Pumped storage power station inlet and outlet body optimization method, system, device and medium

Abstract

This invention belongs to the fields of hydraulic engineering, computational fluid dynamics, and artificial intelligence-aided design technology. It discloses a method, system, equipment, and medium for optimizing the inlet and outlet shapes of pumped storage power stations. The method includes: substituting initial sample points into a fully parameterized geometric model of the flow channel to construct a multi-fidelity calculation sample library; post-processing the multi-fidelity calculation sample library to establish an entropy production objective function aimed at minimizing the total energy dissipation across the entire flow channel; constructing a covariance matrix to generate a co-kriging surrogate model; using the co-kriging surrogate model as a fitness evaluation function, and combining it with the entropy production objective function to output a combination of design variables for optimizing the inlet and outlet shapes of the hydraulic machinery. This invention utilizes non-uniform rational splines to achieve a high degree of freedom in geometric expression, and combines the co-kriging multi-fidelity surrogate model with an improved sparrow search algorithm to achieve global optimization of the flow channel shape.

Description

Technical Field

[0001] This invention relates to the fields of water conservancy and hydropower engineering, computational fluid dynamics (CFD) and artificial intelligence-aided design technology, and in particular to methods, systems, equipment and media for optimizing the shape of the inlet and outlet of pumped storage power stations. Background Technology

[0002] As the global energy structure transitions towards clean and low-carbon energy, pumped storage power stations are hailed as regulators and stabilizers of the power grid due to their large-scale energy storage, frequency and phase regulation, and emergency backup functions. The inlet and outlet, as the throat components connecting the reservoir and the water diversion tunnel, directly affect the circulation efficiency and operational stability of the power station.

[0003] Existing technologies face three main technical bottlenecks in the design and optimization of inlet and outlet flow channels:

[0004] (1) Lack of quantitative evaluation of energy dissipation mechanism: Traditional design methods (such as the standard Bernoulli equation method) usually only focus on the total pressure difference at the inlet and outlet sections (i.e., head loss coefficient). However, the total pressure difference is a macroscopic integral quantity, which cannot reveal the specific location and cause of energy loss inside the flow channel. For complex bidirectional flow (especially diffusion flow under pumping conditions), there are a large number of secondary flows, vortices and flow separations inside the fluid. The traditional pressure difference method cannot distinguish between direct dissipation caused by viscous friction and indirect dissipation caused by turbulent pulsation, resulting in unclear optimization direction. Often, the phenomenon of pressing down the gourd and floating up the ladle occurs. That is, although the total pressure difference is reduced, the local undesirable flow state in the flow channel is aggravated, inducing vibration risk.

[0005] (2) The contradiction between CFD computation cost and accuracy (curse of dimensionality): In order to obtain high-precision flow field details, modern CFD simulation usually requires fine mesh (millions to tens of millions of nodes) and advanced turbulence models (such as SST k-ω or LES). A single high-fidelity (HF) calculation often takes several hours on a high-performance workstation. In the process of intelligent optimization, evolutionary algorithms (such as genetic algorithms and particle swarm algorithms) usually need to call thousands of CFD evaluations. If all HF models are used, the calculation cycle will be as long as several months, which is actually unacceptable in engineering. If a coarse-mesh low-fidelity (LF) model is used, although the calculation is fast, it is very easy to miss key separation flow features, resulting in distortion of optimization results.

[0006] (3) Rigidity and limitations of geometric parameterization: Existing inlet and outlet channel optimization methods mostly adopt orthogonal experimental design combined with simple geometric lines (such as circular arcs or low-order Bézier curves). In terms of geometric parameterization, traditional methods have limited descriptive capabilities and are difficult to handle complex channel topology changes. In particular, low-order curves have global support, that is, moving the control point will change the shape of the entire curve, making it difficult for designers to fine-tune key local features such as the starting inflection point of the diffuser section and the anti-cavitation lip, which easily leads to shape distortion. Moreover, in terms of optimization strategy, the orthogonal experimental design is essentially a discrete sampling method. It can only find the optimal combination among a limited number of preset level values ​​and cannot traverse the continuous design space. This often results in the best solution in the sample rather than the global theoretical best solution, which is very easy to miss the high-performance design points hidden in the parameter range.

[0007] In summary, there is an urgent need for an intelligent optimization design method that can delve into the microscopic dissipation mechanism, effectively balance computational cost and accuracy, and possess a high degree of freedom in local modification capabilities.

[0008] Therefore, how to provide methods, systems, equipment, and media for optimizing the shape of the inlet and outlet of pumped storage power stations is an urgent problem to be solved. Summary of the Invention

[0009] The present invention provides a method, system, equipment and medium for optimizing the shape of the inlet and outlet of a pumped storage power station, in order to solve the problems of the above-mentioned technologies in the prior art.

[0010] According to a first aspect of the present invention, a method for optimizing the shape of the inlet and outlet of a pumped storage power station is provided.

[0011] In one embodiment, the method for optimizing the shape of the inlet and outlet of the pumped storage power station includes:

[0012] Based on the contour lines of the inlet and outlet of the hydraulic machinery, the shape parameters of the inlet and outlet flow channels are determined. The shape parameters are combined with geometric constraints and auxiliary design technology is used to construct a fully parameterized geometric model of the flow channels.

[0013] Substitute the initial sample points into the fully parameterized geometric model of the flow channel to reconstruct the inlet and outlet contour curves. After judging the degree of satisfaction of the reconstruction results and geometric constraints, generate high-fidelity sample sets and low-fidelity sample sets to construct a multi-fidelity calculation sample library.

[0014] The multi-fidelity calculation sample library is post-processed to calculate the direct entropy production rate and turbulent entropy production rate of the inlet and outlet channels, and an entropy production objective function is established with the goal of minimizing the total energy dissipation of the entire channel.

[0015] Based on the autoregressive technique, the correlation mapping relationship between the high-fidelity sample set and the low-fidelity sample set is analyzed, and the process variance is determined according to the mapping relationship to construct the covariance matrix and generate a co-kriging surrogate model.

[0016] The co-Kriging agent model is used as the fitness evaluation function to perform the Levy flight position update process, and combined with the entropy production objective function to output the combination of design variables for optimizing the shape of the hydraulic machinery inlet and outlet.

[0017] According to a second aspect of the present invention, a system for optimizing the shape of the inlet and outlet of a pumped storage power station is provided.

[0018] In one embodiment, a pumped storage power station inlet and outlet shape optimization system includes:

[0019] The geometric model construction unit is used to determine the shape parameters of the inlet and outlet flow channels based on the contour lines of the inlet and outlet of the hydraulic machinery, and to construct a fully parameterized geometric model of the flow channel by combining the shape parameters with geometric constraints using auxiliary design technology.

[0020] The computational sample library construction unit is used to substitute the initial sample points into the fully parameterized geometric model of the flow channel, reconstruct the contour curves of the inlet and outlet, and generate high-fidelity sample sets and low-fidelity sample sets after judging the degree of satisfaction of the reconstruction results and geometric constraints, thus constructing a multi-fidelity computational sample library.

[0021] The entropy production calculation unit is used to post-process the multi-fidelity calculation sample library, calculate the direct entropy production and turbulent entropy production of the inlet and outlet channels, and establish an entropy production objective function with the goal of minimizing the total energy dissipation of the entire channel.

[0022] The surrogate model construction unit is used to analyze the correlation mapping relationship between high-fidelity sample sets and low-fidelity sample sets based on autoregression techniques, and to determine the process variance based on the mapping relationship to construct the covariance matrix and generate a co-kriging surrogate model.

[0023] The design variable combination output unit is used to use the co-Kriging agent model as the fitness evaluation function, perform the Lévy flight position update process, and combine it with the entropy production objective function to output the design variable combination for optimizing the shape of the inlet and outlet of the hydraulic machinery.

[0024] According to a third aspect of the present invention, a computer device is provided.

[0025] In some embodiments, the computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0026] According to a fourth aspect of the present invention, a computer-readable storage medium is provided.

[0027] In one embodiment, a computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the above method.

