A hydropower engineering powerhouse optimization method based on NSGAII-RF parameter prediction

Abstract

This application relates to the field of artificial intelligence technology in water conservancy and hydropower engineering, and in particular to an optimization method for hydropower plant based on NSGAII-RF parameter prediction. The method involves processing multi-dimensional sequence data of the underground power plant design input parameters to form an initial population. After multiple iterations and optimizations, an optimized set is obtained, which is then solved using NSGA-II. Through initialization, mutation, crossover, and selection, a Pareto optimal solution set that meets multi-objective optimization requirements is gradually generated. The Pareto solution set, learned using the random forest algorithm, is used to predict parameters based on a small number of user-input parameters. Based on fitness ranking, a parameterized design scheme for the superstructure is generated. Under the premise of satisfying various constraints, column nodes located in special positions or requiring avoidance areas are effectively excluded. By analyzing the reasonable overlap relationship of the column nodes, the model can accurately calculate the connection relationship between beams and columns, generating a structural beam system formed by these column nodes.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology in water conservancy and hydropower engineering, and in particular to an optimization method for hydropower plant based on NSGAII-RF parameter prediction. Background Technology

[0002] With the advancement of carbon neutrality goals, the design cycle of pumped storage power plants has been significantly shortened compared to traditional hydropower projects. However, the traditional design process for underground pumped storage powerhouses often faces problems such as inconsistencies between drawings and models, and excessively long modeling times. In addition, the complex structure of underground powerhouses, containing many irregular shapes, causes designers to spend a lot of time on repeated iterations.

[0003] Therefore, standardization in factory building design has become crucial for shortening project cycles. How to efficiently optimize the layout of the superstructure, standardize cross-platform information sharing of the substructure, and quickly generate standard-compliant construction drawings has become a pressing issue. The superstructure of the factory building consists of conventional components, but the combination of parameters between these components is highly uncertain. While the locations of large-volume concrete sections in the substructure are relatively fixed, the shapes of the components themselves are highly uncertain.

[0004] Therefore, in order to achieve digital and intelligent design of the factory building, optimization methods are used to study the optimization of the combination relationship of superstructure layout parameters. Summary of the Invention

[0005] The main objective of this application is to provide an optimization method for hydropower plant buildings based on NSGAII-RF parameter prediction. By processing multidimensional sequence data of the input parameters for underground power plant design, an initial population is formed. After multiple iterations and optimizations, the model obtains one or more optimal sets, which are then solved using NSGA-II. This algorithm, through initialization, mutation, crossover, and selection, progressively generates a Pareto optimal solution set that meets multi-objective optimization requirements. The Pareto solution set is learned using the Random Forest (RF) algorithm. Based on this Pareto solution set, and using the RF model, parameter prediction can be performed based on a small number of user-input parameters. Furthermore, based on fitness ranking, a parameterized design scheme for the superstructure can be generated. The proposed optimization model effectively excludes column nodes located in special positions or areas requiring avoidance, while satisfying various constraints. By analyzing the reasonable overlap relationships of column nodes, the model can accurately calculate the beam-column connection relationships, thereby generating a structural beam system formed by these column nodes.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] According to a first aspect of the present invention, the present invention claims protection for a method for optimizing hydropower plant structures based on NSGAII-RF parameter prediction, comprising:

[0008] Obtain the superstructure of the factory building to be optimized, optimize the layout of the superstructure, and generate multiple sets of mechanism layout sequence data using NSGA-II;

[0009] The RF algorithm is used to learn the generated mechanism layout sequence data to generate an RF parameter prediction model for the plant to be optimized.

[0010] Design a user interface to receive data input from users and call the RF parameter prediction model to generate layout scheme data;

[0011] The Revit software API is called to convert the layout scheme data generated by the RF parameter prediction model into a design model. Based on the design model, a recommended optimization scheme is obtained to optimize the design of the factory building to be optimized.

[0012] Furthermore, the step of acquiring the superstructure of the plant to be optimized, optimizing the layout of the superstructure using NSGA-II, and generating multiple sets of mechanism layout sequence data also includes:

[0013] The design input parameters of the superstructure of the building to be optimized are obtained, and the design input parameters are processed into multi-dimensional sequence data to form an initial population.

[0014] The initial population is iterated and optimized multiple times to obtain one or more optimized sets, which are then solved using NSGA-II to obtain multiple sets of facility layout sequence data.

[0015] Furthermore, the step of using the RF algorithm to learn the mechanism layout sequence data and generate an RF parameter prediction model for the plant to be optimized also includes:

[0016] The Pareto solution set generated by calling the Random Forest (RF) model using a Python script;

[0017] Based on the RF model, parameter prediction is performed according to the parameters input by the user to generate the RF parameter prediction model of the plant to be optimized, and the models are sorted according to fitness.

[0018] Furthermore, the step of calling the Revit software API to convert the layout scheme data generated by the RF parameter prediction model into a design model, and then optimizing the design of the factory to be optimized based on the recommended optimization scheme obtained from the design model, further includes:

[0019] The client software calls the Revit software API to generate the design model;

[0020] The RF parameter prediction model is invoked through the user interface. Based on the trained RF parameter prediction model, the input feature parameters are obtained through the user interface and used as the input values ​​of the RF parameter prediction model.

