Wind power wind speed prediction and optimization arrangement method, system and device

Abstract

The invention provides a wind power wind speed prediction and optimization arrangement method, system and device. The wind power wind speed prediction and optimization arrangement method comprises the steps of obtaining wind power wind speed basic data of a target area; constructing a neural network model based on physical information based on the wind power wind speed basic data; acquiring a wind speed at a specific position in a speed field from a neural network model based on physical information, and creating a training data set; training the neural network model based on the physical information based on the training data set; reconstructing wind speed distribution in a speed field through the trained neural network model based on physical information, and predicting to obtain a wind speed prediction value of the wind power plant in the next time period; and optimizing the arrangement of the wind turbine group according to the wind speed distribution in the reconstructed speed field. According to the method, the two-dimensional flow velocity field of the wind power plant can be reconstructed by training one PINN, and accurate wind speed ultra-short-term prediction is realized based on the trained model.

Description

Technical Field

[0001] The present application belongs to the technical field of wind power prediction, and relates to a wind power wind speed prediction method, and in particular to a wind power wind speed prediction and optimization arrangement method, system and device. Background Art

[0002] Wind energy is an unstable energy source with random fluctuations. It is random, intermittent and volatile, which easily leads to the instability of wind power. With the continuous development of wind power, the large-scale access of wind power to the power grid system in the future will inevitably bring challenges to the stability of the power grid system.

[0003] Therefore, it is necessary to accurately predict the output power of wind farms and combine it with efficient regulation of wind turbine systems to ensure reliable operation of the power grid.

[0004] In addition, it is particularly important to reduce various power generation losses during the power generation process of wind farms. Among them, the wake effect is a very important factor in wind farms. The purpose of optimizing the layout of wind turbines is to use the selected models to eliminate the wake effect between units as much as possible within the limited flow field. Generally speaking, the larger the row-to-row distance between wind turbines, the smaller the wake effect and the greater the power generation. However, the distance between rows and ground will also affect the total scale of wind turbine layout.

[0005] Therefore, in recent years, the control strategy and layout optimization of wind farms have been a hot topic in the industry. With the rapid development of artificial neural networks, their ability in time series prediction is quite outstanding.

[0006] In addition, although traditional wind farm modeling methods, such as computational fluid dynamics (CFD) methods, have high accuracy, the existence of multi-scale problems from the atmospheric scale to the wind turbine blade boundary layer scale makes modeling extremely difficult and the computational cost is high.

[0007] Therefore, in the existing technology, the traditional wind farm modeling method has multi-scale problems from the atmospheric scale to the wind turbine blade boundary layer scale, which makes modeling extremely difficult and the calculation cost is high; there are also technical problems such as insufficient accuracy in wind farm wind speed prediction. Summary of the invention

[0008] In view of the shortcomings of the prior art mentioned above, the purpose of the present application is to provide a method, system and device for wind power wind speed prediction and optimal arrangement, which are used to solve the problems in the prior art, such as unstable wind power function, large power generation losses in the wind farm power generation process, and multi-scale problems in traditional wind farm modeling methods from the atmospheric scale to the wind turbine blade boundary layer scale, which makes modeling extremely difficult and the calculation cost high; as well as insufficient wind farm wind speed prediction accuracy.

[0009] To achieve the above-mentioned purpose and other related purposes, in a first aspect, the present application provides a method for predicting and optimizing the arrangement of wind power wind speed, comprising the following steps: obtaining basic wind power wind speed data of a target area; the basic wind power wind speed data comprises: any one or more combinations of wind farm size, location, wind speed, wind direction, temperature, air pressure, terrain, and water depth; constructing a neural network model based on physical information based on the basic wind power wind speed data; obtaining the wind speed at a specific position in a velocity field based on the neural network model based on physical information, and creating a training data set; training the neural network model based on physical information based on the training data set; reconstructing the wind speed distribution in the velocity field through the trained neural network model based on physical information, and predicting the wind speed forecast value of the wind farm in the next time period; optimizing the arrangement of wind turbine groups according to the reconstructed wind speed distribution in the velocity field.

[0010] In an implementation of the first aspect, constructing a neural network model based on physical information based on the wind power wind speed basic data includes the following steps: extracting the wind farm size based on the wind power wind speed basic data; defining a two-dimensional velocity field according to the actual wind farm size; designing a neural network structure based on the two-dimensional velocity field; integrating physical constraints into the loss function of the neural network structure for optimization to obtain a neural network model based on physical information.

[0011] In one implementation of the first aspect, integrating physical constraints into the loss function of the neural network structure for optimization includes the following steps: using automatic differentiation technology to calculate partial derivatives of partial differential equations; using physical equations to integrate physical constraints into the loss function in the neural network; combining the partial derivatives of the partial differential equations with the loss function to obtain a total loss function.

[0012] In an implementation of the first aspect, the physical equation includes: a two-dimensional Navier-Stokes equation, a continuity equation, and a turbulence model; the calculation formula of the physical equation is:

[0013]

[0014] Where ρ represents the fluid density; u and v represent the measured velocity components of the fluid at point (x, y) at time t.

[0015] In an implementation of the first aspect, the calculation formula of the total loss function is:

[0016] L=ω1L d +ω2L f (4)

[0017] L d =u MRMSE +v MRMSE (5)

[0018]

[0019]

[0020] Among them, L represents the total loss function; L d Indicates data item loss; L f represents the physical item loss, i.e., PDE residual loss; ω1 represents the weight of the data item; ω2 represents the weight of the PDE physical item; n represents the total number of laser sensors set; N T represents the total number of time steps; u, v represent the measured velocity components of the fluid at point (x, y) at time t, that is, the measured values ​​at the corresponding positions; represents the predicted value of the corresponding velocity component of the fluid at point (x, y) at time t; MRMSE represents the mean root mean square error; G represents the implicit expression of Navier-Stokes.

[0021] In an implementation of the first aspect, obtaining the wind speed at a specific position in a velocity field based on the physical information-based neural network model and creating a training data set include the following steps: determining a target research area and a specific position of a wind farm; obtaining the wind speed at several specific positions in the velocity field using laser sensors distributed in the wind farm; performing data preprocessing on the wind speed at the specific position in the velocity field to obtain the preprocessed wind speed at the specific position in the velocity field; dividing the continuous time series data into multiple time windows of fixed length; defining a label for each time window; constructing a training data set based on the preprocessed wind speed at the specific position in the velocity field and the corresponding label, and using it as a supervision point for the training model.

