Method for predicting performance of gas turbine based on digital twin technology

Abstract

The invention relates to the field of gas turbine performance prediction, in particular to a method for predicting gas turbine performance based on a digital twin technology. The method comprises the following steps: step 1, establishing a gas turbine mechanism model considering measurement noise according to a gas turbine physical system and historical data; step 2, predicting through the mechanism model to obtain gas turbine gas path performance parameters; 3, training a data model according to the actual measurement result of the gas turbine and the measurement result of the mechanism model, and obtaining a mechanism model error prediction model based on an SOA-BP neural network algorithm; and 4, establishing a hybrid model by using the gas turbine mechanism model and the mechanism model error prediction data model, and the like. Noise characteristics are measured through actual measurement data, and measurement noise is added in a mechanism model, so that a model result is more consistent with an actual value.

Description

Technical Field

[0001] The present invention relates to the field of gas turbine performance prediction, and in particular to a method for predicting gas turbine performance based on digital twin technology. Background Art

[0002] The whole machine performance simulation is an important part of gas turbine simulation, and its main purpose is to predict the performance parameters of the gas turbine through the model, including emissions, cost, life, gas path performance, reliability and other indicators. At present, there are two main ideas for establishing a gas turbine performance prediction model: one is that the mechanism model is mainly based on the thermodynamic relationship of the physical system. Due to the performance degradation of the gas turbine and the personalized characteristics of the component characteristic curve, the simulation accuracy of the mechanism model is difficult to meet the accuracy requirements of the digital twin model; the second is the data model, which predicts the performance of the gas turbine based on historical data and artificial intelligence algorithms. The data model is mainly affected by the quality and quantity of historical data, and the output result may have the problem of overfitting.

[0003] Accurate and fast gas turbine performance simulation models play a vital role in improving gas turbine performance prediction, fault diagnosis and control optimization. Pure gas turbine data-driven models are too affected by the quality and quantity of sensor measurement parameters and have limited generalization capabilities; while traditional mechanism models are too dependent on component characteristic diagrams and have limited model accuracy.

[0004] With the booming development of contemporary artificial intelligence, introducing advanced machine learning algorithms into gas turbine simulation models and building gas turbine performance prediction models can be used in the fields of gas turbine performance monitoring, fault detection, pattern recognition, and health management. This can promote greater development of gas turbines, so it is very important to use physical knowledge and machine learning algorithms to establish gas turbine performance simulation models. Summary of the invention

[0005] In view of the deficiencies of the prior art, the present invention provides a method for predicting gas turbine performance based on digital twin technology. The present invention establishes a data-driven model based on the SOA-BP neural network, and adjusts the output of the mechanism model based on the data-driven model output and the SOA-BP neural network algorithm. The present invention combines the data model, the mechanism model and the component characteristic adaptation to achieve the fusion of data and mechanism. The component characteristic curve in the mechanism model is adjusted by the physical gas turbine operation data, and the measurement noise characteristics are measured by the measured data, and the measurement noise is added to the mechanism model to make the model results more consistent with the actual values.

[0006] The technical solution of the present invention is: a method for predicting gas turbine performance based on digital twin technology, characterized in that it includes the following steps:

[0007] Step 1: Establish a gas turbine mechanism model considering measurement noise based on the gas turbine physical system and historical data, including the following parts:

[0008] Step 11. Use modular modeling method to establish each module in the basic mechanism model of gas turbine.

[0009] Step 12. Using the measured data under different working conditions of the gas turbine to correct the characteristic curves of the compressor and turbine, a characteristic curve closer to the actual compressor and turbine performance is obtained;

[0010] Step 13. Determine the import and export parameters of each module, set the compressor module input and output parameters, the combustion chamber module input and output parameters, the turbine module input and output parameters, the rotor module input and output parameters, and the regenerator module input and output parameters according to the balance relationship between the gas turbine modules, and connect the modules to form a gas turbine mechanism model;

[0011] Step 14. According to the noise characteristics of the gas turbine measurement data, on the basis of the established mechanism model, the measurement noise is added to the predicted values ​​of each measurement parameter to obtain data containing the measurement noise;

[0012] Step 2: predicting the gas turbine gas path performance parameters through the mechanism model; using the ambient temperature and power parameters in the N groups of gas turbine measured data at different powers as inputs to the mechanism model, and predicting the gas turbine gas path parameters through the mechanism model;

[0013] Step 3: training the data model according to the actual measurement results of the gas turbine and the measurement results of the mechanism model, and obtaining the mechanism model error prediction model based on the SOA-BP neural network algorithm;

[0014] Step 31. Obtain N groups of mechanism model prediction error data based on the N groups of measured data and the corresponding N groups of data containing measurement noise obtained based on the mechanism model prediction in step 2;

