Method for optimizing the shape of the transition section of the stilling well of a spillway shaft

Abstract

The application discloses a flood discharge vertical shaft stilling well-bend section body shape unified optimization method, which comprises the following steps: obtaining multi-working condition physical constraints describing the body shape parameters of the flood discharge vertical shaft, the stilling well and the bend section, and constructing a joint feasible region; obtaining a typical flood working condition set; adopting an active sampling strategy to obtain body shape sample data in the joint feasible region for the typical flood working condition set, and iteratively constructing a flow state partition model and a hydraulic index proxy model according to the sample data; the active sampling strategy combines multi-working condition uncertainty and flow state partition boundary information; in the joint feasible region, the flow state partition model and the hydraulic index proxy model are applied to construct and solve a robust optimization model considering cross-working condition flow state stability, and a unified optimization body shape scheme is obtained. Through the introduction of the flow state intermediate layer and the sampling strategy driven by the flow state boundary, the application improves the capturing ability and optimization efficiency of the nonlinear flow state mutation.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of hydraulic structure design and data-driven optimization, and particularly relates to a method for unified optimization of the body shape of a stilling well-turning section of a flood discharge shaft. BACKGROUND

[0002] The flood discharge shaft is a typical high-water-head discharge building and is a core component of a high dam and large reservoir water conservancy and hydropower hub. The stilling well and the turning section at the bottom of the shaft are key positions for energy dissipation and realization of vertical flow state conversion to horizontal flow state of the entire flood discharge shaft. The water flow at this position has strong turbulent, strong shear, and strong mixing characteristics, and the water flow is mixed with a large amount of air and accompanied by spiral vortex and severe rolling. The water flow structure is much more complex than other positions, and the extremely strong three-dimensionality may cause erosion and impact on the structure. The design of the body shape at this position is directly related to the smooth dissipation of the huge water flow energy during flood discharge, the safety of the structure from cavitation vibration, and the protection of the downstream river channel from scour. The stilling well and the turning section are connected in space, and due to the continuous medium characteristics of the water body, the two have overall unity in geometric characteristics and hydraulic performance. However, many previous engineering cases and tests show that the two have a seesaw effect between energy dissipation demand and engineering economy. In view of the above, it is necessary to design the stilling well and the turning section as a whole system, to finely optimize the body shape of the complex three-dimensional hydraulic system of the stilling well-turning section, to ensure that it can safely, stably, and efficiently operate under multiple typical flood conditions in the whole life cycle, to balance the fluid-structure interaction, to achieve the effect of 1+1>2 in energy dissipation and flow regulation, and to effectively improve the overall safety margin and engineering economy of the structure. The method has great engineering significance and urgent practical needs.

[0003] At present, in the body shape design of the stilling well-turning section of the flood discharge shaft, three-dimensional computational fluid dynamics (CFD) numerical simulation has become the mainstream analysis method. Researchers usually rely on CFD combined with physical model tests to evaluate the hydraulic characteristics of a limited number of preset design schemes. On this basis, some studies begin to try to combine surrogate models with multi-objective optimization algorithms. These methods usually use radial basis function (RBF) or Kriging models to fit the functional relationship between the body shape geometric parameters and several key hydraulic indicators (such as energy dissipation rate, bottom impact pressure, and outflow uniformity), and to optimize the solution on this basis to obtain relatively optimal body shape parameters.

[0004] However, the existing optimization method based on the agent model still has technical bottlenecks in characterizing the physical mechanism of high-speed water flow and lacking of cognition of key areas when dealing with such complex high-dimensional nonlinear hydraulic problems. The existing method generally constructs a single-layer black box mapping of the body type-hydraulic index, ignoring the root cause of the nonlinear mutation of the hydraulic index, that is, the transition of the flow pattern (flow state) under the change of water level and flow. For example, a small body adjustment or working condition change may cause the energy dissipater to suddenly change from a stable annular hydraulic jump to an unstable vortex flow. Such qualitative change of flow state cannot be fully characterized by a single and continuous energy dissipation rate or pressure index, resulting in a serious lack of accuracy of the agent model near the flow state mutation boundary. In addition, due to the lack of cognition of the flow state boundary, the existing active sampling strategy (such as sampling based on prediction variance) is also blind, and cannot actively identify and encrypt sampling of these flow state mutation areas which are crucial to the optimization results and structural safety, resulting in that the valuable computing resources are wasted in the smooth area with low information amount, and the key nonlinear boundary cannot be accurately captured by the model due to the sparseness of samples. SUMMARY

[0005] The object of the present application is to provide a flood discharge vertical shaft energy dissipater-bend section body type unified optimization method to solve the above problems existing in the prior art.

[0006] The technical scheme of the present application is a flood discharge vertical shaft energy dissipater-bend section body type unified optimization method, comprising:

[0007] Obtain the multi-condition physical constraints describing the body type parameters of the flood discharge vertical shaft, the energy dissipater and the bend section, and construct a joint feasible region; at the same time, obtain a set of typical flood working conditions;

[0008] An active sampling strategy is used to iteratively obtain body type sample data in the joint feasible region for the set of typical flood working conditions; accordingly, a flow state partition model and a hydraulic index agent model are iteratively constructed; the active sampling strategy integrates multi-condition uncertainty and flow state partition boundary information;

[0009] In the joint feasible region, a robust optimization model considering cross-condition flow state stability is constructed and solved by applying the flow state partition model and the hydraulic index agent model, to obtain a unified optimization body type scheme.

[0010] The present application improves the capture ability of nonlinear flow state mutation and optimization efficiency by introducing a flow state intermediate layer in the coupling system of the flood discharge vertical shaft energy dissipater-bend section and combining a flow state boundary driven sampling strategy. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 A step flow chart of a flood discharge vertical shaft energy dissipater-bend section body type unified optimization method provided by the embodiments of the present application.

[0012] Figure 2 A step flow chart for constructing a flow regime partition model is provided for the embodiments of the present application.

[0013] Figure 3 A step flow chart for obtaining a unified optimization body shape scheme is provided for the embodiments of the present application.

[0014] Figure 4 A step flow chart for constructing a cross-condition flow regime stability evaluation function is provided for the embodiments of the present application.

[0015] Figure 5 A first modification scheme diagram of a flood discharge shaft stilling well-bend section structure is provided for the embodiments of the present application.

[0016] Figure 6 A second modification scheme diagram of a flood discharge shaft stilling well-bend section structure is provided for the embodiments of the present application.

[0017] Wherein the reference signs are: 1, shaft spillway; 2, shaft center; 3, air hole; 4, upper edge elliptic curve; 5, pressure slope section outlet; 6, pressure slope section end; 7, lower edge elliptic curve; 8, stilling well section. DETAILED DESCRIPTION

[0018] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0019] It should be noted that the terms include and have as well as any variations thereof are intended to cover inclusive rather than exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a list of steps or units is not necessarily limited to those steps or units that are clearly listed, but can include other steps or units that are not clearly listed or inherent to such processes, methods, products, or apparatuses.

[0020] It is also found in the research that the existing method is to simply feed the body shape parameters of different samples to the model for training iteration without considering the flow or condition difference, to obtain a group of relatively optimal solutions from the mathematical point of view, and to ignore the engineering and hydrological background and hydraulic conditions such as flow and flow regime of the engineering samples themselves. The obtained results may not match the real flood scene of the engineering, that is, the adaptability is unknown. It may also appear that the body shape is numerically optimal, but it is actually only applicable to an extreme condition and does not match the real frequent flood, resulting in unreasonable applicability of the body shape.

[0021] As Figure 1 shown in the following formula (I), a method for optimizing the body shape of a flood discharge shaft, a stilling well and a turning section is proposed, including the following steps:

[0022] Obtain the multi-condition physical constraints describing the body shape parameters of the flood discharge shaft, the stilling well and the turning section, and construct a joint feasible region; at the same time, obtain a set of typical flood conditions.

[0023] In this embodiment, the body shape parameters refer to key dimensionless parameters for describing the geometric shape of the flood discharge shaft, the stilling well and the turning section, such as the relative depth ratio, the relative length ratio of the stilling well, the curvature coefficient (turning degree) of the turning section, etc. The multi-condition physical constraints refer to the basic physical limitations that the structure must satisfy under different flood flows, such as constraints derived based on the energy equation, the momentum equation, the cavitation safety condition or the submergence limitation condition. The set of typical flood conditions refers to several representative flood scenarios selected from the design flood process, such as the 100-year flood and the 1000-year flood checking condition, each of which corresponds to a set of hydraulic input parameters, such as the inflow and the downstream water level. Specifically, for each typical flood condition, a single-condition physical constraint inequality about the body shape parameters is derived based on the physical principles. By performing mathematical intersection operation on the inequality set of all conditions, the parameter subspace that meets the requirements of all conditions at the same time is obtained, i.e. the joint feasible region, denoted as Ω joint . The joint feasible region constitutes the hard boundary for subsequent optimization, so that any candidate body shape is physically feasible.

[0024] In some optional embodiments, the determination of the set of typical flood conditions can also be based on the flood frequency and the damage degree, and each representative condition is assigned a different weight, which can be used for subsequent robust optimization.

[0025] An active sampling strategy is adopted to iteratively obtain body shape sample data within the joint feasible region for the set of typical flood conditions, and a flow regime partition model and a hydraulic index proxy model are iteratively constructed based on the body shape sample data; the active sampling strategy combines the multi-condition uncertainty and the flow regime partition boundary information to obtain a convergent flow regime partition model and a hydraulic index proxy model.

