A centrifugal pump cavitation performance optimization method and system based on feature selection

Abstract

The application relates to a centrifugal pump cavitation performance optimization method and system based on feature selection, and the method comprises the following steps: taking the cavitation entropy production and the lift of a centrifugal pump as optimization targets, taking the efficiency as a constraint condition, modeling and simulating the centrifugal pump, determining initial optimization variables and an optimization space range; based on the optimization space range, adopting a test design method to extract multiple groups of uniformly distributed variable combination schemes; modeling and simulating the variable combination schemes, obtaining the cavitation entropy production and the lift value, and constructing an initial sample library; training a first approximation model by using the initial sample library, screening out final optimization variables which have a significant influence on the optimization targets, and obtaining a final sample library; training a second approximation model by using the final sample library, optimizing and verifying by using an optimization algorithm, and outputting an optimal parameter combination. The application can better select parameters which have a significant influence on the optimization targets as optimization variables, simultaneously introduces the cavitation entropy production as an optimization target, and further improves the comprehensive optimization effect of cavitation and lift.

Description

Technical Field

[0001] This invention relates to the field of centrifugal pump impeller optimization technology, and in particular to a method and system for optimizing the cavitation performance of centrifugal pumps based on feature selection. Background Technology

[0002] Pumps are among the most typical and widely used hydraulic machines. As a major category of pump products, centrifugal pumps are widely used in industrial production and residential applications. Cavitation is a core issue of widespread concern in the field of hydraulic machinery. Cavitation not only causes a decline in hydraulic performance and excessive vibration and noise, but also leads to cavitation damage to flow components. To ensure stable operation of centrifugal pumps, good cavitation performance is essential. Therefore, developing methods to optimize the cavitation performance of centrifugal pumps and improving their cavitation capabilities is of great practical engineering significance. Traditional cavitation performance optimization uses critical cavitation margin or cavitation volume fraction as optimization targets. However, these targets suffer from drawbacks such as excessively high computational costs or inability to accurately reflect the losses caused by cavitation within the pump. A target that can accurately reflect the energy losses caused by cavitation within the pump while maintaining low computational costs is needed.

[0003] Furthermore, the initially determined optimization variables may include some variables with relatively insignificant influence on the optimization objective. The presence of these variables increases the dimensionality of the optimization space, and an excessive number of irrelevant variables can also reduce the prediction accuracy of the approximate model. To select variables that are more sensitive to the optimization objective from among numerous design variables, the Pearson coefficient is often used as a standard, employing a filtering method to identify variables with a high correlation to the optimization objective. However, the Pearson coefficient primarily focuses on the linear relationship between variables and the objective, neglecting nonlinear relationships and failing to accurately identify parameters with a significant impact on the optimization objective for optimization.

[0004] In summary, in order to further improve the optimization method for centrifugal pump impeller cavitation performance and enhance the optimization effect, it is necessary to develop a comprehensive optimization method for centrifugal pump cavitation and head based on feature selection. Summary of the Invention

[0005] The purpose of this invention is to propose a feature-selection-based method for optimizing the cavitation performance of centrifugal pumps. This method aims to better select parameters that have a significant impact on the optimization objective as optimization variables when comprehensively optimizing the cavitation performance and head of centrifugal pump impellers. At the same time, it introduces an optimization objective with low computational cost that can more accurately reflect the energy loss caused by cavitation in the pump—cavitation entropy production—thereby improving the comprehensive optimization effect of cavitation and head.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A feature-selective method for optimizing the cavitation performance of centrifugal pumps includes:

[0008] Using the cavitation entropy production and head of the centrifugal pump as optimization objectives and efficiency as constraints, a centrifugal pump is modeled and simulated to determine the initial optimization variables and the optimization space range.

