Active flameout prevention control method for aero-retrofit gas turbine

Abstract

An active flameout prevention control method for an aero-reform gas turbine comprises the steps that historical monitoring data are preprocessed, and a combustion dynamical model and a combustion thermodynamic model are established by combining the combustion reaction process and the reaction rate in the gas turbine; the change trend of the reaction rate and the combustion heat release in the gas turbine is used as training features, and a time sequence feature matrix is formed according to the time sequence. Actual flameout data of the gas turbine are obtained, the actual flameout data are marked according to the flameout states of the actual flameout data at different time points, and marked sample data are generated; and constructing a data set, dividing the data set into a training set and a verification set, and training the LSTM deep learning model. And calling a flameout feature capture model to extract combustion features causing the flameout state of the gas turbine in the time sequence according to the current monitoring data, and predicting the flameout probability of the gas turbine based on the combustion features so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds a first threshold value.

Description

Technical Field

[0001] The present invention belongs to the technical field of gas turbine control, and more specifically, relates to an active flameout prevention control method for an aeroderivative gas turbine. Background Art

[0002] Aero-derivative gas turbines are modified aircraft jet engines and equipped with power turbines to make their speed and power meet the needs of land power generation or mechanical traction. The combustion process of gas turbines is highly dependent on the mixture ratio of fuel and air (i.e., air-fuel ratio). Therefore, a suitable air-fuel ratio can ensure the stability of the flame in the combustion chamber and effectively prevent flameout. When the air-fuel ratio is too high (excessive air) or too low (excessive fuel), it may lead to incomplete combustion or flame out.

[0003] At present, the existing aeroderivative gas turbines mainly control flameout by changing their internal structure to allow natural gas intake and supplementary combustion to prevent flameout in the aeroderivative gas turbines. However, this method has a low degree of intelligence and can only perform supplementary combustion or air intake within a specified time or manually. It is difficult to predict when the gas turbine will be at risk of flameout and to actively prevent flameout. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the purpose of the present invention is to solve the above-mentioned defects and further propose an active flameout prevention control method for an aeroderivative gas turbine.

[0005] The present invention adopts the following technical solution.

[0006] A first aspect of the present invention discloses an active flameout prevention control method for an aeroderivative gas turbine, the method comprising:

[0007] Acquire historical monitoring data measured by multiple sensors on the gas turbine, and preprocess the historical monitoring data, wherein the preprocessing includes filtering, denoising and normalization processing;

[0008] Based on the pre-processed historical monitoring data, and in combination with the combustion reaction process and reaction rate in the gas turbine, a combustion kinetic model and a combustion thermodynamic model are established;

[0009] Taking the changing trends of the reaction rate and combustion heat release in the gas turbine as training features, and constructing a time series feature matrix in chronological order based on the training features;

[0010] Acquiring actual flameout data of the gas turbine, and marking the actual flameout data according to flameout states of the actual flameout data at different time points to generate labeled sample data;

[0011] A data set is constructed based on the historical monitoring data, the time series feature matrix, and the labeled sample data, and the data set is divided into a training set and a validation set, and an LSTM deep learning model is trained by the training set and the validation set to obtain a flameout feature capture model;

[0012] The flameout feature capture model is called to extract the combustion features that lead to the flameout state of the gas turbine in the time series according to the current monitoring data, and the flameout probability of the gas turbine is predicted based on the combustion features, so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds a first threshold;

[0013] Among them, the multiple sensors include at least a flow sensor, a temperature sensor and a pressure sensor, and the historical monitoring data includes at least an air-fuel ratio, a combustion chamber pressure, a combustion chamber temperature, a combustion heat release and a reaction rate.

[0014] Furthermore, the reaction equation of the combustion reaction process in the gas turbine is:

[0015] CH4+2O2→CO2+2H2O;

[0016] The reaction rate expression of the combustion reaction process is:

[0017]

[0018] Where A is the frequency factor in the Arrhenius equation, E a is the activation energy in the Arrhenius equation, R is the gas constant, and T is the combustion temperature in K.

[0019] Furthermore, the expression of the combustion thermodynamic model is:

[0020] Q=M f ×(∑(n p ×H p )-∑(n r ×H r ));

[0021] Where, Q is the heat released by combustion, in J / s, N f is the mass flow rate of the fuel, n p With n r represent the molar numbers of products and reactants in the combustion reaction, respectively. p With H r They represent the enthalpy of formation of products and reactants in the combustion reaction, respectively, and the unit is kJ / mol.

[0022] Furthermore, the reaction rate and combustion heat release in the gas turbine are used as training features, and a time series feature matrix is ​​constructed in chronological order based on the training features, including:

[0023] The reaction rate and the combustion heat are sampled in time sequence to obtain the time series characteristic matrix, wherein the expression of the time series characteristic matrix is:

[0024] X t =[V t ,Q t ,F t ];

[0025] Where, X t represents the training features at time point t, Q t is the heat released by combustion at time point t, V t is the reaction rate at time point t, F t is the air-fuel ratio at time t.

[0026] Further, the obtaining of actual flameout data of the gas turbine and marking the actual flameout data according to the flameout status of the actual flameout data at different time points to generate labeled sample data includes:

[0027] Based on the actual flameout data, monitoring data of the gas turbine when it is in a flameout state at any time point is acquired, and the time point in the flameout state and its monitoring data are marked as a first label, and the time point in the non-flameout state and its monitoring data are marked as a second label, to obtain the labeled sample data;

[0028] The label expression of the labeled sample data is:

[0029]

[0030] Where Y t is the label at time point t, 1 is the first label and 0 is the second label.

