Wind power plant virtual laser radar wind measurement method and system

Abstract

The invention relates to a virtual laser radar wind measurement method and system for a wind power plant, and the method comprises the steps: obtaining the actual wind condition data of a first unit and a second unit, carrying out the full-wind-direction wind field CFD simulation to obtain a simulation result, and correction is carried out based on the simulation result and the two groups of actual wind regime data to obtain corrected wind regime data. Then, a prediction model is constructed using the corrected and actual wind regime data. By inputting actual wind regime data and correction data of the second unit to the model, virtual radar prediction data of the second unit can be obtained. According to the invention, through combination of CFD simulation, laser radar data and a machine learning algorithm, radar preset data of a unit position where a laser radar is not installed in a wind power plant is realized. The deployment number of the laser radars is reduced, so that the construction and operation cost of the wind power plant can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a wind farm virtual laser radar wind measurement method and system. Background Art

[0002] Existing wind speed prediction technology has limitations. The intermittent and unpredictable nature of wind speed poses challenges to wind power generation. Although traditional physical models and statistical models have certain applications, they are not accurate enough due to the large amount of calculations and difficulty in handling complex terrain or nonlinear relationships.

[0003] Traditional wind speed measurement equipment, fixed wind towers, have a long history of use in wind farms, but they can only provide wind speed and wind direction data at fixed heights and locations, and cannot capture wind speed changes at different heights and wind directions. For wind farms with complex terrain, fixed wind towers cannot provide comprehensive wind field data and can only observe within a limited range. Although cheaper than lidar equipment, its installation and maintenance costs are still high, and the observation point cannot be moved or flexibly adjusted.

[0004] In order to make up for the problem that fixed wind towers cannot adapt to complex terrain and have a small measurement range, LiDAR is widely used in various wind farms to monitor the incoming wind speed and direction in real time. As a new generation of wind measurement equipment, floating LiDAR can achieve a wide range of wind speed measurement, especially suitable for offshore wind farms or terrain where it is difficult to install fixed towers. It can accurately measure wind speed and direction at a long distance without being restricted by the location of the measurement point. However, the equipment is expensive and the maintenance cost is high, so it is not installed on a large scale, especially at sea or in severe weather conditions. When the waves or wind force change greatly, the stability of the floating equipment will affect the measurement accuracy and the data may be disturbed. Summary of the invention

[0005] The technical problem to be solved by the present invention is the high installation cost of a floating laser radar, and a wind farm virtual laser radar wind measurement method and system are provided.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a wind farm virtual laser radar wind measurement method, the method comprising the following steps: Acquire first actual wind condition data of a first unit and second actual wind condition data of a second unit; wherein the first unit is a unit equipped with a laser radar; and the second unit is a unit not equipped with a laser radar; Carry out CFD simulation of the wind farm in all wind directions for the wind farm to obtain CFD simulation results; Correcting the first actual wind condition data and the second actual wind condition data based on the CFD simulation result to obtain first corrected wind condition data and second corrected wind condition data; constructing a prediction model according to the first actual wind condition data, the first revised wind condition data, the second actual wind condition data and the second revised wind condition data; The second actual wind condition data and the second revised wind condition data are used as input, and based on the prediction model, virtual radar prediction data of the second unit is obtained.

[0007] In one embodiment, performing CFD simulation of a wind farm in all wind directions for the wind farm to obtain CFD simulation results includes: Performing correlation analysis on the first actual wind condition data and the second actual wind condition data according to different wind direction intervals; The first wind turbine that meets the preset correlation threshold is determined, and a CFD simulation of the wind farm in all wind directions is performed on the wind farm to obtain a CFD simulation result.

[0008] In one embodiment, the determining of the first wind turbine generator set that meets the preset correlation threshold and performing CFD simulation of a wind farm in all wind directions for the wind farm comprises: The SST k-ω model is adopted to perform CFD simulation of all wind directions of the wind farm based on the first wind turbine group that meets the preset correlation threshold.

