Dimension reduction collaborative optimization method for features of diversion mechanism

Abstract

The invention discloses a diversion mechanism feature dimension reduction collaborative optimization method, and belongs to the technical field of water turbine optimization. The method specifically comprises the following steps: 1) selecting configuration parameters of a runner and a double-row blade grid, and establishing a first high-dimensional variable library; 2) screening variables in the first high-dimensional variable library based on an elementary effect distribution method to obtain a second high-dimensional variable library; 3) performing dimension reduction processing on variables in the second high-dimensional variable library through matrix decomposition and mapping to obtain a low-dimensional variable library; 4) on the basis of the back propagation neural network model, taking variables in the low-dimensional variable library as input variables, taking performance indexes as output variables, and performing optimization by using a genetic algorithm to obtain an optimal solution; and 5) reflecting the optimal solution back to the original design space to obtain an optimized configuration parameter, and importing the optimized configuration parameter into three-dimensional structure software to generate an optimized runner and double-row cascade entity. According to the method, multi-mode optimization is considered, high-dimensional complex calculation is converted into low-dimensional efficient calculation, the calculation time is greatly shortened, and the optimization efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydraulic turbine optimization, and particularly relates to a collaborative optimization method for feature dimensionality reduction of a guide mechanism. Background Art

[0002] As the core component of a pumped-storage unit, a reversible hydraulic turbine has the ability to rotate bidirectionally and covers two working modes: the pump mode (energy storage) and the hydraulic turbine mode (power generation). The runner is the most critical component of a reversible hydraulic turbine, directly responsible for the two-way efficient conversion of water potential energy and electrical energy. It has a complex and twisted three-dimensional shape, which plays a decisive role in the efficiency of the hydraulic turbine. The double-layer radial water guide structure (i.e., double-row vane cascade) is also an important part of a reversible hydraulic turbine, and its main function is to regulate the flow rate and guide the water flow direction. Its structure is composed of cylindrical blades, which are divided into fixed guide vanes and movable guide vanes. The blade cross-section is an airfoil cross-section. The fixed guide vanes are located outside the movable guide vanes, and the two are collectively referred to as double-ring guide vanes, double-row vane cascades or radial ring vane cascades. The fixed guide vanes are welded to the housing and belong to stationary blades; the movable guide vanes are provided with a central shaft and can adjust the opening angle around the central shaft to realize the regulation of the flow rate and the change of the water flow direction.

[0003] A reversible hydraulic turbine contains multiple key components. During the optimization process, it is necessary to comprehensively consider the collaborative matching between different components. Since there are numerous configuration parameters that affect performance such as efficiency, cavitation, and vibration, especially for the runner, which is a three-dimensional twisted complex surface, the airfoil varies with the blade height. Usually, airfoil cross-sections need to be intercepted at different blade height planes. The more airfoil cross-sections are intercepted, the more parameters are involved. As a result, the number of optimization variable parameters is huge, making the optimization solution process face great difficulties. In the field of feature dimensionality reduction technology, relatively few research results have been achieved at home and abroad. Currently, the research that is relatively in-depth is the spatial mapping technology based on POD modal coefficients, but this method has obvious defects. Its operation steps are cumbersome, and multiple related matrices need to be used for calculation, resulting in the final reduction accuracy being difficult to reach the ideal level and difficult to meet the high requirements for accuracy in actual engineering applications.

[0004] Therefore, there is an urgent need for a method that can simply represent high-dimensional features. On the premise of ensuring accuracy, it can effectively reduce the number of optimization variables, simplify high-dimensional optimization problems, and improve the operation efficiency. Feature dimensionality reduction technology precisely achieves this goal by projecting optimization variables into a new design space to reduce the number of variables. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies in the prior art and provide a collaborative optimization method for feature dimensionality reduction of a guide mechanism.

