Turbine blade inverse problem design method based on machine learning, computer readable storage medium and electronic device

Abstract

The present invention relates to the technical field of turbine blade optimization design, and in particular to a turbine blade inverse problem design method based on machine learning, a computer-readable storage medium, and an electronic device. The turbine blade inverse problem design method based on machine learning includes the following steps: using a first blade geometric parameter as an independent variable and a first blade aerodynamic parameter as a dependent variable to establish a data set from blade geometric parameters to blade aerodynamic parameters; creating a training model for turbine blade inverse problem design; using the data set to train the training model and conduct loss evaluation to determine the inverse problem design model; substituting a second blade aerodynamic parameter into the inverse problem design model to reversely predict the second blade geometric parameter. The present invention can calculate the geometric parameters of the blade according to the aerodynamic parameters of the blade, and at the same time, the turbine blade inverse problem design method based on machine learning also has the advantage of high calculation accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of turbine blade optimization design, and in particular to a turbine blade inverse problem design method based on machine learning, a computer-readable storage medium, and an electronic device. Background Art

[0002] High-efficiency turbine aerodynamic design can effectively reduce engine mass, improve engine efficiency, and enhance engine performance. The inverse problem is part of turbine aerodynamic design, also known as the blade geometry optimization and modification design problem, that is, under the conditions of some aerodynamic performance target parameters in a given flow field, such as velocity distribution, pressure distribution, load distribution, circulation distribution and other two-dimensional parameters, as well as one-dimensional total parameters such as efficiency, power, flow, etc., through the physical relationship between aerodynamic parameters and blade geometry, the blade geometry parameters are obtained through continuous iterative calculation.

[0003] In related technologies, after the target aerodynamic parameters of the turbine blade are given, the traditional high-precision numerical simulation method is usually used to continuously change the blade geometry and perform simulation calculations. The calculation accuracy is high, but the calculation amount is large and the calculation time is long. Although low-precision simulation can quickly obtain calculation results, it cannot be used for accurate flow field analysis. Experimental measurement can reflect the actual situation of the flow field, but the manpower, material resources and time costs are high, and it will be affected by measurement and manufacturing technology errors.

[0004] Machine learning provides a new entry point for the study of inverse problem design of turbine blades. As a subfield of current artificial intelligence technology, machine learning algorithms are algorithms that rely on patterns and statistical inferences. They enable computer systems to complete specific tasks without special programming, that is, soft programming. Researchers use machine learning algorithms to guide computers to train appropriate machine learning models using known data, and use the trained models to make predictions for new data in new situations. As an algorithm model in machine learning, neural networks have strong nonlinear mapping capabilities. It has a multi-layer network structure, each layer consists of multiple neurons, and neurons are connected by weights. The data is divided into training data and test data, and the model is trained and learned with the training data, the weights of the neurons are changed, so that it can obtain regression prediction capabilities and generate a model with generalization capabilities. The test data is data that the neural network has never seen, and is input into the trained model to test its prediction accuracy. The application of machine learning methods in the inverse problem design of turbine blades can quickly and accurately predict the geometric parameters given the aerodynamic parameters of the blades, reduce the amount of calculation, and improve the design efficiency. This is a new attempt outside the traditional numerical simulation calculation methods. Therefore, the following demand arises in this field: Can a blade inverse problem design method be designed? This design method can utilize the regression prediction ability of machine learning to more accurately predict the geometric parameters of the blades based on the aerodynamic parameters of the blades to be designed. Summary of the invention

[0005] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0006] To this end, an embodiment of the present invention proposes an inverse problem design method for turbine blades based on machine learning. The inverse problem design method for turbine blades can calculate the geometric parameters of the blades according to the aerodynamic parameters of the blades. At the same time, the inverse problem design method for turbine blades also has the advantage of high calculation accuracy.

[0007] According to an embodiment of the present invention, a turbine blade inverse problem design method based on machine learning includes the following steps: using a first blade geometric parameter as an independent variable and a first blade aerodynamic parameter as a dependent variable to establish a data set from blade geometric parameters to blade aerodynamic parameters; creating a training model for turbine blade inverse problem design; using the data set to train the training model and conduct loss evaluation to determine the inverse problem design model; substituting the second blade aerodynamic parameter into the inverse problem design model to reversely predict the second blade geometric parameter.

