A multi-working condition modeling method for heavy-duty gas turbine

Abstract

The application provides a heavy-duty gas turbine multi-working condition modeling method and belongs to the field of information technology. The application is based on a heavy-duty gas turbine mechanism structure model, combines actual operation data of the heavy-duty gas turbine, introduces a deep neural network for system identification under a typical working condition, and adopts a quantum particle swarm optimization algorithm to optimize neural network structure parameters, so that high-precision modeling of nonlinear dynamic characteristics of the heavy-duty gas turbine is realized. The application solves the problem of low modeling precision of a single method of the heavy-duty gas turbine, improves model precision and robustness, and effectively provides technical support for field operators. The method can be widely applied in different industrial fields.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of information technology, and relates to a heavy gas turbine thermodynamic principle structure mechanism model, nonlinear modeling and data-based model expression, which is a heavy gas turbine modeling method based on mechanism-data fusion. The present application starts from the heavy gas turbine mechanism structure model, combines the actual operation data of the gas turbine, introduces a deep neural network for system identification under typical working conditions, and uses a quantum particle swarm optimization algorithm to optimize the neural network structure parameters, so as to realize high-precision modeling of the nonlinear dynamic characteristics of the gas turbine. The present application solves the problem of low modeling precision of the single method of the gas turbine, improves the model precision and robustness, and effectively provides technical support for the on-site operators. This method can be widely applied in different industrial fields. BACKGROUND

[0002] As a major breakthrough and development of China, the heavy gas turbine has been listed as one of the eight key tasks of the core competitiveness of the manufacturing industry in the national "14th Five-Year" development plan. The modeling research of the gas turbine focuses on improving the dynamic performance, thermal efficiency and multi-working condition adaptability. The modeling precision directly affects the reliability of the design, control, optimization and fault diagnosis, and even determines its actual engineering application. Therefore, it has important practical significance to construct a high-precision and high-reliability dynamic gas turbine model and study its operating characteristics under variable working conditions.

[0003] Gas turbine modeling is an important support for the development of gas turbine technology, which provides an important research basis for gas turbine design and performance optimization, operation control and intelligentization, as well as fault diagnosis and safety protection. At present, the modeling methods of heavy-duty gas turbine mainly include three types: mathematical modeling based on thermodynamic mechanism, black-box modeling based on actual data, and gray-box modeling combining the two. Mechanism modeling, as a typical modeling method, has been widely applied in gas turbine modeling (Zhang YZ, Liu P, Li Z (2021) Mechanism-data hybrid modeling of gas turbine based on dominant factor method. Engineering Thermophysics, 42(11): 9) (Li JX, Zhou DX, Xiao W, et al (2019) Research on mechanism-data hybrid modeling method of gas turbine. Thermal Power Engineering, 34(12)), which provides a technical support for the research and analysis of gas turbine steady-state performance. In addition, in order to effectively utilize the data generated by the gas turbine and analyze the operating characteristics under multiple conditions, more research starts from data (Rahmoune MBen, Hafaifa A, Kouzou A, et al (2021) Gas turbine monitoring using neural network dynamic nonlinear autoregressive with external exogenous input modelling. Mathematics and Computers in Simulation, 179: 23-47.) (Tan XM, Li W, Shen YH (2022) Research on multi-model fusion modeling method of heavy-duty gas turbine based on data-driven. Gas Turbine Technology, 35(02): 39-46.) to study the nonlinear characteristics of gas turbine. In recent years, more research has been done on the modeling of gas turbine with mechanism and data fusion (Xu M, Liu J, Li M, et al (2022) Improved hybrid modeling method with input and output self-tuning for gas turbine engine. Energy, 238: 121672.) (Bonfiglio A, Cacciacarne S, Invernizzi M, et al (2017) Gas turbine generating units control via feedback linearization approach. Energy, 121: 491-512.).

