Gas turbine performance lightweight modeling method based on thermodynamics

Abstract

The invention relates to the technical field of gas turbine performance prediction, and provides a thermodynamics-based gas turbine performance lightweight modeling method, which comprises the following steps of: 1, acquiring actual operation data of a gas turbine, and preprocessing the operation data; 2, establishing a thermodynamic model of the gas turbine; 3, establishing a gas turbine agent model; and step 4, training the gas turbine proxy model to obtain a trained gas turbine proxy model. According to the method, the influence of external environment conditions and power requirements on working condition operation conditions can be eliminated, and key gas turbine performance parameters can be quickly obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of gas turbine performance monitoring, and in particular to a thermodynamics-based gas turbine performance lightweight modeling method. Background Art

[0002] Gas turbines (GTs) are impeller-driven machines that use gas as a medium. They offer advantages such as low pollution, high thermal efficiency, and excellent peak-shaving performance. They are essential power equipment in key industries and pillar sectors of livelihood, including energy, chemical engineering, metallurgy, electricity, environmental protection, and national defense. Gas turbines are typical nonlinear, multivariable, and multi-operating-condition complex thermal power systems, characterized by extreme nonlinearity and complexity. Furthermore, gas turbines operate at high speeds, high temperatures, and even under highly corrosive conditions, requiring extreme design, precision manufacturing, and reliable operation and maintenance for all components and systems. This involves multidisciplinary, high-precision technology, requiring extensive basic research support and long-term experimental verification and experience accumulation. Real-time performance monitoring remains a major challenge in the gas turbine field. Accurately and timely prediction of gas turbine operating characteristics is crucial for stable operation, optimized control, predictive maintenance, fault diagnosis, and prognosis.

[0003] To accurately simulate the performance of gas turbines operating under all scenarios, nonlinear modeling has become a major research topic in performance simulation. Nonlinear performance simulation modeling methods primarily fall into two categories: mechanism-based and data-driven. The first category, mechanism-based modeling, describes the physical system of the gas turbine based on thermodynamic equilibrium and the law of conservation of energy. It then establishes a complete model of the gas turbine using characteristic curves or equations of the gas turbine's core components. The second category, data-driven system identification modeling, constructs a full-scenario performance simulation model for the gas turbine by analyzing the unit's actual operating data. This method does not require an in-depth understanding of the complex dynamic characteristics of the gas turbine's nonlinear system. Instead, it utilizes various statistical analysis, machine learning, and neural network methods to perform system identification and fitting of the input and output parameters of interest, establishing a simulation model with good accuracy and real-time performance. Data-driven models have become a widely used method for gas turbine performance simulation modeling, and establishing a gas turbine input-output mapping model based on operating data has become a new research approach widely accepted in the industry.

[0004] Mechanistic modeling approaches face challenges with model simplification and error accumulation. Mechanistic models based on physical laws require numerous idealized assumptions about complex systems (e.g., neglecting turbulence effects and inter-component coupling), resulting in reduced model accuracy. Furthermore, mechanistic models rely on a large number of experimentally calibrated parameters (e.g., combustion efficiency and cooling extraction coefficient). However, actual operating conditions fluctuate significantly and exhibit strong nonlinear relationships between parameters. Parameter sensitivity analysis is time-consuming and prone to local optima, impacting the model's applicability.

[0005] Data-driven modeling approaches first face a data quality bottleneck: gas turbines operate in harsh environments (high temperature, high pressure, vibration, and noise), sensor data is susceptible to interference, and sampling frequency is limited, resulting in insufficient or biased training data. Furthermore, complex models (such as long-short-term memory networks) require significant computing power, making it difficult to meet real-time control requirements. Furthermore, data-driven models rely on continuous online learning to adapt to changing operating conditions, increasing system complexity. Summary of the Invention

[0006] The present invention mainly solves the technical problems of high computational complexity, insufficient real-time performance, and weak generalization ability in the existing technology, and proposes a lightweight modeling method for gas turbine performance based on thermodynamics, which uses a neural network structure to reduce the computational load and improve real-time performance. First, a mechanism-based thermodynamic model is established as a performance benchmark, and the basic load operating conditions are "corrected" to the reference operating conditions (ISO operating conditions) to eliminate the influence of external environmental conditions and power demand on the operating conditions. Then, combining the actual operating data and simulation data after data preprocessing, a proxy model based on deep operator networks (DeepONet, DeepOperator Networks) is constructed to quickly obtain key gas turbine performance parameters, such as the corrected compressor, turbine and overall efficiency.

[0007] The present invention provides a method for lightweight modeling of gas turbine performance based on thermodynamics, comprising the following steps:

[0008] Step 1: Acquire actual operating data of the gas turbine and preprocess the operating data;

[0009] Step 2, establishing a gas turbine thermodynamic model;

[0010] Step 201: Build a framework for a gas turbine thermodynamic model; the gas turbine thermodynamic model includes a compressor module, a combustion chamber module, and a turbine module;

[0011] The compressor module uses the characteristic curve interpolation method based on the compressor characteristic curve to convert the speed n, compressor pressure ratio π c , compressor efficiency η c , compressor inlet air flow G c The parameter mapping is a nonlinear function;

[0012] The compressor characteristic curve provides the compressor reduced flow of the compressor module About compressor pressure ratio π c =p2 / p1, compressor reduced speed The functional relationship f1 of the inlet guide vane angle IGV and the compressor efficiency η of the compressor module c About compressor pressure ratio π c=p2 / p1, compressor reduced speed And the functional relationship f2 of the inlet guide vane angle IGV:

[0013]

[0014] The combustion chamber module calculates the temperature rise and pressure loss of fuel-air mixture combustion based on the chemical equilibrium equation; for the ideal gas state equation ρ=p / R g T, taking the derivative of both sides with respect to time t, and combining it with the volume differential equation, we get the differential equation group of its dynamic process as follows:

