Neural network model construction method and device for gas turbine performance prediction

Abstract

This application provides a method for constructing a neural network model for gas turbine performance prediction, which can be applied to the field of artificial intelligence technology. The method includes: acquiring an experimental dataset and a physical model dataset of the gas turbine; training a first physical enhancement model based on the physical model dataset to determine an intermediate physical enhancement model; processing the experimental dataset using the intermediate physical enhancement model to obtain first gas turbine operating parameter prediction data; processing the experimental dataset using a second physical enhancement model to obtain second gas turbine operating parameter prediction data; obtaining a target loss based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and labeled operating parameters from the experimental dataset; and training the second physical enhancement model based on the target loss to obtain a target physical enhancement model for gas turbine performance prediction. This application also provides a device for constructing a neural network model for gas turbine performance prediction.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, specifically to the field of machine learning technology, and more specifically to a method, apparatus, device, medium, and program product for constructing a neural network model for predicting the performance of gas turbines. Background Technology

[0002] In practical engineering, many devices involve multi-physics coupling, exhibiting strong nonlinearity and making performance prediction challenging. Traditional prediction methods fall into two categories: physical model-based and data-driven. Physical model prediction requires solving partial differential equations using the finite element method or finite difference method, which is computationally time-consuming and complex. Data-driven methods rely on data quality and require a large amount of experimental data, but obtaining experimental data for strongly nonlinear devices is difficult. While Physics-Informed Neural Networks (PINNs) combine physical laws with neural networks and are suitable for solving partial differential equations in complex engineering problems, not all practical problems involve partial differential equations, or the coupling of multiple physical fields can lead to complex partial differential equations. For example, the application of PINNs is limited in the field of gas turbine performance prediction. Therefore, a new neural network model construction method is urgently needed to achieve high-precision, fast solutions and small-sample predictions for the performance of strongly nonlinear devices. Summary of the Invention

[0003] In view of the above problems, this application provides a method, apparatus, device, medium and program product for building neural network models for gas turbine performance prediction, which can accelerate the convergence speed of neural networks, improve the accuracy of performance prediction, and save the number of samples used.

[0004] According to the first aspect of this application, a method for constructing a neural network model for predicting gas turbine performance is provided, comprising:

[0005] The experimental dataset and physical model dataset of the gas turbine are obtained. The physical model dataset is determined based on the processing of operating state data related to the gas turbine by the physical model.

[0006] The first physics augmentation model is trained based on the physics model dataset to determine the intermediate physics augmentation model;

[0007] The experimental dataset is processed using the intermediate physics enhancement model to obtain predicted operating parameters for the first gas turbine.

[0008] The experimental dataset was processed using a second physical enhancement model to obtain predicted data for the second gas turbine operating parameters;

[0009] The target loss is obtained based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset; and

[0010] The second physics augmentation model is trained based on the target loss to obtain a target physics augmentation model for gas turbine performance prediction.

[0011] According to an embodiment of this application, obtaining the target loss based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset includes:

[0012] Physical losses are determined based on the first gas turbine operating parameter prediction data and the second gas turbine operating parameter prediction data;

[0013] The experimental loss is determined based on the predicted operating parameters of the second gas turbine and the labeled operating parameters of the experimental dataset; and

[0014] The target loss is determined based on the physical loss, the experimental loss, and the physical loss weight.

[0015] According to an embodiment of this application, training the first physics augmentation model based on the physics model dataset to determine the first intermediate physics augmentation model includes:

[0016] The physical model dataset is input into the first physical enhancement model, and the third gas turbine operating parameter prediction data is output; and

[0017] The first physical augmentation model is trained based on the predicted operating parameters of the third gas turbine and the labeled operating parameters of the physical model dataset.

[0018] According to embodiments of this application, the method further includes:

[0019] The experimental dataset was preprocessed using box plots to obtain the preprocessed experimental dataset.

[0020] According to embodiments of this application, the method further includes:

[0021] The hyperparameters of a physical augmentation neural network model are optimized based on the Bayesian optimization algorithm. The hyperparameters include the number of hidden layers, the number of neurons, the learning rate, the batch size, and the weights of the intermediate physical augmentation model.

[0022] According to embodiments of this application, the method further includes:

[0023] The second physics augmentation model is evaluated using a validation set after a preset training period, and the loss function value on the validation set is calculated.

[0024] If it is determined that the current validation loss is lower than the optimal validation loss, then update the optimal validation loss and save the parameters of the current model; and

[0025] If the loss value on the validation set does not decrease within a preset number of consecutive iterations, the early stopping condition is triggered, and model training is stopped.

[0026] A second aspect of this application provides a method for predicting the performance of a gas turbine based on a physical enhancement model, the method comprising:

[0027] Collect relevant operating status data of the gas turbine;

[0028] Preprocessing of the relevant operating status data of the gas turbine; and

[0029] The preprocessed gas turbine operating status data is input into a pre-trained physical augmentation model to output predicted gas turbine operating parameters, wherein the physical augmentation model is trained according to the method described in the first aspect.