[0015] The deviation calculation formula is: Among them, M measured is the measured value of gas path parameters, M PBM is the PBM model predicted value of the gas path parameter, δ is the deviation;

[0016] Step 32. Select the compressor inlet temperature and the gas turbine load as the input vectors of the SOA-BP neural network model, and the compressor outlet temperature deviation, the combustion chamber outlet temperature deviation, the fuel flow deviation, and the turbine outlet temperature deviation as the SOA-BP neural network output vectors;

[0017] Step 33. Set the ratio of training and validation data in the N groups of data to 8:2, and use the actual measured results of another M groups of repeated experiments in the historical data as test data; perform deviation standardization on the training, validation and test data; the processing formula is as follows:

[0018] y=(δ-δ min ) / (δ max -δ min )

[0019] y is the normalized data, δ is the prediction error of the gas turbine parameter mechanism model, δ max is the maximum prediction error of the gas turbine parameter mechanism model, δ min is the minimum prediction error of the gas turbine parameter mechanism model;

[0020] Step 34. Set the number of neurons in the input layer and output layer of the BP neural network, the size of the hidden layer, the number of weights from the input layer to the hidden layer, and the number of weights from the hidden layer to the output layer;

[0021] Step 35. Set the hyperparameters of the BP neural network, including the number of training times, learning rate, training target minimum error, momentum factor, minimum performance gradient, etc.;

[0022] Step 36. Optimize the initial weights and initial thresholds of the BP neural network using the Seagull optimization algorithm;

[0023] Step 37. Input the training data into the BP neural network for iterative calculation to obtain the mechanism model error prediction model established by using the SOA-BP neural network;

[0024] Step 38. Use the mechanism model error prediction model established by the SOA-BP neural network to predict the M groups of gas turbine mechanism model deviations of the test data, and obtain the prediction value δ′ of the data model according to the prediction deviation;

[0025] Step 4: Establish a hybrid model using the gas turbine mechanism model and the data model for mechanism model error prediction to obtain the predicted values ​​of gas path parameter measurements.

[0026] According to the method for predicting gas turbine performance based on digital twin technology as described above, it is characterized in that: step 4 includes the following process:

[0027] Step 41. Input the compressor inlet temperature and gas turbine load power in the M group of test data into the mechanism model established in step 1 to obtain the prediction result of the mechanism model;

[0028] Step 42. According to the prediction results of the M group mechanism models and the prediction deviation of the M group gas turbine mechanism models obtained in step 3, the prediction value of the gas turbine measured data is calculated, and the formula is as follows:

[0029] Where M′ measured is the predicted value of the mixed model for the gas path parameter measurement, M PBM is the PBM model predicted value of the gas path parameter, and δ′ is the predicted value of the data model.

[0030] According to the method for predicting gas turbine performance based on digital twin technology as described above, it is characterized in that: in step 11, the modules in the basic mechanism model include a compressor module, a combustion chamber module, a turbine module, a rotor module, and a regenerator module;

[0031] The compressor flow and efficiency are obtained through the compressor characteristic curve. The calculation principle equation is expressed as follows:

[0032] G cnp =f C1 [n cnp ,π c ](1)

[0033] η c =f C2 [n cnp ,G cnp ](2)

[0034] Where: G cnp is the standard converted compressor flow, f C1 is the relationship between flow rate, speed and pressure ratio in compressor characteristics, f C2 is the relationship between efficiency, speed and flow rate in compressor characteristics, n cnp is the standard converted compressor efficiency, η c is the compressor efficiency, π c is the compressor pressure ratio, and the corresponding calculation formula is as follows:

[0035]

[0036] Where: n c is the compressor rotor speed, G c is the compressor flow rate, n d is the rated speed of the compressor design, T0 is the inlet temperature of the compressor characteristic line test, T1 is the inlet temperature of the compressor, P0 is the inlet pressure of the compressor characteristic line test, P1 is the inlet pressure of the compressor, and P2 is the outlet pressure of the compressor;

[0037] According to the thermodynamic properties of air and the isentropic process equation, the compressor outlet temperature and compressor power consumption can be obtained.

[0038]

[0039] Where: T2 is the compressor outlet temperature, Nec is the power consumption of the compressor, R is the gas constant of air, c p is the constant pressure specific heat capacity of air;

[0040] The formula for the combustion chamber module is as follows:

[0041]

[0042]

[0043] Where: C vg is the isochoric specific heat capacity of the gas, R g is the gas constant of the fuel gas, ρ is the fuel gas density in the combustion chamber, V is the volume of the combustion chamber, u is the internal energy of the gas, Q is the heat released by fuel combustion, P3 is the gas pressure at the turbine inlet, T3 is the gas temperature at the turbine inlet, G f is the fuel flow added to the combustion chamber, G2 is the air flow at the combustion chamber inlet, and G3 is the gas flow at the combustion chamber outlet; the combustion chamber outlet pressure P3 and outlet temperature T3 can be obtained through the combustion chamber module;

