Gas turbine combustion state real-time monitoring system and method

Abstract

The embodiment of the invention provides a gas turbine combustion state real-time monitoring system and method, and solves the problems that when a plurality of combustion state parameters are integrated into a comprehensive parameter in an existing method, multi-source heterogeneous combustion state parameters cannot be effectively fused, so that the comprehensive parameter has deviation and is inaccurate in evaluation. The method comprises the following steps: acquiring combustion state parameters of a gas turbine, constructing and training a state monitoring model by combining category joint confrontation and a gating circulation unit, identifying and analyzing the combustion state parameters by the state monitoring model, and outputting a comprehensive evaluation value; according to the method, the state monitoring model is constructed and trained by combining the category joint confrontation and the gating circulation unit, so that the state monitoring model accurately identifies and analyzes the combustion state parameters, further outputs the comprehensive evaluation value, can accurately capture the change trend of the combustion state parameters, and improves the evaluation accuracy of the combustion state parameters. Therefore, the potential fatigue problem of the gas turbine can be conveniently found in time, the service life of the gas turbine is prolonged, and sudden faults of equipment are avoided.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of gas turbine monitoring, and in particular to a system and method for real-time monitoring of the combustion state of a gas turbine. Background Art

[0002] The main task of gas turbine combustion diagnosis is to obtain the state parameters of the combustion chamber such as pressure, temperature and composition through effective methods, and then judge the combustion state of the combustion chamber, realize combustion stability, combustion temperature diagnosis, etc., and then cooperate with the gas turbine control system to ensure that the combustion chamber always operates in a safe, stable and low-emission state. By monitoring the state of the combustion chamber in real time, abnormalities can be discovered in time, such as unstable phenomena such as flashback or nozzle extinction, so as to take preventive measures to avoid potential safety risks.

[0003] Chinese patent CN114993689A discloses a method, device, equipment and medium for evaluating the combustion state of a gas turbine. The method includes obtaining various combustion state parameters in the gas turbine, each combustion state parameter at least including exhaust dispersion, combustion chamber dynamic pressure level, NOx emission concentration and CO emission concentration, integrating various combustion state parameters into a comprehensive parameter, calculating the value of the comprehensive parameter each time after the combustion adjustment test is completed, and evaluating the combustion state of the gas turbine according to the value of the comprehensive parameter. However, the existing method cannot effectively integrate multi-source heterogeneous combustion state parameters when integrating several combustion state parameters into one comprehensive parameter, so that the comprehensive parameter has deviations and inaccurate evaluation. In view of the above problems, we propose a real-time monitoring system and method for the combustion state of a gas turbine. Summary of the invention

[0004] The embodiments of the present invention aim to solve at least one of the technical problems existing in the prior art, and provide a system and method for real-time monitoring of the combustion state of a gas turbine.

[0005] In a first aspect, an embodiment of the present invention provides a method for real-time monitoring of combustion status of a gas turbine, the method comprising:

[0006] Obtaining combustion state parameters of the gas turbine, and uploading the combustion state parameters to a state database, where the state database stores the combustion state parameters as time-series historical state parameters;

[0007] Traverse the state database, extract modeling samples from the state database, divide the modeling samples into training sets and test sets, build and train the state monitoring model by combining category joint adversarial and gated recurrent units, and output the trained state monitoring model;

[0008] The state monitoring model is executed with the combustion state parameters as input. The state monitoring model identifies and analyzes the combustion state parameters and outputs a comprehensive evaluation value.

[0009] Determine whether the comprehensive evaluation value exceeds the preset state threshold. If it exceeds the preset state threshold, trigger the combustion warning instruction.

[0010] In some possible embodiments, the method further includes:

[0011] Obtaining a comprehensive assessment value, a host group associated with the comprehensive assessment value, and at least one group of associated units, and calculating a risk assessment value of the associated units associated with the host group based on a conditional value at risk method;

[0012] Determine whether the risk assessment value exceeds the preset risk threshold. If it exceeds the preset risk threshold, trigger the combustion risk instruction.

[0013] In some possible embodiments, the method of combining category joint adversarial and gated recurrent units to construct and train a state monitoring model specifically includes:

[0014] Taking the category joint adversarial model as the initial model, a state monitoring model is built, wherein the initial model consists of a feature extractor, a category discriminator, and a classifier, wherein the feature extractor is connected to the category discriminator, and the category discriminator is connected to the classifier;

[0015] Introducing a data fusion layer into the initial model, the data fusion layer is built based on the gated recurrent unit;

[0016] Get the training set, use the training set as input, set the model iteration training rounds, iteratively train the initial model until convergence, and output the initial model.

[0017] In some possible embodiments, the method of combining category joint adversarial and gated recurrent units to construct and train a state monitoring model specifically further includes:

[0018] Obtain a test set and use the test set to test the model accuracy of the initial model;

[0019] Based on the preset accuracy threshold, determine whether the model accuracy meets the accuracy threshold. If it meets the preset accuracy threshold, output the state monitoring model;

[0020] If the preset accuracy threshold is not met, the cross entropy loss is used to optimize the initial model hyperparameters, and the initial model is iteratively trained again until convergence.