[0015]

[0016] Where: G2, G f G3 and G4 are respectively the air flow rate at the combustion chamber inlet, the fuel flow rate, and the gas flow rate at the combustion chamber outlet; p3 and T3 are respectively the pressure and temperature at the combustion chamber outlet; V is the combustion chamber volume; R g is the fuel gas constant, η B is the combustion efficiency of the combustion chamber; h2, H u and h3 are the enthalpy of air at the combustion chamber inlet, the lower calorific value of fuel, and the enthalpy of gas at the combustion chamber outlet, respectively;

[0017] Introducing the air load parameter as a correction factor, the following combustion chamber combustion efficiency formula is obtained:

[0018] η B =-5.47×10 -11 L 5 +3.98×10 -8 L 4 -8.74×10 -6 L 3 +3.00×10 -4 L 2 -4.57×10 -3 L+99.7

[0019] Where: L is the air load parameter of the combustion chamber, which is defined as follows:

[0020]

[0021] The turbine module is based on the turbine characteristic curve, through the turbine expansion ratio π T and turbine efficiency η T Associated turbine output power P T The turbine characteristic curve diagram provides the turbine equivalent flow of the turbine module About turbine expansion ratio π T =p3 / p4, turbine equivalent speed The functional relationship f3, and the turbine efficiency η of the turbine module T About turbine expansion ratio π T =p3 / p4, turbine equivalent speed Function relationship f4:

[0022]

[0023] Step 202 , performing heat balance calculation on the gas turbine thermodynamic model at the current operating point to obtain heat balance parameters;

[0024] Step 203 , the gas turbine thermodynamic model performs expected value calculation at the current operating point to obtain an expected value of the operating performance;

[0025] Step 204 , calculating and normalizing the gas turbine thermodynamic model parameters to ISO standard reference conditions to obtain a gas turbine thermodynamic model;

[0026] Step 3, establishing a gas turbine proxy model;

[0027] Step 4: Train the gas turbine proxy model to obtain a trained gas turbine proxy model.

[0028] Furthermore, the operating data includes environmental and weather data, thermal parameters, mechanical parameters, control system data and performance indicators;

[0029] The environmental and weather data include ambient temperature T0, ambient pressure p0, and ambient relative humidity H0;

[0030] The thermal parameters include compressor inlet temperature T1, compressor outlet temperature T2, compressor inlet pressure p1, compressor outlet pressure p2, and turbine outlet temperature T4;

[0031] The mechanical parameters include the rotor speed n;

[0032] The control system data includes the inlet guide vane IGV angle and the fuel distribution ratio f;

[0033] The performance indicators include power generation P G , fuel flow G f .

[0034] Furthermore, the step 1 includes the following steps 101 to 104:

[0035] Step 101, performing synchronous alignment on the operating data;

[0036] Step 102: performing outlier detection on the operating data;

[0037] Step 103: Repair missing values in abnormal operation data;

[0038] Step 104: Use the Bayesian wavelet packet denoising method to denoise the operating data.

[0039] Furthermore, step 202 includes the following steps 2021 to 2023:

[0040] Step 2021, defining the operating conditions of the gas turbine thermodynamic model;

[0041] Fixed boundary conditions: ambient temperature T0, ambient pressure p0, ambient relative humidity H0, fuel flow rate G f , speed n, load P G ;

[0042] Initialize state variables: pressure at the inlet and outlet of each component, temperature, and flow rate;

[0043] Step 2022: Establish the energy conservation equations for the compressor module, combustion chamber module, and turbine module:

[0044] The input mechanical work of the compressor module is P c =G c c pa (T2-T1);

[0045] Where G c is the compressor inlet air flow, c pa is the specific heat capacity of air at constant pressure;

[0046] The fuel chemical energy of the combustion chamber module is converted into heat energy Q B =G f ·H u ;

[0047] Where H u It is the lower calorific value of fuel;

[0048] The output power of the turbine module is P T =G T c pg (T3-T4);

[0049] Where G T is the gas flow at turbine inlet, c pg is the specific heat capacity of the gas at constant pressure;

[0050] Step 2023: Combine the energy conservation equations of the compressor module, the combustion chamber module, and the turbine module with the mass conservation equation and the momentum equation to form a nonlinear equation system:

[0051]

[0052] Where: η m for mechanical efficiency;

[0053] The state variables are iteratively solved by the Newton-Raphson method until the residual converges, and the thermal balance parameters are obtained. The thermal balance parameters are the actual compressor inlet air flow, compressor efficiency, turbine efficiency, and combustor outlet temperature at the current operating point.

[0054] Furthermore, the step 203 includes the following steps 2031 to 2033:

[0055] Step 2031, determine the design operating point and environmental conditions:

[0056] Input parameters: ambient temperature T0, ambient pressure p0, fuel lower calorific value H u , target output power P target ;

[0057] Design assumption: Combustion chamber combustion efficiency η B =100%; compressor efficiency η c The optimal value of the compressor characteristic curve is taken from the factory, and the turbine efficiency η T The optimal value is taken from the factory turbine characteristic curve;

[0058] Step 2032: Load the factory compressor characteristic curve and factory turbine characteristic curve, and perform interpolation calculation:

[0059] The compressor characteristic curve is interpolated using the function relationship f1 and the function relationship f2, and the speed n and the compressor inlet air flow G are used to calculate the compressor characteristic curve. c Find out the compressor pressure ratio π c , compressor efficiency η c ;

[0060] The turbine characteristic curve, function relationship f3 and function relationship f4 are interpolated, and the turbine expansion ratio π is used. T Find out the turbine inlet gas flow G T , turbine efficiency η T ;

[0061] Step 2033, iteratively solve the ideal parameters to obtain the expected value of the operating performance:

[0062] Solving variable: Compressor inlet air flow rate G c , compressor pressure ratio π c , turbine expansion ratio π T , fuel flow G f ;

[0063] The expected values of the operating performance are the ideal compressor inlet air flow, compressor efficiency, turbine efficiency, output power, and fuel flow in the current operation.