[0006] The specific technical solution adopted by the present invention is as follows:

[0007] The present invention provides a method for collaborative optimization of the characteristics of a flow guiding mechanism, and the specific steps are as follows:

[0008] S1: Construct a first high-dimensional variable library with the configuration parameters of the runner and the double-row cascade as the parameters to be optimized;

[0009] S2: Based on the elementary effect distribution method, screen the variables in the first high-dimensional variable library, and incorporate the selected important variables into the second high-dimensional variable library;

[0010] S3: Perform dimensionality reduction processing on the variables in the second high-dimensional variable library through matrix decomposition and mapping to obtain a low-dimensional variable library;

[0011] S4: Based on the backpropagation neural network model, use the genetic algorithm to optimize with the variables in the low-dimensional variable library as input variables and the performance indicators as output variables to obtain the optimal solution;

[0012] S5: Inverse map the optimal solution back to the original design space, obtain the optimized configuration parameters and import them into the 3D structure software to generate the optimized runner and double-row cascade entities.

[0013] Preferably, in step S1, the configuration parameters include the shape parameters of the movable guide vanes, the shape parameters of the fixed guide vanes, the relative position relationship parameters between the movable guide vanes and the fixed guide vanes, the axial plane parameters, and the runner blade angle parameters.

[0014] Preferably, the elementary effect distribution method in step S2 is specifically as follows:

[0015] S21: Sample from the design space to which each variable in the first high-dimensional variable library belongs, and all the sampled variable values form a first high-dimensional variable library sample sequence X G1 ; Input the first high-dimensional variable library sample sequence X G1 into the 3D structure software to draw the 3D runner and double-row cascade entities; Use the 3D numerical simulation software to calculate the obtained entities to obtain the performance indicators Y of the first high-dimensional variable library G1 ; The variable values and performance indicators sampled from the first high-dimensional variable library form a matrix [X G1 , Y G1 ;

[0016] S22: Based on [X G1 , Y G1 , calculate the absolute value S1 of the sample mean and the sample standard deviation S2 of the elementary effect d i (x) to measure the influence degree of the variable on the performance indicator; Eliminate the variables with insignificant influence, and incorporate the remaining variables into the second high-dimensional variable library.

[0017] Furthermore, variables with the absolute value of the sample mean S1 less than 5 and variables with the sample standard deviation S2 within the range of [0, 5] are excluded.

[0018] Furthermore, the dimensionality reduction process in step S3 is as follows:

[0019] S31: Using the Latin hypercube sampling method, sample from the design space of each variable in the second highest-dimensional variable library. All the sampled variable values form the second highest-dimensional variable library sample sequence X for generating three-dimensional entities G2 ; Input the second highest-dimensional variable library sample sequence X G2 into the three-dimensional structure software to draw the three-dimensional runner and the double-row cascade entities; Use the three-dimensional numerical simulation software to calculate the obtained entities to obtain the performance indicators Y of the second highest-dimensional variable library G2 ; The variable values and performance indicators sampled from the second highest-dimensional variable library form the matrix [X G2 , Y G2 ;

[0020] S32: Using the singular value decomposition theory, denote the number of variables in the second highest-dimensional variable library as n and the number of sampling times as m. Arrange the sampled variables into a matrix, denoted as the sample matrix input m×n ; Input the sample matrix input m×n and decompose it into the left singular matrix U m×m , the eigenvalue matrix S m×n and the transpose of the right singular matrix V n×n ;

[0021] S33: Perform truncation according to the determined truncation value k j , where k j < n. The specific truncation method is as follows: Retain the first k m×m columns of U j to obtain the truncated left singular matrix Retain the first k m×n columns and the first k j rows of S j to obtain the truncated eigenvalue matrix Retain the first k n×n columns of V j to obtain the truncated right singular matrix

[0022] S34: Reconstruct the output matrix output from the truncated left singular matrix , the eigenvalue matrix and the transpose of the right singular matrix ; m×n ;

[0023] S35: Calculate the truncation error error,

[0024]

[0025] S36: If the truncation error error is greater than a preset threshold, reselect the truncation value k j , and repeat steps S33 to S35 until the truncation error error is less than or equal to the preset threshold; the threshold is selected from 0.01 to 0.05;

[0026] S37: Take the truncated right singular matrix above as the projection coefficient matrix, and project the sample sequence X in the second high-dimensional variable library G2 onto the new low-dimensional design space to obtain a low-dimensional variable library:

[0027]

[0028] where is the sample matrix in the low-dimensional variable library, which is composed of m sample sequences X after dimension reduction to k j dimensions G2 '.