[0028] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0029] This invention introduces entropy production theory to accurately quantify energy dissipation sources, utilizes non-uniform rational B-splines to achieve high-degree-of-freedom geometric representation, and combines a co-Kriging multi-fidelity surrogate model with an improved sparrow search algorithm to achieve global optimization of the flow channel shape while significantly reducing computational costs.

[0030] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0031] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0032] Figure 1 This is a flowchart illustrating a method for optimizing the shape of the inlet and outlet of a pumped storage power station according to an exemplary embodiment;

[0033] Figure 2 This is a schematic diagram of a pumped storage power station inlet and outlet shape optimization system according to an exemplary embodiment.

[0034] Figure 3 This is a schematic diagram of the structure of a computer device according to an exemplary embodiment;

[0035] Figure 4 This is a technical flowchart illustrating a method for optimizing the shape of the inlet and outlet of a pumped storage power station according to an exemplary embodiment.

[0036] Figure 5 This is a schematic diagram illustrating the influence of NURBS curve control points and weighting factors on the flow channel profile in an inlet and outlet shape optimization method for a pumped storage power station, according to an exemplary embodiment.

[0037] Figure 6 This is a cloud map of energy dissipation distribution inside the flow channel calculated based on entropy production theory in the method for optimizing the shape of the inlet and outlet of a pumped storage power station according to an exemplary embodiment.

[0038] Figure 7This is a comparison of the prediction errors between the multi-fidelity Co-Kriging model and the traditional single-fidelity Kriging model in the method for optimizing the shape of the inlet and outlet of a pumped storage power station, according to an exemplary embodiment.

[0039] Figure 8 This is a comparison of the convergence curves of the improved Sparrow Search algorithm and the standard SSA algorithm in the method for optimizing the shape of the inlet and outlet of a pumped storage power station, according to an exemplary embodiment.

[0040] Figure 9 This is a schematic diagram of the functional module architecture and data interaction of the hydraulic mechanical flow channel optimization system in the method for optimizing the shape of the inlet and outlet of a pumped storage power station, according to an exemplary embodiment. Detailed Implementation

[0041] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some portions and features of certain embodiments may be included in or replace portions and features of other embodiments. The scope of the embodiments herein includes the entire scope of the claims and all available equivalents thereof. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0042] The modules in the apparatus or system of this application can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0043] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0044] Figure 1 An embodiment of the method for optimizing the shape of the inlet and outlet of a pumped storage power station according to the present invention is shown.

[0045] In this optional embodiment, the method for optimizing the shape of the pumped storage power station's inlet and outlet includes:

[0046] Step S101: Based on the contour lines of the inlet and outlet of the hydraulic machinery, determine the shape parameters of the inlet and outlet flow channels, and combine the shape parameters with geometric constraints to construct a fully parameterized geometric model of the flow channels using auxiliary design technology.

[0047] Step S102: Substitute the initial sample points into the fully parameterized geometric model of the flow channel, reconstruct the contour curves of the inlet and outlet, and generate a high-fidelity sample set and a low-fidelity sample set after judging the degree of satisfaction of the reconstruction results and geometric constraints, and construct a multi-fidelity calculation sample library.

[0048] Step S103: Post-process the multi-fidelity calculation sample library, calculate the direct entropy production rate and turbulent entropy production rate of the inlet and outlet channels, and establish an entropy production objective function with the goal of minimizing the total energy dissipation of the entire channel.

[0049] Step S104: Analyze the correlation mapping relationship between the high-fidelity sample set and the low-fidelity sample set based on autoregressive techniques, and determine the process variance based on the mapping relationship to construct the covariance matrix and generate a co-kriging surrogate model.

[0050] Step S105: The co-Kriging agent model is used as the fitness evaluation function to perform the Levy flight position update process, and combined with the entropy production objective function to output the design variable combination for the shape optimization of the hydraulic machinery inlet and outlet.

[0051] In this optional embodiment, when determining the shape parameters of the inlet and outlet flow channels based on the contour lines of the inlet and outlet of the hydraulic machinery, and combining the shape parameters with geometric constraints to construct a fully parameterized geometric model of the flow channel using auxiliary design technology, parametric contour lines of the top plates of the inlet and outlet can be constructed on the longitudinal sections of the inlet and outlet of the hydraulic machinery based on non-uniform rational spline technology. The shape parameters of the inlet and outlet flow channels are determined using the parametric contour curves. Based on the shape parameters of the inlet and outlet flow channels, a three-dimensional solid model of the fluid domain is performed using auxiliary design technology, and the top plate curves of the diffuser sections of the inlet and outlet are selected as optimization objects. The degree of the non-uniform rational spline is defined. Based on the definition results, control points are selected to generate geometric constraints including the overall diffusion angle and the local diffusion angle. The geometric constraints are used to constrain the three-dimensional solid modeling results of the fluid domain to complete the optimization of the top plate curves of the diffuser sections of the inlet and outlet. Based on the optimization results, a fully parameterized geometric model of the flow channel is output.

[0052] In this optional embodiment, when substituting the initial sample points into the fully parameterized geometric model of the flow channel to reconstruct the inlet and outlet contour curves, and generating high-fidelity and low-fidelity sample sets after judging the degree of satisfaction of the reconstruction results and geometric constraints, and constructing a multi-fidelity calculation sample library, the inlet and outlet shape optimization problem can be defined as a hypercube, and the control points and their corresponding weight factors can be used as input variables of the hypercube. Simultaneously, the range of values ​​for the input variables can be divided into several equally probable intervals. Input variables are arbitrarily selected from each equally probable interval and shuffled. Initial sample points are generated based on the processing results, and these initial sample points are then substituted into the flow channel. The top plate curves of the diffuser sections of the inlet and outlet are reconstructed using a fully parameterized geometric model. The degree of satisfaction of the reconstructed top plate curves of the diffuser sections with the overall diffusion angle constraints and local diffusion angle constraints is determined. Initial sample points that do not meet the constraints are removed based on the satisfaction level, and initial sample points are re-acquired until the number of initial sample points that meet the constraints reaches the target number, outputting a valid sample set. Based on the valid sample set, a high-fidelity sample set and a low-fidelity sample set are generated using the maximum-minimum distance criterion and a nested sampling strategy. Fluid dynamics simulation is performed based on the high-fidelity sample set and the low-fidelity sample set, and a multi-fidelity calculation sample library is constructed based on the simulation results.

[0053] In this optional embodiment, when generating high-fidelity and low-fidelity sample sets based on an effective sample set using the maximum-minimum distance criterion and a nested sampling strategy, and performing fluid dynamics simulations based on these sets, and constructing a multi-fidelity computational sample library based on the simulation results, the effective sample set containing all sample points can be used as the low-fidelity sample set. Sample points are randomly selected from the low-fidelity sample set and added to the high-fidelity sample set. The dimensional space Euclidean distance of the unselected sample points in the low-fidelity sample set is then calculated; the minimum dimensional space Euclidean distance is then selected. The distance is the minimum distance of randomly selected sample points, and the minimum distance of all unselected sample points is compared. Based on the comparison results, the unselected sample point with the maximum minimum distance is selected and added to the high-fidelity sample set. The unselected sample points are repeatedly selected and added to the high-fidelity sample set until the number of samples in the high-fidelity sample set reaches the preset number, and the high-fidelity sample set is output. The high-fidelity sample set and the low-fidelity sample set are used to determine the unoptimized geometric boundaries of the top plate curves of the diffuser section of the inlet and outlet, and a multi-fidelity calculation sample library is constructed by combining fluid dynamics numerical simulation technology.