[0008] The inverse problem design method for turbine blades based on machine learning according to an embodiment of the present invention can calculate the geometric parameters of the blades according to the aerodynamic parameters of the blades. At the same time, the inverse problem design method for turbine blades also has the advantage of high calculation accuracy.

[0009] In some embodiments, the step of taking the first blade geometric parameter as an independent variable and the first blade aerodynamic parameter as a dependent variable to establish a data set from the blade geometric parameter to the blade aerodynamic parameter comprises: based on the value range of the blade geometric parameter, using a sampling algorithm to sample the value range to obtain a blade shape file composed of the first blade geometric parameter;

[0010] The blade shape file is calculated to obtain the aerodynamic parameters of the first blade; and the first blade geometric parameters are normalized with the aerodynamic parameters of the first blade corresponding to the first blade geometric parameters to obtain the data set.

[0011] In some embodiments, the creation of a training model for inverse problem design of turbine blades includes: creating a 4-layer neural network model consisting of an input layer, a first hidden layer, a second hidden layer and an output layer; setting the input layer and the first hidden layer as RBF layers; setting the second hidden layer and the output layer as fully connected layers.

[0012] In some embodiments, the using of the data set to train the training model and perform loss evaluation on the training model to determine the inverse problem design model includes: dividing the data set into training data and test data, wherein the test data includes the aerodynamic parameters of the third blade and the third blade geometric parameters corresponding to the aerodynamic parameters of the third blade; using the training data to train the 4-layer neural network model and perform loss evaluation on the loss to obtain the inverse problem design model; and using the test data to verify the inverse problem design model to determine the inverse problem design model.

[0013] In some embodiments, the use of the training data to train the 4-layer neural network model and perform loss evaluation to obtain the inverse problem design model includes: performing loss evaluation on the training model using a loss function; iteratively optimizing the loss function using an optimization algorithm; and evaluating the calculation results of the loss evaluation to obtain the inverse problem design model.

[0014] In some embodiments, the use of the test data to verify the inverse problem design model to determine the inverse problem design model includes: substituting the third blade aerodynamic parameter of the test data into the inverse problem design model to obtain the fourth blade geometric parameter; performing denormalization processing on the fourth blade geometric parameter to obtain the fifth blade geometric parameter; establishing a predicted geometric model based on the fifth blade geometric parameter; performing denormalization processing on the third blade geometric parameter of the test data to obtain the sixth geometric parameter; establishing a true geometric model based on the sixth blade geometric parameter; and comparing and evaluating the blade profiles of the predicted geometric model and the true geometric model at the same blade height to determine the inverse problem design model.

[0015] In some embodiments, after the step of substituting the third aerodynamic parameter of the test data into the inverse problem design model to obtain the sixth blade geometric parameter, it also includes: calculating the sixth blade geometric parameter to obtain the fourth aerodynamic parameter; and calculating the average error based on the third aerodynamic parameter and the fourth aerodynamic parameter.

[0016] In some embodiments, the activation functions of the input layer and the first hidden layer are radial basis functions; the activation function of the second hidden layer includes a ReLU function; and the activation function of the output layer includes a sigmoid function.

[0017] In some embodiments, the first blade geometric parameters include installation angle, inlet geometric angle, outlet geometric angle, leading edge pressure surface wedge angle, leading edge suction surface wedge angle, trailing edge wedge angle, rearward bend angle, and leading edge diameter geometric parameters, and the first aerodynamic parameters include efficiency, power, flow rate, outlet relative Mach number, outlet absolute Mach number, outlet relative airflow angle, outlet absolute airflow angle, and reaction degree.

[0018] According to a computer-readable storage medium of an embodiment of the present invention, computer instructions are stored thereon, and when the computer instructions are executed by a processor, the machine learning-based turbine blade inverse problem design method described in any of the above embodiments is implemented.