[0009] Based on the optimized spatial range, an experimental design method is used to extract multiple combinations of uniformly distributed variables;

[0010] Modeling and simulating variable combination schemes to obtain cavitation entropy production and head values, and constructing an initial sample library;

[0011] The first approximate model is trained using the initial sample library, and the final optimization variables that have a significant impact on the optimization objective are selected to obtain the final sample library;

[0012] The second approximate model is trained using the final sample library, and the optimization algorithm is used to find and verify the optimal parameter combination.

[0013] Optionally, the calculation methods for cavitation entropy production, head, and efficiency are as follows:

[0014]

[0015] in, The cavitation entropy production rate within the pump, Let μ be the mass transfer rate during the gas-liquid two-phase phase transition process, μ be the dynamic viscosity of the liquid, and ρ be the mass transfer rate. v Where is the gas density, EPC is the cavitation entropy production inside the pump, T is the Kelvin temperature, H is the head, and P is the gas density. in P out The total pressure at the inlet and outlet of the centrifugal pump are given by ρ, g is the acceleration due to gravity, η is the efficiency, and ρ is the total pressure at the inlet and outlet of the centrifugal pump. l Let Q be the density of the liquid. v Let T be the liquid volumetric flow rate, T be the impeller torque, and ω be the centrifugal pump impeller rotational angular velocity.

[0016] Optionally, the initial optimization variables include:

[0017] Impeller outlet width, blade inlet angle, blade outlet angle, coordinates of control points for blade thickness distribution curve, coordinates of control points for impeller front cover shape in impeller axial projection, coordinates of control points for blade inlet edge in impeller axial projection.

[0018] Optionally, modeling and simulating the variable combination scheme to obtain cavitation entropy production and head values, and constructing an initial sample library includes:

[0019] For each variable combination scheme, 3D modeling of the computational domain, mesh generation, and cavitation numerical simulation are performed; where the computational domain is the water body region where the 3D model of the impeller is located.

[0020] By using CFD-Post to obtain the cavitation entropy production and head values ​​of different schemes from the results of cavitation numerical simulation, an initial sample library is constructed.

[0021] Optionally, a first approximate model is trained using the initial sample library, and the final optimization variables that have a significant impact on the optimization objective are selected. The final sample library includes:

[0022] The first approximate model is trained using the initial sample library. The feature selection method is used to select the features that have a more significant impact on the two optimization objectives. The variables that have the most significant impact on head and cavitation entropy production are merged as the final optimization variables, resulting in the final sample library after removing the design variables with insignificant impact.

[0023] Optionally, feature selection methods can be used to screen out features that have a more significant impact on the two optimization objectives, including:

[0024] XGBoost is selected as the first approximation model. The encapsulation method in feature selection is adopted to couple XGBoost with the NSGA-II algorithm. After NSGA-II selects a feature subset from the full feature set, the XGBoost model is used to evaluate the subset. The fitting accuracy of the model is used as the evaluation standard. The algorithm iterates continuously until the fitting accuracy of the model reaches the preset level or the preset number of iterations is reached. The final feature subset is output as the final optimization variable.

[0025] Optionally, training a second approximation model using the final sample library includes:

[0026] The second approximate model is trained using the final sample library. If the determination coefficient of the trained model is greater than the preset value, the trained model is adopted. If it is less than the preset value, feature selection is performed again, and then the model is trained again until the determination coefficient of the model is greater than the preset value.

[0027] Optionally, optimization algorithms are used to find and verify the following:

[0028] The optimal parameters are obtained by using an optimization algorithm and verified by numerical simulation. If the error between the numerical simulation result and the optimization result is less than a preset threshold, the optimization result is adopted. If it is greater than the preset threshold, the second approximate model is trained and optimized again until the error is less than the preset threshold, and the optimal parameters are output. Efficiency is used as a constraint when using the optimization algorithm for optimization. The second approximate model adopts the TabPFN model.

[0029] A centrifugal pump cavitation performance optimization system based on feature selection, the system comprising: a three-dimensional simulation module, a sampling module, a sample library construction module, a screening module, and an optimal parameter output module;

[0030] The three-dimensional simulation module is used to model and simulate a centrifugal pump with cavitation entropy production and head as optimization objectives and efficiency as constraints, and to determine the initial optimization variables and the optimization space range.