[0026] In the embodiment, the flow regime partition model is different from the traditional hydraulic index, i.e., the dimensionless number Fr (Froude number) generally distinguishes the rapid flow and the slow flow; Re (Reynolds number) generally distinguishes the laminar flow and the turbulent flow, but is to divide the continuous flow field characteristics, such as vortex, negative pressure zone, backflow zone, water jump, water drop, clear full flow alternation and other morphological characteristics, into discrete flow regime categories with clear engineering significance, such as stable flow regime, sub-stable flow regime or unstable flow regime, through clustering and other means. The hydraulic index proxy model is a mathematical model for predicting specific performance indicators, including energy dissipation rate, impact pressure, turbulence intensity and the like. The body shape sample data includes body shape parameters, working conditions, corresponding flow regime categories and hydraulic indexes. The embodiment discards the traditional body shape-index single-layer proxy model, and constructs a body shape-flow regime-index three-layer mapping architecture. Specifically, the flow regime partition model maps the body shape parameters and working conditions into the flow regime category, i.e., R = f(X, C); and the hydraulic index proxy model takes the body shape parameters, working conditions and the flow regime category R determined by the flow regime partition model as inputs to predict the hydraulic index I = g(X, C, R). The flow regime category R as an intermediate physical layer can more accurately reflect the nonlinear mutation of flow, thereby improving the prediction accuracy of the hydraulic index proxy model g. To efficiently construct the above model, the active sampling strategy is used to guide the newly added numerical calculation samples, which can avoid wasting computing resources in the low information area. Exemplarily, the active sampling strategy evaluates the value of the candidate body shape x through the fused sampling value function V(x). V(x) fuses information in multiple dimensions, especially the multi-condition uncertainty and the flow regime partition boundary information, i.e., the area where the hydraulic index proxy model g is not accurate in prediction and the area where the flow regime partition model f is difficult to distinguish the flow regime, i.e., the flow regime mutation zone. The iteration process is as follows: based on a small amount of initial body shape sample data, the initial models of f and g are obtained. The value function V(x) is used to select the body shape point with the highest value in the joint feasible region, and high-precision numerical calculation is performed on it to obtain new body shape sample data. The new data is incorporated into the data set, and the flow regime partition model f and the hydraulic index proxy model g are retrained. Repeat this process until the model accuracy reaches the preset convergence standard, and finally obtain the converged flow regime partition model and the converged hydraulic index proxy model.

[0027] In the joint feasible region, the flow regime partition model and the hydraulic index proxy model are applied to construct and solve a robust optimization model considering the flow regime stability across working conditions, to obtain a unified optimization body shape scheme.

[0028] In this embodiment, the robust optimization model refers to the optimization model not only pursues the optimal average performance, but also pursues the stable performance and safety under all working conditions, especially the most unfavorable working conditions. The converged flow regime partition model f and the hydraulic index proxy model g are used as fast evaluation tools to replace time-consuming high-precision calculations. At the same time, the cross-condition flow regime stability is taken as the core constraint or target of optimization. Specifically, using the converged flow regime partition model f, it can be evaluated whether all typical working conditions c j of any body X can maintain in the stable flow regime category. For example, a constraint S(X) ≥ S min can be constructed, where S(X) is the flow regime stability index calculated based on the classification confidence predicted by the model f. The optimization solving process is: in the joint feasible region Ω joint , taking the body parameter X as the decision variable, taking the multi-condition weighted fitness and the performance of the most unfavorable working condition predicted by the hydraulic index model g as the optimization target, and taking the cross-condition flow regime stability predicted by the flow regime partition model f as the constraint, a multi-objective optimization algorithm is used for solving. The final result of the solution is one or more body parameter combinations that perform well in performance and flow regime stability, that is, the unified optimization body scheme. This scheme can be further supplemented by high-fidelity verification and engineering checking, and the specific structural geometric dimensions are calculated, as the final output.

[0029] In one specific embodiment, several body shape parameters of the stilling well and the turning section are defined as subsequent optimization objects. The design parameters that have greater influence on the energy dissipation, flow regulation and flow state transition effect of the overall structure of the stilling well and the turning section are selected as the body shape optimization parameters. Specifically, the stilling well is a sink-shaped energy dissipator structure with the bottom plate of the vertical well protruding downward, including an upstream wall, a downstream wall, two side walls and a bottom plate. Before optimization, the cross section of the stilling well is circular with the same diameter as the upper vertical well. The main body shape parameters include: the depth h of the stilling well, which is generally 1.5D~2.5D, D being the diameter of the circular cross section of the flood discharge vertical well (the vertical well section parameters belong to the known engineering data), the depth of the stilling well being calculated from the top of the lower wall of the turning section to the bottom plate of the stilling well; further, in order to facilitate comparison of different body shapes and parameter popularization, the dimensionless ratio of the depth of the stilling well to the diameter of the vertical well (depth-diameter ratio d = h / D) is used to represent the required depth of the stilling well, d being between 1.5 and 3. The upstream side width d1 of the cross section of the stilling well, since during flood discharge, the flow passes through the stilling well and the turning section to the downstream side, the upstream side wall is relatively small under water pressure, and considering energy dissipation requirements and engineering excavation amount, the upstream side body shape of the stilling pool can be consistent with the upper vertical well section, i.e. a semicircular shape with a radius equal to the vertical well radius R', at this time d1 = R'. The downstream side width d2 of the cross section of the stilling well, according to the relationship between d2 and the diameter D of the vertical well, three types of body shapes can be defined as follows: when d2 = D, i.e. S = 1, it is defined as a city gate type cross section; when d2 > D, i.e. S > 1, it is defined as a hole type cross section; when d2 < D, i.e. S < 1, it is defined as a horseshoe type cross section. The downstream side length l of the center of the stilling well, from the center of the vertical well to the most downstream side endpoint of the stilling well; the total length of the stilling well = R' + l. Accordingly, the dimensionless number of the cross section extension coefficient of the stilling well, i.e. the expansion degree to the downstream, is defined as T = l / R', T being generally between 1.2 and 2.0. It should be noted that, in the absence of necessity, such as water pressure and other hydraulic indicators being within a reasonable range, the stilling well will not be expanded to the upstream to increase the water body; if the water body needs to be expanded, the stilling well is usually expanded to the downstream side to increase the water body.

[0030] The turning section includes an upper wall surface connection curve, a pressure slope section, an upstream side connecting shaft and a lower stilling well, and a downstream side connecting a drainage hole. Main body size parameters include: a turning section upper wall surface transverse radius a1, a vertical radius b1, a turning angle a1; it should be noted that the starting point of the turning section upper wall surface is the bottom end point of the shaft section, and the end point of the turning section upper wall surface is the end point of the pressure slope section top plate; a turning section lower wall surface transverse radius a2, a vertical radius b2, a turning angle a2; it should be noted that the starting point of the turning section lower wall surface is the top end point of the stilling well, and the end point of the turning section lower wall surface is the end point of the pressure slope section bottom plate. The pressure slope section: the pressure slope top plate length L1, the pressure slope slope ratio i, and the bottom plate length L2 is the same as the horizontal projection length of the top plate. Further, the turning ratio R* of the turning section upper and lower wall surfaces is defined: R*=vertical radius(b) / transverse radius(a). The closer this ratio is to 1, the closer the turning section curve is to a quarter of a circle. The farther the ratio deviates from 1, the closer the arc is to a quarter of an ellipse, whether greater than or less than 1; the farther the ratio is from 1, the higher the degree of flatness or slenderness of the ellipse in the horizontal direction. When R*=1, the turning section is a quarter of a circle, and the vertical radius is equal to the transverse radius; when R*<1, the turning section is a flat elliptical arc, and the vertical radius is less than the transverse radius; when R*>1, the turning section is a high and thin elliptical arc, and the vertical radius is greater than the transverse radius. Preferably, the turning angle a is 90°, that is, the starting and ending points of the turning section are tangent to the shaft wall surface in the vertical direction and the pressure slope or bottom plate in the horizontal direction.

[0031] In one possible embodiment, a joint feasible region is constructed, comprising:

[0032] For each of a set of typical flood conditions, single-condition physical constraint inequalities related to the body size parameters are derived.

[0033] In other words, the body size parameters X are defined as a dimensionless geometric parameter vector; for each of a set of typical flood conditions C={c1, …, c m}c j , single-condition physical constraint inequalities describing the physical limitations of the condition are derived, and the set of single-condition physical constraint inequalities defines a single-condition physical feasible region Ω j corresponding to the condition.

[0034] In one preferred implementation, the single-condition physical constraint inequalities are derived based on at least two of the energy equation, the momentum equation, the cavitation safety condition, and the submergence limitation condition.

[0035] In other words, deriving single-condition physical constraint inequalities comprises: based on the energy equation, the momentum equation, or the cavitation safety condition, establishing body size parameters X in the condition c jThe hydraulic response calculation function is used; the standard control parameters, including the allowable cavitation index and the allowable maximum impact pressure, are read; by comparing the hydraulic response calculation function with the standard control parameters, a single-condition physical constraint inequality is established, which includes at least: σ(X, c j )≥σ allowed ;P(X,c j )≤P max ;where σ(X, c j ) is based on X and c j The calculated cavitation number, σ allowed To standardize the allowable cavitation index in the control parameters; P(X, c j ) is based on X and c j The calculated impact pressure, P max To standardize the maximum allowable impact pressure in the control parameters.

[0036] In this embodiment, it is necessary to organize the basic engineering data and standard control parameters. The basic engineering data includes topographic data of the dam site and downstream river channel, the initial layout of the spillway shafts, and structural reference dimensions; among which, the structural reference dimensions include the shaft diameter D. Standard control parameters define the design red lines, i.e., the boundary values ​​allowed by design specifications or engineering requirements, such as the maximum impact pressure P allowed by the specifications. max Or the allowable cavitation number σ required by engineering specifications. allowed The typical flood case set C = {c1, ..., c2} m} represents scenarios selected from design flood process data. For example, c1 could be a 100-year design scenario, and c2 could be a 1000-year check scenario. Each scenario c j This corresponds to a specific set of hydraulic parameters, such as the inflow rate Q under this operating condition. j and downstream control water level Z j The body size parameters are defined as a dimensionless geometric parameter vector X = [x1, x2, ..., x...]. n The term x1 is used to uniformly describe the shape of the energy dissipation well and the turning section. For example, x1 can be the depth-to-diameter ratio of the energy dissipation well, x2 can be the curvature coefficient of the turning section, and x3 can be the relative length of the bottom of the well.