[0054] In this optional embodiment, when determining the unoptimized geometric boundaries of the inlet and outlet diffuser section top plate curves using high-fidelity and low-fidelity sample sets, and constructing a multi-fidelity calculation sample library using fluid dynamics numerical simulation technology, the control point coordinates and weighting factors corresponding to the high-fidelity and low-fidelity sample sets can be obtained, and the control point coordinates and weighting factors can be imported into 3D software to generate 3D parametric top plate profiles; based on the fully parametric geometric model of the flow channel, the unoptimized fixed geometric boundaries of the inlet and outlet diffuser section top plate curves are determined, combined with sweeping... The technology generates a closed top plate surface and stitches the closed top plate surface with the fixed geometric boundary to generate a three-dimensional fluid domain solid model. The high-fidelity sample set is then finely meshed, and the three-dimensional fluid domain solid model is run in combination with shear stress turbulence technology to perform high-fidelity computational fluid dynamics numerical simulation. The low-fidelity sample set is sparsely meshed, and the three-dimensional fluid domain solid model is run in combination with dual turbulence technology to perform low-fidelity computational fluid dynamics numerical simulation. Based on the high-fidelity and low-fidelity simulation results, the flow field data is output to construct a multi-fidelity computational sample library.

[0055] In this optional embodiment, when analyzing the correlation mapping relationship between the high-fidelity sample set and the low-fidelity sample set based on autoregressive techniques, and determining the process variance and constructing the covariance matrix to generate the co-kriging surrogate model based on the mapping relationship, the high-fidelity sample set and the low-fidelity sample set can be used as a basis. The condensed log-likelihood function of the spatial hyperparameters can be obtained through the maximum likelihood estimation method. The optimal hyperparameter estimate is output based on the condensed log-likelihood function. The optimal hyperparameter estimate is substituted into the analytical expression to determine the final process variance and scaling factor, thereby completing the construction of the multi-fidelity covariance matrix of the co-kriging model. Based on the multi-fidelity covariance matrix, the high-fidelity sample set is divided into several groups to select the training set, and the training set and the low-fidelity sample set are fused to train the co-kriging model. After verifying the global generalization accuracy of the co-kriging model after training, sample points of the high-fidelity sample set are automatically added based on the verification results to ensure the accuracy of the co-kriging model, and the final co-kriging surrogate model is output.

[0056] In this optional embodiment, based on high-fidelity and low-fidelity sample sets, the condensed log-likelihood function of spatial hyperparameters is obtained through maximum likelihood estimation. When outputting the optimal hyperparameter estimate based on the condensed log-likelihood function, the high-fidelity and low-fidelity sample sets can be combined using a Gaussian stochastic process to construct a joint maximum likelihood function for the occurrence probability of multi-fidelity fluid dynamics training samples. The natural logarithm of the joint maximum likelihood function is selected to obtain the log-likelihood function. The partial derivatives of the process variance and scaling factor of the log-likelihood function are calculated, and the partial derivatives are defined as zero after calculation. The analytical expressions of the process variance and scaling factor with respect to spatial hyperparameters are derived. The analytical expressions are substituted into the log-likelihood function to eliminate the process variance and scaling factor, resulting in a condensed log-likelihood function containing only spatial hyperparameters. The condensed log-likelihood function is then maximized using nonlinear unconstrained optimization techniques. Based on the maximization optimization results, the solution corresponding to the condensed log-likelihood function reaching its maximum value is selected as the optimal hyperparameter estimate.

[0057] In this optional embodiment, when using the co-Kriging surrogate model as the fitness evaluation function, performing Lévy flight position update processing, and combining it with the entropy production objective function to output the design variable combination for the body shape optimization of the hydraulic machinery inlet and outlet, several initial positions of individuals can be generated based on cubic chaotic mapping to construct an initial sparrow population. Using the co-Kriging surrogate model as the fitness evaluation function, the fitness of the body shape optimization scheme corresponding to the position of each sparrow individual in the initial sparrow population is calculated to evaluate the quality of the body shape optimization scheme. Based on the quality evaluation results, a random step size following a Lévy distribution is added to determine the non-uniformity coefficient of the flow velocity in the pumping condition. A penalty function is applied to the target value of the entropy production objective function corresponding to the current individual based on the determination results, so that the initial sparrow population moves towards a region that balances both directions. Based on the region movement results, the body shape optimization variable combination corresponding to the body shape of the inlet and outlet flow channels is output. Based on the body shape optimization variable combination, the fully parameterized geometric model of the flow channel is reconstructed, and a fluid dynamics numerical simulation is performed to determine the design variable combination scheme for the body shape optimization of the hydraulic machinery inlet and outlet.

[0058] Figure 2 An embodiment of the pumped storage power station inlet and outlet shape optimization system of the present invention is shown.

[0059] In this optional embodiment, the pumped storage power station inlet and outlet shape optimization system includes:

[0060] The geometric model construction unit 201 is used to determine the shape parameters of the inlet and outlet flow channels based on the contour lines of the inlet and outlet of the hydraulic machinery, and to construct a fully parameterized geometric model of the flow channel by combining the shape parameters with geometric constraints using auxiliary design technology.

[0061] The computational sample library construction unit 202 is used to substitute the initial sample points into the fully parameterized geometric model of the flow channel, reconstruct the contour curves of the inlet and outlet, and generate high-fidelity sample sets and low-fidelity sample sets after judging the degree of satisfaction of the reconstruction results and geometric constraints, thus constructing a multi-fidelity computational sample library.

[0062] Entropy production calculation unit 203 is used to post-process the multi-fidelity calculation sample library, calculate the direct entropy production and turbulent entropy production of the inlet and outlet channels, and establish an entropy production objective function with the goal of minimizing the total energy dissipation of the entire channel.

[0063] The surrogate model construction unit 204 is used to analyze the correlation mapping relationship between the high-fidelity sample set and the low-fidelity sample set based on autoregression technology, and to determine the process variance based on the mapping relationship to construct the covariance matrix and generate a co-kriging surrogate model.

[0064] The design variable combination output unit 205 is used to use the co-Kriging agent model as a fitness evaluation function, perform the Levy flight position update process, and combine it with the entropy production objective function to output the design variable combination for optimizing the shape of the hydraulic machinery inlet and outlet.

[0065] In this optional embodiment, the geometric model building unit 201 includes:

[0066] The flow channel shape parameter determination module is used to construct the parametric contour lines of the top plates of the inlet and outlet of the hydraulic machinery on the longitudinal section of the inlet and outlet based on the non-uniform rational spline technology, and to determine the flow channel shape parameters of the inlet and outlet using the parametric contour curves.

[0067] The optimization object parameter definition module is used to perform three-dimensional solid modeling of the fluid domain based on the inlet and outlet flow channel shape parameters using auxiliary design technology, and selects the top plate curves of the diffuser section of the inlet and outlet as optimization objects, defining the degree of non-uniform rational splines.

[0068] The geometric model output module is used to select control points based on the definition results to generate geometric constraints including the overall diffusion angle and the local diffusion angle. The geometric constraints are used to constrain the three-dimensional solid modeling results of the fluid domain to optimize the top plate curves of the diffusion section of the inlet and outlet. Based on the optimization results, a fully parameterized geometric model of the flow channel is output.

[0069] In this optional embodiment, the computational sample library construction unit 202 includes:

[0070] The equal probability interval division module is used to define the inlet and outlet shape optimization problem as a hypercube, and to use the control points and their corresponding weight factors as input variables of the hypercube. At the same time, the range of values ​​of the input variables is divided into several equal probability intervals.

[0071] The top plate curve reconstruction module is used to arbitrarily select input variables in each equal probability interval, shuffle them, generate initial sample points based on the processing results, and substitute the initial sample points into the fully parameterized geometric model of the flow channel to reconstruct the top plate curves of the diffuser sections of the inlet and outlet.

[0072] The sample set output module is used to determine the degree of satisfaction between the reconstructed inlet and outlet diffuser section top plate curves and the overall and local diffuser angle constraints. Based on the degree of satisfaction, the initial sample points that do not meet the constraints are removed, and the initial sample points are re-acquired until the number of initial sample points that meet the constraints reaches the target number and the valid sample set is output.

[0073] The computational sample construction module is used to generate high-fidelity and low-fidelity sample sets based on the effective sample set, using the maximum-minimum distance criterion and nested sampling strategy. Based on the high-fidelity and low-fidelity sample sets, fluid dynamics simulations are performed, and a multi-fidelity computational sample library is constructed based on the simulation results.