[0022] In an implementation of the first aspect, training the physical information-based neural network model based on the training data set includes the following steps: using an Adams optimizer to train the training data set once to obtain a physical information-based neural network model after the loss value decreases; using an L-BFGS optimizer to perform secondary training until the convergence condition is met to obtain a trained physical information-based neural network model.

[0023] In an implementation of the first aspect, the calculation formula of the emission factor method is: reconstructing the wind speed distribution in the velocity field through a trained physical information-based neural network model, and obtaining the wind speed forecast value of the wind farm in the next time period through prediction, including the following steps: taking the position coordinates of the velocity field as the target point for wind speed reconstruction, inputting them into the trained physical information-based neural network model, and obtaining the wind speed forecast value corresponding to the position coordinates of the velocity field; mapping the wind speed forecast value to the grid of the velocity field to form a reconstructed wind speed distribution map; analyzing the reconstructed wind speed distribution map to obtain the analyzed wind speed distribution result, and optimizing and fine-tuning the trained physical information-based neural network model; obtaining the real-time wind speed of the target position in a certain time period; preprocessing the real-time wind speed in a certain time period to obtain the real-time wind speed in a certain time period after preprocessing; inputting the real-time wind speed in a certain time period after preprocessing into the trained physical information-based neural network model for prediction, and obtaining the wind speed forecast value of the wind farm in the next time period.

[0024] In a second aspect, the present application provides a wind power wind speed prediction and optimization arrangement system, including: an acquisition module, used to obtain basic wind power wind speed data of a target area; the basic wind power wind speed data includes: any one or more combinations of wind farm size, wind speed, wind direction, temperature, air pressure, terrain, and water depth; a model construction module, used to construct a neural network model based on physical information based on the basic wind power wind speed data; a wind speed acquisition module, used to obtain the wind speed at a specific position in a velocity field based on the neural network model based on physical information, and create a training data set; a model training module, used to train the neural network model based on physical information based on the training data set; a wind speed reconstruction module, used to reconstruct the wind speed distribution in the velocity field through the trained neural network model based on physical information, and predict the wind speed of the wind farm in the next time period; an optimization arrangement module, used to optimize the arrangement of wind turbine groups according to the reconstructed wind speed distribution in the velocity field.

[0025] In the last aspect, the present application provides a wind power wind speed prediction and optimization arrangement device, comprising: a processor and a memory. The memory is used to store a computer program; the processor is connected to the memory and is used to execute the computer program stored in the memory, so that the wind power wind speed prediction and optimization arrangement device executes the wind power wind speed prediction and optimization arrangement method.

[0026] As described above, the wind power wind speed prediction and optimization arrangement method, system and device described in the present application have the following beneficial effects:

[0027] (1) The wind speed prediction and optimal arrangement method provided in the present application uses a neural network based on physical information. By adding partial differential equations to the traditional neural network to make it satisfy the laws of physics, the problem of traditional neural network dependence on data and poor robustness can be overcome, and it has great advantages in processing incomplete data or noisy data. At the same time, the PINN adopted in the present application can make full use of the prior information of the physical constraint equation that generates the data, and add it as its own regular constraint to the loss function. It can use a small amount of flow field sampling data to reconstruct the regional flow more accurately. And the model can be used for wind farm modeling, and can provide accurate, ultra-short-term velocity field prediction, while taking into account the computational efficiency, providing a reference for short-term pitch and yaw adjustments, and realizing safe and stable operation of wind power.

[0028] (2) This application simulates the velocity distribution of the entire flow field by installing a laser radar sensor in the wind farm; based on measured data, it is more in line with actual working conditions; at the same time, it predicts the wind speed change over a period of time at the wind turbine position, which can provide a reference for the law analysis of wind power in the wind farm and provide a more comprehensive understanding of the physical characteristics of the wind farm;

[0029] (3) PINN has advantages in solving high-dimensional forward and inversion problems. By combining machine learning technology with a large amount of fluid dynamics data, it can provide accurate velocity field prediction results while having high computational efficiency.

[0030] (4) The accuracy of wind field assessment can be improved by using mesoscale data simulation of laser wind radar and microscale numerical simulation of PINN;

[0031] (5) This application is applicable to scenarios such as offshore wind farms and onshore wind farms and has wide applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Shown is a flow chart of a method for wind speed prediction and optimization arrangement of wind power in one embodiment of the present application.

[0033] Figure 2 Shown is a flow chart of a wind speed prediction method in one embodiment of the wind power wind speed prediction and optimization arrangement method described in this application.

[0034] Figure 3 Shown is a flow chart of S2 in the method for wind power wind speed prediction and optimal arrangement described in this application.

[0035] Figure 4 Shown is a neural network schematic diagram based on physical information in one embodiment of the wind power wind speed prediction and optimization arrangement method described in this application.

[0036] Figure 5Shown is a flow chart of S3 in the method for wind power wind speed prediction and optimal arrangement described in this application.

[0037] Figure 6 Shown is a flow chart of S4 in one embodiment of the wind power wind speed prediction and optimization arrangement method described in the present application.

[0038] Figure 7 Shown is a flow chart of S5 in one embodiment of the wind power wind speed prediction and optimization arrangement method described in the present application.

[0039] Figure 8 Shown is the arrangement of wind farm lidar sensors and wind turbine locations in one embodiment of the wind speed prediction and optimization arrangement method described in this application.

[0040] Fig. 9 Shown is a schematic diagram of the principle structure of a wind power wind speed prediction and optimization arrangement system in one embodiment of the present application.

[0041] Fig.10 Shown is a schematic diagram of the principle structure of a wind power wind speed prediction and optimization arrangement device in one embodiment of the present application.

[0042] Component number description

[0043] 91 Get Module

[0044] 92 Model building modules

[0045] 93 Wind speed acquisition module

[0046] 94 Model training module

[0047] 95 Wind speed reconstruction module

[0048] 96 Optimize the layout of modules

[0049] 101 Processor

[0050] 102 Memory DETAILED DESCRIPTION

[0051] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0052] It should be noted that the illustrations provided in the following embodiments are only used to illustrate the basic concept of the present application in a schematic manner. Therefore, the drawings only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0053] A method, system and device for predicting wind speed and optimizing arrangement of wind power provided in the embodiments of the present application will be described in detail below in conjunction with the drawings in the embodiments of the present application.