[0021] In some possible embodiments, the method of optimizing the initial model hyperparameters using cross entropy loss specifically includes:

[0022] Define the loss function of the initial model as the cross entropy loss function;

[0023] The CIoU component is introduced into the cross entropy loss function, where the CIoU component is expressed as:

[0024]

[0025] Among them, x u For the training set input representation, represents the mean variance of the training set, y u represents the true value of the training set, Represents the predicted value of the training set;

[0026] The loss function is expressed as:

[0027]

[0028] Based on the stochastic gradient descent algorithm, the model parameters with the minimum cross entropy loss are found, and the adjusted model parameters are fed back to the initial model to achieve initial model hyperparameter optimization.

[0029] In some possible embodiments, the method of the state monitoring model identifying and analyzing the combustion state parameters and outputting the comprehensive evaluation value specifically includes:

[0030] Obtain combustion state parameters. The data fusion layer performs Fourier transform on the multi-source heterogeneous combustion state parameters and projects the combustion state parameters into the Fourier domain.

[0031] The data fusion layer introduces the combustion state parameters in the Fourier domain into the frequency domain convolution fusion to obtain the feature fusion set;

[0032] Load the feature fusion set. The feature extractor extracts the input feature fusion set based on the Hi-att network, indexes the feature factors embedded in the feature fusion set through the Hi-att network, enhances the feature factors by feature pooling, and outputs the feature extraction set after the feature factors are enhanced.

[0033] x t =f(x r *[x y :x y+1 ]+b x )(5)

[0034] Among them, x t represents the output representation of the feature extraction set, f is the activation function of the feature extractor, x r represents the input feature fusion set, b x represents the bias vector of the Hi-att network, [x y :x y+1 ] represents the set of characteristic factors;

[0035] Obtain a feature extraction set, and the category discriminator judges the state category to which the feature extraction set belongs based on the naive Bayes algorithm, and outputs a data judgment set;

[0036] Among them, the judgment function of the category discriminator is expressed as:

[0037]

[0038] Among them, x p represents the data judgment set, n represents the size of the feature extraction set, P represents the thickness of the judgment function judgment layer, and α represents the category weight coefficient;

[0039]

[0040] Among them, α represents the category weight coefficient, α0 is the initial weight constant based on principal component analysis, and g represents the coupling coefficient of the naive Bayes algorithm.

[0041] In some possible embodiments, the method for the state monitoring model to identify and analyze combustion state parameters and output a comprehensive evaluation value specifically further includes:

[0042] Obtain a data judgment set, and the classifier calculates the state classification value of the data judgment set based on the attribute attention mechanism;

[0043] The classifier combines the data classification value and the quantile performance algorithm to calculate the comprehensive evaluation value.

[0044] In some possible embodiments, the method for calculating the state classification value specifically includes:

[0045]

[0046] Among them, y p represents the state classification value, score(·) is the attribute attention score function, x m p represents the weight of the mth state level, and k represents the number of state levels;

[0047]

[0048] Among them, tanh is the activation function of the classifier, W x is the attention weight matrix, b p Represents the weight bias term.

[0049] In a second aspect, an embodiment of the present invention provides a gas turbine combustion state real-time monitoring system, which is used to implement the gas turbine combustion state real-time monitoring method as described above, and the gas turbine combustion state real-time monitoring system specifically includes:

[0050] A parameter acquisition module is used to obtain the combustion state parameters of the gas turbine and upload the combustion state parameters to the state database, which stores the combustion state parameters as time-series historical state parameters;

[0051] The model training module is used to traverse the state database, extract modeling samples from the state database, divide the modeling samples into training sets and test sets, build and train the state monitoring model by combining category joint confrontation and gated recurrent units, and output the trained state monitoring model;

[0052] The comprehensive evaluation module takes the combustion state parameters as input and executes the state monitoring model. The state monitoring model identifies and analyzes the combustion state parameters and outputs a comprehensive evaluation value.

[0053] The first judgment module is used to judge whether the comprehensive evaluation value exceeds the preset state threshold, and if it exceeds the preset state threshold, trigger the combustion warning instruction;

[0054] The second judgment module is used to obtain the comprehensive evaluation value, the host group associated with the comprehensive evaluation value, and at least one group of associated units, calculate the risk assessment value of the associated units associated with the host group based on the conditional risk value method, and judge whether the risk assessment value exceeds the preset risk threshold. If it exceeds the preset risk threshold, the combustion risk instruction is triggered.

[0055] In some possible embodiments, the comprehensive evaluation module specifically includes:

[0056] The parameter transformation unit obtains the combustion state parameters. The data fusion layer performs Fourier transformation on the multi-source heterogeneous combustion state parameters and projects the combustion state parameters into the Fourier domain.

