Wind power plant wake flow distribution simulation method and device based on bellows splicing and storage medium

Abstract

The invention discloses a wind power plant wake flow distribution simulation method and device based on air bellow splicing and a storage medium, and the method comprises the steps: simulating and generating a single fan wake flow sample based on an automatic simulation interface; constructing a single-fan wake flow model according to the single-fan wake flow sample; constructing a wind power plant wake flow model according to the single-fan wake flow model and the actual wind power plant distribution condition; based on the wind power plant wake flow model, simulation analysis is conducted on wake flow distribution of an actual wind power plant in an air bellow splicing mode, and the air bellow splicing mode is a mode that the actual wind power plant is divided into a plurality of spliced air bellow. According to the method, the simulation calculation time can be effectively shortened, so that the wind power plant wake flow distribution simulation analysis efficiency can be effectively improved; and the wind energy loss caused by the wake effect in the wind power plant can be effectively reduced, so that the overall output of the wind power plant can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method, a device and a storage medium for simulating the wake distribution of a wind farm based on wind box splicing. Background Art

[0002] As an important renewable energy source, wind energy has a rapid development momentum and broad development prospects around the world. Large-scale wind farms are often composed of dozens or even hundreds of wind turbines arranged side by side to form a network. The wind turbines at the upwind position absorb wind energy to do work, which will reduce the wind speed of the downstream wind turbines. This phenomenon is called the wake effect. The wake effect will greatly affect the power generation efficiency of the downstream wind turbines.

[0003] Existing wind farm wake distribution simulation methods are mainly divided into simulation and analysis of wind farm wake distribution based on computational fluid dynamics mechanism models. However, the computational fluid dynamics mechanism model requires fine discretization of the wind farm. After reading the specific environmental information and wind turbine information of the wind farm, CFD simulation is performed through a large amount of grid calculations to obtain the specific spatiotemporal wind distribution of the entire wind farm. As a result, the numerical simulation of the mechanism model requires extremely long calculation time, and the efficiency of wind farm wake distribution simulation analysis is low. Summary of the invention

[0004] The present invention provides a wind farm wake distribution simulation method, device and storage medium based on wind box splicing, so as to solve the technical problem that the existing wind farm wake distribution simulation method needs to finely discretize the wind farm based on the mechanism model of computational fluid dynamics, resulting in extremely long calculation time for the numerical simulation of the mechanism model and low efficiency of the wind farm wake distribution simulation analysis.

[0005] The present invention provides a wind farm wake distribution simulation method based on wind box splicing, comprising:

[0006] Based on the automated simulation interface, a single wind turbine wake sample is simulated and generated;

[0007] Constructing a single wind turbine wake model according to the single wind turbine wake sample;

[0008] Constructing a wind farm wake model according to the single wind turbine wake model and the actual wind farm distribution;

[0009] Based on the wind farm wake model, a wind box splicing mode is adopted to perform simulation analysis on the wake distribution of an actual wind farm, wherein the wind box splicing mode is a mode of splitting the actual wind farm into a plurality of spliced ​​wind boxes.

[0010] Furthermore, the automated simulation interface includes a batch simulation interface of the SOWFA container and a post-processing interface of the SOWFA container. The simulation based on the automated simulation interface generates a single fan wake sample, including:

[0011] Perform simulation calculations according to the batch simulation interface of the SOWFA image and the SOWFA container to obtain SOWFA batch simulation results; wherein the SOWFA image is constructed according to a pre-constructed Docker text, and the Docker text includes a basic image command, a SOWFA installation command, and a SOWFA dependency command;

[0012] The post-processing interface of the SOWFA container is determined by Paraview, and the SOWFA batch simulation results are processed according to the post-processing interface to obtain a single fan wake sample.

[0013] Further, the single wind turbine wake model is constructed according to the single wind turbine wake sample, and the single wind turbine wake model is constructed, including:

[0014] Determine the state parameters at the next moment according to the current state parameters of the actual wind farm;

[0015] Sampling the single fan wake sample according to a wind box sampling algorithm to obtain a single fan wake data set;

[0016] Based on the single wind turbine wake data set and the next moment state parameters, the single wind turbine wake model is pre-trained and re-trained to construct a single wind turbine wake model.

[0017] Furthermore, determining the state parameters at the next moment according to the current moment state parameters of the actual wind farm includes:

[0018] Determine the current state parameters of the actual wind farm based on the wind box sampling algorithm, and extract the spatial characteristics of the current state parameters;

[0019] Determine the spatial characteristics of the state parameters at the next moment according to the spatial characteristics of the state parameters at the current moment, the incoming flow conditions and the fan parameters;

[0020] The next moment parameters of the actual wind farm are determined according to the spatial characteristics of the next moment state parameters.

[0021] Furthermore, the loss function of the pre-training is:

[0022]

[0023] Among them, L re1 is the loss function of pre-training; mean 0,2,3 is the average value over dimensions 0, 2, and 3; It is the self-predicted tail flow of the Autoencoder network; is the actual wake; t is the time.

[0024] Furthermore, the loss function of the retraining is:

[0025]

[0026] L ls =||Z t+1 +Z t-1 -2Z t || 2

[0027] Among them, L re2 and L ls are all loss functions for retraining; mean 0,2,3 is the average value over dimensions 0, 2, and 3; is the wind field state parameter at time t predicted by the neural network based on the wind field state parameter at time t-1; The actual wind field state parameters determined by the simulation process at time t; is the actual wind field state parameter at time t Wind field parameters mapped by the Autoencoder network; L ls To constrain the neural network to maintain a linear loss function in the latent space; Z t+1 , Z t , Z t-1 They are the characteristic vectors mapped in the latent space by the actual wind field state parameters at time t+1, t, and t-1 respectively.