[0054] See also Figure 1 and Figure 2 , respectively showing a flow chart of a wind power wind speed prediction and optimization arrangement method in one embodiment of the present application and a flow chart of a wind speed prediction method in one embodiment of the wind power wind speed prediction and optimization arrangement method in the present application. Figure 1 and Figure 2 As shown, this embodiment provides a method for predicting wind speed and optimizing the arrangement of wind power.

[0055] The wind power wind speed prediction and optimization arrangement method specifically includes the following steps:

[0056] S1, obtaining basic wind speed data of the target area, wherein the basic wind speed data includes but is not limited to: wind farm size, location information, wind speed data, wind direction, temperature, air pressure, terrain, water depth and other types of data.

[0057] In this embodiment, according to the research requirements, various types of data and information or materials such as location information, meteorological data and terrain data within the wind farm in the target area are collected.

[0058] Specifically, first determine the scope of the target area where data needs to be obtained; then determine the source of the data, such as: selecting official or authoritative agencies (such as: national or local government energy departments, meteorological bureaus, ocean bureaus, scientific research and academic institutions, commercial data providers, etc.), and obtain relevant data through the channels they provide; and statistically organize the obtained data.

[0059] For example: First, determine the specific location and scope of the wind farm for which data needs to be obtained. Then, obtain the precise location (such as longitude and latitude, etc.) and approximate size of the wind farm through map services (such as Google Maps, Amap, etc.) or marine geographic information systems (GIS). Next, obtain wind speed and direction data through acquisition equipment; such as: using buoys, wind towers, meteorological stations and other equipment for field observation to collect real-time wind speed and direction data; using satellite remote sensing technology and numerical simulation methods to obtain wind speed and direction data over a larger range and longer time scale; or purchasing processed and analyzed historical wind speed and direction data sets from data providers. In addition, meteorological data such as temperature and air pressure in the target area can also be obtained from the Meteorological Bureau or relevant agencies. These data are helpful for analyzing the operating environment and performance of the wind farm; using marine geological survey ships, sonar and other equipment to conduct detailed measurements of the seabed topography to obtain information such as water depth and seabed topography. Finally, the collected data is integrated to form complete basic data on wind speed for wind power.

[0060] It is important to ensure the accuracy and reliability of data during data collection, and to conduct multiple verifications and comparisons of important data. At the same time, attention should be paid to the timeliness of data, and the data set should be updated and maintained in a timely manner to reflect the latest wind farm conditions.

[0061] S2, constructing a neural network model based on physical information based on the wind power wind speed basic data. Figure 3 and Figure 4 , respectively showing a flow chart of S2 in the wind power wind speed prediction and optimization arrangement method described in this application and a neural network chart based on physical information in one embodiment of the wind power wind speed prediction and optimization arrangement method described in this application. Figure 3 and Figure 4 As shown, S2 comprises the following steps:

[0062] This application is modeled by using PINN (Physics-Informed Neural Networks).

[0063] PINN uses deep neural networks as nonlinear function approximators. With the help of automatic differentiation technology that is widely used in deep neural networks, the differential form constraints in the partial differential equations to be solved are integrated into the loss function of the neural network for optimization, thereby learning a deep neural network model that approximates the solution of the partial differential equation with physical model constraints.

[0064] S21, extracting the size of the wind farm based on the basic wind speed data;

[0065] S22, defining a two-dimensional velocity field according to the actual size of the wind farm.

[0066] In this embodiment, wind speed data is selected from the wind power wind speed basic data; and the wind speed data is preprocessed to obtain preprocessed wind speed data; combined with defining a two-dimensional grid according to the actual wind field size, the wind sound is simulated through the two-dimensional grid to obtain a two-dimensional velocity field.

[0067] Specifically, wind speed data of the area where the wind farm is located at different time scales are selected from the basic wind power wind speed data; these data should cover the wind farm area and include wind speed changes at different time scales as much as possible. Then clean the collected wind speed data to remove outliers and noise; the data can be interpolated as needed to ensure the integrity and continuity of the data in the wind farm area. Further, analyze the wind farm characteristics, that is, analyze the spatiotemporal variation characteristics of wind speed, including: average wind speed, wind direction, wind speed fluctuations, etc., to identify the dominant wind direction and wind speed change area in the wind farm. Finally, based on the analysis results of the wind farm data, define a two-dimensional grid to simulate the wind farm; the resolution of the grid should be determined according to the research needs and data accuracy, and should usually be fine enough to capture subtle changes in the wind farm; at each grid point, assign corresponding speed values ​​according to the wind speed data to form a two-dimensional velocity field.

[0068] S23, designing a neural network structure based on the two-dimensional velocity field;

[0069] S24, integrating the physical constraints into the loss function of the neural network structure for optimization, and obtaining a neural network model based on physical information. The steps include: using automatic differentiation technology to calculate the partial derivatives of the partial differential equation; using physical equations to integrate the physical constraints into the loss function in the neural network; combining the partial derivatives of the partial differential equation with the loss function to obtain the total loss function. The physical equations include: two-dimensional Navier-Stokes equations, continuity equations and turbulence models.

[0070] In this embodiment, first, the physical equation describing the wind field is determined, and the relationship between variables such as wind speed, wind direction, and air pressure is reflected through the physical direction. Then, according to the requirements of the physical model, the PINN neural network structure is designed; and the constraints and loss functions of the physical model are embedded in the PINN neural network structure.

[0071] Specifically, determine the physical equations that describe the wind field; design the structure of the PINN according to the requirements of the physical model, including: input layer, hidden layer and output layer; the input layer receives the aforementioned two-dimensional velocity field data, the hidden layer includes enough neurons and layers to obtain the nonlinear relationship of the wind speed data; the output layer outputs the predicted wind speed value. Furthermore, in the training process of PINNs, the constraints or loss functions of the physical model are embedded by defining the physical information error term. These constraints can be based on the residual of the physical equation or the data-driven loss function to ensure that the predicted value output by the network conforms to the physical laws.