[0004] The existing gas turbine modeling method has great limitations under complex working conditions, such as simple description of nonlinear characteristics and dynamic changes, limited generalization ability and the like. In view of the above problems, a mechanism-data fusion modeling method is designed, a mechanism model is constructed based on the structure principle and operation principle of the gas turbine, and a fusion model framework is constructed combined with typical working condition operation data, so as to make up for the shortcomings of a single method, improve the model precision and robustness, and has important practical significance. SUMMARY

[0005] The problem solved by the present application is mainly the modeling and optimization of heavy-duty gas turbines under complex working conditions, which involves heavy-duty gas turbine modeling and optimization of mechanism modeling, multi-working condition model identification and mechanism data fusion. In order to solve the above problems, the field data of an industrial gas turbine are analyzed, the measured data are preprocessed, the ideas of modular modeling and mechanism modeling are used, the heavy-duty gas turbine nonlinear mathematical model is built based on the law of conservation of mass and the law of conservation of energy and thermodynamic analysis, and in the process of establishing the mathematical model, the combustion chamber model is introduced to fully reflect the operation process of the gas turbine. Finally, the operation data and the gas turbine mathematical model are combined, the deep neural network model is trained, the QPSO algorithm is introduced to optimize the network structure and hyperparameters, and the mechanism data fusion gas turbine model is constructed. By using the present application, the operation data and the state change can be accurately obtained, the requirements of the model precision of the engineering can be met, and the technical support can be effectively provided for the on-site operators.

[0006] The technical scheme of the present application:

[0007] A heavy-duty gas turbine multi-working condition modeling method, the specific steps are as follows:

[0008] Step 1, obtaining and preprocessing real-time operation data of heavy-duty gas turbine;

[0009] The heavy-duty gas turbine real-time operation data measured by the heavy-duty gas turbine sensor are filtered; the heavy-duty gas turbine real-time operation data include power generation, exhaust temperature, air flow and fuel flow;

[0010] Step 2, modular modeling of heavy-duty gas turbine based on mechanism equation;

[0011] Based on the mass conservation law and the energy conservation law and thermodynamic analysis, the nonlinear mathematical model of heavy-duty gas turbine is built by the idea of modular modeling and mechanism modeling. The combustion chamber pressure and the combustion chamber temperature are selected as the state variables of the nonlinear mathematical model of heavy-duty gas turbine, the air flow and the fuel flow are selected as the control input variables of the nonlinear mathematical model of heavy-duty gas turbine, and the power generation and the exhaust temperature of heavy-duty gas turbine are selected as the output variables of the nonlinear mathematical model of heavy-duty gas turbine. The specific calculation steps are as follows:

[0012] ① Combustion chamber main modeling: In the energy conversion process of the combustion chamber, the mass flow of the working medium satisfies the dynamic balance relationship to ensure the continuity of energy transmission of the heavy-duty gas turbine, which is specifically expressed as follows:

[0013] (1)

[0014] In the formula, and respectively represent the air flow and the fuel flow, kg / s; represents the flue gas flow at the outlet of the combustion chamber, kg / s; represents the combustion chamber volume, m3; is the air density, kg / m3; is the time.

[0015] Combined with the energy conservation law and the basic law of ideal gas, the following equation is obtained:

[0016] (2)

[0017] In the formula, represents the air constant-pressure specific heat capacity, J / kg·K; represents the inlet air temperature of the combustion chamber, K; and respectively represent the low heat value of fuel and the enthalpy value of flue gas at the outlet of the combustion chamber, J / kg; represents the internal energy of the control volume, J / kg, represents the constant-pressure specific heat capacity of flue gas at the outlet of the combustion chamber, J / kg·K; represents the constant of the outlet flue gas, J / kg·K; represents the outlet temperature of the combustion chamber, K.

[0018] Further obtained:

[0019] (3)

[0020] In the formula, is the combustion chamber pressure, is the inlet air temperature of the combustion chamber.