[0021] Based on the quantitative indicators in the data tags, the layout schemes are sorted and recommended, and the optimal ranked recommended scheme is automatically output for designers to make decision-making references and fine-tune parameters.

[0022] Furthermore, the step of obtaining the design input parameters of the superstructure of the building to be optimized, and processing the design input parameters into multidimensional sequence data to form an initial population, further includes:

[0023] A genetic algorithm is used to optimize the design input parameters of the superstructure of the building to be optimized, and design variable objectives are obtained. The design variable objectives include functional objectives, economic objectives, and safety objectives.

[0024] When the design variable objective is the water intake form, the unit rotation and water intake direction are obtained, and the water intake form constraint result is obtained based on different combinations of the unit rotation and water intake direction.

[0025] When the design variable target is the plant size, the unit spacing and main plant span are determined to meet the program setting range according to the hydropower project unit type, and the plant size constraint conditions are obtained.

[0026] When the design variable objective is the beam-column arrangement, the coordinate position of the column is obtained, and the column arrangement constraints are determined based on the center position of the circular pier shroud.

[0027] When the design variable objective is the column span distribution, the span data between columns is obtained, the maximum and minimum span limits of the column spans are determined, and the column span distribution layout constraints are obtained.

[0028] When the design variable target is beam size, the beam's cross-sectional height and width are obtained. According to the structural design code, the beam cross-sectional dimensions meet the minimum structural requirements. The beam constraint conditions are determined based on the beam's cross-sectional width, the ratio of cross-sectional height to width, and the span-to-depth ratio.

[0029] Under the constraints, the economic optimization objective is achieved through parameter combination tuning, and the number of columns to be arranged within the interval and the upper limit and minimum value of the number of columns that can be arranged are obtained.

[0030] By adjusting parameters within a range, combinations with fewer columns can be obtained, maximizing available space.

[0031] By optimizing the span distribution between columns to match appropriate beam and column dimensions, a multi-objective optimization equation set for the factory layout is obtained;

[0032] Based on a multi-objective function, the NSGA-II simulation of the natural selection process is used, and a set of constraints and boundary conditions are combined to encapsulate a multi-parameter input layout logic function. The layout logic function generates a single-unit unit structure column layout network through parameter-driven generation.

[0033] Based on the constraints, the beam-column connection relationship is calculated through the overlap relationship of the column nodes, a structural beam system generated by the association of column nodes is generated, and the beam-column component dimensions are initially estimated through the span data.

[0034] Based on the changes in the input parameters of the arrangement function, multiple combinations of input parameters are generated. According to the arrangement logic function, the input parameters are used as multidimensional sequence data to form an initial population.

[0035] Furthermore, the step of iterating and optimizing the initial population multiple times to obtain one or more optimized sets, and then using NSGA-II to solve for multiple sets of facility layout sequence data, also includes:

[0036] Call the population generation function to create a population containing a specified number of individuals, where the individuals are a random combination of parameters;

[0037] An evaluation function is defined to evaluate the fitness of each individual in the population. The fitness is obtained by calling the layout logic function and calculating the objective function value of the multi-objective analysis based on the layout return value, taking into account the impact of different parameter combinations on the engineering layout.

[0038] Different fitness evaluation components are set for the target number of columns, the target beam size, and the target floor height, and then integrated into a weighted overall fitness evaluation value.

[0039] During the optimization process, users can select the optimal single-target strategy or the optimal combination of combined target strategies through the user interface.

[0040] After setting the initial population, crossover and mutation methods are used to generate new offspring, and a combined mutation strategy is used to mutate the offspring.

[0041] For genes with a defined range of values ​​and relatively regular values, mutation is performed on a single gene using a defined boundary plus step size approach.

[0042] For genes with a definite enumeration value, a random mutation combination is generated by combining the enumeration value with a roulette wheel algorithm.

[0043] For genes with continuous parameter value distribution, a combined mutation method using Gaussian operator and Cauchy operator is employed.

[0044] In the individual mutation process of the genetic algorithm, a roulette wheel selection mechanism is used to perform dynamic selection in a probabilistic manner;

[0045] In the individual crossover stage, a simulated binary crossover strategy is used to combine the selected parents, simulating the genetic characteristics of traditional binary crossover in a continuous variable space.

[0046] After crossover and mutation are completed, all offspring individuals generated in the current generation replace the original population as the basis for the next generation population. The evolutionary process terminates with a fixed number of iterations or a convergence criterion.

[0047] After the optimization process is terminated, the individual with the highest fitness value is selected from the final population as the optimal layout solution for the current problem. Through the iterative optimization mechanism of this genetic algorithm, automatic search and optimal solution approximation in the multi-parameter design space can be achieved, and the optimal parameter combination scheme is finally obtained. Multiple sets of mechanism layout sequence data are obtained by solving with NSGA-II.

[0048] Furthermore, the design of the user interface, which receives user-input data and calls the RF parameter prediction model to generate layout scheme data, also includes:

[0049] Using some parameters as design anchors to drive multi-dimensional parameter combinations, intelligent generation and layout optimization of complex engineering spaces are performed;

[0050] Using key feature parameters as input targets, and combining semantic tags from historical data, we perform multi-parameter combination prediction and intelligent recommendation.

