A method, equipment, and medium for optimizing wind farm layout with multiple solutions based on clustering annealing elite search.

Abstract

This application discloses a method, device, and medium for optimizing wind farm layouts with multiple solutions based on clustering annealing elite search, relating to the field of wind power generation technology. The method includes: constructing an objective function based on wind farm optimization parameters, with the goal of maximizing normalized annual power generation and the constraints stated therein; randomly generating an initial population composed of multiple initial wind farm layouts; using a clustering annealing elite search algorithm, iteratively updating the initial population based on the objective function until a preset termination condition is met, obtaining a target population; dynamically dividing the target population into multiple clusters using a k-means clustering algorithm, obtaining multiple target clusters; within each target cluster, selecting the wind farm layout with the highest objective function value as the elite layout of the corresponding target cluster, obtaining the optimal layout set. This application outputs multiple wind farm layout schemes with different structures but superior performance in a single operation, improving the decision-making flexibility and engineering adaptability of wind farm layout optimization.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and in particular to a method, equipment and medium for optimizing the layout of wind farms with multiple solutions based on clustering annealing elite search. Background Technology

[0002] Optimizing wind farm layout is a key technological aspect of achieving efficient wind energy development and utilization. Its core is to maximize the annual power generation of the entire wind farm by scientifically planning the precise geographical location of each wind turbine within a given geographical boundary and a fixed number of turbines. The technical challenge stems from complex aerodynamic effects: the "wake" generated by upstream turbines significantly reduces the incoming wind speed and mass of downstream turbines, leading to a decrease in overall power output. Therefore, the quality of the wind farm layout directly determines the project's economic benefits.

[0003] To solve this problem, various optimization methods have been developed, mainly including metaheuristic algorithms, local search algorithms, and hybrid and intelligent algorithms. However, these algorithms generally follow the "single optimal solution" paradigm, meaning that no matter how the algorithm is improved, its design goal and output are to guide the search process to converge to a single, optimal or near-optimal layout scheme. This sets the problem as having only one globally optimal solution, leading to the following key drawbacks that cause serious contradictions with real-world engineering needs: Defect 1: Ignoring the inherent multimodal characteristics of the solution space. Due to the nonlinearity of wake interactions, the statistical distribution of wind direction, and the discreteness of wind turbine locations, the response surface of annual power generation in the wind farm layout optimization problem actually exhibits multimodal characteristics. This means that in the feasible solution space, there are objectively multiple locally optimal solutions with different structures but similar performance (i.e., multiple high-performance layout schemes). Existing techniques aim to find one of these peaks, while ignoring other equally important feasible schemes, failing to fully explore and utilize the full potential of the solution space.

[0004] Defect 2: It cannot provide decision-making flexibility for complex engineering constraints and has poor robustness. In actual wind farm development, the final implementation plan needs to meet a large number of non-aerodynamic constraints that are difficult to accurately model or foresee in the early optimization stage, such as: seabed geological conditions of offshore wind farms; spatial control restrictions such as waterways, ecological protection zones, and military zones; the huge difference between cable laying costs and engineering feasibility; and accessibility requirements for operation and maintenance.

[0005] When the "single optimal solution" obtained based on relevant technologies fails due to any of the above-mentioned real-world constraints, decision-makers will face the dilemma of a depleted "solution library" and will be forced to restart optimization or accept a suboptimal solution, resulting in wasted preliminary design work and increased project costs and risks.

[0006] While methods such as "multi-objective optimization" can generate a range of compromise solutions, the resulting diversity stems from trade-offs in the objective function space, such as the trade-off between cost and power generation, rather than the diversity of the decision space structure under a single power generation objective. Therefore, it cannot answer the crucial engineering question: "Given a predetermined power generation target, what are some completely different alternative layouts that can address various engineering obstacles?"

[0007] Therefore, it is necessary to provide a multi-solution wind farm layout optimization method based on clustered annealing elite search to solve the above problems. Summary of the Invention

[0008] The purpose of this application is to provide a method, device and medium for optimizing wind farm layout based on clustering annealing elite search, which outputs multiple wind farm layout schemes with different structures but superior performance in a single optimization, thereby improving the decision-making flexibility and engineering adaptability of wind farm layout optimization.

[0009] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a multi-solution wind farm layout optimization method based on clustered annealing elite search, wherein the multi-solution wind farm layout optimization method based on clustered annealing elite search includes: Obtain wind farm optimization parameters; wind farm optimization parameters include: wind resource data, wind turbine parameters, and constraints. Based on the wind farm optimization parameters, with the goal of maximizing normalized annual power generation and the constraints, an objective function is constructed. An initial population consisting of multiple initial wind farm layouts is randomly generated; each initial wind farm layout includes the planar coordinates of all wind turbines. The clustering annealing elite search algorithm is used to iteratively update the initial population based on the objective function until the preset termination condition is met, thereby obtaining the target population. The k-means clustering algorithm is used to dynamically divide the target population into multiple clusters, resulting in multiple target clusters; Within each target cluster, the wind farm layout with the highest objective function value is selected as the elite layout of the corresponding target cluster, thus obtaining the optimal layout set.

