An adaptive optimization method and system for blade parameters of a rim-driven propeller

Abstract

An adaptive optimization method for rim propeller blade parameters includes the following steps: Step S2: Parametric modeling of the RDT blade using characteristic curves to express it, resulting in at least two RDT blade models; Step S3: Generating several sample schemes for the RDT blade model using the Sobol sequence algorithm, and collecting results for each sample scheme; Step S4: Setting the objective function and constraints, and training the validation model based on the sample schemes and the corresponding collected results, resulting in a trained validation model; Step S5: Combining the validation model, obtaining the optimal RDT blade model using a second-generation non-dominated sorting genetic algorithm, and validating different optimal design schemes using CFD methods, comparing the performance differences between different optimal RDT blade models.

Description

Technical Field

[0001] This invention relates to the field of blade design technology, and in particular to an adaptive optimization method and system for blade parameters of a rim propeller. Background Technology

[0002] Currently, rim-driven thrusters (RDTs) are widely used in marine propulsion systems due to their unique rimless design. The main advantages of rim-driven thrusters include compact structure, reduced vibration and noise, and reduced cavitation. As the core component of a rim-driven thruster, the design quality of the RDT blade directly affects the overall performance of the thruster. However, traditional blade design methods often rely on empirical formulas or fixed design processes, making it difficult to fully consider the impact of different design parameters on blade performance. This can lead to the inability to find the optimal blade shape and parameter combination during the design process, thus limiting the improvement of thruster performance.

[0003] Existing design methods often fail to quickly and accurately determine efficient blade designs during the design phase, resulting in long design cycles, high costs, and final blade designs that may not fully meet practical requirements. Therefore, a method that can automatically optimize blade parameters during the design process, improving design efficiency and blade performance, is urgently needed. Summary of the Invention

[0004] To address the aforementioned shortcomings, the present invention aims to propose an adaptive optimization method and system for blade parameters of a rim-driven propeller, thereby solving the problem in the prior art that requires continuous manual testing to obtain optimized blade parameters and greatly improving the efficiency of obtaining optimization schemes.

[0005] To achieve this objective, the present invention adopts the following technical solution: an adaptive optimization method for blade parameters of a rim-driven propeller, comprising the following steps:

[0006] Step S1: Parametric modeling of the RDT blade is performed by expressing it through characteristic curves, resulting in at least two RDT blade models;

[0007] Step S2: Use the Sobol sequence algorithm to generate several sample schemes for the RDT blade model, and collect the results for each sample scheme;

[0008] Step S3: Set the objective function and constraints, and train the validation model based on several sample schemes and the corresponding collection results to obtain the trained validation model;

[0009] Step S4: Combine the validation model and use the second-generation non-dominated sorting genetic algorithm to obtain the optimal RDT blade model. Then, use the CFD method to validate different optimal design schemes and compare the performance differences between different optimal RDT blade models.

[0010] Preferably, the specific steps of step S1 are as follows:

[0011] Based on the requirements of RDT blade settings, the corresponding characteristic curves were selected. The parametric modeling of the RDT blades was completed by introducing characteristic curves based on the NSRDC-NACA66-mod airfoil. The characteristic curves include B-Spline curves and F-Spline curves.

[0012] Preferably, the specific steps of step S2 are as follows:

[0013] Determine the parameter space for sampling;

[0014] A series of sample points are generated within a defined parameter space based on the Sobol sequence algorithm;

[0015] The open water efficiency was calculated using sample points. Cavitation initiation risk assessment index Total torque coefficient 10 The mutual influence relationship between them; set the mutual influence relationship as the result of the collection.

[0016] Preferably, the objective function and constraints in step S3 are as follows:

[0017] ;

[0018] in For open water efficiency, The minimum pressure coefficient, For emptying, The torque coefficient,

[0019] The values ​​of the control points for the characteristic curve of the blade design parameters are given, where These represent the control points of the characteristic curve for the blade design parameters, which specifically include chord length and thickness. This represents a reference value for the torque coefficient. The value is taken at the point where the blade tip is connected to the rim, where This indicates the camber and pitch at the blade tip where the rim connects to the blade.