[0009] In one embodiment, the virtual radar prediction data of the second wind turbine generator set includes virtual radar data of the second wind turbine generator set to be tested; Based on the CFD simulation result, the first actual wind condition data and the second actual wind condition data are corrected respectively to obtain the first corrected wind condition data and the second corrected wind condition data, including: A second wind turbine set that meets a preset distance and a preset correlation threshold is determined as the second wind turbine set to be tested, and is corrected according to the CFD simulation result and the first actual wind condition data and the second actual wind condition data to obtain first corrected wind condition data and second corrected wind condition data.

[0010] In one embodiment, the virtual radar prediction data of the second unit also includes virtual radar data of other second units; The constructing of a prediction model according to the first actual wind condition data, the first revised wind condition data, the second actual wind condition data and the second revised wind condition data comprises: Establishing a first model for predicting virtual radar data of the second wind turbine to be tested by learning a mapping relationship between the first actual wind condition data and the first revised wind condition data; Establishing a second model for predicting virtual radar data of other second wind turbines by learning the virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second revised wind condition data; The step of taking the second actual wind condition data and the second revised wind condition data as input and obtaining the virtual radar prediction data of the second unit based on the prediction model comprises: Using the actual wind condition data of the second wind turbine as input of the first model to obtain virtual radar data of the second wind turbine to be tested; The virtual radar data of the second wind turbine group to be tested, the second actual wind condition data and the second corrected wind condition data are used as inputs of the second model to obtain the virtual radar data of the other second wind turbine group.

[0011] In one embodiment, the step of using the virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second corrected wind condition data as inputs of the second model to obtain the virtual radar data of the other second turbines comprises: The virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second corrected wind condition data are used as inputs of the second model and the virtual radar data of the other second turbines are obtained using an inverse proportional distance weighting method.

[0012] In one embodiment, constructing a prediction model based on the first actual wind condition data, the first revised wind condition data, the second actual wind condition data, and the second revised wind condition data further includes: Calculating an error evaluation index value of the first model and an error evaluation index value of the second model; Evaluating the prediction accuracy of the first model according to the error evaluation index value of the first model; The prediction accuracy of the second model is evaluated according to the error evaluation index value of the second model.

[0013] In one embodiment, performing correlation analysis on the first actual wind condition data and the second actual wind condition data according to different wind direction intervals includes: Calculating the Pearson correlation coefficient between the first actual wind condition data and the second actual wind condition data in different wind direction intervals; The step of determining the first wind turbine generator set that meets the preset correlation threshold comprises: A first wind turbine group that meets a preset correlation threshold is determined according to the Pearson correlation coefficient.

[0014] In one embodiment, performing correlation analysis on the first actual wind condition data and the second actual wind condition data according to different wind direction intervals includes: Segmenting the first actual wind condition data and the second actual wind condition data based on a preset wind direction interval; According to different wind direction intervals, correlation analysis is performed on the first actual wind condition data and the second actual wind condition data.

[0015] The present application also provides a wind farm virtual laser radar wind measurement system, comprising a processor and a memory storing a computer program, wherein the processor implements the steps of any of the above-mentioned wind farm virtual laser radar wind measurement methods when executing the computer program.