[0029] Furthermore, the performance indicators Y G1 and Y G2 both include the cavitation coefficient σ under the pump operating condition p , the efficiency η under the pump operating condition p , the efficiency η under the turbine operating condition t and the non-blade zone pressure pulsation NVPP under the turbine operating condition t These four performance indicators.

[0030] Furthermore, the specific steps in step S4 are as follows:

[0031] S41: Construct a surrogate model between [X G2 ', Y G2 according to the backpropagation neural network model, where the sample sequence X of the variables in the low-dimensional variable library G2 ' is used as the input variable, and the cavitation coefficient σ under the pump operating condition p , the efficiency η under the pump operating condition p , the efficiency η under the turbine operating condition t and the non-blade zone pressure pulsation NVPP under the turbine operating condition t These four performance indicators are used as the output variables;

[0032] S42: Perform genetic optimization on the surrogate model to obtain the Pareto front X of the low-dimensional variable library G2 '_pareto as the optimal solution.

[0033] Further, the optimal solution in step S5 is mapped back to the original design space as follows:

[0034]

[0035] Where: input_back m×n is the sample matrix after inverse mapping;

[0036] The variable X m×n in the inverse-mapped sample matrix input_back G2 _pareto is used as the optimized configuration parameter.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] (1) Different from the previous local optimization that only considered the configuration parameters of a single component, the goal of the present invention is to achieve the collaborative optimization of the runner and two rows of cascades, and construct the first high-dimensional variable library G1 covering all configuration variables, so as to realize the comprehensive optimization of the overall flow passage of the reversible turbine and improve the overall performance.

[0039] (2) The optimization method provided by the present invention takes into account both the pump and turbine modes; it not only covers cavitation and efficiency, but also incorporates the pressure pulsation index in the non-blade region during the turbine mode into the multi-objective optimization for the first time, making the optimization result achieve a better balance in multi-dimensional performance and broadening the consideration range of the performance optimization of the reversible turbine.

[0040] (3) The optimization method provided by the present invention uses the elementary effect distribution method to screen the variables that significantly affect the performance, and then uses the Latin hypercube sampling combined with matrix decomposition and mapping for dimensionality reduction, converting the high-dimensional complex calculation into a low-dimensional efficient calculation, greatly shortening the calculation time, improving the optimization efficiency, and reducing the consumption of computing resources.

[0041] (4) The optimization method provided by the present invention uses the surrogate model based on the backpropagation neural network as the predictor, takes the genetic algorithm as the optimization means, and takes the neural network fitting model as the optimization goal, which can quickly and accurately find the optimal solution, ensure the reliability and superiority of the optimization result, and provide an efficient and accurate solution for the runner-guide vane collaborative optimization of the reversible turbine. Description of the Drawings

[0042] Figure 1 is the flow chart of the characteristic dimensionality reduction collaborative optimization method for the guide mechanism provided by the present invention;

[0043] Figure 2 is the three-dimensional structure schematic diagram of the runner and the cascade;

[0044] Figure 3 is the control diagram of the Bessel curve of the movable guide vane in the present invention;

[0045] Figure 4 This is the Bessel curve control diagram of the stay vanes in the present invention;

[0046] Figure 5 This is the schematic diagram of the positions of the movable vanes and the stay vanes in the present invention;

[0047] Figure 6 This is the meridional view schematic diagram of the runner and the cascade in the present invention;

[0048] Figure 7 This is the schematic diagram of the runner blades in the present invention;

[0049] Figure 8 This is the comparison diagram of the runner and the double-row cascade entities before and after optimization in Embodiment 1, where (a)-(c) are the runner, the stay vanes, and the movable vanes before optimization respectively; (d)-(f) are the runner, the stay vanes, and the movable vanes after optimization respectively;

[0050] Figure 9 This is the comparison of the runner part before and after optimization in Embodiment 1, where the black part is the runner before optimization, the yellow part is the runner after optimization, (a) is the view from the high-pressure side; (b) is the view from the low-pressure side; (c) is the three-dimensional view;

[0051] Figure 10 This is the comparison diagram of four performances before and after optimization in Embodiment 1, where (a) is the efficiency η under the turbine condition t ; (b) is the efficiency η under the pump condition p ; (c) is the non-vane zone pressure pulsation NVPP under the turbine condition t ; (d) is the cavitation coefficient σ under the pump condition p . Detailed implementation manners

[0052] The present invention will be further described and explained below in conjunction with the drawings and specific implementation manners. The technical features of each implementation manner in the present invention can be combined correspondingly without conflict.