[0020] Preferably, the specific steps for training the verification model based on the collected results in step S3 are as follows:

[0021] Step S31: Use the sample scheme as the input layer variable and the result as the output layer variable;

[0022] Step S32: Use the mean squared error loss function to measure the difference between the predicted value and the actual value;

[0023] For each sample scheme, calculate the square of the difference between the predicted value and the actual value, and then average the results over all sample schemes to obtain the loss function value. Where N is the number of sample schemes, It is the predicted value of the nth sample. It is the actual value of the nth sample;

[0024] Step S33: Based on the loss function value, use the backpropagation algorithm to update the connection weights and biases of the neural network; starting from the output layer, calculate the partial derivative of the loss function with respect to each connection weight and bias, and then adjust the connection weights and biases according to the magnitude and direction of the partial derivatives so that the loss function value gradually decreases;

[0025] Step S34: Repeat steps S32 to S33 until the loss function value is less than the loss threshold;

[0026] Step S35: Obtain the coefficient of determination and determine whether the coefficient of determination is greater than the coefficient threshold. If it is less than the threshold, reconstruct the validation model. If it is greater than the threshold, output the validation model.

[0027] The determination coefficients are as follows:

[0028] ;

[0029] in This represents the average of the true values.

[0030] Preferably, the process of using the second-generation non-dominated sorting genetic algorithm in step S4 is as follows:

[0031] Step S41: First, determine the population individuals. Based on the results and the validation model, determine the initial population individuals and set the initial population size.

[0032] Step S42: For each individual in the population, calculate its fitness value using the validation model according to the set objective function;

[0033] Step S43: Perform non-dominated ranking on the individuals in the population, and divide the individuals into different non-dominated levels according to the relationship of domination and subordination;

[0034] Step S44: Based on the non-dominated sorting and crowding distance calculation results, select a subset of individuals as parents for the next generation; perform crossover on the selected parent individuals to generate new offspring individuals, and perform mutation on the newly generated offspring individuals.

[0035] Step S45: Repeat step S44 above to continuously perform iterative optimization. The iteration process continues until the number of iterations meets the threshold.

[0036] Preferably, the specific CFD operating mode parameters in step S4 include the following steps:

[0037] Obtain the operational procedures required for simulating a rim-driven propeller;

[0038] The simulation process is recorded as a .java file using macro commands.

[0039] By combining the sim source files and .java files, the STAR CCM+ solver is called to process the design schemes in batches.

[0040] Perform computational domain generation, mesh generation, and export the computation results to obtain performance simulation results.

[0041] An adaptive optimization system for blade parameters of a rim-driven propeller, using the aforementioned adaptive optimization method for blade parameters of a rim-driven propeller, includes a modeling module, a sample generation module, a model training module, and a verification module;

[0042] The modeling module is used to perform parametric modeling of RDT blades by expressing them through feature curves, resulting in at least two RDT blade models.

[0043] The sample generation module is used to generate several sample schemes of the RDT blade model using the Sobol sequence algorithm, and to collect the results of each sample scheme.

[0044] The model training module is used to set the objective function and constraints, and to train the validation model based on several sample schemes and the corresponding collection results to obtain the trained validation model.

[0045] The verification module is used to: combine the verification model, use the second-generation non-dominated sorting genetic algorithm to obtain the optimal RDT blade model, and verify different optimal design schemes through CFD methods to compare the performance differences between different optimal RDT blade models.

[0046] One of the above technical solutions has the following advantages or beneficial effects: In this invention, by using parametric modeling and the Sobol sequence algorithm, a large number of sample solutions can be quickly generated, and the results can be collected and analyzed. This greatly improves design efficiency and shortens the R&D cycle. An approximate model is obtained by setting an objective function and constraints to train a BP neural network model, and then a second-generation non-dominated sorting genetic algorithm (NSGA-II) is used to search for the optimal design solution, thereby achieving adaptive optimization of the blade under different working conditions. This method can be applied to the design of different types of RDT blades and can be extended and improved according to actual needs. This helps to promote the continuous development and innovation of related technologies. In this way, the propulsion efficiency and stability of the rim propeller under different working environments can be significantly improved, thereby enhancing the overall performance and reliability. Attached Figure Description

[0047] Figure 1 This is a flowchart of one embodiment of the method of the present invention.