[0016] The implementation of the present invention has the following beneficial effects: The present application obtains the actual wind condition data of the first unit and the second unit, wherein the first unit is a unit equipped with a laser radar, and the second unit is a unit without a laser radar installed, performs a CFD simulation of the wind field in all wind directions to obtain simulation results, and makes corrections based on the simulation results and the two sets of actual wind condition data to obtain the corrected wind condition data. Subsequently, a prediction model is constructed using these corrections and actual wind condition data. By inputting the actual wind condition data and the corrected data of the second unit into the model, the virtual radar prediction data of the second unit can be obtained. The present invention combines CFD simulation, laser radar data and machine learning algorithms to realize the radar preset data for the unit position of the wind farm where the laser radar is not installed. Since the number of laser radars deployed is reduced, the present invention also helps to reduce the construction and operation costs of wind farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 This is a flow chart of an embodiment of a virtual laser radar wind measurement method for a wind farm of the present application; Figure 2 is a flow chart of another embodiment of the wind farm virtual laser radar wind measurement method of the present application; Figure 3 This is a schematic diagram of the wind turbine location and LiDAR measurement points in this application; Figure 4 It is a visual graph of the CFD simulation result data of the wind farm in this application; Figure 5 This is the LSTM network structure framework diagram of this application. DETAILED DESCRIPTION

[0019] The present application is further described in detail below in conjunction with the accompanying drawings and examples. It is particularly noted that the following examples are only used to illustrate the present application, but are not intended to limit the scope of the present application. Similarly, the following examples are only some embodiments of the present application rather than all embodiments, and all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0020] Aiming at the technical problems existing in the background technology, namely, the insufficient wind speed prediction accuracy and the high cost and poor flexibility of wind speed measurement equipment, the present invention provides a wind farm virtual laser radar wind measurement method and system, which predicts the virtual radar data of the laser radar unit not installed by combining CFD computational fluid dynamics simulation, actual laser radar actual measurement data and machine learning algorithm. Thus, the problems of low wind speed prediction accuracy and high measurement cost are solved.

[0021] like Figure 1 As shown, the technical method of the present application provides a virtual laser radar wind measurement method for a wind farm, wherein the method comprises the following steps: Acquire first actual wind condition data of a first unit and second actual wind condition data of a second unit; wherein the first unit is a unit equipped with a laser radar; and the second unit is a unit not equipped with a laser radar; Carry out CFD simulation of wind farms in all wind directions and obtain CFD simulation results; Based on the CFD simulation result and the first actual wind condition data and the second actual wind condition data, correction is performed to obtain first corrected wind condition data and second corrected wind condition data; Constructing a prediction model based on the first actual wind condition data, the first revised wind condition data, the second actual wind condition data, and the second revised wind condition data; The second actual wind condition data and the second revised wind condition data are used as input, and based on the prediction model, virtual radar prediction data of the second unit is obtained.

[0022] It should be noted that the present invention can be applied to offshore wind farms or terrains where it is difficult to install fixed towers. The method first collects the measured data of the wind speed and direction instrument of the wind farm and the measured data of the installed laser radar, and simultaneously obtains the position information and unit parameters of the wind farm. Collect the first actual wind condition data and the second actual wind condition data of the wind farm. These data can be the original data of wind speed and wind direction, or they can be the decomposed wind speed component data (such as ux, uy), where the wind information contains both wind speed and wind direction, and the situations of different incoming wind directions are very different. When performing deep learning modeling, the periodicity of wind direction needs to be considered, such as 0° and 360° are the same. In order to facilitate wind speed prediction, the horizontal wind speed needs to be decomposed, and then modeling and prediction are performed based on the wind speed components. The wind speed decomposition formulas are shown in (1) and (2). The first unit is a unit equipped with a laser radar and anemometer, and the second unit is a unit equipped with anemometer. The first actual wind condition data includes anemometer data and laser radar data, and the second actual wind condition data includes anemometer data. The anemometer data is marked as T, the laser radar data is marked as L, the position where the laser radar is installed is marked as install, and the position where the laser radar is not installed is marked as unknown. The wind speed and direction data of the first unit are: The wind speed and direction data of the second unit, The radar data for the first unit is Virtual radar data of the second unit. Next, carry out CFD simulation of the wind farm in all wind directions and establish a CFD database for the wind farm in all wind directions. , Correction is performed to obtain the corrected wind speed time history information of the entire field. The first model extends it to the entire field by establishing a deep learning relationship between the corrected data and the measured data; the second model further expands on this basis, and establishes a virtual position lidar data intelligent wind measurement technology based on the inverse proportional distance weighting method according to the position distance between the input data and the virtual position to be measured. This technology can realize real-time wind measurement of virtual lidar. The present invention effectively improves the accuracy and flexibility of wind speed measurement in wind farms and reduces costs.

