CN116049637A - Gas-liquid two-phase flow pattern recognition method based on flow noise signals - Google Patents
Gas-liquid two-phase flow pattern recognition method based on flow noise signals Download PDFInfo
- Publication number
- CN116049637A CN116049637A CN202310176230.2A CN202310176230A CN116049637A CN 116049637 A CN116049637 A CN 116049637A CN 202310176230 A CN202310176230 A CN 202310176230A CN 116049637 A CN116049637 A CN 116049637A
- Authority
- CN
- China
- Prior art keywords
- flow
- flow pattern
- pattern recognition
- signal
- imf
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Complex Calculations (AREA)
Abstract
The invention relates to a gas-liquid two-phase flow pattern recognition method based on a flow noise signal, which comprises the following steps: 1) Acquiring a gas-liquid two-phase flow noise signal in a pipeline system, classifying the acquired flow noise signal, and constructing an original data set; 2) Preprocessing an original data set to obtain IMF components of each flow noise signal; 3) Extracting 3 component characteristic parameters of the flow noise signal IMF by using a signal analysis method; 4) The method comprises the steps of processing extracted characteristic parameters to obtain a sample set, dividing the sample set into a training set and a test set, inputting the training set into an SVM training module, training to obtain an SVM flow pattern recognition model of a gas-liquid two-phase flow, and inputting the test set into the SVM flow pattern recognition model to obtain a recognition result; 5) And after extracting characteristic parameters of the unknown flow pattern noise signals, constructing characteristic vectors, and inputting the characteristic vectors into a trained SVM flow pattern recognition model to obtain a flow pattern recognition result. The method of the invention has obviously improved accuracy.
Description
Technical Field
The invention belongs to the technical field of flow pattern recognition, and relates to a gas-liquid two-phase flow pattern recognition method based on a flow noise signal.
Background
Two-phase flow widely exists in various fields such as energy, chemical industry, metallurgy, nuclear industry, aerospace and the like, and as one of the most common two-phase flow forms, gas-liquid two-phase flow widely exists in the development and utilization of deep sea oil gas, mixing transportation and other processes. The flow pattern, namely the flowing state and structure, is the most basic and the most critical parameter in the gas-liquid two-phase flow, is closely related to the mass and heat transfer process, and the complicated flow structure and flow pattern are caused by the different properties and behaviors of the two-phase flow; the correctness of the flow pattern judgment is significant for accurately predicting the pressure drop and heat transfer performance of the two-phase flow, accurately measuring the two-phase flow parameters and guaranteeing the safe operation of the equipment.
The current methods for identifying the flow patterns of the gas-liquid two-phase flow are mainly divided into a flow pattern method, a direct observation method and an indirect analysis method. Direct observation methods such as a visual inspection method, a high-speed image pickup method, and the like; indirect analysis methods include capacitance tomography, ultrasonic measurement, gamma ray, optical, differential pressure signal recognition, and the like. However, the direct observation method is often high in subjectivity and high in requirement on equipment transparency, and is difficult to realize in actual engineering; and gamma rays cause harm to the body of on-site staff due to the existence of radioactivity; the ultrasonic measurement method, the capacitance tomography technology, the optical method and other equipment are complex and expensive to operate, have high cost and poor economical efficiency, and are not suitable for engineering practical application. Therefore, it is very necessary to develop a gas-liquid two-phase flow pattern recognition method with high recognition accuracy, simple operation and strong economy.
In the gas-liquid two-phase flow process, different flow patterns are generated along with different flow rates of the gas-liquid two-phase due to different physical properties of the two phases, and different flow noise signals are generated, wherein the noise signals contain abundant physical and dynamic process information; compared with the flow pattern identification method using pressure, differential pressure signals or phase density image reconstruction, the sampling frequency of the noise signals can reach several megahertz generally, the sampling time of the signals is short, the economy is good, and the method is suitable for identifying the flow pattern of the gas-liquid two-phase flow in the industrial process. Therefore, a new method for identifying flow patterns by using gas-liquid two-phase flow noise signals is developed, and the method becomes a new technical direction.
Disclosure of Invention
The invention aims to provide a gas-liquid two-phase flow pattern recognition method based on a flow noise signal, which solves the problems that the prior art is difficult to accurately and rapidly recognize the flow pattern of a gas-liquid two-phase flow of a pipeline system in the deep sea oil gas exploitation and transportation process, so that the detection timeliness and accuracy of parameters of the gas-liquid two-phase flow and the multiphase flow are insufficient, and the safety of the system is ensured to be insufficient.