[0074] In this optional embodiment, the computational sample construction module includes:

[0075] The Euclidean distance calculation submodule is used to take the effective sample set containing all sample points as the low-fidelity sample set, randomly select sample points from the low-fidelity sample set and add them to the high-fidelity sample set, and calculate the dimensional space Euclidean distance of the unselected sample points in the low-fidelity sample set.

[0076] The high-fidelity sample filtering module is used to filter the minimum distance of randomly selected sample points with the minimum Euclidean distance in the minimum dimension space, compare the minimum distance of all unselected sample points, and select the unselected sample points with the maximum minimum distance based on the comparison results, and add them to the high-fidelity sample set.

[0077] The high-fidelity sample output module is used to repeatedly select unselected sample points and add them to the high-fidelity sample set until the number of samples in the high-fidelity sample set reaches a preset number, and then output the high-fidelity sample set.

[0078] The computational sample library generation module is used to determine the unoptimized geometric boundaries of the top plate curves of the diffuser sections at the inlet and outlet by combining high-fidelity sample sets and low-fidelity sample sets, and to construct a multi-fidelity computational sample library by combining fluid dynamics numerical simulation technology.

[0079] In this optional embodiment, the agent model construction unit 204 includes:

[0080] The estimation output module is used to obtain the condensed log-likelihood function of spatial hyperparameters based on high-fidelity sample sets and low-fidelity sample sets through the maximum likelihood estimation method, and output the optimal hyperparameter estimate based on the condensed log-likelihood function.

[0081] The variance matrix construction module is used to substitute the optimal hyperparameter estimates into the analytical expression to determine the final process variance and scaling factor, so as to complete the construction of the multifidelity covariance matrix of the co-kriging model.

[0082] The model fusion training module is used to divide the high-fidelity sample set into several groups to select the training set based on the multi-fidelity covariance matrix, and then fuse the training set with the low-fidelity sample set to train the co-kriging model.

[0083] The model optimization output module is used to verify the global generalization accuracy of the co-kriging model after training. Based on the verification results, it automatically adds sample points from a high-fidelity sample set to ensure the accuracy of the co-kriging model and outputs the final co-kriging surrogate model.

[0084] like Figures 4 to 9 As shown, to facilitate understanding of the above technical solution of the present invention, the following description further illustrates the above technical solution of the present invention from the perspective of architecture and principle, based on the optimization of the side-type inlet and outlet of a pumped storage power station, as follows:

[0085] Step 1: Construct a NURBS parametric model;

[0086] A three-dimensional geometric model of the inlet and outlet flow channels of hydraulic machinery was constructed using Non-Uniform Rational B-Spline (NURBS) technology. The specific implementation steps were as follows: First, NURBS technology was used to construct parametric profile lines of the inlet / outlet top plates on the longitudinal sections of the inlet / outlet. Based on this profile shape and fixing the remaining volumetric parameters of the flow channels, the model was imported into CAD software for three-dimensional solid modeling of the fluid domain. For the top plate profile of the diffuser section, non-uniform node vectors were used. (For cubic curves), through recursive derivation of basis functions, it is ensured that the curve has G at its endpoints. 1 Continuity, introducing a weighting factor w i Then, the equation of the curve is defined as:

[0087] ;

[0088] In the formula, u represents the normalized node parameter, with a value range of [0,1], representing the parameterized position of the curve along the axis; n represents the number of control points minus one; P i The position of the i-th control point determines the geometry of the curve; w i N represents the weighting factor corresponding to the i-th control point, indicating the degree to which the adjustment curve is applied to that control point, in order to achieve fine-tuning of the local curvature of the flow channel without changing the vertices of the control polygon; i,p (u) represents the p-th order B-spline basis function.

[0089] This example selects the top plate curve of the diffuser section at the inlet / outlet as the optimization object, defines the NURBS curve degree as p=3, and selects 6 control points p0, p1, p2, p3, p4, and p5, where the positions of p0 (inlet) and p5 (outlet) are fixed to match the upstream and downstream geometric boundary conditions. The x-axis values ​​of p1, p2, p3, and p4 are... i The coordinates are evenly distributed along the axis of the flow channel, with the ordinate y = 0. i The initial positions are equidistantly distributed along the vertical direction on the profile. The optimization variables include the ordinates of the four middle control points (p1 to p4) and the corresponding weighting factors w. i There are a total of 8 variables.

[0090] To ensure the engineering feasibility of the optimization results, the following geometric constraints are set:

[0091] Tangential constraint at the connection: Ensure that the tangents at the top plate p0 and p5 are collinear with the connection section (adjustment section and tunnel); Excavation envelope constraint: The movement range of intermediate control points must be limited within the geological excavation envelope to prevent intrusion into the self-stabilizing range of the rock mass; Overall diffusion angle constraint: According to the "Design Code for Pumped Storage Power Stations", the overall diffusion angle of the top plate profile (i.e., the angle between the line connecting p0 and p5 and the horizontal line) is controlled between 3° and 5°; Local diffusion angle constraint: To limit flow separation within the channel, the line connecting any two adjacent control points (i.e., P5 and P6) must be collinear with the horizontal line. i To P i+1 The resulting local diffusion angle is between 0° and 12°, ensuring that the top plate profile will not undergo local abrupt changes.

[0092] Furthermore, based on NURBS-based high-order geometric parameterization, the following technical effects were achieved:

[0093] Abandoning the traditional Bezier curve: Non-uniform rational B-splines (NURBS) are used to parameterize the flow channel profile. NURBS introduces the concepts of weights and knot vectors; Local control capability: NURBS basis functions have local support; moving a control point or adjusting its weight only affects a local region of the curve without disrupting the overall flow regime; Weight adjustment: Even if the coordinates of a control point remain unchanged, increasing the weight of a control point can pull the curve toward that point; conversely, decreasing it will push it away, providing additional degrees of freedom for fine-tuning the flow channel; Precise representation: NURBS can accurately represent conic sections (such as ellipses and hyperbolas), which is crucial for designing transition sections with specific hydraulic characteristics.

[0094] Step 2: Generation and filtering of the initial sample set;

[0095] Parameter space definition: The design space of the optimization problem is defined as an N=8-dimensional hypercube, with input variables X=[y1, y2, y3, y4, w1, w2, w3, w4], where y i w represents the ordinate of the four control points in the middle. i These are the corresponding weighting factors.

[0096] Super Latin Cubic Sampling (LHS): Divide the range of values ​​for each variable into M equally probable intervals (M is 100 in this example), randomly select a value from each interval and shuffle it to generate the initial sample matrix S. initial This ensures that the projection of each dimension satisfies a uniform distribution.

[0097] Geometric constraint filtering: The generated sample points are substituted into the NURBS equation to reconstruct the top plate curve. The global and local diffusion angle constraints are checked, and sample points that do not meet the constraints are highlighted and re-collected until 100 valid sample sets S are obtained. valid .

[0098] Step 3: Multi-fidelity sample stratification and model reconstruction;

[0099] (1) From the filtered valid sample set S valid Using nested sampling to extract subsets:

[0100] Low-fidelity (LF) sample set: contains all M sample points, used to calculate the overall trend of the design space;

[0101] High-fidelity (HF) sample set: A nested sampling strategy is adopted, and based on the maximum-minimum distance criterion, K=10 sample points (covering the boundary and central region) are selected from the above sample set.