[0030] A third aspect of this application provides a neural network model building apparatus for predicting gas turbine performance, comprising:

[0031] The data acquisition module is used to acquire the experimental dataset and physical model dataset of the gas turbine. The physical model dataset is determined based on the physical model processing of operating state data related to the gas turbine.

[0032] An intermediate physics enhancement model training module is used to train a first physics enhancement model based on the physics model dataset to determine an intermediate physics enhancement model.

[0033] The first data processing module uses the intermediate physical enhancement model to process the experimental dataset to obtain the predicted operating parameters of the first gas turbine.

[0034] The second data processing module is used to process the experimental dataset using the second physical enhancement model to obtain the second gas turbine operating parameter prediction data.

[0035] The target loss generation module is used to obtain the target loss based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset; and

[0036] The second physics augmentation model training module is used to train the second physics augmentation model based on the target loss to obtain a target physics augmentation model for gas turbine performance prediction.

[0037] A fourth aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0038] A fifth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0039] A sixth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0040] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0041] Figure 1 A flowchart illustrating a method for constructing a neural network model for predicting gas turbine performance according to an embodiment of this application is shown schematically.

[0042] Figure 2 A schematic diagram of a gas turbine according to an embodiment of this application is shown.

[0043] Figure 3 The diagram illustrates an intermediate physics enhancement model training process according to an embodiment of this application.

[0044] Figure 4 The diagram illustrates a training process for a target physics enhancement model according to an embodiment of this application.

[0045] Figure 5 A schematic diagram showing how the loss function values ​​of the training and validation sets change with the number of training iterations in the target physics augmentation model and the multilayer perceptron neural network;

[0046] Figure 6 A flowchart illustrating a gas turbine performance prediction method based on a physical enhancement model according to an embodiment of this application is shown schematically.

[0047] Figure 7 A schematic diagram illustrates the structural block diagram of a neural network model building apparatus for gas turbine performance prediction according to an embodiment of this application; and

[0048] Figure 8 A block diagram of an electronic device suitable for implementing a neural network model construction method for predicting gas turbine performance is illustrated according to an embodiment of this application. Detailed Implementation

[0049] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0050] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0051] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0052] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0053] Physical information neural networks (PINs) combine physical laws with neural networks to solve partial differential equations in complex engineering problems. By training the NPN according to physical laws, they accelerate convergence and improve performance prediction accuracy. However, in practical engineering problems, not all problems involve partial differential equations, or some device designs involve the coupling of multiple physical fields, resulting in more than one complex and difficult-to-solve partial differential equation. Examples include the Navier-Stokes equations in fluid dynamics and the microscopic and macroscopic behaviors in materials science, thus limiting the application of NPNs.

[0054] In view of the above-mentioned technical problems, embodiments of this application provide a method for constructing a neural network model for gas turbine performance prediction. The method involves acquiring an experimental dataset and a physical model dataset for the gas turbine, wherein the physical model dataset is determined based on the processing of gas turbine-related operating state data by the physical model; training a first physical enhancement model based on the physical model dataset to determine an intermediate physical enhancement model; processing the experimental dataset using the intermediate physical enhancement model to obtain first gas turbine operating parameter prediction data; processing the experimental dataset using a second physical enhancement model to obtain second gas turbine operating parameter prediction data; obtaining a target loss based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset; and training the second physical enhancement model based on the target loss to obtain a target physical enhancement model for gas turbine performance prediction.

[0055] The model building method provided in this application combines physical model data and experimental data to train the model, giving full play to the advantages of both physical models and experimental data, improving the accuracy of gas turbine performance prediction, reducing reliance on large amounts of experimental data, and lowering data acquisition costs. This provides a high-accuracy, fast-solving-speed, and small-sample-number solution for performance prediction of strongly nonlinear devices in practical engineering.

[0056] Figure 1 A flowchart illustrating a method for constructing a neural network model for predicting gas turbine performance according to an embodiment of this application is shown. Figure 2 A schematic diagram of a gas turbine according to an embodiment of this application is shown.

[0057] like Figure 1 As shown, the neural network model construction method for gas turbine performance prediction in this embodiment includes operations S110 to S160.

[0058] During operation of S110, the experimental dataset and physical model dataset of the gas turbine are obtained.

[0059] According to an embodiment of this application, the physical model dataset is determined based on physical model processing and gas turbine-related operating state data.

[0060] In one example, an experimental dataset is first obtained through experiments or data collection to build an experimental database. A physical model of the gas turbine is then established by simplifying assumptions and using empirical formulas or numerical simulation methods. Based on the physical model, input and output features are selected, and simulations are performed using these features to obtain a physical model dataset, thus building a physical model database.