The invention adopts the technical scheme that the gas-liquid two-phase flow pattern recognition method based on the flow noise signals is implemented according to the following steps:
step 2, preprocessing through an original data set to obtain 3 components of IMF components of each flow noise signal;
step 3, extracting characteristic parameters of 3 components of the flow noise signal IMF by using a signal analysis method;
step 4, processing the characteristic parameters extracted in the step 3 to obtain a sample set, dividing the sample set into a training set and a test set, inputting the training set into an SVM training module, training to obtain an SVM flow pattern recognition model of a gas-liquid two-phase flow, inputting the test set into the flow pattern SVM flow pattern recognition model obtained by testing to obtain a flow pattern recognition result, and verifying the recognition accuracy of the trained model;
and 5, extracting the characteristic parameters of the unknown flow pattern noise signals according to the optimal characteristic subsets determined in the step 3, constructing characteristic vectors, and inputting the characteristic vectors into a trained SVM flow pattern recognition model to obtain a flow pattern recognition result.
The beneficial effects of the invention are that the invention comprises the following aspects:
1) The method for identifying the flow pattern of the gas-liquid two-phase flow of the vertical pipe by utilizing the flow noise signal is provided by collecting the flow noise signal of the gas-liquid two-phase flow of the pipeline, the flow noise signal contains very rich dynamic process information, the flow noise signal is more close to the flowing essence, the signal collection is quick, the economy is good, the identification accuracy is high, and the method is more suitable for identifying the flow pattern of the gas-liquid two-phase flow in the industrial process.
2) The genetic algorithm is adopted to optimize the decomposition parameters of the variation mode, so that over decomposition and under decomposition are prevented, the decomposition effect is improved, the noise factors and instability of the original signals are reduced, the data quality is improved, meanwhile, the components are screened by adopting a method of combining the Pearson correlation coefficient and the variance contribution rate, the model training time is shortened, and the recognition flow is more efficient;
3) The RF-RFE algorithm is adopted to carry out data dimension reduction on the feature vector, so that the analysis efficiency is optimized, the subsequent model training and predicting time is reduced, and the speed and instantaneity of the algorithm for identifying the flow pattern are improved;
4) The improved PSO algorithm is utilized to optimize SVM parameters, the local searching capability and the global searching capability can be better balanced in the optimization process, the searching efficiency is higher, the advanced convergence is avoided, the local optimal value is trapped, and meanwhile, the accuracy of the method for identifying the flow pattern is improved. The identification method provided by the invention is simple and quick to operate, good in instantaneity, strong in economy, high in identification accuracy and wide in application range, and provides a certain reference for carrying out gas-liquid two-phase flow pattern identification by using the flowing noise signals and for the subsequent flow type identification method.
Drawings
FIG. 1 is a schematic flow chart of the method of the present invention;
FIG. 2 is a flow noise signal acquired by the method of the present invention;
FIG. 3 is a flow chart of the genetic algorithm optimization variation modal decomposition parameters of the method of the invention;
FIG. 4 is a flow chart of the improved particle swarm algorithm of the invention for optimizing SVM parameters;
FIG. 5 is a graph of classification results of flow pattern recognition on a test set according to an embodiment of the method of the present invention.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
Referring to fig. 1, the gas-liquid two-phase flow pattern recognition method of the present invention is specifically implemented based on a flow noise signal according to the following steps:
The method comprises the following steps: the method comprises the steps of collecting flow noise signals through a hydrophone, continuously adjusting apparent speeds of gas phase and liquid phase of a pipeline, collecting noise signals of gas-liquid two-phase flow of the pipeline under different flow pattern working conditions to obtain an original time sequence S (t), classifying the collected flow noise signals under different flow pattern working conditions to obtain an original data set S, wherein the original time sequence S (t) is shown in FIG. 2 i (t)；
Step 2, passing the original dataset S i (t) performing a pretreatment to obtain IMF components Imf of each flow noise signal after the treatment 1 (t)，Imf 2 (t)，Imf 3 (t) the specific process is as follows:
2.1 Processing the raw data set S acquired in step 1) by means of a wavelet thresholding noise reduction method i (t) decomposing the original signal into layers of coefficients through wavelet decomposition, giving a threshold value, comparing each decomposition coefficient with the threshold value, reserving the decomposition coefficient larger than the threshold value, and carrying out inverse transformation reconstruction on the signal to obtain a flowing noise signal S' (t);
2.2 Referring to fig. 3, an optimization algorithm is applied in combination with a variation modal decomposition method to decompose the flow noise signal S' (t) after the noise reduction process in step 2.1);
in order to determine parameters of the variational modal decomposition to prevent over-decomposition and under-decomposition, the parameters need to be optimized by an intelligent optimization algorithm. The decomposition layer number K and the penalty factor alpha in the decomposition process of the variation mode can have a larger influence on the decomposition result, so that the optimal combination parameters [ K, alpha ] of the variation mode decomposition are globally optimized through a genetic algorithm to obtain the optimal decomposition effect.