[0021] Solving the above equation, we get:

[0022] (4)

[0023] Considering the air compression process as a constant efficiency adiabatic process, we have:

[0024] (5)

[0025] where, T0represents the ambient temperature, K; P0represents the ambient pressure, Pa; η0represents the adiabatic efficiency of the compressor; R represents the ideal gas constant, J / kg·K. The flue gas flow rate at the outlet of the combustion chamber is represented as follows:

[0026] (6)

[0027] where, Q0represents the rated flow rate at the outlet of the combustion chamber, kg / s; P0represents the rated pressure at the outlet of the combustion chamber, Pa; T0represents the rated temperature at the outlet of the combustion chamber, K.

[0028] ② Heavy-duty gas turbine input module: The relationship between the fuel flow rate input by the heavy-duty gas turbine and the fuel valve opening, and the relationship between the air flow rate input by the heavy-duty gas turbine and the air introduction valve opening are both considered as first-order inertia links, which are specifically represented as:

[0029] (7)

[0030] where, τaand τfrespectively represent the time constants of the air introduction valve and the fuel valve actuator, Uaand Ufrespectively represent the control commands of the air flow rate and the gas flow rate, providing target reference values for the fuel valve actuator; represents a complex variable.

[0031] ③ Heavy-duty gas turbine output module: The output of the heavy-duty gas turbine is the power generation and the exhaust gas temperature , and the relationship between the heavy-duty gas turbine output and the state is represented as follows:

[0032] (8)

[0033] (9)

[0034] where, P0represents the rated power generation of the heavy-duty gas turbine, η0represents the adiabatic efficiency of the heavy-duty gas turbine.

[0035] (iv) The nonlinear mathematical model of the heavy-duty gas turbine is as follows:

[0036] (10)

[0037] (11)

[0038] wherein, , and , , and are white noises with mean 0, denote the model output power and exhaust temperature, , denote the measured Gaussian white noise of the output power with mean 0, denote the measured Gaussian white noise of the exhaust temperature with mean 0.

[0039] Step three, heavy-duty gas turbine modeling based on mechanism data fusion;

[0040] Based on the quantum particle swarm optimization deep neural network modeling method, the fusion of mechanism model and data driven method is realized, and the deep neural network model is obtained, and the specific calculation steps are as follows:

[0041] ① Standardize the real-time running data of the heavy-duty gas turbine after pretreatment in step one, and divide it into training set and test set.

[0042] ② Set the initial network structure and hyperparameter range of the deep neural network model, initialize the particle swarm optimization algorithm, including the population size, the maximum number of iterations and the search space dimension. Randomly generate the position vector of each particle in the defined search space, where each dimension value of the position vector directly corresponds to a specific hyperparameter configuration scheme, i.e. the combination of specific values of hidden layer node number, learning rate and training round.

[0043] ③ The parameters represented by the current position of the particle are used for deep neural network model training, and MSE is selected as the fitness function to evaluate the performance of the deep neural network model, which is defined as follows:

[0044] (12)

[0045] wherein, and denote the deep neural network model training output target and output result, denote the data number.

[0046] ④ Based on the fitness evaluation results, determine the individual optimal solution and the global optimal solution for each particle. During each iteration, the motion behavior of the particle swarm is defined as:

[0047] (13)

[0048] In the formula, The point of attraction for particle motion; Indicates the particle's current position; A random number within the interval (0,1); This represents the position of the particle in the previous iteration. and The first The individual optimal position and the population optimal position of each particle; The expansion / contraction coefficient represents the coefficient used to adjust the convergence performance of the quantum particle swarm optimization algorithm. The maximum and minimum values ​​of the expansion / contraction coefficient are expressed as follows: and ; Let be a constant that is randomly distributed in the interval (0,1); This represents the average optimal position of the particle; Indicates the total number of particles; and This represents the maximum number of iterations and the current number of iterations for the quantum particle swarm optimization algorithm.

