CN-121997778-A - Gao Weibei phyllus wind field layout optimization method assisted by reversible neural network model

CN121997778ACN 121997778 ACN121997778 ACN 121997778ACN-121997778-A

Abstract

The invention belongs to the field of wind power, and discloses a Gao Weibei phyllos wind field layout optimization method assisted by a reversible neural network model, which comprises the following steps of step 1, modeling a scene to be laid out and generating a feasible layout scene; and step 2, designing a reversible neural network structure and a loss function with high-dimensional layout information compression, and step 3, performing Gao Weibei Yes wind field layout optimization assisted by the reversible neural network. The invention solves the problem that the traditional method either adopts an analytic wake model to sacrifice precision or adopts a heuristic algorithm to sacrifice efficiency, provides a high-efficiency framework based on Bayesian optimization, adopts a high-fidelity numerical wake model, and gives consideration to improvement of optimization efficiency and result precision.

Inventors

MENG WENCHAO
ZHOU YIHUAN
YANG QINMIN
ZHAO YUHANG
LI YIMING

Assignees

浙江大学

Dates

Publication Date: 20260508
Application Date: 20260409

Claims (9)

1. A Gao Weibei phyllus wind field layout optimization method assisted by a reversible neural network model is characterized by comprising the following steps: Step 1, modeling a scene to be laid out and generating a feasible layout scene: step 2, preprocessing high-dimensional layout information, and designing a structure and a loss function of a reversible neural network model; step 3, performing Gao Weibei phyllus wind field layout optimization assisted by a reversible neural network model, including: In the main optimization stage, embedding a reversible neural network model into an initialization stage in a forward propagation mode, in the initialization stage, uniformly sampling a plurality of layout schemes on a feasible layout data set through a Latin hypercube sampling technology, wherein the layout scheme characterization simultaneously obtains a corresponding low-dimensional characterization and a corresponding objective function value through forward propagation of the reversible neural network model and high-fidelity computational fluid dynamics simulation, thereby constructing an initial input-output pair low-dimensional characterization data set; in the iterative optimization process of the main optimization stage, a reversible neural network model is embedded in a reverse propagation mode, an acquisition function is optimized on a probability agent model constructed in a low-dimensional space, the next evaluation point is selected through balance exploration and utilization, then the evaluation point is paired with a corresponding objective function value, a low-dimensional characterization data set is filled in, the probability agent model is updated on the basis, and then the next iterative optimization is carried out until a stopping condition is met, and an optimal solution is output.
2. The method for optimizing the layout of Gao Weibei casserole wind fields assisted by the reversible neural network model according to claim 1, wherein the modeling of the scene to be laid out in the step 1 includes: determining a representation mode of the position of the fan in the wind field, and representing the position of the fan by using grid coordinates; In grid type characterization, a wind field to be laid out is divided into a plurality of grids, the side length of each grid depends on the rotor diameter D of a fan to be laid out by the wind field, and a set formed by M vertexes is recorded as Namely, the optional positions of N fans to be laid out, x i and y i respectively represent the cis-position of the points where the vertexes are respectively positioned in the directions of the x axis and the y axis, i is the number of the corresponding fan, Representing an integer space, the representation of each layout is written as a set of N vertices satisfying the constraint: , in the formula, A symbol representing a layout scheme; each possible layout must satisfy a minimum distance constraint between fans: , Where d ij represents the distance between fan i and fan j, and d min is the minimum distance between fans determined from the geographic location at the wind farm to be laid out and the selected fan model.
3. The method for optimizing the layout of Gao Weibei casserole wind fields assisted by the reversible neural network model according to claim 2, wherein the modeling of the scene to be laid out in the step 1 further comprises: the objective function defining wind farm layout optimization is annual energy production, AEP: , Wherein D year represents the number of days of the year, W represents the number of wind conditions considered in the wind farm, f j represents the frequency of occurrence of the jth wind condition, Representing the power generation power of the ith fan under the jth wind condition; Modeling the fan by using a brake disc model and calculating wake effects by combining a high-fidelity numerical model RANS, wherein the brake disc model is added into a standard RANS equation in the form of a source term, and the formula is as follows: , Wherein t is time, x i and x j are Cartesian space coordinates, subscripts i and j represent horizontal and vertical directions, respectively, And Representing time-averaged velocity and pressure, ρ and μ being air density and dynamic viscosity, f i representing the source term from the brake disk, respectively, and finally closing the Reynolds stress term using the k- ε model 。
4. A method for optimizing a layout of a Gao Weibei-leaf wind field assisted by a reversible neural network model according to claim 3, wherein the generating a feasible layout scene in the step 1 includes: based on the modeled scene to be laid out, using a backtracking method to automatically generate all feasible layouts under the scene to be laid out, and storing all the feasible layouts into a data set Is a kind of medium.
5. The method for optimizing the layout of Gao Weibei casserole wind fields assisted by the reversible neural network model according to claim 1, wherein the preprocessing of the high-dimensional layout information in the step 2 comprises the following steps: For the layout representation composed of the positions of N fans to be laid out, each vertex in the wind field divided by grids is allocated with a unique integer index, and each layout is represented by recording the index corresponding to the vertex where the fan is placed, so that the dimension of the layout representation is reduced from 2N to N; for the feasible layout data set generated in the step 1 Performing an intrinsic dimensional analysis; Data sets of feasible layout Normalizing to obtain normalized feasible layout data set 。
6. The method for optimizing a Gao Weibei-leaf wind field layout assisted by a reversible neural network model according to claim 5, wherein the reversible neural network model in step 2 includes a plurality of reversible coupling layers; the reversible neural network model learns a certain nonlinear transformation f θ , and inputs the same Mapping to specific hidden spaces of the same dimension And has an easily available inverse mapping The method comprises the following steps of: , Wherein, the Representation by reverse propagation The resulting output is then used to determine, Representing real space.
7. The method for optimizing the layout of Gao Weibei cass wind field assisted by the reversible neural network model according to claim 6, wherein the reversible coupling layer of the reversible neural network model in the step 2 is an affine coupling layer of RNVP, each affine coupling layer comprises 4 sub-networks for learning different nonlinear transformations, and the sub-networks adopt multi-layer perceptron; Output vector to reversible neural network model Performing dimension segmentation: , Wherein, the Representing the hidden variables of the low-dimensional representation, Is the number of intrinsic dimensions of the data set, Representing hidden variables of additional information, during training And (3) performing reverse propagation reconstruction after splicing with the all 0 vectors: , Wherein, the Representing a reconstruction without additional information, Representative and All 0 vectors of the same shape.
8. The method for optimizing a Gao Weibei-leaf wind field layout assisted by a reversible neural network model according to claim 6, wherein the design of the loss function of the reversible neural network model in step 2 includes: integrated loss function L consisting of 3 different loss terms: , in the formula, For the standard reconstruction loss, Represents the reconstruction penalty of reconstruction using only the partitioned low-dimensional representations, The residual loss of the extra information indicates the amount of the stored information in the divided extra dimension, wherein omega 1 、ω 2 and omega 3 are weight coefficients; Wherein, the The calculation mode of (a) is as follows: , in the formula, Representing the 2-norm of the data obtained, Representing the desire; Wherein, the The calculation mode of (a) is as follows: , Wherein, the The calculation mode of (a) is as follows: 。
9. The method for optimizing a Gao Weibei-leaf wind field layout assisted by a reversible neural network model according to claim 1, wherein the step 2 further includes: In the preprocessing stage, the reversible neural network model realizes compression and accurate reconstruction of high-dimensional layout information by pre-training on all feasible layout schemes, and after the training error converges, the reversible neural network model learns a deterministic bijective mapping, wherein the bijective comprises unijective and full-shot properties, and the reversible neural network model is stored at the moment.

