KR20230013357A - Apparatus for predicting pressure fluctuation at gas turbine combustor using artificial neural network and method for predicting pressure fluctuation at gas turbine combustor using the same - Google Patents

Abstract

The present invention relates to a pressure perturbation prediction apparatus in a gas turbine combustor using artificial neural network technique. The pressure perturbation prediction apparatus comprises: a nozzle unit; a fluid inlet unit which introduces first fluid and second fluid at preset flow rates into the nozzle unit, respectively; a combustion unit in which a third fluid formed by mixing of the first fluid and the second fluid in the nozzle unit is introduced to form a flame; a characteristic calculation unit which calculates flow information, which is a characteristic of the third fluid, and flame information, which is a characteristic of the flame; and a prediction unit which receives the flow information and the flame information, calculates frequency information of pressure perturbation and intensity information of pressure perturbation based on a deep learning model, and predicts the pressure perturbation based on the frequency information of pressure perturbation and the intensity information of pressure perturbation. According to the present invention, the frequency information and intensity information of pressure perturbation that occurs during a combustion process can be quickly predicted to design a stable combustor. In addition, according to the present invention, the frequency information and intensity information of pressure perturbation can be predicted without performing a combustion test by using flame-centered interpolation according to an equivalence ratio and flow conditions.

Description

Apparatus and method for predicting pressure perturbation in gas turbine combustor using artificial neural network technique

The present invention relates to an apparatus and method for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique, and more particularly, to an artificial neural network predicting frequency information and intensity information of pressure perturbation occurring in a combustion process through a deep learning model. It relates to an apparatus and method for predicting pressure fluctuation in a gas turbine combustor using a neural network technique.

Recently, as the amount of power generation through new and renewable energy increases rapidly, a power generation method using a gas turbine is in the spotlight as a means to compensate for the intermittence of power generation, which is the biggest drawback of new and renewable energy. In addition, since it uses fuels such as LNG and hydrogen, it has the advantage of being environmentally friendly compared to thermal power generation using coal or oil.

The gas turbine burns fuel in a lean premixed state to reduce exhaust gas emissions. In this case, it has the advantage of reducing exhaust gas emissions and the disadvantage of easy occurrence of combustion instability phenomenon in which large pressure fluctuations occur at the same time. When combustion instability occurs in a gas turbine combustor, it can cause a decrease in efficiency and even damage to the combustor. Therefore, determining whether combustion instability is present and predicting pressure fluctuations that can reduce it is essential for stable power plant operation.

In the case of the numerical method of predicting pressure perturbation occurring in the existing combustion process, there is an advantage in that detailed flow and structural characteristics of the flame can be confirmed by directly calculating the 3D flame shape and flow field. A disadvantage is that it is difficult to quickly predict perturbations. In addition, in the case of the low-order network model using the flame transfer function, since it is simplified and calculated in 1D, it has the feature of quickly obtaining analysis results. In order to predict the pressure perturbation of , there is a disadvantage in that the flame transfer function must be measured under each condition.

Therefore, there is a need for an efficient prediction technology capable of predicting pressure fluctuations in a short time under various conditions.

An object of the present invention is to solve the above-mentioned conventional problems, and to obtain frequency information and intensity information of pressure perturbation occurring during the combustion process of a gas turbine combustor through a prediction model learned based on pressure perturbation data under various conditions. It is to provide an apparatus and method for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique that efficiently predicts.

The above object, according to the present invention, the nozzle unit; a fluid inlet unit for introducing a first fluid and a second fluid at predetermined flow rates into the nozzle unit; a combustion unit in which a flame is formed by introducing a third fluid formed by mixing the first fluid and the second fluid in the nozzle unit; a feature calculation unit for calculating flow information, which is a characteristic of the third fluid, and flame information, which is a characteristic of the flame; And receiving the flow information and the flame information and calculating the frequency information of the pressure perturbation and the intensity information of the pressure perturbation based on the deep learning model to calculate the pressure perturbation based on the frequency information of the pressure perturbation and the intensity information of the pressure perturbation It is achieved by a pressure perturbation predictor in a gas turbine combustor using an artificial neural network technique including a prediction unit for predicting.

In addition, the flow information may include equivalence ratio information, which is a ratio of the flow rate of the first fluid to the flow rate of the second fluid, and outlet flow rate information, which is a rate at which the third fluid is discharged from the nozzle unit.

In addition, the flame information may include flame center information corresponding to the center of a self-emitting image of the flame formed in the combustion unit.

In addition, the prediction unit may use an artificial neural network (ANN) model.

In addition, the first fluid is provided with air, the second fluid is provided with hydrocarbon fuel, and the first fluid is heated to a temperature between 190 ° C and 210 ° C and introduced into the nozzle unit.

In addition, the combustion unit may be provided in a form in which the burner under test is reduced at a predetermined ratio.

In addition, the flame center information may be calculated as a position of a maximum intensity value in the self-emission image of the flame.

Also, the self-luminous image may be captured by a CMOS camera coupled with an OH filter or a CH filter.

In addition, the outlet velocity information is calculated by dividing the sum of the predetermined flow rates by the outlet area of the nozzle unit, measured through particle image velocimetry (PIV), or through a hot-wire anemometer. can be measured

Also, the calculation unit may calculate the frequency information of the pressure perturbation and the intensity information of the pressure perturbation based on Equation 1.