[0061] In one example, such as Figure 2 As shown, a gas turbine includes a compressor, combustion chamber, turbine, and generator. The physical model of a gas turbine is as follows:

[0062] (1) Compressor

[0063] The expression for the compressor's pressure ratio correction factor is shown in equation (1):

[0064] Equation (1)

[0065] in, This is a compression ratio correction factor; This is the compressor speed correction factor; This is the mass flow correction factor for the compressor. , , Related to the speed correction factor, its expression is shown in equations (2) to (4):

[0066] Equation (2)

[0067] Equation (3)

[0068] Equation (4)

[0069] in, and These are empirical constants, 1.8 and 1.5 respectively.

[0070] The expressions for the mass flow rate correction factor and the speed correction factor are shown in equations (5) and (6):

[0071] Equation (5)

[0072] Equation (6)

[0073] in, , and These are the mass flow rate, temperature, and pressure of the compressor inlet air under non-design operating conditions. , and These are the mass flow rate, temperature, and pressure of the compressor inlet air under design operating conditions; and These represent the compressor speeds under non-design operating conditions and design operating conditions, respectively.

[0074] The expression for the isentropic efficiency correction coefficient of the compressor is shown in equation (7):

[0075] Equation (7)

[0076] in, This is the isentropic efficiency correction coefficient for the compressor; The empirical constant is 0.3.

[0077] The expressions for the compressor's pressure ratio and isentropic efficiency under off-design conditions are given in equations (8) and (9):

[0078] Equation (8)

[0079] Equation (9)

[0080] in, and This refers to the compressor's pressure ratio under non-design and design operating conditions. and The isentropic efficiency of the compressor under non-design conditions and design conditions.

[0081] The expression for calculating the enthalpy of the compressor outlet air is shown in equation (10):

[0082] Equation (10)

[0083] in, and This refers to the enthalpy values ​​of the air at the compressor inlet and outlet. This is the air enthalpy at the isentropic point of the compressor outlet.

[0084] The formula for calculating the electrical power consumption of the compressor is shown in equation (11):

[0085] Equation (11)

[0086] in, This refers to the electrical power consumption of the compressor.

[0087] (2) Combustion chamber

[0088] The energy conservation expression for the combustion chamber is shown in equation (12):

[0089] Equation (12)

[0090] in, and These are the enthalpy values ​​of the air outlet of the compressor and the gas inlet of the turbine, respectively. It has the low calorific value of natural gas; This refers to the inlet air flow rate for natural gas. This refers to the combustion chamber efficiency.

[0091] The formula for calculating the combustion chamber inlet air pressure is shown in equation (13):

[0092] Equation (13)

[0093] in, This is the expression for calculating the air pressure at the combustion chamber inlet.

[0094] The expression for calculating the pressure of the gas at the combustion chamber outlet is shown in equation (14):

[0095] Equation (14)

[0096] in, The pressure of the gas exiting the combustion chamber; This is the pressure loss coefficient.

[0097] (3) Turbine

[0098] The formula for calculating the isentropic efficiency correction coefficient of a turbine is shown in equation (15):

[0099] Equation (15)

[0100] in, The isentropic efficiency correction factor for the turbine; The empirical constant is 0.3; and These are the turbine speed correction factor and the mass flow rate correction factor, respectively.

[0101] The formulas for calculating the turbine's speed correction factor and mass flow rate correction factor are given in equations (16) and (17):

[0102] Equation (16)

[0103] Equation (17)

[0104] in, and These are the outlet exhaust gas mass flow rates under non-design and design conditions of the turbine, respectively; and This indicates the inlet gas temperature under non-design and design operating conditions of the turbine. and These are the inlet gas pressures under non-design and design turbine operating conditions, respectively. and These represent the turbine's rotational speeds under design and non-design operating conditions, respectively.

[0105] The formula for calculating the mass flow rate of exhaust gas at the turbine outlet under off-design conditions is given in equation (18):

[0106] Equation (18)

[0107] in, This represents the mass flow rate of natural gas.

[0108] According to the Freuger formula, the flow characteristics of a turbine are expressed as equation (19):

[0109] Equation (19)

[0110] in, , The expansion ratio of the turbine; The exhaust pressure of the turbine under non-design operating conditions.

[0111] The expression for calculating the isentropic efficiency of a turbine under off-design conditions is shown in equation (20):

[0112] Equation (20)

[0113] in, and These are the enthalpy values ​​of the turbine inlet gas and the outlet flue gas, respectively. This is the enthalpy value of the isentropic point of the flue gas at the turbine outlet. This is the enthalpy value of the isentropic point of the flue gas at the turbine outlet.

[0114] Electrical work generated by the turbine :

[0115] The expression for calculating the electrical work generated by the gas turbine is shown in equation (21):

[0116] Equation (21)

[0117] in, The electrical work generated by the gas turbine.

[0118] The calculation of the load factor and power generation efficiency of the gas turbine is shown in equations (22) and (23):

[0119] Equation (22)

[0120] Equation (23)

[0121] in, This refers to the load factor of the gas turbine. This refers to the rated capacity of the gas turbine. This refers to the power generation efficiency of the gas turbine.