[0051] The NSGA-II multi-objective evolutionary algorithm generates more than a number of high-fitness placement parameter combinations. The fitness values ​​of the placement parameter combinations are sorted and design labels are added by combining manual experience annotations to form a training sample set.

[0052] A random forest algorithm is introduced to construct a layout parameter prediction model. The model learns from the training samples by integrating multiple decision trees and integrates the prediction results of each tree in a regression manner.

[0053] During the construction of each decision tree, some features are randomly selected from all feature parameters as candidate subsets, and at each split node, the optimal split feature and the corresponding split point are selected for data partitioning until the preset stopping condition is met.

[0054] The RF parameter prediction model was evaluated and its parameters were tuned using cross-validation.

[0055] The unit spacing at a preset distance is selected as the input feature parameter, and the trained RF parameter prediction model is called to predict the layout scheme, thereby verifying the application effect of the RF parameter prediction model in actual design.

[0056] This application relates to the field of artificial intelligence technology in water conservancy and hydropower engineering, and in particular to an optimization method for hydropower plant based on NSGAII-RF parameter prediction. The method involves processing multi-dimensional sequence data of the underground power plant design input parameters to form an initial population. After multiple iterations and optimizations, an optimized set is obtained, which is then solved using NSGA-II. Through initialization, mutation, crossover, and selection, a Pareto optimal solution set that meets multi-objective optimization requirements is gradually generated. The Pareto solution set, generated using the random forest algorithm, is used to predict parameters based on a small number of user-input parameters. Based on fitness ranking, a parameterized design scheme for the superstructure is generated. Under the premise of satisfying various constraints, column nodes located in special positions or requiring avoidance areas are effectively excluded. By analyzing the reasonable overlap relationship of the column nodes, the model can accurately calculate the beam-column connection relationship and generate a structural beam system formed by these column nodes. Attached Figure Description

[0057] Figure 1 A flowchart illustrating the workflow of a hydropower plant optimization method based on NSGAII-RF parameter prediction, as claimed in this application.

[0058] Figure 2 A schematic diagram of the automatically generated structural layout of a hydropower plant optimization method based on NSGAII-RF parameter prediction, as claimed in an embodiment of this application.

[0059] Figure 3 This is a schematic diagram illustrating the prediction accuracy of a hydropower plant optimization method based on NSGAII-RF parameter prediction, as claimed in an embodiment of this application. Detailed Implementation

[0060] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0061] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0062] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0063] According to a first embodiment of the present invention, the present invention claims protection for a method for optimizing hydropower plant structures based on NSGAII-RF parameter prediction, with reference to Figure 1 ,include:

[0064] Obtain the superstructure of the factory building to be optimized, optimize the layout of the superstructure, and generate multiple sets of mechanism layout sequence data using NSGA-II;

[0065] The RF algorithm is used to learn the generated mechanism layout sequence data to generate an RF parameter prediction model for the plant to be optimized.

[0066] Design a user interface to receive data input from users and call the RF parameter prediction model to generate layout scheme data;

[0067] The Revit software API is called to convert the layout scheme data generated by the RF parameter prediction model into a design model. Based on the design model, a recommended optimization scheme is obtained to optimize the design of the factory building to be optimized.

[0068] In this embodiment, a user interface is designed to receive user-input data. In this example, the designer inputs the unit spacing and the number of units as key parameters. Based on these key parameters, the background program introduces the ball valve hoisting hole and unit dimensions as variation boundaries to generate a layout combination with 17 input parameters, including water inlet method, hood radius, ball valve hoisting hole width, upstream width of the main plant, busbar tunnel direction, number of structural columns, and structural column spacing. The layout parameter combination is optimized using the NSGA-II method. Through evolution and multi-criteria fitness evaluation, nine sets of structural layout parameters with high fitness are obtained as recommended schemes. The RF parameter prediction model is then used to generate layout scheme data. Subsequently, the designer selects the recommended structural layout data in the user interface, thus obtaining a complete functional and structural layout scheme for the underground powerhouse.

[0069] Furthermore, the step of acquiring the superstructure of the plant to be optimized, optimizing the layout of the superstructure using NSGA-II, and generating multiple sets of mechanism layout sequence data also includes:

[0070] The design input parameters of the superstructure of the building to be optimized are obtained, and the design input parameters are processed into multi-dimensional sequence data to form an initial population.

[0071] The initial population is iterated and optimized multiple times to obtain one or more optimized sets, which are then solved using NSGA-II to obtain multiple sets of facility layout sequence data.

[0072] Furthermore, the step of using the RF algorithm to learn the mechanism layout sequence data and generate an RF parameter prediction model for the plant to be optimized also includes:

[0073] The Pareto solution set generated by calling the Random Forest (RF) model using a Python script;

[0074] Based on the RF model, parameter prediction is performed according to the parameters input by the user to generate the RF parameter prediction model of the plant to be optimized, and the models are sorted according to fitness.

[0075] Furthermore, the step of calling the Revit software API to convert the layout scheme data generated by the RF parameter prediction model into a design model, and then optimizing the design of the factory to be optimized based on the recommended optimization scheme obtained from the design model, further includes:

[0076] The client software calls the Revit software API to generate the design model;

[0077] The RF parameter prediction model is invoked through the user interface. Based on the trained RF parameter prediction model, the input feature parameters are obtained through the user interface and used as the input values ​​of the RF parameter prediction model.