[0023] The wind speed decomposition formula is as follows: (1) (2) In the formula, U represents the wind speed; It represents wind direction and is defined as the wind rose diagram, which is the angle between the wind direction and the north wind in the clockwise direction, and is 0° when the north wind is blowing. Ux and Uy represent the wind speed components along the east and north directions, respectively.

[0024] After the wind speed component is predicted, it is necessary to reversely calculate the wind speed and wind direction using formulas (3) and (4).

[0025] (3) (4) Where acos represents the arc cosine function.

[0026] like Figure 2 and Figure 3 As shown, further, CFD simulation of the wind farm in all wind directions is carried out for the wind farm, and the CFD simulation results include: Performing correlation analysis on the first actual wind condition data and the second actual wind condition data in different wind direction intervals; The first wind turbine that meets the preset correlation threshold is determined, and a CFD simulation of the wind farm in all wind directions is performed on the wind farm to obtain a CFD simulation result.

[0027] Specifically, the first actual wind condition data and the second actual wind condition data of the wind farm are collected. When these data are decomposed wind speed component data (such as ux, uy). According to different wind direction intervals, the first actual wind condition data and the second actual wind condition data are subjected to correlation analysis. The correlation analysis can be based on statistical methods, such as the Pearson correlation coefficient or the Spearman rank correlation coefficient, to determine the similarity or consistency between the two sets of data. Through the correlation analysis, the first wind turbine that meets the preset correlation threshold is determined. The data of these units will be used for subsequent CFD simulations to ensure the accuracy and reliability of the simulation results.

[0028] The calculation formula of Pearson correlation coefficient is shown in (5): (5) In the formula, R(X,Y) represents the Pearson correlation coefficient; n represents the length of the data; represents the average value of the sequence X; Represents the mean value of the series Y.

[0029] The calculation of the Pearson correlation coefficient requires that the data satisfy the normal distribution and the gap between the data cannot be too large. When the random variable does not satisfy the normal distribution, that is, does not meet the Pearson correlation coefficient hypothesis test conditions, the Spearman correlation coefficient can be used for evaluation. The calculation formula is as shown in formula (6).

[0030] (6) In the formula, d i It represents the rank difference between two groups of variables after they are sorted from small to large, and n is the length of the data.

[0031] like Figure 4As shown, further, determining the first wind turbine group that meets the preset correlation threshold, and performing a wind farm CFD simulation of all wind directions for the wind farm include: The SST k-ω model is used to perform CFD simulation of all wind directions of the wind farm based on the first wind turbine that meets the preset correlation threshold.

[0032] Specifically, among them Figure 4 As shown, 10-L, 14-L, and 18-L represent the corresponding laser radars. This solution uses the SST k-ω turbulence model when performing computational fluid dynamics (CFD) simulations. This model performs well in simulating turbulent flows in wind farms due to its excellent ability to capture complex flow characteristics. In order to analyze the characteristics of the wind farm more carefully, the wind direction is subdivided into 16 intervals, each of which accurately covers 22.5°, and detailed CFD simulations are performed in each of these wind direction intervals. At the same time, in order to ensure comprehensive coverage and high accuracy of the data, the wind speed range is set to 3 to 12 meters per second (m / s), and 19 groups of different wind speeds are set for simulation at intervals of 0.5m / s. In addition, in terms of the optimization of the CFD model, it is not limited to the use of the SST k-ω model, but other fluid dynamics models such as RANS can also be considered, in order to more comprehensively understand and predict the flow characteristics of the wind farm through a variety of simulation methods.