[0010] In one embodiment, the wind resource data includes: a joint frequency distribution of wind direction and wind speed; wherein the wind direction is discretized into multiple sectors, and the wind speed is discretized into multiple intervals from the cut-in wind speed to the cut-out wind speed. The wind turbine parameters include: wind turbine model, rotor diameter, hub height, thrust coefficient curve, and power curve; The constraints include boundary constraints and minimum spacing constraints; the boundary constraints are used to limit the feasible area of ​​the wind farm layout; the minimum spacing constraints are used to limit the minimum Euclidean distance between any two wind turbines.

[0011] In one embodiment, based on the wind farm optimization parameters, with the goal of maximizing annual power generation and the constraints, an objective function is constructed, specifically including: Using the Gaussian wake model, the downstream wind speed loss of a single wind turbine is calculated based on the rotor diameter and thrust coefficient curves in the turbine parameters. Calculate the incoming wind speed of each wind turbine based on the downstream wind speed loss of each wind turbine. Based on the power curves in the wind turbine parameters, the output power of each wind turbine under the current incoming wind speed is determined, and the output power of all wind turbines in the wind farm is summed to obtain the total power of the wind farm. Based on the joint frequency distribution of wind direction and wind speed in wind resource data, the total power of the wind farm is weighted and summed with the frequency of occurrence of the corresponding operating conditions to obtain the annual power generation. The annual power generation is normalized to obtain the normalized annual power generation; With the goal of maximizing normalized annual power generation and the aforementioned constraints, the objective function is constructed.

[0012] In one embodiment, the expression for normalized annual power generation is: ; in, Layout of wind farms Normalized annual power generation; For wind farm layout; The number of discrete sectors for wind direction; Index of wind direction sector; To cut off the wind speed; To cut in wind speed; This is the index for the discrete interval of wind speed; Layout of wind farms In wind direction Wind speed Total power under operating conditions; Wind direction Wind speed Frequency of occurrence of the operating condition; This represents the theoretical maximum annual power generation without considering wake losses.

[0013] In one embodiment, an initial population consisting of multiple initial wind farm layouts is randomly generated; each initial wind farm layout includes the planar coordinates of all wind turbines, specifically including: Multiple initial wind farm layouts are randomly generated to form an initial population, and a unique family identifier is assigned to each initial wind farm layout; each initial wind farm layout X is an N turb A matrix of size 2, where N turb Given the total number of wind turbines, each row of the matrix stores the planar coordinates of one wind turbine. All initial wind farm layouts satisfy the boundary constraints and minimum spacing constraints in the aforementioned constraints.

[0014] In one embodiment, when "using the clustering annealing elite search algorithm, based on the objective function, to iteratively update the initial population until a preset termination condition is met, thus obtaining the target population", the update process at any current iteration number specifically includes: Based on the geometric similarity of the wind farm layouts in the decision space in the population at the current iteration number, the k-means clustering algorithm is used to dynamically divide the population at the current iteration number into multiple clusters, resulting in multiple clusters at the current iteration number; among them, the number of clusters decreases with the iteration number according to the annealing plan; when the current iteration number is the initial iteration number, the population at the current iteration number is the initial population; Within each cluster at the current iteration number, the wind farm layout with the highest objective function value is selected as the elite layout of the corresponding cluster at the current iteration number, thus obtaining multiple elite layouts at the current iteration number. Assign or inherit a unique family identifier to each elite layout in the current iteration to track the evolutionary lineage of different layout families; The number of offspring is allocated according to the objective function value of each elite layout at the current iteration number, and the new generation population at the current iteration number is obtained according to each elite layout and the corresponding number of offspring at the current iteration number; the new generation population at the current iteration number has the same size as the population at the current iteration number. By utilizing local stochastic search optimization, the layout of each wind farm in the new generation population at the current iteration number is optimized through the objective function to obtain the updated population at the current iteration number; Determine whether the updated population at the current iteration number meets the preset termination condition; If not, then the updated population at the current iteration number will be used as the population at the next iteration number, and the next iteration will proceed. If so, the updated population at the current iteration number will be determined as the target population.

[0015] In one implementation, the number of clusters decreases with each iteration according to the following annealing scheme: ; in, For the first Number of clusters in the number of iterations; It is the initial number of clusters; It is the minimum number of clusters; It is the rate at which the speed of transition from exploration to development is adjusted; This represents the number of iterations.

[0016] In one embodiment, the preset termination condition is that the cumulative running time of the iterative updates reaches the preset total computation time.

[0017] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for optimizing the layout of wind farms based on clustering annealing elite search.

[0018] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned multi-solution wind farm layout optimization method based on clustering annealing elite search.