[0048] Figure 2 This is a schematic diagram of the structure of one embodiment of the system of the present invention. Detailed Implementation

[0049] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0050] In the description of embodiments of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of embodiments of the present invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0051] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0052] like Figures 1-2As shown, an adaptive optimization method for blade parameters of a rim-driven propeller includes the following steps:

[0053] Step S1: Parametric modeling of the RDT blade is performed by expressing it through characteristic curves, resulting in at least two RDT blade models;

[0054] Step S2: Use the Sobol sequence algorithm to generate several sample schemes for the RDT blade model, and collect the results for each sample scheme;

[0055] Step S3: Set the objective function and constraints, and train the validation model based on several sample schemes and the corresponding collection results to obtain the trained validation model;

[0056] Step S4: Combining the validation model, the optimal RDT blade model is obtained using a second-generation non-dominated sorting genetic algorithm. Different optimal design schemes are then validated using CFD methods (Computational Fluid Dynamics, a numerical analysis method that uses computers to simulate fluid flow and heat transfer processes), and the performance differences between different optimal RDT blade models are compared.

[0057] This invention utilizes parametric modeling and the Sobol sequence algorithm to rapidly generate a large number of sample designs, enabling result collection and analysis. This significantly improves design efficiency and shortens the R&D cycle. An approximate model is obtained by training a BP neural network model with a set objective function and constraints. Then, a second-generation non-dominated sorting genetic algorithm (NSGA-II) is used to search for the optimal design scheme, thereby achieving adaptive optimization of the blade under different operating conditions. This method can be applied to different types of RDT blade designs and can be extended and improved according to actual needs. This contributes to the continuous development and innovation of related technologies. In this way, the propulsion efficiency and stability of rim propellers under different working environments can be significantly improved, thereby enhancing overall performance and reliability.

[0058] The following is an example: First, the user can use the feature curve to express the RDT blade for parametric modeling. At this time, multiple different RDT blade models can be obtained, such as blade model 1, blade model 2, blade model 3, ... blade model n, etc.

[0059] Then, the Sobol sequence algorithm can be used to generate schemes for the RDT blade model. For example, in the first sample scheme of blade model 1, the blade length is X1 and the width is Y1. In the second sample scheme, the blade twist is set to X2 and the length is set to Y2. Multiple different sample schemes can be generated using the Sobol sequence algorithm. Simultaneously, corresponding results are collected. For example, in this optimization, if the open water efficiency needs to be optimized, the open water efficiency in different sample schemes is obtained. Then, the corresponding adopted results are obtained.

[0060] The sample schemes and their results are then input into the training model for training. Once trained, the model can output corresponding results based on the sample schemes. Therefore, the optimal design scheme for each RDT blade model can be found using a second-generation non-dominated sorting genetic algorithm. Finally, the performance of each optimal RDT blade model is compared to select the ideal blade design scheme.

[0061] Preferably, the specific steps of step S1 are as follows:

[0062] Based on the requirements of RDT blade settings, the corresponding characteristic curves were selected. The parametric modeling of the RDT blades was completed by introducing characteristic curves based on the NSRDC-NACA66-mod airfoil. The characteristic curves include B-Spline curves and F-Spline curves.

[0063] The B-Spline curve exhibits good local controllability and smoothness, while the F-Spline curve has advantages in fitting certain specific shapes. To further model the parameters to better reflect reality, the actual geometry and design requirements of the blade are considered. If the blade geometry is complex and high-precision fitting is required, the F-Spline curve, which emphasizes fitting complex shapes, can be chosen. Conversely, if local control and smoothness have a greater impact on blade performance, the B-Spline curve can be selected. By introducing a fully parametric modeling method that uses characteristic curves to express the RDT hydraulic components, the geometric characteristics of the blade can be described more accurately. This allows for obtaining more realistic test results during subsequent CFD method verification.