[0053] As a preferred embodiment of the specific implementation manner of the present invention, as Figure 1 shown, a method for collaborative optimization of the feature dimensionality reduction of the flow guiding mechanism is provided, and the specific steps are as follows:

[0054] Step 1: Select the configuration parameters of the runner and the double-row cascade, and establish the first high-dimensional variable library G1.

[0055] Figure 2It is a schematic diagram of the three-dimensional structure of the runner and the blade cascade. The blade cascade is a double-row blade cascade, including movable guide vanes and fixed guide vanes. The selected configuration parameters include the shape parameters of the movable guide vanes, the shape parameters of the fixed guide vanes, the relative position relationship VOC between the movable guide vanes and the fixed guide vanes, the axial plane parameters, and the runner blade angle parameters, as shown in Table 1 specifically.

[0056] Configuration parameters in the first high-dimensional variable library G1 of Table 1

[0057]

[0058]

[0059] Figure 3 and Figure 4 are respectively schematic diagrams of the control points of the Bezier curves of the movable guide vanes (gv) and the fixed guide vanes (sv). The coordinate of each control point of the Bezier curve in the figure contains two values in the X-axis direction and the Y-axis direction. Among them, [X1, X2, X3, X4] represents the equally divided points of the guide vane evenly divided into 4 equal parts in the length direction. The length of the guide vane is a fixed value. Therefore, the abscissas [X1, X2, X3, X4] of the control points are constant values that remain unchanged. The ordinates are the variables to be optimized. sv_1, sv_2, sv_3, sv_4, sv_5, sv_6, sv_7, sv_8 are the ordinates controlling the Bezier curve of the fixed guide vane, and gv_1, gv_2, gv_3, gv_4, gv_5, gv_6, gv_7, gv_8 are the ordinates controlling the Bezier curve of the movable guide vane. The change of the values of these ordinates will cause the change of the contour line of the guide vane. Therefore, by optimizing these 16 variables, the shapes of the fixed guide vane and the movable guide vane can be optimized respectively.

[0060] Figure 5 is a schematic diagram of the positions of the movable guide vane and the fixed guide vane in the embodiment. From Figure 5 it can be seen that γ1 represents the included angle between two adjacent fixed guide vanes, and γ2 represents the included angle between the movable guide vane and the fixed guide vane. VOC is used to represent the relative position relationship between the movable guide vane and the fixed guide vane, that is

[0061] Figure 6 is a schematic diagram of the axial plane of the runner and the blade cascade in the embodiment. D0 represents the distance between the rotation center of the movable guide vane and the central axis of the runner, D1 represents the distance between the outer edge of the runner and the central axis of the runner, D b represents the inner diameter at the lower end of the runner, and D2 represents the outer diameter at the lower end of the runner. The span plane is the evenly divided middle section. There are a total of four span planes, which are respectively denoted as span1, span2, span3, and span4. Span represents the blade height.

[0062] Figure 7β1 represents the low-pressure side blade angle, which is the angle between the low-pressure side blade end and the blade bone line in the circumferential direction; β2 represents the high-pressure side blade angle, which is the angle between the low-pressure side blade end and the blade bone line in the circumferential direction; represents the wrap angle of the runner blade. Based on the above, β 1_ span1 represents the low-pressure side blade angle of blade height surface 1, β 2_ span1 represents the high pressure side blade angle of blade height surface 1, Represents the runner blade wrap angle of blade height surface 1, and so on.

[0063] The above 34 variables are used as the first high-dimensional variable library G1.

