Gas turbine blade parameter prediction method and system

Abstract

The invention discloses a gas turbine blade parameter prediction method and system, and relates to the technical field of gas turbine strength and pneumatic design, and the method comprises the following steps: obtaining a generator and a discriminator, carrying out the iteration of the generator and the discriminator, and obtaining an optimal generator and an optimal discriminator; sampling parameters of the turbine blade model are resampled, resampled data are input into the optimal generator, and optimal blade parameter data are obtained; training the third graph convolutional neural network through the resampling data and the optimal leaf parameter data to obtain a physical field prediction network; and predicting the turbine blade parameters of the gas turbine through the physical field prediction network. According to the method, the high-quality parameter training data is generated based on the high-quality generator and discriminator, so that the finite element calculation data volume can be greatly reduced, the finite element calculation cost is greatly reduced, the data acquisition efficiency is improved, and the construction of the physical field prediction network is accelerated on the premise of ensuring the prediction precision.

Description

Technical Field

[0001] The present invention relates to the technical field of gas turbine strength and aerodynamic design, and particularly to a method and system for predicting parameters of a gas turbine turbine blade. Background Art

[0002] As an important component of a gas turbine, the structural performance of a gas turbine turbine blade is directly related to the working efficiency and service life of the entire device. When the turbine blade is in a harsh working environment of high temperature, high pressure and high stress, it needs to bear complex thermo-mechanical loads. Therefore, the design, manufacture and performance evaluation of turbine blades are particularly important. To ensure the safety, reliability of the device and extend its service life, it is necessary to accurately predict the key parameters of the turbine blade, which is of great significance for optimizing the design, improving the efficiency of the gas turbine and reducing the operating cost.

[0003] Currently, the methods for predicting gas turbine turbine blade parameters mainly adopt artificial intelligence and neural network technologies. These technologies are based on a large amount of experimental data and simulation data to train a deep neural network model, so as to quickly and efficiently complete the prediction of the key performance parameters of the blade. Compared with traditional physical models, the neural network model has a strong non-linear fitting ability and can capture the parameter change rules under complex working conditions. In addition, these artificial intelligence-based methods can also combine actual operation data to achieve online prediction and adaptive optimization, saving time and computing resources while improving the flexibility and applicability of prediction.

[0004] However, there are still some challenges in the process of training a neural network to predict gas turbine turbine blade parameters. The high performance of the neural network model depends on a large amount of high-quality parameter training data, which are currently mainly obtained based on finite element calculations. This process has a large amount of calculations, requires a lot of computing resources and time, so that the generation of a large amount of data has become a bottleneck problem in the prediction process. The large-scale computing requirements not only reduce the efficiency, but also limit the rapid iteration and optimization of the neural network model. Summary of the Invention

[0005] Based on the above-mentioned defects existing in the prior art, the present invention provides a method and system for predicting parameters of a gas turbine turbine blade, which solves the problem that the existing data are currently mainly obtained based on finite element calculations, and this process has a large amount of calculations, requires a lot of computing resources and time, so that the generation of a large amount of data has become a bottleneck in the prediction process.

[0006] The present invention adopts the following technical solutions:

[0007] In the first aspect, the present invention provides a method for predicting parameters of a gas turbine turbine blade, including the following steps:

[0008] Build a turbine blade model of a gas turbine and specify sampling parameters;

[0009] Perform a first sampling on the sampling parameters of the turbine blade model under different conditions to obtain first training data; perform a blade thermal-fluid-solid coupling calculation on the turbine blade model to obtain real data of the blade parameters; use the first training data as the input and the real data as the output to train the first graph convolutional neural network to obtain a generator;

[0010] Perform a second sampling on the sampling parameters of the turbine blade model to obtain second training data, input the second training data into the generator to obtain forged data; train the second graph convolutional neural network with the first training data and the real data as well as the second training data and the forged data to obtain a discriminator;