[0072] For example: first clarify the input and output of the model. For the two-dimensional velocity field in this embodiment, the input is the two-dimensional grid data of the velocity field (such as: the components of the velocity vector in the x and y directions), and the output is the predicted value of the velocity field, the velocity field feature extraction and other physical quantities. Design the neural network structure of PINN, use the deep neural network as a nonlinear function approximator, and use the automatic differentiation technology widely used in deep neural networks to integrate the differential form constraints in the partial differential equations to be solved into the loss function of the neural network for optimization, so as to learn and obtain a deep neural network model that approximates the solution of the partial differential equation with physical model constraints. Then, use the random initialization strategy to initialize the network weights and groups of the PINN neural network. Next, determine the physical constraints that the velocity field should satisfy, and use physical equations to integrate the physical constraints into the loss function in the neural network; combine the partial derivatives of the partial differential equations with the loss function to obtain the total loss function.

[0073] Furthermore, the total loss function includes two parts: data fitting data term loss and physical term loss. The data fitting term is used to measure the difference between the model prediction value and the actual observation value, such as the mean square error; the physical constraint term is used to penalize the prediction results that do not meet the physical constraints, which can be achieved by quantifying the degree of violation of the physical constraints as a penalty term and adding it to the loss function.

[0074] The calculation formula of the physical equation is:

[0075]

[0076] Where ρ represents the fluid density; u and v represent the measured velocity components of the fluid at point (x, y) at time t.

[0077] The calculation formula of the total loss function is:

[0078] L=ω1L d +ω2L f (4)

[0079] L d =u MRMSE +v MRMSE (5)

[0080]

[0081] Among them, L represents the total loss function; L d Indicates data item loss; L f represents the physical item loss, i.e., PDE residual loss; ω1 represents the weight of the data item; ω2 represents the weight of the PDE physical item; n represents the total number of laser sensors set; NT represents the total number of time steps; u, v represent the measured velocity components of the fluid at point (x, y) at time t, that is, the measured values ​​at the corresponding positions; represents the predicted value of the corresponding velocity component of the fluid at point (x, y) at time t; MRMSE represents the mean root mean square error; G represents the implicit expression of Navier-Stokes.

[0082] S3, based on the physical information-based neural network model, obtain the wind speed at a specific location in the velocity field and create a training data set. Figure 5 , which is a flow chart of S3 in the method for wind power wind speed prediction and optimization arrangement described in this application. Figure 5 As shown, S3 comprises the following steps:

[0083] S31, determine the target study area and specific location of the wind farm;

[0084] S32, using laser sensors distributed in the wind farm to obtain wind speeds at specific locations in a number of speed fields;

[0085] S33, performing data preprocessing on the wind speed at the specific position in the velocity field to obtain the preprocessed wind speed at the specific position in the velocity field;

[0086] S34, dividing the continuous time series data into a plurality of time windows of fixed length;

[0087] S35, defining a label for each time window;

[0088] S36, constructing a training data set based on the wind speed at a specific location in the preprocessed velocity field and the corresponding label, and using it as a supervision point for training the model.

[0089] In this embodiment, first, it is necessary to clarify the specific area of ​​the wind farm to be studied and the specific position in the velocity field, where the specific area includes: geographical location, seabed topography, surrounding environment, etc.; the specific position refers to the foundation position of the wind turbine, the boundary of the wind farm or an area with special physical properties, etc.

[0090] Then, laser sensors distributed in the wind farm are used to collect historical wind speed data, meteorological data (such as wind direction, temperature, humidity, air pressure, altitude, etc.) and physical parameters (such as terrain height, ocean current field, temperature gradient, etc.) of these specific locations and their surrounding areas.

[0091] Next, the wind speed at a specific location in the velocity field is cleaned, and outliers, missing values, or unreasonable data points are removed to ensure the quality of the data, thereby obtaining the wind speed at a specific location in the velocity field after preprocessing.

[0092] Then, the continuous time series data is divided into multiple time windows of fixed length. Each time window contains data of multiple time steps, which is used to predict the wind speed at a certain time step in the future. Then, a label is assigned to each time window, that is, the future wind speed value to be predicted, that is, the wind speed at one or more time steps after the time window. Then, the wind speed at a specific location in the preprocessed speed field and the corresponding label are used to construct a training data set, a validation data set, and a test data set, so as to perform cross-validation and final evaluation during the training process.

[0093] S4, training the neural network model based on physical information based on the training data set. Figure 6 , which is a flow chart of S4 in the method for wind power wind speed prediction and optimization arrangement described in this application. Figure 6 As shown, S4 comprises the following steps:

[0094] S41, using Adams optimizer to train the training data set once to obtain a neural network model based on physical information after the loss value is reduced.

[0095] In this embodiment, the weights and biases of the neural network are initialized first; then the parameters of the Adams optimizer are configured; and then the samples in the training data set are used to train the neural network model based on physical information.

[0096] Specifically, the training data set is iterated, and each iteration includes four steps: forward propagation, loss calculation, back propagation, and parameter update. Including:

[0097] In the forward propagation phase, the model’s predictions for a given input are calculated.

[0098] During the loss calculation phase, a composite loss function that incorporates physical constraints is used to evaluate the difference between the predicted and actual values.

[0099] In the back-propagation phase, the weights and biases of the model are updated according to the gradient information of the loss function.

[0100] The Adam optimizer is used to adaptively adjust the learning rate based on the gradient information and update the model parameters.

[0101] It should be noted that during the training process, the changes in the loss function should be monitored regularly to ensure that the loss value is gradually decreasing; at the same time, the validation set can be used to evaluate the generalization ability of the model to avoid overfitting or underfitting.

[0102] S42, using the L-BFGS optimizer for secondary training until the convergence condition is met, and a trained neural network model based on physical information is obtained.

[0103] In this embodiment, the parameters of the L-BFGS optimizer are first adjusted as needed, and the upper limit of the number of iterations is set; then the L-BFGS optimizer is configured, and then training is performed.

[0104] Specifically, configure the L-BFGS optimizer in the deep learning framework and prepare an optimization closure function that will calculate the loss and return the gradient; then, use the L-BFGS optimizer for iterative optimization until the convergence condition is met (such as: the value of the total loss function is less than the preset threshold, etc.); in each iteration, the L-BFGS optimizer will automatically adjust the step size and update the weights and biases of the neural network in the optimal direction. Check whether the convergence condition is met after each iteration. If it is met, stop training; otherwise, continue to the next round of iteration. Finally, the test data set can be used to test and evaluate the model.