The information entropy can be used for representing the information quantity contained when the variable occurs, the smaller the information entropy is, the smaller the signal noise is, the more the effective information is, and the optimization is carried out by taking the local minimum envelope entropy as the fitness function, and the specific steps of optimizing are as follows:
2.2.1 Initializing genetic algorithm parameters, and determining the value range of a decomposition layer number K and a penalty factor alpha in the decomposition process of the variation mode;
2.2.2 Performing variation modal decomposition on the flowing noise signal, and calculating the local minimum envelope entropy of the individual according to the formula (1);
2.2.3 Global searching is carried out by taking the local minimum envelope entropy as a fitness function, and the minimum value is found;
2.2.4 Judging whether the maximum evolution algebra is reached, stopping iteration if the maximum evolution algebra is reached, and outputting an optimal solution; otherwise, continuing to operate;
2.2.5 Selecting, crossing and mutating the population to obtain a next generation population, and returning to the step 2.2.2) calculating the envelope entropy of the individuals of the next generation population to continue operation; until the iteration stop condition is met;
2.2.6 Outputting the optimal parameter combination [ K, alpha ].
The envelope entropy is calculated as follows:
wherein a (t) is the sum of an envelope signal x' (t) obtained by carrying out Hilbert demodulation on a modal component after variation modal decomposition and an original signal x (t), N is the signal length, and E p Is envelope entropy;
2.3 The flow noise signal is decomposed by a variation modal decomposition algorithm to obtain a plurality of IMF components, the signal characteristics of the IMF components are more obvious, and the components and the original signal are calculated according to a cross-correlation methodPearson coefficients, and filtering IMF components by combining variance contribution rate, and reserving 3 components Imf with highest similarity with the original signal 1 (t)、Imf 2 (t)、Imf 3 (t) as a subsequent feature extraction object, noise factors and instability of an original signal are reduced, data quality is improved, and model training time is shortened;
the calculation formula of the Pearson correlation coefficient is:
where cov (X, Y) is the covariance of the original signal and the IMF component, σ x And sigma (sigma) Y Standard deviations of the original signal and IMF components, respectively;
the variance contribution ratio, i.e., the ratio of IMF component variance to original signal variance, is calculated as follows:
wherein X is var 、Y var The IMF component and the variance value of the original signal sequence, respectively.