[0031] Furthermore, the data set is constructed based on the historical monitoring data, the time series feature matrix and the labeled sample data, and the data set is divided into a training set and a validation set, and the LSTM deep learning model is trained by the training set and the validation set to obtain the flameout feature capture model, including:

[0032] The labeled sample data and the time series feature matrix at the same time point are regarded as the same group of data, and a data set is constructed based on the historical monitoring data, the labeled sample data and the time series feature matrix of the same group, so as to divide the data set into a training set and a validation set;

[0033] The time series feature matrix is ​​input into the input layer of the LSTM deep learning model to call the LSTM deep learning model to learn the time series features in the historical monitoring data and capture the dynamic patterns of the reaction rate and combustion heat release in the gas turbine.

[0034] Furthermore, the data set is constructed based on the historical monitoring data, the time series feature matrix and the labeled sample data, and the data set is divided into a training set and a validation set, and the LSTM deep learning model is trained by the training set and the validation set to obtain the flameout feature capture model, and further includes:

[0035] Mapping the model output to a binary classification output through the fully connected layer of the LSTM deep learning model, and inputting the hidden layer output of the LSTM deep learning model into the fully connected layer for linear combination and nonlinear transformation of the training features;

[0036] Among them, the linear combination and nonlinear transformation expressions of the training features are:

[0037] z t =W fc ·h t +b fc ;

[0038] In the formula, z t is the output of the fully connected layer, W fc is the weight matrix of the fully connected layer, h t is the hidden layer output, b fc is the bias vector of the fully connected layer;

[0039] The loss function of the LSTM deep learning model is defined by the binary cross entropy loss function, and the model parameters are optimized by the gradient descent method in the back propagation algorithm to minimize the loss function. The expression of the loss function is:

[0040]

[0041] Where N is the number of samples, Y t is the actual flameout probability at time point t, is the flameout probability at time point t predicted by the model, The value range is [0,1]. is obtained when the Sigmoid activation function is used for the binary classification output.

[0042] The second aspect of the present invention discloses an active flameout prevention control device for an aeroderivative gas turbine, the device comprising:

[0043] A data preprocessing module, used for acquiring historical monitoring data measured by multiple sensors on the gas turbine, and preprocessing the historical monitoring data, wherein the preprocessing includes filtering, denoising and normalization processing;

[0044] A model building module, used to build a combustion kinetic model and a combustion thermodynamic model based on the pre-processed historical monitoring data and in combination with the combustion reaction process and reaction rate in the gas turbine;

[0045] A feature extraction module, used to use the variation trend of the reaction rate and combustion heat release in the gas turbine as training features, and to construct a time series feature matrix in chronological order based on the training features;

[0046] A data labeling module, used for acquiring actual flameout data of the gas turbine, and labeling the actual flameout data according to the flameout status of the actual flameout data at different time points, so as to generate labeled sample data;

[0047] A deep learning model training module is used to construct a data set based on the historical monitoring data, the time series feature matrix and the labeled sample data, and divide the data set into a training set and a validation set, and train the LSTM deep learning model through the training set and the validation set to obtain a flameout feature capture model;

[0048] an anti-stall control module, configured to call the flameout feature capture model to extract the combustion features that lead to the flameout state of the gas turbine in a time series according to current monitoring data, and predict the flameout probability of the gas turbine based on the combustion features, so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds a first threshold;

[0049] Among them, the multiple sensors include at least a flow sensor, a temperature sensor and a pressure sensor, and the historical monitoring data includes at least an air-fuel ratio, a combustion chamber pressure, a combustion chamber temperature, a combustion heat release and a reaction rate.

[0050] The third aspect of the present invention discloses a terminal, including a processor and a storage medium; the storage medium is used to store instructions;

[0051] The processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0052] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described in the first aspect when executed by a processor.

[0053] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages:

[0054] By obtaining the historical monitoring data measured by multiple sensors on the gas turbine and preprocessing the collected historical monitoring data, a combustion kinetic model and a combustion thermodynamic model are established based on the preprocessed historical monitoring data and the combustion reaction process and reaction rate in the gas turbine. Subsequently, the changing trend of the reaction rate and the heat release of combustion in the gas turbine is used as the training feature, and a time series feature matrix is ​​constructed in chronological order based on the training features, and the actual flameout data is marked according to the flameout state of the actual flameout data at different time points to generate labeled sample data. Then, a data set is constructed based on the historical monitoring data, the time series feature matrix, and the labeled sample data, and the data set is divided into a training set and a validation set. The LSTM deep learning model is trained through the training set and the validation set to obtain the flameout feature capture model. Finally, the flameout feature capture model is called to extract the combustion features that cause the flameout state of the gas turbine in the time series according to the current monitoring data, and the flameout probability of the gas turbine is predicted based on the combustion features, so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds the set threshold. This method maintains the stability of the combustion process through real-time monitoring and intelligent control of the air-fuel ratio, thereby actively preventing the gas turbine from stalling, and has a high degree of intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of an active flameout prevention control method for an aeroderivative gas turbine provided by the present invention. DETAILED DESCRIPTION

[0056] The present application is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present application.

[0057] like Figure 1 As shown, in one embodiment, an active flameout prevention control method for an aeroderivative gas turbine includes the following steps:

[0058] Step S110, acquiring historical monitoring data measured by multiple sensors on the gas turbine, and preprocessing the historical monitoring data, wherein the preprocessing includes filtering, denoising and normalization processing.