[0049] ⑤ Repeat steps ③ and ④ until the optimal solution is obtained or the preset maximum number of iterations is reached, then end the quantum particle swarm optimization algorithm.

[0050] Step 4: Train the deep neural network model using the optimal solution to obtain the optimized deep neural network model, and verify the deep neural network model based on the measured data. If the error exceeds the specified error limit, return to Step 3 to retrain the deep neural network model until the error meets the requirements.

[0051] The effects and benefits of this invention are as follows: Based on the deep integration of thermodynamic mechanisms and deep learning, this invention constructs a high-precision simulation model of heavy-duty gas turbines under multiple coupled operating conditions, which can accurately characterize the dynamic characteristics of heavy-duty gas turbines in complex industrial scenarios. When establishing soft measurement models for key performance parameters, the collaborative optimization of mechanism constraints and data-driven approaches significantly improves computational robustness under various operating conditions. An innovative adaptive mechanism for neural network architecture oriented towards industrial data is proposed, enabling the model to retain the interpretability of thermodynamic equations while possessing the strong nonlinear fitting capability of deep learning. Based on verification using actual industrial operation data, this invention forms a mechanism- and data-driven digital twin modeling system for heavy-duty gas turbines, providing highly reliable technical support for unit energy efficiency optimization, fault prediction, and intelligent operation and maintenance. Attached Figure Description

[0052] Figure 1 is a flow chart of the implementation of the present application.

[0053] Figure 2 is a result chart of the power generation prediction.

[0054] Figure 3 is a result chart of the exhaust temperature prediction. DETAILED DESCRIPTION

[0055] In order to better understand the technical solutions of the present application, the present application takes a certain gas turbine as an example, and the embodiments of the present application are described in detail in combination with the drawings. The present application simulates and analyzes different operating conditions of the gas turbine, establishes a model of deep fusion of the gas turbine mechanism and data, and completes the theoretical support for the output predictive maintenance of the gas turbine. According to the method flow shown in the present application, the specific implementation steps of the present application are as follows: Figure 1

[0056] Step one, obtaining and preprocessing real-time operating data of heavy-duty gas turbine;

[0057] The real-time operating data of the heavy-duty gas turbine measured by the sensor of the heavy-duty gas turbine is filtered; the real-time operating data of the heavy-duty gas turbine includes power generation power, exhaust temperature, air flow and fuel flow;

[0058] Step two, modeling of heavy-duty gas turbine based on mechanism equation;

[0059] Based on the mass conservation law and the energy conservation law and the thermodynamic analysis, the heavy-duty gas turbine nonlinear mathematical model is built by the idea of modular modeling and mechanism modeling, the combustion chamber pressure and the combustion chamber temperature are selected as the state quantity of the heavy-duty gas turbine nonlinear mathematical model, the air flow and the fuel flow are selected as the control input quantity of the heavy-duty gas turbine nonlinear mathematical model, and the power generation power and the exhaust temperature of the heavy-duty gas turbine are selected as the output quantity of the heavy-duty gas turbine nonlinear mathematical model. The specific calculation steps are as follows:

[0060] ① Combustion chamber main modeling: in the energy conversion process of the combustion chamber, the mass flow of the working medium satisfies the dynamic balance relationship to ensure the continuity of the energy transmission of the heavy-duty gas turbine, which is specifically expressed as follows:

[0061] (1)

[0062] In the formula, and respectively represent the air flow and the fuel flow, kg / s; represent the flue gas flow at the outlet of the combustion chamber, kg / s; represent the combustion chamber volume, m3;​ is the air density, kg / m3; is the time.

[0063] Combining the law of conservation of energy and the basic law of ideal gas, we get:

[0064] (2)

[0065] where, is the air constant-pressure specific heat capacity, J / kg·K; is the combustion chamber inlet air temperature, K; and are the fuel low heat value and the combustion chamber outlet flue gas enthalpy value, respectively, J / kg; is the control volume internal energy, J / kg, is the combustion chamber outlet flue gas constant-pressure specific heat capacity, J / kg·K; is the outlet flue gas constant, J / kg·K; is the combustion chamber outlet temperature, K.