[0019] An electronic device according to an embodiment of the present invention includes: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to: execute the machine learning-based turbine blade inverse problem design method described in any of the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of a method for inverse problem design of a turbine blade based on machine learning according to some embodiments of the present invention;

[0021] Figure 2A is a flow chart of step S1 in a turbine blade inverse problem design method based on machine learning according to some embodiments of the present invention;

[0022] Figure 2B is a flow chart of step S3 in the inverse problem design method for turbine blades based on machine learning according to some embodiments of the present invention;

[0023] Figure 2C is a flow chart of step S33 in the inverse problem design method for turbine blades based on machine learning according to some embodiments of the present invention;

[0024] Figure 2D is a schematic flow chart of comparing aerodynamic parameters of a third blade and aerodynamic parameters of a fourth blade in a turbine blade inverse problem design method based on machine learning according to some embodiments of the present invention;

[0025] Figure 3 is a schematic diagram of a 4-layer neural network model provided according to some embodiments of the present invention;

[0026] Figure 4 is a comparison diagram of loss function curves during training and testing of a 4-layer neural network model provided in some embodiments of the present invention;

[0027] Figure 5 is a blade profile comparison diagram of a predicted geometric model and a true geometric model at a guide vane root section according to some embodiments of the present invention;

[0028] Figure 6 is a blade profile comparison diagram of a predicted geometric model and a true geometric model at a guide vane top section according to some embodiments of the present invention;

[0029] Figure 7is a blade profile comparison diagram of a predicted geometric model and a true geometric model at a root section of a moving blade according to some embodiments of the present invention;

[0030] Figure 8 is a blade profile comparison diagram of a predicted geometric model and a true geometric model in a middle section of a moving blade according to some embodiments of the present invention;

[0031] Fig. 9 It is a blade profile comparison diagram of the predicted geometric model and the true geometric model at the top section of the moving blade provided according to some embodiments of the present invention. DETAILED DESCRIPTION

[0032] Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0033] like Figure 1 As shown, the inverse problem design method of a turbine blade based on machine learning according to an embodiment of the present invention comprises the following steps:

[0034] S1: Using the first blade geometric parameter as the independent variable and the first blade aerodynamic parameter as the dependent variable, a data set from blade geometric parameter to blade aerodynamic parameter is established.

[0035] In some embodiments, the data set can be pre-constructed, or the first blade geometric parameter can be used as an input condition. The target output of the first blade aerodynamic parameter can be calculated by using a multi-stage S2 flow surface calculation program, so that the first blade geometric parameter and the data set of the first blade aerodynamic parameter can be obtained. In this embodiment, the first blade geometric parameter can include 8 input conditions, and the 8 input conditions are respectively the installation angle, the inlet geometric angle, the outlet geometric angle, the leading edge pressure surface wedge angle, the leading edge suction surface wedge angle, the trailing edge wedge angle, the rear bend angle, and the leading edge diameter geometric parameter. 8 target outputs can be obtained by calculating through the multi-stage S2 flow surface calculation program. The 8 target outputs, i.e., the first aerodynamic parameters, are respectively efficiency, power, flow, outlet relative Mach number, outlet absolute Mach number, outlet relative airflow angle, outlet absolute airflow angle, and reaction degree.

[0036] In some embodiments, Figure 2A As shown, the first blade geometric parameter is used as an independent variable, and the first blade aerodynamic parameter is used as a dependent variable to establish a data set from the blade geometric parameter to the blade aerodynamic parameter, including:

[0037] S11: Based on the value range of the blade geometric parameters, a sampling algorithm is used to sample the value range to obtain a blade shape file composed of the first geometric parameters.

[0038] It should be noted that the blade profile file is a specific blade geometric parameter obtained by a sampling algorithm within the blade geometric value range. Eight specific blade geometric parameters constitute one blade profile file. For example, as shown in Table 1, this embodiment selects five sections of the blade, which are the guide vane root, guide vane top, moving blade root, moving blade middle and moving blade top, and each section contains the eight geometric parameters. The specific values ​​of the eight geometric parameters of different sections can be uniformly sampled within the value range of Table 1 using the Latin hypercube sampling algorithm to obtain multiple blade profile files with eight specific geometric parameters as one blade profile file.

[0039] Table 1 Geometric parameter value range

[0040]

[0041] S12: Calculate the blade shape file to obtain the first aerodynamic parameters.

[0042] It should be noted that, in the calculation process using the multi-stage S2 flow surface calculation program, the turbine blade input boundary conditions of the multi-stage S2 flow surface calculation program are shown in Table 2.

[0043] Table 2S2 Program input boundary conditions

[0044] Boundary conditions Value unit Total inlet pressure 1012340 Pa total inlet temperature 1330 K Outlet static pressure 300000 Pa Outlet radius 295.75 mm

[0045] S13: Normalize the first geometric parameter and the first aerodynamic parameter corresponding to the first geometric parameter to obtain a data set.