[0031] The sampling module is used to extract multiple uniformly distributed variable combinations based on the optimized spatial range and using experimental design methods.

[0032] The sample library construction module is used to model and simulate variable combination schemes, obtain cavitation entropy production and head values, and construct an initial sample library.

[0033] The filtering module is used to train a first approximate model using an initial sample library, filter out the final optimization variables that have a significant impact on the optimization objective, and obtain the final sample library.

[0034] The optimal parameter output module is used to train the second approximate model using the final sample library, optimize and verify it using an optimization algorithm, and output the optimal parameter combination.

[0035] Optionally, the initial optimization variables include:

[0036] Impeller outlet width, blade inlet angle, blade outlet angle, coordinates of control points for blade thickness distribution curve, coordinates of control points for impeller front cover shape in impeller axial projection, coordinates of control points for blade inlet edge in impeller axial projection.

[0037] The beneficial effects of this invention are as follows:

[0038] This invention characterizes the energy loss caused by cavitation within a centrifugal pump by introducing cavitation entropy production, reflecting the cavitation state within the pump. Furthermore, this method requires only a single cavitation numerical simulation, whereas calculating NPSHr requires repeated numerical simulations with decreasing centrifugal pump inlet pressure, resulting in a lengthy computation time. By introducing cavitation entropy production, the problems of insufficient cavitation volume fraction to characterize energy loss due to cavitation and the high computational cost of NPSHr are avoided.

[0039] This invention employs feature selection methods, including encapsulation and embedded methods, to select the variables with the most significant impact on cavitation entropy production and head from the initial optimization variables as the final optimization variables. This method avoids the problem of insufficient consideration of nonlinear relationships when using Pearson coefficients as the standard for selecting optimization variables, and better achieves dimensionality reduction of the optimization space. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a schematic diagram of a centrifugal pump cavitation performance optimization method based on feature selection according to an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of the isosurface of 10% cavitation distribution in the original scheme of this embodiment of the invention;

[0043] Figure 3 This is a schematic diagram of the isosurface of the 10% cavitation distribution in an optimized scheme according to an embodiment of the present invention;

[0044] Figure 4 This is a comparison chart of cavitation characteristic curves before and after optimization in an embodiment of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] like Figure 1 As shown, this embodiment proposes a feature-selection-based method for optimizing the cavitation performance of centrifugal pumps. This embodiment is mainly applied after the initial design of the centrifugal pump hydraulic model to further improve the cavitation performance and head of the centrifugal pump. The steps include:

[0048] S1. Taking the cavitation entropy production and head of the centrifugal pump as optimization objectives and efficiency as constraints, the centrifugal pump is modeled and simulated to determine the initial optimization variables and the optimization space range.

[0049] S2. Based on the optimized spatial range, multiple combinations of uniformly distributed variables are extracted using experimental design methods;

[0050] S3. Model and simulate the variable combination scheme to obtain cavitation entropy production and head value, and construct an initial sample library;

[0051] S4. Train the first approximate model using the initial sample library, select the final optimization variables that have a significant impact on the optimization objective, and obtain the final sample library;

[0052] S5. Train the second approximate model using the final sample library, use the optimization algorithm to find and verify the optimal parameter combination, and output the optimal parameter combination.

[0053] Specifically, in this embodiment, S1. Taking the cavitation entropy production and head of the centrifugal pump as optimization objectives and efficiency as constraints, the centrifugal pump is modeled and simulated to determine the initial optimization variables and optimization space range, including:

[0054] Cavitation entropy production and head were determined as optimization objectives. An optimization operating condition was selected, and in Ansys Workbench, BladeGen, ICEM CFD, TurboGrid, and CFX were used to perform 3D computational domain modeling, mesh generation, and cavitation numerical simulation of the impeller, volute, and inlet pipe in the original design. The initial values ​​of the optimization objectives were calculated according to the following formula, and the initial design variables were determined based on design experience, clarifying the upper and lower boundaries of the design variables. In this example, the optimization operating condition Q was selected. v =1.0Q vd P in =55kPa(P in For the centrifugal pump inlet pressure), a 3D computational domain modeling, mesh generation, and cavitation numerical simulation were performed on the impeller, volute, and inlet pipe in the original design. The cavitation entropy production (EPC), head (H), and efficiency (η) of the original design were calculated using the following formula. Simultaneously, the critical cavitation margin (NPSH) was obtained through multiple numerical simulations. r The values ​​mentioned above are shown in Table 2 below. The impeller outlet width x0, blade inlet angle x1, blade outlet angle x2, control point coordinates of the blade thickness distribution curve x3 and x4, control point coordinates of the impeller front cover shape in the impeller axial projection diagram x5, and control point coordinates of the blade inlet edge in the impeller axial projection diagram x6, x7, x8, and x9 are selected as initial optimization variables. The upper and lower limits of these optimization variables are also defined, and their initial values ​​and limits are shown in Table 1 below.

[0055]

[0056]

[0057] in, The cavitation entropy production rate within the pump, Let μ be the mass transfer rate during the gas-liquid two-phase phase transition process, μ be the dynamic viscosity of the liquid, and ρ be the mass transfer rate. v Where is the gas density, EPC is the cavitation entropy production inside the pump, T is the Kelvin temperature, taken as T = 298.1K; H is the head, P in P outHere, g represents the total pressure at the inlet and outlet of the centrifugal pump, respectively, and g is the acceleration due to gravity, taken as g = 9.8m. 2 / s; η is efficiency, ρ l Let Q be the density of the liquid. v Let T be the liquid volumetric flow rate, T be the impeller torque, and ω be the centrifugal pump impeller rotational angular velocity.

[0058] Table 1 Initial Optimization Variables Table

[0059] Initial optimization variables numerical values upper limit lower limit unit x0 0.031 0.034 0.030 m x1 39.1 -60.8 -40.8 ° x2 27.2 -62.8 -52.8 ° x3 0.0096 0.011 0.008 - x4 0.005 0.007 0.004 - x5 0.008 0.015 0.005 - x6 0.0079 0.01 0 - x7 0.0621 0.062 0.052 - x8 0.0322 0.041 0.031 - x9 0.0709 0.075 0.065 -

[0060] Table 2 Initial values ​​of each parameter

[0061] Performance parameters EPC(W / K) H(m) η(%) <![CDATA[NPSH r (m)]]> numerical values 0.105 32.23 82.9 3.54

[0062] Specifically, in this embodiment, S2. Based on the optimized spatial range, the experimental design method is used to extract multiple uniformly distributed variable combination schemes, including:

[0063] Using experimental design methods, and based on a defined optimization space, combinations of variables are extracted according to a uniform distribution principle to ensure that the extracted variable combinations comprehensively and effectively cover the design space. This example employs Latin hypercube sampling to uniformly generate 120 variable combinations within the design space.

[0064] Specifically, in this embodiment, S3. Modeling and simulating the variable combination scheme to obtain cavitation entropy production and head value, and constructing an initial sample library includes:

[0065] In Ansys Workbench, BladeGen, TurboGrid, and CFX were used to complete the 3D modeling of the computational domain, mesh generation, and cavitation numerical simulation for each variable combination scheme. CFD-Post was used to obtain the cavitation entropy production and head values ​​of different schemes from the results of the cavitation numerical simulation to build an initial sample library. The computational domain was the water body region where the impeller 3D model was located. Since this method does not involve the optimization of the geometric parameters of the volute and inlet pipe, the computational domain and mesh of the volute and inlet pipe models are the same as in the first step.