[0064] Furthermore, the step 2033 includes the following steps 20331 to 20335:

[0065] Step 20331, assuming the initial compressor inlet air flow rate The initial compressor pressure ratio π is obtained by interpolation based on the compressor characteristic curve c and compressor efficiency η c ;

[0066] Step 20332, calculate the compressor outlet temperature:

[0067] (k a is the air specific heat ratio);

[0068] Step 20333, calculate the fuel flow in the combustion chamber:

[0069] Step 20334, from the turbine expansion ratio π T , the turbine inlet gas flow G is obtained by interpolation based on the turbine characteristic curve T and turbine efficiency η T , calculate the turbine output power P T ;

[0070] Step 20335, adjust G c Make the turbine output work satisfy P T =P c +P target .

[0071] Furthermore, the step 204 includes the following steps 2041 to 2043:

[0072] Step 2041, solve the correction factor:

[0073]

[0074] Where x actual is the thermal balance parameter calculated in step 202, x expect The expected value calculated in step 203;

[0075] Step 2042, define the residual:

[0076]

[0077] Step 2043, iteratively adjust the parameters to obtain the gas turbine thermodynamic model:

[0078] Input the efficiency and flow correction factors DMM corresponding to the compressor and turbine into the gas turbine thermodynamic model to preliminarily adjust the fuel flow Run the model and calculate the residuals;

[0079] Termination condition: |R (k) |<∈(e.g., ∈=0.1%);

[0080] After termination, the gas turbine thermodynamic model is obtained.

[0081] Furthermore, the input of the gas turbine proxy model is a branch network and a backbone network;

[0082] The branch network: discrete sampling of the input function u(x), including gas turbine boundary conditions such as ambient temperature T0, ambient relative humidity H0, and thermodynamic parameter distribution;

[0083] The backbone network: historical data y of the data to be predicted;

[0084] The operator fusion calculation process of the gas turbine proxy model is as follows:

[0085] The branch network receives discrete samples of the input function u(x), extracts high-dimensional features through the fully connected layer, and outputs a low-dimensional latent vector b = [B1, B2, B3];

[0086] The backbone network receives the historical data y of the data to be predicted and generates a latent vector t = [T1, T2, T3] of the same dimension as b;

[0087] Multiply the output vectors of the branch network and the trunk network element by element and then calculate the inner product operation G(u)(y)=<b,t> ;

[0088] The output of the gas turbine proxy model is a predicted value; the predicted value includes the corrected calculated compressor efficiency η c.ISO , turbine efficiency η T,ISO , whole machine thermal efficiency HR ISO , whole machine power P ISO .

[0089] The present invention provides a lightweight modeling method for gas turbine performance based on thermodynamics. By establishing a time-series mapping relationship between multiple parameters of the gas turbine, a lightweight proxy model is constructed to achieve rapid simulation of performance under all operating conditions and prediction of component attenuation trends, breaking through the bottleneck of traditional mechanism models that are time-consuming and difficult to update in real time. Based on the thermodynamic degradation mechanism of the gas turbine, a performance simulation correction model based on thermodynamics is constructed. The actual operating parameters are normalized to ISO standard reference conditions through iterative calculations such as thermal balance calculation and expected value calculation, thereby achieving high-precision decoupling correction of the effects of environmental and power coupling under all operating conditions. Sensitive parameters that are strongly correlated with performance attenuation are extracted from multi-dimensional sensor data as input parameters of the branch network to suppress non-correlated noise interference. The core adopts the deep operator network architecture of DeepONET with branch-trunk dual channels to capture the dependency relationship of aerodynamically sensitive parameters (such as compressor inlet air flow and compressor outlet pressure), and combines the corrected historical data of components and overall efficiency as the input of the backbone network to improve prediction accuracy. The generalization characteristics of neural operators are used to achieve rapid prediction of nonlinear responses under all operating conditions.

[0090] The present invention can predict key indicators such as gas turbine efficiency and power at millisecond speeds, support online performance optimization and maintenance decisions, and subsequently provide a high-precision real-time model for intelligent operation and maintenance of gas turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 This is a flow chart for implementing the thermodynamics-based lightweight modeling method for gas turbine performance provided by the present invention;

[0092] Figure 2 It is a schematic diagram of the component-level gas turbine thermodynamic model;

[0093] Figure 3 This is a diagram of the Simulink modeling of the Newton-Raphson solution for the steady-state model;

[0094] Figure 4 This is a schematic diagram of the DeepONet backbone network and its branch network structure;

[0095] Figure 5 This is the architecture diagram of the gas turbine proxy model based on DeepONet;

[0096] Figure 6 It is a schematic diagram of gas turbine operating data;

[0097] Figure 7 This is a schematic diagram showing the impact of different numbers of nodes on the results during the training process;

[0098] Figure 8 This is a schematic diagram of the impact of different activation functions and Dropout rates on the results during the training process;

[0099] Figure 9 This is a schematic diagram of compressor corrected efficiency prediction;

[0100] Figure 10 It is a schematic diagram of turbine corrected efficiency prediction;

[0101] Figure 11 This is a schematic diagram of the prediction of the corrected heat rate of the whole machine;

[0102] Figure 12 It is a schematic diagram of the corrected power prediction of the whole machine. DETAILED DESCRIPTION

[0103] To make the technical problems solved, the technical solutions adopted, and the technical effects achieved by the present invention more clearly apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, rather than all of the contents.

[0104] like Figure 1 As shown, an embodiment of the present invention provides a method for lightweight modeling of gas turbine performance based on thermodynamics, including the following process:

[0105] Step 1: Acquire actual operating data of the gas turbine and preprocess the operating data.

[0106] The operating data includes: environmental and weather data, thermal parameters, mechanical parameters, control system data and performance indicators.