[0037] The derivation process is for each working condition c. j This embodiment establishes a functional relationship between fundamental hydraulic principles and the parameter vector X. It abandons the traditional experience-based design approach and instead uses physical equations to define feasibility. For example, regarding cavitation safety conditions, a calculation function σ can be established based on the energy equation. calc (X, c) j ), used to calculate the size X and working condition c j The actual cavitation number under the given conditions. This constraint is then expressed as σ.calc (X, c) j )≥σ allowed For example, regarding the impact pressure limitation, a calculation function P can be established based on the momentum equation. calc (X, c) j This constraint is used to calculate the maximum impact pressure on the base plate. It is represented by P. calc (X, c) j )≤P max Similarly, for the flooding constraint, a function H can be established. calc (X, c) j )≥H min This is to prevent unwanted flow patterns from occurring during the turning section. Summarizing all the above inequalities forms the operating condition c. j The set of single-condition physical constraint inequalities, which defines the single-condition physical feasible region Ω in the n-dimensional parameter space. j .

[0038] In some alternative implementations, the above calculation function σ calc and P calc It can be obtained through simplified hydraulic formulas, or it can be quickly evaluated through high-precision three-dimensional numerical simulation (CFD) or a trained surrogate model.

[0039] Perform intersection operations on the single-condition physical constraint inequalities for all operating conditions to form a joint feasible region.

[0040] In other words, the joint feasible region Ω is determined by performing an intersection operation on all single-case physical feasible regions. joint =∩ j=1 m Ω j Where j is the operating condition index, m is the total number of typical flood operating conditions, and Ω j Let Ω be the single-condition physical feasible region corresponding to the j-th working condition. joint A joint feasible region that satisfies all operating condition constraints.

[0041] In this embodiment, multiple single-condition physical feasible domains Ω are considered. j Perform a mathematical intersection operation on (j=1, 2, ..., m). The resulting joint feasible region Ω joint It is a subset in the shape parameter space X that simultaneously satisfies the physical constraints of all m typical flood conditions. Its mathematical expression can be written as: Ω joint =∩ j=1 m Ω j. So that any subsequent sampling or optimization within this feasible region will have candidates that are physically safe and feasible. This reduces the search space and improves the efficiency and reliability of subsequent data-driven optimization. In some embodiments, if the constraints of different operating conditions conflict, resulting in Ω joint being empty, a feedback step can also be included, i.e. adjusting the specification control parameters or engineering base data, and re-executing the present embodiment until a non-empty joint feasible region is obtained.

[0042] In an optional embodiment, before iteratively obtaining the body shape sample data, an initial multi-operating condition flow regime and hydraulic index dataset is constructed, specifically:

[0043] An initial body shape parameter sample set is generated within the joint feasible region.

[0044] In the present embodiment, within the high-dimensional (n-dimensional) and possibly irregularly shaped joint feasible region Ω joint , a set of sample points with good spatial filling and representativeness is selected. Preferably, the Latin Hypercube Sampling (LHS) method can be used, so that the projection on each dimension of the body shape parameter X is uniformly distributed, with high sampling efficiency. Alternatively, in other embodiments, methods such as uniform design, orthogonal design, or low-dispersion sequence sampling based on Hammersley sequence can also be used. It should be noted that since the joint feasible region Ω joint is defined by the intersection of inequalities, it can have a complex shape. Therefore, the sampling process can be combined with constraint handling techniques, for example, sampling within a hyperrectangle that contains the joint feasible region Ω joint , and removing sample points that do not satisfy all constraints, i.e. removing sample points that are not within the joint feasible region Ω joint . The initial body shape parameter sample set {X1, X2,..., XN} is output, where N is the number of initial samples, for example N = 50 or 100. N

[0045] For the initial body shape parameter sample set, high-precision three-dimensional numerical calculations are performed under a set of typical flood operating conditions.

[0046] Specifically, input the sample set {X i} and the operating condition set {c j}, perform a full combination calculation process, i.e. for each body shape parameter X i , perform a high-precision three-dimensional numerical calculation under each typical operating condition c j . Preferably, this high-precision calculation uses computational fluid dynamics (CFD) simulation. For example, use the k-ε turbulence model based on the Reynolds time average (RANS) method, and combine the VOF (volume of fluid) method to capture the shape of the free water surface. The boundary conditions of the calculation model are determined according to the operating condition c​j Set, for example, the inlet is set to flow Q j , the outlet is set to water level Z j Output N x m high-fidelity flow field raw data sets, where N is the number of samples and m is the number of working conditions; Each data set contains detailed velocity field, pressure field and free water surface position field data.

[0047] Extract multi-condition flow pattern raw data and hydraulic index raw data to construct an initial multi-condition flow pattern and hydraulic index data set.

[0048] In this embodiment, N x m flow field raw data sets are input. Unstructured flow field data is converted into structured features and labels that can be used for subsequent machine learning. Specifically, hydraulic index raw data, which is the training label for the subsequent hydraulic index proxy model g(), is calculated. For example, for each flow field, its total energy dissipation rate η, maximum impact pressure P max_calc on the bottom plate, outflow velocity unevenness coefficient, etc. are calculated. Multi-condition flow pattern raw data, which is the training feature for the subsequent flow pattern partition model f(), is extracted. This process includes calculating vortex intensity, negative pressure zone volume, backflow zone geometric boundary, water jump shape parameters, water drop position, and whether clear full flow appears, etc. The body shape parameter X i , working condition c j , corresponding flow pattern raw data and hydraulic index raw data are associated to form an initial multi-condition flow pattern and hydraulic index data set.

[0049] In an exemplary embodiment, the flow pattern partition model and the hydraulic index proxy model are iteratively constructed, including:

[0050] A flow pattern partition model R = f(X, C) is constructed for mapping body shape parameters and working conditions to discrete flow pattern categories; Where X represents the input body shape parameters; C represents the working condition; f() represents the flow pattern partition model; R is a discrete flow pattern category.

[0051] As shown in Figure 2 , in a preferred implementation, the flow pattern partition model is constructed, including:

[0052] Flow pattern features representing flow patterns are extracted from the multi-condition flow field data contained in the body shape sample data to form a flow pattern feature data set.

[0053] In this embodiment, physical features that can quantitatively describe flow patterns, i.e. flow pattern features, are extracted from N x m high-fidelity flow field raw data as the basis for subsequent flow pattern partitioning; Where high-fidelity flow field raw data includes velocity field, pressure field, free water surface position field, etc. Preferably, flow pattern features include but are not limited to the following: one, negative pressure area feature, used to represent the risk of cavitation and potential instability of flow. Specifically, a pressure threshold can be set, for example, the absolute pressure Pthreshold = 10000 Pa, traverse all grid cells in the computational domain, and count the total volume V threshold of cells with pressure lower than P neg , V neg is the flow regime feature. Two, vortex intensity feature, used to characterize the vortex intensity in the shaft or the vortex structure in the stilling well. Specifically, the spatial average or maximum of vorticity ω or the statistics of streamline curvature can be calculated on the key section of the shaft or stilling well as the vortex intensity feature. Three, ring-shaped hydraulic jump morphology feature, used to characterize the energy dissipation morphology in the stilling well. Preferably, the free water surface position Y(x) at α = 0.5 determined by the VOF method is used to automatically locate the start and end points of the hydraulic jump by analyzing the gradient ΨY / Ψx or curvature of the water surface line Y(x), so as to calculate the relative length L jump or relative thickness H jump of the hydraulic jump as the flow regime feature, where Ψ is the partial derivative and x is the horizontal coordinate along the main flow direction. Four, impact point drift feature, used to characterize the stability of the main flow impact point. Specifically, the spatial coordinates (x, z) of the pressure maximum point (impact point) on the stilling well floor are located, and the spatial standard deviation (drift range) in time sequence is counted as the characterization feature of flow regime stability. The above calculated multiple flow regime features (V neg , max(|ω|), L jump , spatial standard deviation) are combined into an n-dimensional flow regime feature vector V f . This operation is performed for each body type-case combination (X i , c j ), and the dimensions of V f are normalized to form the flow regime feature dataset.

[0054] Perform clustering analysis on the flow regime feature dataset to define the flow regime categories and generate flow regime category labels for the body type samples; use the body type parameters and working conditions in the body type sample data as input and the flow regime category labels as output to train the flow regime partition model.

[0055] In this embodiment, the flow regime is automatically divided into several categories with clear engineering significance by unsupervised learning using the flow regime feature dataset, and a supervised classification model f is trained based on this. Specifically, unsupervised clustering analysis is performed. Preferably, the K-means clustering algorithm is used. The input of the clustering algorithm is the normalized flow regime feature vector V f, the distance metric preferably adopts Euclidean distance. Preferably, the elbow method or silhouette score can be used to evaluate the clustering effect under different clustering numbers K. For example, when the silhouette score is the highest when K = 3, it is determined that the flow regime is divided into 3 categories. The clustering results are reviewed and named by engineering. Those skilled in the art can analyze the K = 3 clustering clusters in combination with the visualization results of the flow field, and name them as flow regime category labels R with engineering significance. For example: R1 = stable flow regime, corresponding to low V neg , stable L jump , R2 = metastable flow regime, R3 = unstable flow regime, corresponding to high V neg or water jump form fragmentation. Train the flow regime partition model f, which is a supervised classification task, the input of the model is the body parameter X and the working condition C, and the output is the flow regime category label R. Preferably, support vector machine (SVM), random forest or deep neural network can be used as the classifier. After training, the flow regime partition model R = f(X, C) is obtained. The model f can predict the flow regime category R to which any unknown body- working condition combination belongs. In some preferred embodiments, the flow regime partition model f not only outputs the predicted category R k , but also outputs the classification confidence of the prediction, that is, p(R = R k | X, C).