[0064] Step 2: Perform elementary effect distribution analysis on the variables in the first high-dimensional variable library G1, remove the variables that have little impact on the target performance, and obtain the second high-dimensional variable library G2. The details are as follows:

[0065] S21: sampling from the design space to which each variable in the first high-dimensional variable library G1 belongs, and all the sampled variable values constitute the first high-dimensional variable library G1 sample sequence X for generating a three-dimensional entity G1 The principles of the above sampling are as follows:

[0066] D=[0,1] k , x i ∈D

[0067] Where D represents the design space to which the design variables belong. For simplicity, all variables are normalized to [0,1]. k represents the dimension of the design space, that is, the number of design variables. i represents the design variables in the design space, i=1,...,k.

[0068]

[0069] P represents the degree of division of the variable value range. For example, when p=5,

[0070] The design space of each variable refers to the possible value range of each variable as a parameter to be optimized, and the value range can be determined based on engineering experience and practical application constraints.

[0071] The first high-dimensional variable library G1 sample sequence X G1 Input the three-dimensional structure software to draw the three-dimensional rotor and double-row leaf sedge entities; use the three-dimensional numerical simulation software to calculate the obtained entities and obtain the first high-dimensional variable library G1 performance index Y G1 ; The matrix composed of the variable values and performance indicators extracted from the first high-dimensional variable library G1 is recorded as [X G1 ,YG1 .

[0072] Performance indicators Y of the first high-dimensional variable library G1 G1 It includes four performance indicators, namely the cavitation coefficient σ under pump operating conditions p , the efficiency η under pump operating conditions p , the efficiency η under turbine operating conditions t and the non-vane area pressure pulsation NVPP under turbine operating conditions t These four performance indicators

[0073] Among them, the cavitation coefficient σ under pump operating conditions p is calculated as follows

[0074]

[0075] In the formula

[0076] Δh represents the degree of head drop under pump operating conditions, with the unit of m

[0077] H represents the head, with the unit of m, and g represents the acceleration due to gravity, with the unit of m / s 2 ;

[0078] λ1 represents the dynamic pressure drop coefficient of the water flow around the blade, dimensionless, taking values between 0.2 and 0.4

[0079] ω1 represents the inlet relative flow velocity, with the unit of rad / s

[0080] λ2 represents the comprehensive loss coefficient of the water flow before entering the blade, dimensionless, taking values between 1.0 and 1.4

[0081] v1 represents the inlet absolute flow velocity, with the unit of m / s

[0082] The efficiency η under pump operating conditions p is calculated as follows

[0083]

[0084] In the formula

[0085] ρ represents the density of the fluid medium, generally the density of water, taking 1000 kg / m 3 .

[0086] Q represents the flow rate, with the unit of m 3 / s

[0087] P represents the shaft input power, with the unit of W

[0088] The efficiency η under turbine operating conditions t is calculated as follows

[0089]

[0090] In the formula:

[0091] M is the output torque, with the unit of N · m;

[0092] w is the rotational angular velocity, with the unit of rad / s;

[0093] The pressure pulsation in the vane - less zone under the turbine condition, NVPP t is calculated as follows:

[0094] NVPPt t = P max – P min

[0095] where P max and P min represent the maximum and minimum values of the time - series pressure pulsation data within a finite time, respectively.

[0096] S22: Calculate the sample mean and sample standard deviation of the elementary effect d G1 ,Y G1 (x) based on [X i (x) to measure the influence degree of variables on the performance index. The elementary effect distribution refers to analyzing the cross - effects between different variables and quantifying the importance degree of variables to the target. The calculation method of the elementary effect d i (x) is as follows:

[0097]

[0098] In the formula: d i (x) represents the result of the elementary effect of x i , and y(x) represents the performance index of the runner and the double - row cascade entity represented by x i , which is obtained through numerical simulation.

[0099] S23: Calculate the sample mean and sample standard deviation based on the above d i (x). Denote the absolute value of the sample mean as S1 and the sample standard deviation as S2; eliminate the variables with the absolute value of the sample mean S1 less than 5 and the variables with insignificant influence within the range of [0,5] for the sample standard deviation S2. Incorporate the remaining variables into the second - highest - dimensional variable library G2. At this time, the dimension drops from the k - dimension in the first - highest - dimensional variable library G1 to the n - dimension in the second - highest - dimensional variable library G2.