[0019] According to the specific embodiments provided in this application, this application has the following technical effects: This application discloses a method, device, and medium for optimizing wind farm layouts with multiple solutions based on clustered annealing elite search. By dynamically clustering, structurally similar layouts are physically isolated into multiple clusters. Elite layouts are independently retained within each cluster, and their lineage is tracked. This allows different layout patterns to evolve in parallel without mutually exclusive evolution, ultimately outputting multiple elite layouts from different clusters. This generates multiple structurally distinct and high-performance wind farm layout schemes in a single optimization, overcoming the limitation of related technologies that can only output a single optimal solution. When encountering constraints in actual engineering, such as geological conditions, environmental restrictions, and cable laying costs that are difficult to model in advance, decision-makers can select a scheme that meets the actual constraints from this set of optimal layouts, thereby significantly improving the decision-making flexibility and engineering adaptability of wind farm layout optimization. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A schematic diagram of the multi-solution wind farm layout optimization method based on clustering annealing elite search provided in an embodiment of this application; Figure 2A schematic diagram of the power curve and thrust coefficient curve of an NREL 5 MW wind turbine provided for an embodiment of this application; Figure 3 This is a schematic diagram of a wind rose example provided in one embodiment of this application; Figure 4 This is a schematic diagram of the iteration curves for various wind rose cases provided in an embodiment of this application; Figure 5 A schematic diagram comparing the annual power generation of wind farm layout based on the method of this application and the genetic random search algorithm in case R1, provided as an embodiment of this application; Figure 6 A schematic diagram illustrating the sensitivity of normalized annual power generation to the ratio of initial to final cluster number under the C1, R0, R1, and R2 cases provided in an embodiment of this application; Figure 7 A schematic diagram illustrating the sensitivity of normalized annual power generation to population size under the C1, R0, R1, and R2 cases provided in an embodiment of this application; Figure 8 A schematic diagram illustrating the sensitivity of normalized annual power generation to local search time under the C1, R0, R1, and R2 cases provided in an embodiment of this application; Figure 9 A schematic diagram illustrating the sensitivity of normalized annual power generation (AEP) to degradation rate under the C1, R0, R1, and R2 cases provided in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0022] 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 some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] This application provides a multi-solution wind farm layout optimization method based on clustering annealing elite search, which is executed by a processor in a computing device. This method aims to generate multiple high-quality wind farm layout schemes that differ in the decision space (i.e., layout structure) but exhibit similar optimal performance on the objective function in a single optimization run, providing a flexible pool of alternative solutions for practical engineering decisions. Traditional wind farm layout optimization methods typically only seek a single globally optimal solution, lacking flexibility when facing unforeseen constraints in actual deployment (such as geological conditions, cable routing, environmental factors, and regulatory restrictions). This application, by introducing dynamic clustering, elite preservation, and lineage tracking mechanisms, effectively maintains and utilizes population diversity during the optimization process.

[0024] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] In one exemplary embodiment, such as Figure 1 As shown, a multi-solution wind farm layout optimization method based on clustered annealing elite search is provided, including the following steps: Wherein: Step S1: Obtain wind farm optimization parameters; wind farm optimization parameters include: wind resource data, wind turbine parameters, and constraints.

[0026] As an optional implementation, the wind resource data includes: a joint frequency distribution of wind direction and wind speed; wherein the wind direction is discretized into multiple sectors, and the wind speed is discretized into multiple intervals from the cut-in wind speed to the cut-out wind speed.

[0027] The wind turbine parameters include: wind turbine model, rotor diameter, hub height, thrust coefficient curve, and power curve.

[0028] The constraints include boundary constraints and minimum spacing constraints; the boundary constraints are used to limit the feasible area of ​​the wind farm layout; the minimum spacing constraints are used to limit the minimum Euclidean distance between any two wind turbines.

[0029] Specifically, wind resource data consists of wind rose diagrams obtained from meteorological towers or reanalysis data (such as ERA5), which include different wind directions. and wind speed joint frequency distribution .wind direction Typically discretized as Each sector (e.g., when wind direction is discrete intervals) At that time, the number of wind sector areas Wind speed discretization is based on the cut-in wind speed. Cut-out wind speed Several intervals (such as wind speed discrete intervals) ).

[0030] Determining wind turbine parameters: A specific model of wind turbine is selected, such as the 5 MW (NREL 5MW) reference wind turbine from a certain country's National Renewable Energy Laboratory. Taking the NREL 5MW as an example, its rotor diameter is known. Hub height 90m, thrust coefficient curve and power curve .like Figure 2 As shown, the power curve defines the output power of the fan at a specific wind speed, and at the cut-in wind speed ( The following and cut-out wind speed ( The above output is zero.

[0031] The constraints include boundary constraints and minimum spacing constraints. The boundary constraints define the feasible region of the wind farm. It is usually a rectangular or polygonal area.