[0066] Further, we get:

[0067] (3)

[0068] where, is the combustion chamber pressure, is the combustion chamber inlet air temperature.

[0069] Solving the above equation, we get:

[0070] (4)

[0071] Considering the air compression process as an adiabatic process with constant efficiency, we have:

[0072] (5)

[0073] where, is the ambient temperature, K; is the ambient pressure, Pa; is the adiabatic efficiency of the compressor; is the ideal gas constant, J / kg·K. The combustion chamber outlet flue gas flow is expressed as:

[0074] (6)

[0075] where, is the combustion chamber outlet rated flow, kg / s; is the combustion chamber outlet rated pressure, Pa; is the combustion chamber outlet rated temperature, K.

[0076] ② Heavy-duty gas turbine input module: The relationship between the fuel flow rate input to the heavy-duty gas turbine and the fuel valve opening, and the relationship between the air flow rate input to the heavy-duty gas turbine and the air introduction valve opening are both regarded as first-order inertia links, which are specifically expressed as:

[0077] (7)

[0078] In the formula, respectively represent the time constant of the air introduction valve and the fuel valve actuator, respectively represent the control command of the air flow rate and the gas flow rate, and provide a target reference value for the fuel valve actuator; represents a complex variable.

[0079] ③ Heavy-duty gas turbine output module: The heavy-duty gas turbine output is the power generation and the exhaust temperature , and the relationship between the heavy-duty gas turbine output and the state is expressed in the following form:

[0080] (8)

[0081] (9)

[0082] In the formula, is the rated power generation of the heavy-duty gas turbine, is the adiabatic efficiency of the heavy-duty gas turbine.

[0083] ④ The heavy-duty gas turbine nonlinear mathematical model is as follows:

[0084] (10)

[0085] (11)

[0086] In the formula, , and , , and are white noises with a mean of 0, represents the model output power generation and exhaust temperature, , represents the measured Gaussian white noise of the output power generation with a mean of 0, represents the measured Gaussian white noise of the exhaust temperature with a mean of 0.

[0087] Step three, heavy-duty gas turbine modeling based on mechanism data fusion;

[0088] The modeling method based on quantum particle swarm optimization deep neural network realizes the fusion of mechanism model and data-driven method, and the specific calculation steps are as follows:

[0089] ①Standardize the real-time operation data of heavy gas turbine after pretreatment in step one, and divide it into training set and test set.

[0090] ②Set the initial network structure and hyperparameter range of the deep neural network model, initialize the particle swarm of the quantum particle swarm optimization algorithm, including the population size, the maximum number of iterations and the search space dimension. Randomly generate the position vector of each particle in the defined search space, where each dimension value of the position vector directly corresponds to a specific hyperparameter configuration scheme, i.e. the combination of specific values of hidden layer node number, learning rate and training round.

[0091] ③The parameters represented by the current position of the particle are used for deep neural network model training, and MSE is selected as the fitness function for evaluating the performance of the deep neural network model, which is defined as follows:

[0092] (12)

[0093] In the formula, and respectively represent the training output target and the output result of the deep neural network model, represents the number of data.

[0094] ④According to the fitness evaluation results, determine the individual optimal solution and global optimal solution of each particle. In each iteration process, the motion behavior of the particle swarm is defined as:

[0095] (13)

[0096] In the formula, represents the attractor of particle motion; represents the current position of the particle; is a random number in the interval (0, 1); is the position of the particle in the last iteration; and are the individual optimal position and population optimal position of the th particle, respectively; represents the expansion and contraction coefficient, which is used to adjust the convergence performance of the quantum particle swarm optimization algorithm. The maximum and minimum values of the expansion and contraction coefficient are represented as and , which usually satisfy ; is a constant randomly distributed in the interval (0, 1); represents the average optimal position of the particle; represents the total number of particles; and represents the maximum number of iterations of the quantum particle swarm optimization algorithm and the current number of iterations.