[0033] Based on the first wind turbine that meets the preset correlation threshold, a CFD simulation of the wind farm in all wind directions is performed. During the simulation, the interaction between wind turbines and the impact of wake on the performance of downstream wind turbines are fully considered, and the classic Jensen wake model can be used for evaluation. CFD simulation is performed on each group of working conditions (a total of 304 groups, consisting of 16 wind direction intervals and 19 wind speed groups) to obtain flow field data, including wind speed, wind direction, flow field distribution and other information.

[0034] Further, performing correlation analysis on the first actual wind condition data and the second actual wind condition data under different wind direction intervals includes: Calculate the Pearson correlation coefficient of the first actual wind condition data and the second actual wind condition data under different wind direction intervals; Determining a first wind turbine generator set that meets a preset correlation threshold includes: A first wind turbine group that meets a preset correlation threshold is determined according to the Pearson correlation coefficient.

[0035] Specifically, after calculating the Pearson correlation coefficients of the first actual wind condition data and the second actual wind condition data (from different wind turbine locations or different time points), these correlation coefficients need to be compared and analyzed. By constructing a list or matrix containing the correlation strength between each wind turbine location and the virtual lidar location, the correlation strength can be intuitively identified. Subsequently, based on the preset correlation threshold, the wind turbine locations with the strongest correlation with the virtual lidar location are screened out. The data of these units are more suitable as inputs to the deep learning model because of their high similarity with the virtual lidar data.

[0036] Further, performing correlation analysis on the first actual wind condition data and the second actual wind condition data under different wind direction intervals includes: Segmenting the first actual wind condition data and the second actual wind condition data based on a preset wind direction interval; According to different wind direction intervals, correlation analysis is performed on the first actual wind condition data and the second actual wind condition data.

[0037] Specifically, according to the preset wind direction interval (for example, every 22.5° is divided into an interval), the original records of the first actual wind condition data and the second actual wind condition data are respectively assigned to the corresponding wind direction intervals to achieve data segmentation. Then, within each wind direction interval, the Pearson correlation coefficient or the Spearman correlation coefficient is selected to calculate the correlation strength between the two sets of data according to the distribution of the data, so as to obtain the correlation coefficient value in each interval. After this step is completed, the correlation coefficient values ​​of all intervals are summarized and compared to reveal the correlation change trend between the two sets of data in different wind direction intervals, and to identify which intervals have more significant or weaker correlation. This information is of great reference value for guiding subsequent deep learning model training, optimizing model performance, and performing performance evaluation.

[0038] Furthermore, the virtual radar prediction data of the second unit includes the virtual radar data of the second wind turbine to be tested; correction is performed based on the CFD simulation results and the first actual wind condition data and the second actual wind condition data to obtain the first revised wind condition data and the second revised wind condition data, including: determining the second wind turbine that meets the preset distance and the preset correlation threshold as the second wind turbine to be tested, and correcting it according to the CFD simulation results and the first actual wind condition data and the second actual wind condition data to obtain the first revised wind condition data and the second revised wind condition data.

[0039] Specifically, according to the preset distance and the preset correlation threshold, the second wind turbine to be tested is determined. The data of these units will be used for comparison and correction with the CFD simulation results. Correction processing is performed based on the CFD simulation results, the first actual wind condition data and the second actual wind condition data. The correction process may include steps such as data interpolation, data fusion and error correction to ensure the accuracy and consistency of the data. Through the correction, the first corrected wind condition data and the second corrected wind condition data are obtained. Among them, the first corrected wind condition data includes the corrected data of the first wind speed and wind direction instrument and the corrected data of the first laser radar. The second corrected wind condition data includes the corrected data of the second wind speed and wind direction instrument, and the corrected wind condition data is used to perform performance evaluation, optimization design and operation and maintenance management of the wind farm. For example, the power output and wind speed attenuation of the unit under different wind directions and wind speeds can be analyzed to optimize the layout and parameter settings of the unit. The corrected data can also be used for training and verification of deep learning models to improve the accuracy and reliability of wind speed prediction.