[0044] The equation for the turbine module is as follows:

[0045] G tnp =f T1 [n tnp ,π t ] (8)

[0046] η t =f T2 [n tnp ,G tnp ] (9)

[0047] f T1 is the relationship between flow rate, speed and pressure ratio in turbine characteristics, f T2 The relationship between efficiency, speed and flow rate in turbine characteristics. The turbine standard equivalent flow rate G can be obtained according to the turbine characteristic line. tnp and efficiency η t , (8)-(9):

[0048]

[0049] Where: n tnp is the standard converted turbine rotor speed, n t is the turbine rotor speed, n d is the turbine design rated speed, T3 is the turbine inlet temperature, and P3 is the turbine inlet pressure; T 30 The inlet temperature of the turbine characteristic line design point is 930℃, P 30 The turbine characteristic line is designed with an inlet pressure of 390 kPa, G tnpis the turbine standard equivalent flow rate, G t is the turbine working fluid flow rate;

[0050] According to the gas dynamics equation, the turbine outlet temperature and turbine power consumption can be obtained.

[0051]

[0052]

[0053] Where: T4 is the turbine outlet temperature, N et To do work for the turbine, R g is the gas constant of the fuel gas, c pg is the constant pressure specific heat capacity of the gas;

[0054] The rotor is the component that connects the compressor and the turbine; according to the law of momentum, the differential equation of the rotor's moment of inertia is:

[0055]

[0056] Where n is the rotor speed in r / min, dE / dt is the rotor residual power, Ne l is the load power, N et For turbine work, N ec is the power consumption of the compressor; the gas turbine rotor speed can be obtained according to formula (13);

[0057] The regenerator model is simplified to the following formula:

[0058] T aout =f[π c ]+T2 (14)

[0059] P aout =P2σ G (15)

[0060] Where: T aout represents the regenerator outlet air temperature, f[π c According to the fitting relationship between the inlet and outlet air temperature difference and the pressure ratio of the gas turbine regenerator, P aout represents the regenerator outlet air pressure, σ G represents the total pressure recovery coefficient of the regenerator; the outlet air temperature and outlet air pressure of the regenerator are obtained according to (14)-(15).

[0061] According to the method for predicting gas turbine performance based on digital twin technology as described above, it is characterized in that: the mechanism model parameters in step 13 include: (1) compressor module input parameters: compressor inlet air pressure P1, compressor outlet air pressure P2, compressor inlet air temperature T1 and compressor speed n; compressor module output parameters: compressor outlet temperature T2, compressor power consumption N ec and compressor air flow G c ;

[0062] (2) Combustion chamber module input parameters: inlet air flow rate G c , outlet gas flow G t , fuel quantity G f , combustion chamber inlet temperature T aout ; Combustion chamber module output parameters: combustion chamber outlet gas pressure P3, combustion chamber inlet air pressure P aout , combustion chamber outlet gas temperature T3;

[0063] (3) Turbine module input parameters: turbine inlet gas pressure P3, turbine inlet gas temperature T3, turbine speed n t , turbine outlet gas pressure P1; turbine module output parameters: turbine outlet temperature T4, turbine working fluid flow G t , turbine work N et ;

[0064] (4) Rotor module input parameters: turbine work N et , compressor power consumption, load power Ne l ;The output parameter of the rotor module is the rotor speed n;

[0065] (5) Regenerator module input parameters: compressor outlet air pressure P2, compressor outlet temperature T2, compressor outlet air pressure Regenerator module output parameters: regenerator outlet air temperature T aout , regenerator outlet air pressure P aout .

[0066] According to the method for predicting gas turbine performance based on digital twin technology as described above, it is characterized in that: the data M containing measurement noise in step 14 noisy The calculation formula is:

[0067] M noisy =M nf +σ * R (17)

[0068] Where M nf is the noise-free value of the measurement result calculated by the mechanism model, σ * represents the variance of the measurement value of the relevant sensor; R is a random matrix.

[0069] The beneficial effects of the present invention are: combining data models, mechanism models and component characteristic adaptation to achieve the integration of data and mechanism. The component characteristic curves in the mechanism model are adjusted by the physical gas turbine operation data, and the measurement noise characteristics are measured by the measured data, and the measurement noise is added to the mechanism model to make the model results more consistent with the actual values. A data-driven model is established based on the SOA-BP neural network, and the output results of the mechanism model are adjusted based on the output of the data-driven model. Compared with the pure mechanism model, the method of the present invention can reduce the average absolute error of the performance parameter simulation by 1.2%, and the maximum average absolute error by 2.2%. Compared with the pure data model, the gas turbine performance prediction method of the present invention reduces the average absolute error of the performance parameter simulation by 2.5% and the maximum average absolute error by 1.9% under the condition of ambient temperature changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a verification of the gas turbine performance prediction results of the present invention.