[0046] It should be noted that, in this embodiment, the normalization formula may be:

[0047]

[0048] In the above formula, x ** is the parameter in the normalized data set; x max is the upper limit of the range of values ​​of x; min is the lower limit of the value range of x; x is the first geometric parameter and the first aerodynamic parameter that need to be normalized.

[0049] S2: Create a training model for the inverse problem design of turbine blades.

[0050] It should be noted that, in the present invention, there is no restriction on the construction and use platform of the training model. For example, it can be constructed on a matlab platform or on other platforms. In this embodiment, the training model uses the Keras machine learning library to build the model, and its backend is the TensorFlow machine learning platform.

[0051] It should be noted that in the present invention, there is no limitation on the algorithm of the training model. For example, the training model can be established based on the BP neural network algorithm, or based on the BRF neural network algorithm, or based on the BP neural network and the BRF neural network, or based on the random forest algorithm.

[0052] In some embodiments, the training model can be constructed based on the BP neural network and the BRF neural network, as shown in FIG. Figure 3 As shown in the figure, the training model is constructed by a 4-layer neural network model. The 4-layer neural network includes an input layer, a first hidden layer, a second hidden layer and an output layer from the data set input end to the calculation result output end. The input layer and the first hidden layer are RBF layers, and the activation functions of the input layer and the first hidden layer are radial basis functions. The second hidden layer and the output layer are fully connected layers. The activation function of the second hidden layer includes the ReLU function, and the activation function of the output layer includes the sigmoid function.

[0053] S3: Use the dataset to train the training model and evaluate the loss to determine the inverse problem design model.

[0054] It should be noted that during the training process, there is no limitation on the types of loss function and optimization algorithm of the training model. For example, the loss function can be the mean absolute error or the mean square error. In some embodiments, the loss function is defined by the mean square error. The optimization algorithm can be the Nesterov gradient acceleration method or the Adam optimizer. In some embodiments, the optimization algorithm uses the Adam optimizer.

[0055] In some embodiments, the weight calculation parameter betas of the input layer and the first hidden layer is set to 0.2, the number of neural units in the input layer is 400, the number of neural units in the first hidden layer is 400, the number of neural units in the second hidden layer is 256, the number of neural units in the output layer is 40, the number of training cycles (epochs) of the training model is 50,000, and the mini-batch size is 256. It can be understood that the above description is only exemplary and is not limited to this in some embodiments.

[0056] Figure 2B Schematic diagram of the process of step S3 in the inverse problem design method of turbine blades based on machine learning according to some embodiments of the present invention; Figure 2B As shown, the method of using the data set to train the training model and evaluate the loss to determine the inverse problem design model includes:

[0057] S31: Divide the data set into training data and test data, wherein the test data includes aerodynamic parameters of the third blade and third geometric parameters corresponding to the aerodynamic parameters of the third blade.

[0058] It should be noted that there is no restriction on the number and proportion of training data and test data. In some embodiments, a total of 9781 sets of data are selected as data sets, of which 8803 sets of data are used as training data for training the training model to give it generalization ability, and 978 sets of data are used as test data. The test data, as data that the training model has not seen, is used to verify the model performance.

[0059] S32: Use the training data to train the 4-layer neural network model and evaluate the loss to obtain the inverse problem design model.

[0060] S33: Use the test data to verify the inverse problem design model to determine the inverse problem design model.

[0061] Figure 4 4 is a comparison diagram of loss function curves during training and testing of a 4-layer neural network model provided according to some embodiments of the present invention; Figure 4 As shown in the figure, the thick solid line and the thin solid line represent the average mean square error (MSE) of the training data and the test data respectively. The final average mean square error (MSE) of the training is 0.00038, and the average mean square error (MSE) of the test is 0.00035. It can be seen that the overall prediction error of the model is very small, the loss of the test data is the same as the loss of the training data, and there is no overfitting phenomenon, which shows that the inverse problem design model has good prediction ability.