[0064] In one embodiment of the present invention, the NSRDC-NACA66-mod airfoil is used for parametric modeling because the NSRDC-NACA66-mod airfoil is widely used and can better meet the design requirements of rim propeller blades. Although other airfoil types may also have some applicability, the NSRDC-NACA66-mod has abundant experimental data, strong application capabilities, and is suitable for most scenario requirements.

[0065] It is worth mentioning that when selecting the B-Spline curve, in order to prevent the default straight closed airfoil curve in the parametric modeling software from causing an uneven trailing edge connection, it is necessary to define the range of the blade in the circumferential (U direction) and radial (V direction) directions to extract the suction and pressure surfaces. Then, the guide edge and trailing edge between the two surfaces are filled using the Fillet command. At the same time, the blade tip part near the hub is filled using a Feature, and the Feature at the blade tip is compiled in a custom way.

[0066] Preferably, the specific steps of step S2 are as follows:

[0067] Determine the parameter space for sampling; that is, determine which blade parameters need to be sampled. These parameters may include blade geometry, shape parameters, etc.

[0068] A series of sample points are generated within a defined parameter space based on the Sobol sequence algorithm; it is worth mentioning that each sample point needs to correspond to a set of RDT blade models.

[0069] The open water efficiency was calculated using sample points. Cavitation initiation risk assessment index Total torque coefficient 10 The mutual influence relationship between them; set the mutual influence relationship as the result of the collection.

[0070] Preferably, the objective function and constraints in step S3 are as follows:

[0071] ;

[0072] in For open water efficiency, The minimum pressure coefficient, For emptying, The torque coefficient,

[0073] The values ​​of the control points for the characteristic curve of the blade design parameters are given, where These represent the control points of the characteristic curve for the blade design parameters, which specifically include chord length and thickness. This represents a reference value for the torque coefficient. The value is taken at the point where the blade tip is connected to the rim, where This indicates the camber and pitch at the blade tip where the rim connects to the blade.

[0074] Preferably, the specific steps for training the verification model based on the collected results in step S3 are as follows:

[0075] Step S31: Use the sample scheme as the input layer variable and the result as the output layer variable;

[0076] In one embodiment of the present invention, the neural network for training and validating the model includes the number of neurons in the input layer, hidden layers, and output layer. The neural network consists of an input layer, two hidden layers, and an output layer. The connection weights and biases of the neural network are initialized. For each sample data, the design parameter values ​​of the input layer are passed to the hidden layer through the connection weights. The output values ​​of the hidden layer are then passed to the output layer through the connection weights. Similarly, the input values ​​of the neurons in the output layer are calculated, and the predicted values ​​of the output layer (predicted open-water efficiency, cavitation initiation risk assessment index, total thrust coefficient, and total torque coefficient) are obtained through an activation function.

[0077] Step S32: Use the mean squared error loss function to measure the difference between the predicted value and the actual value;

[0078] For each sample scheme, calculate the square of the difference between the predicted value and the actual value, and then average the results over all sample schemes to obtain the loss function value. Where N is the number of sample schemes, It is the predicted value of the nth sample. It is the actual value of the nth sample;

[0079] Step S3: Based on the loss function value, use the backpropagation algorithm to update the connection weights and biases of the neural network; starting from the output layer, calculate the partial derivative of the loss function with respect to each connection weight and bias, and then adjust the connection weights and biases according to the magnitude and direction of the partial derivatives so that the loss function value gradually decreases;

[0080] Step S34: Repeat steps S32 to S33 until the loss function value is less than the loss threshold;

[0081] Step S35: Obtain the coefficient of determination and determine whether the coefficient of determination is greater than the coefficient threshold. If it is less than the threshold, reconstruct the validation model. If it is greater than the threshold, output the validation model.

[0082] The determination coefficients are as follows:

[0083] ;

[0084] in This represents the average of the true values.

[0085] Preferably, the process of using the second-generation non-dominated sorting genetic algorithm in step S4 is as follows:

[0086] Step S41: First, determine the population individuals. Based on the results and the validation model, determine the initial population individuals and set the initial population size.

