors and column vectors of a matrix to form a feature vector containing bubble sound main feature ID, forming an instant contour form matrix A representing a bubble sound main physical sound state, and inputting a matrix C obtained by correcting the matrix A into a trained turbine cavitation recognition model to output cavitation judgment results. According to the method, the mixed sound spectrum of the gun sound spectrum and the special pulsation spectrum in the primary cavitation data of the water turbine is extracted and collected, and the mixed sound spectrum is used as a feature vector of cavitation identification and is input into a training model, so that intelligent identification of the primary cavitation phenomenon of the water turbine by a machine can be realized.

Inventors

HAN WENFU
ZHANG LIANG
SUN XIAOXIA
ZHOU JIAN
NI JINBING
ZHAO YIFENG
GUI ZHONGHUA
DING JINGHUAN
XIAO WEI
LI DONGKUO
LU WEIFU

Assignees

国网新源控股有限公司
国网新源控股有限公司抽水蓄能技术经济研究院
东方电气集团东方电机有限公司

Dates

Publication Date: 20260508
Application Date: 20221129
Priority Date: 20221117

Claims (8)

1. The intelligent recognition method for the cavitation phenomenon of the water turbine based on the combined spectrum feature extraction is characterized by comprising the following steps of: s1, collecting noise data when cavitation occurs in a turbine runner model; s2, decomposing the mixed wave signals in the noise data into a plurality of single wave signals, and extracting a first main frequency signal corresponding to firecracker-like sound and a second main frequency signal corresponding to special pressure pulsation sound of the water turbine; S3, calculating statistical characteristics of the first main frequency signal and the second main frequency signal respectively, obtaining relevant analysis index parameters and carrying out normalization processing; S4, mapping the index parameters after normalization processing to different element positions of row vectors and column vectors of the matrix to form PCSV vectors containing cavitation bubble sound main feature IDs; S5, giving different weights to index parameters of different main frequency signals in PCSV vectors for weighted average treatment, and then arranging and combining the obtained new vectors to find a contour form matrix B which can most represent the main physical sound state of the section of noise data; S6, processing the immediately acquired water turbine bubble sound data, inputting the processed water turbine bubble sound data into a water turbine cavitation bubble sound recognition model, and outputting cavitation recognition results.
2. The intelligent recognition method for the cavitation phenomenon of the water turbine based on the combined spectrum feature extraction as claimed in claim 1, wherein the step S6 specifically comprises the following steps: s61, collecting instant noise data of a turbine runner; S62, decomposing a mixed wave signal in the instant noise data into a plurality of single wave signals, respectively calculating the statistical characteristics of the single wave signals, obtaining relevant analysis index parameters and carrying out normalization processing; s63, mapping each index parameter after normalization processing to different element positions of a row vector and a column vector of a matrix to form an instant feature vector containing a bubble sound main feature ID; S64, giving different weights to index parameters of different single wave signals in the instant feature vector for weighted average treatment, and then rearranging and combining the obtained new vector to find an instant contour form matrix A which can most represent the main physical sound state of the noise data; S65, multiplying each index parameter in the instant contour form matrix A by a correction value to obtain a correction contour form matrix C containing a plurality of new vectors; s66, inputting the corrected outline form matrix C into a pre-trained water turbine cavitation bubble recognition model, and outputting cavitation discrimination results.
3. The intelligent recognition method for the cavitation phenomenon of the water turbine based on the combined spectral feature extraction according to claim 2, wherein if the corrected contour form matrix C is equal to the corrected contour form matrix D, the output result is that cavitation occurs in the water turbine runner, and otherwise, the output result is that cavitation does not occur.
4. The intelligent recognition method for the cavitation phenomenon of the water turbine based on the combined spectral feature extraction according to claim 1 or 2 is characterized in that the related analysis index parameters comprise a time domain index parameter, a power spectral density index parameter and a frequency domain index parameter, wherein the time domain index parameter comprises a peak-to-peak value vpp, a quarter frequency probability PVpH, a standard deviation St, a Ku, a Sk and an information entropy H, and the frequency domain index parameter comprises a center of gravity frequency PsdFc.
5. The intelligent recognition method for the cavitation phenomenon of the water turbine based on the combined spectrum feature extraction, which is characterized in that the weight assignment range of index parameters belonging to the first main frequency signal is 0.6-0.8.
6. The intelligent recognition method for the cavitation phenomenon of the water turbine based on the combined spectrum feature extraction according to claim 1 or 2, wherein the 0-mean normalization method is used for carrying out normalization transformation on related analysis index parameters.
7. The intelligent recognition method for the cavitation phenomenon of the water turbine based on the combined spectral feature extraction according to claim 1 or 2, wherein a 0.97 confidence probability assignment function or a mixing sampling function is used for decomposing a mixed wave signal in noise data.
8. The intelligent recognition method for the cavitation phenomenon of the water turbine based on the combined spectrum feature extraction according to claim 1 or 2, wherein the correction value is a discrimination coefficient or interval, and the correction value is determined by combining an empirical parameter and cavitation test data of the water turbine.