[0105] S5, through the trained neural network model based on physical information, reconstruct the wind speed distribution in the speed field and predict the wind speed forecast value of the wind farm in the next period. Figure 7 , which is a flow chart of S5 in the method for wind power wind speed prediction and optimization arrangement described in this application. Figure 7 As shown, S5 comprises the following steps:

[0106] S51, taking the position coordinates of the velocity field as the target point for wind speed reconstruction, and inputting them into the trained neural network model based on physical information, to obtain the wind speed prediction value corresponding to the position coordinates of the velocity field;

[0107] S52, mapping the wind speed prediction value onto a grid of the velocity field to form a reconstructed wind speed distribution map;

[0108] S53, analyzing the reconstructed wind speed distribution map to obtain the analyzed wind speed distribution result, and optimizing and fine-tuning the trained neural network model based on physical information;

[0109] S54, obtaining the real-time wind speed at the target location within a certain period of time;

[0110] S55, preprocessing the real-time wind speed within a certain period of time to obtain the real-time wind speed within the certain period of time after preprocessing;

[0111] S56, inputting the real-time wind speed in a certain period after the preprocessing into the trained neural network model based on physical information for prediction, and obtaining the predicted wind speed value of the wind farm in the next period.

[0112] In this embodiment, first, the position coordinates of the velocity field are used as the target points for wind speed reconstruction and are input into the trained neural network model based on physical information. The model performs forward propagation according to the input data and outputs the wind speed prediction values ​​at the corresponding positions. These prediction values ​​can reflect the wind speed distribution of each point in the velocity field. Then, the wind speed prediction values ​​are mapped to the grid of the velocity field to form a reconstructed wind speed distribution map. Next, the reconstructed wind speed distribution map is analyzed to check whether it conforms to physical laws and actual observations; if deviations or inconsistencies are found, the model can be fine-tuned or retrained to improve the accuracy of wind speed reconstruction. At the same time, the input data, network structure, parameter settings, etc. of the model can also be optimized according to the reconstruction results to further improve the performance of the model.

[0113] Furthermore, the real-time wind speed of the target location within a certain period of time is obtained. The real-time wind speed within a certain period of time is then cleaned and normalized; the processed real-time wind speed within a certain period of time is then input into the trained neural network model based on physical information, the prediction time range is set, and the wind speed prediction value within the time period is output. Finally, error indicators (such as mean square error, mean absolute error, etc.) are used to quantify the prediction effect. According to the evaluation results, the parameters of the model are adjusted to improve the prediction accuracy.

[0114] S6, optimizing the arrangement of the wind turbine group according to the wind speed distribution in the reconstructed velocity field.

[0115] In this embodiment, a fan control strategy is formulated based on the wind speed prediction value; the fan control strategy is used to adjust the fan in real time through a control system.

[0116] Specifically, based on the predicted wind speed data, combined with the actual situation of the wind farm (such as wind turbine type, quantity, layout, etc.), a wind turbine control strategy is formulated. The wind turbine control strategy includes but is not limited to: adjusting the wind turbine's power output, blade angle, speed and other parameters to maximize the wind farm's power generation efficiency and reduce wear, etc. Then, the formulated wind turbine control strategy is applied to the corresponding wind farm, and the wind turbine is adjusted in real time through the control system. Next, the operation of the wind farm after the implementation of the control strategy is monitored, including power generation, wind turbine operating status, etc.; then, based on the monitoring results, the control strategy is evaluated and adjusted to continuously optimize the operating efficiency of the wind farm.

[0117] It should be noted that the data trained in this application is provided by the coherent Doppler wind laser radar distributed in the wind farm. The coherent wind laser radar uses a single-frequency narrow-linewidth laser to emit laser pulses, which interact with aerosol particles in the atmosphere to generate echo signals with Doppler frequency shifts. The atmospheric echo signals collected by the telescope system are coherently beat with the intrinsic laser signal to complete the extraction of Doppler frequency shift information, and the three-dimensional wind field information of the atmosphere is obtained by the vector wind speed inversion technology. The results of the mesoscale numerical simulation are used as the supervision point for PINN training. The mesoscale numerical simulation can fit the characteristics of airflow, air pressure, wind field, etc. in a larger area, but it cannot accurately evaluate the local wind flow conditions in the wind farm. PINN obtains accurate internal area conditions through microscale simulation by adding the fluid mechanics control equation to the loss function, similar to the principle of CFD. Finally, the data observed by the laser wind radar and the results of the PINN numerical simulation are combined to improve the accuracy of the evaluation and reflect the real wind field conditions.

[0118] PINN implements physical constraints by adding physical information to the loss function, such as physical formulas, boundary conditions or initial conditions. During the training process, appropriate weights are selected to minimize the loss function, thereby obtaining a model with a small error between the actual value and the actual value and in accordance with the laws of physics. Through the trained model, the data of wind speed at a specific location in the next period of time is obtained. Finally, by optimizing the wind turbine control strategy, it provides guarantee for the intelligent operation and maintenance of wind farms. By reconstructing the flow field based on measured data, the influence of wind turbine location arrangement on the power generation efficiency of wind farms can also be studied.

[0119] In addition, the PINN proposed in this application is based on the reproduction of the velocity distribution in the two-dimensional wind field. Subsequently, multiple neural networks can be established according to different elevations to describe the entire three-dimensional wind field, which is more accurate and reasonable.

[0120] The following is an explanation using the wind speed prediction of a wind farm in a certain area as an example.

[0121] See also Figure 8 , showing the wind farm lidar sensor and wind turbine position arrangement in one embodiment of the wind power wind speed prediction and optimization arrangement method described in this application.

[0122] First, in a certain area, lidar sensors are deployed within a certain distance in front of and behind the wind turbines to monitor the wind speed and pressure distribution at their locations, serving as supervision points for neural network learning.

[0123] Then, a feedforward neural network is built, and the definition domain is constructed with a two-dimensional velocity field. The neural network takes the position parameter horizontal coordinate x, vertical coordinate y, and time parameter t as input, and the velocity parameters horizontal horizontal u, horizontal vertical v, and pressure p as output. The size of the definition domain can be set according to the size of the wind farm. Take a wind farm as an example, with a 15Mw wind turbine, a wind turbine hub height of 150m, and a wind rotor diameter of 242m. The definition domain size can be set to 10.6Km×4Km.