[0059] Among them, the multiple sensors include at least a flow sensor, a temperature sensor and a pressure sensor, and the historical monitoring data includes at least air-fuel ratio, combustion chamber pressure, combustion chamber temperature, combustion heat release and reaction rate.

[0060] Step S120: establishing a combustion kinetics model and a combustion thermodynamics model based on the preprocessed historical monitoring data and in combination with the combustion reaction process and reaction rate in the gas turbine.

[0061] In some embodiments, the present invention provides an active flameout prevention control method for an aeroderivative gas turbine, wherein the reaction equation of the combustion reaction process in the gas turbine is:

[0062] CH4+2O2→CO2+2H2O.

[0063] Among them, the reaction rate expression of the combustion reaction process is:

[0064]

[0065] Where A is the frequency factor in the Arrhenius equation, E a is the activation energy in the Arrhenius equation, R is the gas constant, and T is the combustion temperature in K.

[0066] It should be noted that the reaction rate describes the speed of the combustion process and is the core dynamic indicator of combustion stability. In flameout prevention control, the reaction rate is directly related to combustion efficiency. This is because the reaction rate indicates the speed at which the fuel reacts with oxygen. Therefore, a higher reaction rate can generate more sufficient heat to maintain the temperature of the combustion chamber, thereby preventing the probability of flameout. However, if the reaction rate is too low, the combustion reaction may not be able to generate enough heat, and the temperature will drop rapidly, resulting in an increased probability of flameout. Therefore, controlling the appropriate reaction rate can ensure that the combustion reaction continues and reduce the probability of flameout.

[0067] Secondly, the reaction rate is affected by the fuel-air ratio and temperature, and usually depends on the fuel-air ratio and combustion temperature. In actual operation, fluctuations in the fuel-air ratio and temperature will cause changes in the reaction rate, so monitoring and controlling the reaction rate can help adjust the supply of fuel and air in time to ensure the stability of the combustion process.

[0068] In addition, the reaction rate is a dynamic parameter that can reflect the rate at which heat is generated during the combustion process. By monitoring the changing trend of the reaction rate in real time, potential flameout risks can be discovered early. For example, when the reaction rate shows a downward trend, the fuel supply or air supply can be automatically adjusted to increase the reaction rate and maintain combustion stability. Therefore, the reaction rate is an important parameter in flameout prevention control. It can dynamically reflect the intensity and stability of the combustion reaction and can be used to determine whether the combustion process in the gas turbine is in a stable state.

[0069] In some embodiments, the present invention provides an active flameout prevention control method for an aeroderivative gas turbine, wherein the expression of the combustion thermodynamic model is:

[0070] Q=M f ×(∑(n p ×H p )-∑(n r ×Hr )).

[0071] Where, Q is the heat released by combustion, unit is H / s, M f is the mass flow rate of the fuel, n p With n r represent the molar numbers of products and reactants in the combustion reaction, respectively. p With H r They represent the enthalpy of formation of products and reactants in the combustion reaction, respectively, and the unit is kJ / mol.

[0072] It should be noted that the heat released by combustion refers to the energy output generated during the combustion process, which directly affects the temperature and stability of the combustion chamber. In flameout prevention control, the heat released by combustion is the energy released by the fuel during the combustion process. If the heat released is insufficient, the temperature of the combustion chamber will drop, and the combustion process may be difficult to maintain, resulting in an increased probability of flameout. Especially when there is a sudden change in load or unstable fuel supply, the fluctuation of heat release may cause unstable combustion and increase the risk of flameout.

[0073] Secondly, the heat release of combustion is affected by the fuel-air ratio and the reaction rate. When the fuel-air ratio is close to the optimal value and the reaction rate is appropriate, the heat release of combustion can usually reach a relatively ideal state, ensuring the stability of the combustion process. Therefore, monitoring the heat release of combustion can help understand whether the combustion process is in an efficient and stable state. In anti-stall control, the heat release can be regarded as an indirect indicator of combustion stability. If the heat release fluctuates greatly, it indicates that the combustion process may be in an unstable state. The heat release of combustion can be increased by adjusting the fuel-air ratio or the reaction rate, so that the combustion chamber maintains sufficient heat, thereby preventing the probability of flameout from increasing.

[0074] Step S130, using the changing trends of the reaction rate and combustion heat release in the gas turbine as training features, and constructing a time series feature matrix in chronological order based on the training features.

[0075] In some embodiments, the present invention provides an active flameout prevention control method for an aero-derivative gas turbine, which uses the change trend of the reaction rate and combustion heat release in the gas turbine as training features, and constructs a time series feature matrix in chronological order based on the training features, specifically including the following steps:

[0076] Step S131, sampling the reaction rate and combustion heat release in time sequence to obtain a time series feature matrix, wherein the expression of the time series feature matrix is:

[0077] X t =[V t ,Q t ,F t ].

[0078] Where, X t represents the training features at time point t, Q t is the heat released by combustion at time point t, V t is the reaction rate at time point t, F t is the air-fuel ratio at time t.

[0079] Step S140, obtaining actual flameout data of the gas turbine, and marking the actual flameout data according to the flameout status of the actual flameout data at different time points to generate labeled sample data.