[0122] Based on the physical model of the gas turbine, it is determined that the gas turbine load factor (f), compressor air inlet temperature (T1) and pressure (P1), and turbine exhaust pressure (P4) have a significant impact on the gas turbine performance. Therefore, these relevant operating state indicators of the gas turbine are selected as input features, including the compressor inlet air mass flow rate (m1), natural gas mass flow rate (m6), and power generation efficiency (η). gen The turbine outlet flue gas temperature (T4) and mass flow rate (m4) are used as output features, i.e., the operating parameters of the gas turbine. The experimental dataset and the physical model dataset provide a comprehensive data foundation for subsequent model training. The experimental dataset reflects the performance of the gas turbine in actual operation, while the physical model dataset is generated based on physical laws, ensuring the scientific validity and reliability of the data. This helps the model learn the performance characteristics of the gas turbine from different perspectives, laying the foundation for improving prediction accuracy.

[0123] In operation S120, the first physics augmentation model is trained based on the physics model dataset to determine the intermediate physics augmentation model.

[0124] In one example, the physical model dataset is input into a first physics augmentation model for training. By adjusting the model parameters to make the prediction results close to the labeled running parameters of the physical model dataset, an intermediate physics augmentation model is determined. The first physics augmentation model can be, for example, a multilayer perceptron (MLP). This enables the first physics augmentation model to learn the physical characteristics of the gas turbine and fit and predict data according to physical laws. Through this training process, the model initially possesses the ability to predict gas turbine performance, and because it incorporates physical knowledge, it reduces reliance on large amounts of experimental data, thus lowering data acquisition costs.

[0125] In operation S130, the experimental dataset is processed using the intermediate physics enhancement model to obtain the predicted operating parameters of the first gas turbine.

[0126] According to an embodiment of this application, a box plot is used to preprocess the experimental dataset to obtain a preprocessed experimental dataset.

[0127] In operation S140, the experimental dataset is processed using the second physical enhancement model to obtain the second gas turbine operating parameter prediction data;

[0128] In operation S150, the target loss is obtained based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset.

[0129] In operation S160, the second physical augmentation model is trained based on the target loss to obtain a target physical augmentation model for gas turbine performance prediction.

[0130] In one example, before inputting the experimental dataset into the intermediate physics augmentation model trained by operation S120, i.e., after collecting the raw experimental data of the gas turbine, the dataset is preprocessed using box plots. The box plots identify outliers in the data, which may be due to measurement errors, abnormal operating conditions, etc. These outliers are then removed, resulting in a preprocessed experimental dataset that better reflects the state distribution of a gas turbine during normal operation.

[0131] In one example, an experimental dataset (containing input parameters of the gas turbine under different operating conditions, such as load rate, compressor air inlet temperature, and pressure) is input into this intermediate physics augmentation model. The model performs calculations and inferences based on previously learned physical laws and data characteristics, outputting the first set of predicted gas turbine operating parameters, such as predicted power generation efficiency and turbine outlet temperature. These predicted data reflect the intermediate physics augmentation model's preliminary predictions of gas turbine performance. A second physics augmentation model is then constructed, with a structure that is either identical or different from the intermediate model. The same experimental dataset is input into this second physics augmentation model. At this point, the model has not been fully trained; it processes the experimental data based on its initialization parameters and outputs the second set of predicted gas turbine operating parameters. This set of data represents the second physics augmentation model's initial predictions of gas turbine performance without considering the constraints of the physical model dataset.

[0132] In one example, the target loss is obtained based on the predicted operating parameters of the first gas turbine, the predicted operating parameters of the second gas turbine, and the labeled operating parameters of the experimental dataset. By comprehensively considering the prediction results of the physical model, the prediction results of the experimental data, and the actual labeled values, the target loss can accurately evaluate the model's predictive performance. The target loss reflects the difference between the model's predicted values ​​and the true values, providing crucial feedback information for further training and optimization of the model. It helps to identify the errors in the model's predictions, thereby guiding the direction of model improvement.

[0133] In one example, the target loss is used as the optimization objective, and optimization algorithms (such as gradient descent) are employed to adjust the parameters of the second physics augmentation model. In each training iteration, the model parameters are updated based on the target loss, gradually decreasing the target loss value. After multiple rounds of iterative training, the training process stops when the target loss reaches a preset threshold or when a certain number of iterations are completed. The resulting second physics augmentation model is the target physics augmentation model. It is optimized under the dual constraints of physical laws and experimental data, enabling more accurate prediction of gas turbine performance parameters and providing a more reliable basis for gas turbine performance evaluation, fault diagnosis, and optimized operation.

[0134] The target physics enhancement model constructed through the embodiments of this application has higher accuracy, better generalization ability and faster convergence speed in predicting gas turbine performance, which meets the needs of high precision, fast solution and small sample size for gas turbine performance prediction in practical engineering applications.

[0135] Figure 3 The diagram illustrates an intermediate physics enhancement model training process according to an embodiment of this application.