[0011] Perform a third sampling on the sampling parameters of the turbine blade model to obtain third training data, input the third training data into the generator to obtain generated data; input the third training data and the generated data into the discriminator to obtain probability values; determine the loss function value of the discriminator based on the probability values, and iterate the generator and the discriminator through the loss function value to obtain an optimal generator and an optimal discriminator;

[0012] Perform resampling on the sampling parameters of the turbine blade model and input the resampled data into the optimal generator to obtain optimal blade parameter data; use the resampled data as the input and the optimal blade parameter data as the output to train the third graph convolutional neural network to obtain a physical field prediction network; predict the gas turbine turbine blade parameters through the physical field prediction network.

[0013] Preferably, model the solid domain and the fluid domain of the turbine blade of the gas turbine to obtain a turbine blade model; the sampling parameters include operating conditions parameters, machining parameters, and material parameters.

[0014] Preferably, perform a blade thermal-fluid-solid coupling calculation on the turbine blade model through a pneumatic batch file to obtain real data of the blade parameters, and the blade parameters include a pressure field and a temperature field.

[0015] Preferably, the step of using the first training data as the input and the real data as the output to train the first graph convolutional neural network to obtain a generator includes the following steps:

[0016] Perform mesh division on the turbine blade model;

[0017] Establish an adjacency matrix through the meshed turbine blade model;

[0018] Obtain the neighborhood node set of each node in the first graph convolutional neural network through the adjacency matrix;

[0019] For each node, aggregate the neighborhood features from its neighboring nodes to obtain the aggregated feature vector of the current node; the neighborhood features include the first training data and the real data;

[0020] Combine the aggregated feature vector of the current node with the node's own features to obtain the updated features;

[0021] Use the first graph convolutional neural network with updated node features as the generator;

[0022] The output of the generator is as follows:

[0023] Y recon = G(X, Z = 0; θ1);

[0024] In the formula, Y is the output of the generator G, recon is the reconstruction, θ1 is the learnable parameter of the generator, X is the first training data, and Z is the noise;

[0025] The reconstruction loss of the generator is as follows:

[0026]

[0027] In the formula, L recon is the reconstruction loss, N is the number of sample points, is the square of the second norm, Y real，i is the i-th real data, is the i-th generated data;

[0028] Update the generator parameters through gradient descent.

[0029] Preferably, training the second graph convolutional neural network with the first training data and the real data, as well as the second training data and the forged data, includes the following steps:

[0030] Make the true label 1 for the first training data and the real data, and make the false label 0 for the second training data and the forged data;

[0031] Input the first training data, the real data, and the true label 1, as well as the second training data, the forged data, and the false label 0 into the second graph convolutional neural network;

[0032] Obtain the node features H (L) from the second graph convolutional network, and obtain the global graph features H graph through average pooling:

[0033] H graph = meanpool(H (L) );

[0034] Average pooling to each feature dimension:

[0035]

[0036] Where h N(v)，k is the representation of the pooled node v in the k-th feature dimension, |N(v)| is the neighborhood size of node v, h u,k is the k-th dimensional feature value of node u, d is the number of dimensions, and N(ν) is the neighborhood, which contains the set of all nodes directly connected to node v;

[0037] Use an activation function to calculate the output probability;

[0038] Use the trained second graph convolutional neural network as the discriminator;

[0039] The discriminator's discrimination output for real data is as follows:

[0040] D(Y real ) = D(X, Y real ; θ2);

[0041] Where D(Y real ) is the output of the discriminator for real data Y real , and θ2 is the learnable parameter of the discriminator;

[0042] The discriminator's discrimination output for forged data is as follows:

[0043] D(Y real ) = D(G(X, Z = 0; θ1); θ2);

[0044] Where D(Y recon ) is the output of the discriminator for forged data .