[0080] In some embodiments, the present invention provides an aeroderivative gas turbine active flameout control method, which obtains actual flameout data of the gas turbine and marks the actual flameout data according to the flameout state of the actual flameout data at different time points to generate labeled sample data, specifically comprising the following steps:

[0081] Step S141, based on the actual flameout data, obtain the monitoring data of the gas turbine when it is in the flameout state at any time point, and mark the time point in the flameout state and its monitoring data as the first label, and mark the time point when it is not in the flameout state and its monitoring data as the second label, so as to obtain labeled sample data.

[0082] Among them, the label expression for annotating sample data is:

[0083]

[0084] Where Y t is the label at time point t, 1 is the first label and 0 is the second label.

[0085] Step S150, constructing a data set based on historical monitoring data, time series feature matrix and labeled sample data, and dividing the data set into a training set and a validation set, training the LSTM deep learning model with the training set and the validation set to obtain a flameout feature capture model.

[0086] In some embodiments, the present invention provides an aero-derivative gas turbine active flameout control method, which constructs a data set based on historical monitoring data, a time series feature matrix, and labeled sample data, and divides the data set into a training set and a validation set. The LSTM deep learning model is trained by the training set and the validation set to obtain a flameout feature capture model, which specifically includes the following steps:

[0087] Step S151, taking the labeled sample data and the time series feature matrix at the same time point as the same group of data, and constructing a data set based on the historical monitoring data, labeled sample data and the time series feature matrix of the same group, so as to divide the data set into a training set and a validation set.

[0088] Step S152, input the time series feature matrix into the input layer of the LSTM deep learning model to call the LSTM deep learning model to learn the time series features in the historical monitoring data and capture the dynamic patterns of the reaction rate and combustion heat release in the gas turbine.

[0089] In some embodiments, the present invention provides an aero-derivative gas turbine active flameout control method, which constructs a data set based on historical monitoring data, a time series feature matrix, and labeled sample data, and divides the data set into a training set and a validation set. The LSTM deep learning model is trained by the training set and the validation set to obtain a flameout feature capture model, which specifically includes the following steps:

[0090] Step S153, mapping the model output to the binary classification output through the fully connected layer of the LSTM deep learning model, and inputting the hidden layer output of the LSTM deep learning model into the fully connected layer for linear combination and nonlinear transformation of the training features.

[0091] Among them, the linear combination and nonlinear transformation expressions of training features are:

[0092] z t =W fc ·h t +b fc .

[0093] In the formula, z t is the output of the fully connected layer, W fc is the weight matrix of the fully connected layer, h t is the hidden layer output, b fc is the bias vector of the fully connected layer.

[0094] Step S154, defining the loss function of the LSTM deep learning model by the binary cross entropy loss function, and optimizing the model parameters by the gradient descent method in the back propagation algorithm to minimize the loss function. The expression of the loss function is:

[0095]

[0096] Where N is the number of samples, Y t is the actual flameout probability at time point t, is the flameout probability at time point t predicted by the model, The value range is [0,1]. It is obtained when the Sigmoid activation function is used for binary classification output.

[0097] Step S160, calling the flameout feature capture model to extract the combustion features that lead to the flameout state of the gas turbine in the time series according to the current monitoring data, and predicting the flameout probability of the gas turbine based on the combustion features, so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds a first threshold.

[0098] It should be noted that the air-fuel ratio (i.e., fuel-air ratio, the mass ratio of fuel to air) is the basis for the stability of the combustion process in the gas turbine. In the gas turbine, the air-fuel ratio directly determines the basis of combustion. Therefore, for flameout prevention control, the air-fuel ratio is the key to its control. When the air-fuel ratio is in the optimal range, the degree of mixing of fuel and oxygen can reach a better level, so that the combustion process can be carried out efficiently, and the heat generated by combustion can be sufficient to maintain the temperature of the combustion chamber in the gas turbine. This is because if the fuel-air ratio is too low (i.e., excess air and insufficient fuel), the combustion reaction will be inhibited, the temperature will decrease, making it difficult to maintain combustion, and the probability of flameout will increase. If the fuel-air ratio is too high (i.e., excess fuel and insufficient air), the combustion will be incomplete, which will lead to low combustion efficiency and decreased reaction rate, and the probability of flameout will increase.

[0099] Secondly, the operating conditions of gas turbines are variable, so different loads and operating conditions have different requirements for fuel and air. By adjusting the fuel-air ratio in real time, better combustion conditions can be maintained under different operating conditions, thereby reducing the probability of flameout. In addition, the gas turbine control system can control the fuel-air ratio in real time by adjusting the fuel flow and air flow to adapt to the changing needs during the combustion process. Compared with other parameters, the fuel-air ratio is a parameter that can be adjusted quickly, which facilitates the rapid response of anti-flameout control. Therefore, the fuel-air ratio is a more core parameter in anti-flameout control, which directly determines the stability of the combustion reaction and has good adjustability.

[0100] The above-mentioned active flameout prevention control method for an aero-derivative gas turbine obtains historical monitoring data measured by multiple sensors on the gas turbine, pre-processes the collected historical monitoring data, and establishes a combustion kinetic model and a combustion thermodynamic model based on the pre-processed historical monitoring data and in combination with the combustion reaction process and reaction rate in the gas turbine. Subsequently, the changing trend of the reaction rate and the combustion heat release in the gas turbine is used as a training feature, and a time series feature matrix is ​​constructed in chronological order based on the training features, and the actual flameout data is marked according to the flameout state of the actual flameout data at different time points to generate labeled sample data. Then, a data set is constructed based on the historical monitoring data, the time series feature matrix, and the labeled sample data, and the data set is divided into a training set and a validation set. The LSTM deep learning model is trained through the training set and the validation set to obtain a flameout feature capture model. Finally, the flameout feature capture model is called to extract the combustion features that cause the flameout state of the gas turbine in the time series according to the current monitoring data, and the flameout probability of the gas turbine is predicted based on the combustion features, so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds the set threshold. This method maintains the stability of the combustion process through real-time monitoring and intelligent control of the air-fuel ratio, thereby actively preventing the gas turbine from stalling, and has a high degree of intelligence.