[0032] Minimum Spacing Constraint: The minimum Euclidean distance between any two wind turbines, typically set as follows: To prevent mechanical interference, i.e. , For the first Location coordinates of the typhoon generator For the first The location coordinates of the typhoon turbines, and the layout of the entire wind farm, are represented by Cartesian coordinates.

[0033] Step S2: Based on the wind farm optimization parameters, construct an objective function with the goal of maximizing normalized annual power generation and the constraints. As an optional implementation method, step S2 specifically includes: Step S21: Using the Gaussian wake model, the downstream wind speed loss of a single wind turbine is calculated based on the rotor diameter and thrust coefficient curves in the turbine parameters.

[0034] Step S22: Calculate the incoming wind speed of each wind turbine based on the downstream wind speed loss of each wind turbine.

[0035] Step S23: Based on the power curve in the wind turbine parameters, determine the output power of each wind turbine under the current incoming wind speed, and sum the output power of all wind turbines in the wind farm to obtain the total power of the wind farm.

[0036] Step S24: Based on the joint frequency distribution of wind direction and wind speed in the wind resource data, the total power of the wind farm is weighted and summed with the frequency of occurrence of the corresponding operating conditions to obtain the annual power generation.

[0037] Step S25: Normalize the annual power generation to obtain normalized annual power generation.

[0038] Step S26: With the goal of maximizing the normalized annual power generation and the constraints mentioned above, the objective function is constructed.

[0039] As an optional implementation, in step S26, the expression for the normalized annual power generation is: (1) in, Layout of wind farms Normalized annual power generation; For wind farm layout; The number of discrete sectors for wind direction; Index of wind direction sector; To cut off the wind speed; To cut in wind speed; This is the index for the discrete interval of wind speed; Layout of wind farms In wind direction Wind speed Total power under operating conditions; Wind direction Wind speed Frequency of occurrence of the operating condition; This represents the theoretical maximum annual power generation without considering wake losses.

[0040] Specifically, the wake model for a single wind turbine is defined: the downstream wind speed loss is calculated using a Gaussian wake model. For wind turbines located upstream... At the downstream point Speed ​​loss caused by for: (2) in, For incoming air velocity; downstream point to the fan The projection of the distance in the direction of the incoming flow; For this downstream point The vertical distance to the centerline of the wake; The wake width increases linearly with the downstream distance, satisfying the following equation (3): (3) in, The rotor diameter; The wake attenuation coefficient is related to the ambient turbulence intensity. Relatedly, in this embodiment, a constant is used. , , It is related to the thrust coefficient The relevant parameters satisfy the following equation (4): (4) Indicates the distance downstream of the wind turbine d The maximum reduction in wind speed along the centerline of the wake is expressed as: (5) Specifically, calculate the combined wake effect of multiple wind turbines: for downstream wind turbines Its incoming wind speed For free-flowing wind speed Subtract all upstream wind turbines The square root of the sum of the squares of the velocity deficit caused by its location is given by the following formula: (6) in, This represents the total number of all upstream wind turbines.

[0041] Specifically, for a given wind direction and wind speed Wind conditions, in the layout The total power of a wind farm is the sum of the output power of all wind turbines, as expressed below: (7) in, This represents the total number of wind turbines. For wind farm layout in In wind direction Wind speed Incoming air velocity under operating conditions This refers to the output power of a single fan.

[0042] Annual Energy Production (AEP) is obtained by frequency-weighted summation of power under all wind direction and wind speed conditions, and then normalized to obtain the normalized annual energy production as shown in equation (1).

[0043] Step S3: Randomly generate an initial population consisting of multiple initial wind farm layouts; each initial wind farm layout includes the planar coordinates of all wind turbines.

[0044] As an optional implementation, step S3 specifically includes: Step S31: Randomly generate multiple initial wind farm layouts to form an initial population, and assign a unique family identifier to each initial wind farm layout; each initial wind farm layout X is an N turb A matrix of size 2, where N turb Given the total number of wind turbines, each row of the matrix stores the planar coordinates of one wind turbine.

[0045] Step S32: All initial wind farm layouts satisfy the boundary constraints and minimum spacing constraints in the aforementioned constraints.

[0046] Specifically, randomly generated An initial wind farm layout constitutes the initial population. Layout of each wind farm It is The matrix ( (Total number of wind turbines), storing the planar coordinates of all wind turbines. And assign a unique family identifier to each wind turbine, i.e. , This represents the transpose of the matrix, which also implicitly contains information about the number of wind turbines. Furthermore, all initial wind farm layouts must satisfy defined boundary constraints and minimum spacing constraints to ensure feasibility.

[0047] For the current population Each wind farm layout Calculate its objective function value, i.e., the normalized annual power generation. .

[0048] Step S4: Using the Clustering Annealing Elite Search (CAES) algorithm, the initial population is iteratively updated based on the objective function until the preset termination condition is met, thus obtaining the target population.