[0107] The environmental and weather data include environmental temperature T0, environmental pressure p0, and environmental relative humidity H0.

[0108] The thermal parameters include compressor inlet temperature T1, compressor outlet temperature T2, compressor inlet pressure p1, compressor outlet pressure p2, and turbine outlet temperature T4.

[0109] The mechanical parameters include the rotor speed n.

[0110] The control system data includes adjustment signals such as the inlet guide vane IGV angle and the fuel distribution ratio f.

[0111] The performance indicators include power generation P G , fuel flow G f .

[0112] The step 1 includes the following steps 101 to 104:

[0113] Step 101: Synchronize and align the gas turbine operating data.

[0114] To address the asynchronous sampling problem of multi-source sensors (temperature, pressure, flow, etc.), sliding window interpolation or dynamic time warping (DTW) is used to unify the timestamp.

[0115] Step 102: Perform outlier detection on the operating data.

[0116] Use machine learning models (isolation forest, AutoEncoder reconstruction error) to identify data anomalies and distinguish real faults from noise interference.

[0117] Step 103: Repair missing values in abnormal operation data.

[0118] Based on the correlation of working conditions, random forest regression prediction is used to fill in the randomly missing data to avoid the steady-state deviation of traditional linear interpolation.

[0119] Step 104: Use the Bayesian wavelet packet denoising method to denoise the operating data.

[0120] The non-stationary vibration signal is decomposed into high-frequency details and low-frequency approximate components according to the optimal basis (entropy criterion) to refine the noise distribution characteristics.

[0121] The posterior probability distribution of each subband coefficient is estimated based on Markov Chain Monte Carlo (MCMC) and the shrinkage threshold is adaptively determined to retain the effective impulse components while suppressing high-frequency noise.

[0122] Step 2: Establish a gas turbine thermodynamic model.

[0123] Step 201: Build a framework for a gas turbine thermodynamic model.

[0124] like Figure 2 As shown in the figure, a single-shaft gas turbine mainly consists of three subsystems: compressor, combustor, and turbine. In addition, a gas turbine usually includes some auxiliary components, such as the intake system, exhaust system, and secondary air system, which are used to cool and protect the gas turbine components.

[0125] Modular modeling of gas turbines decouples complex systems into independent components (such as compressors, combustion chambers, etc.), supports parallel development and parametric debugging, and significantly improves model reusability and maintenance efficiency; and establishes a gas turbine thermodynamic model for component-level performance simulation.

[0126] Each component of the gas turbine thermodynamic model is based on the law of conservation of energy and mass, and is a mechanism model. The gas turbine thermodynamic model includes: a compressor module, a combustion chamber module and a turbine module.

[0127] The compressor module uses the characteristic curve interpolation method based on the compressor characteristic curve to convert the speed n, compressor pressure ratio π c, compressor efficiency η c , compressor inlet air flow G c The parameter mapping is a nonlinear function; the compressor characteristic curve provides the compressor reduced flow of the compressor module About compressor pressure ratio π c =p2 / p1, compressor reduced speed The functional relationship f1 of the inlet guide vane angle IGV and the compressor efficiency η of the compressor module c About compressor pressure ratio π c =p2 / p1, compressor reduced speed And the functional relationship f2 of the inlet guide vane angle IGV:

[0128]

[0129] The compressor reduced flow rate and compressor efficiency can be obtained through the functional relationship of compressor pressure ratio, compressor reduced speed and IGV. By determining only three of the five parameters, the working state of the compressor can be fully determined.

[0130] The combustion chamber module calculates the temperature rise and pressure loss of fuel-air mixture combustion based on the chemical equilibrium equation. The key to combustion chamber modeling is the trend of combustion chamber outlet temperature change and combustion chamber pressure loss change. Combining the principles of conservation of mass and energy, for the ideal gas state equation ρ=p / R g T, and take the derivative of both sides with respect to time t. Combined with the volume differential equation, the differential equation system of its dynamic process is expressed as follows:

[0131]

[0132] Where: G2, G f G3 and G4 are respectively the air flow rate at the combustion chamber inlet, the fuel flow rate, and the gas flow rate at the combustion chamber outlet; p3 and T3 are respectively the pressure and temperature at the combustion chamber outlet; V is the combustion chamber volume; R g is the fuel gas constant, η B is the combustion efficiency of the combustion chamber; h2, H u and h3 are the enthalpy of air at the combustion chamber inlet, the lower calorific value of fuel and the enthalpy of gas at the combustion chamber outlet, respectively.

[0133] Introducing the air load parameter as a correction factor, the following combustion chamber combustion efficiency formula is obtained:

[0134] η B =-5.47×10 -11 L 5 +3.98×10 -8 L 4 -8.74×10 -6 L 3+3.00×10 -4 L 2 -4.57×10 -3 L+99.7

[0135] Where: L is the air load parameter of the combustion chamber, which is defined as follows:

[0136]

[0137] The turbine module is based on the turbine characteristic curve, through the turbine expansion ratio π T and turbine efficiency η T Associated turbine output power P T , and integrates the effect of cooling air flow on work. Similar to the compressor module, the turbine characteristic curve provides the turbine equivalent flow of the turbine module About turbine expansion ratio π T =p3 / p4, turbine equivalent speed The functional relationship f3, and the turbine efficiency η of the turbine module T About turbine expansion ratio π T =p3 / p4, turbine equivalent speed Function relationship f4:

[0138]

[0139] Based on the gas turbine thermodynamic model framework, a gas turbine thermodynamic simulation model is constructed in the Simulink simulation platform. Through numerical solution, the key parameters under typical operating conditions (including but not limited to: compressor outlet temperature, compressor outlet pressure, compressor inlet air flow, combustor outlet temperature, combustor outlet flow, gas turbine output power, compressor efficiency, turbine efficiency, thermal efficiency, etc.) are obtained, and finally a complete gas turbine thermodynamic simulation model is established.