[0056] A hydraulic index proxy model I = g(X, C, R) is constructed for mapping the body parameters, working conditions, and flow regime categories determined by the flow regime partition model together into the hydraulic index; wherein g() is the hydraulic index proxy model; I is the hydraulic index.

[0057] In this embodiment, a proxy model g capable of predicting the hydraulic index I is constructed; wherein the hydraulic index I includes the energy dissipation rate η and the impact pressure P. The input of the model g is not only the body parameter X and the working condition C, but also the flow regime category R as an independent input feature. In other words, the mathematical form of the model is: I = g(X, C, R). Since the response of the hydraulic index I is different under different flow regimes R, the physical law is different, that is, the function g has different function forms under different R. The single-layer model of the traditional I = g(X, C) attempts to fit this highly nonlinear and abrupt physical process with a continuous function, so the accuracy is limited. This embodiment introduces an intermediate layer R to decompose the complex fitting task into classification f first and then segmented regression g, which is more reasonable in physics and has higher prediction accuracy. Specifically, the proxy model g preferably adopts multi-output Gaussian process regression (GPR) or radial basis function (RBF) network. The input of the training data set is (X i , c j , R i ), and the output is (I1, I2,..., In).m ), where R i This is the sample (X) i c j ) Determined flow state category label, I m Let m represent the m-th hydraulic index. Ultimately, two models are obtained: the flow regime partitioning model R = f(X, C) and the hydraulic index surrogate model I = g(X, C, R).

[0058] In an exemplary embodiment, hydraulic evaluation indicators, i.e., hydraulic indices, are defined for stilling wells and bend sections. Specifically, the following five main evaluation indicators are defined to evaluate the rationality of the shape of stilling wells and bend sections from the perspectives of energy dissipation effect, rectification effect, and drainage effect, including: the overall energy dissipation rate of the stilling well and bend section η = (E1-E2) / E1; where E1 is the total head of the reservoir area in meters; E2 is the total head of the outlet section of the bend section in meters; the energy dissipation rate is preferably at least greater than 35%, which is considered to be basically reasonable; greater than 50% is preferable. The maximum impact pressure P of the stilling well bottom plate: P = (F max -F min ) / F avg ;where F max F min F avg These represent the maximum, minimum, and average hydrostatic pressure on the bottom plate, respectively. Preferably, the impact pressure on the bottom plate should be less than 10m pressure head. The coefficient of non-uniformity of pulsating pressure distribution on the walls of the stilling well and turning sections, δ, is preferably less than 0.5 to prevent stress concentration along the path. The calculation formula is: μ Cp =1 / n ∑ i=1 n C pi ;σ Cp =sqrt(1 / n ∑ i=1 n (C pi -μ Cp ) 2 ); δ=σ Cp / μ Cp Where n represents the total number of measuring points; C pi The root mean square value of the pulsating pressure at each measuring point is expressed in units of ×9.81 kPa; μ Cp The average pulsating pressure represents the average level of pulsation intensity across the entire wall surface; σ Cp This reflects the degree of dispersion between the pulsation intensity value at each measuring point and the average level. σ CpThe greater, the greater the difference in pulse intensity of each point, the more uneven the distribution; the uneven coefficient δ is the ratio of the standard deviation to the average value, a dimensionless number that can more accurately reflect the relative dispersion degree of the data. The relative height Hr of the annular hydraulic jump in the shaft, the water body passing through the stilling well and the turning section needs to be able to raise the water level in the shaft to a certain submerged water depth, that is, the water body needs to be able to be blocked to more than the turning section, but it does not affect the discharge capacity, and it cannot affect the discharge capacity of the shaft. Preferably, the water level is above the top of the upper wall of the turning section, but it is appropriate to affect the discharge capacity of the funnel section of the upper shaft. The value range of the annular hydraulic jump height Hc is generally greater than 2 times the depth of the stilling well h and less than the difference between the height of the shaft circular pipe section and the depth of the stilling well, and the calculation formula is: 2h < Hc < H / 2; wherein H is the total fall of the shaft, and the unit is m, from the shaft inlet to the bottom plate of the stilling well; h is the depth of the stilling well, and the unit is m. Preferably, H > 7h ~ 9h, and Hr = Hc / h, then the value range of the relative height of the annular hydraulic jump: 2 < Hr < 3.5. The outflow uneven coefficient U of the turning section end: the ratio of the maximum flow velocity V max of the turning section end outlet section to the average flow velocity V avg , can characterize the uniformity of the flow velocity distribution of the section, and reveal whether the water flow is uniform, stable, and whether it is easy to appear flow state variation, such as water wing and other adverse phenomena. Preferably, the flow velocity distribution at the entrance of the water-retreating tunnel is required to meet: the flow velocity uneven coefficient is not more than 2.

[0059] According to one aspect of the present application, since the initial sampling is blind, and the resources of high-precision three-dimensional numerical calculation are very valuable. Therefore, it is necessary to concentrate the limited calculation resources into the parameter area most valuable to model convergence. Specifically, an active sampling strategy is adopted to iteratively obtain body type sample data, including:

[0060] Generating candidate body type parameter points that have not been calculated in the joint feasible region.

[0061] Specifically, in the joint feasible region Ω joint , a set of candidate body type parameter points {x} is generated, which is much larger in number than the initial sample, for example, the number of candidate points N candidate = 10000. Using the currently converged f and g models, the 10000 candidate points x are predicted under all m typical working conditions c j . For each (x, c j ) combination, the following is obtained: the predicted flow state category R j = f(x, c j ) and its classification confidence p j = p(R = R j | x, c j ); the predicted hydraulic index I j = g(x, cj , R j ).

[0062] constructing an active sampling merit function to evaluate the candidate body parameter point.

[0063] In other words, constructing an active sampling merit function to evaluate the candidate body parameter point.

[0064] In a preferred implementation, constructing an active sampling merit function comprises:

[0065] evaluating the candidate body parameter point's multi-condition comprehensive fitness prediction uncertainty using the hydraulic index proxy model; evaluating the candidate body parameter point's flow regime partition boundary distance using the flow regime partition model; and evaluating the candidate body parameter point's cross-condition flow regime category prediction consistency in the set of typical flood conditions using the flow regime partition model.

[0066] fusing the multi-condition comprehensive fitness prediction uncertainty, the flow regime partition boundary distance, and the cross-condition flow regime category prediction consistency to determine the active sampling merit function V(x) = w1·U(x) + w2·D(x) + w3·C(x); wherein V(x) is the active sampling merit score of the candidate body parameter point x; U(x) is the multi-condition comprehensive fitness prediction uncertainty; D(x) is the flow regime partition boundary distance; C(x) is the cross-condition flow regime category prediction consistency; w1, w2, w3 are preset weight coefficients for fusing the three dimensions.

[0067] Specifically, a comprehensive active sampling merit score V(x) is calculated for each candidate body x. The higher V(x) is, the greater the value of conducting CFD calculation at the point x for improving the model accuracy. V(x) is preferably constructed as a weighted fusion of three sub-items: V(x) = w1·U(x) + w2·D(x) + w3·C(x); wherein w1, w2, w3 are preset weight coefficients for balancing the preferences of different sampling strategies. The multi-condition comprehensive fitness prediction uncertainty U(x) is used to quantify the model's g's uncertainty in its prediction results. If the model g adopts Gaussian Process Regression (GPR), U(x) can be preferably defined as the posterior variance var(I(x)) predicted by the model g at the point x, where I(x) is the predicted value of the hydraulic index at the point x. If other models are adopted, U(x) can be estimated by calculating the standard deviation of multiple model prediction results through Ensemble or Bootstrap methods. The area with high U(x) is the knowledge blind area of the model g.

[0068] The flow regime partition boundary distance D(x) quantifies the degree of uncertainty of the flow regime partition model f, i.e. how close the candidate point x is to the flow regime transition boundary. The model f is most difficult to predict and most likely to make mistakes near the partition boundary, which is the area that needs to be strengthened the most. In an exemplary embodiment, the flow regime partition boundary distance is evaluated by: predicting the flow regime class and the corresponding classification confidence of the candidate body parameter point x in each of the set of typical flood operating conditions using the flow regime partition model f; comparing the classification confidence with a pre-defined flow regime boundary decision threshold to quantify the closeness of the candidate body parameter point x to the flow regime partition boundary; taking the closeness of the flow regime partition boundary as the flow regime partition boundary distance D(x) = max j∈{1，2，…，m} {D j (x)}; D j (x) = 1 - max k {p(R k | x, c j )}; where D(x) is the flow regime partition boundary distance of the candidate body x; D j (x) is the flow regime boundary closeness of the candidate body x in a single operating condition c j ; j is the operating condition index; m is the total number of typical flood operating conditions; p(R k | x, c j ) is the classification confidence of x belonging to class k in c j ; max k {} means taking the maximum classification confidence among all flow regime classes k; max k {p k} is the confidence that the classifier is most likely to predict. 1 - max k {p k}, i.e. D j (x) is the uncertainty or entropy of the classifier, when this value is high, it means that the classifier cannot clearly distinguish which class x belongs to, i.e. x is near the flow regime partition boundary. max j {} means taking the value closest to the flow regime boundary among all operating conditions. Preferably, the calculation of D(x) is based on the classification confidence p j . For a single operating condition c j , its boundary distance D j (x) can be defined as: D j (x) = 1 - max k {p(R k | x, c j )}; where k goes through all flow regime classes. D j (x) means: if the model f is very confident, e.g. p = 0.99, then D j = 0.01, far from the boundary; if the model f is extremely confused, e.g. p = 0.5 between two classes, then Dj = 0.5, at the boundary. For multiple operating conditions, D(x) preferably takes the most unfavorable case among all operating conditions, i.e. D(x) = max j {D j (x)}.