[0087] Step S42: For each individual in the population, calculate its fitness value using the validation model according to the set objective function;

[0088] The validation model quickly predicts the corresponding output variable values ​​(open water efficiency, cavitation initiation risk assessment index, etc.) based on the individual's design parameters. These predicted values ​​serve as the basis for evaluating the individual's performance, thereby determining the individual's fitness value.

[0089] Step S43: Perform non-dominated ranking on the individuals in the population, and divide the individuals into different non-dominated levels according to the dominance and subordination relationship; in order to maintain the diversity of the population, calculate the crowding distance of each individual.

[0090] Step S44: Based on the non-dominated sorting and crowding distance calculation results, select a subset of individuals as parents for the next generation of the population; perform crossover on the selected parent individuals to generate new offspring individuals, thereby increasing population diversity. Perform mutation on the newly generated offspring individuals to further increase population diversity and search capability;

[0091] Step S45: Repeat step S44 above to continuously perform iterative optimization. The iteration process continues until the number of iterations meets the threshold.

[0092] Finally, the optimal individual is selected from the final population as the optimal design scheme for the hydraulic components of the rim propeller. The optimal individual is the one with the best performance in the objective function space under the given constraints, and its corresponding design parameters are the parameters of the optimal design scheme.

[0093] Preferably, the specific CFD operating mode parameters in step S4 include the following steps:

[0094] Obtain the operational procedures required for simulating a rim-driven propeller;

[0095] The simulation process is recorded as a .java file using macro commands.

[0096] By combining the sim source files and .java files, the STAR CCM+ solver is called to process the design schemes in batches.

[0097] Perform computational domain generation, mesh generation, and export the computation results to obtain performance simulation results.

[0098] An adaptive optimization system for blade parameters of a rim-driven propeller, using the aforementioned adaptive optimization method for blade parameters of a rim-driven propeller, includes a modeling module, a sample generation module, a model training module, and a verification module;

[0099] The modeling module is used to perform parametric modeling of RDT blades by expressing them through feature curves, resulting in at least two RDT blade models.

[0100] The sample generation module is used to generate several sample schemes of the RDT blade model using the Sobol sequence algorithm, and to collect the results of each sample scheme.

[0101] The model training module is used to set the objective function and constraints, and to train the validation model based on several sample schemes and the corresponding collection results to obtain the trained validation model.

[0102] The verification module is used to: combine the verification model, use the second-generation non-dominated sorting genetic algorithm to obtain the optimal RDT blade model, and verify different optimal design schemes through CFD methods to compare the performance differences between different optimal RDT blade models.

[0103] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0104] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An adaptive optimization method for blade parameters of a rim-driven propeller, characterized in that, Includes the following steps: Step S1: Parametric modeling of the RDT blade is performed by expressing it through characteristic curves, resulting in at least two RDT blade models; Step S2: Use the Sobol sequence algorithm to generate several sample schemes for the RDT blade model, and collect the results for each sample scheme; Step S3: Set the objective function and constraints, and train the validation model based on several sample schemes and the corresponding collection results to obtain the trained validation model; The objective function and constraints in step S3 are as follows: ； in For open water efficiency, Minimum pressure coefficient, For emptying, The torque coefficient, The values ​​of the control points for the characteristic curve of the blade design parameters are given, where These represent the control points of the characteristic curve for the blade design parameters, which specifically include chord length and thickness. This represents a reference value for the torque coefficient. The value is taken at the point where the blade tip is connected to the rim, where This indicates the camber and pitch at the blade tip where the rim connects; Step S4: Combine the validation model and use the second-generation non-dominated sorting genetic algorithm to obtain the optimal RDT blade model. Then, use the CFD method to validate different optimal design schemes and compare the performance differences between different optimal RDT blade models.

2. The adaptive optimization method for blade parameters of a rim-driven propeller according to claim 1, characterized in that, The specific steps of step S1 are as follows: Based on the requirements of RDT blade settings, the corresponding characteristic curves were selected. The parametric modeling of the RDT blades was completed by introducing characteristic curves based on the NSRDC-NACA66-mod airfoil. The characteristic curves include B-Spline curves and F-Spline curves.