[0078] Based on the quantitative indicators in the data tags, the layout schemes are sorted and recommended, and the optimal ranked recommended scheme is automatically output for designers to make decision-making references and fine-tune parameters.

[0079] Furthermore, the step of obtaining the design input parameters of the superstructure of the building to be optimized, and processing the design input parameters into multidimensional sequence data to form an initial population, further includes:

[0080] A genetic algorithm is used to optimize the design input parameters of the superstructure of the building to be optimized, and design variable objectives are obtained. The design variable objectives include functional objectives, economic objectives, and safety objectives.

[0081] When the design variable objective is the water intake form, the unit rotation and water intake direction are obtained, and the water intake form constraint result is obtained based on different combinations of the unit rotation and water intake direction.

[0082] When the design variable target is the plant size, the unit spacing and main plant span are determined to meet the program setting range according to the hydropower project unit type, and the plant size constraint conditions are obtained.

[0083] When the design variable objective is the beam-column arrangement, the coordinate position of the column is obtained, and the column arrangement constraints are determined based on the center position of the circular pier shroud.

[0084] When the design variable objective is the column span distribution, the span data between columns is obtained, the maximum and minimum span limits of the column spans are determined, and the column span distribution layout constraints are obtained.

[0085] When the design variable target is beam size, the beam's cross-sectional height and width are obtained. According to the structural design code, the beam cross-sectional dimensions meet the minimum structural requirements. The beam constraint conditions are determined based on the beam's cross-sectional width, the ratio of cross-sectional height to width, and the span-to-depth ratio.

[0086] Under the constraints, the economic optimization objective is achieved through parameter combination tuning, and the number of columns to be arranged within the interval and the upper limit and minimum value of the number of columns that can be arranged are obtained.

[0087] By adjusting parameters within a range, combinations with fewer columns can be obtained, maximizing available space.

[0088] By optimizing the span distribution between columns to match appropriate beam and column dimensions, a multi-objective optimization equation set for the factory layout is obtained;

[0089] Based on a multi-objective function, the NSGA-II simulation of the natural selection process is used, and a set of constraints and boundary conditions are combined to encapsulate a multi-parameter input layout logic function. The layout logic function generates a single-unit unit structure column layout network through parameter-driven generation.

[0090] Based on the constraints, the beam-column connection relationship is calculated through the overlap relationship of the column nodes, a structural beam system generated by the association of column nodes is generated, and the beam-column component dimensions are initially estimated through the span data.

[0091] Based on the changes in the input parameters of the arrangement function, multiple combinations of input parameters are generated. According to the arrangement logic function, the input parameters are used as multidimensional sequence data to form an initial population.

[0092] In this embodiment, a genetic algorithm is used to optimize multiple parameters, including the water inlet method, hood radius, ball valve hoisting hole width, unit left-side width, unit spacing, upper width, main plant total span, and busbar tunnel direction. These parameters are crucial for the design and layout of the project, directly affecting its efficiency and cost.

[0093] Assume the design variables include unit spacing, the proportion of structural joints in a single unit, the span of the main plant, the ratio of upper to lower space in the main plant, the location and width of busbar tunnels, the water inlet type, the location and size of ball valve lifting holes, the location and size of ball valve adjusting holes, the location and size of columns, the beam arrangement, and the beam cross-sectional dimensions. The layout objectives are categorized into functional, economic, and safety objectives.

[0094] Regarding the water inlet method, let's set... It is a water inlet type. To change the direction of the generator set (forward or reverse). The direction of water inflow (direction or angle). Then the water inflow method. According to and Different combinations have the following constraints:

[0095] If the unit is rotating in the forward direction and water is being introduced in the forward direction, then F=1:

[0096] (1)

[0097] If the unit is operating in reverse and water is being introduced in the forward direction, then =2:

[0098] (2)

[0099] If the unit is rotating forward and water is being introduced at an angle, then =3:

[0100] (3)

[0101] If the unit is operating in reverse and water is being introduced at an angle, then =4:

[0102] (4)

[0103] For the plant dimensions, the width of a single generating unit is the unit spacing *j*, and the span is the total span of the main plant *S*. Using (0,0) as the origin, the plant is divided in the width direction into a left width *j1*, a right width *j2*, an upper span *S1* above the origin, and a lower span *S2* below the origin. Based on the type of hydropower generating units, the unit spacing and the main plant span meet the program-defined range *j*. min j max S min S max Therefore, the following constraints can be obtained:

[0104] (5)

[0105] The beam-column layout should ensure that the columns avoid the opening area. Let x be the value of the column layout. i y i Let x and y be the positions of the i-th pillar, (x... circle ,y circle () is the center position of the circular windshield of the machine pier. To preset the safety radius, [x rectmin ,x rectmax ] and [y rectmin ,y rectmax Let x and y be the x and y ranges of the rectangular hole, respectively. Then the arrangement of the columns satisfies:

[0106] (6)

[0107] (7)

[0108] For the span distribution of the columns, s i Let be the span between the i-th column and the (i+1)-th column, and the column span must satisfy the maximum span s. max and minimum span s min Based on these constraints, the following layout constraints are obtained:

[0109] (8)

[0110] For beam dimensions, h i Let b be the cross-sectional height of the i-th beam. i Let s be the cross-sectional width of the i-th beam. i Let h be the span of the i-th beam. According to the structural design code, the beam cross-sectional dimensions should meet the minimum structural requirements. According to the reinforced concrete structure code, the beam cross-section h...i The height should not be less than twice the beam width to ensure sufficient rigidity. The beam's cross-sectional width is b. i It should not be less than 200mm. The ratio of the beam's section height to its width (h) i / b i The span should ideally be between 2.5 and 4, and the span-to-height ratio should not be less than 2.0. This yields the following constraints:

[0111] (9)

[0112] Under the above constraints, the economic optimization objective is achieved through parameter combination tuning. Assuming the columns avoid structural openings, the number of columns arranged within the interval is n, and the upper limit of the number of columns that can be arranged within the interval is n. max The minimum number of columns that can be placed is n. min By adjusting parameters within a range, combinations with fewer columns are obtained to maximize available space. By optimizing the span distribution between columns to match appropriate beam and column dimensions, the floor height of the i-th floor arrangement interval is H. i To maximize the space available for the installation of electromechanical pipelines.

[0113] Based on a multi-objective function, the NSGA-II (Non-Dominated Sorting Genetic Algorithm II) is used to solve the multi-objective optimization problem. By simulating the natural selection process, various constraints and boundary conditions are incorporated, and a multi-parameter input layout logic function is encapsulated. This layout logic function generates a single-unit unit structural column layout network through parameter-driven generation.

[0114] Based on constraints, column nodes falling within openings or avoidance areas, such as those of machine pier hoods, ball valve lifting holes, and busbar openings, are excluded. The beam-column connection relationship is calculated based on the overlap relationships of the column nodes, thereby generating a structural beam system associated with the column nodes. The dimensions of the beam-column components are then initially estimated using span data. Figure 2 As shown.

[0115] The layout function includes 10 input parameters: water inlet method, outer radius of the hood, width of the ball valve hoisting hole, center position of the ball valve hoisting hole, left-side width of the unit, widening of the side units, unit spacing, upper span of the main plant, total span of the main plant, and direction of the busbar tunnel. Based on variations in these input parameters, more than 105 combinations can be generated. The layout logic function treats these input parameters as multi-dimensional sequence data, forming an initial population.

[0116] Furthermore, the step of iterating and optimizing the initial population multiple times to obtain one or more optimized sets, and then using NSGA-II to solve for multiple sets of facility layout sequence data, also includes:

[0117] Call the population generation function to create a population containing a specified number of individuals, where the individuals are a random combination of parameters;

[0118] An evaluation function is defined to evaluate the fitness of each individual in the population. The fitness is obtained by calling the layout logic function and calculating the objective function value of the multi-objective analysis based on the layout return value, taking into account the impact of different parameter combinations on the engineering layout.

[0119] Different fitness evaluation components are set for the target number of columns, the target beam size, and the target floor height, and then integrated into a weighted overall fitness evaluation value.

[0120] During the optimization process, users can select the optimal single-target strategy or the optimal combination of combined target strategies through the user interface.

[0121] After setting the initial population, crossover and mutation methods are used to generate new offspring, and a combined mutation strategy is used to mutate the offspring.

[0122] For genes with a defined range of values ​​and relatively regular values, mutation is performed on a single gene using a defined boundary plus step size approach.

[0123] For genes with a definite enumeration value, a random mutation combination is generated by combining the enumeration value with a roulette wheel algorithm.

[0124] For genes with continuous parameter value distribution, a combined mutation method using Gaussian operator and Cauchy operator is employed.

[0125] In the individual mutation process of the genetic algorithm, a roulette wheel selection mechanism is used to perform dynamic selection in a probabilistic manner;

[0126] In the individual crossover stage, a simulated binary crossover strategy is used to combine the selected parents, simulating the genetic characteristics of traditional binary crossover in a continuous variable space.

[0127] After crossover and mutation are completed, all offspring individuals generated in the current generation replace the original population as the basis for the next generation population. The evolutionary process terminates with a fixed number of iterations or a convergence criterion.

[0128] After the optimization process is terminated, the individual with the highest fitness value is selected from the final population as the optimal layout solution for the current problem. Through the iterative optimization mechanism of this genetic algorithm, automatic search and optimal solution approximation in the multi-parameter design space can be achieved, and the optimal parameter combination scheme is finally obtained. Multiple sets of mechanism layout sequence data are obtained by solving with NSGA-II.