[0028] Furthermore, the wind farm wake model is constructed according to the single wind turbine wake model and the actual wind farm distribution, including:

[0029] Determine a wind box model according to the actual distribution of the wind farm, wherein the wind box model includes several wind turbine wind box models and / or a wind turbine-free wind box model;

[0030] Determining the boundary flow condition of the actual wind farm based on the wind box model;

[0031] A wind farm wake model of an actual wind farm is determined according to the single wind turbine wake model and the boundary incoming flow condition.

[0032] The present invention also provides a wind farm wake distribution simulation device based on wind box splicing, comprising:

[0033] Single wind turbine wake sample generation module, used to simulate and generate single wind turbine wake samples based on the automated simulation interface;

[0034] A single wind turbine wake model construction module, used to construct a single wind turbine wake model according to the single wind turbine wake sample;

[0035] A wind farm wake model building module is used to build a wind farm wake model according to the single wind turbine wake model and the actual wind farm distribution;

[0036] The wake distribution simulation analysis module is used to simulate and analyze the wake distribution of an actual wind farm based on the wind farm wake model by adopting a wind box splicing mode, wherein the wind box splicing mode is a mode of splitting the actual wind farm into a plurality of spliced ​​wind boxes.

[0037] The present invention also provides a terminal device, comprising: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, the wind farm wake distribution simulation method based on wind box splicing as described above is implemented.

[0038] The present invention also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned wind farm wake distribution simulation method based on wind box splicing.

[0039] The present invention generates a single wind turbine wake sample and constructs a single wind turbine wake model based on an automated simulation interface, and on this basis, constructs a wind farm wake model in combination with the actual wind farm distribution, so as to simulate and analyze the wake distribution of the actual wind farm using a risk splicing mode. The wake distribution simulation result of the entire wind farm can be accurately obtained without finely discretizing the wind farm and performing a large number of grid calculations for CFD simulation, which can effectively reduce the simulation calculation time, thereby effectively improving the efficiency of the simulation analysis of the wind farm wake distribution; and the embodiments of the present invention can effectively reduce the wind energy loss caused by the wake effect in the wind farm, thereby effectively improving the overall output of the wind farm.

[0040] Furthermore, the present invention groups several related wind turbines into multiple groups of wind turbine models that interfere with each other, by grouping them in pairs. This can ignore the complex spatial structure that may exist between wind turbines with wake correlation, thereby converting the complex multi-wind turbine wind field problem into a dual-wind turbine model problem, thereby effectively simplifying the model and enhancing the versatility of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a method for simulating the wake distribution of a wind farm based on wind box splicing provided by an embodiment of the present invention;

[0042] Figure 2is a schematic diagram of the principle of a single wind turbine wake model provided by an embodiment of the present invention;

[0043] Figure 3 is a schematic diagram of the principle of a wind turbine-free wake model provided by an embodiment of the present invention;

[0044] Figure 4 It is a schematic diagram of the principle of bellows splicing provided by an embodiment of the present invention;

[0045] Figure 5 It is a structural schematic diagram of a wind farm wake distribution simulation device based on wind box splicing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0047] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0048] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0049] See also Figure 1 The present invention provides a wind farm wake distribution simulation method based on wind box splicing, comprising:

[0050] S1. Based on the automated simulation interface, simulate and generate a single fan wake sample;

[0051] In an embodiment of the present invention, the automated simulation interface may be an automated simulation interface of SOWFA, including a batch simulation interface and a post-processing interface, wherein SOWFA (Simulator for Wind Farm Applications) is a high-fidelity wind farm simulation engine developed by the Renewable Energy Laboratory based on the OpenFOAM computational fluid dynamics library.

[0052] S2. Construct a single wind turbine wake model based on the single wind turbine wake sample;

[0053] In the embodiment of the present invention, a single wind turbine wake model may be constructed according to a single wind turbine wake sample and a preset simulation NS equation.

[0054] S3. Construct a wind farm wake model based on the single wind turbine wake model and the actual wind farm distribution;

[0055] S4. Based on the wind farm wake model, a wind box splicing mode is used to simulate and analyze the wake distribution of the actual wind farm, wherein the wind box splicing mode is a mode in which the actual wind farm is split into several spliced ​​wind boxes.

[0056] The embodiment of the present invention generates a single wind turbine wake sample and constructs a single wind turbine wake model based on an automated simulation interface, and on this basis, constructs a wind farm wake model in combination with the actual wind farm distribution, so as to simulate and analyze the wake distribution of the actual wind farm using a risk splicing mode. The wake distribution simulation result of the entire wind farm can be accurately obtained without finely discretizing the wind farm and performing a large number of grid calculations for CFD simulation, which can effectively reduce the simulation calculation time, thereby effectively improving the efficiency of the wind farm wake distribution simulation analysis.

[0057] In one embodiment, the automated simulation interface includes a batch simulation interface of the SOWFA container and a post-processing interface of the SOWFA container. Step S1, based on the automated simulation interface, simulates and generates a single fan wake sample, including:

[0058] S11, performing simulation calculation according to the batch simulation interface of the SOWFA image and the SOWFA container to obtain the SOWFA batch simulation result; wherein the SOWFA image is constructed according to the pre-built Docker text, and the Docker text includes the basic image command, the SOWFA installation command and the SOWFA dependency command;

[0059] In the embodiment of the present invention, a Docker text may be constructed based on Docker, that is, Docker may be used to manage the SOWFA container, so as to generate a SOWFA image that is easy to transplant and deploy according to the Docker text.