[0100] Step 3: Perform dimensionality reduction processing on the variables in the second - highest - dimensional variable library G2 through matrix decomposition and mapping to obtain the low - dimensional variable library G2'. Specifically as follows:

[0101] S31: Use the Latin Hypercube Sampling (LHS) method to sample from the design space of each variable in the second high-dimensional variable library G2. All the sampled variable values form the second high-dimensional variable library G2 sample sequence X for generating the three-dimensional entity. G2 . Latin Hypercube Sampling is a stratified sampling technique used to generate sample points in a multi-dimensional space to ensure that the sample points are evenly distributed throughout the space, thereby improving the efficiency and representativeness of sampling and providing a more concise input that can reflect the data characteristics for subsequent dimensionality reduction processing.

[0102] Input the second high-dimensional variable library G2 sample sequence Y G2 into the three-dimensional structure software to draw the three-dimensional runner and the double-row cascade entity; use the three-dimensional numerical simulation software to calculate the obtained entity to obtain the performance index Y of the high-dimensional variable library G2. G2 . Y G2 and Y G1 both contain the same 4 performance indicators. The matrix formed by the variable values and performance indicators extracted from the second high-dimensional variable library G2 is denoted as [X G2 , Y G2 .

[0103] S32: Using the singular value decomposition in matrix decomposition theory, denote the number of variables in the second high-dimensional variable library G2 as n and the number of sampling times as m. Arrange the sampled variables into a matrix, denoted as the sample matrix input m×n ; decompose the sample matrix input m×n into the product of three matrices:

[0104] input m×n = U m×m ×S m×n ×(V n×n ) T

[0105] In the formula: U m×m represents the left singular matrix, S m×n represents the eigenvalue matrix, and V n×n is the right singular matrix;

[0106] S33: Select a cut-off value k j , and use k j to cut off the matrix. Use k j as the number of variables after dimensionality reduction, k j < n. The specific cut-off method is as follows: Retain the first k m×m columns of U j , and denote the truncated left singular matrix as Retain the first k m×n columns of S j and the first kj Row, denote the truncated eigenvalue matrix as Keep the first k columns of V n×n and denote the truncated right singular matrix as j

[0107] S34: Use the truncated left singular matrix eigenvalue matrix and the transpose of the right singular matrix to reconstruct the output matrix output m×n :

[0108]

[0109] S35: Calculate the truncation error error:

[0110]

[0111] S36: If the truncation error error is greater than the preset threshold, reselect the truncation value k j and repeat steps S33 to S35 until the truncation error error is less than or equal to the preset threshold. The threshold is selected between 0.01 and 0.05.

[0112] S37: Take the above truncated right singular matrix V n×kj as the projection coefficient matrix, and project the sample sequence X G2 in the second high-dimensional variable library G2 onto the new low-dimensional design space to obtain the low-dimensional variable library G2':

[0113]

[0114] where is the sample matrix in the low-dimensional variable library G2'.

[0115] j is composed of m sample sequences X G2 ' after dimensionality reduction to k dimensions.

[0115] k j is used as the new number of variables after dimensionality reduction. Since k j < n < k, dimensionality reduction is thus achieved.

[0116] Step Four: Surrogate model and genetic optimization, specifically as follows:

[0117] S41: Construct a surrogate model between [X G2 ', Y G2 according to the backpropagation neural network model, where the sample sequence X G2 ' of the variables in the low-dimensional variable library G2' is used as the input variable, and the cavitation coefficient σ under the pump operating conditionsp 、The efficiency η under the pump operating condition p 、The efficiency η under the turbine operating condition t and the non-vane zone pressure pulsation NVPP under the turbine operating condition t These four performance indicators are used as output variables;

[0118] S42: Genetically optimize the surrogate model to obtain the Pareto front X G2 '_pareto of the low-dimensional variable library G2' as the optimal solution.

[0119] Step Five: Generate the optimized runner and double-row cascade entities.