[0069] Some body x may behave as R1 under operating condition c1, i.e. indicating stability, but jump to R3 under operating condition c2, i.e. indicating instability. Such body x with inconsistent flow regime across operating conditions is a key obstacle for robust optimization, and is also a region that needs to be highlighted by the model. The cross-operating condition flow regime category prediction consistency C(x) can be defined as a score, for example, the total number of categories that the body x is predicted to be under m operating conditions, the more categories, the higher the score of C(x). j

[0070] According to the evaluation results, select new body sample, iteratively obtain body sample data.

[0071] It can also be said that according to the active sampling value function, select new body sample, iteratively obtain body sample data.

[0072] In this embodiment, according to the V(x) score, sort the 10000 candidate points, and select the K body parameter points with the highest V(x) as the new body samples of this round. For the K new body samples, perform high-precision three-dimensional numerical calculation under all m operating conditions, and extract the flow regime features and hydraulic indicators to form an incremental data set. Further, incorporate the incremental data set into the initial multi-operating condition flow regime and hydraulic indicator data set to form an updated data set with larger scale and richer information. On this basis, repeat the steps including feature extraction, clustering analysis, training of the model f and g to obtain the updated flow regime partition model f update and the hydraulic indicator proxy model g update . Perform convergence determination. For example, evaluate the accuracy of f update and g update on an independent validation set. If the accuracy improvement compared with the last round is less than a preset threshold, for example 1%, or the total iteration rounds reach the upper limit, it is considered that the model has converged. If not converged, take f update and g update as the new current model, and return to start a new round of active sampling. If converged, the iteration process is terminated. The final f update and g update are the converged flow regime partition model and the converged hydraulic indicator proxy model.

[0073] As shown in FIG. 1, in one embodiment of the present application, a unified optimization body scheme is obtained, including: Figure 3

[0074] ​​Using a hydraulic index proxy model, a multi-condition weighted fitness function and a worst-case performance function are constructed.

[0075] In this embodiment, a convergent surrogate model is used to construct the various components of the robust optimization mathematical model. Specifically, the objective function for optimization is constructed using the convergent hydraulic index surrogate model I = g(X, C, R) and R provided by the f model. Preferably, two objective functions are constructed: a multi-condition weighted fitness function J. weighted (X) is used to measure the average performance level of body type X under all m typical operating conditions. For example, the hydraulic indicators I such as energy dissipation rate and impact pressure predicted by g are normalized and weighted, and then weighted and averaged according to the operating condition weights to obtain J. weighted (X). Performance function J under the most unfavorable operating condition. worst (X) is used to measure the worst performance of body size X across all m operating conditions, ensuring the robustness of the optimization. For example, J worst (X) = max j {P calc (X, c) j )}, that is, take the maximum value of the predicted impact pressure among all working conditions.

[0076] A cross-condition flow stability evaluation function is constructed using a flow regime partitioning model.

[0077] like Figure 4 As shown, in a preferred implementation, a cross-condition flow stability evaluation function is constructed, including:

[0078] By applying the flow regime partitioning model, the flow regime category and classification confidence of any body shape parameter under each condition in the typical flood condition set are predicted;

[0079] Based on the degree of consistency of the predicted flow regime category across different operating conditions, or based on the minimum classification confidence of the predicted stable flow regime, the cross-operating condition flow regime stability evaluation function S(X) = min j∈{1，2，…，m} {p(R=R stable |X, c j )}; where j is the working condition index; m is the total number of typical flood working conditions; c j For the j-th typical flood condition; R stable Let R be the target stable flow regime category; R be the discrete flow regime category; p(| ) be the classification confidence score; p(R=R stable |X, c j () indicates that body size X is under working condition c j The classification confidence level of the flow that is predicted to be a stable flow state.

[0080] Specifically, using the convergent flow regime partitioning model R = f(X, C), the key constraints for optimization are constructed, namely, the cross-condition flow regime stability evaluation function S(X). S(X) is preferably defined as: the body shape X in all m typical flood conditions c j The following is predicted by model f to be a steady flow state (denoted as R). stable The minimum classification confidence score of X is calculated as follows: S(X) = min j∈{1，2，…，m} {p(R=R stable |X, c j )};where X is the body shape parameter vector, c j Let be the j-th typical working condition, and p(| ) be the classification confidence score output by the flow regime partitioning model f. The cross-working condition flow regime stability evaluation function S(X) represents the degree to which the body shape X is closest to the flow regime change boundary among all the working conditions it faces. By constraining S(X) in the optimization, the final body shape scheme can be made to have sufficient safety margin under all working conditions and always maintain a stable flow pattern.

[0081] Using body shape parameters as decision variables, within the joint feasible region, the optimization objective is to minimize the weighted fitness function under multiple operating conditions and the performance function under the worst operating condition. X∈Ωjoint {J weighted (X), J worst (X)}, and with the cross-condition flow stability evaluation function as the constraint condition S(X)≥S min The solution yields a unified optimized body shape scheme; where X is the optimization decision variable, i.e., the body shape parameter; Ω joint The joint feasible region; min indicates taking the minimum value; J weighted (X) is the multi-condition weighted fitness function; J worst S(X) is the performance function under the most unfavorable operating condition; S(X) is the flow stability evaluation function across operating conditions; S min This is the preset minimum allowable flow stability threshold.

[0082] In this embodiment, the constructed functions are assembled to form a complete multi-objective constrained optimization model. The preferred mathematical form of this robust optimization model is: min X∈Ωjoint {J weighted (X), J worst (X)},st S(X)≥S min Where min indicates that the optimization objective is to minimize J weighted and J worst The smaller the value, the better; X is the decision variable, i.e., the body shape parameter vector; Ω joint It represents the joint feasible region, serving as the boundary constraint for the decision variable; st represents the constraint condition; S min It is a preset minimum allowable flow stability threshold, such as S.min = 0.75, which means the confidence of the body shape being predicted as stable flow regime under any operating condition should not be lower than 75%. The mathematical model is solved by a multi-objective optimization algorithm. At each step of the optimization iteration, the algorithm generates a new candidate body shape X, and calculates its J weighted , J worst and S(X) by calling the models f and g to evaluate its fitness and feasibility. The final output of the optimization is the Pareto front solution set of the multi-objective problem, i.e. the multi-condition optimization body shape candidate set. Each body shape in the set is the preferred solution that reaches a non-dominated balance between the average performance J weighted and the worst performance J worst under the premise of meeting the flow regime stability constraint S(X).

[0083] In further embodiments, further comprising:

[0084] Performing high-precision three-dimensional numerical calculation on the multi-condition optimization body shape candidate set obtained by solving the robust optimization model to obtain a multi-condition optimization body shape verification hydraulic index data set.

[0085] In the present embodiment, the input is a multi-condition optimization body shape candidate set, for example, 20 preferred body shapes on the Pareto front. Although the surrogate models f and g have converged, there may still be prediction bias, and the final candidate solution needs to be physically verified by high-precision calculation. Specifically, representative body shapes are selected from the candidate set, for example, 5 are uniformly selected, and high-precision three-dimensional numerical calculation (CFD) is performed again on the 5 body shapes under all m typical flood operating conditions. The output is a physically real multi-condition optimization body shape verification flow field data set and a multi-condition optimization body shape verification hydraulic index data set for the 5 candidate body shapes.

[0086] Comparing the multi-condition optimization body shape verification hydraulic index data set with the surrogate model prediction results of the hydraulic index surrogate model; when the verification data and the prediction results have large deviation, the multi-condition optimization body shape verification hydraulic index data set is incorporated into the multi-condition flow regime and hydraulic index data set for correcting the surrogate model.

[0087] In this embodiment, the high-fidelity validation results are cross-verified with the predicted results of the surrogate model. Specifically, the validation hydrodynamic index dataset (CFD true value) is compared with the predicted results of the surrogate model predicted by the model g. If the deviation between the two is within an acceptable engineering precision range, for example, 5%, it is proved that the surrogate models f and g are accurate in the key solution space region, and the next step of screening can be entered. If a larger deviation is found, for example, more than 15%, the model correction mechanism is triggered. The specific operation is: the validation flow field dataset and the validation hydrodynamic index dataset are merged into the updated dataset, which is equivalent to forced and high-value active sampling for the optimal solution region. The f and g are iteratively trained for an additional round of training using the expanded dataset to correct the deviation of the model near the optimal solution. After the model is corrected, the optimization solving can be re-executed.

[0088] Further, it also includes: using the multi-condition optimization body shape validation hydrodynamic index dataset to correct the performance of each candidate body shape in the multi-condition optimization body shape candidate set; for each candidate body shape in the multi-condition optimization body shape candidate set, calculating a multi-condition comprehensive fitness score and extracting its most unfavorable condition performance and flow state stability degree; based on the multi-condition comprehensive fitness score, sorting and combining with the engineering safety margin requirement to screen the unified optimization body shape scheme from the multi-condition optimization body shape candidate set.

[0089] Specifically, input the reliable multi-condition optimization body shape candidate set and its corresponding high-fidelity or high-precision predicted performance data after verification or correction and re-optimization. For each candidate body shape, calculate its multi-condition comprehensive fitness score. The score can be a comprehensive index, for example, based on the condition weight, the J weighted (X), J worst (X) and S(X) are weighted and summed or multi-attribute decision sorted. At the same time, the most unfavorable condition performance J worst and the flow state stability degree S(X) of each candidate body shape are extracted. Based on the multi-condition comprehensive fitness score, sorting is performed while strictly examining whether the most unfavorable condition performance and the flow state stability degree meet the engineering safety margin requirement. For example, among the top several body shapes in the score, the body shape with the minimum J worst and the maximum S(X) margin is selected. The selected body shape is the final output unified optimization body shape scheme, which is the dimensionless geometric parameter vector X optimal .

[0090] In one possible embodiment, after obtaining the unified optimization body shape scheme, it further includes:

[0091] The unified optimization body shape dimensionless parameters are converted into structural geometric design parameters.