[0040] Further, the virtual radar prediction data of the second unit also includes virtual radar data of other second units; Constructing a prediction model based on the first actual wind condition data, the first revised wind condition data, the second actual wind condition data, and the second revised wind condition data includes: establishing a first model for predicting virtual radar data of a second wind turbine to be tested by learning a mapping relationship between the first actual wind condition data and the first revised wind condition data; establishing a second model for predicting virtual radar data of other second wind turbines by learning the virtual radar data of the second wind turbine to be tested, the second actual wind condition data, and the second revised wind condition data; Taking the second actual wind condition data and the second revised wind condition data as input, based on the prediction model, obtaining the virtual radar prediction data of the second unit includes: taking the actual wind condition data of the second unit as the input of the first model to obtain the virtual radar data of the second wind turbine to be tested; taking the virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second revised wind condition data as the input of the second model to obtain the virtual radar data of other second units.

[0041] like Figure 5 As shown, specifically, the LSTM model is used to implement the specific functions of the first model and the second model. The LSTM model is a long short-term memory network model: The LSTM model introduces memory blocks in the hidden layer, including input gate, output gate and forget gate, which not only solves the inherent defects of recursive neural networks, but also can process longer time series. Figure 2The basic structure of the LSTM model is shown, including the input layer, output layer and recursive hidden layer. Each recursive hidden layer consists of multiple storage modules, which contain multiple self-connected storage units, and each storage module has three gates. The calculation of each neuron in the LSTM model is shown in formulas (7)-(15).

[0042] (7) (8) (9) (10) (11) (12) (13) (14) (15) In the formula, i t ,、f t and t Respectively represent the input gate, forget gate and output gate; W i , W f , W o , W c and W hy represents the weight matrix of each layer; b i 、b f 、b o 、b c and b y represents the deviation vector; c t Represents the new state of the cell unit; represents the gate excitation function; tanh is the hyperbolic tangent function; y t Indicates the final prediction result.

[0043] The Bayesian algorithm is used to optimize the model hyperparameters. This algorithm is more robust when searching for the global optimal value of non-convex problems. The Bayesian optimization algorithm is shown in formula (16): (16) In the formula, Indicates candidate set; X * represents the optimal parameter set; f (X) represents the recognition function.

[0044] Specifically, first collect and preprocess the actual wind condition data of the second unit to ensure the accuracy and completeness of the data, and use the CFD simulation results to compare and correct the first and second actual wind condition data. Then, build the first model and the second model based on LSTM, where the first model uses the LSTM model's ability to capture the long-term dependencies of time series data, combined with the learned mapping relationship between the first actual wind condition data and the first corrected wind condition data, to predict the virtual radar data of the second wind turbine to be tested. The second model further integrates the virtual radar data predicted by the first model, the actual wind condition data of the second unit, and the corrected data. In the data analysis and prediction of wind farms, detailed data preprocessing was first performed. This step includes collecting wind speed and wind direction data and lidar data from each machine position, interpolating missing data to ensure data integrity and continuity, and standardizing all features to the range of 0 to 1 to improve the training efficiency of subsequent models. After preprocessing, the data set is divided into a training set (70%), a validation set (20%), and a test set (10%). Subsequently, the architecture of the LSTM model was defined, in which the input layer is responsible for receiving the preprocessed wind speed and wind direction data and the lidar data, the hidden layer is composed of multiple LSTM units to capture long-term dependencies in time series data, and the output layer is responsible for outputting the predicted wind speed and wind direction data or lidar data.