Step 3, extracting 3 components Imf of the flow noise signal IMF processed in step 2 by using a signal analysis method 1 (t)、Imf 2 (t)、Imf 3 The characteristic parameters of (t) are as follows:
3.1 Extracting the characteristics of the IMF components reserved in the step 2, and extracting the characteristic parameters of each component, including kurtosis, root mean square value, standard deviation, hilbert marginal spectrum energy and permutation entropy, to form an original characteristic data set;
the extraction method of kurtosis, root mean square value, standard deviation, hilbert marginal spectrum energy and permutation entropy is as follows:
first, the kurtosis is calculated as follows:
wherein n represents the length of the signal sequence and μ represents the average value of the signal sequence;
second, the root mean square value is calculated as follows:
wherein n represents the length of the signal sequence;
third, the standard deviation is calculated as follows:
wherein n represents the length of the signal sequence and μ represents the average value of the signal sequence;
fourth, the Hilbert marginal spectral energy solution is as follows:
a1 Performing hilbert transformation on the noise signal IMF component obtained after the component is decomposed and filtered by the variation mode in the step 2 to obtain a hilbert spectrum, namely:
wherein RC represents the operation of taking the real part, A i (t) represents the instantaneous amplitude, F i (t) represents an instantaneous frequency;
a2 Solving the hilbert marginal spectrum, the calculation formula is as follows:
a3 The hilbert marginal spectral energy is calculated as follows:
wherein F1 and F2 are frequency intervals of the Hilbert marginal spectrum respectively;
fifth, the solution process of permutation entropy is as follows:
b1 For the screened IMF component (denoted as X (1), X (2) … … X (n)), an embedding dimension m and a time delay t are set delay ；
B2 By embedding dimension m and time delay t delay Reconstructing the phase space of each element extension of the original signal to obtain K=n- (m-1) t delay Sub-sequences, wherein each sub-sequence is denoted Y (i), Y (i) =x (i), X (i+t) delay )……X(i+(m-1)t delay )；
B3 (ii) internally ascending order of each Y (i);
b4 Calculating probability P of each probability occurrence j ；
B5 Calculating the permutation entropy of these probabilities:
3.2 Using RF-RFE algorithm to sort the feature importance of the feature parameters of each component extracted in the step 2.3) to obtain the optimal feature subset, constructing feature vectors, performing normalization processing, further performing data dimension reduction, optimizing analysis efficiency,
the method uses an RF-RFE algorithm to perform data dimension reduction, and mainly comprises two steps of feature importance sorting and recursive feature elimination, wherein the specific process is as follows:
c1 Extracting X training samples from the original characteristic data set by a Bootstrap resampling method, wherein each sample has N characteristics, and the non-extracted samples are called OOB data;
c2 Initializing i=1, creating a decision tree T i ；
C3 Training T for the ith sample i Calculating the classification accuracy A of the ith OOB data i ；
C4 For feature F in OOB dataset j Random noise interference is carried out to obtain a new OOB data set, and classification is calculated againDetermination rate A ij ；
C5 Repeating steps C2) to C4) for i=2, 3 … …);
c6 Feature F) j The importance of (2) is calculated by the following formula:
c7 Ordering the calculated feature importance, and obtaining the classification precision of the initial feature subset by using a cross verification method;
c8 Deleting a feature with the minimum importance from the current feature subset, and calculating the feature importance of the new feature subset and the classification precision under cross verification;
c9 Recursively repeating the step C8) until the feature subset is an empty set, and taking the subset with the highest classification accuracy as the optimal feature subset.
Step 4, processing the characteristic parameters extracted in the step 3 to obtain a sample set, dividing the sample set into a training set and a test set, inputting the training set into an SVM training module, training to obtain an SVM flow pattern recognition model of a gas-liquid two-phase flow, inputting the test set into the SVM flow pattern recognition model obtained by testing as an input parameter, obtaining a flow pattern recognition result, and verifying the recognition accuracy of the trained model;
the specific process is as follows: extracting characteristic parameters of the acquired noise signals according to the optimal characteristic subsets obtained after optimization in the step 3, constructing characteristic vectors through normalization processing, and adding one-dimensional elements serving as labels in a sample set according to a flow pattern to which the signals belong to so as to indicate the attribute of the set of characteristic vectors, wherein 1 represents bubble flow, 2 represents bullet flow and 3 represents annular flow; randomly dividing the processed data into a training set and a testing set, wherein the data of the training set is used for training an SVM flow pattern recognition model; the data of the test set is used as input to verify the accuracy of the trained SVM flow pattern recognition model.