[0092] In this embodiment, the input unified optimization body shape scheme is the optimal dimensionless geometric parameter vector Xoptimal and engineering base data set, especially the structural reference dimension, e.g. the reference diameter D of the flood discharge shaft. During the inverse transformation, the shape parameters are defined as a dimensionless vector X, e.g. x1 = H / D (depth of the stilling basin relative to the diameter), x2 = L / D (length of the stilling basin relative to the diameter), where L is the specific length of the stilling basin. According to X optimal = [x 1，optimal , x 2，optimal ,..., x m，optimal ], the specific geometric dimensions with engineering scales are inversely calculated. For example, if x optimal = 1.5 in X 1，optimal and the reference diameter D = 10 meters, then the inversely calculated specific depth of the stilling basin H = x 1， optimal ·D = 1.5·10 = 15 meters. By performing such inverse calculation on all components in X optimal , a complete set of structural geometric design parameters are obtained, including the specific length, width, floor elevation, turning radius, etc. (in meters or degrees) of the stilling basin, which are used to guide the subsequent engineering drawing.

[0093] Perform unified optimization shape for multi-condition engineering level checking.

[0094] In this embodiment, the input is the set of structural geometric design parameters and the set of typical flood conditions. On the real engineering scale, the final selected design scheme is checked for the last time with high precision to verify its performance. Specifically, based on the parameter set, a full-size three-dimensional geometric model and a numerical calculation model are reconstructed. On this full-size model, engineering-level fluid dynamics calculations are performed for the set of typical flood conditions and necessary checking conditions, such as other flow conditions between the typical conditions. This calculation is usually the highest precision and is used for final confirmation. The output includes the engineering checking flow field and index data set containing the flow field distribution, water level line shape and key hydraulic indicators under each condition.

[0095] Perform engineering checking results and multi-condition physical constraint comparison.

[0096] Specifically, the engineering checking flow field and index data set, and the original set of multi-condition physical constraints, i.e. P max , σ allowed , etc. are input. This step is a verification process of item-by-item comparison. For example, the maximum impact pressure P check under the checking condition is extracted from the data set and compared with the allowable maximum impact pressure P max to determine whether P check ≤ P max. By checking all key indicators for all operating conditions, an engineering check conclusion dataset is formed; the key indicators include energy dissipation rate, cavitation risk, water surface oscillation, outflow uniformity, etc. In some optional embodiments, if it is found that some indicators, for example, P check , slightly exceed the constraints, but are still within the engineering acceptable range or can be solved by fine-tuning the structure, small adjustments to the structure geometry parameter set can also be included to obtain a revised structure geometry parameter set, and the impact of the revised parameters on the hydraulic indicators and flow patterns of each operating condition is recorded to evaluate the safety margin and stability of the final design.

[0097] Output the structure design parameters and the check report.

[0098] In this embodiment, the key results of all the preceding steps are summarized. Specifically, the structure geometry parameter set or the revised structure geometry parameter set is summarized with the engineering check conclusion dataset. A complete structure design parameter and check report is formed. The report preferably includes: design basis, source of joint feasible region, method description of flow pattern partitioning and active sampling, dimensionless parameters of unified optimization body shape and final structure geometry parameters, and multi-condition engineering check results. The report provides complete and traceable technical support for engineering design decisions.

[0099] In one possible embodiment, the construction of the flow regime partition model and the hydraulic index proxy model can also be for: reading an initial multi-working condition flow regime and hydraulic index data set, from which the three-dimensional velocity field, pressure field and free water surface position field corresponding to each individual size parameter and each typical flood working condition combination are extracted to form an original multi-working condition flow field original data subset; for the multi-working condition flow field original data subset, the coordinate system, time step and grid topology are uniformly processed, the flow field results under different sizes and different working conditions are mapped to a unified reference coordinate system and a unified cross-section position, while the obvious numerical abnormal points and convergence failure working conditions are removed, to obtain a multi-working condition flow field pretreatment data set for subsequent feature calculation, which continues to correspond one-to-one with the original size dimensionless geometric parameters and working condition parameters. Reading the multi-working condition flow field pretreatment data set, combining the engineering experience of shaft stilling well and turning section flow regime control, determining the feature types for representing the flow regime, including shaft vortex intensity distribution features, vortex statistics features near the stilling well bottom plate, negative pressure region volume and spatial position, annular hydraulic jump length and relative height, main impact point spatial coordinates on the bottom plate and drift range, outflow cross-section velocity field velocity uniformity index, etc., giving specific calculation methods and value positions for each type of feature, forming a set of flow regime feature definition rule set describing feature name, calculation formula, cross-section position and statistical method. Reading the flow regime feature definition rule set and the multi-working condition flow field pretreatment data set, for each size and working condition combination, according to the flow regime feature definition rule set, calculating vortex intensity statistics, negative pressure region volume, annular hydraulic jump length and relative height, impact point coordinates and drift range, outflow cross-section velocity non-uniformity index and other flow regime features item by item, arranging the obtained multiple feature values in a fixed order to form a feature vector, and binding the feature vector with the corresponding size dimensionless geometric parameters and working condition parameters to form a multi-working condition flow regime feature original data set containing multiple size-working condition-feature vector combinations. Reading the multi-working condition flow regime feature original data set, normalizing each feature dimension according to its physical dimension and typical value range, for example, using linear normalization based on empirical upper and lower limits or standardization based on statistical standard deviation, mapping features of different dimensions and different orders of magnitude to a unified numerical interval, thereby eliminating the dimensional differences between features; after normalization, the normalized feature vector and the corresponding size dimensionless geometric parameters and working condition parameters are kept in one-to-one correspondence to form a complete multi-working condition flow regime feature data set.The multi-working condition flow state characteristic data set and the initial multi-working condition flow state and hydraulic index data set are matched according to the combination key of the body shape dimensionless geometric parameter and the working condition parameter, the normalized characteristic vector is combined with the calculated energy dissipation rate, the maximum impact pressure of the bottom plate, the wall surface pulsating pressure unevenness coefficient, the relative height of the annular hydraulic jump and the outflow velocity unevenness coefficient and other hydraulic indexes to form a complete multi-working condition flow state and hydraulic index characteristic data set, which contains both the characteristic information describing the flow state form and the hydraulic performance index, and is the common basic data for subsequent flow state partition and multi-working condition hydraulic index proxy model training.

[0100] The normalized flow state feature vectors are selected as clustering inputs, and all samples of the body type and working condition combinations are subjected to unsupervised clustering analysis by distance measurement or density measurement method, so as to naturally divide the samples into several flow state clustering clusters, and each clustering cluster corresponds to a typical flow state mode; the clustering results are recorded as the clustering numbers of each sample, and a flow state clustering result data set including the corresponding relationship between the sample index and the clustering number is formed. The flow state clustering result data set and the corresponding multi-working condition flow state and hydraulic index feature data set are read, and the clustering clusters are manually reviewed by engineering and technical personnel in combination with the flow state feature mean value, hydraulic index level and typical flow field image of each cluster in the clustering result, the clustering clusters are merged or split according to the engineering significance of stable flow state, sub-stable flow state and unstable flow state, and the engineering flow state category corresponding to each clustering cluster is determined; after the review and adjustment are completed, the final flow state category label is assigned to each sample, and a flow state category label set is formed, which records the engineering flow state category corresponding to each body type and working condition combination. The body type dimensionless geometric parameters and working condition parameters in the multi-working condition flow state and hydraulic index feature data set and the flow state category label set are read, the samples are divided into a training subset and a verification subset according to a preset proportion, the body type dimensionless geometric parameters and working condition parameters are taken as inputs, and the flow state category label is taken as output to construct a flow state classification data set; on this basis, a suitable classification algorithm is selected to construct an initial flow state partition model, and the parameters are trained through the training subset, so that the model can predict the corresponding flow state category according to the body type dimensionless geometric parameters and working condition parameters, and the initial flow state partition model for predicting the flow state category and its probability information is obtained after the training is completed. The initial flow state partition model and the verification subset data are read, the flow state category is predicted on the verification subset, the prediction result is compared with the real flow state category label, the classification accuracy, confusion matrix and recall rate of each category are calculated, the class probability output by the model is extracted as the flow state classification confidence, and the confidence distribution is counted; whether the initial flow state partition model meets the accuracy requirement is judged according to these evaluation indexes, when the model performance meets the preset standard, the initial flow state partition model is determined as a converged candidate multi-working condition flow state partition model, and a flow state partition model performance evaluation result including classification accuracy data and confidence statistical data is formed, which provides a basis for measuring the flow state boundary distance by using the classification confidence in the subsequent active sampling process.

[0101] The multi-working condition flow regime and hydraulic index characteristic data set and the flow regime category label set are read, the dimensionless geometric parameters of the body shape, the working condition parameters, and the corresponding flow regime category information are taken as inputs, and the corresponding energy dissipation rate, the maximum impact pressure of the bottom plate, the wall surface fluctuating pressure non-uniformity coefficient, the relative height of the annular hydraulic jump, and the outflow velocity non-uniformity coefficient and other hydraulic indexes are taken as outputs, to construct a multi-working condition hydraulic index training data set containing the input-output mapping relationship, and the data set is divided into a training subset and a verification subset according to a preset proportion. The multi-working condition hydraulic index training data set is read, and according to the number of input variables, the number of output indexes, and the expected fitting accuracy, one or more proxy model forms suitable for multi-output regression are selected, for example, a multi-output proxy model based on a radial basis function, and the structure parameters of the proxy model are configured, including the number of basis function centers, the basis function width, the regularization coefficient, etc., to form an initial structure parameter configuration set of the proxy model, which determines the specific form and complexity of the initial multi-working condition hydraulic index proxy model. The multi-working condition hydraulic index training data set and the initial structure parameter configuration set are read, and the multi-working condition hydraulic index proxy model is trained on the training subset, and the least square method or the regularization method is used to fit the mapping relationship from input to output, to obtain an initial multi-working condition hydraulic index proxy model for predicting the multi-working condition hydraulic index; the change of the training error with iterations is recorded during the training process, to form a hydraulic index proxy model training error sequence. The initial multi-working condition hydraulic index proxy model and the verification subset data are read, the dimensionless geometric parameters of the body shape, the working condition parameters, and the flow regime category in the verification subset are input into the proxy model, to obtain the corresponding predicted hydraulic indexes, which are compared with the real hydraulic indexes in the verification subset, the fitting accuracy indexes of each hydraulic index are calculated, such as the determination coefficient, the mean square error, and the maximum relative error, etc., and these accuracy indexes are summarized to form an initial proxy model accuracy evaluation result.