[0049] As an optional implementation, in step S4, when "using the clustering annealing elite search algorithm, based on the objective function, iteratively updating the initial population until a preset termination condition is met to obtain the target population", the update process at any current iteration number specifically includes: Step S41: Based on the geometric structural similarity of each wind farm layout in the decision space in the population at the current iteration number, the k-means clustering algorithm is used to dynamically divide the population at the current iteration number into multiple clusters, resulting in multiple clusters at the current iteration number; wherein, the number of clusters decreases with the iteration number according to the annealing plan; when the current iteration number is the initial iteration number, the population at the current iteration number is the initial population.

[0050] Step S42: Within each cluster at the current iteration number, select the wind farm layout with the highest objective function value as the elite layout of the corresponding cluster at the current iteration number, thereby obtaining multiple elite layouts at the current iteration number.

[0051] Step S43: Assign or inherit a unique family identifier for each elite layout in the current iteration number in order to track the evolutionary lineage of different layout families.

[0052] Step S44: Allocate the number of offspring according to the objective function value of each elite layout under the current iteration number, and obtain the new generation population under the current iteration number according to each elite layout and the corresponding number of offspring under the current iteration number; the new generation population under the current iteration number has the same size as the population under the current iteration number.

[0053] Step S45: Utilize local stochastic search optimization to optimize the layout of each wind farm in the new generation population at the current iteration number through the objective function, thereby obtaining the updated population at the current iteration number.

[0054] Step S46: Determine whether the updated population at the current iteration number meets the preset termination condition.

[0055] Step S47: If not, use the updated population of the current iteration as the population of the next iteration and proceed with the next iteration.

[0056] Step S48: If yes, then the updated population at the current iteration number is determined as the target population.

[0057] As an optional implementation, in step S41, the number of clusters decreases with each iteration according to the following annealing scheme: (8) in, For the first Number of clusters in the number of iterations; This is the initial number of clusters (e.g., 64); It is the minimum number of clusters (e.g., 13); It is the decay rate (e.g., 1) that adjusts the speed of the transition from exploration to development. The number of iterations is denoted by . This strategy allows the algorithm to maintain a large number of clusters in the early stages to broadly explore the search space, while in the later stages it focuses on a few of the most promising clusters for in-depth development.

[0058] As an optional implementation, the preset termination condition is that the cumulative running time of the iterative updates reaches the preset total computation time.

[0059] Specifically, a dynamic clustering and annealing mechanism is applied to the population, and intra-cluster elite selection and lineage tracing are performed, as follows: 1) Dynamic clustering: Using the k-means clustering algorithm, based on the geometric similarity of the coordinate vectors of the wind farm layout in the decision space, the current population is grouped... Divided into Cluster This operation groups layouts with similar structures into a single category, with each cluster representing a potential wind farm layout "pattern" or "family".

[0060] 2) Annealing strategy: Number of clusters It's not fixed, but rather varies over time (optimization algebra). The annealing process is reduced according to the annealing plan, as shown in equation (8).

[0061] 3) Selection of elites within the cluster: In each cluster Internally, the layout with the highest AEP (Advanced Per Element) is selected as the "elite" of this cluster. : (9) in, The cluster index is obtained by performing clustering using the k-means clustering algorithm.

[0062] 4) Genealogical tracing: For each newly selected elite Assign or inherit a unique family identifier. If the elite is appearing for the first time, assign a new family identifier; if it is a "descendant" of a previous elite (generated through local search), inherit its parent's family identifier. This step is used to explicitly track the evolutionary history and survival status of different layout "families" throughout the optimization process. Its role is to quantify and monitor the diversity level of the population and prevent the search from converging to a single pattern too early.

[0063] 5) Based on elite-based population regeneration, a new generation of population is generated: Utilize the selected elites Collective to build the next generation of population At the same time, maintain population size Unchanged. Elite They will be replicated a certain number of times to populate the new population, and each elite will be allocated a certain number of offspring. The number of offspring is directly proportional to the fitness of the cluster it belongs to (average AEP or the AEP of the elite itself), meaning that the better the performance of the elite, the more offspring it contributes. Specifically, this can be allocated using the following formula: (10) Or make appropriate integer adjustments to satisfy the following equation (11): (11) Subsequently, it was filled with copies of these elites or their slightly perturbed variants. That is, the new generation population = all cluster elites + their clones.

[0064] 6) Perform local random search optimization on each individual in the new generation population: For the new generation of populations For each individual layout, the processor executes a local, constrained stochastic search process to attempt to improve its annual power generation (AEP), i.e., within a given local computation time. Within the loop, perform the following operations: randomly select a fan in the layout; randomly generate a movement direction and step size; attempt to move the fan to a new position and check if the new position violates the boundary constraints or minimum spacing constraints; if the new layout after the move has a higher AEP, accept the move and update the layout; otherwise, reject the move and retain the original layout.