Referring to fig. 4, in order to determine the penalty parameter c and the kernel parameter g in the training process, an improved particle swarm algorithm is used to optimize the parameters in the training process, which specifically includes the following steps:
d1 Using the classification accuracy in the cross-validation sense as the adaptability function of algorithm optimization;
d2 Initializing population and particle speed, position and maximum iteration number, bringing population particle information into fitness function to calculate population initial fitness, and taking initial particle fitness value as individual extremum P best The individual history optimal fitness value is taken as global optimal extremum g best ；
D3 Updating the individual optima and global optima extremum of particle positions and velocities, particle populations, and updating the corresponding extremum when particle velocities and positions exceed the original extremum by:
wherein w represents inertial weight, and varies with evolution times, and can better balance global searching capability and local searching capability max ,w min Taking 0.9 and 0.4 respectively; c 1 ,c 2 For acceleration factor, generally 2.0 is taken; r is (r) 1 ，r 2 Is a random number between (0, 1), the subscript i, j represents the ith particle and the jth dimension, t represents the number of iterations, t max Representing a maximum evolution algebra;
d4 Performing self-adaptive mutation operation, initializing the current position of a mutation particle by taking a mutation factor v calculated by the following formula as probability, wherein the mutation factor v is larger in early-stage mutation factor, searching in a larger area with stronger global searching capability, preventing local optimization from being trapped, and the mutation factor v is smaller in later-stage mutation factor, concentrating on search near excellent individuals, so that the algorithm convergence rate is increased;
in the formula, v is taken min ,v max Respectively 0.1 and 0.7, t represents the iteration number, t max Representing a maximum evolution algebra;
d5 Calculating a mutated particle fitness value, comparing the current particle population fitness value with an individual extremum and a global optimal extremum, and updating the individual optimal and global optimal extremum of the particle population;
d6 Judging whether the iteration termination condition is met, outputting the optimal parameter combination if the iteration termination condition is met, otherwise returning to the step D3), and until the iteration termination condition is met.
And 5, extracting the characteristic parameters of the unknown flow pattern noise signals according to the optimal characteristic subsets determined in the step 3, constructing characteristic vectors, and inputting the characteristic vectors into a trained SVM flow pattern recognition model to obtain a flow pattern recognition result.
And (3) experimental verification:
the two-phase medium adopted in the experiment is air and water respectively, the apparent speed adjusting range of the gas phase is 0.1-5 m/s, and the apparent speed adjusting range of the liquid phase is 0.1-2 m/s. The hydrophone is used for collecting two-phase flowing noise signals, the sampling frequency of the signals is 51.2kHz, the apparent flow velocity and flow pattern of gas and liquid phases are regulated, 150 samples are obtained through experiments, and the sampling frequency is 4:1, randomly dividing a training set and a testing set, namely, taking 120 samples as the training set to train an SVM flow pattern recognition model, and using the rest 30 samples for verifying the accuracy of the recognition model; as shown in FIG. 5, the training SVM flow pattern recognition model tested on the test set had a recognition accuracy of 96.7%, i.e., only one sample was recognized as erroneous.
Although the original noise signal can not reach the sampling frequency of several MHz due to the hardware limitation of the data acquisition card in the example, the sampling frequency is much higher than that of the prior methods of applying more pressure signals, conductance probes, phase density imaging and the like, and the signal acquisition is simple and convenient to realize.
Therefore, the method can realize accurate identification of the flow pattern of the gas-liquid two-phase flow, is based on the design of the flow noise signal, has high signal sampling frequency and low requirements on equipment, is convenient to realize, can realize quick identification of the flow pattern, and is more suitable for application in the industrial field.
Claims (5)
1. The gas-liquid two-phase flow pattern recognition method based on the flow noise signal is characterized by comprising the following steps of:
step 1, acquiring flow noise signals of gas-liquid two-phase flow in a pipeline system, classifying the acquired flow noise signals under different flow pattern working conditions, and constructing an original data set;
step 2, preprocessing an original data set to obtain 3 components of IMF components of each flow noise signal;
step 3, extracting characteristic parameters of 3 components of the flow noise signal IMF by using a signal analysis method;
step 4, processing the characteristic parameters extracted in the step 3 to obtain a sample set, dividing the sample set into a training set and a test set, inputting the training set into an SVM training module, training to obtain an SVM flow pattern recognition model of a gas-liquid two-phase flow, inputting the test set into the flow pattern SVM flow pattern recognition model obtained by testing to obtain a flow pattern recognition result, and verifying the recognition accuracy of the trained model;
and 5, extracting the characteristic parameters of the unknown flow pattern noise signals according to the optimal characteristic subsets determined in the step 3, constructing characteristic vectors, and inputting the characteristic vectors into a trained SVM flow pattern recognition model to obtain a flow pattern recognition result.