[0097] ⑤ Repeat steps ③ and ④ until the optimal solution is obtained or the upper limit of the preset number of iterations is reached, and the quantum particle swarm optimization algorithm is ended.

[0098] Step four, train the deep neural network model using the optimal solution, obtain the optimized deep neural network model, and verify the deep neural network model according to the measured data. If the error is higher than the specified upper limit of error, return to step three to retrain the deep neural network model until the error meets the requirements.

[0099] Table 1 is the effect comparison of the method of the application and other modeling methods

[0100]

[0101] Table 1 gives the effect comparison of the method of the application and other modeling methods. Analysis can be obtained that the prediction effect of the output power and the exhaust temperature under the comprehensive low working condition is the best. It can more accurately capture the internal relationship between the model input and output, and the prediction accuracy is higher than other prediction models in RMSE, MSE, MSE and MAPE, which can more accurately fit the dynamic characteristics of the gas turbine. The prediction effects of the power generation power and the exhaust temperature are shown in Figure 2 and Figure 3 The prediction results of the application are highly consistent with the actual output values, and the prediction accuracy and fitting effect are better than PSO-DNN and traditional DNN model, which can more accurately reflect the output change trend of the heavy-duty gas turbine.

Claims

1. A method of multi-operating condition modeling of a heavy duty gas turbine, characterized by, The specific steps are as follows: Step one, obtaining and preprocessing real-time operation data of heavy-duty gas turbine; Step two, modular modeling of heavy-duty gas turbine based on mechanism equation: based on the ideas of modular modeling and mechanism modeling, the nonlinear mathematical model of heavy-duty gas turbine is built based on the law of conservation of mass and the law of conservation of energy and thermodynamic analysis, the pressure of the combustion chamber and the temperature of the combustion chamber are selected as the state variables of the nonlinear mathematical model of heavy-duty gas turbine, the air flow and the fuel flow are selected as the control input variables of the nonlinear mathematical model of heavy-duty gas turbine, and the power generation and the exhaust temperature of the heavy-duty gas turbine are selected as the output variables of the nonlinear mathematical model of heavy-duty gas turbine to build the nonlinear mathematical model of heavy-duty gas turbine; The nonlinear mathematical model of heavy-duty gas turbine is as follows: (10) (11) wherein In the meantime , , and are white noise with mean 0, denotes the model output power generation and exhaust temperature, , denotes the measured Gaussian white noise of the output power generation with mean 0, denotes the measured Gaussian white noise of the exhaust temperature with mean 0; Pc is the combustion chamber pressure, Tc is the combustion chamber outlet temperature, K; and mair and mfuel are the air and fuel flow rates, respectively, kg / s; Cp is the outlet flue gas constant, J / kg K; Cv is the combustion chamber outlet flue gas constant volume specific heat capacity, J / kg K; Vc is the combustion chamber volume, m3; mrc is the combustion chamber outlet rated flow rate, kg / s; Prc is the combustion chamber outlet rated pressure, Pa; Tc is the combustion chamber outlet rated temperature, K; Cp is the air constant pressure specific heat capacity, J / kg K; Tatm is the ambient temperature, K; Patm is the ambient pressure, Pa; is the compressor adiabatic efficiency; LHV is the fuel lower heating value, J / kg; Prg is the heavy duty gas turbine rated power generation, is the heavy duty gas turbine adiabatic efficiency; Step three, modeling of heavy-duty gas turbine based on mechanism data fusion: based on the modeling method of quantum particle swarm optimization deep neural network, the fusion of mechanism model and data-driven method is realized, and a deep neural network model is obtained; Step four, training the deep neural network model with the optimal solution to obtain an optimized deep neural network model, and verifying the deep neural network model according to the measured data; if the error is higher than the specified upper limit of error, return to step three to retrain the deep neural network model until the error meets the requirements.