[0129] In this embodiment, the objective equation of the plant superstructure optimization model is listed. This objective equation aims to improve the economic efficiency of the design scheme through three main objectives: reducing the number of columns, reducing beam cross-sectional dimensions, and increasing the plant's net clearance. Specifically, the main plant span and unit spacing must be within the range set by the program, such as Smin ≤ S ≤ Smax and jmin ≤ j ≤ jmax. Furthermore, the maximum span smax and minimum span smin of the columns also need to be strictly controlled. Construction parameter constraints cover the location and avoidance requirements of the lifting holes, ensuring the smooth implementation of the structural design during actual construction. rcircle is the radius of the pier hood, and xcircle and ycircle are the coordinates of the pier hood's center. By calculating the distance between the column center coordinates and the hood's center coordinates, and using the radius as a criterion, columns and beams falling within the pier hood's range can be excluded. Under these constraints, the beam's cross-sectional height hi, beam cross-sectional width bi, and beam span si must all comply with the corresponding specifications and standards. To optimize the column layout while avoiding structural openings, the number of columns to be placed within a specific interval is n. The maximum number of columns that can be placed within this interval is nmax, and the minimum number is nmin. Furthermore, each floor containing the columns is designated Hi to meet the functional requirements of each floor.

[0130] By calling the population generation function, a population containing a specified number of individuals is created. In the population, the individuals are random combinations of parameters, thus ensuring the diversity and randomness of the population.

[0131] An evaluation function is defined to assess the fitness of each individual in the population. This fitness is calculated by calling the layout logic function and taking into account the impact of different parameter combinations on the project layout, thus arriving at a fitness value. For the evaluation of the fitness value, different fitness evaluation components are set for the objectives of column number, beam size, and floor height, and these are integrated into a weighted overall fitness evaluation value. During the optimization process, the program allows users to select the optimal combination of a single objective strategy or a combination of objective strategies through the user interface.

[0132] Based on the initial population, new offspring are generated using crossover and mutation methods. Due to the different ranges and physical meanings of the input parameters, a combined mutation strategy is employed. For genes with definite value ranges, such as the inter-unit spacing (range 21-24) with relatively regular values, mutation is performed on a single gene using a defined boundary and step size. For genes with definite enumerated values, random mutation combinations are generated using a combination of enumerated values ​​and a roulette wheel algorithm. For genes with continuous parameter value distributions, a combined mutation method using Gaussian and Cauchy operators is employed.

[0133] Gaussian mutation (GM) is an optimization strategy that uses normally distributed random numbers to act on the original position vector to generate new positions. Most mutation operators are distributed around the original positions, which is equivalent to performing a neighborhood search within a small range. Here, x represents the individual value before mutation, and mutation(x) represents the individual value after mutation. As shown in Equation 12, Gaussian mutation has good local search capabilities, but its ability to guide individuals out of local optimal solutions is weak, which is not conducive to global convergence.

[0134] (10)

[0135] Cauchy variation originates from the Cauchy distribution of a continuous probability distribution. Its main characteristics are a small peak at zero and a slow decrease from the peak to zero, resulting in a more uniform variation range. Here, x represents the individual value before mutation, and mutation(x) represents the individual value after mutation. The mutation formula is shown in 5-15, which represents a random number within the interval (0,1).

[0136] (11)

[0137] In the individual mutation process of the genetic algorithm, this paper employs a roulette wheel selection mechanism to dynamically select between Cauchy mutation and Gaussian mutation in a probabilistic manner. Cauchy mutation has a greater tail perturbation capability, which helps to escape local optima traps, while Gaussian mutation is suitable for local fine-grained search. By introducing an independent mutant mechanism, that is, retaining some mutant individuals that did not participate in the crossover in each generation, the diversity of the population and the global optimization capability of the algorithm can be further improved, thereby effectively mitigating the risk of the algorithm getting trapped in local optima.

[0138] In the individual crossover phase, a simulated binary crossover (SBX) strategy is employed to combine the selected parents. This crossover method simulates the genetic characteristics of traditional binary crossover in a continuous variable space, possessing the ability to retain superior genes and explore new solutions. The new individuals generated through this method not only retain some of the excellent traits of their parents but also expand the search space, significantly enhancing the diversity and evolutionary potential of the population.

[0139] After crossover and mutation, all offspring individuals generated in the current generation replace the original population, forming the basis of the next generation. The entire evolutionary process terminates with a fixed number of iterations or a convergence criterion. As iterations continue, the overall fitness level of the population gradually increases, and individual solutions tend towards the global optimum or a near-optimal solution that meets the accuracy requirements.

[0140] After the optimization process terminates, the individual with the highest fitness value is selected from the final population, which is the optimal placement solution for the current problem. Through the iterative optimization mechanism of this genetic algorithm, automatic search and optimal solution approximation in a multi-parameter design space can be achieved, ultimately obtaining the optimal parameter combination scheme. This satisfies the multiple optimization requirements of structural rationality, spatial layout efficiency, and performance indicators in engineering design, providing an efficient and feasible technical path for numerical optimization and intelligent decision-making of complex engineering systems.

[0141] Furthermore, the design of the user interface, which receives user-input data and calls the RF parameter prediction model to generate layout scheme data, also includes:

[0142] Using some parameters as design anchors to drive multi-dimensional parameter combinations, intelligent generation and layout optimization of complex engineering spaces are performed;

[0143] Using key feature parameters as input targets, and combining semantic tags from historical data, we perform multi-parameter combination prediction and intelligent recommendation.

[0144] The NSGA-II multi-objective evolutionary algorithm generates more than a number of high-fitness placement parameter combinations. The fitness values ​​of the placement parameter combinations are sorted and design labels are added by combining manual experience annotations to form a training sample set.