{Formula 1}

(The x is the input value, the y is the frequency information of the pressure perturbation or the intensity information of the pressure perturbation, the w is a weight matrix between each layer of the artificial neural network model, and the b is the bias value in each measurement matrix, where f and g are activation functions)

According to the present invention, according to the present invention, the flow information, which is a characteristic of the third fluid formed by mixing the first and second fluids in the nozzle unit, and the frequency information of pressure perturbation according to the flame information, which is a characteristic of the flame formed in the combustion unit In the apparatus for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique in which strength information of and pressure perturbation is learned, the first fluid and the second fluid of predetermined flow rates respectively flow into the nozzle unit by the fluid inlet unit Fluid inlet step to be; a combustion step in which the third fluid formed by mixing the first fluid and the second fluid in the nozzle unit flows into a combustion unit to form the flame; a feature calculation step in which the flow information and the flame information are calculated by a feature calculation unit; And a prediction step of receiving the flow information and the flame information and calculating the frequency information of the pressure perturbation and the intensity information of the pressure perturbation by the prediction unit based on the deep learning model to predict the pressure perturbation artificial neural network technique including It is achieved by a pressure fluctuation prediction method in a gas turbine combustor using

In addition, the flow information may include equivalence ratio information, which is a ratio of the flow rate of the first fluid to the flow rate of the second fluid, and outlet flow rate information, which is a rate at which the third fluid is discharged from the nozzle unit.

In addition, the flame information may include flame center information corresponding to the center of a self-emitting image of the flame formed in the combustion unit.

In addition, the predicting step may use an artificial neural network (ANN) model.

In addition, the first fluid is provided with air, the second fluid is provided with hydrocarbon fuel, and the first fluid is heated to a temperature between 190 ° C and 210 ° C and introduced into the nozzle unit.

In addition, the combustion unit may be provided in a form in which the burner under test is reduced at a predetermined ratio.

In addition, the flame center information may be calculated as a position of a maximum intensity value in the self-emission image of the flame.

Also, the self-luminous image may be captured by a CMOS camera coupled with an OH filter or a CH filter.

In addition, the outlet velocity information is calculated by dividing the sum of the predetermined flow rates by the outlet area of the nozzle unit, measured through particle image velocimetry (PIV), or through a hot-wire anemometer. can be measured

In the calculating step, the frequency information of the pressure perturbation and the intensity information of the pressure perturbation may be calculated based on Equation 1.

{Formula 1}

(The x is the input value, the y is the frequency information of the pressure perturbation or the intensity information of the pressure perturbation, the w is a weight matrix between each layer of the artificial neural network model, and the b is the bias value in each measurement matrix, where f and g are activation functions)

According to the present invention, it is possible to rapidly predict frequency information and intensity information of pressure perturbations generated in the combustion process so that a stable combustor can be designed.

In addition, according to the present invention, it is possible to predict frequency information and intensity information of pressure perturbation without performing a combustion test through flame center interpolation according to equivalence ratio and flow conditions.

On the other hand, the effects of the present invention are not limited to the effects mentioned above, and various effects may be included within a range apparent to those skilled in the art from the contents to be described below.

1 shows a pressure fluctuation prediction apparatus in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention as a whole,
Figure 2 shows flow conditions of the first fluid and the second fluid of the pressure perturbation predictor in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention,
3 shows definitions of outlet flow velocity information and flame center information calculated by the feature calculation unit of the pressure perturbation predictor in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention,
4 is a block diagram of a deep learning model of a prediction unit of a pressure perturbation predictor in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention;
5 is a flowchart of a pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention;
FIG. 6 shows results of comparing frequency information and intensity information of pressure perturbation predicted by a method for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention with experimental values.

Hereinafter, some embodiments of the present invention will be described in detail through exemplary drawings. In adding reference numerals to components of each drawing, it should be noted that the same components have the same numerals as much as possible even if they are displayed on different drawings.

And, in describing the embodiments of the present invention, if it is determined that a detailed description of a related known configuration or function hinders understanding of the embodiments of the present invention, the detailed description will be omitted.

Also, terms such as first, second, A, B, (a), and (b) may be used to describe components of an embodiment of the present invention. These terms are only used to distinguish the component from other components, and the nature, order, or order of the corresponding component is not limited by the term.

Hereinafter, an apparatus for predicting pressure fluctuation in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 shows a pressure fluctuation prediction device in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention as a whole, and FIG. 2 is a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention Figure 3 shows the flow rate conditions of the first fluid and the second fluid of the pressure perturbation predicting device in , Figure 3 is a pressure perturbation predicting device in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention Characteristic calculation Figure 4 is a deep learning model configuration of the prediction unit of the pressure perturbation predictor in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention. It shows the figure.

As shown in FIG. 1, the apparatus 100 for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention includes a nozzle unit 110, a fluid inlet unit 120, and a combustion unit ( 130), a feature calculation unit 140, and a prediction unit 150.

The above-described feature calculation unit 140 and prediction unit 150 may be implemented as a single terminal such as a PC (Personal Computer), or may be provided as separate devices depending on functions and purposes.

The nozzle unit 110 has a configuration in which the first fluid f1 and the second fluid f2 are mixed, and may be provided as a nozzle of a coaxial type or a jet-in-cross type, and the fluid is sprayed in a jet or swirl form. can In addition, the nozzle unit 110 of the apparatus 100 for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention may be provided as a partial pre-mixed nozzle. In addition, the nozzle unit 110 of the apparatus 100 for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention may be provided as a swirl nozzle of GE's GE7EA gas turbine combustor. However, it is not necessarily limited to the nozzle of the above-described shape, and may be provided in any shape as long as the above-described third fluid (f3) can be formed. In addition, the nozzle unit 110 may flow the first fluid f1 and the second fluid f2 into the combustion unit 130 to be described later without mixing them.

In addition, the nozzle unit 110 mixes the first fluid f1 and the second fluid f2 introduced from the fluid inlet 120 to be described later to form a third fluid f3, and the formed third fluid ( f3) is introduced into the combustion unit 130 to be described later.