[0066] Specifically, in this embodiment, S4. Training a first approximate model using the initial sample library, selecting the final optimization variables that have a significant impact on the optimization objective, and obtaining the final sample library includes:

[0067] An approximate model I was trained using an initial sample library. Feature selection methods were employed to identify features with the most significant impact on both optimization objectives. The variables most significantly affecting head and cavitation entropy production were then combined as the final optimization variables, resulting in a sample library after removing insignificant design variables. In this example, XGBoost was selected as the approximate model I. An encapsulation method within the feature selection process was used, coupling XGBoost with the NSGA-II algorithm. After NSGA-II selected a subset of features from the full feature set, the XGBoost model was used to evaluate this subset. The model's fitting accuracy was used as the evaluation criterion. The algorithm iterated until the model's fitting accuracy reached a certain level or a certain number of iterations were completed, outputting the final feature subset as the final optimization variables. Using this method, the variables with the most significant impact on cavitation entropy production and head in this example were [x0, x1, x6, x7, x8] and [x0, x1, x2, x7, x9], respectively. The variables that have the most significant impact on head and cavitation entropy production are merged, and the final optimized variables are shown in Table 3 below. Variables with weak impact are removed from the sample library to obtain the final sample library.

[0068] Table 3 Final Optimization Variables Table

[0069] Final optimization variables numerical values upper limit lower limit unit x0 0.031 0.034 0.030 m x1 39.1 -60.8 -40.8 ° x2 27.2 -62.8 -52.8 ° x6 0.0079 0.01 0 - x7 0.0621 0.062 0.052 - x8 0.0322 0.041 0.031 - x9 0.0709 0.075 0.065 -

[0070] Specifically, in this embodiment, S5. Training the second approximate model using the final sample library, optimizing and verifying it using an optimization algorithm, and outputting the optimal parameter combination includes:

[0071] Using the final sample database, train approximate model II. If the trained model r 2 If the coefficient is greater than 0.9, it can be used; if it is less than 0.9, feature selection is performed again, followed by training again, until r... 2 The coefficient is greater than 0.9. After obtaining the approximate model II that meets the requirements, the optimal parameters are obtained using an optimization algorithm, and then verified using numerical simulation. If the error between the numerical simulation result and the optimization result is less than 5%, it is considered reasonable, and the optimization result can be adopted. If it is greater than 5%, the approximate model II is trained and optimized again until the error is less than 5%, and the optimal parameters are output. In order to prevent the efficiency of the optimal solution from being lower than that of the original solution, efficiency is used as a constraint when using the optimization algorithm. In this example, the TabPFN model is selected as the approximate model II in this step, and the cavitation entropy production prediction model and the head prediction model are trained using the finally obtained sample library, respectively. 2The coefficients are 0.91 and 0.97, respectively. The NSGA-II optimization algorithm is used to find the optimal parameter combination, and the results are verified by numerical simulation. The deviation between the simulation results and the optimization results is less than 5%, which indicates that the optimal parameters can be adopted. The obtained optimal parameters are related to cavitation entropy production, head, efficiency, and NPSH. r The values ​​are shown in the table below. Comparison reveals that through optimization, the pump's cavitation entropy production decreased by 9.9%, the head increased by 10.46%, and the efficiency increased by 1.1%, resulting in a reduction in NPSH. r It decreased by approximately 27.40%.

[0072] Table 4 Comparison of variable and parameter values ​​before and after optimization

[0073]

[0074] The isosurfaces of 10% cavitation distribution inside the impeller before and after optimization are as follows: Figure 2 , Figure 3 As shown, the cavitation distribution area has been reduced to a certain extent, which indirectly verifies the optimization effect.

[0075] The cavitation characteristic curves of the centrifugal pump before and after optimization are as follows: Figure 4 As shown, it can be observed that, on the one hand, NPSH r To some extent, the pump's cavitation performance was improved, and on the other hand, the head increased, thus enhancing the pump's work capacity. Figure 4 The red dot indicates that the head drops by 3% at this point. 1atm H 1atm This is the head value of the centrifugal pump when the inlet pressure is 1 atmosphere.