[0140] In the Simulink platform, a gas turbine thermodynamic model is constructed using module components (Library) and custom S-Functions. The compressor and turbine characteristic curve data tables are called in real time through the MATLAB Function module, supporting dynamic interpolation calculations. The input and output of each subsystem (compressor, combustor, turbine) are integrated through the signal bus to ensure the consistency and scalability of variable transmission. Steady-state modeling must meet the equilibrium conditions of the simultaneous equations of conservation of mass, conservation of momentum, and conservation of energy.

[0141] Step 202: The gas turbine thermodynamic model performs heat balance calculation at the current operating point to obtain heat balance parameters. Step 202 includes the following steps 2021 to 2023:

[0142] The heat balance calculation of the current operating point refers to the calculation of all energy inputs (such as fuel chemical energy Q B ), output power P G A systematic analysis of the heat loss is conducted by establishing an energy balance equation for the flow of working fluids (air and gas). Based on the known measurable parameters (pressure, temperature, flow) at the inlet and outlet of each component, the calculation process for unknown parameters or measurement deviation parameters is solved. In the Simulink simulation environment, the heat balance calculation and solution are achieved through the following steps:

[0143] Step 2021: Define the operating conditions of the gas turbine thermodynamic model.

[0144] Fixed boundary conditions: ambient temperature T0, ambient pressure p0, ambient relative humidity H0, fuel flow rate G f , speed n, load P G wait.

[0145] Initialize state variables: inlet pressure, outlet pressure, temperature, and flow rate of each component (such as compressor inlet temperature T1, compressor outlet temperature T2, compressor inlet pressure p1, compressor outlet pressure p2, and turbine outlet temperature T4).

[0146] Step 2022: Establish the energy conservation equations for the compressor module, combustion chamber module, and turbine module:

[0147] Energy flow by component:

[0148] The input mechanical work of the compressor module is P c =G c c pa (T2-T1);

[0149] Where: G c is the compressor inlet air flow, c pa is the specific heat of air at constant pressure.

[0150] The fuel chemical energy of the combustion chamber module is converted into heat energy Q B =G f ·H u ;

[0151] Where: H u It is the lower calorific value of fuel.

[0152] The output power of the turbine module is P T =G T c pg (T3-T4);

[0153] Where: G Tis the gas flow at turbine inlet, c pg is the specific heat capacity of the gas at constant pressure.

[0154] Step 2023: Combine the energy conservation equations of the compressor module, combustor module, and turbine module with the mass conservation equation (flow continuity) and momentum equation (pressure ratio-flow characteristics) to form a nonlinear equation system:

[0155]

[0156] Where: η m For mechanical efficiency.

[0157] The state variables are solved iteratively by the Newton-Raphson method until the residual converges and the thermal balance parameters are obtained; the Newton-Raphson solution model for thermal balance calculation is as follows: Figure 3 The heat balance parameters are the actual compressor inlet air flow, compressor efficiency, turbine efficiency, and combustor outlet temperature at the current operating point.

[0158] Step 203: The gas turbine thermodynamic model performs expected value calculation at the current operating point to obtain the expected value of the operating performance. Step 203 includes the following steps 2031 to 2033:

[0159] The expected value calculation refers to the determination of the theoretical performance parameters (such as flow, pressure, temperature, output power, efficiency, etc.) that the unit should achieve under specific operating conditions based on the established gas turbine thermodynamic model and the compressor and turbine design characteristic curves provided by the manufacturer through numerical solution. These theoretical parameters represent the optimal performance of the unit under standard health conditions and ideal operating conditions. In the Simulink simulation environment, the expected value calculation solution is implemented through the following steps:

[0160] Step 2031, determine the design operating point and environmental conditions:

[0161] Input parameters: ambient temperature T0, ambient pressure p0, fuel lower calorific value H u , target output power P target .

[0162] Design assumption: Combustion chamber combustion efficiency η B =100% (complete combustion); compressor efficiency η c The optimal value taken from the factory compressor characteristic curve, turbine efficiency η T Optimum value taken from the factory turbine characteristic diagram.

[0163] Step 2032: Load the factory compressor characteristic curve and factory turbine characteristic curve, and perform interpolation calculation:

[0164] The compressor characteristic curve is interpolated using the function relationship f1 and the function relationship f2, and the speed n and the compressor inlet air flow G are used to calculate the compressor characteristic curve. c Find out the compressor pressure ratio π c , compressor efficiency η c ;

[0165] Turbine characteristic curve: through the turbine expansion ratio π T Find out the turbine inlet gas flow G T , turbine efficiency η T .

[0166] Step 2033, iteratively solve the ideal parameters to obtain the expected value of the operating performance:

[0167] Solving variable: Compressor inlet air flow rate G c , compressor pressure ratio π c , turbine expansion ratio π T , fuel flow G f .

[0168] The expected values of the operating performance are the ideal compressor inlet air flow, compressor efficiency, turbine efficiency, output power, and fuel flow in the current operation.

[0169] Iterative algorithm: Based on the alternating correction steps of the characteristic curve and the theoretical equation, step 2033 includes the following steps 20331 to 20335:

[0170] Step 20331, assuming the initial compressor inlet air flow rate The initial compressor pressure ratio π is obtained by interpolation based on the compressor characteristic curve c and compressor efficiency η c ;

[0171] Step 20332, calculate the compressor outlet temperature:

[0172] (k a is the air specific heat ratio);

[0173] Step 20333, calculate the fuel flow in the combustion chamber:

[0174] Step 20334, from the turbine expansion ratio π T , the turbine inlet gas flow G is obtained by interpolation based on the turbine characteristic curve T and turbine efficiency η T , calculate the turbine output power P T ;

[0175] Step 20335, adjust G c Make the turbine output work satisfy P T=P c +P target .