2. The gas-liquid two-phase flow pattern recognition method based on the flow noise signal according to claim 1, wherein the method is characterized by comprising the following steps: in step 1, the specific process is that,
the method comprises the steps of collecting flow noise signals through a hydrophone, continuously adjusting apparent speeds of gas phase and liquid phase of a pipeline, collecting noise signals of gas-liquid two-phase flow of the pipeline under different flow pattern working conditions to obtain an original time sequence S (t), and classifying the collected flow noise signals under different flow pattern working conditions to obtain an original data set S i (t)。
3. The gas-liquid two-phase flow pattern recognition method based on the flow noise signal according to claim 1, wherein the method is characterized by comprising the following steps: in step 2, the specific process is as follows:
2.1 Processing the raw data set S acquired in step 1) by means of a wavelet thresholding noise reduction method i (t) decomposing the original signal into layers of coefficients through wavelet decomposition, giving a threshold value, comparing each decomposition coefficient with the threshold value, reserving the decomposition coefficient larger than the threshold value, and carrying out inverse transformation reconstruction on the signal to obtain a flowing noise signal S' (t);
2.2 Applying an optimization algorithm and a variation modal decomposition method to decompose the flow noise signal S' (t) subjected to noise reduction treatment in the step 2.1);
the decomposition layer number K and the penalty factor alpha in the decomposition process of the variation mode can have a larger influence on the decomposition result, so that the optimal combination parameters [ K, alpha ] of the variation mode decomposition are globally optimized through a genetic algorithm, and the optimization is performed by taking the local minimum envelope entropy as an fitness function, wherein the specific steps of optimizing are as follows:
2.2.1 Initializing genetic algorithm parameters, and determining the value range of a decomposition layer number K and a penalty factor alpha in the decomposition process of the variation mode;
2.2.2 Performing variation modal decomposition on the flowing noise signal, and calculating the local minimum envelope entropy of the individual according to the formula (1);
2.2.3 Global searching is carried out by taking the local minimum envelope entropy as a fitness function, and the minimum value is found;
2.2.4 Judging whether the maximum evolution algebra is reached, stopping iteration if the maximum evolution algebra is reached, and outputting an optimal solution; otherwise, continuing to operate;
2.2.5 Selecting, crossing and mutating the population to obtain a next generation population, and returning to the step 2.2.2) calculating the envelope entropy of the individuals of the next generation population to continue operation; until the iteration stop condition is met;
2.2.6 Outputting the optimal parameter combination [ K, alpha ];
the envelope entropy is calculated as follows:
wherein a (t) is a post-mode decomposition model through a variation modeThe sum of the envelope signal x' (t) obtained by Hilbert demodulation of the state component and the original signal x (t), N is the signal length, E p Is envelope entropy;
2.3 The flow noise signal is decomposed by a variation modal decomposition algorithm to obtain a plurality of IMF components, the Pearson coefficients of each component and the original signal are calculated according to a cross correlation method, the IMF components are screened by combining the variance contribution rate, and the 3 components Imf with the highest similarity with the original signal are reserved 1 (t)、Imf 2 (t)、Imf 3 (t) as a subsequent feature extraction object;
the calculation formula of the Pearson correlation coefficient is:
where cov (X, Y) is the covariance of the original signal and the IMF component, σ x And sigma (sigma) Y Standard deviations of the original signal and IMF components, respectively;
the variance contribution ratio, i.e., the ratio of IMF component variance to original signal variance, is calculated as follows:
wherein X is var 、Y var The IMF component and the variance value of the original signal sequence, respectively.