[0065] 7) Determine whether the optimization termination condition is met. Determine whether the preset termination conditions have been met. The termination conditions in this application are based on the total computation time. Instead of relying on convergence metrics, this ensures fairness in the allocation of computational resources across cases of varying complexity. If the termination condition is not met, i.e., the running time has elapsed... Then let Then return to perform dynamic clustering and enter the next round of optimization iteration.

[0066] Step S5: The k-means clustering algorithm is used to dynamically divide the target population into multiple clusters, resulting in multiple target clusters.

[0067] Step S6: Within each target cluster, select the wind farm layout with the highest objective function value as the elite layout of the corresponding target cluster, thereby obtaining the optimal layout set.

[0068] Specifically, it outputs multiple sets of optimal layouts and their lineage information: when the preset termination condition is met, it outputs the final result. At this point, the result is not a single layout, but a set of multiple high-performance layouts from different clusters selected from the final population. (The elite of each final cluster, M≤ Furthermore, this method can provide the phylogenetic history of these layouts, showcasing the diversity and origins of alternatives to decision-makers.

[0069] Furthermore, to verify the effectiveness of the proposed multi-solution wind farm layout optimization method based on clustered annealing elite search, this embodiment compares the method with a related technology (Genetic Random Search (Genetic RS)) under the same hardware platform and parallel settings. The experiments were conducted under the same total computation time (50,000 seconds) and initialization conditions. Figure 3 Optimize the five typical wind rose cases (C0, C1, R0, R1, R2) shown. Figure 3 (a) in the text is case C0, with operating conditions of 0° in one direction and a single wind speed of 12m / s; Figure 3 (b) in the text is case C1, with operating conditions of 12 m / s wind speed in all directions; Figure 3 (c) in the text represents the R0 case, Horns Rev 1 offshore wind farm (55.53°N, 7.91°E); Figure 3 (d) in the text represents case R1, Jaisalmer onshore wind farm (26.92°N, 70.90°E). Figure 3 (e) in the example is the R2 case, the Muppandal onshore wind farm (8.25°N, 77.59°E).

[0070] Experimental results show that the genetic random search algorithm outputs only one optimal layout per run, while the method in this application stably outputs at least six elite layouts from different clusters per run, forming an "optimal solution set". Specific quantitative results are as follows: a) Performance equivalence: such as Figure 4 As shown, in cases R0, R1, and R2, the difference between the normalized annual power generation (AEP) of the six elite layouts (i.e., Cluster-1 to Cluster-6) and the highest AEP value found by the genetic random search algorithm in a single run is less than 0.5%, indicating that all schemes in this solution set have top-level annual power generation AEP performance. Figure 4 In the figure, (a) is the iteration curve for case C0. Figure 4 In the figure, (b) is the iteration curve for case C1. Figure 4 In the figure, (c) represents the iteration curve for the R0 case. Figure 4 In the figure, (d) represents the iteration curve for case R1. Figure 4 In the equation (e), the iteration curve for the R2 case is represented.

[0071] b) Structural dissimilarity: Structural differences between layouts are quantified using Hungarian matching distance, such as Figure 5 As shown, Figure 5 This diagram illustrates the comparison of annual power generation of wind farm layouts using the proposed method and the genetic random search algorithm in the R1 case. The wind farm layouts output by the proposed method are represented by blue dots, while those output by the genetic random search algorithm are represented by gray dots. The orange bars represent the improvement rate of annual power generation. Figure 5 (a) to (f) in the diagrams show a comparison of the annual power generation of the first to sixth elite layouts in case R1, based on the method of this application and the genetic random search algorithm. It can be seen that in case R1, the average distance between each pair of wind farm layouts in the output solution set is significant. For example, the wind farm layouts of Cluster-1 and Cluster-6 exhibit complementary spatial arrangements, with distinctly different turbine distributions in the central areas, directly demonstrating the diversity of wind farm layout structures generated by the method of this application.

[0072] c) Superior performance: such as Figure 5 As shown, the annual power generation (AEP) of all layouts in this solution set is higher than or equal to the performance of a single optimal solution obtained by independently running the genetic random search algorithm.

[0073] The technical basis for the above effect lies in the dynamic clustering and annealing mechanism applied to the population, along with intra-cluster elite selection and lineage tracing. Dynamic clustering divides the population into groups based on spatial structure. Each cluster physically isolates and protects different layout patterns, preventing a single pattern from monopolizing the market; annealing plan control. From initial value decay to This enables an automatic transition from extensive exploration to focused development; intra-cluster elite selection and elite-based regeneration continuously allocate computing resources to the best representatives of each mode, ensuring that multiple modes are optimized in parallel to high performance, which results in the final population naturally retaining representative solutions of multiple high-performance peaks.