[0060] S12. Use Paraview to determine the post-processing interface of the SOWFA container, process the SOWFA batch simulation results according to the post-processing interface, and obtain a single fan wake sample.

[0061] It should be noted that SOWFA is a high-fidelity wind farm simulation engine developed by the Renewable Energy Laboratory based on the OpenFOAM computational fluid dynamics library. As a member of the mechanism model, after reading the specific environmental information and wind turbine information of the wind farm, SOWFA performs CFD simulation through a large amount of grid calculations to accurately obtain the specific spatiotemporal wind distribution of the entire wind farm. SOWFA is a wind farm simulation tool with powerful functions, but it is difficult to apply due to its complex operation and complicated parameters. In this embodiment, Docker is first used to implement the virtualized container management of SOWFA to form a SOWFA image that is easy to transplant and deploy. Then the batch simulation interface of the SOWFA container is designed and implemented, and parameter editing, calculation process management, and batch simulation are implemented in combination with the SOWFA image to obtain the SOWFA batch simulation results. Finally, a post-processing interface is constructed based on Paraview to process the SOWFA batch simulation results into data samples (single wind turbine wake samples) that meet business needs.

[0062] In the embodiment of the present invention, the SOWFA image can be loaded on any operating system that supports Docker to build a SOWFA container. Secondly, the embodiment of the present invention develops a batch simulation interface to connect the container and manage the parameters and calculation process of SOWFA. Through this batch simulation interface, batch simulation can be realized. Finally, a Python script is written to export the batch simulation results, and the batch simulation results are post-processed using the post-processing interface of the ParaView library, and the required sample data set is constructed.

[0063] In the embodiment of the present invention, SOWFA is installed in a Docker container and communicates with the host server via SSH. Compared with traditional virtual machines, Docker has many advantages such as fast startup speed, efficient resource call, flexible resource sharing between container and host, simple deployment and migration, etc. Therefore, building a Docker image of SOWFA is intuitive and can bring great convenience. The steps for configuring the SOWFA image are as follows:

[0064] First, pull the complete OpenFoam Docker image and download the scotch installation package. After configuring the system path, create a Docker container using the OpenFoam Docker image. Second, start and enter the container. Download the SOWFA installation package and install SOWFA. Set the SOWFA_DIR and OpenFoam environment variables. Next, install OpenSSH-Server. Start the SSH service and set the connection username and password. Finally, build a Docker image using the current container.

[0065] After building the Docker image of SOWFA, you can use this image to create a Docker container; by mounting a folder in the host server into the container, you can achieve file sharing; by setting up port mapping between the host and the container, you can remotely and directly control the operations in the container through SSH. In order to control the computing process of SOWFA in the container, the embodiment of the present invention can also develop an SSHConnecter module, which can execute commands and transmit data between Docker and the host.

[0066] In the embodiment of the present invention, the SOWFA wind farm simulation can be divided into two stages: an atmospheric boundary layer (ABL) simulation stage and a wind turbine generator simulation stage.

[0067] The purpose of the ABL simulation phase is to simulate a fully developed turbulent flow field that does not contain wind turbines. After the simulation is completed, boundary data for multiple time steps are extracted to provide boundary conditions for the second phase of the simulation. For each operating condition with a specific wind speed and direction, the ABL simulation only needs to be performed once. The simulation results can be reused for the simulation of all wind turbines in the same wind farm under the same operating conditions.

[0068] The wind turbine simulation stage is based on the aforementioned fully developed turbulent flow field. The simplified model of the wind turbine is added to simulate the operation of the wind turbine in the eddy field. In this stage, the wind turbine is added to the ABL case and the wind farm is re-simulated. The simplified wind turbine models include the action line model (ALM) and the action disk model (ADM). The difference between the two models is that the ALM uses a more detailed three-blade model to represent the wind turbine, while the ADM uses a disk model. In order to meet different needs, Python APIs have been developed for both models.

[0069] This embodiment uses Paraview as the post-processing platform for the example. Paraview has a python interface, and its recording function can be used to generate batch scripts to process the massive examples generated previously. The examples generated in this embodiment are all in the format of (600, 600, 300). Paraview scripts are used for processing, the wind speed data of the examples are extracted and a csv file is generated as a learning sample for the single wind turbine proxy model. Data in csv format cannot be directly called by neural networks, so the embodiment of the present invention separately writes a python script to process the csv file, and saves the data in the csv file as a .npy file according to the geometric relationship of the example image, in preparation for subsequent deep learning.

[0070] In summary, the embodiments of the present invention can effectively solve the problems of difficult SOWFA installation, complex parameter setting, and cumbersome calculation process management. The proposed batch simulation tool is an important supplement to SOWFA. It can be foreseen that this tool can promote the application of SOWFA in actual power system engineering scenarios.

[0071] In one embodiment, step S2, constructing a single wind turbine wake model according to a single wind turbine wake sample, constructing a single wind turbine wake model includes:

[0072] S21, determining the state parameters at the next moment according to the current state parameters of the actual wind farm;

[0073] In an embodiment of the present invention, the time term and the space term of the preset simulation NS equation can be decoupled by using a preset neural network to convert the problem of solving the partial differential equation into the problem of solving the ordinary differential equation. The preset neural network includes an encoder and a compiler, the encoder is used to extract the spatial characteristics of the state parameters at the current moment according to the state parameters of the wind farm at the current moment; the compiler determines the parameters at the next moment according to the spatial characteristics of the state parameters at the next moment.