[0102] The joint feasible parameter range and the current calculated body shape sample set, including the initial body shape parameter sample set and the newly added body shape parameter sample set, are read, and the space-filling method is used to generate body shape parameter points in the joint feasible parameter range that have not been subjected to three-dimensional numerical calculation, and to ensure that these candidate body shapes maintain a certain distance from the existing samples in the dimensionless parameter space, so as to avoid excessive concentration of sampling, thereby obtaining a candidate body shape parameter set covering the blank area of the joint feasible parameter range. The candidate body shape parameter set, the set of typical flood working conditions, the converged candidate multi-working condition flow regime partition model, and the initial multi-working condition hydraulic index proxy model are read, and for each body shape parameter point in the candidate body shape parameter set, the converged candidate flow regime partition model is called to predict the flow regime category and its classification confidence under each typical flood working condition, and the initial multi-working condition hydraulic index proxy model is called to predict the corresponding multi-working condition hydraulic indexes such as energy dissipation rate, maximum impact pressure on the bottom plate, wall surface fluctuating pressure non-uniformity coefficient, relative height of annular hydraulic jump, and outflow velocity non-uniformity coefficient. The predicted flow regime categories, classification confidences, and hydraulic indexes under all working conditions are arranged according to the combination of body shape and working condition, forming a candidate body shape multi-working condition prediction performance dataset. The candidate body shape multi-working condition prediction performance dataset is read to evaluate the accuracy of the initial proxy model, and the error information of each hydraulic index in the initial proxy model accuracy evaluation result is used to map the error level to the prediction uncertainty estimate of the corresponding hydraulic index. On this basis, the predicted hydraulic indexes of each candidate body shape under all typical flood working conditions are summarized, and the overall uncertainty of the multi-working condition weighted fitness prediction is calculated. The classification confidence of each candidate body shape under each working condition is compared with the preset flow regime boundary judgment threshold using the classification confidence output by the flow regime partition model, and the proximity of the candidate body shape to the flow regime partition boundary is obtained, and the change of the flow regime category between different working conditions of the candidate body shape is counted, thereby calculating the flow regime consistency index across working conditions. Through the above operations, the active sampling feature index dataset describing the multi-working condition prediction uncertainty, flow regime boundary distance, and flow regime consistency degree of each candidate body shape is obtained. The active sampling feature index dataset and the working condition weight information in the set of typical flood working conditions are read, and the multi-working condition prediction uncertainty, cross-working condition flow regime consistency, and proximity to the relative flow regime boundary of each candidate body shape are weighted and fused according to the preset weight rules, such as assigning a higher weight to the multi-working condition prediction uncertainty, assigning a higher value to the samples close to the flow regime partition boundary, and giving additional importance scores to the samples with larger cross-working condition flow regime changes, and finally calculating the comprehensive active sampling value score for each candidate body shape. The active sampling value index set is formed by arranging all candidate body shapes and their value scores.

[0103] The active sampling value index set is read, and a number of candidate body parameter points are selected in order from high to low of the active sampling value score, to ensure that the selected body type has good distribution in the dimensionless parameter space and takes into account the high uncertainty area and the area close to the flow state partition boundary, to construct the new body parameter sample set that needs to be calculated by high-precision three-dimensional numerical calculation in this round. The new body parameter sample set, the typical flood working condition set, and the three-dimensional body geometry model construction method and boundary condition configuration rules are read, and for each new body parameter sample under each typical flood working condition, a corresponding three-dimensional body geometry model and numerical calculation model are constructed, and the same inlet flow boundary, outlet water level boundary and wall condition as the original sample are set, and three-dimensional numerical calculation is performed to obtain the corresponding velocity field, pressure field and free water surface shape, and these results are arranged into an incremental multi-working-condition flow field original data set; the multiple hydraulic indicators of the new sample under each working condition are calculated, and the incremental multi-working-condition hydraulic indicator original data set is arranged. The initial multi-working-condition flow state and hydraulic indicator data set or the converged multi-working-condition flow state and hydraulic indicator data set formed by the last iteration, as well as the incremental multi-working-condition flow field original data set and the incremental multi-working-condition hydraulic indicator original data set are read, and the incremental flow field data and the incremental hydraulic indicator data are appended to the original data set according to the body parameter and the working condition parameter, and the feature preprocessing and feature merging operations are repeated for the new sample, and the flow state features and the hydraulic indicators of the incremental sample are included in the unified data structure to form an updated multi-working-condition flow state and hydraulic indicator data set containing the original sample and the new sample. The updated multi-working-condition flow state and hydraulic indicator data set is updated, and the feature extraction and flow state partition model training process are performed again based on the updated data set to obtain a new updated multi-working-condition flow state partition model and a corresponding updated flow state partition model performance evaluation result; at the same time, the multi-working-condition hydraulic indicator proxy model is retrained using the updated multi-working-condition flow state and hydraulic indicator data set to obtain a new updated multi-working-condition hydraulic indicator proxy model and a corresponding updated proxy model precision evaluation result to reflect the improvement of the new sample on the flow state division and the hydraulic indicator prediction precision.The performance evaluation results of the updated flow regime partition model and the accuracy evaluation results of the updated agent model are read, and the flow regime classification accuracy, the multi-working condition hydraulic index prediction accuracy, and the active sampling round are comprehensively evaluated according to a preset convergence judgment criterion. When the accuracy indexes of the two types of models change by less than a preset threshold in the latest number of iterations, and the influence of the newly added samples brought by the active sampling on the overall multi-working condition adaptability evaluation result has been weakened to an acceptable level, the current updated multi-working condition flow regime partition model and the updated multi-working condition hydraulic index agent model are determined as the converged multi-working condition flow regime partition model and the converged multi-working condition hydraulic index agent model, the updated multi-working condition flow regime and hydraulic index dataset is determined as the converged multi-working condition flow regime and hydraulic index dataset, and all the newly added body parameter samples in the current round and previous rounds are uniformly arranged into an updated body sample set. If the convergence condition is not met, the updated multi-working condition flow regime and hydraulic index dataset and the updated model are taken as new initial states, and the loop is re-executed until the convergence condition is met.

[0104] According to an aspect of the present application, a flood discharge shaft stilling well-bend section body optimization method can further comprise: performing preliminary numerical simulation on the whole flood discharge shaft to calculate the flow field and water phase distribution, determine the influence weight of N typical floods on the stilling well and the bend section, and extract the corresponding flow characteristics, including the inflow Q and the outlet water level H of the bend section t , as the control conditions for subsequent sample CFD calculation. The initial value range of the body parameters is determined by combining engineering experience and specifications, and the upper limit, average value, and lower limit of each parameter are respectively used for preliminary numerical simulation and comparison with the original scheme flow regime to further narrow down the sampling value range of the body parameter set. Within the range, M design parameter samples are generated by using Latin hypercube sampling or orthogonal experiment method, the local models of the stilling well and the bend section of each sample are constructed, and the flow characteristics obtained in the foregoing step are input as control conditions to simplify the calculation. For the N typical flood conditions that have been determined, CFD modeling calculation is performed on the M samples (total calculation amount is MxN), and the key hydraulic indexes are extracted.

[0105] According to the correlation between the body parameters of the stilling well and the bend section and the hydraulic indexes, a high-precision RBF agent model is fitted to map the law of body parameters and hydraulic performance. Based on the RBF model, a multi-objective optimization model is established for each of the N typical flood conditions, the Pareto front is solved by the NSGA-II multi-objective algorithm, and the set of optimal solutions of the body parameters corresponding to the hydraulic indexes under each typical flood condition is obtained. For the set of optimal body parameters of each of the N conditions, the hydraulic indexes under the remaining (N-1) typical flood conditions are calculated for each body, and the weighted total score is calculated by combining the previously determined flood frequency influence weight, to obtain the comprehensive flow adaptability score of each body for each frequency flood, and the body with the optimal comprehensive flow adaptability is selected as the final optimized scheme.

[0106] In a specific embodiment, the spillway shaft stilling well-bend section body shape unified optimization method of the present application is used to adjust the spillway shaft stilling well-bend section. Specifically, as shown in Figure 5 and Figure 6 , for the spillway shaft of a certain pumped storage power station project, two modification schemes, i.e. the first modification scheme and the second modification scheme, are set for the overall structure of the lower stilling well of the spillway shaft and the L-shaped bend section connecting with the drainage tunnel. The black line in the bottom drawing is the original scheme, and the position of the stilling well-bend section structure in the entire spillway of the spillway shaft is shown by the purple schematic line. The first modification scheme: the upper and lower wall surface short circular arc curves of the L-shaped bend section are changed to elliptic curves, the upper edge elliptic curve equation is x 2 / 6.3 2 +(2.5-y) 2 / 2.5 2 =1, and the lower edge elliptic curve equation is x 2 / 6.0 2 +(3.5-y) 2 / 3.5 2 =1; the pressure slope section is appropriately lengthened, the total length from the shaft center to the end of the pressure slope section is 19.3m, and the outlet height of the pressure slope section is 2m; the stilling well cross section is a circle with a diameter of 5m, and the depth of the stilling well is about 10.5m from the elevation of the first end of the pressure slope section bottom plate, and the bottom plate of the well is lowered to an elevation of 683.64m. The second modification scheme: the upper and lower wall surface elliptic curves are adjusted, the upper edge elliptic curve equation is x 2 / 6.0 2 +(3.5-y) 2 / 3.5 2 =1, and the lower edge elliptic curve equation is x 2 / 6.2 2 +(3-y) 2 / 3 2 =1; the pressure slope section is further lengthened, the total length from the shaft center to the end of the pressure slope section is 30.65m, and the outlet height of the pressure slope section is 2m; the stilling well is lengthened by 2m to the downstream side, the cross section is expanded from a circle to a horseshoe shape with a head-to-tail distance of 7m; the depth of the stilling well is 12.0m from the elevation of the first end of the pressure slope section bottom plate, and the bottom plate of the well is lowered to an elevation of 682.77m.