[0076] This embodiment introduces cavitation entropy production, thereby reducing the computational cost in the centrifugal pump cavitation optimization process and more accurately describing the cavitation state within the pump, using cavitation entropy production and head as optimization objectives. Multiple design variables of the centrifugal pump impeller are selected as initial optimization variables. An experimental design method is used to uniformly extract design schemes within the optimization space, and cavitation numerical simulations are performed on each scheme to obtain an initial sample library. A feature selection method is used to screen variables that significantly affect the optimization objectives as optimization variables, avoiding the problem of insufficient consideration of the nonlinear relationship between optimization variables and optimization objectives when using Pearson coefficients as the standard for selecting optimization variables, thus better achieving dimensionality reduction of the optimization space. Subsequently, after removing optimization variables with weak influence, the sample library is obtained. An approximate model is trained, and an optimization algorithm is selected to find the optimal result. Numerical simulation is used for verification. If the error between the numerical simulation result and the optimization result is less than 5%, it is considered reasonable, and the optimization result can be adopted. If it is greater than 5%, the approximate model II is trained and optimized again until the error is less than 5%, at which point the optimal parameters are output. Through optimization, the optimal solution, compared to the original solution, reduced cavitation entropy production in the pump by 9.9%, increased head by 10.46%, increased efficiency by 1.1%, and improved NPSH. r It decreased by approximately 27.40%.

[0077] This embodiment also proposes a centrifugal pump cavitation performance optimization system based on feature selection. The system includes: a three-dimensional simulation module, a sampling module, a sample library construction module, a screening module, and an optimal parameter output module.

[0078] The three-dimensional simulation module is used to model and simulate a centrifugal pump with cavitation entropy production and head as optimization objectives and efficiency as constraints, and to determine the initial optimization variables and the optimization space range.

[0079] The sampling module is used to extract multiple uniformly distributed combinations of variables based on the optimized spatial range and using experimental design methods.

[0080] The sample library construction module is used to model and simulate variable combination schemes, obtain cavitation entropy production and head values, and build an initial sample library.

[0081] The filtering module is used to train the first approximate model using the initial sample library, filter out the final optimization variables that have a significant impact on the optimization objective, and obtain the final sample library;

[0082] The optimal parameter output module is used to train the second approximate model using the final sample library, optimize and verify the model using an optimization algorithm, and output the optimal parameter combination.

[0083] Furthermore, the initial optimization variables include:

[0084] Impeller outlet width, blade inlet angle, blade outlet angle, coordinates of control points for blade thickness distribution curve, coordinates of control points for impeller front cover shape in impeller axial projection, coordinates of control points for blade inlet edge in impeller axial projection.

[0085] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for optimizing the cavitation performance of a centrifugal pump based on feature selection, characterized in that, include: Using the cavitation entropy production and head of the centrifugal pump as optimization objectives and efficiency as constraints, a centrifugal pump is modeled and simulated to determine the initial optimization variables and the optimization space range. Based on the optimized spatial range, an experimental design method is used to extract multiple combinations of uniformly distributed variables; Modeling and simulating variable combination schemes to obtain cavitation entropy production and head values, and constructing an initial sample library; The first approximate model is trained using the initial sample library, and the final optimization variables that have a significant impact on the optimization objective are selected to obtain the final sample library; The second approximate model is trained using the final sample library, and the optimization algorithm is used to find and verify the optimal parameter combination. The calculation methods for cavitation entropy production, head, and efficiency are as follows: in, The cavitation entropy production rate within the pump, The mass transfer rate during the gas-liquid two-phase phase transition process is denoted as . μ For the dynamic viscosity of the liquid, For gas density, For cavitation entropy production inside the pump, T Kelvin temperature, H For Yang Cheng, P in , P out These are the total pressures at the inlet and outlet of the centrifugal pump, respectively. g It is the acceleration due to gravity. η For efficiency, ρ l For the density of the liquid, Q v The volumetric flow rate of the liquid. T For impeller torque, ω The angular velocity of the centrifugal pump impeller; The initial optimization variables include: Impeller outlet width, blade inlet angle, blade outlet angle, coordinates of control points for blade thickness distribution curve, coordinates of control points for impeller front cover shape in impeller axial projection, coordinates of control points for blade inlet edge in impeller axial projection; Modeling and simulating variable combination schemes to obtain cavitation entropy production and head values, and constructing an initial sample library including: For each variable combination scheme, 3D modeling of the computational domain, mesh generation, and cavitation numerical simulation are performed; where the computational domain is the water body region where the 3D model of the impeller is located. By using CFD-Post to obtain the cavitation entropy production and head values ​​of different schemes from the results of cavitation numerical simulation, an initial sample library is constructed.