[0074] In the embodiment of the present invention, the NS equations of the mechanism model describe the physical equations of the flow problem, wherein the mass conservation equation is:

[0075]

[0076] The momentum conservation equation is:

[0077]

[0078] Where u is the wind speed vector; p is the pressure distribution; ρ is the air density; is the gradient sign (gas momentum is affected by pressure gradient); t is time; The gas viscosity term that prevents relative sliding between fluid layers is generally written as (Shear stress is proportional to velocity gradient).

[0079] In an embodiment of the present invention, a preset neural network includes an encoder and a compiler. The encoder and the decoder will extract the spatial characteristics of the wind farm state parameters, fix the spatial correlation of the fluid field to the encoder neural network, and extract the wind speed distribution of the actual wind farm state parameters as a feature vector that can characterize all the information of the wind farm at that moment. After the encoder extracts all the spatial characteristics of the wind farm state parameters, the model will perform temporal deduction in the latent space of the feature vector learned by the encoder according to the incoming flow conditions, wind turbine parameters and other conditions to obtain the feature vector of the wind field at the next moment, and translate it through the decoder network to obtain the actual wind speed distribution of the wind field at the next moment. In the above manner, in the embodiment of the present invention, the neural network fixes the spatial characteristics to the encoder and decoder networks, and only performs time-based deduction in the latent space where the feature vector is located, thereby realizing the decoupling of time and space of the wind field.

[0080] Regarding the encoder's extraction of wind field state parameter features, the principle is that the wind speed distribution of the wake usually shows a certain regularity. Taking the steady-state wake model as an example, the parameter model summarizes the Gaussian distribution law of wind speed and the law of yaw angle and wake deviation through empirical observation. There is information redundancy in high-dimensional wind speed state variables. There is a certain regular relationship between the wind speeds at adjacent grid points in space. The universal approximation theorem of the embodiment of the present invention ensures that a neural network with enough hidden units and linear output layers can represent any arbitrary function. Therefore, during the model training process, the encoder will find the general laws of the wind field, summarize the spatial information of the model through the laws, and thus extract the wind field state parameters with a large amount of data as characteristic parameters containing all wind field information.

[0081] After the spatial characteristics are fixed, it can be seen from the comparison with the NS equation that only the momentum equation remains in the system of equations The characteristics of the quality equation have been completely summarized in relation to time. In the latent space constructed by the encoder based on the spatial characteristics, the characteristic vectors extracted from the wind field state parameters only have time correlation. It can be pointed out that in the latent space, the NS equation is transformed from a partial differential equation to a first-order ordinary differential equation that is only related to time. In the latent space, the characteristic vectors extracted from the wind field state parameters at each time should satisfy the characteristics of linear deduction. The present invention designs a linear neural network in the latent space, introduces parameter conditions, constrains the time step deduction process in the latent space to be a linear deduction, and finally achieves the goal of using neural network simulation.

[0082] The simulation idea of ​​the neural network in the embodiment of the present invention is to use the state parameter X of the wind farm at time t as the basis. t , boundary flow condition V t , fan parameters U t As input, predict the wind field state parameters at time t+1 In this way, the development process of the wind farm wake at a later time is deduced based on the state parameters of the wind farm at the initial moment, so that the wind farm wake model of the actual wind farm can be determined.

[0083] In the embodiment of the present invention, the state parameters at the next moment may be determined according to the current state parameters of the actual wind farm:

[0084] S211, determining the current state parameters of the actual wind farm based on the wind box sampling algorithm, and extracting the spatial characteristics of the current state parameters;

[0085] S212, determining the spatial characteristics of the state parameters at the next moment according to the spatial characteristics of the state parameters at the current moment, the incoming flow conditions and the fan parameters;

[0086] S213: Determine the next moment parameters of the actual wind farm according to the spatial characteristics of the state parameters at the next moment.

[0087] In the embodiment of the present invention, the input samples of the neural network can be taken in the form of wind box sampling. The reason for selecting wind box sampling is that the wind speed distribution of the wind field is highly coupled in time and space, and wind box sampling is required to retain the spatiotemporal information of the wind field.

[0088] In the process of sampling the three-dimensional wind field, the neural network cannot obtain accurate wake propagation time and coordinates because the wake propagation time is highly coupled with space. Specifically, under the same sampling time interval, the wake propagation distance of different wind conditions is different; under the same sampling distance, the wake propagation time of different wind conditions is different. In the actual wind field, the wind speed at high altitudes is higher than that at low altitudes, and the same sampling distance cannot take into account the wind speed difference on the vertical coordinate.

[0089] To this end, the embodiment of the present invention adopts a wind box sampling method, which divides the wind field into fine grids and uses a high-density uniform sampling method at fixed time intervals to solve the problem that different wind speeds propagate at different distances at the same time interval. Wind box sampling ensures the retention of wake propagation information details, ensures that the input samples of the neural network fully contain the spatiotemporal characteristics of the NS equation, and ensures the information basis for the subsequent feature extraction of the neural network.

[0090] In an embodiment of the present invention, the simulation data may be sampled according to a wind box sampling method, and the obtained wind box data (single wind turbine wake data set) may be flattened into a plane according to spatial logic as learning data input for a neural network.

[0091] In order to achieve better training results of the single wind turbine wake model based on neural network, the training process of the neural network is divided into two parts: pre-training and re-training.

[0092] Pre-training uses 36,900 sets of wind speed data from all samples at each time point as input and trains the Autoencoder network itself as output. The goal of pre-training is to constrain the Autoencoder network in advance to a latent space that can express features, constrain the parameter values ​​of the neural network in advance for formal training, reduce the complexity of formal training, and make it easier for the neural network to be optimized to the target position.