[0124] Automatic differentiation is then used to calculate the partial derivatives required for the residual of the partial differential equation.

[0125] Next, physical constraints are added to build a neural network based on physical information. The physical equations in this application refer to the two-dimensional Navier-Stokes equations, the continuity equation, and the turbulence model.

[0126] The calculation formula of the physical equation is:

[0127]

[0128] Where ρ represents the fluid density; u and v represent the measured velocity components of the fluid at point (x, y) at time t.

[0129] Then, the partial derivative residual loss is combined with the data loss to get the total loss.

[0130] That is, the calculation formula of the total loss function is:

[0131] L=ω1L d +ω2L f

[0132] L d =u MRMSE +v MRMSE

[0133]

[0134] Among them, L represents the total loss function; L d Indicates data item loss; L f represents the physical item loss, i.e., PDE residual loss; ω1 represents the weight of the data item; ω2 represents the weight of the PDE physical item; n represents the total number of laser sensors set; N T represents the total number of time steps; u, v represent the measured velocity components of the fluid at point (x, y) at time t, that is, the measured values ​​at the corresponding positions; represents the predicted value of the corresponding velocity component of the fluid at point (x, y) at time t; MRMSE represents the mean root mean square error; G represents the implicit expression of Navier-Stokes.

[0135] Next, the neural network is trained, and the weights and biases are continuously updated until the value of the total loss function meets the predetermined requirements (less than the preset value, such as:). The model is first trained using the Adams optimizer to achieve a rapid decrease in the loss value, and then trained using the L-BFGS optimizer until convergence.

[0136] Based on the training model, the two-dimensional velocity field can be reproduced, and the effective wind speed at the wind turbine rotor in the short term can be quickly predicted.

[0137] Finally, the fan control can be optimized according to the wind speed changes in the next time period.

[0138] The wind speed prediction and optimal arrangement method provided in the present application simulates the velocity distribution of the entire flow field through the laser radar sensor set in the wind farm. Based on measured data, it is more in line with actual working conditions; it predicts the wind speed changes within a period of time under the wind turbine position, provides a reference for the regular analysis of wind power in the wind farm, and can more comprehensively understand the physical characteristics of the wind farm. The use of laser wind radar mesoscale data simulation and PINN microscale numerical simulation can improve the accuracy of wind farm assessment; at the same time, by combining machine learning technology with a large amount of fluid dynamics data, accurate velocity field prediction results are provided, and the calculation efficiency is high. Similarly, the present application can use a small amount of flow field sampling data to reconstruct regional flow more accurately; and the PINN model can be used for wind farm modeling, which can provide accurate, ultra-short-term velocity field predictions, while taking into account calculation efficiency, and provide a reference for short-term pitch and yaw adjustments, so as to achieve safe and stable operation of wind power.

[0139] The protection scope of the wind power wind speed prediction and optimization arrangement method described in the embodiment of the present application is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing or replacing steps in the prior art based on the principles of the present application are included in the protection scope of the present application.

[0140] The embodiment of the present application also provides a wind power wind speed prediction and optimization arrangement system, which can implement the wind power wind speed prediction and optimization arrangement method described in the present application, but the implementation device of the wind power wind speed prediction and optimization arrangement method described in the present application includes but is not limited to the structure of the wind power wind speed prediction and optimization arrangement system listed in the present embodiment. All structural deformations and replacements of the prior art made according to the principles of the present application are included in the protection scope of the present application.

[0141] The wind speed prediction and optimization arrangement system for wind power provided in this embodiment will be described in detail below with reference to the drawings.

[0142] This embodiment provides a wind power wind speed prediction and optimization arrangement system, including:

[0143] See also Fig. 9 , which is a schematic diagram of the principle structure of the wind speed prediction and optimization arrangement system described in this application in one embodiment. Fig. 9 As shown, the wind power wind speed prediction and optimization arrangement system includes: an acquisition module 91, a model construction module 92, a wind speed acquisition module 93, a model training module 94, a wind speed reconstruction module 95 and an optimization arrangement module 96.

[0144] The acquisition module 91 is used to acquire basic wind speed data of the target area; the basic wind speed data of the wind power includes but is not limited to: wind farm size, wind speed, wind direction, temperature, air pressure, terrain, water depth and other data.

[0145] In this embodiment, according to research requirements, various types of data and information or materials such as location information, meteorological data and terrain data in the wind farm within the target area are collected to obtain basic wind speed data.

[0146] Specifically, first determine the scope of the target area where data needs to be obtained; then determine the source of the data, such as: selecting official or authoritative agencies (such as: national or local government energy departments, meteorological bureaus, ocean bureaus, scientific research and academic institutions, commercial data providers, etc.), and obtain relevant data through the channels they provide; and statistically organize the obtained data.

[0147] The model building module 92 is connected to the acquisition module 91 and is used to build a neural network model based on physical information based on the basic wind speed data of wind power.

[0148] In this embodiment, the size of the wind farm is extracted based on the basic wind speed data; and a two-dimensional velocity field is defined according to the actual size of the wind farm.

[0149] Select wind speed data from the wind power wind speed basic data; perform data preprocessing on the wind speed data to obtain preprocessed wind speed data; define a two-dimensional grid according to the actual wind field size, simulate the wind sound through the two-dimensional grid, and obtain a two-dimensional velocity field; design a neural network structure based on the two-dimensional velocity field; integrate the physical constraints into the loss function of the neural network structure for optimization to obtain a neural network model based on physical information. The following steps are included: use automatic differentiation technology to calculate the partial derivatives of the partial differential equation; use physical equations to integrate the physical constraints into the loss function in the neural network; combine the partial derivatives of the partial differential equation with the loss function to obtain the total loss function. Wherein, the physical equations include: two-dimensional Navier-Stokes equations, continuity equations and turbulence models.

[0150] In this embodiment, wind speed data is selected from the wind power wind speed basic data; and the wind speed data is preprocessed to obtain preprocessed wind speed data; combined with defining a two-dimensional grid according to the actual wind field size, the wind sound is simulated through the two-dimensional grid to obtain a two-dimensional velocity field. The physical equation describing the wind field is determined, and the relationship between variables such as wind speed, wind direction, and air pressure is reflected through the physical direction. Then, according to the requirements of the physical model, the PINN neural network structure is designed; and the constraints and loss functions of the physical model are embedded in the PINN neural network structure.