[0101] In a specific embodiment, the present invention provides an active flameout prevention control method for an aero-derivative gas turbine. The combustion process of the gas turbine is highly dependent on the mixture ratio of fuel and air, that is, the air-fuel ratio. A suitable air-fuel ratio can ensure the flame stability in the combustion chamber and prevent flameout. When the air-fuel ratio is too high (excess air) or too low (excess fuel), it may lead to incomplete combustion or flame out. Therefore, by real-time monitoring and adjustment of the air-fuel ratio, the stability of the combustion process can be maintained and flameout can be actively prevented.

[0102] In this embodiment, steps 1 to 4 are included:

[0103] Step 1: Sensor selection and installation High-precision sensors (such as flow sensors, temperature sensors, etc.) are installed at key locations of the gas turbine to monitor parameters such as fuel-air ratio (air-fuel ratio), combustion chamber pressure, combustion chamber temperature, combustion heat release, reaction rate, etc. in real time.

[0104] Step 2: Historical data collection and preprocessing:

[0105] Specifically, the high-precision sensors installed in step 1 are used to continuously collect monitoring data of the key parameters involved in step 1 in real time, and the collected monitoring data are preprocessed by filtering, denoising and normalization to ensure the accuracy and consistency of the data.

[0106] Step 3: Establish the combustion kinetics model and the combustion thermodynamics model:

[0107] Specifically, the process of establishing the combustion dynamics model is as follows:

[0108] First, the chemical reaction process is clarified. The main reaction equation of combustion is expressed as:

[0109] CH4+2O2→CO2+2H2O.

[0110] This chemical reaction equation is used to describe the reaction of fuel and oxygen to produce carbon dioxide and water.

[0111] Secondly, the reaction rate V equation of the above chemical reaction process is expressed as:

[0112]

[0113] Wherein, A is the frequency factor in the Arrhenius equation. In this embodiment, its value can be 10 7 m 3 / mol·s,E a is the activation energy in the Arrhenius equation, which depends on the chemical reaction conditions and can be 200KJ / mol. R is the gas constant, which is about 8.314J / (mol·K). T is the combustion temperature in K.

[0114] Establishment of combustion thermodynamic model:

[0115] Q=M f ×(∑(n p ×H p )-∑(n r ×H r )).

[0116] Where, Q is the heat released by combustion, in J / s, M f is the mass flow rate of the fuel, n p With n r represent the molar numbers of combustion products and reactants, respectively, H p With H r They represent the enthalpy of formation of the products and reactants of combustion, respectively, and the unit is kJ / mol.

[0117] Step 4, deep learning model training:

[0118] Specifically, it includes steps 41 to 44:

[0119] Step 41, feature extraction:

[0120] The changing trends of reaction rate V and combustion heat release Q are used as training features. To this end, V and Q are sampled by time to form a time series feature matrix as follows:

[0121] X t=[V t ,Q t ,F t ].

[0122] Where, X t represents the training features at time point t, Q t is the heat released by combustion at time point t, V t is the reaction rate at time point t, F t is the air-fuel ratio at time t.

[0123] Step 42, generation of target label:

[0124] Get the actual flameout data labeling sample of the gas turbine. If the parameters after a certain time point t cause flameout, then mark it as flameout state (set to 1) at that time point, otherwise mark it as normal state (set to 0). Form the label Y t :

[0125]

[0126] Among them, [X t ,Y t ] will be treated as a set of data.

[0127] Step 43, data set segmentation:

[0128] The dataset is divided into a training set and a validation set, with a common ratio of 80% for training and 20% for validation. This dataset split helps evaluate the generalization ability of the model.

[0129] In this embodiment, an LSTM (Long Short-Term Memory) network may be selected to process time series data because it is suitable for capturing the time dependency of the gas turbine system.

[0130] The model structure and the functions of each structure include:

[0131] Input layer: receives the time series feature matrix X t .

[0132] LSTM hidden layer: The LSTM hidden layer is the core part of the LSTM network, including the forget gate, input gate, and output gate, which are used to learn time series features and capture the dynamic patterns of reaction rate and combustion heat release.

[0133] Fully connected layer: maps the LSTM output to a binary classification output. The output h of the LSTM hidden layer t , after the fully connected layer, the linear combination and nonlinear transformation of the features are performed, and the calculation formula is:

[0134] z t =W fc ·h t +bfc .

[0135] In the formula, z t is the output of the fully connected layer, W fc is the weight matrix of the fully connected layer, whose parameters are learned from the training data, b fc is the bias vector of the fully connected layer, whose parameters are learned from the training data.

[0136] Output layer: Use the Sigmoid activation function to limit the output to [0,1] for classification (i.e. output ).

[0137]

[0138] In the formula, The value range is [0,1].

[0139] In this embodiment, a binary cross entropy loss function may be used to define the loss of the model to optimize the flameout classification accuracy. The loss function expression is:

[0140]

[0141] Where N is the number of samples, Y t is the actual flameout probability at time point t, is the flameout probability at time point t predicted by the model.