[0057] The parameter fusion unit introduces the combustion state parameters in the Fourier domain into the frequency domain convolution fusion based on the data fusion layer to obtain a feature fusion set;

[0058] Feature extraction layer, loads the feature fusion set, and the feature extractor extracts the input feature fusion set based on the Hi-att network, indexes the feature factors embedded in the feature fusion set through the Hi-att network, enhances the feature factors by feature pooling, and outputs the feature extraction set after the feature factors are enhanced;

[0059] The parameter judgment unit is used to obtain a feature extraction set. The category discriminator judges the state category to which the feature extraction set belongs based on the naive Bayes algorithm and outputs a data judgment set.

[0060] The comprehensive evaluation unit is used to obtain the data judgment set. The classifier calculates the state classification value of the data judgment set based on the attribute attention mechanism. The classifier calculates the comprehensive evaluation value by combining the data classification value and the quantile performance algorithm.

[0061] Compared with the prior art, the embodiments of the present invention mainly have the following beneficial effects:

[0062] In the embodiment of the present invention, the state monitoring model is constructed and trained by combining the category joint confrontation and the gated loop unit, so that the state monitoring model can accurately identify and analyze the combustion state parameters, and then output a comprehensive evaluation value. The state monitoring model can accurately capture the change trend of the combustion state parameters, so as to facilitate the timely discovery of potential fatigue problems of the gas turbine, extend the service life of the gas turbine, and avoid sudden equipment failures. The existing method overcomes the problem that when integrating several combustion state parameters into one comprehensive parameter, it cannot effectively integrate the combustion state parameters of multiple sources and heterogeneity, thereby causing the comprehensive parameter to have deviations and inaccurate evaluations.

[0063] In an embodiment of the present invention, a state monitoring model and a training method are provided. When the state detection model is constructed, a category joint adversarial model is used as the initial model, so that the combustion state parameters can be feature fused, extracted and discriminated respectively, ensuring that multi-source heterogeneous combustion state parameters can be effectively fused, so that the model performs well when processing combustion state parameters with uncertainty and noise.

[0064] In the embodiment of the present invention, the combustion state parameters in the Fourier domain are introduced into the frequency domain convolution fusion through the data fusion layer, thereby realizing the effective fusion of multi-source heterogeneous combustion state parameters, and further facilitating the state monitoring model to identify and analyze the combustion state parameters. On the one hand, the frequency domain convolution fusion can help to more accurately identify abnormal conditions in the combustion process and provide strong data support for fault diagnosis. On the other hand, the combustion state parameters are fused with other related data in the frequency domain to achieve information complementarity and improve the comprehensiveness of data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0066] Figure 1 It is a schematic diagram of the implementation flow of the method for real-time monitoring of the combustion state of a gas turbine provided by the present invention;

[0067] Figure 2 It is a schematic diagram of the implementation flow of the method for constructing and training a state monitoring model by combining category joint confrontation and gated recurrent units provided by the present invention;

[0068] Figure 3 It is a schematic diagram of a process for implementing a method for optimizing the initial model hyperparameters using cross entropy loss provided by the present invention;

[0069] Figure 4It is a schematic diagram of a flow chart of a method for realizing a state monitoring model provided by the present invention to identify and analyze combustion state parameters and output a comprehensive evaluation value;

[0070] Figure 5 It is a structural schematic diagram of a gas turbine combustion state real-time monitoring system provided by the present invention;

[0071] Figure 6 It is a structural schematic diagram of the comprehensive evaluation module provided by the present invention. DETAILED DESCRIPTION

[0072] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0073] Unless otherwise specified, the technical terms or scientific terms used in the embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. "Including" or "comprising" used in the embodiments of the present invention neither limit the shapes, numbers, steps, actions, operations, components, originals and / or their groups mentioned, nor exclude the appearance or addition of one or more other different shapes, numbers, steps, actions, operations, components, originals and / or their groups, or the addition of these. In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number and order of the indicated technical features. Thus, the features defined as "first" and "second" can explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0074] Unless otherwise specifically stated, the relative arrangement of the components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention. At the same time, it should be understood that for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship, and the techniques, methods and devices known to ordinary technicians in the relevant fields may not be discussed in detail, but where appropriate, the techniques, methods and devices shown should be considered as part of the authorized specification. In all examples shown and discussed here, any specific other examples may have different values. It should be noted that similar symbols and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0075] In the description of the embodiments of the present invention, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in the embodiments of the present invention and the features of the different embodiments or examples, without contradiction.

[0076] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described here.