[0120] Multiply the optimal solution by and map it back to the original design space, specifically as follows:

[0121]

[0122] where: input_back m×n is the sample matrix after the inverse mapping; is the sample matrix in the low-dimensional variable library G2'; is the truncated right singular matrix.

[0123] The variables X m×n in the sample matrix input_back G2 _pareto after the inverse mapping are used as the optimized configuration parameters and imported into the 3D structure software to generate the optimized runner and double-row cascade entities.

[0124] Example 1

[0125] In this example, using the above optimization method, 34 variables including the above movable guide vane shape parameters, fixed guide vane shape parameters, relative position relationship parameters between the movable guide vane and the fixed guide vane, axial plane parameters, and runner blade angle parameters are selected for optimization. The comparison before and after optimization is Figures 8 - 10 as shown.

[0126] As Figure 8 and Figure 9 shown, after optimization, the blade wrap angle of the runner expands, the high-pressure side diameter D1 decreases, and D b and D2 remain basically unchanged. After optimization, the thickness distribution of the fixed guide vane in the length direction remains basically unchanged, but the curvature diameters at both ends of the optimized fixed guide vane slightly decrease. After optimization, the thickness distribution of the movable guide vane in the length direction changes, and the position of the maximum thickness moves from the illustrated position A to position B, and the curvature diameters at both ends basically remain unchanged.

[0127] Perform 3D simulation calculations on the optimized runner and double-row cascade entities to obtain the cavitation coefficient σ under the pump operating conditionp 、Efficiency η under pump operating conditions p 、Efficiency η under turbine operating conditions t and the non-blade zone pressure pulsation NVPP under turbine operating conditions t ,and compare with that before optimization. As Figure 10 shown, after optimization, the efficiencies under both turbine and pump operating conditions have been improved. In the small flow rate section (0.2~1.0Q / Q d ), the pressure pulsation has been significantly reduced, the cavitation effect under pump operating conditions has been reduced, and the cavitation performance has been improved.

[0128] The embodiments described above are only a preferred solution of the present invention, but they are not intended to limit the present invention. Those of ordinary skill in the relevant technical fields can still make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all technical solutions obtained by means of equivalent replacement or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for collaborative optimization of feature dimension reduction of a flow guiding mechanism, characterized in that The specific steps are as follows: S1: Construct a first high-dimensional variable library with the configuration parameters of the runner and the double-row cascade as the parameters to be optimized; S2: Screen the variables in the first high-dimensional variable library based on the elementary effect distribution method, and incorporate the selected important variables into the second high-dimensional variable library; S3: Perform dimensionality reduction processing on the variables in the second high-dimensional variable library through matrix decomposition and mapping to obtain a low-dimensional variable library; S4: Based on the backpropagation neural network model, use the genetic algorithm to optimize with the variables in the low-dimensional variable library as input variables and the performance index as the output variable to obtain the optimal solution; S5: Inverse-map the optimal solution back to the original design space, obtain the optimized configuration parameters, and import them into 3D structural software to generate the optimized runner and double-row cascade entities.

2. The collaborative optimization method for reducing the dimension of the characteristics of the diversion mechanism according to claim 1, wherein In step S1, the configuration parameters include the shape parameters of the movable guide vanes, the shape parameters of the fixed guide vanes, the relative position relationship parameters between the movable guide vanes and the fixed guide vanes, the axial plane parameters, and the runner blade angle parameters.

3. The collaborative optimization method for reducing the dimensionality of the flow guiding mechanism features according to claim 1, characterized in that In step S2, the specific elementary effect distribution method is as follows: S21: Sample from the design space to which each variable in the first high-dimensional variable library belongs. All the extracted variable values form the first high-dimensional variable library sample sequence X for generating a three-dimensional entity G1 ; Input the first high-dimensional variable library sample sequence X G1 into a 3D structure software to draw a 3D runner and a double-row cascade entity; Use a 3D numerical simulation software to calculate the obtained entity to obtain the performance index Y of the first high-dimensional variable library G1 ; The variable values and performance indexes extracted from the first high-dimensional variable library form a matrix [X G1 , Y G1 ; S22: Based on [X G1 , Y G1 calculate the elementary effect d i the absolute value S1 of the sample mean and the sample standard deviation S2 of (x), measure the influence degree of the variable on the performance index, eliminate the variables with insignificant influence, and incorporate the remaining variables into the second highest-dimensional variable library.