[0142] Step 44, model parameter optimization:

[0143] Use the back-propagation algorithm to optimize the model parameters through the gradient descent method to minimize the loss function. During the training process, you need to set the learning rate α (usually in the range of 0.001 to 0.01) and the batch size (commonly 32 or 64. In deep learning training, usually not all data is used for parameter update. Instead, the mini-batch gradient descent method (Mini-Batch Gradient Descent) is used, that is, the data is divided into small batches, and only one batch of data is used for each update. The commonly used batch size is 32 or 64).

[0144] In this embodiment, the implicit state h t It can help extract patterns that may cause flameout in time series. During the training process, LSTM continuously adjusts the weights so that the implicit state h t The characteristic patterns related to the flameout event are gradually captured, and a characteristic sequence that can reflect the current system state is generated.

[0145] The pattern extraction process includes:

[0146] Generation of input feature sequence: The key parameters of the gas turbine (such as reaction rate and combustion heat release) are input into the LSTM network in time series.

[0147] Accumulation of hidden states: Each time step generates a hidden state h t , which combines the combustion characteristics of the current time step and the previous time step.

[0148] Analysis of implicit states: By analyzing the historical implicit states at the time of flameout, potential flameout risk characteristic patterns are extracted.

[0149] In addition, in order to identify the risk of flameout in actual operation, it is necessary to set a risk threshold based on the trained model. The risk threshold can be set by observing historical flameout data and selecting a probability value as the flameout risk threshold. For example, if the threshold is set to 0.7, it means that if If it exceeds 0.7, the system will consider that the gas turbine is at a high risk of flameout and can control the air-fuel ratio to actively prevent flameout.

[0150] In the flameout prediction model of gas turbines, flameout is usually a low-probability event, which leads to an imbalanced distribution of positive and negative samples in the training data. If the standard loss function is used directly for training, the model may tend to predict "no flameout" to maximize the overall accuracy, thereby ignoring the prediction accuracy of flameout. This situation may cause the model to perform poorly in the detection of low-probability events, making it difficult to accurately identify the risk of flameout.

[0151] To solve this problem, the model can be corrected based on error feedback so that it pays more attention to the prediction of flameout events. The specific method is to calculate the relative error rate and use it as an indicator for correcting the model. The relative error rate can not only reflect the prediction error of the model, but also indirectly reflect the model's ability to capture low-probability events. By incorporating the relative error rate as part of the input features into the model training, the model can more effectively identify flameout patterns, thereby enhancing its attention to flameout events.

[0152] In addition, adding a weighted factor of the relative error rate to the loss function can further enhance the model's sensitivity to flameout events. By increasing the weight of samples with higher error rates, the model will also conduct more in-depth learning of these samples to correct predictions with larger errors, thereby gradually improving the ability to predict low-probability events during the training process. In this way, the loss function can further improve the model's prediction accuracy for flameout events while reducing the risk of false positives and false negatives.

[0153] In this embodiment, the model correction process based on error feedback includes steps A1 to A3:

[0154] Step A1, using the historical data obtained by the sensor, define the error E:

[0155] E=Y′-Y.

[0156] Where Y′ is the actual observed value, Y is the model calculated value, and Y′ or Y can represent the reaction rate V or the combustion heat Q.

[0157] Step A2: Compare the corrected model output with the new experimental data and calculate the relative error rate E r :

[0158]

[0159] In the formula, ε1, ε2, ε3 can take values ​​of 0.3, 0.2 and 0.5 respectively.

[0160] Step A3, re-modify the loss function based on the relative error rate.

[0161]

[0162] An active flameout prevention control device for an aeroderivative gas turbine provided by the present invention is described below. The active flameout prevention control device for an aeroderivative gas turbine described below and the active flameout prevention control method for an aeroderivative gas turbine described above can be referred to each other.

[0163] In one embodiment, an aero-derivative gas turbine active flameout prevention control device includes a data preprocessing module, a model building module, a feature extraction module, a data labeling module, a deep learning model training module, and an flameout prevention control module.

[0164] The data preprocessing module is used to obtain the historical monitoring data measured by multiple sensors on the gas turbine and preprocess the historical monitoring data. The preprocessing includes filtering, denoising and normalization.

[0165] The model building module is used to establish a combustion kinetic model and a combustion thermodynamic model based on the pre-processed historical monitoring data and in combination with the combustion reaction process and reaction rate in the gas turbine.

[0166] The feature extraction module is used to use the changing trends of the reaction rate and combustion heat release in the gas turbine as training features, and to construct a time series feature matrix in chronological order based on the training features.

[0167] The data labeling module is used to obtain the actual flameout data of the gas turbine and label the actual flameout data according to the flameout status of the actual flameout data at different time points to generate labeled sample data.

[0168] The deep learning model training module is used to construct a data set based on historical monitoring data, time series feature matrix and labeled sample data, and divide the data set into a training set and a validation set. The LSTM deep learning model is trained with the training set and the validation set to obtain the flameout feature capture model.

[0169] The anti-stall control module is used to call the flameout feature capture model to extract the combustion features that lead to the flameout state of the gas turbine in the time series according to the current monitoring data, and predict the flameout probability of the gas turbine based on the combustion features, so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds a first threshold.

[0170] Among them, the multiple sensors include at least a flow sensor, a temperature sensor and a pressure sensor, and the historical monitoring data includes at least air-fuel ratio, combustion chamber pressure, combustion chamber temperature, combustion heat release and reaction rate.