[0093] During the pre-training process, the samples are blurred and mask noise is added to increase the diversity of the samples. The robustness of the model is increased through the defects of the samples, and the missing data is used to ensure that the feature vectors in the latent space can accurately represent the characteristics of the wind farm state parameters and ensure the completeness of the latent space. Based on the goal of pre-training, the pre-training loss function is designed.

[0094] Next, retraining is performed with 369 samples of wind speed data at t = 20002s and the yaw angle and incoming flow conditions of the wind turbine at subsequent moments as input, iterating 99 times in the latent space, and decoding all iterative results through the Decoder network and comparing them with the actual results to verify the accuracy of the model. All predicted wake results obtained during the iteration process are compared with the simulation results of the mechanism model, and the retraining loss function is designed.

[0095] In the embodiment of the present invention, the current state data of the actual wind farm may be determined according to the wind turbine sampling algorithm.

[0096] S22, sampling the single fan wake samples according to the wind box sampling algorithm to obtain a single fan wake data set;

[0097] S23. Based on the single wind turbine wake data set and the state parameters at the next moment, the single wind turbine wake model is pre-trained and re-trained to construct a single wind turbine wake model.

[0098] See also Figure 2 , which is a schematic diagram of the principle of a single wind turbine wake model provided in an embodiment of the present invention.

[0099] In one embodiment, the loss function of pre-training is:

[0100]

[0101] Among them, L re1 is the loss function of pre-training; mean 0,2,3 is the average value over dimensions 0, 2, and 3; It is the self-predicted tail flow of the Autoencoder network; is the actual wake; t is the time.

[0102] In one embodiment, the loss function for retraining is:

[0103]

[0104] L ls =||Z t+1 +Z t-1 -2Z t || 2

[0105] Among them, L re2 and L ls are all loss functions for retraining; mean 0,2,3 is the average value over dimensions 0, 2, and 3; is the wind field state parameter at time t predicted by the neural network based on the wind field state parameter at time t-1; The actual wind field state parameters determined by the simulation process at time t; is the actual wind field state parameter at time t Wind field parameters mapped by the Autoencoder network; L ls To constrain the neural network to maintain a linear loss function in the latent space; Z t+1 , Z t , Z t-1 They are the characteristic vectors mapped in the latent space by the actual wind field state parameters at time t+1, t, and t-1 respectively.

[0106] In one embodiment, step S3, constructing a wind farm wake model according to a single wind turbine wake model and actual wind farm distribution, includes:

[0107] S31. Determine a wind box model according to the actual distribution of the wind farm, wherein the wind box model includes several wind turbine wind box models and / or a wind turbine-free wind box model;

[0108] S32, determining the boundary flow conditions of the actual wind farm based on the wind box model;

[0109] S33. Determine a wind farm wake model of an actual wind farm according to a single wind turbine wake model and boundary flow conditions.

[0110] In an embodiment of the present invention, a large-scale wind farm is often composed of dozens or even hundreds of wind turbines arranged side by side to form a network, and the wake model of a single wind turbine is obviously not applicable to actual projects. In this embodiment, based on the training foundation that has been completed for a small wind farm, the wind farm simulated by the mechanism model is expanded to a wind farm that contains multiple wind turbines and is more similar to an actual wind farm. Through the study of multi-wind turbine wind farms, a wind box model design is proposed, and the expanded wind farm is divided into a wind box with wind turbines and a wind box without wind turbines. By constructing a neural network for the two wind box models, the actual wind farm is restored by splicing the wind boxes, thereby achieving the versatility of the model.

[0111] Specifically, in a large wind farm, under normal inflow conditions, not all the wakes of the upstream wind turbines will affect the downstream wind turbines. Only the wind turbines in the upstream wind turbine wake radiation area will be truly affected by the wake. At the beginning of the design and construction of a large wind farm, the inflow conditions at the construction site in years will be studied, and the wind turbines to be built will be arranged according to the actual situation. Many wind turbines in actual wind farms are not affected by the wake. Based on the initial planning of wind farm construction, the wind turbines that are actually affected by the wake are often small wind farms consisting of only a few wind turbines. Therefore, in the wake study process of a large wind farm, it is only necessary to model the wake of several wind turbines that are interfering in front and behind.

[0112] In the embodiment of the present invention, the actual wind farm can be divided into the space near the wind turbine and the area without wind turbines, so as to divide the large wind farm into a wind farm composed of multiple wind boxes. Among them, the wind box model in the embodiment of the present invention designs two wind box modes, including a wind box mode with wind turbines and a wind box mode without wind turbines. The two wind box modes can be combined into any form of wind farm by splicing. Among them, the aerodynamic description of the wind box without wind turbines can also be expressed by the NS equation.

[0113]

[0114] Unlike the bellows with fans, since the bellows does not contain fans, the above equations do not contain the volume force term of the fans, and the form is more concise. Therefore, a neural network structure similar to that of the bellows with fans can be used to construct the bellows model.

[0115] See also Figure 3 , which is a schematic diagram of the principle of a wake model without an extension provided by an embodiment of the present invention. For the wind box itself, its overall wind speed characteristics are still extracted by using the wind speed distribution in the wind box at time t. For the fan parameters, this item does not exist in the wind box without a fan, so the wake model without a fan corresponding to the wind box without a fan mode deletes this part of the neural network during the design process.

[0116] During the training of the single wind turbine model, the boundary flow condition of the wind field is the naturally distributed turbulent wind, which can be considered to remain unchanged on average over a long time scale, which is very different from the wind box model. The focus of the multi-wind box model is on the simulation of the wind turbine wake. Therefore, for the wind box model, the wind field flow is no longer a stable atmospheric flow, but an uneven flow containing a variety of possible wind turbine wakes.