[0077] When the existing method integrates several combustion state parameters into one comprehensive parameter, it cannot effectively integrate the combustion state parameters of multiple sources and heterogeneity, so that the comprehensive parameters have deviations and inaccurate evaluation. In view of the above problems, we propose a real-time monitoring system and method for the combustion state of a gas turbine. In short, when the method is executed, the combustion state parameters of the gas turbine are first obtained, and the state monitoring model is constructed and trained in combination with the category joint confrontation and the gated cycle unit. Then, the state monitoring model identifies and analyzes the combustion state parameters, outputs a comprehensive evaluation value, and finally obtains the comprehensive evaluation value, the host group associated with the comprehensive evaluation value, and at least one group of associated units. The risk evaluation value of the associated unit associated with the host group is calculated based on the conditional risk value method. In the embodiment of the present invention, by combining the category joint confrontation and the gated cycle unit to construct and train the state monitoring model, the state monitoring model can accurately identify and analyze the combustion state parameters, and then output the comprehensive evaluation value. The state monitoring model can accurately capture the change trend of the combustion state parameters, so as to facilitate the timely discovery of potential fatigue problems of the gas turbine, extend the service life of the gas turbine, and avoid sudden equipment failures. This overcomes the problem that the existing method cannot effectively integrate multiple heterogeneous combustion state parameters when integrating several combustion state parameters into one comprehensive parameter, resulting in deviations in the comprehensive parameters and inaccurate evaluation.

[0078] The embodiment of the present invention provides a method for real-time monitoring of combustion status of a gas turbine. Figure 1 The figure shows a schematic diagram of the process flow of the gas turbine combustion state real-time monitoring method, which specifically includes:

[0079] Step S10, obtaining the combustion state parameters of the gas turbine, and uploading the combustion state parameters to a state database, where the state database stores the combustion state parameters as time-series historical state parameters.

[0080] It should be noted that the combustion state parameters include but are not limited to combustion stability parameters, combustion emission parameters, and unit operation parameters. The combustion stability parameters include fire detection intensity, exhaust temperature parameters, and flame state images, while the unit operation parameters include but are not limited to gas turbine facility power, load, voltage, and current values.

[0081] Step S20, traverse the state database, extract modeling samples from the state database, divide the modeling samples into a training set and a test set, build and train a state monitoring model by combining category joint adversarial and gated recurrent units, and output the trained state monitoring model.

[0082] Step S30, using the combustion state parameters as input, executing the state monitoring model, the state monitoring model identifies and analyzes the combustion state parameters, and outputs a comprehensive evaluation value.

[0083] Step S40, determining whether the comprehensive evaluation value exceeds a preset state threshold.

[0084] Step S50: If the preset state threshold is exceeded, a combustion warning instruction is triggered.

[0085] It should be noted that the preset state threshold can be 0.35-0.4. When the comprehensive evaluation value of the host group exceeds the preset state threshold, the combustion warning instruction is triggered. The combustion warning instruction is presented to the background monitor and the user end based on the communication protocol.

[0086] Step S60, obtaining a comprehensive evaluation value, a host group associated with the comprehensive evaluation value, and at least one group of associated machine groups, and calculating a risk evaluation value of an associated machine group associated with the host group based on a conditional value at risk method.

[0087] Step S70, determining whether the risk assessment value exceeds a preset risk threshold.

[0088] Step S80: If the preset risk threshold is exceeded, a combustion risk instruction is triggered.

[0089] In this embodiment, the preset risk threshold is set to 0.25-0.3, and the associated unit in this embodiment refers to a unit that is associated with and connected to the main unit and works in coordination with the main unit. The present invention evaluates the gas turbine by collecting the main unit parameters, which can improve the system's load and parameter identification response capabilities.

[0090] In the embodiment of the present invention, the state monitoring model is constructed and trained by combining the category joint confrontation and the gated loop unit, so that the state monitoring model can accurately identify and analyze the combustion state parameters, and then output a comprehensive evaluation value. The state monitoring model can accurately capture the change trend of the combustion state parameters, so as to facilitate the timely discovery of potential fatigue problems of the gas turbine, extend the service life of the gas turbine, and avoid sudden equipment failures. The existing method overcomes the problem that when integrating several combustion state parameters into one comprehensive parameter, it cannot effectively integrate the combustion state parameters of multiple sources and heterogeneity, thereby causing the comprehensive parameter to have deviations and inaccurate evaluations.

[0091] The embodiment of the present invention provides a method for constructing and training a state monitoring model by combining category joint confrontation and gated recurrent unit. Figure 2 The schematic diagram of the implementation flow of the method for constructing and training a state monitoring model by combining category joint confrontation and gated recurrent unit is shown. The method for constructing and training a state monitoring model by combining category joint confrontation and gated recurrent unit specifically includes:

[0092] Step S101, using the category joint adversarial model as the initial model, building a state monitoring model, wherein the initial model consists of a feature extractor, a category discriminator, and a classifier, the feature extractor is connected to the category discriminator, and the category discriminator is connected to the classifier.

[0093] Step S102, introducing a data fusion layer into the initial model, where the data fusion layer is constructed based on a gated recurrent unit.

[0094] Step S103, obtaining a training set, taking the training set as input, setting model iterative training rounds, iteratively training the initial model until convergence, and outputting the initial model.