[0171] In this embodiment, the present invention provides an aeroderivative gas turbine active flameout control device, and the reaction equation of the combustion reaction process in the gas turbine is:

[0172] CH4+2O2→CO2+2H2O.

[0173] Among them, the reaction rate expression of the combustion reaction process is:

[0174]

[0175] Where A is the frequency factor in the Arrhenius equation, E a is the activation energy in the Arrhenius equation, R is the gas constant, and T is the combustion temperature in K.

[0176] In this embodiment, the present invention provides an aeroderivative gas turbine active flameout control device, and the expression of the combustion thermodynamic model is:

[0177] Q=M f ×(∑(n p ×H p )-∑(n r ×H r )).

[0178] Where, Q is the heat released by combustion, in J / s, M f is the mass flow rate of the fuel, n p With n r represent the molar numbers of products and reactants in the combustion reaction, respectively. p With H r They represent the enthalpy of formation of products and reactants in the combustion reaction, respectively, and the unit is kJ / mol.

[0179] In this embodiment, the present invention provides an aeroderivative gas turbine active flameout control device, and the feature extraction module is specifically used for:

[0180] The reaction rate and combustion heat are sampled in time sequence to obtain a time series characteristic matrix, where the expression of the time series characteristic matrix is:

[0181] Xt=[Vt,Qt,Ft].

[0182] Where Xt represents the training feature at time point t, Qt is the combustion heat release at time point t, Vt is the reaction rate at time point t, and Ft is the air-fuel ratio at time point t.

[0183] In this embodiment, the present invention provides an aero-derivative gas turbine active flameout prevention control device, and the data annotation module is specifically used for:

[0184] Based on the actual flameout data, the monitoring data of the gas turbine when it is in the flameout state at any time point is obtained, and the time point when it is in the flameout state and its monitoring data are marked as the first label, and the time point when it is not in the flameout state and its monitoring data are marked as the second label, so as to obtain labeled sample data.

[0185] Among them, the label expression for annotating sample data is:

[0186]

[0187] Where Yt is the label at time point t, 1 is the first label, and 0 is the second label.

[0188] In this embodiment, the present invention provides an aeroderivative gas turbine active flameout control device, and the deep learning model training module is specifically used for:

[0189] The labeled sample data and the time series feature matrix at the same time point are regarded as the same group of data, and a data set is constructed based on the historical monitoring data, labeled sample data and time series feature matrix of the same group, so as to divide the data set into a training set and a validation set.

[0190] The time series feature matrix is ​​input into the input layer of the LSTM deep learning model to call the LSTM deep learning model to learn the time series features in the historical monitoring data and capture the dynamic patterns of reaction rate and combustion heat release in the gas turbine.

[0191] In this embodiment, the present invention provides an aeroderivative gas turbine active flameout control device, and the deep learning model training module is specifically used for:

[0192] The model output is mapped to the binary classification output through the fully connected layer of the LSTM deep learning model, and the hidden layer output of the LSTM deep learning model is input into the fully connected layer for linear combination and nonlinear transformation of the training features.

[0193] Among them, the linear combination and nonlinear transformation expressions of training features are:

[0194] zt=Wfc·ht+bfc.

[0195] Where zt is the output of the fully connected layer, Wfc is the weight matrix of the fully connected layer, ht is the output of the hidden layer, and bfc is the bias vector of the fully connected layer.

[0196] The loss function of the LSTM deep learning model is defined by the binary cross entropy loss function, and the model parameters are optimized by the gradient descent method in the back propagation algorithm to minimize the loss function. The expression of the loss function is:

[0197]

[0198] Where N is the number of samples, Yt is the actual flameout probability at time point t, is the flameout probability at time point t predicted by the model, The value range is [0,1]. It is obtained when the Sigmoid activation function is used for binary classification output.

[0199] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0200] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0201] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0202] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0203] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0204] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0205] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0206] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0207] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An active flameout prevention control method for an aeroderivative gas turbine, characterized in that: The method comprises: Acquire historical monitoring data measured by multiple sensors on the gas turbine, and preprocess the historical monitoring data, wherein the preprocessing includes filtering, denoising and normalization processing; Based on the pre-processed historical monitoring data, and in combination with the combustion reaction process and reaction rate in the gas turbine, a combustion kinetic model and a combustion thermodynamic model are established; Taking the changing trends of the reaction rate and combustion heat release in the gas turbine as training features, and constructing a time series feature matrix in chronological order based on the training features; Acquiring actual flameout data of the gas turbine, and marking the actual flameout data according to flameout states of the actual flameout data at different time points to generate labeled sample data; A data set is constructed based on the historical monitoring data, the time series feature matrix, and the labeled sample data, and the data set is divided into a training set and a validation set, and an LSTM deep learning model is trained by the training set and the validation set to obtain a flameout feature capture model; The flameout feature capture model is called to extract the combustion features that lead to the flameout state of the gas turbine in the time series according to the current monitoring data, and the flameout probability of the gas turbine is predicted based on the combustion features, so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds a first threshold; Among them, the multiple sensors include at least a flow sensor, a temperature sensor and a pressure sensor, and the historical monitoring data includes at least an air-fuel ratio, a combustion chamber pressure, a combustion chamber temperature, a combustion heat release and a reaction rate.

2. The active flameout prevention control method for an aeroderivative gas turbine according to claim 1, characterized in that: The reaction equation of the combustion reaction process in the gas turbine is: CH4+2O2→CO2+2H2O; The reaction rate expression of the combustion reaction process is: Where A is the frequency factor in the Arrhenius equation, E a is the activation energy in the Arrhenius equation, R is the gas constant, and T is the combustion temperature in K.