[0045] Preferably, the loss function of the generator is specifically as follows:

[0046] L gen = λ recon L recon + λ G L G ;

[0047] Where L gen is the total generator loss, λ recon and λ G are hyperparameters, and L G is the adversarial loss;

[0048] The loss function of the discriminator is specifically as follows:

[0049]

[0050] Where L Dis the loss of the discriminator, D(X i , Y i ) is the output probability of the discriminator, y i is the label of the discriminator, X i is the i-th third sampling parameter, Y i is the i-th generated data;

[0051] Iterating the generator and the discriminator through the loss function value specifically includes the following steps:

[0052] Comparing the loss function value with a target threshold;

[0053] When the loss function value is greater than the target threshold, iterating the generator and the discriminator based on the Adam optimizer, otherwise outputting the optimal learning parameters of the generator and the discriminator.

[0054] Preferably, training the third graph convolutional neural network specifically includes the following steps:

[0055] Using the resampled data, the optimal blade parameter data, the first training data, and the real data of the blade parameters as the training set, where the training set is as follows:

[0056] (X + X, Y + Y real ) → (X new , Y new );

[0057] In the formula, is the resampled data, X is the first training data, is the optimal blade parameter data, Y real is the real data, X new is the input of the training set, Y new is the output of the training set;

[0058] Training through the following formula:

[0059] Y new = F(X new ; θ3);

[0060] In the formula, is the physical field prediction network, and θ3 is the learnable parameter of the physical field prediction network.

[0061] In a second aspect, the present invention provides a gas turbine turbine blade parameter prediction system, including:

[0062] A construction module for constructing a turbine blade model of a gas turbine and specifying sampling parameters;

[0063] The generator training module is used to perform a first sampling on the sampling parameters of the turbine blade model under different conditions to obtain first training data; perform blade thermo-fluid-solid coupling calculation on the turbine blade model to obtain the real data of the blade parameters; use the first training data as the input and the real data as the output to train the first graph convolutional neural network to obtain a generator;

[0064] The discriminator training module is used to perform a second sampling on the sampling parameters of the turbine blade model to obtain second training data, input the second training data into the generator to obtain forged data; train the second graph convolutional neural network with the first training data and the real data as well as the second training data and the forged data to obtain a discriminator;

[0065] The iteration module is used to perform a third sampling on the sampling parameters of the turbine blade model to obtain third training data, input the third training data into the generator to obtain generated data; input the third training data and the generated data into the discriminator to obtain a probability value; determine the loss function value of the discriminator based on the probability value, and iterate the generator and the discriminator through the loss function value to obtain an optimal generator and an optimal discriminator;

[0066] The prediction module is used to resample the sampling parameters of the turbine blade model and input the resampled data into the optimal generator to obtain optimal blade parameter data; use the resampled data as the input and the optimal blade parameter data as the output to train the third graph convolutional neural network to obtain a physical field prediction network; predict the parameters of the gas turbine turbine blade through the physical field prediction network.

[0067] Compared with the prior art, at least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0068] The present invention first performs multiple samplings on the sampling parameter data of the turbine blade model under different conditions, and simultaneously performs the calculation of the blade thermal-fluid-solid coupling corresponding to the sampling data of this time to obtain the real data of the blade parameters. The first graph convolutional neural network is trained with the primary sampling parameter data and the blade parameter data to obtain the generator. The second graph convolutional neural network is trained with the secondary sampling data and the data generated by the generator to obtain the discriminator. The constructed generator and discriminator are optimized again with the data of the third sampling to obtain the optimal generator and the optimal discriminator. Finally, the optimal generator and the optimal discriminator are used to quickly generate the required data volume of the physical field prediction network. In the entire training process of the generator and the discriminator of the present invention, only a small amount of calculations of the blade thermal-fluid-solid coupling are required. Based on the calculation results of the thermal-fluid-solid coupling and the sampling data, high-quality generators and discriminators are generated. Based on the high-quality generators and discriminators, high-quality parameter training data is generated, which can greatly reduce the amount of finite element calculation data, greatly reduce the finite element calculation cost, improve the data acquisition efficiency, and accelerate the construction of the physical field prediction network on the premise of ensuring the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0070] Figure 1 It is a flowchart of a method for predicting the parameters of a gas turbine turbine blade according to the present invention;