[0117] See also Figure 4, which is a schematic diagram of the principle of wind box splicing provided in an embodiment of the present invention. The wind box model needs to consider the splicing problem between multiple wind boxes, and how to accurately describe the boundary conditions of the wind box flow becomes an important issue in the research process. The wind speed distribution in three-dimensional space is not uniform. This is especially true for wake flows that actually contain multiple turbulences. For the uniform atmospheric flow, the fluid at the same vertical interface at a certain moment will change over time. If there is a wake at the interface, the changes in the propagation process will be more complicated.

[0118] Therefore, when considering the wake effect, the wind field sampled by the wind box cannot only give a specific interface to characterize all the characteristics of the boundary flow when giving the boundary conditions. To solve this problem, starting from the basis of the mechanism model, the NS equation is discretized to obtain:

[0119]

[0120] Among them, the subscript of the parameter represents the position of the parameter after spatial discretization, and the superscript represents the time after temporal discretization. is the wind speed at point i at time n+1; is the wind speed at point i at time n; is the pressure at point i+1 / 2 at time n; is the pressure at i-1 / 2 at time n; μ is the dynamic viscosity of the fluid; is the wind speed at point i-1 at time n.

[0121] According to the discrete form, it can be seen from the difference format of the NS equation that the state parameter of the wind speed at (x0, y0, z0) at time t depends on the state parameter of the wind speed at (x0-u x Δt,y0-u y Δt,z0-u z For the discretized wind box data, this means that the data of the boundary condition selection area needs to include the envelope surface of the wind speed range within the Δt time. In other words, it is necessary to include the farthest distance that can be reached within the Δt time at the highest wind speed in the interface. Based on the above analysis, this embodiment adopts a method of extracting features from multiple interfaces as boundary conditions, and all the information of the boundary conditions is included through the numerical discretization of multiple interfaces.

[0122] The boundary conditions of the rear wind box will be the last few air outlet interfaces of the front wind box as the boundary conditions v t, input into the neural network model. The neural network responsible for boundary flow condition analysis will extract the boundary condition features of several wind outlet interfaces, map them into the latent space, and complete the deduction of the overall state parameters of the wind field. From a physical point of view, the principle of such a design is that within the Δt time, the wake of the interface outside the selected interface cannot be propagated to the rear wind box, so the wind speed data has no correlation with the rear wind box. The characteristics of all wind speed interfaces that may propagate the wind speed to the rear wind box have been processed by the boundary condition neural network.

[0123] Given the standard atmospheric flow, SOWFA is used to generate samples of two wind turbines with a size of (1800m, 600m, 300m) and a spacing of 1200m. The forward input of the wind box is the standard atmospheric flow, and the internal wind turbines have different yaw angles. The entire wind field is divided into three wind boxes. With the two wind turbines as the center and the three-dimensional wind speed data of 300m before and after, 738 single-wind turbine wind box samples with a size of (600m, 600m, 300m) are constructed, and 369 wind boxes without wind turbines with a size of (600m, 600m, 300m) are selected based on the 600m spacing between the two wind turbine wind boxes. A sampling point is selected every 20m, and the three-dimensional wind speed data at the data point is extracted. A total of (31, 31, 16) data points are obtained for each wind box.

[0124] A total of 369 dual-fan samples were generated, with wind speeds ranging from 8 to 16 m / s, incoming wind speeds from the west, and yaw angles of the fans ranging from 250 to 290. After the turbulent wind field was fully developed for 20,000 s, wind speed samples were extracted at intervals of 2 s from 20,002 to 20,200 s, with a total of 100 sample data.

[0125] Since the training data of the bellows model consists of three bellows, and considering that the first bellows has a uniform incoming flow, the boundary conditions are given in a different way from the subsequent bellows, so the actual model training method is to use a mode of separate training of the three bellows. In this embodiment, an independent bellows model is constructed for each bellows, and the neural network structure of each bellows is still relatively complex. The actual training process involves multiple iterations. In order to achieve better training results, the training process of the neural network is divided into two parts: pre-training and re-training.

[0126] The pre-training of the three wind boxes is based on the samples corresponding to each wind box. A total of 36,900 sets of wind speed data for each wind box at each time point are used as input, and the Autoencoder network is trained as output. Similar to the pre-training goal of the single wind turbine wake model, the pre-training will constrain the Autoencoder network to a latent space that can express features in advance, constrain the parameter values ​​of the neural network in advance for re-training, reduce the complexity of re-training, and make it easier for the neural network to be optimized to the target position.

[0127] During the pre-training process, the samples of the three wind boxes were blurred and mask noise was added to increase the diversity of the samples. It is expected that the defects of the samples can increase the robustness of the model, and the missing data can ensure that the feature vectors in the latent space can accurately represent the characteristics of the wind field state parameters to ensure the completeness of the latent space.

[0128] A total of 369 training samples were involved in the formal training of the three wind boxes, of which 80% were used as training data and 20% were used as test data. During the deduction training process, the wind speed state parameters of the wind field at t=20002s were used as the basis. The first wind box uses fixed parameters as the incoming flow conditions, and the second and third wind boxes use the last three air outlet interfaces of the previous wind box as the incoming flow conditions. The yaw angles of the fans inside the first and third wind boxes are used as fan parameters, and the second wind box has no fan parameters. Based on the initial wind field state parameters and subsequent parameter conditions, the wind field iterates 99 times in the latent space formed by the Autoencoder network, and all iterative results are decoded by the Decoder network and compared with the actual results to verify the accuracy of the model. The errors of all predicted wake results obtained in the iterative process are compared with the simulation results of the mechanism model. The loss function of the model design is the same as that of the single wind turbine wake model.