[0136] According to an embodiment of this application, training a first physical augmentation model based on the physical model dataset to determine an intermediate physical augmentation model includes: inputting the physical model dataset into the first physical augmentation model and outputting third gas turbine operating parameter prediction data; and training the first physical augmentation model based on the third gas turbine operating parameter prediction data and the labeled operating parameters of the physical model dataset.

[0137] Combination Figure 3 The training process shown includes the following steps 1-6:

[0138] Step 1: Initialize parameters. Normalize the original dataset and divide it into training, validation, and test sets. Initialize weights and biases, and determine the batch training size. The normalization formula used is shown in Equation (24).

[0139] Equation (24)

[0140] in, Let q represent the features of the s-th sample.

[0141] Step 2: Forward Propagation. The outputs of the hidden and output layers are calculated sequentially using activation functions, weights, and biases. The expression for forward propagation is shown in equation (25):

[0142]

[0143] Equation (25)

[0144] in, Output before activation; As weight; Output is generated after activation; Let be the activation function; the superscript l represents the l-th layer; the subscript ij represents the i-th neuron in the l-th layer. The modified linear unit function (ReLU) is chosen as the activation function, expressed as equation (26):

[0145] Equation (26)

[0146] Step 3: Calculate the loss function. The mean squared error (MSE) is chosen as the loss function, and its expression is given by equation (27):

[0147] Equation (27)

[0148] in, and These represent the labeled operating parameters of the physical model dataset and the predicted operating parameters of the third gas turbine, respectively; N is the number of samples.

[0149] Step 4: Backpropagation. Calculate the gradients of each parameter and perform backpropagation. The gradient expressions for the weights and biases are given by equation (28):

[0150]

[0151] Equation (28)

[0152] Step 5: Parameter Update. Update the weights and biases, expressed as equation (29):

[0153]

[0154] Equation (29)

[0155] in, Let be the learning rate, and u represent the u-th iteration.

[0156] Step 6: Iterative calculation. Repeat the above steps until the required number of iterations is reached.

[0157] According to an embodiment of this application, after a preset training period, the second physical augmentation model is evaluated using a validation set, and the loss function value on the validation set is calculated; if it is determined that the current validation loss is lower than the optimal validation loss, the optimal validation loss is updated, and the parameters of the current model are saved; and if the loss value on the validation set does not decrease within a preset number of consecutive iterations, an early stopping condition is triggered, and model training is stopped.

[0158] In one example, for a multilayer perceptron neural network model, early stopping is used to prevent overfitting and improve prediction accuracy. The expression for early stopping can be found in equation (30):

[0159]

[0160] Equation (30)

[0161] in, This represents the number of times the loss function value changes after iterations. The superscript ep indicates the ep-th iteration; the superscript c indicates the number of iterations that have changed; and es indicates the number of early stopping iterations. Training stops if the validation set loss has not significantly decreased by the es-th iteration.

[0162] Figure 4 The diagram illustrates a target physics enhancement model training process according to an embodiment of this application.

[0163] like Figure 4 As shown, the preprocessed experimental dataset is normalized. The normalized training data is then input into the intermediate physics augmentation model and the second physics augmentation model to output the first and second gas turbine operating parameter prediction data. The target loss is obtained based on the first, second, and labeled operating parameters of the experimental dataset. The gradient of the target loss function with respect to the network weights and biases is calculated using the backpropagation algorithm. The intermediate physics augmentation model parameters, physics loss weights, and second physics augmentation model parameters are updated based on the calculated gradient. The forward propagation, loss function calculation, backpropagation, and parameter update process is repeated until a predetermined number of iterations is reached or a preset convergence condition is met.

[0164] According to an embodiment of this application, obtaining the target loss based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset includes: determining the physical loss based on the first gas turbine operating parameter prediction data and the second gas turbine operating parameter prediction data; determining the experimental loss based on the second gas turbine operating parameter prediction data and the labeled operating parameters of the experimental dataset; and determining the target loss based on the physical loss, the experimental loss, and the physical loss weight.

[0165] In one example, the difference between the predicted operating parameters of the first gas turbine and the predicted operating parameters of the second gas turbine is first calculated to obtain the physical loss, which reflects the deviation between the prediction results of the intermediate physical enhancement model and the prediction results of the second physical enhancement model. At the same time, the difference between the predicted operating parameters of the second gas turbine and the labeled operating parameters of the experimental dataset is calculated to obtain the experimental loss, which reflects the deviation between the prediction results of the second physical enhancement model and the actual data. Then, according to the preset physical loss weight (e.g., the weight value determined by experience or optimization algorithm), the physical loss and the experimental loss are weighted and summed to obtain the final target loss. The target loss comprehensively considers the prediction results of the physical model and the actual situation of the experimental data, and is used to measure the overall prediction performance of the model. The expression of the target loss function is shown in equation (31):

[0166] Equation (31)

[0167]

[0168]

[0169] in, The loss function for the Physics-enhanced Neural Network Model (PENN); and The mean squared errors are obtained based on the intermediate physics enhancement model and the new experimental dataset, respectively. for The weights; These are the predictions from the intermediate physics enhancement model; These are the predicted values ​​obtained by training on the experimental dataset.