[0071] Figure 2 It is a schematic diagram of the construction process of the physical field prediction network according to the present invention;

[0072] Figure 3 It is a schematic diagram of the construction process of the generator according to the present invention;

[0073] Figure 4 It is a schematic diagram of the construction process of the discriminator according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0074] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0075] Refer to Figure 1, an embodiment of the present invention provides a method for predicting the parameters of a gas turbine turbine blade, which specifically includes the following steps:

[0076] S1: Model the solid and fluid of the gas turbine turbine blade to obtain a turbine blade model, including the solid domain of the turbine blade and the flow passage area where the fluid flows through the turbine blade. Mesh the turbine blade model to obtain multiple mesh nodes.

[0077] S2: Establish boundary conditions.

[0078] Under different conditions, perform sampling parameter sampling and blade thermo-fluid-solid coupling calculation on the turbine blade model respectively to obtain multiple sample points and corresponding boundary conditions (blade parameter data), that is, the first training data.

[0079] The sampling parameters are generally relevant parameters that can determine the change of blade parameters and can be set in finite element calculations. Specify the sampling parameters and the range of parameter changes to determine the calculation boundary conditions. The range of changes gives specific upper and lower deviations according to the actual situation. The sampling parameters include operating conditions parameters, machining parameters, and material parameters. The operating conditions parameters include the total temperature at the mainstream gas inlet, the total pressure at the mainstream gas inlet, the flow rate at the single-channel outlet, the total temperature at the cooling gas inlet, the flow rate of the single-channel cooling gas, and the rotational speed, etc. The machining parameters include the absolute gas flow angle at the moving blade inlet, and the material parameters include the specific heat capacity of the blade material, the thermal conductivity of the blade material, the density of the blade material, the linear expansion coefficient of the blade material, and the elastic modulus of the blade material, etc.

[0080] Use the Latin hypercube method to sample the sampling parameters under different conditions (different operating conditions, different machining conditions, and different material conditions).

[0081] Let the generated sample points be {x (1) , x (2) ,..., x (N)}, x is the sampling parameter sample, N is the number of sample points, and each point d is the dimension, and one dimension represents a type of parameter. Ensure that the projection of each dimension i satisfies:

[0082]

[0083] where is the i-th dimensional sample point in the k-th sample, is the lower bound of the i-th dimension, that is, the minimum value of the sample point, and Δx i is the length of the i-th dimensional subspace, and the calculation method is: is the randomly shuffled interval index, which are non-repeating for different k and are used to divide the sub-intervals of the i-th dimension. The interval is defined as: Each sub - interval corresponds to an index, ranging from 1 to N. That is, when it is the first sub - interval; when it is the last sub - interval. Combine N groups of intervals in all dimensions to generate N sample points.

[0084] Build a node attribute matrix. Use a pneumatic batch file to perform blade thermal - fluid - structure interaction calculation on the turbine blade model. The blade thermal - fluid - structure interaction calculation is a finite - element calculation, and the node attribute matrix is calculated, that is, the parameter data of the gas - turbine turbine blade, including pressure - field and temperature - field data, etc.

[0085] Build an adjacency matrix. For unstructured grids, its connection relationship can be expressed as G:

[0086] G = {N, E} (2);

[0087] where N is the set of grid nodes, each node represents a computational unit, and when input into the graph convolutional neural network, it represents a point on the adjacency matrix, and the nodes of the adjacency matrix correspond to the grid nodes, and E is the connection relationship between grid nodes.