[0071] Figure 2 is the gas turbine performance prediction error of the present invention.

[0072] Figure 3 This is to verify the generalization ability of the gas turbine performance prediction model of the present invention. DETAILED DESCRIPTION

[0073] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0074] This patent proposes a method for establishing a hybrid model, which combines data models, mechanism models and component characteristic adaptation to achieve the integration of data and mechanism. The component characteristic curves in the mechanism model are adjusted by the physical gas turbine operation data, a data-driven model is established based on the SOA-BP neural network, the output of the mechanism model is adjusted based on the data-driven model output, and a two-way data flow between the data-driven model and the mechanism model is established. Compared with the pure mechanism model, the proposed method can reduce the average absolute error of performance parameter simulation by 1.2% and the maximum average absolute error by 2.2%. Compared with the pure data model, the proposed DT method reduces the average absolute error of performance parameter simulation by 2.5% and the maximum average absolute error by 1.9% under the condition of ambient temperature changes. The results show that the digital twin method developed based on the hybrid model can effectively and reliably predict gas turbine performance and help gas turbine operation optimization and health management decisions.

[0075] The following steps are involved:

[0076] Step 1: Establish a gas turbine mechanism model considering measurement noise based on the gas turbine physical system and historical data, including the following parts:

[0077] Step 11. Use the classic modular modeling method to establish each module in the basic mechanism model of the gas turbine, including the compressor module, combustion chamber module, turbine module, rotor module, and regenerator module.

[0078] The compressor flow and efficiency are obtained through the compressor characteristic curve, and the corresponding calculation principle equations are expressed as shown in equations (1)-(2).

[0079] G cnp =f C1 [n cnp ,π c ] (1)

[0080] η c =f C2 [n cnp ,G cnp ](2)

[0081] Where: G cnp is the standard converted compressor flow, f C1 is the relationship between flow rate, speed and pressure ratio in compressor characteristics, f C2 is the relationship between efficiency, speed and flow rate in compressor characteristics, n cnp is the standard converted compressor efficiency, η c is the compressor efficiency, π c is the compressor pressure ratio, and the corresponding calculation formula is as follows.

[0082]

[0083] Where: n c is the compressor rotor speed, G c is the compressor flow rate, n d is the design rated speed of the compressor, 51000r / min, T0 is the inlet temperature of the compressor characteristic line test, 288.15K, T1 is the inlet temperature of the compressor, P0 is the inlet pressure of the compressor characteristic line test, 101.325kPa, P1 is the inlet pressure of the compressor, and P2 is the outlet pressure of the compressor.

[0084] According to the thermodynamic properties of air and the isentropic process equation, the compressor outlet temperature and compressor power consumption can be obtained.

[0085]

[0086] Where: T2 is the compressor outlet temperature, N ec is the power consumption of the compressor, R is the gas constant of air, c p is the specific heat of air at constant pressure.

[0087] The combustion chamber mainly provides the turbine with high-temperature and high-pressure gas after the mixture of compressed air and fuel is burned. According to the principles of conservation of mass and conservation of energy, the combustion chamber module takes into account the volume inertia of the combustion chamber and ignores the thermal inertia of the combustion chamber. The formula is as follows:

[0088]

[0089] Where: C vg is the isochoric specific heat capacity of the gas, R g is the gas constant of the fuel gas, ρ is the fuel gas density in the combustion chamber, V is the volume of the combustion chamber, u is the internal energy of the gas, Q is the heat released by fuel combustion, P3 is the gas pressure at the turbine inlet, T3 is the gas temperature at the turbine inlet, G f is the fuel flow added to the combustion chamber, G2 is the combustion chamber inlet air flow, and G3 is the combustion chamber outlet gas flow. The combustion chamber outlet pressure P3 and outlet temperature T3 can be obtained through the combustion chamber module.

[0090] The principle equation of the turbine module is shown below:

[0091] G tnp =f T1 [n tnp ,π t ] (8)

[0092] η t =f T2 [n tnp ,G tnp ] (9)

[0093] f T1 is the relationship between flow rate, speed and pressure ratio in turbine characteristics, f T2 The relationship between efficiency, speed and flow rate in turbine characteristics. The turbine standard equivalent flow rate G can be obtained according to the turbine characteristic line. tnp and efficiency η t , (8)-(9):

[0094]

[0095] Where: n tnp is the standard converted turbine rotor speed, n t is the turbine rotor speed, n d T is the turbine design rated speed, T3 is the turbine inlet temperature, and P3 is the turbine inlet pressure. 30 The inlet temperature of the turbine characteristic line design point is 930℃, P 30 The turbine characteristic line is designed with an inlet pressure of 390 kPa, G tnp is the turbine standard equivalent flow rate, G t is the turbine working fluid flow rate.