[0170] According to an embodiment of this application, the hyperparameters of a physical augmentation neural network model are optimized based on a Bayesian optimization algorithm, wherein the hyperparameters include the number of hidden layers, the number of neurons, the learning rate, the batch size, and the weights of the intermediate physical augmentation model.

[0171] In one example, the hyperparameters of a physics-enhanced neural network model can be optimized using Bayesian optimization. First, the search space for hyperparameters is determined; for example, the number of hidden layers can be chosen between 3 and 5, and the number of neurons between 50 and 200. Then, the Bayesian optimization algorithm is used to explore this search space, predicting the model performance under different combinations of hyperparameters by constructing a surrogate model. The combination of hyperparameters with better performance is then selected. These hyperparameters include the number of hidden layers, the number of neurons, the learning rate, the batch size, and the weights of intermediate physics-enhanced models.

[0172] The target physics augmentation model (PENN) in this embodiment is configured as follows: 70% of the original dataset is randomly selected as the training set, 15% of the remaining dataset is randomly selected as the validation set, and the remaining dataset is used as the test set. The training iterations are 5000, with 50 early stopping iterations, meaning that training stops if the loss function of the validation set does not improve within 50 training iterations. To demonstrate the superiority of the model construction method provided in this application, the prediction performance of PENN is compared with that of three other prediction methods: Random Forest (RF), Support Vector Regression (SVR), and Multilayer Perceptron Neural Network (MLP). To ensure fairness in the comparison, the hyperparameters of RF, SVR, and MLP are optimized using a Bayesian optimization algorithm; the specific parameters are shown in Table 1.

[0173] Table 1. Optimal hyperparameter values ​​for each prediction method obtained by the Bayesian optimization algorithm.

[0174]

[0175] Figure 5 This diagram illustrates how the loss function values ​​of the training and validation sets change with the number of training iterations in a target physics augmentation model and a multilayer perceptron neural network.

[0176] like Figure 5 As shown, the loss function value of the PENN validation set did not improve after 982 training sessions, and the loss function value of the MLP validation set did not improve after 471 training sessions. Therefore, the PENN and MLP validation sets terminated at 1032 and 521 training sessions, respectively.

[0177] Table 2 compares the prediction accuracy of different prediction methods. Looking at the MSE of the Target Physics Augmentation Model (PENN) on the test set, it is 53.18%, 99.17%, and 35.05% lower than MLP, RF, and SVR, respectively. This means that, under the same sample data, the method provided in this application provides the model with the highest prediction accuracy and the strongest generalization ability.

[0178] Table 2 Comparison of Performance and Prediction Accuracy of Different Prediction Methods

[0179]

[0180] To demonstrate the advantage of the proposed model building method in saving sample quantity, 1100, 1000, and 900 samples were randomly selected from the training set for training, and their performance prediction accuracy was compared with that of SVR and MLP. The number of samples in the test set remained unchanged. Table 4 compares the prediction accuracy of PENN with the original sample quantity (1704 samples) for both small sample sets and SVR and MLP. It can be seen that PENN can achieve high performance prediction accuracy even with a small sample quantity. For example, looking at the test set, when the training set sample quantity is 1100, the MSE of the PENN test set is reduced by 9.24% and 34.58% compared to the original sample quantity of SVR and MLP test sets. This indicates that compared to SVR and MLP, the proposed model building method can save at least 35.45% of the sample quantity.

[0181] Table 3. Comparison of prediction accuracy between PENN with small sample size and SVR and MLP with original sample size.

[0182]

[0183] Figure 6 A flowchart illustrating a gas turbine performance prediction method based on a physical enhancement model according to an embodiment of this application is shown.

[0184] like Figure 6 As shown, this includes operations S610 to S630.

[0185] The S610 is used to collect relevant operating status data of the gas turbine.

[0186] In operation S620, the relevant operating status data of the gas turbine are preprocessed.

[0187] In operation S630, the preprocessed gas turbine-related operating status data is input into a pre-trained physical augmentation model to output gas turbine operating parameter prediction data. The physical augmentation model is trained according to the aforementioned neural network model construction method for gas turbine performance prediction.

[0188] In one example, relevant operating status data of the gas turbine during actual operation or when performance prediction is needed, such as real-time load, ambient temperature, and pressure, are collected. This data undergoes preprocessing, including data cleaning and normalization. The preprocessed data is then input into a pre-trained physical augmentation model. Based on its learned knowledge, the model outputs predicted operating parameters of the gas turbine, such as power generation efficiency, air mass flow rate, and emissions, providing a basis for gas turbine operation monitoring, fault diagnosis, and performance optimization.