[0107] In summary, the body shape optimization method of the spillway shaft stilling well-bend section of the application is based on the derivation and construction of the joint feasible region under multiple working conditions as the hard boundary of optimization. A three-layer mapping proxy model of body shape-flow state-index is constructed: a flow state partition model is constructed through flow state feature extraction and cluster analysis, which is used to map the body shape and working condition into discrete flow state categories; the body shape, working condition and flow state category are collectively used as input to construct a high-precision hydraulic index proxy model. An active sampling strategy is adopted, which integrates the uncertainty of multiple working conditions, the boundary distance of flow state partition and the consistency of cross-working condition flow state, to iteratively update the model. In the joint feasible region, a robust optimization model considering the stability of cross-working condition flow state is solved to obtain the unified optimization body shape scheme.

[0108] The application proposes a three-layer mapping architecture of body shape-flow state-index, which divides the complex continuous flow field into discrete flow state categories with clear engineering significance through flow state feature extraction and unsupervised clustering. On this basis, two cascaded models are constructed: a flow state partition model R and a hydraulic index model I with flow state category R as explicit input. The complex nonlinear fitting task is divided into classification and conditional regression, which enables the model to physically and accurately capture and express the flow state mutation boundary, solving the bottleneck of insufficient accuracy of traditional models near the flow state transition point. An active sampling strategy driven by flow state boundary is also proposed. Instead of relying solely on the prediction variance of the hydraulic index to guide sampling, a fused sampling value function V(x) is constructed. This function introduces two key terms: flow state partition boundary distance D(x) and cross-working condition flow state category prediction consistency C(x). This enables the sampling strategy to actively and intelligently identify and encrypt the flow state boundary regions that are most confusing to the model f and the regions where the flow state jumps between working conditions. This enables the accurate allocation of computing resources to the most critical parameter space positions for capturing flow state mutations and ensuring robustness, solving the problem of sampling blindness and improving optimization efficiency.

[0109] The above describes the preferred embodiments of the application in detail, but the application is not limited to the specific details in the above embodiments. Within the technical concept range of the application, various equivalent transformations of the technical solutions of the application can be made, and these equivalent transformations all belong to the protection range of the application.

Claims

1. A method for optimizing the shape of a stilling well-bend section of a flood release shaft, characterized in that, The application relates to a method for optimizing the body shape of a spillway shaft, comprising the following steps: acquiring multi-condition physical constraints describing the body shape parameters of the spillway shaft, the stilling well and the turning section, and constructing a joint feasible region; meanwhile, a set of typical flood working conditions is acquired; adopting an active sampling strategy to iteratively acquire body shape sample data in the joint feasible region for the set of typical flood working conditions; a flow state partition model and a hydraulic index proxy model are iteratively constructed according to the body shape sample data; the active sampling strategy fuses multi-condition uncertainty and flow state partition boundary information; in the joint feasible region, the flow state partition model and the hydraulic index proxy model are applied to construct and solve a robust optimization model considering cross-condition flow state stability, and a unified optimization body shape scheme is obtained; iteratively constructing the flow state partition model and the hydraulic index proxy model, comprising: constructing a flow state partition model R=f(X,C), which is used for mapping the body shape parameters and the working conditions into discrete flow state categories; wherein X represents the input body shape parameters; C represents the working conditions; f() represents the flow state partition model; R is the discrete flow state category; constructing a hydraulic index proxy model I=g(X,C,R), which is used for mapping the body shape parameters, the working conditions and the flow state category determined by the flow state partition model into a hydraulic index; wherein g() is the hydraulic index proxy model; I is the hydraulic index; obtaining the unified optimization body shape scheme, comprising: using the hydraulic index proxy model to construct a multi-condition weighted fitness function and a least favorable working condition performance function; using the flow state partition model to construct a cross-condition flow state stability evaluation function; Body shape parameters are taken as decision variables, a joint feasible region is taken as the optimization objective min X∈Ωjoint {J weighted worst (X)} and a cross-condition flow state stability evaluation function S(X)≥S min is taken as a constraint condition, and a unified optimization body shape scheme is obtained; wherein X is an optimization decision variable, i.e. a body shape parameter; Ω joint is a joint feasible region; min indicates taking a minimum value; J weighted (X) is a multi-condition weighted fitness function; J worst (X) is a most unfavorable condition performance function; S(X) is a cross-condition flow state stability evaluation function; S min is a preset allowed minimum flow state stability threshold.​ iteratively acquiring the body shape sample data, comprising: generating candidate body shape parameter points that have not been calculated in the joint feasible region; constructing an active sampling value function to evaluate the candidate body shape parameter points; according to the evaluation results, selecting new body shape samples to iteratively acquire the body shape sample data; constructing the active sampling value function, comprising: using the hydraulic index proxy model to evaluate the multi-condition comprehensive fitness prediction uncertainty of the candidate body shape parameter points; using the flow state partition model to evaluate the flow state partition boundary distance of the candidate body shape parameter points; using the flow state partition model to evaluate the cross-condition flow state category prediction consistency of the candidate body shape parameter points in the set of typical flood working conditions; fusing the multi-condition comprehensive fitness prediction uncertainty, the flow state partition boundary distance and the cross-condition flow state category prediction consistency to determine the active sampling value function V(x)=w1·U(x)+w2·D(x)+w3·C(x); wherein V(x) is the active sampling value score of the candidate body shape parameter point x; U(x) is the multi-condition comprehensive fitness prediction uncertainty; D(x) is the flow state partition boundary distance; C(x) is the cross-condition flow state category prediction consistency; w1, w2 and w3 are preset weight coefficients for fusing the three dimensions; constructing the joint feasible region, comprising: The body size parameter X is defined as a dimensionless geometric parameter vector; for each working condition c m in a typical flood working condition set C = {c1,..., c j , a single-working-condition physical constraint inequality describing the physical limit of the working condition is derived, and a single-working-condition physical constraint inequality set defines a single-working-condition physical feasible region Ω j corresponding to the working condition; The joint feasible region Ω is determined by performing intersection operation on all single-condition physical feasible regions joint =∩ j=1 m Ω j ; where j is the index of the condition, m is the total number of typical flood conditions, Ω j is the single-condition physical feasible region corresponding to the jth condition, Ω joint is the joint feasible region satisfying all condition constraints.

2. The method of claim 1, wherein, constructing the flow state partition model, comprising: extracting flow state features representing flow patterns from multi-condition flow field data contained in the body shape sample data to form a flow state feature data set; performing cluster analysis on the flow state feature data set to define flow state categories and generate flow state category labels for the body shape samples; taking the body shape parameters and the working conditions in the body shape sample data as inputs and taking the flow state category labels as outputs, the flow state partition model is trained.

3. The method of claim 1, wherein, Constructing the flow regime stability evaluation function across working conditions, including: Applying the flow regime partition model to predict the flow regime category and classification confidence of any body parameter under each working condition in the typical flood working condition set; A flow regime stability evaluation function S(X) = min j∈{1，2，…，m} {p(R = R stable | X, c j j)} ; where j is the index of the operating condition; m is the total number of typical flood operating conditions; c j is the jth typical flood operating condition; R stable is the target stable flow regime category; R is the discrete flow regime category; p( | ) is the classification confidence; p(R = R stable | X, c j j) represents the classification confidence that the body X is predicted to be in the stable flow regime at the operating condition c j .

4. The method of claim 1, wherein, Evaluating the flow regime partition boundary distance, including: Using the flow regime partition model to predict the flow regime category and corresponding classification confidence of the candidate body parameter point under each working condition in the typical flood working condition set; Comparing the classification confidence with the preset flow regime boundary judgment threshold to quantify the proximity of the candidate body parameter point to the flow regime partition boundary; The proximity of the flow regime partition boundary is taken as the flow regime partition boundary distance D(x) = max j∈{1，2，…，m} {D j (x)}; D j (x) = 1 - max k {p(R k | x, c j )}; where D(x) is the flow regime partition boundary distance for candidate body type x; D j (x) is the flow regime boundary proximity for candidate body type x at a single operating condition c j ; j is the operating condition index; m is the total number of typical flood operating conditions; p(R k | x, c j ) is the classification confidence that x belongs to class k at c j ; max k {} means taking the maximum classification confidence among all flow regime classes k; max j {} means taking the value closest to the flow regime boundary among all operating conditions.

5. The method of claim 1, wherein, Deriving single-working-condition physical constraint inequalities, including: Based on energy equation, momentum equation or cavitation safety condition, the hydraulic response calculation function of body shape parameter X under working condition c j is established. Reading the specification control parameters including the allowable cavitation index and the allowable maximum impact pressure; By comparing the hydraulic response calculation function with the specification control parameter, a single working condition physical constraint inequality is established, the single working condition physical constraint inequality at least comprising: σ(X, c j ) ≥ σ allowed ; P(X, c j ) ≤ P max ; wherein σ(X, c j ) is the cavitation number calculated according to X and c j , σ allowed is the allowable cavitation index in the specification control parameter; P(X, c j ) is the impact pressure calculated according to X and c j , P max is the allowable maximum impact pressure in the specification control parameter.

Images