[0074] Furthermore, the method described in this application exhibits excellent robustness to core hyperparameters, with performance fluctuations far below the tolerance requirements of engineering applications. Sensitivity analysis of key control parameters yields the following quantification results: a) such as Figure 6 As shown, Figure 6 Figures (a) through (d) illustrate the sensitivity of normalized annual power generation to the initial / final cluster ratio for cases C1, R0, R1, and R2, respectively. When the initial / final cluster ratio varies significantly between 50% / 10% and 25% / 1%, the standard deviation of normalized annual power generation is generally below 0.002 in the four complex wind cases (C1, R0, R1, and R2), with the maximum fluctuation being less than 0.5%.

[0075] b) such as Figures 7-9 As shown, Figure 7 In the diagrams (a) to (d), the normalized annual power generation is sensitive to the ratio of the initial to the final cluster number under the C1, R0, R1, and R2 cases, respectively. Figure 8 In the diagrams (a) to (d), the normalized annual power generation sensitivity to population size is shown for cases C1, R0, R1, and R2, respectively. Figure 9 Figures (a) to (d) illustrate the sensitivity of normalized annual power generation to local search time for cases C1, R0, R1, and R2, respectively. It can be seen that the final AEP fluctuation is less than 2% regardless of changes in other key parameters such as population size (32-256), local search time (150-800 seconds), and decay rate (0.5-3.0). Under all test parameter configurations, the method in this application consistently outperforms the genetic random search algorithm.

[0076] The technical basis for this effect lies in the robustness inherent in the stable structure of the CAES framework. The "dynamic clustering and annealing mechanism" constitutes a smooth, adaptive diversity management loop, rather than relying on brittle parameter thresholds. Even if a parameter setting tends to converge prematurely, the clustering mechanism will force the retention of multiple clusters to maintain exploration potential; conversely, if overexploration occurs, the annealing mechanism will gradually shrink the focus. This structured diversity maintenance makes the algorithm's performance insensitive to precise parameter settings.

[0077] Furthermore, while achieving the aforementioned multimodal search and output capabilities, the additional computational overhead of the proposed method is negligible, and its computational efficiency is comparable to that of the genetic random search algorithm. Under the same hardware platform and parallel settings, comparing the single-generation runtime of the proposed method and the genetic random search algorithm, within a 300-second time budget per generation, the total additional overhead from the newly added core operation in the proposed method is only about 0.08 seconds, accounting for approximately 0.027% of the total computation time. Therefore, under a fixed total optimization budget, the two algorithms can complete the same number of effective optimization generations.

[0078] The technological basis for this effect lies in the optimized design of the computational cost distribution. All innovative mechanisms are executed after calculating the annual power generation of each layout in the population, and only perform lightweight operations on low-dimensional coordinate data, without increasing the number of calculations for the annual power generation AEP. This demonstrates that the improvement in multimodal capabilities is achieved through innovative algorithmic architecture rather than simply piling on computational resources.

[0079] Beneficial effects: The proposed method for optimizing wind farm layout based on clustered annealing elite search achieves significant comprehensive progress in the specific technical field of wind farm layout optimization through a series of technical improvements, including the introduction of dynamic clustering and annealing mechanisms, intra-cluster elite selection and lineage tracking, and elite-based population regeneration strategies. It can simultaneously provide multiple high-performance and structurally distinct layout schemes, exhibits strong robustness, and maintains high computational efficiency. It effectively solves the challenges posed by model uncertainty and subsequent constraint changes in practical engineering to wind farm layout, and has outstanding industrial practical value.

[0080] In one exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a multi-solution wind farm layout optimization method based on clustering annealing elite search.

[0081] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements a multi-solution wind farm layout optimization method based on clustering annealing elite search.

[0082] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 10As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a multi-solution wind farm layout optimization method based on clustered annealing elite search.

[0083] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0084] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0085] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0086] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based wind power generation logic devices, etc., but are not limited to these.

[0087] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0088] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A multi-solution wind farm layout optimization method based on clustering annealing elite search, characterized in that, The multi-solution wind farm layout optimization method based on clustered annealing elite search includes: Obtain wind farm optimization parameters; wind farm optimization parameters include: wind resource data, wind turbine parameters, and constraints. Based on the wind farm optimization parameters, with the goal of maximizing normalized annual power generation and the constraints, an objective function is constructed. An initial population consisting of multiple initial wind farm layouts is randomly generated; each initial wind farm layout includes the planar coordinates of all wind turbines. The clustering annealing elite search algorithm is used to iteratively update the initial population based on the objective function until the preset termination condition is met, thus obtaining the target population. The k-means clustering algorithm is used to dynamically divide the target population into multiple clusters, resulting in multiple target clusters; Within each target cluster, the wind farm layout with the highest objective function value is selected as the elite layout of the corresponding target cluster, thus obtaining the optimal layout set.

2. The multi-solution wind farm layout optimization method based on clustering annealing elite search according to claim 1, characterized in that, Wind resource data includes: the joint frequency distribution of wind direction and wind speed; where wind direction is discretized into multiple sectors, and wind speed is discretized into multiple intervals from cut-in wind speed to cut-out wind speed; The wind turbine parameters include: wind turbine model, rotor diameter, hub height, thrust coefficient curve, and power curve; The constraints include boundary constraints and minimum spacing constraints; the boundary constraints are used to limit the feasible area of ​​the wind farm layout; the minimum spacing constraints are used to limit the minimum Euclidean distance between any two wind turbines.