[0189] It should be understood that the neural network model construction method for gas turbine performance prediction provided in this application embodiment, besides having significant application value in the field of gas turbines, can also be widely applied to other fields involving complex physical systems and strong nonlinear relationships. For example, in the field of energy prediction: performance prediction of aircraft engines, steam turbines, internal combustion engines, etc., helps optimize energy utilization efficiency; in the field of industrial equipment maintenance technology: performance prediction of chemical process equipment, power transformers, etc., allows for advance maintenance planning and reduces unexpected downtime; in the field of new energy technology: it helps improve the efficiency and reliability of new energy equipment such as fuel cells and wind turbines; in the field of electronics and automation: it can be applied to robot motion control and electronic equipment heat dissipation performance prediction, enhancing system stability and efficiency; in the field of biomedical engineering technology: it can be used for medical equipment performance prediction, improving equipment accuracy and reliability, etc.

[0190] Based on the aforementioned neural network model construction method for gas turbine performance prediction, this application also provides a neural network model construction device for gas turbine performance prediction. The following will be combined with... Figure 7 The device is described in detail.

[0191] Figure 7 A schematic block diagram of a neural network model building apparatus for predicting gas turbine performance is shown according to an embodiment of this application.

[0192] like Figure 7 As shown, the neural network model building device 700 for gas turbine performance prediction in this embodiment includes a data acquisition module 710, an intermediate physical enhancement model training module 720, a first data processing module 730, a second data processing module 740, a target loss generation module 750, and a second physical enhancement model training module 760.

[0193] The data acquisition module 710 is used to acquire experimental datasets and physical model datasets of the gas turbine. The physical model dataset is determined based on physical model processing of operating state data related to the gas turbine. In one embodiment, the data acquisition module 710 can be used to perform the operation S110 described above, which will not be repeated here.

[0194] The intermediate physics enhancement model training module 720 is used to train the first physics enhancement model based on the physics model dataset to determine the intermediate physics enhancement model, which will not be described in detail here. In one embodiment, the intermediate physics enhancement model training module 720 can be used to perform the operation S120 described above, which will not be described in detail here.

[0195] The first data processing module 730 is used to process the experimental dataset using the intermediate physics enhancement model to obtain predicted data of the first gas turbine operating parameters. In one embodiment, the first data processing module 730 can be used to perform the operation S130 described above, which will not be repeated here.

[0196] The second data processing module 740 is used to process the experimental dataset using the second physical enhancement model to obtain the second gas turbine operating parameter prediction data. In one embodiment, the second data processing module 740 can be used to perform the operation S140 described above, which will not be repeated here.

[0197] The target loss generation module 750 is used to obtain the target loss based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset. In one embodiment, the target loss generation module 750 can be used to perform the operation S150 described above, which will not be repeated here.

[0198] The second physics augmentation model training module 760 is used to train the second physics augmentation model based on the target loss to obtain a target physics augmentation model for gas turbine performance prediction. In one embodiment, the second physics augmentation model training module 760 can be used to perform the operation S160 described above, which will not be repeated here.

[0199] According to embodiments of this application, any multiple modules among the data acquisition module 710, intermediate physical enhancement model training module 720, first data processing module 730, second data processing module 740, target loss generation module 750, and second physical enhancement model training module 760 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 710, intermediate physical enhancement model training module 720, first data processing module 730, second data processing module 740, target loss generation module 750, and second physical enhancement model training module 760 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the data acquisition module 710, the intermediate physical enhancement model training module 720, the first data processing module 730, the second data processing module 740, the target loss generation module 750, and the second physical enhancement model training module 760 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0200] Figure 8 A block diagram of an electronic device suitable for implementing a neural network model construction method for predicting gas turbine performance is illustrated according to an embodiment of this application.

[0201] like Figure 8 As shown, an electronic device 800 according to an embodiment of this application includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage portion 808 into a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0202] RAM 803 stores various programs and data required for the operation of electronic device 800. Processor 801, ROM 802, and RAM 803 are interconnected via bus 804. Processor 801 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 802 and / or RAM 803. It should be noted that the programs may also be stored in one or more memories other than ROM 802 and RAM 803. Processor 801 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0203] According to embodiments of this application, the electronic device 800 may further include an input / output (I / O) interface 805, which is also connected to a bus 804. The electronic device 800 may also include one or more of the following components connected to the input / output (I / O) interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output (I / O) interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 810 as needed so that computer programs read from it can be installed into the storage section 808 as needed.

[0204] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0205] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 802 and / or RAM 803 and / or one or more memories other than ROM 802 and RAM 803 described above.

[0206] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the neural network model construction method for gas turbine performance prediction provided in the embodiments of this application.