[0088] The input of the adjacency matrix A in the graph convolutional network is defined as:

[0089]

[0090] S3: Reconstruction training:

[0091] Refer to Figure 3 , and use the above - established training set, which includes boundary conditions, node attribute matrix, and adjacency matrix. Train the graph convolutional neural network (GCN) in the generator. The GCN network uses the SAGEConv convolutional layer, and the aggregation and update process can be expressed in two steps:

[0092] The first step: Neighborhood feature aggregation.

[0093] Obtain the set of neighborhood nodes N(ν) of node ν through the adjacency matrix, N(ν) = {u|A νu = 1}.

[0094] Aggregate neighborhood features from the neighborhood N(ν) of node ν:

[0095]

[0096] where represents the feature vector or representation related to node v and all its neighborhood nodes N(ν), Denote the value or feature generated at the l-th layer related to node u, including boundary conditions and node attributes. u is a node adjacent to node v. The aggregation function AGG describes how to aggregate the features of the neighborhood N(ν), which can be implemented by mean aggregation as shown below:

[0097]

[0098] Step 2: Update the features of the central node.

[0099] Combine the aggregated neighborhood features with the self-features of the central node to generate the updated features. The expression is as follows:

[0100]

[0101] where, denotes the weight of the linear transformation of the node's own features, denotes the weight of the linear transformation of the neighborhood aggregation information, denotes the feature representation of node ν at the l-th layer. The non-linear activation function σ(·) is GeLU, and its mathematical definition is:

[0102]

[0103] At this stage, the input noise Z is zero, and the output is generated only depending on the sample point X:

[0104] Y recon = G(X, Z = 0; θ1) (8);

[0105] where, Y represents the output of the generator G, θ1 is the learnable parameter of the generator, and recon represents reconstruction.

[0106] The reconstruction loss is represented by the mean squared error (MSE):

[0107]

[0108] where, refers to the square of the two-norm.

[0109] Update the generator parameter θ1 through gradient descent:

[0110]

[0111] where, η1 is the learning rate of the generator.

[0112] Through the training of the reconstruction loss, the generator learns from the parameter X to the temperature field Y (1) and the pressure field Y (2)The basic mapping relationship. That is, the generator has learned the basic mapping relationship from the sampling results and the adjacency matrix to the temperature field and the pressure field.

[0113] S4: Use the above-mentioned Latin hypercube method and the given range to perform secondary sampling on the sampling parameters to generate a new parameter set X. Input it into the above generator to obtain the corresponding pressure field and temperature field parameters. Use the pressure field and temperature field parameters obtained by the above network to replace the random noise Y generated by the traditional normal distribution fake , which can be closer to the task objective. This data is called forged data.

[0114] S5: In the present invention, the discriminator is also composed of a graph convolutional neural network, and its main purpose is to determine the authenticity probability of the data. Make the data Y generated by the generator fake with a label of 0 (fake data), and the training set data Y real with a label of 1 (real data).

[0115] D(X,Y real ) → 1 (11);

[0116] D(X,Y fake ) → 0 (12);

[0117] S6: Refer to Figure 4 , and train the discriminator. The network of the discriminator still uses the SAGEConv convolutional layer, which is a binary classification model. The activation function uses GeLU. The output of the final layer is a single probability value, which is used to represent the authenticity of the input data. Generate the last layer of node features H from the graph convolutional network (L) , and obtain the full graph feature H through average pooling graph :

[0118] H graph = meanpool(H (L) ) (13);

[0119] Average pooling is specific to each feature dimension:

[0120]

[0121] where |N(v)| represents the neighborhood size of node v, and h u,k represents the k-dimensional eigenvalue of node u.

[0122] The final layer uses the activation function (Sigmoid) to calculate the output probability:

[0123] D(G,X) = σ(W out H graph + b out ) (15);

[0124] Among them: W out is the output weight, and b out is the bias.

[0125] The activation function Sigmoid is expressed as:

[0126]

[0127] Among them, x ∈ R refers to any real number input, and e -x refers to the natural exponential function. The Sigmoid function maps the input from -∞ to ∞ to a continuous value within the interval (0, 1).