2. The centrifugal pump cavitation performance optimization method based on feature selection according to claim 1, characterized in that, The first approximate model is trained using the initial sample library. The final optimization variables that significantly affect the optimization objective are then selected, resulting in the final sample library, which includes: The first approximate model is trained using the initial sample library. The feature selection method is used to select the features that have a more significant impact on the two optimization objectives. The variables that have the most significant impact on head and cavitation entropy production are merged as the final optimization variables, resulting in the final sample library after removing the design variables with insignificant impact.

3. The centrifugal pump cavitation performance optimization method based on feature selection according to claim 2, characterized in that, Feature selection methods were used to identify features that had a more significant impact on the two optimization objectives, including: XGBoost is selected as the first approximation model. The encapsulation method in feature selection is adopted to couple XGBoost with the NSGA-II algorithm. After NSGA-II selects a feature subset from the full feature set, the XGBoost model is used to evaluate the subset. The fitting accuracy of the model is used as the evaluation standard. The algorithm iterates continuously until the fitting accuracy of the model reaches the preset level or the preset number of iterations is reached. The final feature subset is output as the final optimization variable.

4. The centrifugal pump cavitation performance optimization method based on feature selection according to claim 1, characterized in that, Training the second approximation model using the final sample library includes: The second approximate model is trained using the final sample library. If the determination coefficient of the trained model is greater than the preset value, the trained model is adopted. If it is less than the preset value, feature selection is performed again, and then the model is trained again until the determination coefficient of the model is greater than the preset value.

5. The centrifugal pump cavitation performance optimization method based on feature selection according to claim 1, characterized in that, Optimization and verification using optimization algorithms include: The optimal parameters are obtained by using an optimization algorithm and verified by numerical simulation. If the error between the numerical simulation result and the optimization result is less than a preset threshold, the optimization result is adopted. If it is greater than the preset threshold, the second approximate model is trained and optimized again until the error is less than the preset threshold, and the optimal parameters are output. Efficiency is used as a constraint when using the optimization algorithm for optimization. The second approximate model adopts the TabPFN model.

6. A centrifugal pump cavitation performance optimization system based on feature selection, characterized in that, The system for implementing the method as described in any one of claims 1-5 includes: a three-dimensional simulation module, a sampling module, a sample library construction module, a screening module, and an optimal parameter output module; The three-dimensional simulation module uses the cavitation entropy production and head of the centrifugal pump as optimization objectives and efficiency as constraints to model and simulate the centrifugal pump, and determine the initial optimization variables and optimization space range. The sampling module is used to extract multiple uniformly distributed variable combinations based on the optimized spatial range and using experimental design methods. The sample library construction module is used to model and simulate variable combination schemes, obtain cavitation entropy production and head values, and construct an initial sample library. The filtering module is used to train a first approximate model using an initial sample library, filter out the final optimization variables that have a significant impact on the optimization objective, and obtain the final sample library. The optimal parameter output module is used to train the second approximate model using the final sample library, optimize and verify it using an optimization algorithm, and output the optimal parameter combination.

7. The centrifugal pump cavitation performance optimization system based on feature selection according to claim 6, characterized in that, The initial optimization variables include: Impeller outlet width, blade inlet angle, blade outlet angle, coordinates of control points for blade thickness distribution curve, coordinates of control points for impeller front cover shape in impeller axial projection, coordinates of control points for blade inlet edge in impeller axial projection.

Images