Description

Gao Weibei phyllus wind field layout optimization method assisted by reversible neural network model Technical Field The invention relates to the field of wind power, in particular to a Gao Weibei Yes wind field layout optimization method assisted by a reversible neural network model. Background Wind energy plays an important role in global energy conversion as a clean renewable energy source. With the continuous expansion of the wind farm, in actual operation, wake effects inevitably exist between fans. Wake flow refers to the phenomenon of speed weakening and turbulence enhancement of airflow caused by an upstream fan in the running process, and the effect can obviously reduce the effective incoming flow wind speed of a downstream fan, increase load fluctuation and further influence the power output and the service life of equipment of the whole wind power plant. Therefore, wind farm layout optimization is an important topic in research and engineering practice in order to mitigate the adverse effects of wake effects. Through reasonably designing the spatial distribution of the fans in the field, the generated energy or income can be maximized under given wind resources and topography conditions, and various constraints are considered. Existing methods and research works can be divided into two classes according to the wake models used, one class is based on analytical wake models, such as Jensen models, gaussian models, etc. The model has the advantages of high calculation efficiency, less parameter requirement, easy integration into an optimization framework and the like by simplifying the description of speed attenuation and turbulence diffusion of wake flow through physical approximation and an empirical formula. Many studies have used gradient-based algorithms or intelligent heuristic algorithms to solve optimization problems by using such wake models to explicitly or implicitly model the objective function of the wind farm design, resulting in an optimal wind farm layout scheme. However, since such wake models are built on a strong simplifying assumption, it is difficult to accurately describe the heterogeneous inflow conditions and flow field characteristics under multi-wake superposition, and prediction accuracy is limited, which may result in a large difference between the final result and the actual situation. Another class of methods uses numerical wake models based on computational fluid dynamics, such as the ranolae Navier-Stokes equation, large vortex modeling, etc., to model wake interactions with high fidelity. The model can accurately and comprehensively describe the speed field, turbulence characteristics and evolution process of the wake flow of the fan, and has higher prediction precision. However, numerical models are described by partial differential equations controlling fluid behavior, it is difficult to explicitly construct a layout-optimized objective function based on the partial differential equations, and gradient-based algorithms cannot be directly used for solving such problems, in addition, research has been conducted to utilize heuristic algorithms for convergence, but the numerical models themselves are high in calculation cost, the calculation resources and time required for single simulation are expensive, and the time required for convergence can be extremely long due to the combination of computationally intensive heuristic algorithms. In conclusion, the analytical model is suitable for large-scale rapid optimization, but has limited precision, and the numerical model has high precision and excessive calculation load. How to balance between high-precision modeling and efficient optimization calculation becomes an important challenge in wind farm layout optimization research and application. Disclosure of Invention The invention aims to provide a Gao Weibei phyllus wind field layout optimization method assisted by a reversible neural network model so as to solve the problems in the background technology. In order to achieve the above purpose, the present invention provides the following technical solutions: a Gao Weibei phyllus wind field layout optimization method assisted by a reversible neural network model comprises the following steps: Step 1, modeling a scene to be laid out and generating a feasible layout scene: step 2, preprocessing high-dimensional layout information, and designing a structure and a loss function of a reversible neural network model; step 3, performing Gao Weibei phyllus wind field layout optimization assisted by a reversible neural network model, including: In the main optimization stage, embedding a reversible neural network model into an initialization stage in a forward propagation mode, in the initialization stage, uniformly sampling a plurality of layout schemes on a feasible layout data set through a Latin hypercube sampling technology, wherein the layout scheme characterization simultaneously obtains a corresponding low-dimensional cha