[0176] Step 204 : Calculate and normalize the parameters of the gas turbine thermodynamic model to ISO standard reference conditions to obtain the gas turbine thermodynamic model.

[0177] The purpose of normalizing to ISO standard reference conditions is to eliminate interference from environmental variables and convert actual operating data into equivalent values under standard conditions (typically 15°C, 1 atm, 60% relative humidity) to facilitate performance evaluation and benchmarking. The following is an iterative calculation step for normalizing to ISO standard reference conditions based on the above expected value and the difference between the thermal balance ratio:

[0178] Step 2041, solve the correction factor:

[0179]

[0180] Where: x actual is the thermal balance parameter calculated in step 202, x expect is the expected value calculated in step 203.

[0181] Step 2042, define the residual:

[0182]

[0183] Step 2043, iteratively adjust the parameters to obtain the gas turbine thermodynamic model:

[0184] Input the efficiency and flow correction factors DMM corresponding to the compressor and turbine into the gas turbine thermodynamic model to preliminarily adjust the fuel flow Run the model and calculate the residuals.

[0185] Termination condition: |R (k) |<∈(e.g., ∈=0.1%).

[0186] After termination, the gas turbine thermodynamic model is obtained.

[0187] Step 3: Establish a gas turbine proxy model.

[0188] DeepONET (Deep Operator Network) is a deep learning-based proxy model construction method designed for efficient approximation of input-output operators of complex physical systems. Its core idea is to extend the scalar mapping capability of traditional neural networks to function space mapping, and to achieve high-precision modeling of high-dimensional, nonlinear, and multi-physics field coupled systems by decoupling the nonlinear relationship between input functions and prediction variables. Figure 4As shown in Figure 2, DeepONET adopts a branch-trunk dual-channel structure to process the mapping of input function and prediction component respectively.

[0189] The present invention establishes a gas turbine proxy model based on DeepONET, specifically, Figure 5 As shown, the input of the gas turbine proxy model is a branch network and a backbone network;

[0190] The branch network (Branch Net): discrete sampling of the input function u(x), including gas turbine boundary conditions such as ambient temperature T0, ambient relative humidity H0, and thermodynamic parameter distribution such as compressor pressure ratio π c =p2 / p1, compressor inlet air flow G c , fuel flow G f , operating point load P G .

[0191] The trunk network (Trunk Net): the historical data y of the data to be predicted (such as the corrected calculated compressor efficiency η c，ISO , turbine efficiency η T，ISO , whole machine thermal efficiency HR ISO , whole machine power P ISO )

[0192] The operator fusion calculation process of the gas turbine proxy model is as follows:

[0193] The branch network receives discrete samples of the input function u(x), extracts high-dimensional features through the fully connected layer, and outputs a low-dimensional latent vector b = [B1, B2, B3];

[0194] The backbone network receives the historical data y of the data to be predicted and generates a latent vector t = [T1, T2, T3] of the same dimension as b;

[0195] Multiply the output vectors of the branch network Branch and the trunk network element by element and then calculate the inner product operation G(u)(y)=<b,t> .

[0196] The output of the gas turbine proxy model is a predicted value; the predicted value includes the corrected calculated compressor efficiency η c，ISO , turbine efficiency η T，ISO , whole machine thermal efficiency HR ISO , whole machine power P ISO .

[0197] Step 4: Train the gas turbine proxy model to obtain a trained gas turbine proxy model.

[0198] The model uses a cross-validation strategy, splitting the dataset into a training set (70% for model weight updates), a validation set (20% for monitoring model generalization and triggering early stopping and learning rate decay), and a test set (10% for final evaluation of model performance to ensure no data leakage). During the training phase, weights are iteratively updated using the Adam optimizer, with the training set loss calculated and backpropagated after each round. During the validation phase, parameters are frozen, and the mean squared error (MSE) is monitored to assess generalization. A learning rate decay strategy (initial value 0.001, decay factor 0.5) is used for dynamic adjustment of the learning rate.

[0199] The trained gas turbine proxy model is used to be integrated into the gas turbine digital twin system to output efficiency and power parameters in real time, monitor performance degradation, and support operation and maintenance decisions.

[0200] The present invention is illustrated below by way of example:

[0201] The input variables are ambient temperature, ambient relative humidity, compressor inlet flow, gas flow, and measured gas turbine output power. Figure 6 This is the result of data cleaning from March 1, 2010 to April 27, 2010.

[0202] The proxy model uses a cross-validation strategy, splitting the dataset into training, validation, and test sets at a ratio of 7:2:1. During the training phase, the weights are iteratively updated using the Adam optimizer, with the training set loss calculated and backpropagated for each round. During the validation phase, the parameters are frozen and the mean squared error (MSE) is monitored to assess generalization ability. A learning rate decay strategy (initial value 0.001, decay factor 0.5) is used. The effect of the number of hidden layer nodes on the results is tested, as shown in the following example. Figure 7 As shown. When the number of nodes increases to 128, the MSE drops to 0.0422%, but when it increases to 150, the MSE rises back to 0.0587% due to overfitting. Comparing eLU and ReLU functions, it is found that the MSE of the eLU function test set is lower. Preferably, the eLU activation strategy is adopted. After adding the Dropout layer (ratio 0.1-0.5), the overfitting phenomenon is significantly suppressed (the training and validation MSE are reduced). Figure 8 As shown, a too low ratio (<0.2) leads to underfitting (MSE increases). Preferably, the dropout rate is dynamically adjusted (0.3-0.5) based on the L2 norm of the inter-layer weights, taking into account both regularization strength and information integrity.

[0203] The gas turbine proxy model built based on DeepONET shows excellent time series dynamic capture capability and generalization performance on the test dataset. The data includes the corrected compressor efficiency ( Figure 9 ), turbine efficiency ( Figure 10 ), heat rate ( Figure 11 ), the whole machine output power ( Figure 12 ) Four indicators, it can be seen from the figure that the results predicted by DeepONET are consistent with the actual results and the error is small.