[0102] (2) The specific steps of the nested sampling strategy are as follows:

[0103] Given M low-fidelity sample sets S LF As a candidate sample pool; the first sample point is randomly selected from the candidate sample pool and added to the high-fidelity sample set S. HF In the middle, all design variables in the candidate sample pool are mapped and normalized to the dimensionless interval [0,1] to eliminate the coordinate variable y. i and weighting factor w i The distance calculation bias is caused by differences in units and orders of magnitude. Then, the remaining unselected sample points in the candidate sample pool are calculated. N-dimensional space Euclidean distance The calculation formula is:

[0104] ;

[0105] In the formula, N represents the total dimensions of the design variables, which is 8 in this example. This represents the normalized parameter value of the unselected candidate sample point in the j-th dimension. This represents the normalized parameter value of the selected high-fidelity sample points in the j-th dimension; calculate... To all selected points After calculating the distances, the minimum value among them is taken as the minimum distance of the candidate sample point, denoted as D. min (X c )=min{ d (X c ,X s )},∀X s ∈S HF Compare the minimum distances of all unselected sample points, select the sample point whose minimum distance is the maximum value, and add it to the high-fidelity sample set S. HF In the middle; repeat the generation until a high-fidelity sample set S is reached. HF With the number of samples reaching the preset K=10, this strategy can ensure that the extracted high-fidelity samples are evenly distributed throughout the design space, effectively covering the boundaries and central areas of the design space, and avoiding local data aggregation.

[0106] (3) Based on each sample vector X i The corresponding top plate curves are obtained, and a three-dimensional fluid domain model of the inlet / outlet is constructed based on this for CFD calculations:

[0107] The control point coordinates y of the sample points generated in (2) i and weighting factor w i Import the data into UG NX software, and use the software's artistic spline curve function to generate a curve that satisfies G based on the specified node vectors, control point coordinates, and corresponding weighting factors. 1 or G 2 A continuous, highly smooth, three-dimensional parametric top plate profile is constructed. Based on the drawings, the unoptimized fixed geometric boundary parts of the inlet / outlet are constructed, and the surface is generated using the sweep function. The closed top plate surface and the fixed boundary surface are then stitched together to form a three-dimensional fluid domain solid model. The model is imported into ICEM CFD software, the boundary conditions of the model are set, and the mesh is generated. During the mesh generation process, the boundary layer is densified according to different fidelity requirements.

[0108] (4) Based on the fidelity of the sample set and in order to balance computational cost and accuracy, a differentiated numerical simulation strategy is adopted, where y + y is a dimensionless wall distance, a dimensionless parameter characterizing the relative distance between the first-layer grid node and the solid wall of the flow channel in a CFD computational grid. + The numerical value determines the resolution of the boundary layer mesh and the selection criteria for the turbulence model:

[0109] LF simulation: Sparse gridding (grid number < 3 × 10⁻⁶) for low-fidelity samples 6 And y + >30, quickly obtain the calculation results;

[0110] HF simulation: High-fidelity samples are divided into fine grids (grid number > 8 × 10⁻⁶). 6 And y + ≈1, to calculate high-precision results.

[0111] Step 4: Calculate the entropy production objective function;

[0112] Based on the second law of thermodynamics and the Reynolds-averaged time (RANS) assumption, the local entropy yield in incompressible turbulent flow It consists of two parts: Direct Entropy Production (EPDD) caused by the time-averaged velocity field and Turbulent Entropy Production (EPTD) caused by the fluctuating velocity field.

[0113] Direct entropy production (EPDD) originates from the molecular viscosity of the fluid and the time-averaged velocity gradient of the flow field. According to the constitutive equations of fluid mechanics, its calculation formula is as follows:

[0114] ;

[0115] The turbulence model uses the following empirical formula to accurately calculate this term:

[0116] ;

[0117] in, .

[0118] Overall objective function: In this embodiment, the objective is to minimize the total energy dissipation across the entire flow channel. For the above two items throughout the computational domain The sum of volume integrals within:

[0119] ;

[0120] The first term is the direct entropy production caused by the time-averaged velocity gradient (EPDD), and the second term is the turbulent entropy production caused by fluctuating velocity (EPTD). This represents the volume of the flow channel computational domain; Indicates fluid dynamic viscosity, Thermodynamic temperature; express The empirical constant for the turbulence model is set to 0.09. Indicates fluid density; Indicates the characteristic frequency of turbulence; Represents turbulent kinetic energy; Let the square of the time-averaged velocity deformation tensor modulus be denoted as . This represents the target value of entropy production. Represents the direct entropy production rate. Represents the turbulent entropy production rate. Represents the differential symbol. , , They represent , , The time-averaged velocity component in the direction, , , These represent the three coordinate components in the Cartesian coordinate system. It represents a partial differential.

[0121] Compared to the traditional differential pressure method, this indicator can accurately locate high-energy-consuming areas (i.e., entropy-generating hot spots) inside the flow channel caused by flow separation and vortex breaking.

[0122] Therefore, this embodiment no longer relies solely on the total pressure difference, but instead, based on the second law of thermodynamics, introduces the entropy production rate as the core evaluation index. Based on the Reynolds average (RANS) assumption, the irreversible energy loss in the flow channel is decomposed into the following two parts, and the system automatically calculates the sum of their volume integrals over the entire domain as the optimization objective:

[0123] Direct Entropy Production (EPD): Originating from the molecular viscosity of the fluid and the velocity gradient of the time-averaged velocity field, it represents the viscous frictional dissipation caused by the time-averaged shear motion between fluid layers. This invention calculates this term precisely based on the dynamic viscosity of the fluid and the square of the velocity deformation tensor modulus.

[0124] Turbulent Entropy Production (EPTD): Originating from the turbulent fluctuating velocity field, it represents the energy loss during the breakup of large-scale eddies into smaller-scale eddies. Specifically, it relates to the SST used in this embodiment. Turbulence models, using empirical formulas This indicator, when calculated precisely, can accurately capture the region of strong turbulent dissipation caused by flow separation within the flow channel.

[0125] The optimization objective is to change the traditional objective of minimizing pressure difference to minimizing the total entropy production within the entire fluid domain (i.e., the sum of direct entropy production and turbulent entropy production). Since the hydraulic losses in the inlet and outlet channels mainly originate from the large-scale vortex dissipation caused by flow separation in the diffuser section, this evaluation system can more accurately locate and suppress hot spots (high energy loss zones) inside the channel than the pressure difference method, thereby effectively eliminating undesirable flow regimes.

[0126] Step 5: Co-Kriging model training;

[0127] Construct the covariance matrix C, and solve for the hyperparameters using 100 sets of LF data and 10 sets of HF data through maximum likelihood estimation (MLE). and scaling factor The specific optimization derivation steps are as follows: Based on the Gaussian random process assumption, the observation data of M low-fidelity samples and K high-fidelity samples are combined to construct a joint maximum likelihood function of the occurrence probability of multi-fidelity CFD training samples. To simplify the multiplication and differentiation calculations in the nonlinear optimization process, the natural logarithm of the joint likelihood function is taken to obtain the log-likelihood function. ; respectively for the log-likelihood function Process variance and scaling factor By taking the partial derivatives and setting them to zero, the variance and scaling factor can be derived. Regarding space hyperparameters The analytical expression is obtained by substituting it back into the original log-likelihood function and eliminating these variables, resulting in a function containing only the unknown hyperparameters. The condensed log-likelihood function is obtained; a nonlinear unconstrained optimization algorithm is used to maximize the condensed log-likelihood function. When the log-likelihood function reaches its maximum value, the solution at this point is the optimal estimate of the spatial correlation hyperparameters. Substituting the obtained optimal hyperparameters back into the analytical expression, the final scaling factor is calculated. Combine process variance to complete the multifidelity covariance matrix. Construction and training of proxy models.

[0128] The covariance matrix of the co-kriging model It has a block structure:

[0129] ;

[0130] in, and These represent the low-fidelity and high-fidelity sample sets, respectively. and Let these represent the spatial correlation functions of the low-fidelity model and the difference model, respectively. and This represents the process variance.