[0062] Figure 2C Schematic diagram of the process of step S33 in the inverse problem design method of turbine blades based on machine learning according to some embodiments of the present invention; Figure 2C As shown, the method of verifying the inverse problem design model by using the test data to determine the inverse problem design model includes:

[0063] S331: Substitute the aerodynamic parameters of the third blade of the test data into the inverse problem design model to obtain the geometric parameters of the fourth blade.

[0064] S332: Perform denormalization processing on the fourth blade geometric parameters to obtain the fifth blade geometric parameters.

[0065] It should be noted that, in this embodiment, the denormalization formula is:

[0066] x=x * ×(x max -x min )+x min

[0067] In the above formula, x is the fifth geometric parameter after denormalization, x *is the fourth geometric parameter obtained based on the test data; x max is the upper limit of the range of values ​​of x; min The lower limit of the range of values ​​for x.

[0068] S333: Establish a predictive geometric model based on the fifth blade geometric parameters.

[0069] S334: Denormalize the third blade geometric parameter of the test data to obtain a sixth geometric parameter.

[0070] S335: Establish a true value geometric model based on the sixth blade geometric parameters.

[0071] S336: Compare and evaluate the blade profiles of the predicted geometric model and the true geometric model at the same blade height to determine the inverse problem design model.

[0072] Comparison of blade profiles of the predicted geometry model and the true geometry model in five design sections, such as Figures 5 to 9 As shown in the figure, the five sections of the predicted geometric model and the true geometric model have a very high degree of overlap in the leading edge, trailing edge, pressure surface profile, and suction surface profile. It can be considered that the inverse problem design model has achieved a high-precision prediction of the blade geometry.

[0073] Figure 2D is a flow chart of comparing the aerodynamic parameters of the second blade and the aerodynamic parameters of the third blade in the inverse problem design method of a turbine blade based on machine learning according to some embodiments of the present invention; Figure 2D ,and Figure 2C The difference is that in the inverse problem design method of turbine blades based on machine learning in this embodiment, after substituting the second aerodynamic parameter of the test data into the inverse problem design model to obtain the third blade geometric parameter step, the method further includes:

[0074] S337: Calculate the sixth blade geometric parameter to obtain a fourth aerodynamic parameter;

[0075] It should be noted that the sixth blade geometric parameters can be substituted into the multi-stage S2 flow surface calculation program to obtain the fourth aerodynamic parameters.

[0076] S338: Calculate an average error based on the third aerodynamic parameter and the fourth aerodynamic parameter.

[0077] Table 3 is a comparison of 6 groups of aerodynamic parameters of the third blade and the calculated aerodynamic parameters of the fourth blade, to prove whether the aerodynamic parameters of the fourth blade are close to the aerodynamic parameters of the third blade, wherein the upper row of the same example is the third aerodynamic parameters.

[0078] Table 3 Calculation results of the aerodynamic parameters of the third blade and the fourth blade

[0079]

[0080] The average errors of the 6 groups of aerodynamic parameters of the third blade and the calculated aerodynamic parameters of the fourth blade in Table 3 are calculated, and the results are as follows: the average errors of efficiency, power, flow, outlet relative Mach number, outlet absolute Mach number, outlet relative airflow angle, outlet absolute airflow angle, and reaction degree are 0.15%, 0.57%, 0.82%, 0.67%, 1.81%, 0.32%, 1.37%, and 2.13%, respectively. Based on the overall average error of the above 6 groups of examples of 0.98%, it can be considered that the aerodynamic parameters of the third blade are very close to the aerodynamic parameters of the fourth blade. Since the aerodynamic parameters of the fourth blade are calculated based on the geometric parameters of the third blade, and the geometric parameters of the fourth blade are the results predicted based on the inverse problem design model, the inverse problem design model prediction of the inverse problem design method of turbine blades based on machine learning is more accurate.

[0081] S4: Substitute the aerodynamic parameters of the second blade into the inverse problem design model to reversely predict the geometric parameters of the second blade.

[0082] It should be noted that the second blade aerodynamic parameters are blade aerodynamic parameters that match the blade to be designed, and the second blade geometric parameters, as target output, are geometric parameters that are similar to the blade to be designed.

[0083] A specific embodiment of a computer-readable storage medium of the present invention is a computer-readable storage medium having computer instructions stored thereon, which implement the machine learning-based turbine blade inverse problem design method when the computer instructions are executed by a processor.