[0096] According to the gas dynamics equation, the turbine outlet temperature and turbine power consumption can be obtained.

[0097]

[0098] Where: T4 is the turbine outlet temperature, N et To do work for the turbine, R g is the gas constant of the fuel gas, c pg is the constant pressure specific heat capacity of the gas.

[0099] The rotor is the component that connects the compressor and the turbine. According to the law of momentum, the differential equation of the rotor's moment of inertia is:

[0100]

[0101] Where n is the rotor speed in r / min, dE / dt is the rotor residual power, Ne l is the load power, N et For turbine work, N ec is the power consumption of the compressor. According to formula (13), the gas turbine rotor speed can be obtained.

[0102] The regenerator model is simplified to the following formula:

[0103] T aout =f[π c ]+T2 (14)

[0104] P aout =P2σ G (15)

[0105] Where: T aout represents the regenerator outlet air temperature, f[π c According to the fitting relationship between the inlet and outlet air temperature difference and the pressure ratio of the gas turbine regenerator, P aout represents the regenerator outlet air pressure, σ G represents the total pressure recovery coefficient of the regenerator. According to (14)-(15), the outlet air temperature and outlet air pressure of the regenerator can be obtained:

[0106] Step 12. Use the measured data under different operating conditions of the gas turbine to correct the characteristic curves of the compressor and turbine to obtain characteristic curves that are closer to the actual compressor and turbine performance, thereby improving the calculation accuracy of the compressor and turbine modules.

[0107] Step 13. Determine the import and export parameters of each module, set the compressor module input and output parameters, combustion chamber module input and output parameters, turbine module input and output parameters, rotor module input and output parameters, and regenerator module input and output parameters according to the balance relationship between the gas turbine modules, and connect the modules to form a gas turbine mechanism model.

[0108] (1) Input parameters of the compressor module: compressor inlet air pressure P1, compressor outlet air pressure P2, compressor inlet air temperature T1 and compressor speed n; output parameters of the compressor module: compressor outlet temperature T2, compressor power consumption N ec and compressor air flow G c ;

[0109] (2) Combustion chamber module input parameters: inlet air flow rate G c , outlet gas flow G t , fuel quantity G f , combustion chamber inlet temperature T aout ; Combustion chamber module output parameters: combustion chamber outlet gas pressure P3, combustion chamber inlet air pressure P aout , combustion chamber outlet gas temperature T3;

[0110] (3) Turbine module input parameters: turbine inlet gas pressure P3, turbine inlet gas temperature T3, turbine speed n t , turbine outlet gas pressure P1; turbine module output parameters: turbine outlet temperature T4, turbine working fluid flow G t , turbine work N et ;

[0111] (4) Rotor module input parameters: turbine work N et , compressor power consumption, load power Ne l ;The output parameter of the rotor module is the rotor speed n;

[0112] (5) Regenerator module input parameters: compressor outlet air pressure P2, compressor outlet temperature T2, compressor outlet air pressure Regenerator module output parameters: regenerator outlet air temperature T aout , regenerator outlet air pressure P aout

[0113] Step 14. According to the noise characteristics of the gas turbine measurement data, on the basis of the established mechanism model, the measurement noise is added to the predicted value of each measurement parameter to obtain data containing measurement noise. The gas turbine shaft speed measurement belongs to a frequency signal. The present invention does not consider the noise of the speed sensor. For the temperature, pressure and flow sensor noise, a normal distribution with a mean of zero and a predefined variance is considered. When the predefined variance is determined for each measurement, data containing measurement noise can be generated by formula (17).

[0114] M noisy =M nf +σ * R (17)

[0115] Where Mnf is the noise-free value of the measurement result calculated by the mechanism model. The predefined variance σ of each measurement * Represents the variance of the measurement value of the relevant sensor. R is a random matrix that follows a standard normal distribution and has a size similar to M. noisy Same. noisy Refers to the measurement value containing random errors. Referring to the literature data, the predefined measurement noise variance of each sensor used in the present invention is shown in Table 1.

[0116] Table 1. Noise settings for measuring sensors

[0117]

[0118] Step 2: Predict gas turbine gas path performance parameters through mechanism model

[0119] The ambient temperature and power parameters in N groups of gas turbine measured data at different powers obtained in the experiment are used as the input of the mechanism model. 1024 groups of gas turbine gas path parameters are predicted by the mechanism model, that is, 1024 groups of data containing measurement noise are obtained, including compressor outlet temperature, combustion chamber outlet temperature, turbine outlet temperature, fuel flow rate and other parameters.

[0120] Step 3: Train the data model according to the actual measurement results of the gas turbine and the measurement results of the mechanism model, and obtain the mechanism model error prediction model based on the SOA-BP neural network algorithm.