[0128] The discriminator's discrimination output for real data is as follows:

[0129] D(Y real ) = D(X, Y real ; θ2) (17);

[0130] In the formula, D(Y real ) is the discriminator's output for real data Y real , and θ2 is the discriminator's learnable parameter;

[0131] The discriminator's discrimination output for forged data is as follows:

[0132] D(Y recon ) = D(G(X, Z = 0; θ1); θ2) (18);

[0133] In the formula, D(Y recon ) is the discriminator's output for forged data .

[0134] S7: Determine the loss function. The discriminator's loss function L D (θ2) is:

[0135] The discriminator's loss function L D is:

[0136]

[0137] Among them, D(X i , Y i ) represents the output probability of the discriminator for sample i, and y i is the discriminator's label, with the real sample being 1 and the forged sample being 0.

[0138] Update the discriminator parameter θ2 through gradient descent:

[0139]

[0140] Among them, η2 is the discriminator learning rate.

[0141] The convergence condition is:

[0142] Among them, is a hyperparameter representing the target value of the loss function. Different loss functions have different target values as convergence conditions.

[0143] The adversarial loss L of the generator G is:

[0144]

[0145] The total generator loss L gen is:

[0146] L gen = λ recon L recon + λ G L G (22);

[0147] Among them, λ recon and λ G are hyperparameters that respectively control the weights of the reconstruction loss and the adversarial loss.

[0148] The optimization objectives of both the generator and the discriminator are to minimize their own loss functions L gen and L D . Both are trained using the Adam optimizer:

[0149]

[0150] S8: After the above training steps, continuously update the learnable parameters θ1 and θ2 of the generator and the discriminator, and gradually improve their capabilities. Until the quality of the samples generated by the generator is high enough and the adversarial network reaches an approximately convergent state.

[0151] S9: Fix the learnable parameter θ1 of the generator, and use this trained network to create several sets of data, called high-quality forged data (X, Y).

[0152] S10: Refer to Figure 2 , and jointly use the high-quality forged dataset (X, Y) and the original training set (X, Y real ) as the input of the graph prediction network to train the physical field prediction network.

[0153] The training set is expressed as:

[0154] (X + X, Y + Y real ) → (X new , Y new ) (25);

[0155] The training method is expressed as:

[0156] Y new = F(X new ; θ3) (26);

[0157] Among them, θ3 is the learnable parameter of the physical field prediction network.

[0158] The network uses the SAGEConv convolutional layer, and the GELU activation function is used for each layer except the last layer. The last layer uses SAGEConv to output the physical field predicted by the graph convolution.

[0159] The loss function uses MSELoss:

[0160]

[0161] Among them, Y new,i represents the target value of the i-th sample, and Y i represents the predicted value of the i-th sample.

[0162] The Adam optimizer is used to optimize the loss function to minimize its value.

[0163] S11: After determining the parameter θ3, the physical field prediction network F is obtained, which has the prediction function.

[0164] Based on the same concept, the present invention also provides a gas turbine turbine blade parameter prediction system, including a construction module, a generator training module, a discriminator training module, an iteration module, and a prediction module.

[0165] The construction module is used to construct the turbine blade model of the gas turbine and specify the sampling parameters.

[0166] The generator training module is used to perform a first sampling on the sampling parameters of the turbine blade model under different conditions to obtain the first training data; perform blade thermal-fluid-solid coupling calculation on the turbine blade model to obtain the real data of the blade parameters; use the first training data as the input and the real data as the output to train the first graph convolutional neural network to obtain the generator.

[0167] The discriminator training module is used to perform a second sampling on the sampling parameters of the turbine blade model to obtain the second training data, input the second training data into the generator to obtain the forged data; train the second graph convolutional neural network through the first training data and the real data, and the second training data and the forged data to obtain the discriminator.