The fluid inlet 120 is configured to introduce the first fluid f1 and the second fluid f2 at predetermined flow rates into the nozzle unit 110, respectively, and may be provided as a mass flow controller (MFC). there is.

Here, according to the fluid inlet 120 provided as a mass flow meter, the above-described first fluid f1 and the above-described second fluid f2 can be supplied to the nozzle unit 110 at a constant flow rate, so that the equivalent ratio information described later and the outlet flow rate information to be described later can be kept constant.

In addition, the pressure at the rear of the above-described fluid inlet 120 is set to normal pressure to 3 atmospheric pressure in consideration of the effect of increasing the pressure inside the combustion unit 130 described later by combustion, and the front pressure is set higher than the above-mentioned rear pressure. So that the fluid supply is made smoothly, it is preferable to set the pressure to 4 to 6 atm in order to prevent the fluid inlet 120 from being damaged due to excessive high pressure. Here, a pressure regulator may be provided in front of the mass flow meter to set the front pressure of the fluid inlet 120 to 4 to 6 atm.

The fluid inlet 120 may be used for the first fluid f1 and the second fluid f2, respectively, and may deliver a predetermined flow rate to the nozzle unit 110 described above. In addition, the fluid inlet unit 120 may transmit the flow rate values of the first fluid f1 and the second fluid f2 to the feature calculation unit 140 to be described later, and the feature calculation unit 140 to be described below may transmit the first fluid f1 and the second fluid f2. The equivalent ratio information can be calculated through the flow rate ratio of the fluid f1 and the second fluid f2.

Here, the first fluid f1 may be provided with air, and the second fluid f2 may be provided with hydrocarbon fuel. More preferably, the second fluid f2 may be prepared with methane.

In addition, the first fluid (f1) may be introduced after being heated to a temperature between 190 ° C and 210 ° C before flowing into the nozzle unit 110 to simulate the high-temperature environment of a commercial gas turbine combustor, more preferably 200 It can be introduced by heating to a temperature of °C.

As shown in FIG. 2, in order to learn the deep learning model of the predictor 150 of the apparatus 100 for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention, a fluid inlet unit 120 may set flow conditions of the first fluid f1 and the second fluid f2 flowing into the nozzle unit 110 to be described later in various ways. Here, slpm is an abbreviation of standard liter per minute and means the flow rate discharged per minute at standard 0℃ and 1 atmospheric pressure.

The combustion unit 130 has a configuration in which a flame is formed, and the third fluid f3 is introduced from the nozzle unit 110 described above.

Here, the combustion unit 130 is provided in a form in which the burner under test is reduced at a predetermined ratio. Here, the combustor to be tested may be GE7EA, a gas turbine combustor of GE, and a preset ratio may be provided at a ratio of 1/3. However, it is not necessarily limited to the above-described combustor, and may be provided in any shape as long as the above-described flame can be formed.

According to this, there is an advantage in that the characteristics of the combustor under test can be identified through the Lab-scale combustor, so that the experiment can be performed safely with a small flow rate.

Here, the combustion unit 130 may be divided into a visible combustion module capable of visualizing flame and a duct-type combustion module unable to visualize flame. Here, the visualization combustion module is provided with a transparent quartz window so that the inside of the combustor can be seen, and the duct type combustion module is provided with all surfaces made of steel so that the interior cannot be seen. In addition, the visualization combustion module may cool the quartz window by injecting high-pressure air from the outside of the visualization combustion module in order to prevent the quartz window from being broken by heat.

In addition, the duct-type combustion module can prevent a temperature increase through water cooling by forming a water jacket through a double wall structure outside the duct-type combustion module in order to prevent deformation or damage caused by heat. In addition, a plurality of ports capable of measuring temperature or pressure may be formed on a side surface of the duct-type combustion module.

Since a general gas turbine combustor has a circular cross section, it is preferable that both the above-described visible combustion module and the duct type combustion module have a circular cross section. However, it is common to use a laser and a camera to visualize the flame, but when the visualization combustion module has a circular cross section, measurement is difficult due to refraction of light. Therefore, the visible combustion module can be designed to have a square cross section.

When the visualization combustion module has a square cross-section and the duct-type combustion module has a circular cross-section, the acoustic characteristics inside the combustion unit 130 may be changed due to a rapid change in cross-sectional area, according to an embodiment of the present invention. Between the visualization combustion module and the duct-type combustion module of the pressure perturbation prediction device 100 in a gas turbine combustor using an artificial neural network technique, one side cross section is circular and the other side cross section is square, and the inside has one side and the other side so that there is no rapid cross-sectional change. A connection module provided in the form of a loft on the side may be provided.

A plug nozzle may be provided at the rear end of the combustion unit 130 to partially shield the exit area of the combustion unit 130 to form an acoustically closed boundary, and the plug nozzle covers the area of the combustion unit 130 by 80% to 80%. 95% can be shielded. More preferably, the plug nozzle may shield the outlet area of the combustion unit 130 so that the maximum speed at the exit of the combustion unit 130 is Mach 0.1 or higher. In addition, in order to change the longitudinal resonant frequency inside the combustion unit 130, the aforementioned plug nozzle may be moved in the axial direction of the combustion unit 130 by the traverse motor.

A cooling water passage may be provided inside the aforementioned plug nozzle to prevent the plug nozzle from being deformed or damaged by heat.

An exhaust gas discharge unit through which exhaust gas is discharged to the outside may be formed at a rear end of the combustion unit 130, and a silencer may be provided in the exhaust gas discharge unit to reduce noise caused by exhaust gas discharge.

In addition, the above-described combustion unit 130 may have a circular cross section.

The feature calculator 140 is a component that calculates flow information, which is a characteristic of the above-described third fluid (f3), and flame information, which is a characteristic of the above-described flame.