Description

Hydraulic turbine cavitation intelligent recognition method based on combined spectrum feature extraction Technical Field The invention relates to a method for identifying cavitation of a water turbine, in particular to an intelligent method for identifying cavitation of a water turbine based on combination spectrum feature extraction. Background Primary cavitation of a water turbine refers to the phenomenon that the local pressure in liquid is reduced to a critical value, the contained gas nuclei are rapidly increased, and cavitation begins to occur. The noise and vibration are accompanied, so that the power generation efficiency is reduced, the output is reduced, and the hydraulic vibration is aggravated. The service life of the water turbine is influenced, and the safe operation of a water power station and a power grid is threatened. Therefore, the identification of the cavitation of the water turbine has important significance for the operation safety of the hydropower station and the power grid, and the identification of the cavitation phenomenon of the water turbine is carried out in the actual operation of the industry at present in a manual judgment mode. The flow state of the vortex belt of the draft tube and the water outlet side of the rotating wheel is observed manually through the draft tube straight cone section made of transparent organic glass. In the test process, a flash frequency instrument (a light-dark alternative light source with adjustable frequency) is used for polishing, the frequency of the flash frequency instrument is adjusted to be equal to or close to the rotating speed frequency of the model water turbine, and the rotating wheel blade which seems to be stationary or rotates slowly can be clearly seen by naked eyes, so that the cavitation condition of the water outlet edge of the rotating wheel blade can be observed. The method has very high requirements on staff, and generally, at least a person with about ten years of working experience can observe and judge whether cavitation exists or not. The method is high in subjectivity and low in accuracy and efficiency. In the prior art, there is a method for identifying cavitation acoustic signals of a water turbine by a big data learning mode, for example, patent CN113255848a discloses a method for identifying cavitation acoustic signals of a water turbine based on big data learning. The method comprises the steps of obtaining multiple neural network models based on big data learning, carrying out time sequence clustering based on multiple operation conditions under multiple output conditions of a hydraulic turbine unit by utilizing an SOM neural network through extracting acoustic signal time sequence data of the hydraulic turbine unit, screening feature quantities of stable conditions under the healthy condition of the hydraulic turbine unit, carrying out feature screening of multiple measuring points under the stable conditions of the hydraulic turbine unit by introducing a random forest algorithm, extracting optimal feature measuring points and optimal feature subsets with higher sensitivity to a prediction model, finally establishing a health state prediction model by using a gating circulation unit, and judging whether initial cavitation exists in equipment or not and carrying out early warning reminding through self-adaption assessment of the sum of dynamic tolerances of the multiple measuring points. By researching the characteristics of cavitation noise of the water turbine, the characteristics of related phenomena of cavitation of the water turbine are required to be known, and information contained in the data is accurately and comprehensively analyzed from sample data, so that the analysis result of test data is combined with the principle of cavitation phenomenon of the water turbine, different stages of cavitation of the water turbine can be effectively distinguished, and diagnosis and identification of cavitation of a runner of a primary model are completed. The mode of predicting and outputting short-time stable working condition information in the future in advance by screening the characteristic quantity of the stable working condition of the turbine unit in the healthy state is deficient in accuracy and recognition efficiency of primary cavitation phenomenon recognition of the turbine. When cavitation of the water turbine occurs, tiny bubbles are generated at the outer edges of the blades, and multi-state noise is generated at the same time, and because the water turbine has atypical physical sound state characteristics, the test technology based on the current physical sound state configuration cannot accurately identify whether cavitation occurs to the model rotating wheel or not through the sound state of the water turbine. Disclosure of Invention The invention aims to solve the problems of the prior art water turbine cavitation identification technology, and provides an intelligent water turbine cavitation p