[0145] A random forest algorithm is introduced to construct a layout parameter prediction model. The model learns from the training samples by integrating multiple decision trees and integrates the prediction results of each tree in a regression manner.

[0146] During the construction of each decision tree, some features are randomly selected from all feature parameters as candidate subsets, and at each split node, the optimal split feature and the corresponding split point are selected for data partitioning until the preset stopping condition is met.

[0147] The RF parameter prediction model was evaluated and its parameters were tuned using cross-validation.

[0148] The unit spacing at a preset distance is selected as the input feature parameter, and the trained RF parameter prediction model is called to predict the layout scheme, thereby verifying the application effect of the RF parameter prediction model in actual design.

[0149] In this embodiment, during the layout design of underground powerhouses in hydropower projects, a small number of parameters are often used as design anchors to drive multi-dimensional parameter combinations, thereby achieving intelligent generation and layout optimization of complex engineering spaces. To meet the requirements of globally generative automated design driven by few or single parameters, key feature parameters need to be used as input targets, combined with semantic tags of historical layout data, to achieve prediction and intelligent recommendation of multi-parameter combinations.

[0150] To construct an efficient data-driven prediction mechanism, over 4000 high-fitness placement parameter combinations were first generated based on the NSGA-II multi-objective evolutionary algorithm. These combinations were then sorted by fitness values ​​and labeled with design tags based on human experience to form a training sample set. Building upon this, a Random Forest algorithm was introduced to construct the placement parameter prediction model. This method learns from the training samples by integrating multiple decision trees and integrates the prediction results of each tree using a regression approach, thereby improving prediction accuracy and the model's generalization ability.

[0151] During the construction of each decision tree, the algorithm randomly selects a subset of features from all feature parameters as candidate subsets. At each split node, it selects the optimal splitting feature and the corresponding splitting point for data partitioning until preset stopping conditions (such as maximum depth, minimum number of samples, etc.) are met. To further improve model performance, K-fold cross-validation is used to evaluate the model and fine-tune its parameters, ensuring its generalization ability on the test set. Figure 3 As shown, the accuracy of the optimized model can stably reach over 95%.

[0152] To verify the model's effectiveness in practical design, a unit spacing of 24m was selected as the input feature parameter, and the trained prediction model was used to predict the layout scheme. Among the multiple recommended combinations output by the model, the top-ranked data combinations showed a high degree of similarity to the manually designed data, demonstrating good spatial logic and engineering adaptability. This fully reflects the model's ability to predict multi-dimensional layout results under single-parameter driving.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0154] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0155] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A hydropower engineering plant optimization method based on NSGAII-RF parameter prediction, characterized in that, include: Obtain the superstructure of the factory building to be optimized, optimize the layout of the superstructure, and generate multiple sets of mechanism layout sequence data using NSGA-II; The RF algorithm is used to learn the generated mechanism layout sequence data to generate an RF parameter prediction model for the plant to be optimized. Design a user interface to receive data input from users and call the RF parameter prediction model to generate layout scheme data; The Revit software API is called to convert the layout scheme data generated by the RF parameter prediction model into a design model. Based on the design model, a recommended optimization scheme is obtained to optimize the design of the factory building to be optimized.

2. The method for predicting and optimizing the hydropower plant based on NSGAII-RF parameters according to claim 1, characterized in that, The process of acquiring the superstructure of the plant to be optimized, optimizing the layout of the superstructure using NSGA-II, and generating multiple sets of mechanism layout sequence data also includes: The design input parameters of the superstructure of the building to be optimized are obtained, and the design input parameters are processed into multi-dimensional sequence data to form an initial population. The initial population is iterated and optimized multiple times to obtain one or more optimized sets, which are then solved using NSGA-II to obtain multiple sets of facility layout sequence data.

3. The method for optimization of hydropower plant based on NSGAII-RF parameter prediction according to claim 1, characterized in that, The step of using the RF algorithm to learn the mechanism layout sequence data and generate an RF parameter prediction model for the plant to be optimized also includes: The Pareto solution set generated by calling the Random Forest (RF) model using a Python script; Based on the RF model, parameter prediction is performed according to the parameters input by the user to generate the RF parameter prediction model of the plant to be optimized, and the models are sorted according to fitness.

4. The method of Claim 1, wherein, The step of calling the Revit software API to convert the layout data generated by the RF parameter prediction model into a design model, and then optimizing the design of the factory to be optimized based on the recommended optimization scheme obtained from the design model, further includes: The client software calls the Revit software API to generate the design model; The RF parameter prediction model is invoked through the user interface. Based on the trained RF parameter prediction model, the input feature parameters are obtained through the user interface and used as the input values ​​of the RF parameter prediction model. Based on the quantitative indicators in the data tags, the layout schemes are sorted and recommended, and the optimal ranked recommended scheme is automatically output for designers to make decision-making references and fine-tune parameters.