Here, the flow information is equivalent ratio information, which is the ratio of the flow rate of the first fluid f1 and the flow rate of the second fluid f2 described above, and the third fluid f3 is discharged from the nozzle unit 110 described above. It may include outlet flow rate information, which is the speed at which the However, it is not necessarily limited thereto, and it may be provided with any values as long as the characteristics of the above-described third fluid (f3) can be clearly represented.

Here, the flame information may include flame center information that is the center of the above-described flame formed in the above-described combustion unit 130 . However, it is not necessarily limited thereto, and it may be provided with any value as long as it can clearly represent the characteristics of the above-described flame.

Here, the equivalence ratio information is calculated as the ratio of the flow rates of the second fluid f2 and the first fluid f1, and the high equivalence ratio information means that the flow rate ratio of the fuel is high. The flow rates of the first fluid f1 and the second fluid f2 may be transmitted from the fluid inlet 120 to the feature calculator 140 or may be directly input by the user.

The outlet flow rate information means the flow rate at the outlet of the nozzle unit 110, and is calculated by dividing the flow rate of the third fluid f3 by the outlet area of the nozzle unit 110, or is calculated by particle image velocimetry (PIV). ) or a hot-wire anemometer.

Here, the flow rate of the third fluid f3 may be calculated by adding the flow rate of the first fluid f1 and the flow rate of the second fluid f2.

In addition, when PIV is used, after mixing the particles with at least one of the first fluid (f1) or the second fluid (f2), the double pulse laser generates two laser pulses at very short intervals to determine the movement distance of the particles. Thus, the outlet velocity information can be measured. Here, Al2O3, TiO2, or ZrO2 may be used as the fine particles, and fine particles having a diameter of 0.5 um to 2 um may be used. In addition, the double pulse laser may be provided as a laser that emits double pulses at 10 Hz to 10 kHz. At this time, the laser may be provided as a Nd:YAG laser.

In addition, when using a hot wire velocimetry, the hot wire velocimetry should be installed at the outlet of the nozzle unit 110, and in this case, since the outlet velocity information cannot be measured in the state where the flame is formed, the third fluid (f3) in the state where the flame is not formed Outlet flow velocity information can be measured through the velocity of the flow when it is injected into the combustion unit 130 through the nozzle unit 110 only.

As shown in FIG. 3, the flame center information may be calculated as a position of the maximum intensity value or a weighted average position of the intensity in the image of the flame. Here, the image of the flame may be a direct image of the flame captured by a camcorder or a DSLR, or may be a self-luminous image.

Here, the self-luminous image may be captured by a CMOS camera coupled with an OH filter or a CH filter. When a flame is formed, an OH or CH signal is generated in the flame, and accordingly, the OH or CH signal indicates a location where a chemical reaction takes place. Therefore, the position of the maximum intensity of the OH or CH signal means the position where the chemical reaction is most active, and thus means the center of the flame.

Here, the CMOS camera can measure the image of the flame at high speed, and the CMOS camera can be used to take a self-luminous image of the flame at 10 kHz.

The above-described flame information may be used by interpolating flame information according to the above-described flow information.

Therefore, the above-described center of the flame may be derived through the self-luminous image of the flame, but by interpolating the position of the flame center information according to the above-described equivalence ratio information and the above-mentioned exit velocity information, the flame center information can be obtained without performing an actual combustion experiment. location can be predicted. When frequency information and intensity information of pressure perturbation are predicted by the prediction unit 150 to be described later using the position of the predicted flame center information, frequency information and intensity information of pressure perturbation can be obtained without performing an actual combustion experiment It is possible to derive a combustor design plan that can predict and avoid combustion instability.

The prediction unit 150 receives the above-described flow information and the above-described flame information and calculates frequency information of pressure perturbation and intensity information of pressure perturbation based on a deep learning model to predict pressure perturbation. Artificial neural network (Artificial neural network (ANN) model.

In addition, the above-described prediction unit 150 may calculate frequency information of the pressure perturbation and intensity information of the pressure perturbation based on Equation 1.

{Formula 1}

(The above-mentioned x is the input value, the above-mentioned y is the above-mentioned frequency information of the pressure perturbation or the above-mentioned intensity information of the pressure perturbation, the above-mentioned w is the weight matrix between each layer of the above-mentioned artificial neural network model, and the above-mentioned b is The above-mentioned deflection value matrix in each measurement, the above-mentioned f and the above-mentioned g are activation functions)

Here, the intensity information of pressure perturbation described above may be intensity information of pressure perturbation generated inside the combustion unit 130 during a combustion process.

As shown in FIG. 4, the prediction unit 150 of the apparatus 100 for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention provides equivalence ratio information, flame center information, and exit flow velocity information It receives the input and outputs the frequency information of the pressure perturbation and the intensity information of the pressure perturbation.

Here, since the prediction unit 150 uses an artificial neural network model, intermediate layer parameters are derived using equivalence ratio information, flame center information, and exit flow velocity information, which are input values, and output values are derived through the derived intermediate layer parameters. In the process of deriving the middle layer parameters, the accuracy of prediction can be improved by putting a weight on each input value.

In addition, the apparatus 100 for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention receives equivalence ratio information, flame center information, and exit flow velocity information as input values, and uses an artificial neural network model to generate an output value Since the frequency information and intensity information of phosphorus pressure perturbation are predicted, it is possible to predict the frequency information and intensity information of pressure perturbation in a shorter time than the existing numerical analysis method through 3D flame shape and flow field, and the shape and flow of the actual flame Because it uses conditions and deep learning models, it has the advantage of higher prediction accuracy compared to 1D low-order network models.