[0207] When the computer program is executed by the processor 801, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0208] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 809, and / or installed from a removable medium 811. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0209] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 809, and / or installed from the removable medium 811. When the computer program is executed by the processor 901, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0210] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0211] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0212] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0213] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A method for constructing a neural network model for gas turbine performance prediction, characterized by, The method includes: Experimental datasets are obtained through experiments or data collection, and physical models of gas turbines are established using empirical formulas or numerical simulation methods by simplifying assumptions. Input and output features are selected based on the physical model. The physical model dataset is obtained by simulation based on the physical model, the input features, and the output features. The gas turbine load rate, compressor air inlet temperature, compressor air inlet pressure, and turbine exhaust pressure are selected as input features, and the compressor inlet air mass flow rate, natural gas mass flow rate, power generation efficiency, turbine outlet flue gas temperature, and turbine outlet flue gas mass flow rate are selected as output features. The experimental dataset reflects the performance of the gas turbine in actual operation, and the physical model dataset is generated based on physical laws. The first physics augmentation model is trained based on the physics model dataset to determine the intermediate physics augmentation model; The experimental dataset is processed using the intermediate physics enhancement model to obtain predicted operating parameters for the first gas turbine. The experimental dataset was processed using a second physical enhancement model to obtain predicted data for the second gas turbine operating parameters; The target loss is obtained based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset; and The second physics augmentation model is trained based on the target loss to obtain a target physics augmentation model for gas turbine performance prediction. The step of obtaining the target loss based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset includes: The difference between the predicted operating parameters of the first gas turbine and the predicted operating parameters of the second gas turbine is calculated to obtain the physical loss; The difference between the predicted operating parameters of the second gas turbine and the labeled operating parameters of the experimental dataset is calculated to obtain the experimental loss; and The physical loss and the experimental loss are weighted and summed according to the preset physical loss weights to obtain the target loss.

2. The method of claim 1, wherein, The step of training the first physics augmentation model based on the physics model dataset to determine the intermediate physics augmentation model includes: The physical model dataset is input into the first physical enhancement model, and the third gas turbine operating parameter prediction data is output; and The first physical augmentation model is trained based on the predicted operating parameters of the third gas turbine and the labeled operating parameters of the physical model dataset.

3. The method according to claim 1, characterized in that, The method further includes: The experimental dataset was preprocessed using box plots to obtain the preprocessed experimental dataset.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: The hyperparameters of a physical augmentation neural network model are optimized based on the Bayesian optimization algorithm. The hyperparameters include the number of hidden layers, the number of neurons, the learning rate, the batch size, and the weights of the intermediate physical augmentation model.

5. The method according to claim 4, characterized in that, The method further includes: The second physics augmentation model is evaluated using a validation set after a preset training period, and the loss function value on the validation set is calculated. If it is determined that the current validation loss is lower than the optimal validation loss, then update the optimal validation loss and save the parameters of the current model; and If the loss value on the validation set does not decrease within a preset number of consecutive iterations, the early stopping condition is triggered, and model training is stopped.

6. A method for predicting gas turbine performance based on a physical enhancement model, characterized in that, The method includes: Collect relevant operating status data of the gas turbine; Preprocessing of the relevant operating status data of the gas turbine; and The preprocessed gas turbine operating status data is input into a pre-trained physical augmentation model to output predicted gas turbine operating parameters, wherein the physical augmentation model is trained by the method according to any one of claims 1 to 5.

7. A neural network model construction device for predicting gas turbine performance, characterized in that, The device includes: The data acquisition module is used to acquire experimental datasets through experiments or data collection, and to establish a physical model of the gas turbine using empirical formulas or numerical simulation methods by simplifying assumptions. Input and output features are selected based on the physical model, and the physical model dataset is obtained through simulation based on the physical model, the input features, and the output features. The gas turbine load rate, compressor air inlet temperature, compressor air inlet pressure, and turbine exhaust pressure are selected as input features, and the compressor inlet air mass flow rate, natural gas mass flow rate, power generation efficiency, turbine outlet flue gas temperature, and turbine outlet flue gas mass flow rate are selected as output features. The experimental dataset reflects the performance of the gas turbine in actual operation, and the physical model dataset is generated based on physical laws. An intermediate physics enhancement model training module is used to train a first physics enhancement model based on the physics model dataset to determine an intermediate physics enhancement model. The first data processing module is used to process the experimental dataset using the intermediate physical enhancement model to obtain the first gas turbine operating parameter prediction data. The second data processing module is used to process the experimental dataset using the second physical enhancement model to obtain the second gas turbine operating parameter prediction data. The target loss generation module is used to obtain the target loss based on the first gas turbine operating parameter prediction data, the second gas turbine operating parameter prediction data, and the labeled operating parameters of the experimental dataset; and The second physics augmentation model training module is used to train the second physics augmentation model based on the target loss to obtain a target physics augmentation model for gas turbine performance prediction. The target loss generation module is also used to calculate the difference between the predicted operating parameters of the first gas turbine and the predicted operating parameters of the second gas turbine to obtain the physical loss; calculate the difference between the predicted operating parameters of the second gas turbine and the labeled operating parameters of the experimental dataset to obtain the experimental loss; and perform a weighted summation of the physical loss and the experimental loss according to the preset physical loss weight to obtain the target loss.

8. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 5.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 5.

Images