[0121] Step 31. Based on the N groups of measured data and the corresponding N groups of data containing measurement noise obtained based on the mechanism model prediction in step 2, N groups of mechanism model prediction error data are obtained. The error data is the deviation, including compressor outlet temperature deviation, combustion chamber outlet temperature deviation, turbine outlet temperature deviation, and fuel flow deviation.

[0122] The deviation calculation formula is: Among them, M measured is the measured value of gas path parameters, M PBM is the PBM model predicted value of the gas path parameter, and δ is the deviation.

[0123] Step 32. Select the compressor inlet temperature and gas turbine load as the input vectors of the SOA-BP neural network model, and the compressor outlet temperature deviation, the combustion chamber outlet temperature deviation, the fuel flow deviation, and the turbine outlet temperature deviation as the SOA-BP neural network output vectors.

[0124] Step 33. Set the ratio of training and validation data in the N groups of data to 8:2, and use the actual test results of another M groups of repeated experiments in the historical data as test data. Perform deviation standardization on the training, validation and test data. The processing formula is as follows:

[0125] y=(δ-δ min ) / (δ max -δ min )

[0126] y is the normalized data, δ is the prediction error of the gas turbine parameter mechanism model, δ max is the maximum prediction error of the gas turbine parameter mechanism model, δ min It is the minimum prediction error of the gas turbine parameter mechanism model.

[0127] Step 34. Set the number of neurons in the input layer and output layer of the BP neural network, the size of the hidden layer, the number of weights from the input layer to the hidden layer, and the number of weights from the hidden layer to the output layer;

[0128] Step 35. Set the hyperparameters of the BP neural network, including the number of training times, learning rate, training target minimum error, momentum factor, minimum performance gradient, etc.

[0129] Step 36. Optimize the initial weights and initial thresholds of the BP neural network using the Seagull optimization algorithm.

[0130] Step 37. Input the training data into the BP neural network for iterative calculation to obtain the mechanism model error prediction model established using the SOA-BP neural network.

[0131] Step 38. Based on the mechanism model error prediction model established by using the SOA-BP neural network, the M groups of gas turbine mechanism model deviations of the test data are predicted, including the compressor outlet temperature prediction deviation σ′(T2), the combustion chamber outlet temperature prediction deviation σ′(T3), the fuel flow prediction deviation σ′(G f ), turbine outlet temperature prediction deviation σ′(T4).

[0132] Step 4: Establish a hybrid model using the gas turbine mechanism model and the data model for mechanism model error prediction to obtain the predicted values ​​of gas path parameter measurements.

[0133] Step 41. Input the compressor inlet temperature and gas turbine load power in the M group of test data into the mechanism model established in step 1 to obtain the prediction results of the mechanism model, including parameters such as compressor outlet temperature, combustion chamber outlet temperature, turbine outlet temperature, fuel flow rate, etc.

[0134] Step 42. Based on the prediction results of the above M groups of mechanism models and the prediction deviations of the M groups of gas turbine mechanism models obtained in step 3, the prediction values ​​of the gas turbine measured data are calculated. The calculation formula is as follows:

[0135] Where M′ measured is the predicted value of the mixed model for the gas path parameter measurement, MPBM is the PBM model predicted value of the gas path parameter, and δ′ is the predicted value of the data model.

[0136] Figure 1 The gas turbine performance prediction result of the present invention is verified. f -PBM+DDM is the fuel flow prediction result of the hybrid model, G f -PBM is the fuel flow prediction result of the mechanism model, G f -measured is the measured result of fuel flow; P2-PBM+DDM is the predicted result of compressor outlet pressure of hybrid model, P2-PBM is the predicted result of compressor outlet pressure of mechanism model, P2-measured is the measured value of compressor outlet pressure; T3-PBM+DDM is the predicted result of combustion chamber outlet temperature of hybrid model, T3-PBM is the predicted result of combustion chamber outlet temperature of mechanism model, T3-measured is the measured value of combustion chamber outlet temperature; T4-PBM+DDM is the predicted result of turbine outlet temperature of hybrid model, T4-PBM is the predicted result of turbine outlet temperature of mechanism model, T4-measured is the measured value of turbine outlet temperature.

[0137] Figure 2 The gas turbine performance prediction error of the present invention is visualized.

[0138] Figure 3 This is the verification result of the generalization ability of the gas turbine performance prediction model of the present invention.

[0139] Table 2 shows the results of the adaptability verification of different models under environmental changes.