The artificial neural network technique according to an embodiment of the present invention including the nozzle unit 110, the fluid inlet unit 120, the combustion unit 130, the feature calculation unit 140, and the prediction unit 150 as described above According to the apparatus 100 for predicting pressure perturbation in the used gas turbine combustor, it is possible to quickly predict frequency information and intensity information of pressure perturbation generated in the combustion process so that a stable combustor can be designed. In addition, it is possible to predict frequency information and intensity information of pressure perturbation without performing a combustion test through flame center interpolation according to equivalence ratio and flow conditions.

Hereinafter, a pressure fluctuation prediction method in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

5 is a flowchart of a pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention, and FIG. 6 is a flow chart of a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention. It shows the results of comparing the frequency information and intensity information of the pressure perturbation predicted by the pressure perturbation prediction method with the experimental values.

As shown in FIG. 5, the method for predicting pressure perturbation in a gas turbine combustor (S100) using an artificial neural network technique according to an embodiment of the present invention includes a fluid introduction step (S110), a combustion step (S120), a feature calculation step (S130), including a prediction step (S140).

The fluid inlet step (S110) is a configuration in which the first fluid (f1) and the second fluid (f2) of a predetermined flow rate are respectively introduced into the nozzle unit 110 by the fluid inlet 120, where the fluid inlet ( 120) may be provided as a mass flow controller (MFC).

Here, the nozzle unit 110 may be provided as a nozzle of a coaxial type or a jet-in-cross method in which the first fluid f1 and the second fluid f2 are mixed, and the fluid is in a jet or swirl form. can be sprayed. In addition, the nozzle unit 110 of the method for predicting pressure perturbation in a gas turbine combustor (S100) using an artificial neural network technique according to an embodiment of the present invention may be provided as a partial pre-mixed nozzle. In addition, the nozzle unit 110 of the pressure perturbation prediction method (S100) in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention may be provided as a swirl nozzle of GE's GE7EA gas turbine combustor. However, it is not necessarily limited to the nozzle of the above-described shape, and may be provided in any shape as long as the above-described third fluid (f3) can be formed. In addition, the nozzle unit 110 may flow the first fluid f1 and the second fluid f2 into the combustion unit 130 to be described later without mixing them.

In addition, the nozzle unit 110 mixes the first fluid (f1) and the second fluid (f2) introduced from the fluid introduction step (S110) to form a third fluid (f3), and the formed third fluid (f3) is introduced into the combustion unit 130.

Here, according to the fluid introduction step (S110) using a mass flow meter, the above-described first fluid (f1) and the above-described second fluid (f2) can be supplied to the combustion unit 130 at a constant flow rate, so that equivalent ratio information to be described later and the outlet flow rate information to be described later can be kept constant.

In addition, the pressure at the rear of the above-described fluid inlet 120 is set to normal pressure to 3 atmospheric pressure in consideration of the effect of increasing the pressure inside the combustion unit 130 described later by combustion, and the front pressure is set higher than the above-mentioned rear pressure. So that the fluid supply is made smoothly, it is preferable to set the pressure to 4 to 6 atm in order to prevent the fluid inlet 120 from being damaged due to excessive high pressure. Here, a pressure regulator may be provided in front of the mass flow meter to set the front pressure of the fluid inlet 120 to 4 to 6 atm.

In the fluid inlet step (S110), the fluid inlet 120 may be used for the first fluid f1 and the second fluid f2, respectively, and may deliver a predetermined flow rate to the combustion unit 130 to be described later. In addition, the fluid introduction step (S110) may transfer the flow rate values of the first fluid (f1) and the second fluid (f2) to the feature calculation step (S130) to be described later, and the feature calculation step (S130 to be described later) The equivalent ratio information can be calculated through the flow rate ratio of the fluid f1 and the second fluid f2.

Here, the first fluid f1 may be provided with air, and the second fluid f2 may be provided with hydrocarbon fuel. More preferably, the second fluid f2 may be prepared with methane.

In addition, the first fluid (f1) may be introduced after being heated to a temperature between 190 ° C and 210 ° C before flowing into the nozzle unit 110 to simulate the high-temperature environment of a commercial gas turbine combustor, more preferably 200 It can be introduced by heating to a temperature of °C.

As shown in FIG. 2, fluid introduction step to learn the deep learning model of the prediction step (S140) of the pressure perturbation prediction method (S100) in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention In (S110), flow rate conditions of the first fluid f1 and the second fluid f2 flowing into the nozzle unit 110 to be described later may be set in various ways. Here, slpm is an abbreviation of standard liter per minute and means the flow rate discharged per minute at standard 0℃ and 1 atmospheric pressure.

In the combustion step (S120), a flame is formed in the combustion unit 130, and the third fluid f1 and the second fluid f2 described above are mixed by the nozzle unit 110 described above. The fluid f3 flows into the combustion unit 130 to form a flame.

Here, the combustion unit 130 is provided in a form in which the burner under test is reduced at a predetermined ratio. Here, the combustor to be tested may be GE7EA, a gas turbine combustor of GE, and a preset ratio may be provided at a ratio of 1/3. However, it is not necessarily limited to the above-described combustor, and may be provided in any shape as long as the above-described flame can be formed.

According to this, there is an advantage in that the characteristics of the combustor under test can be identified through the Lab-scale combustor, so that the experiment can be performed safely with a small flow rate.

Here, the combustion unit 130 may be divided into a visible combustion module capable of visualizing flame and a duct-type combustion module unable to visualize flame. Here, the visualization combustion module is provided with a transparent quartz window so that the inside of the combustor can be seen, and the duct type combustion module is provided with all surfaces made of steel so that the interior cannot be seen. In addition, the visualization combustion module may cool the quartz window by injecting high-pressure air from the outside of the combustion unit 130 in order to prevent the quartz window from being broken by heat.