3. The adaptive optimization method for blade parameters of a rim-driven propeller according to claim 1, characterized in that, The specific steps of step S2 are as follows: Determine the parameter space for sampling; A series of sample points are generated within a defined parameter space based on the Sobol sequence algorithm; The open water efficiency was calculated using sample points. Cavitation initiation risk assessment index Total torque coefficient 10 The mutual influence relationship between them; setting the mutual influence relationship as the result of the data collection. Minimum pressure coefficient, For emptying, This is the torque coefficient.

4. The adaptive optimization method for blade parameters of a rim-driven propeller according to claim 1, characterized in that, The specific steps for training the verification model based on the collected results in step S3 are as follows: Step S31: Use the sample scheme as the input layer variable and the result as the output layer variable; Step S32: Use the mean squared error loss function to measure the difference between the predicted value and the actual value; For each sample scheme, calculate the square of the difference between the predicted value and the actual value, and then average the results over all sample schemes to obtain the loss function value. Where N is the number of sample schemes, It is the predicted value of the nth sample. It is the actual value of the nth sample; Step S33: Based on the loss function value, use the backpropagation algorithm to update the connection weights and biases of the neural network; starting from the output layer, calculate the partial derivative of the loss function with respect to each connection weight and bias, and then adjust the connection weights and biases according to the magnitude and direction of the partial derivatives so that the loss function value gradually decreases; Step S34: Repeat steps S32 to S33 until the loss function value is less than the loss threshold; Step S35: Obtain the coefficient of determination and determine whether the coefficient of determination is greater than the coefficient threshold. If it is less than the threshold, reconstruct the validation model. If it is greater than the threshold, output the validation model. The determination coefficients are as follows: ； in This represents the average of the true values.

5. The adaptive optimization method for blade parameters of a rim-driven propeller according to claim 1, characterized in that, The specific process of using the second-generation non-dominated sorting genetic algorithm in step S4 is as follows: Step S41: First, determine the population individuals. Based on the results and the validation model, determine the initial population individuals and set the initial population size. Step S42: For each individual in the population, calculate its fitness value using the validation model according to the set objective function; Step S43: Perform non-dominated ranking on the individuals in the population, and divide the individuals into different non-dominated levels according to the relationship of domination and subordination; Step S44: Based on the non-dominated sorting and crowding distance calculation results, select a subset of individuals as parents for the next generation; perform crossover on the selected parent individuals to generate new offspring individuals, and perform mutation on the newly generated offspring individuals. Step S45: Repeat step S44 above to continuously perform iterative optimization. The iteration process continues until the number of iterations meets the threshold.

6. The adaptive optimization method for blade parameters of a rim-driven propeller according to claim 1, characterized in that, The specific CFD operating mode parameters in step S4 include the following steps: Obtain the operational procedures required for simulating a rim-driven propeller; The simulation process is recorded as a .java file using macro commands. By combining the sim source files and .java files, the STAR CCM+ solver is called to process the design schemes in batches. Perform computational domain generation, mesh generation, and export the computation results to obtain performance simulation results.

7. An adaptive optimization system for blade parameters of a rim-driven propeller, using the adaptive optimization method for blade parameters of a rim-driven propeller as described in any one of claims 1 to 6, characterized in that, It includes a modeling module, a sample generation module, a model training module, and a validation module; The modeling module is used to perform parametric modeling of RDT blades by expressing them through feature curves, resulting in at least two RDT blade models. The sample generation module is used to generate several sample schemes of the RDT blade model using the Sobol sequence algorithm, and to collect the results of each sample scheme. The model training module is used to set the objective function and constraints, and to train the validation model based on several sample schemes and the corresponding collection results to obtain the trained validation model. The verification module is used to: combine the verification model, use the second-generation non-dominated sorting genetic algorithm to obtain the optimal RDT blade model, and verify different optimal design schemes through CFD methods to compare the performance differences between different optimal RDT blade models.

Images