3. The multi-solution wind farm layout optimization method based on clustering annealing elite search according to claim 2, characterized in that, Based on the aforementioned wind farm optimization parameters, and with the goal of maximizing annual power generation, and constrained by the aforementioned conditions, an objective function is constructed, specifically including: Using the Gaussian wake model, the downstream wind speed loss of a single wind turbine is calculated based on the rotor diameter and thrust coefficient curves in the turbine parameters. Calculate the incoming wind speed of each wind turbine based on the downstream wind speed loss of each wind turbine. Based on the power curves in the wind turbine parameters, the output power of each wind turbine under the current incoming wind speed is determined, and the output power of all wind turbines in the wind farm is summed to obtain the total power of the wind farm. Based on the joint frequency distribution of wind direction and wind speed in wind resource data, the total power of the wind farm is weighted and summed with the frequency of occurrence of the corresponding operating conditions to obtain the annual power generation. The annual power generation is normalized to obtain the normalized annual power generation; With the goal of maximizing normalized annual power generation and the aforementioned constraints, the objective function is constructed.

4. The multi-solution wind farm layout optimization method based on clustered annealing elite search according to claim 3, characterized in that, The expression for normalized annual power generation is: ； in, Layout of wind farms Normalized annual power generation; For wind farm layout; The number of discrete sectors for wind direction; Index of wind direction sector; To cut off the wind speed; To cut in wind speed; This is the index for the discrete interval of wind speed; Layout of wind farms In wind direction Wind speed Total power under operating conditions; Wind direction Wind speed Frequency of occurrence of the operating condition; This represents the theoretical maximum annual power generation without considering wake losses.

5. The multi-solution wind farm layout optimization method based on clustering annealing elite search according to claim 2, characterized in that, An initial population consisting of multiple initial wind farm layouts is randomly generated; each initial wind farm layout includes the planar coordinates of all wind turbines, specifically: Multiple initial wind farm layouts are randomly generated to form an initial population, and a unique family identifier is assigned to each initial wind farm layout; each initial wind farm layout X is an N turb A matrix of size 2, where N turb Given the total number of wind turbines, each row of the matrix stores the planar coordinates of one wind turbine. All initial wind farm layouts satisfy the boundary constraints and minimum spacing constraints in the aforementioned constraints.

6. The multi-solution wind farm layout optimization method based on clustered annealing elite search according to claim 1, characterized in that, When "using the clustering annealing elite search algorithm, based on the objective function, iteratively updating the initial population until a preset termination condition is met to obtain the target population", the update process at any current iteration number specifically includes: Based on the geometric similarity of the wind farm layouts in the decision space in the population at the current iteration number, the k-means clustering algorithm is used to dynamically divide the population at the current iteration number into multiple clusters, resulting in multiple clusters at the current iteration number; among them, the number of clusters decreases with the iteration number according to the annealing plan; when the current iteration number is the initial iteration number, the population at the current iteration number is the initial population; Within each cluster at the current iteration number, the wind farm layout with the highest objective function value is selected as the elite layout of the corresponding cluster at the current iteration number, thus obtaining multiple elite layouts at the current iteration number. Assign or inherit a unique family identifier to each elite layout in the current iteration to track the evolutionary lineage of different layout families; The number of offspring is allocated according to the objective function value of each elite layout at the current iteration number, and the new generation population at the current iteration number is obtained according to each elite layout and the corresponding number of offspring at the current iteration number; the new generation population at the current iteration number has the same size as the population at the current iteration number. By utilizing local stochastic search optimization, the layout of each wind farm in the new generation population at the current iteration number is optimized through the objective function to obtain the updated population at the current iteration number; Determine whether the updated population at the current iteration number meets the preset termination condition; If not, then the updated population at the current iteration number will be used as the population at the next iteration number, and the next iteration will proceed. If so, the updated population at the current iteration number will be determined as the target population.

7. The multi-solution wind farm layout optimization method based on clustered annealing elite search according to claim 6, characterized in that, The number of clusters decreases with each iteration according to the following annealing schedule: ； in, For the first Number of clusters in the number of iterations; It is the initial number of clusters; It is the minimum number of clusters; It is the rate at which the speed of transition from exploration to development is adjusted; This represents the number of iterations.

8. The multi-solution wind farm layout optimization method based on clustered annealing elite search according to claim 1, characterized in that, The preset termination condition is that the cumulative running time of the iterative updates reaches the preset total computation time.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the multi-solution wind farm layout optimization method based on clustering annealing elite search as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the multi-solution wind farm layout optimization method based on clustering annealing elite search as described in any one of claims 1-8.

Images