[0095] It should be noted that, in this embodiment, the ratio of the training set to the test set can be 4:1 or 7:3, and the training rounds of the model can be 200-1000 times.

[0096] Step S104, obtaining a test set, and using the test set to test the model accuracy of the initial model.

[0097] Step S105: judging whether the model accuracy meets the accuracy threshold based on the preset accuracy threshold.

[0098] In this embodiment, the preset accuracy threshold may be 0.95-0.98.

[0099] Step S106: if the preset accuracy threshold is met, output the state monitoring model.

[0100] Step S107: If the preset accuracy threshold is not met, the cross entropy loss is used to optimize the initial model hyperparameters, and the initial model is iteratively trained again until convergence.

[0101] In an embodiment of the present invention, a state monitoring model and a training method are provided. When the state detection model is constructed, a category joint adversarial model is used as the initial model, so that the combustion state parameters can be feature fused, extracted and discriminated respectively, ensuring that multi-source heterogeneous combustion state parameters can be effectively fused, so that the model performs well when processing combustion state parameters with uncertainty and noise.

[0102] The embodiment of the present invention provides a method for optimizing the initial model hyperparameters using cross entropy loss. Figure 3 The schematic diagram of the implementation process of the method for optimizing the initial model hyperparameters by using the cross entropy loss is shown. The method for optimizing the initial model hyperparameters by using the cross entropy loss specifically includes:

[0103] Step S1071, define the loss function of the initial model as a cross entropy loss function.

[0104] Step S1072, introducing the CIoU component into the cross entropy loss function, where the CIoU component is expressed as:

[0105]

[0106] Among them, x u For the training set input representation, represents the mean variance of the training set, y u represents the true value of the training set, Represents the predicted value of the training set;

[0107] The loss function is expressed as:

[0108]

[0109] Step S1073, based on the stochastic gradient descent algorithm, find the model parameters with the minimum cross entropy loss, and feed the adjusted model parameters back to the initial model to achieve initial model hyperparameter optimization.

[0110] In the embodiment of the present invention, by optimizing the initial model hyperparameters using cross entropy loss, the convergence speed and accuracy of the model can be improved.

[0111] The embodiment of the present invention provides a method for identifying and analyzing combustion state parameters using a state monitoring model and outputting a comprehensive evaluation value. Figure 4 The schematic diagram of the implementation flow of the method for identifying and analyzing the combustion state parameters and outputting the comprehensive evaluation value by the state monitoring model is shown. The method for identifying and analyzing the combustion state parameters and outputting the comprehensive evaluation value by the state monitoring model specifically includes:

[0112] Step S201, obtaining combustion state parameters, the data fusion layer performs Fourier transform on the multi-source heterogeneous combustion state parameters, and projects the combustion state parameters into the Fourier domain.

[0113] Step S202: the data fusion layer introduces the combustion state parameters in the Fourier domain into the frequency domain convolution fusion to obtain a feature fusion set.

[0114] Step S203, loading the feature fusion set, the feature extractor performs feature extraction representation on the input feature fusion set based on the Hi-att network, indexes the feature factors embedded in the feature fusion set through the Hi-att network, enhances the feature factors by feature pooling, and outputs the feature extraction set after the feature factors are enhanced;

[0115] x t =f(x r *[x y :x y+1 ]+b x )(5)

[0116] Among them, x t represents the output representation of the feature extraction set, f is the activation function of the feature extractor, x r represents the input feature fusion set, b x represents the bias vector of the Hi-att network, [x y :x y+1 ] represents a set of characteristic factors.

[0117] Step S204, obtaining a feature extraction set, and the category discriminator judges the state category to which the feature extraction set belongs based on the naive Bayes algorithm, and outputs a data judgment set;

[0118] Among them, the judgment function of the category discriminator is expressed as:

[0119]

[0120] Among them, x p represents the data judgment set, n represents the size of the feature extraction set, P represents the thickness of the judgment layer of the judgment function, in this embodiment, the thickness of the judgment layer is set to 1-8, and α represents the category weight coefficient;

[0121]

[0122] Wherein, α represents the category weight coefficient, α0 is the initial weight constant based on principal component analysis, the initial weight constant is set to 1-6, and g represents the coupling coefficient of the naive Bayes algorithm. In this embodiment, the coupling coefficient is set to 0.2-0.6.

[0123] Step S205, obtaining a data judgment set, and the classifier calculates the state classification value of the data judgment set based on the attribute attention mechanism.

[0124] Step S206: The classifier calculates a comprehensive evaluation value by combining the data classification value and the quantile performance algorithm.

[0125] In this embodiment, the method for calculating the status classification value specifically includes:

[0126]

[0127] Among them, y p represents the state classification value, score(·) is the attribute attention score function, x m p represents the weight of the mth state level, k represents the number of state levels. It should be noted that the number of state levels can be 1-8;

[0128]

[0129] Among them, tanh is the activation function of the classifier, W x is the attention weight matrix, b p Represents the weight bias term.