2. A method of multi-operating condition modeling of a heavy duty gas turbine as claimed in claim 1, wherein, The step one is specifically: filtering the heavy-duty gas turbine real-time operation data measured by the heavy-duty gas turbine sensor; the heavy-duty gas turbine real-time operation data includes power generation, exhaust temperature, air flow and fuel flow.

3. A method of multi-operating condition modeling of a heavy duty gas turbine as claimed in claim 2, wherein, The step two is specifically: ① Combustion chamber modeling: in the energy conversion process of the combustion chamber, the mass flow of the working medium satisfies the dynamic balance relationship to ensure the continuity of energy transmission of the heavy-duty gas turbine, which is specifically expressed as follows: (1) wherein represents the flue gas flow rate at the outlet of the combustion chamber, kg / s; is the air density, kg / m3; is the time; Combined with the law of conservation of energy and the basic law of ideal gas, we get: (2) wherein Tin represents the combustion chamber inlet air temperature, K; Hout represents the combustion chamber outlet flue gas enthalpy, J / kg; Econt represents the control volume internal energy, J / kg; Further get: (3) wherein Tairin is the combustion chamber inlet air temperature; Solving the above equation, we get: (4) Regarding the air compression process as a constant efficiency adiabatic process, we have: (5) wherein R is the ideal gas constant, J / kg-K; and the flue gas flow rate at the combustor exit is expressed as: (6) ② Heavy-duty gas turbine input module: the relationship between the fuel flow and the fuel valve opening of the heavy-duty gas turbine input, and the relationship between the air flow and the air introduction valve opening of the heavy-duty gas turbine input are regarded as first-order inertia links, which are specifically expressed as follows: (7) wherein respectively represent the time constant of the air intake valve and the fuel valve actuator, respectively represent the control commands of the air flow and the gas flow, providing target reference values for the fuel valve actuator; denotes a complex variable; • Heavy-duty gas turbine output module: heavy-duty gas turbine output is power output and exhaust temperature The relationship between heavy-duty gas turbine output and state is expressed in the following form: (8) (9)。 4. A method of multi-operating condition modeling of a heavy duty gas turbine as claimed in claim 3, wherein, The step three is specifically: ① Standardizing the heavy-duty gas turbine real-time operation data preprocessed in step one, and dividing it into training set and test set; ② Set the initial network structure and hyperparameter range of the deep neural network model, initialize the quantum particle swarm optimization algorithm particle group, including group size, maximum iteration number and search space dimension; randomly generate the position vector of each particle in the defined search space, where each position vector dimension value directly corresponds to a specific hyperparameter configuration scheme, i.e. the combination of specific values of hidden layer node number, learning rate and training rounds; ③ Use the parameters represented by the current position of the particle for deep neural network model training, and select MSE as the fitness function for evaluating the performance of the deep neural network model, which is defined as follows: (12) In the formula, and respectively represent a deep neural network model training output target and an output result, represents the number of data; (4) According to the fitness evaluation results, the individual optimal solution and the global optimal solution of each particle are determined, and the motion behavior of the particle swarm in each iteration process is defined as: (13) wherein, represents an attractor of particle motion; represents the current position of a particle; is a random number in the interval (0, 1); is the position of the particle in the last iteration; and are the individual optimal position and the population optimal position of the th particle, respectively; represents a stretching-shrinking coefficient, which is used to adjust the convergence performance of the quantum particle swarm optimization algorithm, the maximum and minimum values of the stretching-shrinking coefficient are represented as and , respectively; is a constant randomly distributed in the interval (0, 1); represents the average optimal position of the particles; represents the total number of particles; and represent the maximum number of iterations and the current number of iterations of the quantum particle swarm optimization algorithm; (5) Steps 3 and 4 are repeated until the optimal solution is obtained or the preset upper limit of the iteration number is reached, and the quantum particle swarm optimization algorithm is ended.

Images