[0131] Validation: Leave-one-out cross-validation was used to ensure the multiple correlation coefficient of the surrogate model. If the accuracy is insufficient, the Expectation-Maximum Improvement (EI) criterion is used to automatically add HF sample points at the point of maximum error. The specific steps of leave-one-out cross-validation are as follows: Divide the K high-fidelity samples into K groups. In the i-th round of validation (i=1,2,…,K), the i-th sample is extracted as the validation set, and the remaining K-1 samples are used as the training set. The co-Kriging surrogate model is retrained using the K-1 training points and all low-fidelity data. At this time, the input features of the model are the 8 variables defined in step two. The trained Co-Kriging model is used to predict the objective function of the extracted i-th sample. And calculate its value y compared with the CFD calculated value. i Prediction error between Repeat the above steps K times to iterate through all high-fidelity samples, and calculate the global generalization accuracy based on the following formula:

[0132] ;

[0133] like If the model accuracy is reasonable, proceed to the next optimization step. Then, at the location with the largest mean square error in the design space, i.e. the location of the most uncertain set shape in the model, new high-fidelity sample points are adaptively added, and the CFD calculation results are used to update the sample library until the accuracy meets the requirements.

[0134] Furthermore, a Co-Kriging surrogate model was constructed, utilizing a large amount of low-fidelity data (LF, coarse grid, fast computation) to capture the overall trend of the objective function; and a small amount of high-fidelity data (HF, fine grid, slow computation) to correct the model's magnitude error. An autoregressive architecture was then established. The mathematical correlation, in which Indicates the scaling factor. Indicates a high-fidelity process. Indicates a low-fidelity process. The difference between the two is represented by a Gaussian stochastic process; the model calibrates the trend bias of the low-fidelity data by minimizing the mean square error (MSE) and maximizes the information gain. With this fusion, the prediction accuracy close to that of the full high-precision simulation can be obtained with only about 1 / 10 of the computational cost of the high-precision simulation.

[0135] The objective function also includes a two-way operating condition coupling penalty term:

[0136] ;

[0137] in, and These represent the total entropy output under turbine and pump operating conditions, respectively. This represents the penalty function, applied to the non-uniformity of the outlet velocity distribution in the diffuser section under pump operating conditions. When the threshold is exceeded, the value increases exponentially.

[0138] Step Six: ISSA Global Optimization;

[0139] Population initialization: using Cubic chaotic mapping Generate the initial positions of 50 sparrows.

[0140] The specific mathematical expression for initializing the sparrow population is as follows:

[0141] ;

[0142] in, Indicates the first The values ​​of the chaotic variables are within the range of... , This represents the control parameter, with values ​​ranging from 2.595 to 3.

[0143] Iterative Search: The discoverer (accounting for 20%) is responsible for the global search; if fitness stagnates, it performs a Levi flight jump. ;

[0144] The improved formula for updating the discoverer's location is as follows: ,in, Indicates the step size control factor. This represents point-to-point multiplication. This represents a random path that follows a Lévy distribution, used to enhance the population's ability to escape local optima in the later stages of algorithm iteration.

[0145] Joiners (80%) follow the discoverers but make minor adjustments locally; watchdogs (10%) are randomly generated to prevent the algorithm from getting stuck in local optima; bidirectional constraint: when evaluating fitness, if the velocity non-uniformity coefficient of a design under pumping conditions is... Then apply a 10% threshold to its entropy production target value. 4 The penalty function forces the population to move to a region that balances both aspects.

[0146] Furthermore, to address the premature convergence problem of the standard SSA algorithm, two improvements can be introduced:

[0147] Cubic chaotic mapping initialization: The ergodicity of the chaotic sequence is used to generate an initial population, ensuring that the design scheme is uniformly distributed in the solution space.

[0148] Levy Flight Perturbation: A random step size following a Levy distribution is added to the discoverer position update formula. Levy Flight has the characteristic of alternating long and short step sizes, which can help the algorithm escape the local optimum trap and perform extensive global search.

[0149] Addition Strategy: After each ISSA iteration, calculate the co-kriging prediction variance at the current optimal solution; if the prediction variance is greater than a preset threshold, call high-fidelity CFD to calculate the point, update the result to the high-fidelity sample library, and retrain the surrogate model.

[0150] Step 7: Results Analysis;

[0151] In the comparison of flow regime and entropy production distribution characteristics, Figure 7 This paper presents a comparison of the entropy productivity (LTEPD) distribution and streamlines within the longitudinal profile of the flow channel before and after optimization, under controlled operating conditions (pumping conditions). Before optimization (a): as shown in the image. Figure 7 As shown above, significant large-scale flow separation and recirculation vortexes were observed behind the top plate of the initially designed diffuser section. The streamlines were violently curled and broken at this point, resulting in a reduction in the effective area of ​​the flow cross section. Accompanying this recirculation structure, this region and the downstream wake region exhibited a high-intensity entropy production distribution (bright yellow / green areas in the figure), indicating that there is severe turbulent energy dissipation at this point.

[0152] After optimization (b): such as Figure 7 As shown below, after NURBS parameterization and ISSA algorithm optimization in this embodiment, the top plate profile of the flow channel underwent adaptive fine-tuning (exhibiting an S-shaped downstream flow characteristic). The optimized streamlines smoothly transition along the wall, and the original large-scale backflow vortices on the top plate are completely eliminated, achieving ideal wall-attached flow. The overall color of the flow field is mainly dark blue (low-entropy production region), with only a very thin boundary layer dissipation near the wall, indicating that the mechanical energy loss inside the fluid has been effectively suppressed.

[0153] Table 1, a quantitative performance index analysis table, details the comparison of entropy production values ​​for each component under bidirectional operating conditions before and after optimization.

[0154] Table 1. Comparison of entropy production values ​​for each component under bidirectional operating conditions before and after optimization.

[0155]

[0156] The weighted total revenue is calculated based on equal weights under bidirectional operating conditions, and the weighted total entropy output of the flow channel is significantly reduced from 2297.62 to [a lower value]. The overall energy loss decreased by 23.00%.

[0157] The pumping operation condition represents the worst hydraulic performance in pumping and diffusion. The total entropy production decreased from 3938.87 to 2841.60, a significant reduction of 27.86%. Specifically, the turbulent entropy production (pulsating entropy production), representing vortex dissipation intensity, decreased from 3647.21 to 2781.95, a reduction of 23.72%. This data is consistent with... Figure 7 The phenomenon of the disappearance of the mid-flow vortex is highly consistent, quantitatively confirming the excellent suppression ability of the method of the present invention for turbulent vortices.

[0158] Although the total entropy production in the power generation mode increases slightly (from 656.37 to 696.72, an increase of about 6.1%), considering that the original energy consumption base of the pumping mode is about 6 times that of the power generation mode (3938.87 vs 656.37), the strategy adopted in this embodiment is to sacrifice a very small amount of non-control mode performance in exchange for a huge performance improvement in the control mode. This strategy achieves a significant reduction in weighted total energy consumption at the level, and significantly improves the overall operating economy and flow stability of the power plant.

[0159] Furthermore, this embodiment uses entropy production analysis to intuitively demonstrate the hot spot region of energy loss, transforming the optimization design from black-box trial and error to targeted treatment. The Co-Kriging model utilizes inexpensive data as a foundation and expensive data for calibration, reducing computational costs by more than 90% while ensuring physical realism. Relying on the local support characteristics and weight factor adjustment function of NURBS technology, this embodiment breaks the coupling constraint between overall smoothness and local features in traditional curve generation methods, realizing independent degree of freedom control of key parts of the inlet and outlet, significantly improving the adaptability and success rate of channel shape optimization under complex boundary conditions. At the same time, the ISSA algorithm, which integrates chaos and Lévy flight, effectively avoids getting trapped in local optima and can discover innovative topological solutions that traditional gradient algorithms cannot find.

[0160] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores static and dynamic information data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0161] Those skilled in the art will understand that Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0162] In addition, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0163] In addition, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0164] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0165] This invention is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this invention is limited only by the appended claims.