[0204] The DeepONET model compares the four prediction variables of gas turbines with traditional deep learning methods, including nonlinear autoregressive exogenous models (NARX), long short-term memory networks (LSTM), and gated recurrent units (GRU). The comparison results are shown in Table 1 below. The prediction error of DeepONET is significantly lower than that of traditional models, and the prediction time is also around 10ms, which meets the needs of real-time monitoring. In summary, the DeepONET agent model outperforms traditional data-driven methods in terms of accuracy, dynamic response, and engineering applicability, providing an efficient and reliable solution for gas turbine digital twins, health management, and control optimization. In the future, it can be integrated with multi-model data through transfer learning to further expand its cross-platform generalization capabilities.

[0205] Table 1 Comparison results between DeepONET model and traditional deep learning algorithms

[0206]

[0207]

[0208] The present invention targets sensor signal anomalies in high-temperature, high-pressure, and noisy environments, and performs data preprocessing operations including outlier processing, missing value filling, and Bayesian wavelet packet denoising to improve data quality and model generalization capabilities. By real-time acquisition of environmental parameters (temperature, relative humidity, pressure) and grid load instructions, a performance simulation correction model based on thermodynamics is constructed. The actual operating parameters are normalized to ISO standard reference conditions through iterative calculations such as heat balance calculation and expected value calculation, achieving high-precision decoupling correction of the effects of environmental and power coupling under all operating conditions, and supporting gas turbine performance degradation assessment and optimization control. The gas turbine proxy modeling method based on the deep operator network (DeepONet) decouples the input function and variable mapping relationship through a branch-trunk dual-channel architecture, and uses the generalization characteristics of neural operators to achieve rapid prediction of nonlinear responses under all operating conditions. Compared with traditional proxy models, it has strong generalization capabilities and migration adaptability, and can accurately approximate gas turbine performance simulation with a small number of samples. At the same time, it supports millisecond-level real-time simulation, providing an efficient computing kernel for dynamic optimization and digital twin construction of complex systems.

[0209] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications to the technical solutions described in the above embodiments, or equivalent replacement of some or all of the technical features therein, do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for lightweight modeling of gas turbine performance based on thermodynamics, characterized in that: The following processes are included: Step 1: Acquire actual operating data of the gas turbine and preprocess the operating data; Step 2, establishing a gas turbine thermodynamic model; Step 201, building a framework of a gas turbine thermodynamic model; The gas turbine thermodynamic model includes: a compressor module, a combustion chamber module and a turbine module; The compressor module uses the characteristic curve interpolation method based on the compressor characteristic curve to convert the speed n, compressor pressure ratio π c , compressor efficiency η c , compressor inlet air flow G c The parameter mapping is a nonlinear function; The compressor characteristic curve provides the compressor reduced flow of the compressor module About compressor pressure ratio π c =p2 / p1, compressor reduced speed The functional relationship f1 of the inlet guide vane angle IGV and the compressor efficiency η of the compressor module c About compressor pressure ratio π c =p2 / p1, compressor reduced speed And the functional relationship f2 of the inlet guide vane angle IGV: The combustion chamber module calculates the temperature rise and pressure loss of fuel-air mixture combustion based on the chemical equilibrium equation; for the ideal gas state equation ρ=p / R g T, taking the derivative of both sides with respect to time t, and combining it with the volume differential equation, we get the differential equation group of its dynamic process as follows: Where: G2, G f G3 and G4 are respectively the air flow rate at the combustion chamber inlet, the fuel flow rate, and the gas flow rate at the combustion chamber outlet; p3 and T3 are respectively the pressure and temperature at the combustion chamber outlet; V is the combustion chamber volume; R g is the fuel gas constant, η B is the combustion efficiency of the combustion chamber; h2, H u and h3 are the enthalpy of air at the combustion chamber inlet, the lower calorific value of fuel, and the enthalpy of gas at the combustion chamber outlet, respectively; Introducing the air load parameter as a correction factor, the following combustion chamber combustion efficiency formula is obtained: η B =-5.47×10 -11 L 5 +3.98×10 -8 L 4 -8.74×10 -6 L 3 +3.00×10 -4 L 2 -4.57×10 -3 L+99.7 Where: L is the air load parameter of the combustion chamber, which is defined as follows: The turbine module is based on the turbine characteristic curve, through the turbine expansion ratio π T and turbine efficiency η T Associated turbine output power P T The turbine characteristic curve diagram provides the turbine equivalent flow of the turbine module About turbine expansion ratio π T =p3 / p4, turbine equivalent speed The functional relationship f3, and the turbine efficiency η of the turbine module T About turbine expansion ratio π T =p3 / p4, turbine equivalent speed Function relationship f4: Step 202 , performing heat balance calculation on the gas turbine thermodynamic model at the current operating point to obtain heat balance parameters; Step 203 , the gas turbine thermodynamic model performs expected value calculation at the current operating point to obtain an expected value of the operating performance; Step 204 , calculating and normalizing the gas turbine thermodynamic model parameters to ISO standard reference conditions to obtain the gas turbine thermodynamic model; Step 3, establishing a gas turbine proxy model; Step 4: Train the gas turbine proxy model to obtain a trained gas turbine proxy model.

2. The method for lightweight modeling of gas turbine performance based on thermodynamics according to claim 1, characterized in that: The operating data includes environmental and weather data, thermal parameters, mechanical parameters, control system data and performance indicators; The environmental and weather data include ambient temperature T0, ambient pressure p0, and ambient relative humidity H0; The thermal parameters include compressor inlet temperature T1, compressor outlet temperature T2, compressor inlet pressure p1, compressor outlet pressure p2, and turbine outlet temperature T4; The mechanical parameters include the rotor speed n; The control system data includes the inlet guide vane IGV angle and the fuel distribution ratio f; The performance indicators include power generation P G , fuel flow G f .