In addition, the duct-type combustion module can prevent a temperature increase through water cooling by forming a water jacket through a double wall structure outside the combustion unit 130 in order to prevent deformation or damage due to heat. In addition, a plurality of ports capable of measuring temperature or pressure may be formed on a side surface of the duct-type combustion module.

Since a general gas turbine combustor has a circular cross section, it is preferable that both the above-described visible combustion module and the duct type combustion module have a circular cross section. However, it is common to use a laser and a camera to visualize the flame, but when the visualization combustion module has a circular cross section, measurement is difficult due to refraction of light. Therefore, the visible combustion module can be designed to have a square cross section.

When the visualization combustion module has a square cross-section and the duct-type combustion module has a circular cross-section, the acoustic characteristics inside the combustion unit 130 may be changed due to a rapid change in cross-sectional area, according to an embodiment of the present invention. Visualization of Pressure Perturbation Prediction Method (S100) in Gas Turbine Combustor Using Artificial Neural Network Technique Between the combustion module and the duct-type combustion module, one side cross section is circular and the other side cross section is square, and the inside has one side and the other side so that there is no rapid cross-sectional change. A connection module provided in the form of a loft on the side may be provided.

A plug nozzle may be provided at the rear end of the combustion unit 130 to partially shield the exit area of the combustion unit 130 to form an acoustically closed boundary, and the plug nozzle covers the area of the combustion unit 130 by 80% to 80%. 95% can be shielded. More preferably, the plug nozzle may shield the outlet area of the combustion unit 130 so that the maximum speed at the exit of the combustion unit 130 is Mach 0.1 or higher. In addition, in order to change the longitudinal resonant frequency inside the combustion unit 130, the aforementioned plug nozzle may be moved in the axial direction of the combustion unit 130 by the traverse motor.

A cooling water passage may be provided inside the aforementioned plug nozzle to prevent the plug nozzle from being deformed or damaged by heat.

An exhaust gas discharge unit through which exhaust gas is discharged to the outside may be formed at a rear end of the combustion unit 130, and a silencer may be provided in the exhaust gas discharge unit to reduce noise caused by exhaust gas discharge.

In addition, the above-described combustion unit 130 may have a circular cross section.

The feature calculation step (S130) is a step in which flow information, which is a feature of the above-described third fluid (f3), and flame information, which is a feature of the above-described flame, are calculated by the feature calculator 140.

Here, the flow information is equivalent ratio information, which is the ratio of the flow rate of the first fluid f1 and the flow rate of the second fluid f2 described above, and the third fluid f3 is discharged from the nozzle unit 110 described above. It may include outlet flow rate information, which is the speed at which the However, it is not necessarily limited thereto, and it may be provided with any values as long as the characteristics of the above-described third fluid (f3) can be clearly represented.

Here, the flame information may include flame center information that is the center of the above-described flame formed in the above-described combustion unit 130 . However, it is not necessarily limited thereto, and it may be provided with any value as long as it can clearly represent the characteristics of the above-described flame.

Here, the equivalence ratio information is calculated as the ratio of the flow rates of the second fluid f2 and the first fluid f1, and the high equivalence ratio information means that the flow rate ratio of the fuel is high. The flow rates of the first fluid f1 and the second fluid f2 may be transferred from the above-described fluid introduction step S110 to the feature calculation step S130 or may be directly input by the user.

The outlet flow rate information means the flow rate at the outlet of the nozzle unit 110, and is calculated by dividing the flow rate of the third fluid f3 by the outlet area of the nozzle unit 110, or is calculated by particle image velocimetry (PIV). ) or a hot-wire anemometer.

Here, the flow rate of the third fluid f3 may be calculated by adding the flow rate of the first fluid f1 and the flow rate of the second fluid f2.

In addition, when PIV is used, after mixing the particles with at least one of the first fluid (f1) or the second fluid (f2), the double pulse laser generates two laser pulses at very short intervals to determine the movement distance of the particles. Thus, the outlet velocity information can be measured. Here, Al2O3, TiO2, or ZrO2 may be used as the fine particles, and fine particles having a diameter of 0.5 um to 2 um may be used. In addition, the double pulse laser may be provided as a laser that emits double pulses at 10 Hz to 10 kHz. At this time, the laser may be provided as a Nd:YAG laser.

In addition, when using a hot wire velocimetry, the hot wire velocimetry should be installed at the outlet of the nozzle unit 110, and in this case, since the outlet velocity information cannot be measured in the state where the flame is formed, the third fluid (f3) in the state where the flame is not formed Outlet flow velocity information can be measured through the velocity of the flow when it is injected into the combustion unit 130 through the nozzle unit 110 only.

As shown in FIG. 3, the flame center information may be calculated as a position of the maximum intensity value or a weighted average position of the intensity in the image of the flame. Here, the image of the flame may be a direct image of the flame captured by a camcorder or a DSLR, or may be a self-luminous image.

Here, the self-luminous image may be captured by a CMOS camera coupled with an OH filter or a CH filter. When a flame is formed, an OH or CH signal is generated in the flame, and accordingly, the OH or CH signal indicates a location where a chemical reaction takes place. Therefore, the position of the maximum intensity of the OH or CH signal means the position where the chemical reaction is most active, and thus means the center of the flame.

Here, the CMOS camera can measure the image of the flame at high speed, and the CMOS camera can be used to take a self-luminous image of the flame at 10 kHz.

The above-described flame information may be used by interpolating flame information according to the above-described flow information.