[0130] In the embodiment of the present invention, the combustion state parameters in the Fourier domain are introduced into the frequency domain convolution fusion through the data fusion layer, thereby realizing the effective fusion of multi-source heterogeneous combustion state parameters, and further facilitating the state monitoring model to identify and analyze the combustion state parameters. On the one hand, the frequency domain convolution fusion can help to more accurately identify abnormal conditions in the combustion process and provide strong data support for fault diagnosis. On the other hand, the combustion state parameters are fused with other related data in the frequency domain to achieve information complementarity and improve the comprehensiveness of data analysis.

[0131] On the other hand, the embodiment of the present invention also provides a gas turbine combustion state real-time monitoring system, Figure 5 The structure diagram of the gas turbine combustion state real-time monitoring system is shown, and the gas turbine combustion state real-time monitoring system specifically includes:

[0132] The parameter acquisition module 100 is used to obtain the combustion state parameters of the gas turbine and upload the combustion state parameters to the state database, which stores the combustion state parameters as time-series historical state parameters;

[0133] The model training module 200 is used to traverse the state database, extract modeling samples from the state database, divide the modeling samples into a training set and a test set, build and train a state monitoring model by combining the category joint confrontation and the gated recurrent unit, and output the trained state monitoring model;

[0134] The comprehensive evaluation module 300 takes the combustion state parameters as input, executes the state monitoring model, and the state monitoring model identifies and analyzes the combustion state parameters and outputs a comprehensive evaluation value;

[0135] The first judgment module 400 is used to judge whether the comprehensive evaluation value exceeds a preset state threshold, and if it exceeds the preset state threshold, trigger a combustion warning instruction;

[0136] The second judgment module 500 is used to obtain the comprehensive evaluation value, the host group associated with the comprehensive evaluation value, and at least one group of associated units, calculate the risk evaluation value of the associated units associated with the host group based on the conditional risk value method, and judge whether the risk evaluation value exceeds the preset risk threshold. If it exceeds the preset risk threshold, the combustion risk instruction is triggered.

[0137] In this embodiment, the parameter acquisition module 100, the model training module 200, the comprehensive evaluation module 300, the first judgment module 400, and the second judgment module 500 realize data interaction through 5G communication or DTU communication connection. In the embodiment of the present invention, a real-time monitoring system for the combustion state of a gas turbine is provided. The real-time monitoring system for the combustion state of a gas turbine is composed of a parameter acquisition module 100, a model training module 200, a comprehensive evaluation module 300, a first judgment module 400, and a second judgment module 500. The parameter acquisition module 100, the model training module 200, the comprehensive evaluation module 300, the first judgment module 400, and the second judgment module 500 cooperate to facilitate the timely discovery of potential fatigue problems of the gas turbine, extend the service life of the gas turbine, and avoid sudden equipment failures. The existing method overcomes the problem that when integrating several combustion state parameters into one comprehensive parameter, it cannot effectively integrate the combustion state parameters of multiple sources and heterogeneity, thereby causing the comprehensive parameter to have deviations and inaccurate evaluations.

[0138] The embodiment of the present invention provides a comprehensive evaluation module. Figure 6 The schematic diagram of the structure of the comprehensive evaluation module is shown, and the comprehensive evaluation module 300 specifically includes:

[0139] The parameter transformation unit 310 obtains the combustion state parameters, and the data fusion layer performs Fourier transformation on the multi-source heterogeneous combustion state parameters, and projects the combustion state parameters into the Fourier domain;

[0140] The parameter fusion unit 320 introduces the combustion state parameters in the Fourier domain into the frequency domain convolution fusion based on the data fusion layer to obtain a feature fusion set;

[0141] The feature extraction layer 330 loads the feature fusion set. The feature extractor performs feature extraction representation on the input feature fusion set based on the Hi-att network, indexes the feature factors embedded in the feature fusion set through the Hi-att network, enhances the feature factors by feature pooling, and outputs the feature extraction set after the feature factors are enhanced.

[0142] The parameter judgment unit 340 is used to obtain a feature extraction set, and the category discriminator judges the state category to which the feature extraction set belongs based on the naive Bayes algorithm, and outputs a data judgment set;

[0143] The comprehensive evaluation unit 350 is used to obtain a data judgment set. The classifier calculates the state classification value of the data judgment set based on the attribute attention mechanism. The classifier calculates the comprehensive evaluation value by combining the data classification value and the quantile performance algorithm.

[0144] On the other hand, an embodiment of the present invention further provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the method of any one of the above embodiments is implemented.

[0145] As another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer program instructions, and the computer program instructions can be executed by a processor. When the computer program instructions are executed, the method of any one of the above embodiments is implemented.