4. The gas-liquid two-phase flow pattern recognition method based on the flow noise signal according to claim 1, wherein the method is characterized by comprising the following steps: in step 3, the specific process is as follows:
3.1 Extracting the characteristics of the IMF components reserved in the step 2, and extracting the characteristic parameters of each component, including kurtosis, root mean square value, standard deviation, hilbert marginal spectrum energy and permutation entropy, to form an original characteristic data set; the extraction methods of kurtosis, root mean square value, standard deviation, hilbert marginal spectrum energy and permutation entropy are respectively as follows:
first, the kurtosis is calculated as follows:
wherein n represents the length of the signal sequence and μ represents the average value of the signal sequence;
second, the root mean square value is calculated as follows:
wherein n represents the length of the signal sequence;
third, the standard deviation is calculated as follows:
wherein n represents the length of the signal sequence and μ represents the average value of the signal sequence;
fourth, the Hilbert marginal spectral energy solution is as follows:
a1 Performing hilbert transformation on the noise signal IMF component obtained after the component is decomposed and filtered by the variation mode in the step 2 to obtain a hilbert spectrum, namely:
wherein RC represents the operation of taking the real part, A i (t) represents the instantaneous amplitude, F i (t) represents an instantaneous frequency;
a2 Solving the hilbert marginal spectrum, the calculation formula is as follows:
a3 The hilbert marginal spectral energy is calculated as follows:
wherein F1 and F2 are frequency intervals of the Hilbert marginal spectrum respectively;
fifth, the solution process of permutation entropy is as follows:
b1 For the selected IMF component, denoted by X (1), X (2) … … X (n), the embedding dimension m and the time delay t are set delay ；
B2 By embedding dimension m and time delay t delay Reconstructing the phase space of each element extension of the original signal to obtain K=n- (m-1) t delay Sub-sequences, wherein each sub-sequence is denoted Y (i), Y (i) =x (i), X (i+t) delay )……X(i+(m-1)t delay )；
B3 (ii) internally ascending order of each Y (i);
b4 Calculating probability P of each probability occurrence j ；
B5 Calculating the permutation entropy of these probabilities:
3.2 Using RF-RFE algorithm to sort the feature importance of the feature parameters of each component extracted in the step 2.3) to obtain the optimal feature subset, constructing feature vector, normalizing,
the method uses an RF-RFE algorithm to perform data dimension reduction, and comprises two steps of feature importance sorting and recursive feature elimination, wherein the process is as follows:
c1 Extracting X training samples from the original characteristic data set by a Bootstrap resampling method, wherein each sample has N characteristics, and the non-extracted samples are called OOB data;
c2 Initializing i=1, creating a decision tree T i ；
C3 Training T for the ith sample i Meter (D)Calculating the classification accuracy A of the ith OOB data i ；
C4 For feature F in OOB dataset j Random noise interference is carried out to obtain a new OOB data set, and classification accuracy A is calculated again ij ；
C5 Repeating steps C2) to C4) for i=2, 3 … …);
c6 Feature F) j The importance of (2) is calculated by the following formula:
c7 Ordering the calculated feature importance, and obtaining the classification precision of the initial feature subset by using a cross verification method;
c8 Deleting a feature with the minimum importance from the current feature subset, and calculating the feature importance of the new feature subset and the classification precision under cross verification;
c9 Recursively repeating the step C8) until the feature subset is an empty set, and taking the subset with the highest classification accuracy as the optimal feature subset.
5. The gas-liquid two-phase flow pattern recognition method based on the flow noise signal according to claim 1, wherein the method is characterized by comprising the following steps: in step 4, the specific process is that,
extracting characteristic parameters of the collected noise signals according to the optimal characteristic subsets obtained after optimization in the step 3, constructing characteristic vectors through normalization processing, and adding one-dimensional elements serving as labels in a sample set according to a flow pattern to which the signals belong to so as to indicate the attribute of the set of characteristic vectors, wherein 1 represents bubble flow, 2 represents bullet flow and 3 represents annular flow; randomly dividing the processed data into a training set and a testing set, wherein the data of the training set is used for training an SVM flow pattern recognition model; the data of the test set is used as input to verify the accuracy of the trained SVM flow pattern recognition model,
in the training process of SVM flow pattern recognition model, in order to determine the punishment parameter c and the value of the kernel function parameter g in the training process, an improved particle swarm algorithm is used for optimizing the parameters, and the process is as follows:
d1 Using the classification accuracy in the cross-validation sense as the adaptability function of algorithm optimization;
d2 Initializing population and particle speed, position and maximum iteration number, bringing population particle information into fitness function to calculate population initial fitness, and taking initial particle fitness value as individual extremum P best The individual history optimal fitness value is taken as global optimal extremum g best ；