Therefore, the above-described center of the flame may be derived through the self-luminous image of the flame, but by interpolating the position of the flame center information according to the above-described equivalence ratio information and the above-mentioned exit velocity information, the flame center information can be obtained without performing an actual combustion experiment. location can be predicted. When the frequency information and the intensity information of the pressure perturbation are predicted by the prediction step (S140) described later using the position of the predicted flame center information, the frequency information and the intensity information of the pressure perturbation can be obtained without performing an actual combustion experiment It is possible to derive a combustor design plan that can predict and avoid combustion instability.

In the prediction step (S140), the prediction unit 150 receives the above-described flow information and the above-described flame information, calculates the frequency information of the pressure perturbation and the intensity information of the pressure perturbation based on the deep learning model to predict the pressure perturbation As a result, an artificial neural network (ANN) model is used.

Also, in the above-described predicting step (S140), the frequency information of the pressure perturbation and the intensity information of the pressure perturbation may be calculated based on Equation 1.

{Formula 1}

(The above-mentioned x is the input value, the above-mentioned y is the above-mentioned frequency information of the pressure perturbation or the above-mentioned intensity information of the pressure perturbation, the above-mentioned w is the weight matrix between each layer of the above-mentioned artificial neural network model, and the above-mentioned b is The above-mentioned deflection value matrix in each measurement, the above-mentioned f and the above-mentioned g are activation functions)

Here, the intensity information of pressure perturbation described above may be intensity information of pressure perturbation generated inside the combustion unit 130 during a combustion process.

As shown in FIG. 4, the prediction step (S140) of the pressure perturbation prediction method (S100) in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention is equivalence ratio information, flame center information, and exit flow velocity information It receives the input and outputs the frequency information of the pressure perturbation and the intensity information of the pressure perturbation.

Here, since the prediction step (S140) uses an artificial neural network model, intermediate layer parameters are derived using equivalence ratio information, flame center information, and exit flow velocity information, which are input values, and output values are derived through the derived intermediate layer parameters. Here, in the process of deriving the middle layer parameters, the accuracy of prediction can be improved by putting a weight on each input value.

In addition, the pressure perturbation prediction method (S100) in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention receives equivalence ratio information, flame center information, and exit flow velocity information as input values, and uses an artificial neural network model to generate an output value Since the frequency information and intensity information of phosphorus pressure perturbation are predicted, it is possible to predict the frequency information and intensity information of pressure perturbation in a shorter time than the existing numerical analysis method through 3D flame shape and flow field, and the shape and flow of the actual flame Because it uses conditions and deep learning models, it has the advantage of higher prediction accuracy compared to 1D low-order network models.

The artificial neural network model of the prediction step (S140) of the pressure perturbation prediction method (S100) in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention may be learned with 190 data samples. 6 shows frequency information of pressure perturbation according to 34 conditions predicted through a pressure perturbation prediction method (S100) in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention learned with 190 data samples, and It shows the intensity information result of pressure perturbation.

As shown in FIG. 6 (a), the pressure perturbation prediction method (S100) in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention predicts frequency information of pressure perturbation for 34 data samples As a result, the R-square of the predicted value and the actual value of the frequency information of the pressure perturbation is 0.9971, and very accurate prediction is performed.

As shown in FIG. 6 (a), the pressure perturbation prediction method (S100) in a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention predicts intensity information of pressure perturbation for 34 data samples As a result, the R-square of the predicted value and the actual value of the intensity information of pressure perturbation is 0.9204, and very accurate prediction is performed.

In a gas turbine combustor using an artificial neural network technique according to an embodiment of the present invention including the above-described fluid introduction step (S110), combustion step (S120), feature calculation step (S130), and prediction step (S140) According to the pressure perturbation prediction method ( S100 ), frequency information and intensity information of pressure perturbation can be predicted so that a stable combustor can be designed. In addition, it is possible to predict frequency information and intensity information of pressure perturbation without performing a combustion test through flame center interpolation according to equivalence ratio information and flow conditions.

In the above, even though all the components constituting the embodiment of the present invention have been described as being combined or operated as one, the present invention is not necessarily limited to these embodiments. That is, within the scope of the object of the present invention, all of the components may be selectively combined with one or more to operate.

In addition, terms such as "comprise", "comprise" or "having" described above mean that the corresponding component may be present unless otherwise stated, and thus exclude other components. It should be construed as being able to further include other components. All terms, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs, unless defined otherwise. Commonly used terms, such as terms defined in a dictionary, should be interpreted as being consistent with the contextual meaning of the related art, and unless explicitly defined in the present invention, they are not interpreted in an ideal or excessively formal meaning.

In addition, the above description is merely an example of the technical idea of the present invention, and various modifications and variations can be made to those skilled in the art without departing from the essential characteristics of the present invention.

Therefore, the embodiments disclosed in the present invention are not intended to limit the technical spirit of the present invention, but to explain, and the scope of the technical spirit of the present invention is not limited by these embodiments. The protection scope of the present invention should be construed according to the claims below, and all technical ideas within the equivalent range should be construed as being included in the scope of the present invention.