[0168] The iterative module is used to perform three samplings on the sampling parameters of the turbine blade model to obtain the third training data, input the third training data into the generator to obtain generated data; input the third training data and the generated data into the discriminator to obtain probability values; determine the loss function value of the discriminator based on the probability values, and iterate the generator and the discriminator through the loss function value to obtain the optimal generator and the optimal discriminator.

[0169] The prediction module is used to resample the sampling parameters of the turbine blade model and input the resampled data into the optimal generator to obtain the optimal blade parameter data; use the resampled data as the input and the optimal blade parameter data as the output to train the third graph convolutional neural network to obtain the physical field prediction network; predict the gas turbine turbine blade parameters through the physical field prediction network.

[0170] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0171] Obviously, those skilled in the art can make various changes and deformations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and deformations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and deformations.

Claims

1. A method for predicting parameters of a gas turbine turbine blade, characterized in that, It includes the following steps: Construct a turbine blade model of a gas turbine and specify sampling parameters; Perform a first sampling on the sampling parameters of the turbine blade model under different conditions to obtain first training data; Perform blade thermal-fluid-solid coupling calculation on the turbine blade model to obtain real data of blade parameters; Use the first training data as input and the real data as output to train the first graph convolutional neural network to obtain a generator; Perform a second sampling on the sampling parameters of the turbine blade model to obtain second training data, and input the second training data into the generator to obtain forged data; Use the first training data and real data, as well as the second training data and forged data to train the second graph convolutional neural network to obtain a discriminator; Perform a third sampling on the sampling parameters of the turbine blade model to obtain third training data, and input the third training data into the generator to obtain generated data; Input the third training data and the generated data into the discriminator to obtain a probability value; determine the loss function value of the discriminator based on the probability value, and iterate the generator and the discriminator through the loss function value to obtain an optimal generator and an optimal discriminator; Resample the sampling parameters of the turbine blade model, and input the resampled data into the optimal generator to obtain optimal blade parameter data; Use the resampled data as input and the optimal blade parameter data as output to train the third graph convolutional neural network to obtain a physical field prediction network; predict the turbine blade parameters of the gas turbine through the physical field prediction network.

2. The parameter prediction method for a gas turbine turbine blade according to claim 1, characterized in that Model the solid domain and fluid domain of the turbine blade of the gas turbine to obtain a turbine blade model; the sampling parameters include operating conditions parameters, machining parameters, and material parameters.

3. A method for predicting the parameters of a gas turbine turbine blade according to claim 1, characterized in that, Perform blade thermal-fluid-solid coupling calculation on the turbine blade model through a pneumatic batch file to obtain real data of blade parameters, and the blade parameters include a pressure field and a temperature field.

4. A gas turbine turbine blade parameter prediction method according to claim 1, characterized in that The step of using the first training data as input and the real data as output to train the first graph convolutional neural network to obtain a generator includes the following steps: Perform mesh division on the turbine blade model; Establish an adjacency matrix through the mesh-divided turbine blade model; Obtain the neighborhood node set of each node in the first graph convolutional neural network through the adjacency matrix; For each node, aggregate the neighborhood features from its neighborhood nodes to obtain the aggregated feature vector of the current node; the neighborhood features include the first training data and the real data; Combine the aggregated feature vector of the current node with the self-feature of the node to obtain the updated feature; Use the first graph convolutional neural network with updated node features as the generator; The output of the generator is as follows: Y recon = G(X, Z = 0; θ1); In the formula, Y is the output of the generator G, recon is the reconstruction, θ1 is the learnable parameter of the generator, X is the first training data, and Z is the noise; The reconstruction loss of the generator is as follows: Where, L recon is the reconstruction loss, N is the number of sample points, is the square of the second norm, Y real，i is the i-th real data, is the i-th generated data; Update the generator parameters through gradient descent.