[0045] After the model architecture was determined, a series of hyperparameters were set for the LSTM model, including the number of delays, the number of hidden layers, the number of hidden units, the dropout rate, the learning rate, the maximum number of iterations, the correction frequency, the number of corrections allowed, and the minimum batch training size. These parameters jointly control the training process and performance of the model. Next, the LSTM model was compiled, the mean square error was selected as the loss function, and the Adam optimizer was used. In order to further optimize the performance of the model, the Bayesian optimization method was used to fine-tune the hyperparameters. The model was trained and evaluated through multiple iterations until the predetermined number of iterations was reached or the performance indicators were no longer significantly improved. During the model training phase, the LSTM model was trained using the training set data and regularly verified using the validation set data to adjust the model parameters and improve performance. At the same time, in order to prevent overfitting, the model was corrected every certain number of steps during the training process. After the training was completed, the performance of the model was fully evaluated using the validation set and test set data, and error indicators such as the root mean square error, mean absolute error, and correlation coefficient were calculated to ensure that the model can also show good generalization ability on unseen data. Finally, the trained model was applied to the actual wind farm data to generate virtual lidar data. Specifically, two models were constructed: the first model used the mapping relationship between the corrected wind speed and wind direction instrument data of the installed lidar position and the measured wind speed and wind direction instrument data to generate virtual lidar data for positions without lidar installed; the second model further integrated the prediction results of the first model, the actual wind condition data of the second unit and the corrected data, and used the LSTM model's ability to capture complex relationships to predict the virtual radar data of other positions without lidar installed. In the input and output processing, the input layer, input gate, forget gate, recursive hidden layer, output layer and output gate of the LSTM model were designed to ensure the accuracy and reliability of the prediction results. In addition, the Bayesian algorithm was used to optimize the model hyperparameters, further improving the performance and accuracy of the model. In summary, this application successfully achieved accurate prediction of the virtual radar data of the second unit by constructing a model prediction system based on the long short-term memory network (LSTM). This achievement provides solid data support and decision-making basis for performance evaluation, optimization design and operation and maintenance management of wind farms. It is worth noting that the choice of deep learning models is not limited to the LSTM model. Instead, other types of models, such as convolutional neural networks (CNN), can be explored as alternatives or supplements based on the specific needs and characteristics of different wind farms in order to achieve better prediction results.

[0046] Furthermore, the virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second revised wind condition data are used as inputs of the second model to obtain virtual radar data of other second turbines, including: The virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second corrected wind condition data are used as inputs of the second model and the virtual radar data of other second turbines are obtained by using an inverse proportional distance weighting method.

[0047] Specifically, the core idea of ​​the first model is to build a deep learning relationship between the corrected data and the measured data, and extend this relationship to other related data sets. Similarly, the second model is also built based on the deep learning relationship between the corrected data and the measured data, and further extends this relationship to a wider range of application scenarios.

[0048] In the specific implementation, the inverse proportional distance weighting method is used to optimize the intelligent wind measurement technology of virtual position lidar data according to the distance between the two model input data and the location to be predicted. This technology can realize the real-time wind measurement function of virtual lidar.

[0049] When it is necessary to predict the virtual radar data of other second wind turbines, the virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second corrected wind condition data are used as the input of the second model. On this basis, the inverse proportional distance weighting method is used to comprehensively consider the distance relationship between different data points and the position to be predicted, so as to more accurately predict the virtual radar data of other second wind turbines.

[0050] Such an approach not only improves the prediction accuracy, but also enhances the generalization ability of the model, making it possible to more effectively utilize existing data resources and provide more reliable support for the operation and management of wind farms.

[0051] Further, constructing a prediction model based on the first actual wind condition data, the first revised wind condition data, the second actual wind condition data, and the second revised wind condition data also includes: Calculate the error evaluation index value of the first model and the error evaluation index value of the second model; Evaluating the prediction accuracy of the first model according to the error evaluation index value of the first model; The prediction accuracy of the second model is evaluated according to the error evaluation index value of the second model.