[0146] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program instructions / modules corresponding to the real-time monitoring method for the combustion state of a gas turbine in the embodiment of the present application. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the use of the real-time monitoring method for the combustion state of a gas turbine, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the local module via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0147] The processor executes various functional applications and data processing of the server by running non-volatile software programs, instructions and modules stored in the memory, that is, realizes the real-time monitoring method of the combustion state of the gas turbine of the above method embodiment.

[0148] Finally, it should be noted that the computer-readable storage medium (e.g., memory) herein may be a volatile memory or a nonvolatile memory, or may include both a volatile memory and a nonvolatile memory. As an example and not by way of limitation, a nonvolatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. A volatile memory may include a random access memory (RAM), which may act as an external cache memory. As an example and not by way of limitation, RAM may be obtained in a variety of forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to include, but are not limited to, these and other suitable types of memory.

[0149] In summary, the present invention provides a real-time monitoring system and method for the combustion state of a gas turbine. When the method is executed, the combustion state parameters of the gas turbine are first obtained, and a state monitoring model is constructed and trained by combining category joint confrontation and gated loop units. Then, the state monitoring model identifies and analyzes the combustion state parameters, and outputs a comprehensive evaluation value. Finally, the comprehensive evaluation value, a host group associated with the comprehensive evaluation value, and at least one group of associated units are obtained, and the risk assessment value of the associated units associated with the host group is calculated based on the conditional risk value method.

[0150] In the embodiment of the present invention, the state monitoring model is constructed and trained by combining the category joint confrontation and the gated loop unit, so that the state monitoring model can accurately identify and analyze the combustion state parameters, and then output a comprehensive evaluation value. The state monitoring model can accurately capture the change trend of the combustion state parameters, so as to facilitate the timely discovery of potential fatigue problems of the gas turbine, extend the service life of the gas turbine, and avoid sudden equipment failures. The existing method overcomes the problem that when integrating several combustion state parameters into one comprehensive parameter, it cannot effectively integrate the combustion state parameters of multiple sources and heterogeneity, thereby causing the comprehensive parameter to have deviations and inaccurate evaluations.

[0151] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0152] In the several embodiments provided in the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the above-mentioned units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be in the form of telecommunication or other forms.

[0153] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0154] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for real-time monitoring of combustion status of a gas turbine, characterized in that: The gas turbine combustion state real-time monitoring method comprises: Obtaining combustion state parameters of the gas turbine, and uploading the combustion state parameters to a state database, where the state database stores the combustion state parameters as time-series historical state parameters; Traverse the state database, extract modeling samples from the state database, divide the modeling samples into training sets and test sets, build and train the state monitoring model by combining category joint adversarial and gated recurrent units, and output the trained state monitoring model; The state monitoring model is executed with the combustion state parameters as input. The state monitoring model identifies and analyzes the combustion state parameters and outputs a comprehensive evaluation value. Determine whether the comprehensive evaluation value exceeds the preset state threshold. If it exceeds the preset state threshold, trigger the combustion warning instruction.

2. The method for real-time monitoring of combustion status of a gas turbine according to claim 1, characterized in that: The method further comprises: Obtaining a comprehensive assessment value, a host group associated with the comprehensive assessment value, and at least one group of associated units, and calculating a risk assessment value of an associated unit associated with the host group based on a conditional value at risk method; Determine whether the risk assessment value exceeds the preset risk threshold. If it exceeds the preset risk threshold, trigger the combustion risk instruction.

3. The method for real-time monitoring of combustion status of a gas turbine according to claim 2, characterized in that: The method of combining category joint adversarial and gated recurrent units to construct and train a state monitoring model specifically includes: Taking the category joint adversarial model as the initial model, a state monitoring model is built, wherein the initial model consists of a feature extractor, a category discriminator, and a classifier, wherein the feature extractor is connected to the category discriminator, and the category discriminator is connected to the classifier; Introducing a data fusion layer into the initial model, the data fusion layer is built based on the gated recurrent unit; Get the training set, use the training set as input, set the model iteration training rounds, iteratively train the initial model until convergence, and output the initial model.

4. The method for real-time monitoring of combustion status of a gas turbine according to claim 3, characterized in that: The method of combining category joint adversarial and gated recurrent units to construct and train a state monitoring model specifically includes: Obtain a test set and use the test set to test the model accuracy of the initial model; Based on the preset accuracy threshold, determine whether the model accuracy meets the accuracy threshold. If it meets the preset accuracy threshold, output the state monitoring model; If the preset accuracy threshold is not met, the cross entropy loss is used to optimize the initial model hyperparameters, and the initial model is iteratively trained again until convergence.

5. The method for real-time monitoring of combustion status of a gas turbine according to claim 4, characterized in that: The method of optimizing the initial model hyperparameters using cross entropy loss specifically includes: Define the loss function of the initial model as the cross entropy loss function; The CIoU component is introduced into the cross entropy loss function, where the CIoU component is expressed as: Among them, x u For the training set input representation, represents the mean variance of the training set, y u represents the true value of the training set, Represents the predicted value of the training set; The loss function is expressed as: Based on the stochastic gradient descent algorithm, the model parameters with the minimum cross entropy loss are found, and the adjusted model parameters are fed back to the initial model to achieve initial model hyperparameter optimization.