5. A method for predicting gas turbine turbine blade parameters according to claim 4, characterized in that, The step of using the first training data and real data, as well as the second training data and forged data to train the second graph convolutional neural network includes the following steps: Make a true label 1 for the first training data and real data, and make a false label 0 for the second training data and forged data; Input the first training data, real data, true label 1, the second training data, forged data, and fake label 0 into the second graph convolutional neural network; Obtain the node feature H from the second graph convolutional network (L) , and obtain the global graph feature H through average pooling graph : H graph = meanpool(H (L) ); Average pool to each feature dimension: where h N(v)k is the representation of the pooled node v in the k-th feature dimension, |N(v)| is the neighborhood size of node v, and h u,k is the feature value of node u in the k-th dimension, d is the number of dimensions, and N(ν) is the neighborhood, which contains the set of all nodes directly connected to node v; Use an activation function to calculate the output probability; Take the trained second graph convolutional neural network as the discriminator; The discriminator's discrimination output for real data is as follows: D(Y real ) = D(X,Y real ; θ2); where D(Y real ) is the output of the discriminator for the real data Y real , and θ2 is the learnable parameter of the discriminator; The discriminator's discrimination output for forged data is as follows: D(Y recon ) = D(G(X, Z = 0; θ1); θ2); where D(Y recon ) is the output of the discriminator for the forged data .

6. The parameter prediction method for a gas turbine turbine blade according to claim 5, wherein The loss function of the generator is specifically as follows: L gen = λ recon L recon + λ G L G ; where L gen is the total generator loss, λ recon and λ G are hyperparameters, and L G is the adversarial loss; The loss function of the discriminator is specifically as follows: Where L D is the loss of the discriminator, D(X i , Y i ) is the output probability of the discriminator, y i is the label of the discriminator, X i is the i-th third sampling parameter, and Y i is the i-th generated data; Iterate the generator and discriminator through the loss function value, which specifically includes the following steps: Compare the loss function value with the target threshold; When the loss function value is greater than the target threshold, iterate the generator and discriminator based on the Adam optimizer, otherwise output the optimal learning parameters of the generator and discriminator.

7. A method for predicting parameters of a gas turbine turbine blade according to claim 1, characterized in that The training of the third graph convolutional neural network specifically includes the following steps: Use the resampled data, optimal blade parameter data, the first training data, and the real data of the blade parameters as the training set, where the training set is as follows: (X+X,Y+Y rea l )→(X new ,Y new ); In the formula, is the resampled data, X is the first training data, is the optimal blade parameter data, Y real is the real data, X new is the input of the training set, Y new is the output of the training set; Train through the following formula: Y new = F(X new ; θ3); In the formula, is the physical field prediction network, and θ3 is the learnable parameter of the physical field prediction network.

8. A gas turbine turbine blade parameter prediction system, characterized in that, Include: A construction module for constructing a turbine blade model of a gas turbine and specifying sampling parameters; A generator training module for sampling the sampling parameters of the turbine blade model once under different conditions to obtain the first training data; Perform blade thermal fluid-solid coupling calculation on the turbine blade model to obtain the real data of the blade parameters; Use the first training data as the input and the real data as the output to train the first graph convolutional neural network to obtain the generator; A discriminator training module for sampling the sampling parameters of the turbine blade model twice to obtain the second training data, and input the second training data into the generator to obtain forged data; Train the second graph convolutional neural network through the first training data and real data, as well as the second training data and forged data to obtain the discriminator; An iteration module for sampling the sampling parameters of the turbine blade model three times to obtain the third training data, and input the third training data into the generator to obtain generated data; Input the third training data and the generated data into the discriminator to obtain a probability value; Determine the loss function value of the discriminator based on the probability value, and iterate the generator and discriminator through the loss function value to obtain the optimal generator and optimal discriminator; A prediction module for resampling the sampling parameters of the turbine blade model and inputting the resampled data into the optimal generator to obtain the optimal blade parameter data; Use the resampled data as the input and the optimal blade parameter data as the output to train the third graph convolutional neural network to obtain the physical field prediction network; Predict the gas turbine turbine blade parameters through the physical field prediction network.

Images