100: Pressure perturbation prediction device in gas turbine combustor using artificial neural network technique according to an embodiment of the present invention
110: nozzle part
120: fluid inlet
130: combustion unit
140: feature calculation unit
150: prediction unit
S100: Pressure perturbation prediction method in gas turbine combustor using artificial neural network technique according to an embodiment of the present invention
S110: fluid inflow step
S120: Combustion step
S130: feature calculation step
S140: prediction step
f1: first fluid
f2: Second fluid
f3: 3rd fluid

Claims (20)

nozzle unit;
a fluid inlet unit for introducing a first fluid and a second fluid at predetermined flow rates into the nozzle unit;
a combustion unit in which a flame is formed by introducing a third fluid formed by mixing the first fluid and the second fluid in the nozzle unit;
a feature calculation unit for calculating flow information, which is a characteristic of the third fluid, and flame information, which is a characteristic of the flame; and
Predict the pressure perturbation based on the frequency information of the pressure perturbation and the intensity information of the pressure perturbation by calculating the frequency information of the pressure perturbation and the intensity information of the pressure perturbation based on the deep learning model by receiving the flow information and the flame information An apparatus for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique comprising a prediction unit to do.
The method of claim 1,
The flow information,
Gas turbine using an artificial neural network technique, characterized in that it includes equivalence ratio information, which is the ratio of the flow rate of the first fluid to the flow rate of the second fluid, and outlet flow rate information, which is the rate at which the third fluid is discharged from the nozzle unit A device for predicting pressure fluctuations in combustors.
The method of claim 1,
The flame information,
A pressure perturbation predictor in a gas turbine combustor using an artificial neural network technique, characterized in that it includes flame center information corresponding to the center of the self-luminous image of the flame formed in the combustion unit.
The method of claim 1,
The prediction unit,
An apparatus for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique, characterized in that using an artificial neural network (ANN) model.
The method of claim 1,
The first fluid,
provided by air,
The second fluid,
It is prepared with hydrocarbon fuel,
The first fluid,
A pressure perturbation predictor in a gas turbine combustor using an artificial neural network technique, characterized in that it is heated to a temperature between 190 ° C and 210 ° C and introduced into the nozzle unit.
The method of claim 1,
the combustion unit,
A pressure perturbation prediction device in a gas turbine combustor using an artificial neural network technique, characterized in that the test combustor is provided in a reduced form at a predetermined ratio.
The method of claim 1,
The flame center information,
A pressure perturbation predictor in a gas turbine combustor using an artificial neural network technique, characterized in that it is calculated as the position of the maximum intensity value in the self-luminous image of the flame.
The method of claim 7,
The self-luminous image,
An apparatus for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique, characterized in that it is photographed by a CMOS camera coupled with an OH filter or a CH filter.
The method of claim 1,
The outlet flow information,
Artificial neural network, characterized in that calculated by dividing the predetermined sum of flow rates by the outlet area of the nozzle unit, measured through particle image velocimetry (PIV), or measured through hot-wire anemometer Pressure fluctuation prediction device in gas turbine combustor using method.
The method of claim 1,
The calculation unit,
An apparatus for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique, characterized in that for calculating the frequency information of the pressure perturbation and the intensity information of the pressure perturbation based on Equation 1.
{Formula 1}

(The x is the input value, the y is the frequency information of the pressure perturbation or the intensity information of the pressure perturbation, the w is a weight matrix between each layer of the artificial neural network model, and the b is the bias value in each measurement matrix, where f and g are activation functions)
The frequency information of the pressure perturbation and the intensity information of the pressure perturbation according to the flow information, which is a characteristic of the third fluid formed by mixing the first and second fluids in the nozzle unit, and the flame information, which is a characteristic of the flame formed in the combustion unit, are learned. In the apparatus for predicting pressure perturbation in a gas turbine combustor using an artificial neural network technique,
a fluid inlet step in which the first fluid and the second fluid at predetermined flow rates are respectively introduced into the nozzle unit by the fluid inlet unit;
a combustion step in which the third fluid formed by mixing the first fluid and the second fluid in the nozzle unit flows into a combustion unit to form the flame;
a feature calculation step in which the flow information and the flame information are calculated by a feature calculation unit; and
Using artificial neural network technique including a prediction step in which frequency information and intensity information of pressure perturbation are calculated by the predictor based on a deep learning model by receiving the flow information and the flame information to predict pressure perturbation A method for predicting pressure fluctuations in gas turbine combustors.
The method of claim 11,
The flow information,
Gas turbine using an artificial neural network technique, characterized in that it includes equivalence ratio information, which is the ratio of the flow rate of the first fluid to the flow rate of the second fluid, and outlet flow rate information, which is the rate at which the third fluid is discharged from the nozzle unit A method for predicting pressure fluctuations in combustors.
The method of claim 11,
The flame information,
A pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique, characterized in that it includes flame center information corresponding to the center of the self-luminous image of the flame formed in the combustion unit.
The method of claim 11,
The prediction step is
Pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique, characterized in that using an artificial neural network (ANN) model.
The method of claim 11,
The first fluid,
provided by air,
The second fluid,
It is prepared with hydrocarbon fuel,
The first fluid,
Pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique, characterized in that heated to a temperature between 190 ℃ and 210 ℃ introduced into the nozzle unit.
The method of claim 11,
the combustion unit,
A pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique, characterized in that the test combustor is provided in a reduced form at a predetermined ratio.
The method of claim 11,
The flame center information,
Pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique, characterized in that calculated as the position of the maximum intensity value in the self-luminous image of the flame.
The method of claim 17
The self-luminous image,
A pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique, characterized in that it is photographed by a CMOS camera coupled with an OH filter or a CH filter.
The method of claim 11,
The outlet flow information,
Artificial neural network, characterized in that calculated by dividing the predetermined sum of flow rates by the outlet area of the nozzle unit, measured through particle image velocimetry (PIV), or measured through hot-wire anemometer Method for Predicting Pressure Perturbation in Gas Turbine Combustor Using Method.
The method of claim 11,
In the calculation step,
A pressure perturbation prediction method in a gas turbine combustor using an artificial neural network technique, characterized in that for calculating the frequency information of the pressure perturbation and the intensity information of the pressure perturbation based on Equation 1.
{Formula 1}

(The x is the input value, the y is the frequency information of the pressure perturbation or the intensity information of the pressure perturbation, the w is a weight matrix between each layer of the artificial neural network model, and the b is a bias value in each measurement matrix, where f and g are activation functions)

Images