3. The method for lightweight modeling of gas turbine performance based on thermodynamics according to claim 2, characterized in that: The step 1 includes the following steps 101 to 104: Step 101, performing synchronous alignment on the operating data; Step 102: performing outlier detection on the operating data; Step 103: Repair missing values in abnormal operation data; Step 104: Use the Bayesian wavelet packet denoising method to denoise the operating data.

4. The method for lightweight modeling of gas turbine performance based on thermodynamics according to claim 1, characterized in that: The step 202 includes the following steps 2021 to 2023: Step 2021, defining the operating conditions of the gas turbine thermodynamic model; Fixed boundary conditions: ambient temperature T0, ambient pressure p0, ambient relative humidity H0, fuel flow rate G f , speed n, load P G ; Initialize state variables: pressure at the inlet and outlet of each component, temperature, and flow rate; Step 2022: Establish the energy conservation equations for the compressor module, combustion chamber module, and turbine module: The input mechanical work of the compressor module is P c =G c c pa (T2-T1); Where G c is the compressor inlet air flow, c pa is the specific heat capacity of air at constant pressure; The fuel chemical energy of the combustion chamber module is converted into heat energy Q B =G f ·H u ; Where H u It is the lower calorific value of fuel; The output power of the turbine module is P T =G T c pg (T3-T4); Where G T is the gas flow at turbine inlet, c pg is the specific heat capacity of gas at constant pressure; Step 2023: Combine the energy conservation equations of the compressor module, the combustion chamber module, and the turbine module with the mass conservation equation and the momentum equation to form a nonlinear equation system: Where: η m for mechanical efficiency; The state variables are iteratively solved by the Newton-Raphson method until the residual converges, and the thermal balance parameters are obtained. The thermal balance parameters are the actual compressor inlet air flow, compressor efficiency, turbine efficiency, and combustor outlet temperature at the current operating point.

5. The method for lightweight modeling of gas turbine performance based on thermodynamics according to claim 4, characterized in that: The step 203 includes the following steps 2031 to 2033: Step 2031, determine the design operating point and environmental conditions: Input parameters: ambient temperature T0, ambient pressure p0, fuel lower calorific value H u , target output power P target ; Design assumption: Combustion chamber combustion efficiency η B =100%; compressor efficiency η c The optimal value of the compressor characteristic curve is taken from the factory, and the turbine efficiency η T The optimal value is taken from the factory turbine characteristic curve; Step 2032: Load the factory compressor characteristic curve and factory turbine characteristic curve, and perform interpolation calculation: The compressor characteristic curve is interpolated using the function relationship f1 and the function relationship f2, and the speed n and the compressor inlet air flow G are used to calculate the compressor characteristic curve. c Find out the compressor pressure ratio π c , compressor efficiency η c ; The turbine characteristic curve, function relationship f3 and function relationship f4 are interpolated, and the turbine expansion ratio π is used. T Find out the turbine inlet gas flow G T , turbine efficiency η T ; Step 2033, iteratively solve the ideal parameters to obtain the expected value of the operating performance: Solving variable: Compressor inlet air flow rate G c , compressor pressure ratio π c , turbine expansion ratio π T , fuel flow G f ; The expected values of the operating performance are the ideal compressor inlet air flow, compressor efficiency, turbine efficiency, output power, and fuel flow in the current operation.

6. The method for lightweight modeling of gas turbine performance based on thermodynamics according to claim 5, characterized in that: The step 2033 includes the following steps 20331 to 20335: Step 20331, assuming the initial compressor inlet air flow rate The initial compressor pressure ratio π is obtained by interpolation based on the compressor characteristic curve c and compressor efficiency η c ; Step 20332, calculate the compressor outlet temperature: (k a is the air specific heat ratio); Step 20333, calculate the fuel flow in the combustion chamber: Step 20334, from the turbine expansion ratio π T , the turbine inlet gas flow G is obtained by interpolation based on the turbine characteristic curve T and turbine efficiency η T , calculate the turbine output power P T ; Step 20335, adjust G c Make the turbine output work satisfy P T =P c +P target .

7. The method for lightweight modeling of gas turbine performance based on thermodynamics according to claim 6, characterized in that: The step 204 includes the following steps 2041 to 2043: Step 2041, solve the correction factor: Where x actual is the thermal balance parameter calculated in step 202, x expect The expected value calculated in step 203; Step 2042, define the residual: Step 2043, iteratively adjust the parameters to obtain the gas turbine thermodynamic model: Input the efficiency and flow correction factors DMM corresponding to the compressor and turbine into the gas turbine thermodynamic model to preliminarily adjust the fuel flow Run the model and calculate the residuals; Termination condition: |R (k) |<∈(e.g., ∈=0.1%); After termination, the gas turbine thermodynamic model is obtained.

8. The method for lightweight modeling of gas turbine performance based on thermodynamics according to claim 1, characterized in that: The input of the gas turbine proxy model is a branch network and a backbone network; The branch network: discrete sampling of the input function u(x), including gas turbine boundary conditions such as ambient temperature T0, ambient relative humidity H0, and thermodynamic parameter distribution; The backbone network: historical data y of the data to be predicted; The operator fusion calculation process of the gas turbine proxy model is as follows: The branch network receives discrete samples of the input function u(x), extracts high-dimensional features through a fully connected layer, and outputs a low-dimensional latent vector b = [B1, B2, B3]; The backbone network receives the historical data y of the data to be predicted and generates a latent vector t = [T1, T2, T3] of the same dimension as b; Multiply the output vectors of the branch network and the trunk network element by element and then calculate the inner product operation G(u)(y)=<b,t> ; The output of the gas turbine proxy model is a predicted value; the predicted value includes the corrected calculated compressor efficiency η c.ISO , turbine efficiency η T,ISO , whole machine thermal efficiency HR ISO , whole machine power P ISO .

Images