6. The method for real-time monitoring of combustion status of a gas turbine according to claim 1, characterized in that: The method of the state monitoring model identifying and analyzing the combustion state parameters and outputting the comprehensive evaluation value specifically includes: Obtain combustion state parameters. The data fusion layer performs Fourier transform on the multi-source heterogeneous combustion state parameters and projects the combustion state parameters into the Fourier domain. The data fusion layer introduces the combustion state parameters in the Fourier domain into the frequency domain convolution fusion to obtain the feature fusion set; Load the feature fusion set. The feature extractor extracts the input feature fusion set based on the Hi-att network, indexes the feature factors embedded in the feature fusion set through the Hi-att network, enhances the feature factors by feature pooling, and outputs the feature extraction set after the feature factors are enhanced. Among them, x t represents the output representation of the feature extraction set, f is the activation function of the feature extractor, x r represents the input feature fusion set, b x represents the bias vector of the Hi-att network, represents a set of characteristic factors; Obtain a feature extraction set, and the category discriminator judges the state category to which the feature extraction set belongs based on the naive Bayes algorithm, and outputs a data judgment set; Among them, the judgment function of the category discriminator is expressed as: Among them, x p represents the data judgment set, n represents the size of the feature extraction set, P represents the thickness of the judgment function judgment layer, and α represents the category weight coefficient; Among them, α represents the category weight coefficient, α0 is the initial weight constant based on principal component analysis, and g represents the coupling coefficient of the naive Bayes algorithm.

7. The method for real-time monitoring of combustion status of a gas turbine according to claim 6, characterized in that: The method of identifying and analyzing the combustion state parameters by the state monitoring model and outputting a comprehensive evaluation value specifically includes: Obtain a data judgment set, and the classifier calculates the state classification value of the data judgment set based on the attribute attention mechanism; The classifier combines the data classification value and the quantile performance algorithm to calculate the comprehensive evaluation value.

8. The method for real-time monitoring of combustion status of a gas turbine according to claim 7, characterized in that: The method for calculating the state classification value specifically includes: Among them, y p represents the state classification value, score(·) is the attribute attention score function, x m p represents the weight of the mth state level, and k represents the number of state levels; Among them, tanh is the activation function of the classifier, W x is the attention weight matrix, b p Represents the weight bias term.

9. A gas turbine combustion state real-time monitoring system, used to implement the gas turbine combustion state real-time monitoring method according to any one of claims 1 to 8, characterized in that: The gas turbine combustion state real-time monitoring system specifically comprises: A parameter acquisition module is used to obtain the combustion state parameters of the gas turbine and upload the combustion state parameters to the state database, which stores the combustion state parameters as time-series historical state parameters; The model training module is used to traverse the state database, extract modeling samples from the state database, divide the modeling samples into training sets and test sets, build and train the state monitoring model by combining category joint confrontation and gated recurrent units, and output the trained state monitoring model; The comprehensive evaluation module takes the combustion state parameters as input and executes the state monitoring model. The state monitoring model identifies and analyzes the combustion state parameters and outputs a comprehensive evaluation value. The first judgment module is used to judge whether the comprehensive evaluation value exceeds the preset state threshold, and if it exceeds the preset state threshold, trigger the combustion warning instruction; The second judgment module is used to obtain the comprehensive evaluation value, the host group associated with the comprehensive evaluation value, and at least one group of associated units, calculate the risk assessment value of the associated units associated with the host group based on the conditional risk value method, and judge whether the risk assessment value exceeds the preset risk threshold. If it exceeds the preset risk threshold, the combustion risk instruction is triggered.

10. The gas turbine combustion state real-time monitoring system according to claim 9, characterized in that: The comprehensive evaluation module specifically includes: The parameter transformation unit obtains the combustion state parameters. The data fusion layer performs Fourier transformation on the multi-source heterogeneous combustion state parameters and projects the combustion state parameters into the Fourier domain. The parameter fusion unit introduces the combustion state parameters in the Fourier domain into the frequency domain convolution fusion based on the data fusion layer to obtain a feature fusion set; Feature extraction layer, loads the feature fusion set, and the feature extractor extracts the input feature fusion set based on the Hi-att network, indexes the feature factors embedded in the feature fusion set through the Hi-att network, enhances the feature factors by feature pooling, and outputs the feature extraction set after the feature factors are enhanced; The parameter judgment unit is used to obtain a feature extraction set. The category discriminator judges the state category to which the feature extraction set belongs based on the naive Bayes algorithm and outputs a data judgment set. The comprehensive evaluation unit is used to obtain the data judgment set. The classifier calculates the state classification value of the data judgment set based on the attribute attention mechanism. The classifier calculates the comprehensive evaluation value by combining the data classification value and the quantile performance algorithm.

Images