[0151] Specifically, determine the physical equations that describe the wind field; design the structure of the PINN according to the requirements of the physical model, including: input layer, hidden layer and output layer; the input layer receives the aforementioned two-dimensional velocity field data, the hidden layer includes enough neurons and layers to obtain the nonlinear relationship of the wind speed data; the output layer outputs the predicted wind speed value. Furthermore, in the training process of PINNs, the constraints or loss functions of the physical model are embedded by defining the physical information error term. These constraints can be based on the residual of the physical equation or the data-driven loss function to ensure that the predicted value output by the network conforms to the physical laws.

[0152] The wind speed acquisition module 93 is used to acquire the wind speed at a specific position in the velocity field based on the physical information-based neural network model and create a training data set.

[0153] In this embodiment, a target research area and a specific location of a wind farm are determined; wind speeds at specific locations in a plurality of velocity fields are obtained using laser sensors distributed in the wind farm; data preprocessing is performed on the wind speeds at specific locations in the velocity field to obtain preprocessed wind speeds at specific locations in the velocity field; continuous time series data are divided into a plurality of time windows of fixed length; a label is defined for each time window; a training data set is constructed based on the preprocessed wind speeds at specific locations in the velocity field and the corresponding labels, and is used as a supervision point for a training model.

[0154] The model training module 94 is used to train the physical information based neural network model based on the training data set.

[0155] In this embodiment, the training data set is trained once using the Adams optimizer to obtain a neural network model based on physical information after the loss value decreases. The L-BFGS optimizer is used for secondary training until the convergence condition is met to obtain a trained neural network model based on physical information.

[0156] In this embodiment, the weights and biases of the neural network are initialized first; then the parameters of the Adams optimizer are configured; and then the samples in the training data set are used to train the neural network model based on physical information. Then the parameters of the L-BFGS optimizer are adjusted as needed, and the upper limit of the number of iterations is set; then the L-BFGS optimizer is configured, and then training is performed.

[0157] Specifically, the training dataset is iterated, and each iteration includes four steps: forward propagation, loss calculation, back propagation, and parameter update. Including: In the forward propagation stage, the predicted value of the model under a given input is calculated. In the loss calculation stage, a composite loss function containing physical constraints is used to evaluate the difference between the predicted value and the actual value. In the back propagation stage, the weights and biases of the model are updated according to the gradient information of the loss function. The Adam optimizer is used to adaptively adjust the learning rate according to the gradient information and update the model parameters.

[0158] Specifically, configure the L-BFGS optimizer in the deep learning framework and prepare an optimization closure function that will calculate the loss and return the gradient; then, use the L-BFGS optimizer for iterative optimization until the convergence condition is met (such as: the value of the total loss function is less than the preset threshold, etc.); in each iteration, the L-BFGS optimizer will automatically adjust the step size and update the weights and biases of the neural network in the optimal direction. Check whether the convergence condition is met after each iteration. If it is met, stop training; otherwise, continue to the next round of iteration. Finally, the test data set can be used to test and evaluate the model.

[0159] The wind speed reconstruction module 95 is used to reconstruct the wind speed distribution in the speed field through a trained neural network model based on physical information, and predict the wind speed of the wind farm in the next period of time.

[0160] In this embodiment, the position coordinates of the velocity field are used as the target point for wind speed reconstruction and are input into the trained neural network model based on physical information to obtain the wind speed prediction value corresponding to the position coordinates of the velocity field; the wind speed prediction value is mapped to the grid of the velocity field to form a reconstructed wind speed distribution map; the reconstructed wind speed distribution map is analyzed to obtain the analyzed wind speed distribution result, and the trained neural network model based on physical information is optimized and fine-tuned; the real-time wind speed of the target position within a certain time period is obtained; the real-time wind speed within a certain time period is preprocessed to obtain the real-time wind speed within a certain time period after preprocessing; the real-time wind speed within a certain time period after preprocessing is input into the trained neural network model based on physical information for prediction to obtain the wind speed prediction value of the wind farm in the next time period.

[0161] The optimized arrangement module 96 is used to optimize the arrangement of the wind turbine group according to the wind speed distribution in the reconstructed speed field.

[0162] In this embodiment, a fan control strategy is formulated based on the wind speed prediction value; the fan control strategy is used to adjust the fan in real time through a control system.

[0163] The wind speed prediction and optimal arrangement model for wind power generation can build a wind speed prediction and optimal arrangement system, which can reasonably utilize big data and combine machine learning methods to achieve efficient prediction of wind speed distribution in wind farms, and provide ideas for efficient operation and maintenance of wind farms.

[0164] It should be noted that it should be understood that the division of the various modules of the above system is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the x module can be a separately established processing element, or it can be integrated in a certain chip of the above system for implementation. In addition, it can also be stored in the memory of the above system in the form of program code, and called and executed by a certain processing element of the above system. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0165] The above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital signal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0166] See also Fig.10 , which is a schematic diagram of the principle structure of the wind speed prediction and optimization arrangement device described in this application in one embodiment. Fig.10As shown, this embodiment provides a wind power wind speed prediction and optimized arrangement device, and the wind power wind speed prediction and optimized arrangement device includes: a processor 101 and a memory 102; the memory 102 is used to store computer programs; the processor 101 is connected to the memory 102, and is used to execute the computer program stored in the memory 102, so that the wind power wind speed prediction and optimized arrangement device performs the various steps of the wind power wind speed prediction and optimized arrangement method as described above.

[0167] Preferably, the memory may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0168] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0169] In summary, the wind power wind speed prediction and optimization arrangement method, system and device provided in this application have the following beneficial effects:

[0170] The wind speed prediction and optimal arrangement method provided in the present application simulates the velocity distribution of the entire flow field through the laser radar sensor set in the wind farm. Based on measured data, it is more in line with actual working conditions; it predicts the wind speed changes within a period of time under the wind turbine position, provides a reference for the regular analysis of wind power in the wind farm, and can more comprehensively understand the physical characteristics of the wind farm. The use of laser wind radar mesoscale data simulation and PINN microscale numerical simulation can improve the accuracy of wind farm assessment; at the same time, by combining machine learning technology with a large amount of fluid dynamics data, accurate velocity field prediction results are provided, and the calculation efficiency is high. Similarly, the present application can use a small amount of flow field sampling data to reconstruct regional flow more accurately; and the PINN model can be used for wind farm modeling, which can provide accurate, ultra-short-term velocity field predictions, while taking into account calculation efficiency, and provide a reference for short-term pitch and yaw adjustments, so as to achieve safe and stable operation of wind power.