D3 Updating the individual optima and global optima extremum of particle positions and velocities, particle populations, and updating the corresponding extremum when particle velocities and positions exceed the original extremum by:
wherein w represents inertial weight, and varies with the number of evolutions, w max ,w min Respectively take 0.9 and 0.4; c 1 ,c 2 Is an acceleration coefficient; r is (r) 1 ，r 2 Is a random number between (0, 1), the subscript i, j represents the ith particle and the jth dimension, t represents the number of iterations, t max Representing a maximum evolution algebra;
d4 Performing adaptive mutation operation, initializing the current position of the mutation particles by taking the mutation factor v calculated by the following formula as probability, and accelerating the convergence speed of the algorithm;
in the formula, v min ,v max Respectively 0.1 and 0.7, t is the iteration number, t max Is the maximum evolution algebra;
d5 Calculating a mutated particle fitness value, comparing the current particle population fitness value with an individual extremum and a global optimal extremum, and updating the individual optimal and global optimal extremum of the particle population;
d6 Judging whether the iteration termination condition is met, outputting the optimal parameter combination if the iteration termination condition is met, otherwise returning to the step D3), and until the iteration termination condition is met.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310176230.2A CN116049637A (en) | 2023-02-28 | 2023-02-28 | Gas-liquid two-phase flow pattern recognition method based on flow noise signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310176230.2A CN116049637A (en) | 2023-02-28 | 2023-02-28 | Gas-liquid two-phase flow pattern recognition method based on flow noise signals |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116049637A true CN116049637A (en) | 2023-05-02 |
Family
ID=86118298
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310176230.2A Pending CN116049637A (en) | 2023-02-28 | 2023-02-28 | Gas-liquid two-phase flow pattern recognition method based on flow noise signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116049637A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117476039A (en) * | 2023-12-25 | 2024-01-30 | 西安理工大学 | Acoustic signal-based primary cavitation early warning method for water turbine |
-
2023
- 2023-02-28 CN CN202310176230.2A patent/CN116049637A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117476039A (en) * | 2023-12-25 | 2024-01-30 | 西安理工大学 | Acoustic signal-based primary cavitation early warning method for water turbine |
| CN117476039B (en) * | 2023-12-25 | 2024-03-08 | 西安理工大学 | Acoustic signal-based primary cavitation early warning method for water turbine |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Multiscale local features learning based on BP neural network for rolling bearing intelligent fault diagnosis | |
| CN108710777B (en) | Diversified anomaly detection identification method based on multi-convolution self-coding neural network | |
| CN111751133B (en) | Intelligent fault diagnosis method of deep convolutional neural network model based on non-local mean embedding | |
| Xin et al. | Fault diagnosis of wheelset bearings in high-speed trains using logarithmic short-time Fourier transform and modified self-calibrated residual network | |
| Hasan et al. | A fault diagnosis framework for centrifugal pumps by scalogram-based imaging and deep learning | |
| Wu et al. | Complex system fault diagnosis based on a fuzzy robust wavelet support vector classifier and an adaptive Gaussian particle swarm optimization | |
| Motlagh et al. | Power quality disturbances recognition using adaptive chirp mode pursuit and grasshopper optimized support vector machines | |
| CN101984340A (en) | Method for identifying flow type of soft grain two-phase turbulence based on artificial intelligence | |
| Arbin et al. | Comparative analysis between k-means and k-medoids for statistical clustering | |
| CN109147817B (en) | Denoising frequency characteristic extraction method based on variation-limited Boltzmann machine | |
| CN116049637A (en) | Gas-liquid two-phase flow pattern recognition method based on flow noise signals | |
| CN116894187A (en) | Gear box fault diagnosis method based on deep migration learning | |
| CN111665050B (en) | Rolling bearing fault diagnosis method based on clustering K-SVD algorithm | |
| Cai et al. | EMD and GNN-AdaBoost fault diagnosis for urban rail train rolling bearings | |
| CN114676733A (en) | Fault diagnosis method for complex supply and delivery mechanism based on sparse self-coding assisted classification generation type countermeasure network | |
| KR20200061307A (en) | Earthquake classification method and apparatus using deep learning | |
| CN113884300A (en) | Rolling bearing fault diagnosis method for deep anti-migration learning | |
| Tu et al. | A coupling model of multi-feature fusion and multi-machine learning model integration for defect recognition | |
| CN110222610A (en) | A kind of method and device of Modulation recognition | |
| CN113791389A (en) | Unknown radar radiation source individual identification method and system | |
| CN117473414A (en) | Bearing fault position identification method based on low-noise time-frequency image | |
| Hamdaoui et al. | Improved signal processing for bearing fault diagnosis in noisy environments using signal denoising, time–frequency transform, and deep learning | |
| CN112183260A (en) | One-way valve fault diagnosis method based on total variation noise reduction and RQA | |
| CN115356599B (en) | Multi-mode urban power grid fault diagnosis method and system | |
| CN117037841A (en) | Acoustic signal hierarchical cavitation intensity identification method based on hierarchical transition network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |