SVR combustion instability prediction method and system

Abstract

The invention discloses an SVR combustion instability prediction method and system, and relates to the field of energy power and intelligent control, and the method comprises the steps: obtaining a fault combustion flame chemiluminescence signal, and generating a signal data set comprising a training data set and a test data set; establishing a support vector regression model based on the training data set; evaluating the support vector regression model by using the test data set, and adjusting according to an evaluation result to obtain a prediction model; using the prediction model to predict to-be-predicted sample data; the historical flame chemiluminescence signal data of the gas turbine under different working conditions are adopted as data input and can be used for predicting the working state of the gas turbine after a certain time delay, and the model can perform algorithm processing on the flame chemiluminescence signal measured by the gas turbine and predict the change trend of the future moment. The occurrence time of the unstable combustion can be predicted, and enough time is reserved for the control process of the unstable combustion.

Description

Technical Field

[0001] The present invention relates to the fields of energy power and intelligent control, and in particular to a method and system for predicting SVR combustion instability. Background Art

[0002] Combustion instability is a common problem in modern gas turbines and aircraft engines using lean premixed combustion technology. Thermoacoustic instability not only increases combustion pollutants but also causes localized flameouts. Long-term operation can even damage the gas turbine structure and reduce the life of the combustion chamber. Therefore, the question of what signal to use and how to predict the occurrence of thermoacoustic oscillations has become a research focus. Because thermoacoustic oscillations are induced by numerous factors, involving multiple disciplines and influencing mechanisms, it is very difficult to propose a reasonable combustion instability prediction model. In practical engineering, the commonly used pressure pulsation signal suffers from data monitoring lags due to its installation location being far from the measurement point. Therefore, it is necessary to develop a high-precision, universal, and accurate combustion instability prediction model.

[0003] Compared to traditional pressure signals, combustion chamber flame chemiluminescence signals have a faster response speed when a fault occurs. This is because combustion instability is mainly caused by three influencing factors: 1) fluctuations in combustion velocity, entropy waves, or flame heat release, 2) acoustic pressure pulsations in the combustion chamber, and 3) fluctuations in the flow field within the combustion chamber. These factors may also cause changes in the flame structure, making the flame structure characteristic signal more predictive than the pressure pulsation signal and responding faster. Compared with other time series prediction models, the Support Vector Regression (SVR) algorithm can use less data for model generation, and has higher computational efficiency and accuracy. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing combustion instability prediction method has the problem of data monitoring lag in the pressure pulsation signal acquisition and lacks a high-precision and universal prediction model.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting SVR combustion instability, comprising:

[0008] Acquire a faulty combustion flame chemiluminescence signal and generate a signal data set including a training data set and a test data set;

[0009] Based on the training data set, a support vector regression model is established;

[0010] Use the test data set to evaluate the support vector regression model and make adjustments based on the evaluation results to obtain a prediction model;

[0011] Use the prediction model to make predictions on the sample data to be predicted.

[0012] As a preferred method for predicting SVR combustion instability, the following is a proposed method:

[0013] The method of obtaining the faulty combustion flame chemiluminescence signal and generating a signal data set including a training data set and a test data set comprises:

[0014] The chemiluminescence signals of faulty combustion flames under different working conditions are collected, and the one-dimensional flame chemiluminescence signal time series is converted into a matrix form to obtain the correlation information between the data.

[0015] As a preferred method for predicting SVR combustion instability, the following is a proposed method:

[0016] The support vector regression model is established based on the training data set, including:

[0017] Select appropriate kernel function to establish support vector regression model;

[0018] The grid search optimization method is used to optimize the parameters of the penalty factor C and the kernel function coefficient g. The verification classification accuracy of the training set under each set of C and g is calculated, and the set of C and g that makes the verification classification accuracy the highest is selected as the optimal parameter.

[0019] As a preferred method for predicting SVR combustion instability, the following is a proposed method:

[0020] The grid search optimization method is used to optimize the parameters of the penalty factor C and the kernel function coefficient g, including:

[0021] Take the training data set as the original data set, randomly divide the data set into K subsets of equal size, use each subset data as a validation set, and use the remaining K-1 groups of subset data as training sets to obtain K models. The average classification accuracy of the final validation set of the K models is used as the performance indicator of the classifier under this K-CV. The parameter combination that gives the highest performance indicator is selected as the optimal parameter.

[0022] As a preferred method for predicting SVR combustion instability, the following is a proposed method:

[0023] The support vector regression model is evaluated using the test data set and adjusted according to the evaluation results to obtain a prediction model including:

[0024] Based on the support vector regression model and the test dataset, select evaluation indicators to measure model performance;

[0025] Input the test data set into the support vector regression model to obtain the corresponding prediction results;

[0026] According to the selected evaluation indicators, calculate the evaluation indicator values ​​of the model on the test data set.

[0027] As a preferred method for predicting SVR combustion instability, the following is a proposed method:

[0028] The method of evaluating the support vector regression model using the test data set and adjusting the prediction model based on the evaluation results further includes:

[0029] If the evaluation index is lower than the preset threshold, the support vector regression model is adjusted. The adjustment direction includes reselecting the kernel function, expanding or narrowing the search range of C and g, or adjusting the K value and the maximum number of iterations.

[0030] The process of establishing and evaluating the support vector regression model is repeated until the evaluation index of the support vector regression model on the test data set reaches a preset threshold, at which point a prediction model is obtained.

[0031] As a preferred method for predicting SVR combustion instability, the following is a proposed method:

[0032] The use of the prediction model to predict the sample data to be predicted includes:

[0033] Denormalize the prediction results to restore them to the scale of the original data;

[0034] The denormalized prediction results are analyzed to determine the occurrence time of combustion instability or other related states. The prediction results are applied to the actual combustion instability control process to provide a basis for control decision-making.

[0035] In a second aspect, an embodiment of the present invention provides an SVR combustion instability prediction system, comprising:

[0036] A data set acquisition module is used to acquire the chemiluminescence signal of the faulty combustion flame and generate a signal data set including a training data set and a test data set;

[0037] Regression model building module, used to build a support vector regression model based on the training data set;

[0038] The prediction model generation module is used to evaluate the support vector regression model using the test data set and make adjustments based on the evaluation results to obtain the prediction model;

[0039] The prediction module is used to use the prediction model to predict the sample data to be predicted.

[0040] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0041] memory and processor;

[0042] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the SVR combustion instability prediction method as described in any embodiment of the present invention.

[0043] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the SVR combustion instability prediction method.

[0044] The present invention utilizes historical flame chemiluminescence signal data from a gas turbine under various operating conditions as input to predict the operating state of the gas turbine after a certain time delay. The model then processes the measured flame chemiluminescence signal to predict future trends. This allows for the prediction of the onset of combustion instability, allowing sufficient time for the control process to address this issue. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0046] Figure 1 is an overall flow chart of the SVR combustion instability prediction method of the present invention;

[0047] Figure 2 1. A graph showing the test value and predicted value of a combustion pressure pulsation curve in a simulation example of the SVR combustion instability prediction method of the present invention;

[0048] Figure 3 is a combustion pressure pulsation prediction curve error diagram in a simulation example of the SVR combustion instability prediction method of the present invention;

[0049] Figure 4 It is a graph showing the test value and predicted value of the combustion flame chemiluminescence signal curve in a simulation example of the SVR combustion instability prediction method of the present invention;

[0050] Figure 5 It is a graph of the error in the prediction curve of the combustion flame chemiluminescence signal in the simulation example of the SVR combustion instability prediction method of the present invention. DETAILED DESCRIPTION

[0051] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0052] Example 1, with reference to Figure 1 , which is a first embodiment of the present invention, provides a method for predicting SVR combustion instability, comprising:

[0053] S1: Acquire the chemiluminescence signal of the faulty combustion flame and generate a signal dataset including a training dataset and a test dataset;

[0054] S2: Based on the training data set, a support vector regression model is established;

[0055] S3: Use the test data set to evaluate the support vector regression model and make adjustments based on the evaluation results to obtain a prediction model;

[0056] S4: Use the prediction model to predict the sample data to be predicted.

[0057] It should be noted that the flame chemiluminescence signal-based support vector regression prediction model constructed by executing steps S1-S4 above can effectively predict combustion instability using the flame chemiluminescence signal of faulty combustion. This prediction model can accurately predict the sample data to be predicted, issuing early warnings for the control process of combustion instability, allowing operators sufficient time to take appropriate measures such as adjusting combustion parameters and optimizing combustion strategies, thereby effectively avoiding the occurrence of combustion instability and improving the safety, stability, and operational efficiency of combustion equipment. This method has significant application value and promotional significance in the field of operation monitoring and control of gas turbines and other related combustion equipment.

[0058] Example 2, reference Figure 1 , which is an embodiment of the present invention, provides an SVR combustion instability prediction method based on the previous embodiment, including:

[0059] In this embodiment, obtaining the faulty combustion flame chemiluminescence signal and generating a signal dataset including a training dataset and a test dataset in the above step S1 includes:

[0060] The chemiluminescence signals of faulty combustion flames under different working conditions are collected, and the one-dimensional flame chemiluminescence signal time series is converted into a matrix form to obtain the correlation information between the data.

[0061] Specifically, let the time series x={x1,x2,…,x n}, for the value x at time t t , which corresponds to the information of any previous p moments, that is, when the input vector x p ={x1,x2,…,x p}, we can predict x t , where p is the embedding dimension. Similarly, taking the matrix X consisting of the input vectors as input, we can output a continuous time series, represented as a one-dimensional vector Y, which can be expressed as:

[0062]

[0063] Where τ is the delay time, which can be changed to meet the needs of the actual prediction model; the values ​​of X and Y are normalized to the range [0,1] to increase the resolution of the data.

[0064] Divide the data set into training data set and test data set; if necessary, perform preprocessing such as data cleaning and removing bad values ​​on abnormal data to ensure the quality and reliability of the data.

[0065] In another possible embodiment, when converting the one-dimensional flame chemiluminescence signal time series into a matrix form, wavelet transform can be used to replace simple time series mapping. Wavelet transform has the characteristics of multi-resolution analysis and can decompose the signal into components of different frequencies and extract local features in the signal. The specific operation is to carry out wavelet decomposition to the collected chemiluminescence signal, select a suitable wavelet basis (such as Daubechies wavelet) and decomposition level, and decompose the signal into approximate components and detail components of different scales. These components are then combined into data in matrix form, which can better capture the fluctuation and mutation information in the signal.

[0066] In another possible implementation, after normalizing the data, principal component analysis (PCA) can be used to reduce the data's dimensionality. PCA can identify the main characteristic directions in the data, projecting high-dimensional data into a low-dimensional space while retaining the data's primary information. For the dataset that has undergone matrix transformation and normalization, its covariance matrix is ​​calculated, and then the eigenvalues ​​and eigenvectors of the covariance matrix are solved. The principal components corresponding to the eigenvectors with larger eigenvalues ​​are selected, and the original data are projected onto these principal components to obtain the reduced-dimensional data. This can reduce redundant information in the data and improve the efficiency of subsequent model training.

[0067] In this embodiment, establishing a support vector regression model based on the training data set in step S2 includes:

[0068] The radial basis kernel function (RBF) is selected to establish the support vector regression model, which is expressed as:

[0069]

[0070] Among them, x i and x j are the input vector and the center vector, ||x i -x j || represents the Euclidean distance between them, g is the kernel parameter, when x i Stay away from x j When , the function value is very small.

[0071] The grid search optimization method is used to optimize the parameters of the penalty factor C and the kernel function coefficient g. The verification classification accuracy of the training set under each set of C and g is calculated, and the set of C and g that gives the highest verification classification accuracy is selected as the optimal parameter.

[0072] Specifically, the training dataset is used as the original dataset, and the dataset is randomly divided into K subsets of equal size. Each subset data is used as a validation set, and the remaining K-1 groups of subset data are used as training sets to obtain K models. The average classification accuracy of the final validation set of these K models is used as the performance indicator of the classifier under this K-CV, and the parameter combination that gives the highest performance indicator is selected as the optimal parameter.

[0073] Finally, we select an appropriate value for K (e.g., 3), a maximum number of iterations (e.g., 20), a search range for C (e.g., 0-100), and a search range for g (e.g., 0-100). After multiple calculations, we determine the specific values ​​for C and g (e.g., 4.0 for C and 0.8 for the kernel function coefficient g). Furthermore, the loss function parameter is typically set to 0.01 based on experience to control the regression bandwidth.

[0074] In this embodiment, K is preferably 3, the maximum number of iterations is 20, the search range of C is 0-100, and the search range of g is 0-100. After multiple calculations, the size of C is 4.0, and the size of the kernel function coefficient g is 0.8.

[0075] In another possible implementation, a genetic algorithm can be used instead of the grid search optimization method for parameter optimization. Specifically, a set of parameter individuals, including a penalty factor C and a kernel function coefficient g, is randomly initialized, with each individual representing a possible parameter combination. A fitness function is then defined based on the model's validation classification accuracy on the training set, and each individual is evaluated. Through genetic operations such as selection, crossover, and mutation, new parameter individuals are continuously generated until the individual with the highest fitness value, i.e., the optimal parameter combination, is found.

[0076] In another possible implementation, a particle swarm optimization algorithm can be used instead of a grid search optimization method for parameter optimization. Specifically, each particle represents a set of parameters (C and g). The particle moves in the parameter space, updating its position based on its own historical optimal position and the historical optimal position of the swarm. Through continuous iteration, the particles gradually converge to the optimal solution. In practical applications, appropriate parameters such as the number of particles, number of iterations, and learning factor are set to enable the algorithm to efficiently find the optimal values ​​of C and g.

[0077] In this embodiment, the support vector regression model is evaluated using the test data set in step S3, and is adjusted according to the evaluation results to obtain a prediction model including:

[0078] Based on the support vector regression model and the test dataset, select appropriate evaluation metrics to measure model performance.

[0079] Common evaluation metrics include mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination.

[0080] The test data set is input into the support vector regression model to obtain the corresponding prediction results.

[0081] According to the selected evaluation indicators, calculate the evaluation indicator values ​​of the model on the test data set;

[0082] If the evaluation index shows that the model performance does not meet expectations, that is, the evaluation index is lower than the preset threshold, the support vector regression model is adjusted. The adjustment direction includes reselecting the kernel function and re-optimizing the parameters (expanding or narrowing the search range of C and g, or adjusting the K value and the maximum number of iterations).

[0083] The process of establishing and evaluating the support vector regression model is repeated until the evaluation index of the support vector regression model on the test data set reaches a preset threshold, at which point a prediction model is obtained.

[0084] In this embodiment, the above step S4 uses the prediction model to predict the sample data to be predicted, including:

[0085] Since the prediction results are obtained based on normalized data, it is necessary to perform a denormalization operation on the prediction results to restore them to the scale of the original data;

[0086] The denormalized prediction results are analyzed to determine the occurrence time of combustion instability or other related states. The prediction results are applied to the actual combustion instability control process to provide a basis for control decisions, such as taking measures in advance to adjust combustion parameters to avoid the occurrence of combustion instability.

[0087] Example 3. The above is a schematic diagram of the SVR combustion instability prediction method of this embodiment. It should be noted that the technical solutions of the SVR combustion instability prediction system and the above-mentioned SVR combustion instability prediction method are based on the same concept. For details not described in detail in the technical solution of the SVR combustion instability prediction system of this embodiment, please refer to the description of the technical solution of the above-mentioned SVR combustion instability prediction method.

[0088] This embodiment further provides a system based on the SVR combustion instability prediction method, including:

[0089] A data set acquisition module is used to acquire the chemiluminescence signal of the faulty combustion flame and generate a signal data set including a training data set and a test data set;

[0090] Regression model building module, used to build a support vector regression model based on the training data set;

[0091] The prediction model generation module is used to evaluate the support vector regression model using the test data set and make adjustments based on the evaluation results to obtain the prediction model;

[0092] The prediction module is used to use the prediction model to predict the sample data to be predicted.

[0093] This embodiment further provides a computing device applicable to the SVR combustion instability prediction method, including:

[0094] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the SVR combustion instability prediction method proposed in the above embodiment.

[0095] This embodiment further provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the SVR combustion instability prediction method proposed in the above embodiment is implemented.

[0096] The storage medium proposed in this embodiment and the SVR combustion instability prediction method proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0097] Example 4, with reference to Figure 2-Figure 5 , which is an embodiment of the present invention, provides an SVR combustion instability prediction method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0098] 2000 consecutive data points were extracted from the collected combustion pressure pulsation data to establish a training set. The embedding dimension p was set to 10, and the delay time τ was set to 50ms. The data was converted into a matrix according to the data processing method proposed in the modeling theory, resulting in a total of 1990 data sets. The input X is a 10-dimensional vector group, and the output Y is a continuous time series consisting of 1990 data points. The built model is applied to any experimental acquisition sequence that has not participated in the model training. The prediction curve is as follows: Figure 2 As shown in the figure, the test value is the combustion pressure pulsation data collected experimentally, the predicted value is the output result of the prediction model, and the time t=0 is defined as the time when the model starts to predict, that is, the starting point of the one-dimensional vector Y. Figure 3 It can be seen that the combustion pressure pulsation prediction curve based on the support vector machine is basically consistent with the experimentally measured pressure curve, and the maximum relative error of the prediction value (error / average of the measured absolute value) is 4.2%.