[0084] A specific embodiment of an electronic device of the present invention, the electronic device comprises: a processor; a memory for storing processor executable instructions; wherein the processor is configured to: execute the machine learning-based turbine blade inverse problem design method.

[0085] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0086] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory. Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and cannot be construed as limitations on the present invention, and those of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for inverse problem design of turbine blades based on machine learning, characterized in that: include: Taking the first blade geometric parameter as an independent variable and the first blade aerodynamic parameter as a dependent variable, a data set from blade geometric parameter to blade aerodynamic parameter is established; Create a training model for inverse design of turbine blades; Using the data set to train the training model and perform loss evaluation to determine an inverse problem design model; Substituting the aerodynamic parameters of the second blade into the inverse problem design model to reversely predict the geometric parameters of the second blade; Wherein, the first blade geometric parameters include installation angle, inlet geometric angle, outlet geometric angle, leading edge pressure surface wedge angle, leading edge suction surface wedge angle, trailing edge wedge angle, back bend angle, leading edge diameter geometric parameters, and the first blade aerodynamic parameters include efficiency, power, flow rate, outlet relative Mach number, outlet absolute Mach number, outlet relative airflow angle, outlet absolute airflow angle and reaction degree; The first blade geometric parameter is a geometric parameter of a blade cross section, and the blade cross section includes a guide vane root, a guide vane top, a moving blade root, a moving blade middle, and a moving blade top; The method of using the first blade geometric parameter as an independent variable and the first blade aerodynamic parameter as a dependent variable to establish a data set from the blade geometric parameter to the blade aerodynamic parameter includes: Based on the value range of the geometric parameters of each cross section, a Latin hypercube sampling algorithm is used to uniformly sample values ​​to obtain a blade shape file composed of the first blade geometric parameters; Calculating the blade shape file to obtain the aerodynamic parameters of the first blade; Normalizing the first blade geometric parameter and the first blade aerodynamic parameter corresponding to the first blade geometric parameter to obtain the data set; The method of creating a training model for inverse problem design of turbine blades comprises: Create a 4-layer neural network model consisting of an input layer, a first hidden layer, a second hidden layer, and an output layer; Set the input layer and the first hidden layer to RBF layers; Set the second hidden layer and output layer to fully connected layers; The step of training the training model and evaluating the loss using the data set to determine the inverse problem design model comprises: Dividing the data set into training data and test data, wherein the test data includes aerodynamic parameters of the third blade and geometric parameters of the third blade corresponding to the aerodynamic parameters of the third blade; Using the training data to train and evaluate the loss of the 4-layer neural network model to obtain the inverse problem design model; Verifying the inverse problem design model using the test data to determine the inverse problem design model; The method of using the training data to train the four-layer neural network model and conduct loss evaluation to obtain the inverse problem design model includes: Use the loss function to evaluate the loss of the training model; Iteratively optimize the loss function using an optimization algorithm; Evaluating the calculation results of the loss assessment to obtain the inverse problem design model; The step of verifying the inverse problem design model by using the test data to determine the inverse problem design model includes: Substituting the aerodynamic parameters of the third blade of the test data into the inverse problem design model to obtain the geometric parameters of the fourth blade; Denormalizing the fourth blade geometric parameters to obtain the fifth blade geometric parameters; Establishing a prediction geometry model based on the fifth blade geometry parameter; Denormalize the third blade geometric parameter of the test data to obtain a sixth geometric parameter; Based on the sixth blade geometric parameters, a true value geometric model is established; The predicted geometric model and the true geometric model are compared and evaluated on the blade profile at the same blade height to determine the inverse problem design model.

2. The inverse problem design method for turbine blades based on machine learning according to claim 1, characterized in that: After the step of substituting the third aerodynamic parameter of the test data into the inverse problem design model to obtain the sixth blade geometric parameter, the method further includes: Calculating the sixth blade geometric parameters to obtain fourth aerodynamic parameters; An average error is calculated based on the third aerodynamic parameter and the fourth aerodynamic parameter.

3. The inverse problem design method for turbine blades based on machine learning according to claim 2, characterized in that: The activation functions of the input layer and the first hidden layer are both radial basis functions; the activation function of the second hidden layer includes a ReLU function; and the activation function of the output layer includes a sigmoid function.

4. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the method according to any one of claims 1 to 3 is implemented.

5. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to execute the method according to any one of claims 1 to 3.

Images