Claims

1. A method for optimizing the size of the intake and outlet of a pumped storage power plant, characterized by, include: Based on the contour lines of the inlet and outlet of the hydraulic machinery, the shape parameters of the inlet and outlet flow channels are determined. The shape parameters are combined with geometric constraints and auxiliary design technology is used to construct a fully parameterized geometric model of the flow channels. Substitute the initial sample points into the fully parameterized geometric model of the flow channel to reconstruct the inlet and outlet contour curves. After judging the degree of satisfaction of the reconstruction results and geometric constraints, generate high-fidelity sample sets and low-fidelity sample sets to construct a multi-fidelity calculation sample library. The multi-fidelity calculation sample library is post-processed to calculate the direct entropy production rate and turbulent entropy production rate of the inlet and outlet channels, and an entropy production objective function is established with the goal of minimizing the total energy dissipation of the entire channel. Based on the autoregressive technique, the correlation mapping relationship between the high-fidelity sample set and the low-fidelity sample set is analyzed, and the process variance is determined according to the mapping relationship to construct the covariance matrix and generate a co-kriging surrogate model. The co-Kriging agent model is used as the fitness evaluation function to perform the Levy flight position update process, and combined with the entropy production objective function to output the combination of design variables for optimizing the shape of the hydraulic machinery inlet and outlet.

2. The method of claim 1, wherein, The process of determining the shape parameters of the inlet and outlet flow channels based on the contour lines of the hydraulic machinery inlet and outlet, and then constructing a fully parametric geometric model of the flow channels by combining the shape parameters with geometric constraints using computer-aided design techniques includes: Based on the non-uniform rational spline technique, parametric contour lines of the top plates of the inlet and outlet of the hydraulic machinery are constructed on the longitudinal section of the inlet and outlet. The body shape parameters of the inlet and outlet flow channels are determined by using the parametric contour curves. Using computer-aided design techniques, a three-dimensional solid model of the fluid domain is created based on the shape parameters of the inlet and outlet channels. The top plate curves of the diffuser sections of the inlet and outlet are selected as optimization objects, and the degree of non-uniform rational splines is defined. Based on the definition results, control points are selected to generate geometric constraints including the overall diffusion angle and the local diffusion angle. The geometric constraints are used to constrain the three-dimensional solid modeling results of the fluid domain to optimize the top plate curves of the diffusion sections of the inlet and outlet. Based on the optimization results, a fully parameterized geometric model of the flow channel is output.

3. The method of claim 1, wherein, The process of substituting initial sample points into the fully parameterized geometric model of the flow channel, reconstructing the inlet and outlet contour curves, and generating high-fidelity and low-fidelity sample sets after judging the degree of satisfaction of the reconstruction results and geometric constraints, and constructing a multi-fidelity calculation sample library includes: The problem of optimizing the shape of the inlet and outlet is defined as a hypercube, and the control points and their corresponding weight factors are used as input variables of the hypercube. At the same time, the range of values ​​of the input variables is divided into several equal probability intervals. In each equal probability interval, input variables are randomly selected and arranged. Initial sample points are generated based on the processing results. The initial sample points are then substituted into the fully parameterized geometric model of the flow channel to reconstruct the top plate curves of the diffuser section of the inlet and outlet. Determine the degree of satisfaction between the reconstructed inlet and outlet diffuser section top plate curves and the overall and local diffuser angle constraints. Based on the degree of satisfaction, remove initial sample points that do not meet the constraints and re-collect initial sample points until the number of initial sample points that meet the constraints reaches the target number and output the effective sample set. Based on the effective sample set, high-fidelity sample sets and low-fidelity sample sets are generated using the maximum-minimum distance criterion and nested sampling strategy. Fluid dynamics simulations are then performed based on the high-fidelity sample sets and low-fidelity sample sets, and a multi-fidelity computational sample library is constructed based on the simulation results.

4. The method for optimizing the intake body of a pumped storage power plant according to claim 3, characterized in that, Based on the effective sample set, high-fidelity and low-fidelity sample sets are generated using the maximum-minimum distance criterion and a nested sampling strategy. Fluid dynamics simulations are then performed based on these high-fidelity and low-fidelity sample sets. A multi-fidelity computational sample library is constructed based on the simulation results, including: The effective sample set containing all sample points is used as the low-fidelity sample set. Sample points are randomly selected from the low-fidelity sample set and added to the high-fidelity sample set. The dimensional space Euclidean distance of the unselected sample points in the low-fidelity sample set is calculated. The minimum dimensional space Euclidean distance is selected as the minimum distance of randomly selected sample points, and the minimum distance of all unselected sample points is compared. Based on the comparison results, the unselected sample points with the maximum minimum distance are selected and added to the high-fidelity sample set. Repeatedly select unselected sample points and add them to the high-fidelity sample set until the number of samples in the high-fidelity sample set reaches the preset number, and then output the high-fidelity sample set. The high-fidelity sample set and the low-fidelity sample set are used to determine the unoptimized geometric boundaries of the top plate curves of the diffuser section at the inlet and outlet, and a multi-fidelity calculation sample library is constructed by combining fluid dynamics numerical simulation technology.

5. The method of claim 4, wherein, The step of determining the unoptimized geometric boundaries of the top plate curves of the diffuser sections at the inlet and outlet by combining high-fidelity and low-fidelity sample sets, and constructing a multi-fidelity calculation sample library by combining fluid dynamics numerical simulation technology includes: Obtain the control point coordinates and weight factors corresponding to the high-fidelity sample set and the low-fidelity sample set, and import the control point coordinates and weight factors into the 3D software to generate the 3D parametric top plate profile. Based on the fully parameterized geometric model of the flow channel, the fixed geometric boundaries of the top plate curves of the diffuser section of the inlet and outlet are determined. The closed top plate surface is generated by combining the sweeping technique, and the closed top plate surface and the fixed geometric boundary are stitched together to generate a three-dimensional fluid domain solid model. The high-fidelity sample set is processed by fine mesh generation, and high-fidelity computational fluid dynamics numerical simulation is performed by running a three-dimensional fluid domain solid model in combination with shear stress turbulence technology. The low-fidelity sample set is divided into sparse grids, and a three-dimensional fluid domain solid model is run using both high-fidelity and low-fidelity turbulence techniques to perform low-fidelity computational fluid dynamics numerical simulations. Based on the high-fidelity and low-fidelity simulation results, the flow field data is output to construct a multi-fidelity computational sample library.

6. The method of claim 5, wherein, The expression for the entropy production objective function is: ； ； In the formula, This represents the target value of entropy production. This represents the volume of the computational domain of the inlet and outlet flow channels. Represents the direct entropy production rate. Represents the turbulent entropy production rate. Represents the differential symbol. Indicates the dynamic viscosity of a fluid. Represents the thermodynamic temperature of a fluid. Let represent the square of the time-averaged velocity deformation tensor modulus. This represents the empirical constants of the turbulence techniques used by both parties. Indicates fluid density, Represents the characteristic frequency of turbulence. Represents turbulent kinetic energy. , , They represent , , The time-averaged velocity component in the direction, , , These represent the three coordinate components in the Cartesian coordinate system. It represents a partial differential.

7. The method for optimizing the inlet and outlet shapes of a pumped storage power station according to claim 1, characterized in that, The method of analyzing the correlation mapping relationship between high-fidelity sample sets and low-fidelity sample sets based on autoregressive techniques, and determining the process variance and constructing the covariance matrix to generate a co-kriging surrogate model based on the mapping relationship includes: Based on high-fidelity and low-fidelity sample sets, the condensed log-likelihood function of spatial hyperparameters is obtained by the maximum likelihood estimation method, and the optimal hyperparameter estimate is output based on the condensed log-likelihood function. Substitute the optimal hyperparameter estimates into the analytical expression to determine the final process variance and scaling factor, thereby completing the construction of the multifidelity covariance matrix of the co-kriging model; Based on the multifidelity covariance matrix, the high-fidelity sample set is divided into several groups to select the training set, and the training set is fused with the low-fidelity sample set to train the co-kriging model. After verifying the global generalization accuracy of the co-kriging model after training, sample points of a high-fidelity sample set are automatically added based on the verification results to ensure the accuracy of the co-kriging model, and the final co-kriging surrogate model is output.