[0099] 2000 consecutive data were randomly selected from the collected combustion flame chemiluminescence signal data to establish a training set. The other parameter settings were the same as those of the pressure data above. The predicted and measured values ​​of the flame chemiluminescence signal were obtained as shown in the figure. Figure 4 As shown in Figure 2, it can be seen that the predicted value curve of combustion flame chemiluminescence signal based on support vector machine is basically consistent with the experimental curve. Figure 5 As shown, the maximum relative error of the predicted value is 2.5%.

[0100] Therefore, for the same prediction model, the flame chemiluminescence signal generated during the combustion process provides better prediction results than pressure pulsation data. This is primarily because when collecting chemiluminescence signals, high-speed cameras and other equipment capture the full-scale flame, while pressure pulsation data is collected using a single-point pressure sensor, which may not capture all data features.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting SVR combustion instability, characterized in that: include: Acquire a faulty combustion flame chemiluminescence signal and generate a signal data set including a training data set and a test data set; Based on the training data set, a support vector regression model is established; Use the test data set to evaluate the support vector regression model and make adjustments based on the evaluation results to obtain a prediction model; Use the prediction model to make predictions on the sample data to be predicted.

2. The SVR combustion instability prediction method according to claim 1, wherein: The method of obtaining the faulty combustion flame chemiluminescence signal and generating a signal data set including a training data set and a test data set comprises: The chemiluminescence signals of faulty combustion flames under different working conditions are collected, and the one-dimensional flame chemiluminescence signal time series is converted into a matrix form to obtain the correlation information between the data.

3. The SVR combustion instability prediction method according to claim 2, wherein: The support vector regression model is established based on the training data set, including: Select appropriate kernel function to establish support vector regression model; The grid search optimization method is used to optimize the parameters of the penalty factor C and the kernel function coefficient g. The verification classification accuracy of the training set under each set of C and g is calculated, and the set of C and g that makes the verification classification accuracy the highest is selected as the optimal parameter.

4. The SVR combustion instability prediction method according to claim 3, wherein: The grid search optimization method is used to optimize the parameters of the penalty factor C and the kernel function coefficient g, including: Take the training data set as the original data set, randomly divide the data set into K subsets of equal size, use each subset data as a validation set, and use the remaining K-1 groups of subset data as training sets to obtain K models. The average classification accuracy of the final validation set of the K models is used as the performance indicator of the classifier under this K-CV. The parameter combination that gives the highest performance indicator is selected as the optimal parameter.

5. The SVR combustion instability prediction method according to claim 4, wherein: The support vector regression model is evaluated using the test data set and adjusted according to the evaluation results to obtain a prediction model including: Based on the support vector regression model and the test dataset, select evaluation indicators to measure model performance; Input the test data set into the support vector regression model to obtain the corresponding prediction results; According to the selected evaluation indicators, calculate the evaluation indicator values ​​of the model on the test data set.

6. The SVR combustion instability prediction method according to claim 5, wherein: The method of evaluating the support vector regression model using the test data set and adjusting the prediction model based on the evaluation results further includes: If the evaluation index is lower than the preset threshold, the support vector regression model is adjusted. The adjustment direction includes reselecting the kernel function, expanding or narrowing the search range of C and g, or adjusting the K value and the maximum number of iterations. The process of establishing and evaluating the support vector regression model is repeated until the evaluation index of the support vector regression model on the test data set reaches a preset threshold, at which point a prediction model is obtained.

7. The SVR combustion instability prediction method according to claim 6, wherein: The use of the prediction model to predict the sample data to be predicted includes: Denormalize the prediction results to restore them to the scale of the original data; The denormalized prediction results are analyzed to determine the occurrence time of combustion instability or other related states. The prediction results are applied to the actual combustion instability control process to provide a basis for control decision-making.

8. An SVR combustion instability prediction system, using the method according to any one of claims 1 to 7, characterized in that: include: A data set acquisition module is used to acquire the chemiluminescence signal of the faulty combustion flame and generate a signal data set including a training data set and a test data set; The regression model building module is used to build a support vector regression model based on the training data set; The prediction model generation module is used to evaluate the support vector regression model using the test data set and make adjustments based on the evaluation results to obtain the prediction model; The prediction module is used to use the prediction model to predict the sample data to be predicted.

9. A computing device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images