4. The collaborative optimization method for reducing the dimensionality of the diversion mechanism features according to claim 3, characterized in that Eliminate the variables with the absolute value of the sample mean S1 less than 5 and the variables with the sample standard deviation S2 in the range of [0, 5].

5. The collaborative optimization method for reducing the dimensionality of the guiding mechanism features according to claim 3, wherein In step S3, the dimensionality reduction processing is as follows: S31: Use the Latin hypercube sampling method to sample from the design space of each variable in the second high-dimensional variable library. All the sampled variable values form the second high-dimensional variable library sample sequence X for generating three-dimensional entities. G2 ; Input the second high-dimensional variable library sample sequence X G2 into the three-dimensional structure software to draw the three-dimensional runner and double-row cascade entities; Use the three-dimensional numerical simulation software to calculate the obtained entities to obtain the performance indicators Y of the second high-dimensional variable library. G2 ; The variable values and performance indicators sampled from the second high-dimensional variable library form the matrix [X G2 , Y G2 ; S32: Using the singular value decomposition theory, denote the number of variables in the second highest-dimensional variable library as n, the number of sampling times as m, and arrange the sampled variables into a matrix, denoted as the sample matrix input m×n ; Decompose the sample matrix input m×n into the left singular matrix U m×m , the eigenvalue matrix S m×n and the transpose of the right singular matrix V n×n . S33: According to the determined truncation value k j perform truncation, where k j < n, and the specific truncation method is as follows: Retain the first k m×m columns of U j to obtain the truncated left singular matrix Retain the first k m×n columns and the first k j rows of S j to obtain the truncated eigenvalue matrix Retain the first k n×n columns of V j to obtain the truncated right singular matrix S34: Reconstruct the output matrix output using the truncated left singular matrix eigenvalue matrix and the transpose of the right singular matrix ; m×n ; S35: Calculate the truncation error error: S36: If the truncation error error is greater than a preset threshold, reselect the truncation value k j , and repeat steps S33 to S35 until the truncation error error is less than or equal to the preset threshold; the threshold is selected from 0.01 to 0.05; S37: Take the truncated right singular matrix mentioned above as the projection coefficient matrix, and project the sample sequence X G2 in the second high-dimensional variable library onto the new low-dimensional design space to obtain a low-dimensional variable library: In the formula is the sample matrix in the low-dimensional variable library, which is composed of m sample sequences X j ' after dimensionality reduction to k G2 dimensions.

6. The method for collaborative optimization of the characteristics of the diversion mechanism according to claim 5, characterized in that Performance index Y G1 and Y G2 both include the cavitation coefficient σ under pump operating conditions p , the efficiency η under pump operating conditions p , the efficiency η under turbine operating conditions t and the non-vane region pressure pulsation NVPP under turbine operating conditions t These are four performance indices 7. The collaborative optimization method for reducing the dimensionality of the diversion mechanism features according to claim 6, characterized in that, In step S4, it is specifically as follows: S41: Construct a surrogate model between [X G2 ’, Y G2 , where the sample sequence X G2 ’ of the variables in the low-dimensional variable library is used as the input variable, and the cavitation coefficient σ p under the pump operating condition, the efficiency η p under the pump operating condition, the efficiency η t under the turbine operating condition, and the non-blade zone pressure pulsation NVPP t under the turbine operating condition are used as the output variables; S42: Genetically optimize the proxy model to obtain the Pareto front X G2 ’_pareto of the low-dimensional variable library as the optimal solution.

8. The method for collaborative optimization of the characteristics of the flow guiding mechanism according to claim 7, wherein In step S5, the inverse mapping of the optimal solution back to the original design space is specifically as follows: where: input_back m×n is the sample matrix after the inverse mapping; The sample matrix input_back after inverse mapping m×n The variable X in G2 _pareto is used as the optimized configuration parameter.

Images