8. The method for optimizing the shape of the inlet and outlet of a pumped storage power station according to claim 7, characterized in that, The process involves using high-fidelity and low-fidelity sample sets as a basis, obtaining the condensed log-likelihood function of spatial hyperparameters through maximum likelihood estimation, and outputting the optimal hyperparameter estimates based on the condensed log-likelihood function. Based on Gaussian random processes, a joint maximum likelihood function for the occurrence probability of multi-fidelity fluid dynamics training samples is constructed by combining high-fidelity and low-fidelity sample sets. The natural logarithm of the joint maximum likelihood function is then selected to obtain the log-likelihood function. Find the partial derivatives of the process variance and scaling factor of the log-likelihood function, and define the partial derivatives as zero after finding them. Derive the analytical expressions of the process variance and scaling factor with respect to the spatial hyperparameters. Substituting the analytical expression into the log-likelihood function eliminates the process variance and scaling factor, resulting in a condensed log-likelihood function containing only spatial hyperparameters. Then, nonlinear unconstrained optimization techniques are used to maximize the condensed log-likelihood function. When the condensed log-likelihood function reaches its maximum value based on the maximization optimization result, the corresponding solution is taken as the optimal hyperparameter estimate.

9. The method for optimizing the shape of the inlet and outlet of a pumped storage power station according to claim 1, characterized in that, The process of using the co-Kriging agent model as a fitness evaluation function to perform Lévy flight position update processing, and combining it with the entropy production objective function to output the design variable combination for optimizing the shape of the hydraulic machinery inlet and outlet, includes: The initial positions of several individuals are generated based on cubic chaotic mapping to construct an initial sparrow population. The fitness of the body size optimization scheme corresponding to the position of each sparrow individual in the initial sparrow population is calculated using the co-kriging agent model as the fitness evaluation function, and the quality of the body size optimization scheme is evaluated. Based on the evaluation results of superiority and inferiority, a random step size following the Lévy distribution is added to determine the non-uniformity coefficient of the flow velocity in the pumping condition. Based on the determination results, a penalty function is applied to the target value of the entropy production objective function corresponding to the current individual, so as to make the initial sparrow population move to the area that takes both sides into account. Based on the regional movement results, the output of the body shape optimization variable combination corresponding to the body shape of the inlet and outlet flow channels is used to reconstruct the fully parameterized geometric model of the flow channel based on the body shape optimization variable combination. Fluid dynamics numerical simulation is then performed to determine the body shape optimization design variable combination scheme of the hydraulic machinery inlet and outlet.

10. A pumped storage power station inlet and outlet shape optimization system, characterized in that, include: The geometric model construction unit is used to determine the shape parameters of the inlet and outlet flow channels based on the contour lines of the inlet and outlet of the hydraulic machinery, and to construct a fully parameterized geometric model of the flow channel by combining the shape parameters with geometric constraints using auxiliary design technology. The computational sample library construction unit is used to substitute the initial sample points into the fully parameterized geometric model of the flow channel, reconstruct the contour curves of the inlet and outlet, and generate high-fidelity sample sets and low-fidelity sample sets after judging the degree of satisfaction of the reconstruction results and geometric constraints, thus constructing a multi-fidelity computational sample library. The entropy production calculation unit is used to post-process the multi-fidelity calculation sample library, calculate the direct entropy production and turbulent entropy production of the inlet and outlet channels, and establish an entropy production objective function with the goal of minimizing the total energy dissipation of the entire channel. The surrogate model construction unit is used to analyze the correlation mapping relationship between high-fidelity sample sets and low-fidelity sample sets based on autoregression techniques, and to determine the process variance based on the mapping relationship to construct the covariance matrix and generate a co-kriging surrogate model. The design variable combination output unit is used to use the co-Kriging agent model as the fitness evaluation function, perform the Lévy flight position update process, and combine it with the entropy production objective function to output the design variable combination for optimizing the shape of the inlet and outlet of the hydraulic machinery.

11. The pumped storage power station inlet and outlet shape optimization system according to claim 10, characterized in that, The geometric model construction unit includes: The flow channel shape parameter determination module is used to construct the parametric contour lines of the top plates of the inlet and outlet of the hydraulic machinery on the longitudinal section of the inlet and outlet based on the non-uniform rational spline technology, and to determine the flow channel shape parameters of the inlet and outlet using the parametric contour curves. The optimization object parameter definition module is used to perform three-dimensional solid modeling of the fluid domain based on the inlet and outlet flow channel shape parameters using auxiliary design technology, and selects the top plate curves of the diffuser section of the inlet and outlet as optimization objects, defining the degree of non-uniform rational splines. The geometric model output module is used to select control points based on the definition results to generate geometric constraints including the overall diffusion angle and the local diffusion angle. The geometric constraints are used to constrain the three-dimensional solid modeling results of the fluid domain to optimize the top plate curves of the diffusion section of the inlet and outlet. Based on the optimization results, a fully parameterized geometric model of the flow channel is output.

12. The pumped storage power station inlet and outlet shape optimization system according to claim 10, characterized in that, The computational sample library construction unit includes: The equal probability interval division module is used to define the inlet and outlet shape optimization problem as a hypercube, and to use the control points and their corresponding weight factors as input variables of the hypercube. At the same time, the range of values ​​of the input variables is divided into several equal probability intervals. The top plate curve reconstruction module is used to arbitrarily select input variables in each equal probability interval, shuffle them, generate initial sample points based on the processing results, and substitute the initial sample points into the fully parameterized geometric model of the flow channel to reconstruct the top plate curves of the diffuser sections of the inlet and outlet. The sample set output module is used to determine the degree of satisfaction between the reconstructed inlet and outlet diffuser section top plate curves and the overall and local diffuser angle constraints. Based on the degree of satisfaction, the initial sample points that do not meet the constraints are removed, and the initial sample points are re-acquired until the number of initial sample points that meet the constraints reaches the target number and the valid sample set is output. The computational sample construction module is used to generate high-fidelity and low-fidelity sample sets based on the effective sample set, using the maximum-minimum distance criterion and nested sampling strategy. Based on the high-fidelity and low-fidelity sample sets, fluid dynamics simulations are performed, and a multi-fidelity computational sample library is constructed based on the simulation results.

13. The pumped storage power station inlet and outlet shape optimization system according to claim 12, characterized in that, The computational sample construction module includes: The Euclidean distance calculation submodule is used to take the effective sample set containing all sample points as the low-fidelity sample set, randomly select sample points from the low-fidelity sample set and add them to the high-fidelity sample set, and calculate the dimensional space Euclidean distance of the unselected sample points in the low-fidelity sample set. The high-fidelity sample filtering module is used to filter the minimum distance of randomly selected sample points with the minimum Euclidean distance in the minimum dimension space, compare the minimum distance of all unselected sample points, and select the unselected sample points with the maximum minimum distance based on the comparison results, and add them to the high-fidelity sample set. The high-fidelity sample output module is used to repeatedly select unselected sample points and add them to the high-fidelity sample set until the number of samples in the high-fidelity sample set reaches a preset number, and then output the high-fidelity sample set. The computational sample library generation module is used to determine the unoptimized geometric boundaries of the top plate curves of the diffuser sections at the inlet and outlet by combining high-fidelity sample sets and low-fidelity sample sets, and to construct a multi-fidelity computational sample library by combining fluid dynamics numerical simulation technology.

14. The pumped storage power station inlet and outlet shape optimization system according to claim 10, characterized in that, The proxy model construction unit includes: The estimation output module is used to obtain the condensed log-likelihood function of spatial hyperparameters based on high-fidelity sample sets and low-fidelity sample sets through the maximum likelihood estimation method, and output the optimal hyperparameter estimate based on the condensed log-likelihood function. The variance matrix construction module is used to substitute the optimal hyperparameter estimates into the analytical expression to determine the final process variance and scaling factor, so as to complete the construction of the multifidelity covariance matrix of the co-kriging model. The model fusion training module is used to divide the high-fidelity sample set into several groups to select the training set based on the multi-fidelity covariance matrix, and then fuse the training set with the low-fidelity sample set to train the co-kriging model. The model optimization output module is used to verify the global generalization accuracy of the co-kriging model after training. Based on the verification results, it automatically adds sample points from a high-fidelity sample set to ensure the accuracy of the co-kriging model and outputs the final co-kriging surrogate model.

15. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 9.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

Images