3. The active flameout prevention control method for an aeroderivative gas turbine according to claim 2, characterized in that: The expression of the combustion thermodynamic model is: Q=M f ×(∑(n p ×H p )-∑(n r ×H r )); Where, Q is the heat released by combustion, in J / s, M f is the mass flow rate of the fuel, n p With n r represent the molar numbers of products and reactants in the combustion reaction, respectively. p With H r They represent the enthalpy of formation of products and reactants in the combustion reaction, respectively, and the unit is kJ / mol.

4. The active flameout prevention control method for an aeroderivative gas turbine according to claim 3, characterized in that: The reaction rate and combustion heat release change trends in the gas turbine are used as training features, and a time series feature matrix is ​​constructed in chronological order based on the training features, including: The reaction rate and the combustion heat are sampled in time sequence to obtain the time series characteristic matrix, wherein the expression of the time series characteristic matrix is: X t =[V t ,Q t ,F t ]; Where, X t represents the training features at time point t, Q t is the heat released by combustion at time point t, V t is the reaction rate at time point t, F t is the air-fuel ratio at time t.

5. The active flameout prevention control method for an aeroderivative gas turbine according to claim 4, characterized in that: The obtaining of actual flameout data of the gas turbine and marking the actual flameout data according to flameout states of the actual flameout data at different time points to generate labeled sample data includes: Based on the actual flameout data, monitoring data of the gas turbine when it is in a flameout state at any time point is acquired, and the time point in the flameout state and its monitoring data are marked as a first label, and the time point in the non-flameout state and its monitoring data are marked as a second label, to obtain the labeled sample data; The label expression of the labeled sample data is: Where Y t is the label at time point t, 1 is the first label and 0 is the second label.

6. The active flameout prevention control method for an aeroderivative gas turbine according to claim 5, characterized in that: The method constructs a data set based on the historical monitoring data, the time series feature matrix and the labeled sample data, and divides the data set into a training set and a validation set, and trains the LSTM deep learning model through the training set and the validation set to obtain a flameout feature capture model, including: The labeled sample data and the time series feature matrix at the same time point are regarded as the same group of data, and a data set is constructed based on the historical monitoring data, the labeled sample data and the time series feature matrix of the same group, so as to divide the data set into a training set and a validation set; The time series feature matrix is ​​input into the input layer of the LSTM deep learning model to call the LSTM deep learning model to learn the time series features in the historical monitoring data and capture the dynamic patterns of the reaction rate and combustion heat release in the gas turbine.

7. The active flameout prevention control method for an aeroderivative gas turbine according to claim 6, characterized in that: The method further comprises: constructing a data set based on the historical monitoring data, the time series feature matrix and the labeled sample data, dividing the data set into a training set and a validation set, training the LSTM deep learning model through the training set and the validation set to obtain the flameout feature capture model; Mapping the model output to a binary classification output through the fully connected layer of the LSTM deep learning model, and inputting the hidden layer output of the LSTM deep learning model into the fully connected layer for linear combination and nonlinear transformation of the training features; Among them, the linear combination and nonlinear transformation expressions of the training features are: z t =W fc ·h t +b fc ; In the formula, z t is the output of the fully connected layer, W fc is the weight matrix of the fully connected layer, h t is the hidden layer output, b fc is the bias vector of the fully connected layer; The loss function of the LSTM deep learning model is defined by the binary cross entropy loss function, and the model parameters are optimized by the gradient descent method in the back propagation algorithm to minimize the loss function. The expression of the loss function is: In the formula, N is the number of samples, Y is t is the actual flameout probability at time point t, is the flameout probability at time point t predicted by the model, The value range is [0,1]. is obtained when the Sigmoid activation function is used for the binary classification output.

8. An active flameout prevention control device for an aeroderivative gas turbine, characterized in that: The device comprises: A data preprocessing module, used for acquiring historical monitoring data measured by multiple sensors on the gas turbine, and preprocessing the historical monitoring data, wherein the preprocessing includes filtering, denoising and normalization processing; A model building module, used to build a combustion kinetic model and a combustion thermodynamic model based on the pre-processed historical monitoring data and in combination with the combustion reaction process and reaction rate in the gas turbine; A feature extraction module, used to use the variation trend of the reaction rate and combustion heat release in the gas turbine as training features, and to construct a time series feature matrix in chronological order based on the training features; A data labeling module, used for acquiring actual flameout data of the gas turbine, and labeling the actual flameout data according to the flameout status of the actual flameout data at different time points, so as to generate labeled sample data; A deep learning model training module is used to construct a data set based on the historical monitoring data, the time series feature matrix and the labeled sample data, and divide the data set into a training set and a validation set, and train the LSTM deep learning model through the training set and the validation set to obtain a flameout feature capture model; an anti-stall control module, configured to call the flameout feature capture model to extract the combustion features that lead to the flameout state of the gas turbine in a time series according to current monitoring data, and predict the flameout probability of the gas turbine based on the combustion features, so as to actively prevent flameout by controlling the air-fuel ratio when the flameout probability exceeds a first threshold; Among them, the multiple sensors include at least a flow sensor, a temperature sensor and a pressure sensor, and the historical monitoring data includes at least an air-fuel ratio, a combustion chamber pressure, a combustion chamber temperature, a combustion heat release and a reaction rate.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images