5. The method for optimization of hydropower plant based on NSGAII-RF parameter prediction according to claim 2, characterized in that, The step of obtaining the design input parameters of the superstructure of the building to be optimized, and processing the design input parameters into multidimensional sequence data to form an initial population, further includes: A genetic algorithm is used to optimize the design input parameters of the superstructure of the building to be optimized, and design variable objectives are obtained. The design variable objectives include functional objectives, economic objectives, and safety objectives. When the design variable objective is the water intake form, the unit rotation and water intake direction are obtained, and the water intake form constraint result is obtained based on different combinations of the unit rotation and water intake direction. When the design variable target is the plant size, the unit spacing and main plant span are determined to meet the program setting range according to the hydropower project unit type, and the plant size constraint conditions are obtained. When the design variable objective is the beam-column arrangement, the coordinate position of the column is obtained, and the column arrangement constraints are determined based on the center position of the circular pier shroud. When the design variable objective is the column span distribution, the span data between columns is obtained, the maximum and minimum span limits of the column spans are determined, and the column span distribution layout constraints are obtained. When the design variable target is beam size, the beam's cross-sectional height and width are obtained. According to the structural design code, the beam cross-sectional dimensions meet the minimum structural requirements. The beam constraint conditions are determined based on the beam's cross-sectional width, the ratio of cross-sectional height to width, and the span-to-depth ratio. Under the constraints, the economic optimization objective is achieved through parameter combination tuning, and the number of columns to be arranged within the interval and the upper limit and minimum value of the number of columns that can be arranged are obtained. By adjusting parameters within a range, combinations with fewer columns can be obtained, maximizing available space. By optimizing the span distribution between columns to match appropriate beam and column dimensions, a multi-objective optimization equation set for the factory layout is obtained; Based on a multi-objective function, the NSGA-II is used to simulate the natural selection process. Various constraints and boundary conditions are combined, and a multi-parameter input layout logic function is encapsulated. The layout logic function generates a single-unit unit structure column layout network through parameter-driven generation. Based on the constraints, the beam-column connection relationship is calculated through the overlap relationship of the column nodes, a structural beam system generated by the association of column nodes is generated, and the beam-column component dimensions are initially estimated through the span data. Based on the changes in the input parameters of the arrangement logic function, multiple combinations of input parameters are generated. According to the arrangement logic function, the input parameters are used as multidimensional sequence data to form an initial population.

6. The method according to claim 5, wherein, The process of iterating and optimizing the initial population multiple times to obtain one or more optimized sets, and then using NSGA-II to solve for multiple sets of facility layout sequence data, further includes: Call the population generation function to create a population containing a specified number of individuals, where the individuals are a random combination of parameters; An evaluation function is defined to evaluate the fitness of each individual in the population. The fitness is obtained by calling the layout logic function and calculating the objective function value of the multi-objective analysis based on the layout return value, taking into account the impact of different parameter combinations on the engineering layout. Different fitness evaluation components are set for the target number of columns, the target beam size, and the target floor height, and then integrated into a weighted overall fitness evaluation value. During the optimization process, users can select the optimal single-target strategy or the optimal combination of combined target strategies through the user interface. After setting the initial population, crossover and mutation methods are used to generate new offspring, and a combined mutation strategy is used to mutate the offspring. For genes with a defined range of values ​​and relatively regular values, mutation is performed by defining the boundary of a single gene and then using a step size. For genes with a definite enumeration value, a random mutation combination is generated by combining the enumeration value with a roulette wheel algorithm. For genes with continuous parameter value distribution, a combined mutation method using Gaussian operator + Cauchy operator is employed; In the individual mutation process of the genetic algorithm, a roulette wheel selection mechanism is used to perform dynamic selection in a probabilistic manner; In the individual crossover stage, a simulated binary crossover strategy is used to combine the selected parents, simulating the genetic characteristics of traditional binary crossover in a continuous variable space. After crossover and mutation are completed, all offspring individuals generated in the current generation replace the original population as the basis for the next generation population. The evolutionary process terminates with a fixed number of iterations or a convergence criterion. After the optimization process is terminated, the individual with the highest fitness value is selected from the final population as the optimal layout solution for the current problem. Through the iterative optimization mechanism of this genetic algorithm, automatic search and optimal solution approximation in the multi-parameter design space can be achieved, and the optimal parameter combination scheme is finally obtained. Multiple sets of mechanism layout sequence data are obtained by solving with NSGA-II.

7. The method of Claim 1, wherein, The user interface design receives user-input data, calls the RF parameter prediction model to generate layout scheme data, and also includes: Using some parameters as design anchors to drive multi-dimensional parameter combinations, intelligent generation and layout optimization of complex engineering spaces are performed; Using key feature parameters as input targets, and combining semantic tags from historical data, we perform multi-parameter combination prediction and intelligent recommendation. Multiple sets of high-fitness layout parameter combinations are generated based on the NSGA-II multi-objective evolutionary algorithm. The fitness values ​​of the layout parameter combinations are sorted and design labels are added in combination with manual experience annotation to form a training sample set. A random forest algorithm is introduced to construct a layout parameter prediction model. The model learns from the training samples by integrating multiple decision trees and integrates the prediction results of each tree in a regression manner. During the construction of each decision tree, some features are randomly selected from all feature parameters as candidate subsets, and at each split node, the optimal split feature and the corresponding split point are selected for data partitioning until the preset stopping condition is met. The RF parameter prediction model was evaluated and its parameters were tuned using cross-validation. The unit spacing at a preset distance is selected as the input feature parameter, and the trained RF parameter prediction model is called to predict the layout scheme, thereby verifying the application effect of the RF parameter prediction model in actual design.

Images