[0140] Table 2 Verification results of adaptability of different models under environmental changes

[0141]

[0142]

Claims

1. A method for predicting gas turbine performance based on digital twin technology, characterized in that: The following steps are involved: Step 1: Establish a gas turbine mechanism model considering measurement noise based on the gas turbine physical system and historical data, including the following parts: Step 11. Use modular modeling method to establish each module in the basic mechanism model of gas turbine. Step 12. Using the measured data under different working conditions of the gas turbine to correct the characteristic curves of the compressor and turbine, a characteristic curve closer to the actual compressor and turbine performance is obtained; Step 13. Determine the import and export parameters of each module, set the compressor module input and output parameters, the combustion chamber module input and output parameters, the turbine module input and output parameters, the rotor module input and output parameters, and the regenerator module input and output parameters according to the balance relationship between the gas turbine modules, and connect the modules to form a gas turbine mechanism model; Step 14. According to the noise characteristics of the gas turbine measurement data, on the basis of the established mechanism model, the measurement noise is added to the predicted values ​​of each measurement parameter to obtain data containing the measurement noise; Step 2: predicting the gas turbine gas path performance parameters through the mechanism model; using the ambient temperature and power parameters in the N groups of gas turbine measured data at different powers as inputs to the mechanism model, and predicting the gas turbine gas path parameters through the mechanism model; Step 3: training the data model according to the actual measurement results of the gas turbine and the measurement results of the mechanism model, and obtaining the mechanism model error prediction model based on the SOA-BP neural network algorithm; Step 31. Obtain N groups of mechanism model prediction error data based on the N groups of measured data and the corresponding N groups of data containing measurement noise obtained based on the mechanism model prediction in step 2; The deviation calculation formula is: Among them, M measured is the measured value of gas path parameters, M PBM is the PBM model predicted value of the gas path parameter, δ is the deviation; Step 32. Select the compressor inlet temperature and the gas turbine load as the input vectors of the SOA-BP neural network model, and the compressor outlet temperature deviation, the combustion chamber outlet temperature deviation, the fuel flow deviation, and the turbine outlet temperature deviation as the SOA-BP neural network output vectors; Step 33. Set the ratio of training and validation data in the N groups of data to 8:2, and use the actual measured results of another M groups of repeated experiments in the historical data as test data; perform deviation standardization on the training, validation and test data; the processing formula is as follows: y=(δ-δ min ) / (d max -d min ) y is the normalized data, δ is the prediction error of the gas turbine parameter mechanism model, δ max is the maximum prediction error of the gas turbine parameter mechanism model, δ min is the minimum prediction error of the gas turbine parameter mechanism model; Step 34. Set the number of neurons in the input layer and output layer of the BP neural network, the size of the hidden layer, the number of weights from the input layer to the hidden layer, and the number of weights from the hidden layer to the output layer; Step 35. Set the hyperparameters of the BP neural network, including the number of training times, learning rate, training target minimum error, momentum factor, minimum performance gradient, etc.; Step 36. Optimize the initial weights and initial thresholds of the BP neural network using the Seagull optimization algorithm; Step 37. Input the training data into the BP neural network for iterative calculation to obtain the mechanism model error prediction model established by using the SOA-BP neural network; Step 38. Use the mechanism model error prediction model established by the SOA-BP neural network to predict the M groups of gas turbine mechanism model deviations of the test data, and obtain the prediction value δ′ of the data model according to the prediction deviation; Step 4: Establish a hybrid model using the gas turbine mechanism model and the data model for mechanism model error prediction to obtain the predicted values ​​of gas path parameter measurements.

2. The method for predicting gas turbine performance based on digital twin technology according to claim 1, characterized in that: Step 4 includes the following process: Step 41. Input the compressor inlet temperature and gas turbine load power in the M group of test data into the mechanism model established in step 1 to obtain the prediction result of the mechanism model; Step 42. According to the prediction results of the M group mechanism models and the prediction deviation of the M group gas turbine mechanism models obtained in step 3, the prediction value of the gas turbine measured data is calculated, and the formula is as follows: Where M′ measured is the predicted value of the mixed model for the gas path parameter measurement, M PBM is the PBM model predicted value of the gas path parameter, and δ′ is the predicted value of the data model.