[0129] The implementation of the embodiments of the present invention has the following beneficial effects:

[0130] The embodiment of the present invention generates a single wind turbine wake sample and constructs a single wind turbine wake model based on an automated simulation interface, and on this basis, constructs a wind farm wake model in combination with the actual wind farm distribution, so as to simulate and analyze the wake distribution of the actual wind farm using a risk splicing mode. The wake distribution simulation result of the entire wind farm can be accurately obtained without finely discretizing the wind farm and performing a large number of grid calculations for CFD simulation, which can effectively reduce the simulation calculation time, thereby effectively improving the efficiency of the simulation analysis of the wind farm wake distribution; and the embodiment of the present invention can effectively reduce the wind energy loss caused by the wake effect in the wind farm, thereby effectively improving the overall output of the wind farm.

[0131] Furthermore, the embodiment of the present invention groups the related wind turbines into multiple groups of wind turbine models that interfere with each other by twos in front and behind. This can ignore the complex spatial structure that may exist between the wind turbines with wake correlation, thereby converting the complex multi-wind turbine wind field problem into a dual-wind turbine model problem, thereby effectively simplifying the model and enhancing the versatility of the model.

[0132] See also Figure 5 Based on the same inventive concept as the above embodiment, the present invention also provides a wind farm wake distribution simulation device based on wind box splicing, comprising:

[0133] A single wind turbine wake sample generation module 10 is used to simulate and generate a single wind turbine wake sample based on an automated simulation interface;

[0134] A single wind turbine wake model building module 20 is used to build a single wind turbine wake model according to a single wind turbine wake sample;

[0135] A wind farm wake model building module 30 is used to build a wind farm wake model according to a single wind turbine wake model and actual wind farm distribution;

[0136] The wake distribution simulation analysis module 40 is used to simulate and analyze the wake distribution of an actual wind farm based on the wind farm wake model by adopting a wind box splicing mode, wherein the wind box splicing mode is a mode of splitting the actual wind farm into a plurality of spliced ​​wind boxes.

[0137] In one embodiment, the automated simulation interface includes a batch simulation interface of the SOWFA container and a post-processing interface of the SOWFA container, and the single fan wake sample generation module 10 is further used for:

[0138] Perform simulation calculations according to the batch simulation interface of the SOWFA image and the SOWFA container to obtain the SOWFA batch simulation results; wherein the SOWFA image is constructed according to the pre-built Docker text, and the Docker text includes the base image command, the SOWFA installation command and the SOWFA dependency command;

[0139] The post-processing interface of the SOWFA container is determined by Paraview. The SOWFA batch simulation results are processed according to the post-processing interface to obtain a single fan wake sample.

[0140] In one embodiment, the single wind turbine wake model building module 20 is further used to:

[0141] Determine the state parameters at the next moment according to the current state parameters of the actual wind farm;

[0142] The single fan wake samples are sampled according to the wind box sampling algorithm to obtain a single fan wake data set;

[0143] Based on the single wind turbine wake data set and the state parameters at the next moment, the single wind turbine wake model is pre-trained and re-trained to construct a single wind turbine wake model.

[0144] In one embodiment, determining the state parameters at the next moment according to the current moment state parameters of the actual wind farm includes:

[0145] Determine the current state parameters of the actual wind farm based on the wind box sampling algorithm and extract the spatial characteristics of the current state parameters;

[0146] Determine the spatial characteristics of the state parameters at the next moment according to the spatial characteristics of the state parameters at the current moment, the incoming flow conditions and the fan parameters;

[0147] The next moment parameters of the actual wind farm are determined according to the spatial characteristics of the state parameters at the next moment.

[0148] In one embodiment, the loss function of pre-training is:

[0149]

[0150] Among them, L re1 is the loss function of pre-training; mean 0,2,3 is the average value over dimensions 0, 2, and 3; It is the self-predicted tail flow of the Autoencoder network; is the actual wake; t is the time.

[0151] In one embodiment, the loss function for retraining is:

[0152]

[0153] L ls =||Z t+1 +Z t-1 -2Z t || 2

[0154] Among them, L re2 and L ls are all loss functions for retraining; mean 0,2,3 is the average value over dimensions 0, 2, and 3; is the wind field state parameter at time t predicted by the neural network based on the wind field state parameter at time t-1; The actual wind field state parameters determined by the simulation process at time t; is the actual wind field state parameter at time t Wind field parameters mapped by the Autoencoder network; L ls To constrain the neural network to maintain a linear loss function in the latent space; Z t+1 , Z t , Z t-1 They are the characteristic vectors mapped in the latent space by the actual wind field state parameters at time t+1, t, and t-1 respectively.

[0155] In one embodiment, the wake distribution simulation analysis module 40 is further used to:

[0156] Determine a wind box model according to the actual distribution of the wind farm, wherein the wind box model includes several wind turbine wind box models and / or a wind turbine-free wind box model;

[0157] Determine the boundary flow conditions of the actual wind farm based on the wind box model;

[0158] According to the single wind turbine wake model and boundary flow conditions, the wind farm wake model of the actual wind farm is determined.

[0159] Accordingly, an embodiment of the present invention also provides a terminal device, comprising: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the wind farm wake distribution simulation method based on wind box splicing as any one of the above embodiments.

[0160] The terminal device of this embodiment includes: a processor, a memory, and a computer program and computer instructions stored in the memory and executable on the processor. When the processor executes the computer program, each step in the above embodiment 1 is implemented, for example Figure 1 Alternatively, when the processor executes the computer program, the functions of each module / unit in the above device embodiment are implemented, such as the wind farm wake model building module 30 .