[0171] The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. A method for predicting wind speed and optimizing the arrangement of wind power, characterized in that: The following steps are involved: Obtaining basic wind speed data of the target area; the basic wind speed data includes: any one or more combinations of wind farm size, location, wind speed, wind direction, temperature, air pressure, terrain, and water depth; Constructing a neural network model based on physical information based on the wind power wind speed basic data; Based on the physical information-based neural network model, obtain the wind speed at a specific location in the velocity field and create a training data set; Training the neural network model based on physical information based on the training data set; Through the trained neural network model based on physical information, the wind speed distribution in the speed field is reconstructed, and the wind speed forecast value of the wind farm in the next period is predicted; The arrangement of wind turbines is optimized according to the wind speed distribution in the reconstructed velocity field.

2. The method for wind power wind speed prediction and optimization arrangement according to claim 1, characterized in that: Constructing a neural network model based on physical information based on the wind power wind speed basic data includes the following steps: Extracting wind farm size based on the wind power wind speed basic data; According to the actual wind farm size, define a two-dimensional velocity field; Designing a neural network structure based on the two-dimensional velocity field; The physical constraints are integrated into the loss function of the neural network structure for optimization to obtain a neural network model based on physical information.

3. The method for wind power wind speed prediction and optimization arrangement according to claim 2 is characterized in that: Integrating physical constraints into the loss function of the neural network structure for optimization includes the following steps: Use automatic differentiation techniques to calculate partial derivatives of partial differential equations; Use physical equations to incorporate physical constraints into the loss function in the neural network; The partial derivatives of the partial differential equation are combined with the loss function to obtain the total loss function.

4. The method for wind power wind speed prediction and optimization arrangement according to claim 3 is characterized in that: The physical equations include: two-dimensional Navier-Stokes equations, continuity equations and turbulence models; The calculation formula of the physical equation is: Where ρ represents the fluid density; u and v represent the measured velocity components of the fluid at point (x, y) at time t.

5. The method for wind power wind speed prediction and optimization arrangement according to claim 3 is characterized in that: The calculation formula of the total loss function is: L=ω1L d +ω2L f Among them, L represents the total loss function; L d Indicates data item loss; L f represents the physical item loss, i.e., PDE residual loss; ω1 represents the weight of the data item; ω2 represents the weight of the PDE physical item; n represents the total number of laser sensors set; N T represents the total number of time steps; u, v represent the measured velocity components of the fluid at point (x, y) at time t, that is, the measured values ​​at the corresponding positions; Represents the predicted value of the corresponding velocity component of the fluid at point (x, y) at time t; MRMSE represents the mean root mean square error; G represents the implicit expression of Navier-Stokes.

6. The method for wind power wind speed prediction and optimization arrangement according to claim 1, characterized in that: Obtaining the wind speed at a specific location in the velocity field based on the physical information-based neural network model and creating a training data set includes the following steps: Identify target study areas and specific locations for wind farms; Using laser sensors distributed in the wind farm, wind speeds at specific locations in several velocity fields are obtained; Performing data preprocessing on the wind speed at the specific position in the velocity field to obtain the preprocessed wind speed at the specific position in the velocity field; Divide continuous time series data into multiple time windows of fixed length; Define labels for each time window; A training data set is constructed based on the wind speed at a specific position in the preprocessed velocity field and the corresponding label, and is used as a supervision point for the training model.

7. The method for wind power wind speed prediction and optimization arrangement according to claim 1, characterized in that: Training the neural network model based on physical information based on the training data set comprises the following steps: The training data set is trained once using an Adams optimizer to obtain a neural network model based on physical information after the loss value is reduced; The L-BFGS optimizer is used for secondary training until the convergence condition is met, and a trained neural network model based on physical information is obtained.

8. The method for wind power wind speed prediction and optimization arrangement according to claim 1, characterized in that: Reconstructing the wind speed distribution in the speed field through the trained neural network model based on physical information, and obtaining the wind speed forecast value of the wind farm in the next period through prediction includes the following steps: The position coordinates of the velocity field are used as the target points for wind speed reconstruction and input into the trained neural network model based on physical information to obtain the wind speed prediction value corresponding to the position coordinates of the velocity field. Mapping the wind speed prediction value onto a grid of the velocity field to form a reconstructed wind speed distribution map; Analyzing the reconstructed wind speed distribution map to obtain analyzed wind speed distribution results, and optimizing and fine-tuning the trained neural network model based on physical information; Get the real-time wind speed at the target location within a certain period of time; Preprocessing the real-time wind speed in a certain period of time to obtain the real-time wind speed in the certain period of time after preprocessing; The real-time wind speed in a certain period after the preprocessing is input into the trained neural network model based on physical information for prediction, so as to obtain the predicted wind speed value of the wind farm in the next period.

9. A wind power wind speed prediction and optimization arrangement system, characterized in that: include: An acquisition module is used to obtain basic wind speed data of the target area; The basic wind speed data of wind power generation includes: any one or more combinations of wind farm size, wind speed, wind direction, temperature, air pressure, terrain, and water depth; A model building module, used to build a neural network model based on physical information based on the basic wind speed data of wind power; A wind speed acquisition module, used to acquire the wind speed at a specific location in the velocity field based on the physical information-based neural network model, and create a training data set; A model training module, used for training the neural network model based on physical information based on the training data set; The wind speed reconstruction module is used to reconstruct the wind speed distribution in the speed field through the trained neural network model based on physical information, and predict the wind speed of the wind farm in the next period; The optimized arrangement module is used to optimize the arrangement of wind turbine groups according to the wind speed distribution in the reconstructed velocity field.

10. A wind power wind speed prediction and optimization arrangement device, characterized in that: include: Processor and memory; The memory is used to store computer programs; The processor is connected to the memory and is used to execute the computer program stored in the memory so that the wind power wind speed prediction and optimization arrangement device executes the wind power wind speed prediction and optimization arrangement method according to any one of claims 1 to 8.

Images