3. The method for predicting gas turbine performance based on digital twin technology according to claim 1 or 2, characterized in that: In step 11, the modules in the basic mechanism model include a compressor module, a combustion chamber module, a turbine module, a rotor module, and a regenerator module; The compressor flow and efficiency are obtained through the compressor characteristic curve. The calculation principle equation is expressed as follows: G cnp =f C1 [n cnp ,p c ] (1) or c =f C2 [n cnp ,G cnp ](2) Where: G cnp is the standard converted compressor flow, f C1 is the relationship between flow rate, speed and pressure ratio in compressor characteristics, f C2 is the relationship between efficiency, speed and flow rate in compressor characteristics, n cnp is the standard converted compressor efficiency, η c is the compressor efficiency, π c is the compressor pressure ratio, and the corresponding calculation formula is as follows: Where: n c is the compressor rotor speed, G c is the compressor flow rate, n d is the rated speed of the compressor design, T0 is the inlet temperature of the compressor characteristic line test, T1 is the inlet temperature of the compressor, P0 is the inlet pressure of the compressor characteristic line test, P1 is the inlet pressure of the compressor, and P2 is the outlet pressure of the compressor; According to the thermodynamic properties of air and the isentropic process equation, the compressor outlet temperature and compressor power consumption can be obtained. Where: T2 is the compressor outlet temperature, N ec is the power consumption of the compressor, R is the gas constant of air, c p is the constant pressure specific heat capacity of air; The formula for the combustion chamber module is as follows: Where: C vg is the isochoric specific heat capacity of the gas, R g is the gas constant of the fuel gas, ρ is the fuel gas density in the combustion chamber, V is the volume of the combustion chamber, u is the internal energy of the gas, Q is the heat released by fuel combustion, P3 is the gas pressure at the turbine inlet, T3 is the gas temperature at the turbine inlet, G f is the fuel flow added to the combustion chamber, G2 is the air flow at the combustion chamber inlet, and G3 is the gas flow at the combustion chamber outlet; the combustion chamber outlet pressure P3 and outlet temperature T3 can be obtained through the combustion chamber module; The equation for the turbine module is as follows: G tnp =f T1 [n tnp ,p t ] (8) or t =f T2 [n tnp ,G tnp ] (9) f T1 is the relationship between flow rate, speed and pressure ratio in turbine characteristics, f T2 The relationship between efficiency, speed and flow rate in turbine characteristics. The turbine standard equivalent flow rate G can be obtained according to the turbine characteristic line. tnp and efficiency η t , (8)-(9): Where: n tnp is the standard converted turbine rotor speed, n t is the turbine rotor speed, n d is the turbine design rated speed, T3 is the turbine inlet temperature, and P3 is the turbine inlet pressure; T 30 The inlet temperature of the turbine characteristic line design point is 930℃, P 30 The turbine characteristic line is designed with an inlet pressure of 390 kPa, G tnp is the turbine standard equivalent flow rate, G t is the turbine working fluid flow rate; According to the gas dynamics equation, the turbine outlet temperature and turbine power consumption can be obtained. Where: T4 is the turbine outlet temperature, N et To do work for the turbine, R g is the gas constant of the fuel gas, c pg is the constant pressure specific heat capacity of the gas; The rotor is the component that connects the compressor and the turbine; according to the law of momentum, the differential equation of the rotor's moment of inertia is: Where n is the rotor speed in r / min, dE / dt is the rotor residual power, Ne l is the load power, N et For the turbine to do work, N ec is the power consumption of the compressor; the gas turbine rotor speed can be obtained according to formula (13); The regenerator model is simplified to the following formula: T aout =f[π c ]+T2 (14) P aout =P2σ G (15) Where: T aout represents the regenerator outlet air temperature, f[π c According to the fitting relationship between the inlet and outlet air temperature difference and the pressure ratio of the gas turbine regenerator, P aout represents the regenerator outlet air pressure, σ G represents the total pressure recovery coefficient of the regenerator; the outlet air temperature and outlet air pressure of the regenerator are obtained according to (14)-(15).

4. The method for predicting gas turbine performance based on digital twin technology according to claim 1 or 2, characterized in that: The mechanism model parameters in step 13 include: (1) compressor module input parameters: compressor inlet air pressure P1, compressor outlet air pressure P2, compressor inlet air temperature T1 and compressor speed n; compressor module output parameters: compressor outlet temperature T2, compressor power consumption N ec and compressor air flow G c ; (2) Combustion chamber module input parameters: inlet air flow rate G c , outlet gas flow G t , fuel quantity G f , combustion chamber inlet temperature T aout ; Combustion chamber module output parameters: combustion chamber outlet gas pressure P3, combustion chamber inlet air pressure P aout , combustion chamber outlet gas temperature T3; (3) Turbine module input parameters: turbine inlet gas pressure P3, turbine inlet gas temperature T3, turbine speed n t , turbine outlet gas pressure P1; turbine module output parameters: turbine outlet temperature T4, turbine working fluid flow G t , turbine work N et ; (4) Rotor module input parameters: turbine work N et , compressor power consumption, load power Ne l ;The output parameter of the rotor module is the rotor speed n; (5) Regenerator module input parameters: compressor outlet air pressure P2, compressor outlet temperature T2, compressor outlet air pressure Regenerator module output parameters: regenerator outlet air temperature T aout , regenerator outlet air pressure P aout .

5. The method for predicting gas turbine performance based on digital twin technology according to claim 1 or 2, characterized in that: The data M containing measurement noise in step 14 noisy The calculation formula is: M noisy =M nf +s * R (17) Where M nf is the noise-free value of the measurement result calculated by the mechanism model, σ * represents the variance of the measurement value of the relevant sensor; R is a random matrix.

Images