[0161] Exemplarily, the computer program can be divided into one or more modules / units, one or more modules / units are stored in a memory and executed by a processor to complete the present invention. One or more modules / units can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in a terminal device. For example, the wind farm wake model construction module 30 is used to construct a wind farm wake model based on a single wind turbine wake model and the actual wind farm distribution.

[0162] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the schematic diagram is merely an example of a terminal device and does not constitute a limitation on the terminal device. The terminal device may include more or fewer components than shown in the diagram, or may combine certain components, or different components. For example, the terminal device may also include an input / output device, a network access device, a bus, etc.

[0163] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0164] The memory can be used to store computer programs and / or modules. The processor realizes various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile terminal, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0165] Among them, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0166] Accordingly, an embodiment of the present invention further provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a wind farm wake distribution simulation method based on wind box splicing as described in any one of the above embodiments.

[0167] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A wind farm wake distribution simulation method based on wind box splicing, characterized in that: include: Based on the automated simulation interface, a single wind turbine wake sample is simulated and generated; Constructing a single wind turbine wake model according to the single wind turbine wake sample; Constructing a wind farm wake model according to the single wind turbine wake model and the actual wind farm distribution; Based on the wind farm wake model, a wind box splicing mode is adopted to perform simulation analysis on the wake distribution of an actual wind farm, wherein the wind box splicing mode is a mode of splitting the actual wind farm into a plurality of spliced ​​wind boxes.

2. The wind farm wake distribution simulation method based on wind box splicing according to claim 1 is characterized in that: The automated simulation interface includes a batch simulation interface of the SOWFA container and a post-processing interface of the SOWFA container. The method of simulating and generating a single fan wake sample based on the automated simulation interface includes: Perform simulation calculations according to the batch simulation interface of the SOWFA image and the SOWFA container to obtain SOWFA batch simulation results; wherein the SOWFA image is constructed according to a pre-constructed Docker text, and the Docker text includes a basic image command, a SOWFA installation command, and a SOWFA dependency command; The post-processing interface of the SOWFA container is determined by Paraview, and the SOWFA batch simulation results are processed according to the post-processing interface to obtain a single fan wake sample.

3. The wind farm wake distribution simulation method based on wind box splicing according to claim 1 is characterized in that: The step of constructing a single wind turbine wake model according to the single wind turbine wake sample comprises: Determine the state parameters at the next moment according to the current state parameters of the actual wind farm; Sampling the single fan wake sample according to a wind box sampling algorithm to obtain a single fan wake data set; Based on the single wind turbine wake data set and the next moment state parameters, the single wind turbine wake model is pre-trained and re-trained to construct a single wind turbine wake model.

4. The method for simulating the wind farm wake distribution based on wind box splicing according to claim 3, characterized in that: The determining of the state parameters at the next moment according to the current moment state parameters of the actual wind farm comprises: Determine the current state parameters of the actual wind farm based on the wind box sampling algorithm, and extract the spatial characteristics of the current state parameters; Determine the spatial characteristics of the state parameters at the next moment according to the spatial characteristics of the state parameters at the current moment, the incoming flow conditions and the fan parameters; The next moment parameters of the actual wind farm are determined according to the spatial characteristics of the next moment state parameters.

5. The method for simulating the wind farm wake distribution based on wind box splicing according to claim 3, characterized in that: The loss function of the pre-training is: Among them, L re1 is the loss function of pre-training; mean 0,2,3 is the average value over dimensions 0, 2, and 3; It is the self-predicted tail flow of the Autoencoder network; is the actual wake; t is the time.

6. The method for simulating the wind farm wake distribution based on wind box splicing according to claim 3, characterized in that: The loss function of the retraining is: L ls =‖Z t+1 +Z t-1 -2Z t ‖ 2 Among them, L re2 and L ls are all loss functions for retraining; mean 0,2,3 is the average value over dimensions 0, 2, and 3; is the wind field state parameter at time t predicted by the neural network based on the wind field state parameter at time t-1; The actual wind field state parameters determined by the simulation process at time t; is the actual wind field state parameter at time t Wind field parameters mapped by the Autoencoder network; L ls To constrain the neural network to maintain a linear loss function in the latent space; Z t+1 , Z t , Z t-1 They are the characteristic vectors mapped in the latent space by the actual wind field state parameters at time t+1, t, and t-1 respectively.

7. The method for simulating the wind farm wake distribution based on wind box splicing according to any one of claims 1 to 6, characterized in that: The step of constructing a wind farm wake model according to the single wind turbine wake model and the actual wind farm distribution includes: Determine a wind box model according to the actual distribution of the wind farm, wherein the wind box model includes several wind turbine wind box models and / or a wind turbine-free wind box model; Determining the boundary flow condition of the actual wind farm based on the wind box model; A wind farm wake model of an actual wind farm is determined according to the single wind turbine wake model and the boundary incoming flow condition.

8. A wind farm wake distribution simulation device based on wind box splicing, characterized in that: include: Single wind turbine wake sample generation module, used to simulate and generate single wind turbine wake samples based on the automated simulation interface; A single wind turbine wake model construction module, used to construct a single wind turbine wake model according to the single wind turbine wake sample; A wind farm wake model building module is used to build a wind farm wake model according to the single wind turbine wake model and the actual wind farm distribution; The wake distribution simulation analysis module is used to simulate and analyze the wake distribution of an actual wind farm based on the wind farm wake model by adopting a wind box splicing mode, wherein the wind box splicing mode is a mode of splitting the actual wind farm into a plurality of spliced ​​wind boxes.

9. A terminal device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for simulating the wind farm wake distribution based on wind box splicing according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the wind farm wake distribution simulation method based on wind box splicing according to any one of claims 1 to 7.

Images