[0052] Specifically, in order to accurately measure the performance of the short-term wind speed prediction model, a variety of error evaluation indicators are used, including the correlation coefficient R, the root mean square error RMSE, the mean absolute percentage error MAPE, and the determination coefficient R². The calculation formulas of these indicators are given in formula (17), formula (18), and formula (19). Among these indicators, the smaller the values ​​of RMSE and MAPE, the smaller the difference between the prediction results and the actual observation values; and the closer the values ​​of R and R² are to 1, the higher the prediction accuracy of the model. It is worth noting that the MAPE indicator is not applicable when the denominator may be zero, because this will lead to calculation errors. Among them.

[0053] The correlation coefficient R, the root mean square error RMSE are as shown in formula (17), the mean absolute percentage error MAPE is as shown in formula (18), and the determination coefficient R2 is as shown in formula (19). The calculation formulas are as follows.

[0054] (17) (18) (19) In the formula, Represents Y i The predicted value of Represents Y i The average value of .

[0055] The smaller the RMSE and MAPE values ​​are, the higher the R and R 2 The closer it is to 1, the higher the prediction accuracy of the model. It should be noted that the MAPE error indicator is not applicable when the denominator may be zero.

[0056] In the process of building the prediction model, based on the first actual wind data, the first corrected wind data, the second actual wind data and the second corrected wind data, not only the model training and prediction are carried out, but also the error evaluation index values ​​of the first model and the second model are calculated. By comparing these index values, the prediction accuracy of the two models can be objectively evaluated. Specifically, the error evaluation index value of the first model reflects its accuracy in predicting the first corrected wind data using the first actual wind data; while the error evaluation index value of the second model reveals its accuracy in predicting the virtual radar data of other second units after integrating the virtual radar prediction data of the second wind turbine to be tested, the second actual wind data and the second corrected wind data. Such an evaluation process provides a comprehensive and accurate model performance measurement standard, which helps to continuously optimize and improve the prediction model.

[0057] In the hypothetical case of a wind farm, in view of the fact that laser radar has been installed at position 18 and laser radar has not been installed at position 14, the wind speed and direction instrument data (T_18) and laser radar data (L_18) of position 18, as well as the wind speed and direction instrument data (T_14) of position 14 were first collected. Subsequently, the SST k-ω model was used to perform CFD numerical simulations of all wind directions and wind speeds of the wind farm, and a wind farm full wind direction CFD database containing the flow field distribution under different wind speeds and wind directions was generated. The classic Jensen wake model was used to evaluate the wake effect of the fleet. On this basis, combined with the measured wind data (T_18 and L_18) of position 18 and the flow field simulation results in the CFD database, the corrected full-field wind speed time history information was generated, including the corrected wind speed and direction instrument data (T_18 correction) and corrected laser radar data (L_18 correction) of position 18. Next, a deep learning relationship between the corrected wind speed and direction instrument data (T_18 corrected) of the No. 18 position and the measured wind speed and direction instrument data (T_18) was established. By training the LSTM model (i.e., the second model), the function of using the wind speed and direction instrument data (T_14) of the No. 14 position to predict its virtual lidar data (L_14) was realized. In order to further improve the prediction accuracy, the generated L_14 data was corrected in combination with the flow field simulation results in the CFD database to obtain the L_14 corrected data. This series of steps not only demonstrates the implementation process of the virtual lidar intelligent wind measurement technology, but also reflects the effective method of using the CFD database and deep learning model to improve the wind speed prediction accuracy of wind farms.

[0058] The present invention also provides a wind farm virtual laser radar wind measurement system, comprising a processor and a memory storing a computer program, wherein the processor implements any of the steps of the wind farm virtual laser radar wind measurement method described above when executing the computer program.

[0059] The present invention combines computational fluid dynamics (CFD) simulation with deep learning technology to build a virtual LiDAR system, thus achieving high efficiency and low cost in wind speed prediction. The present invention covers multiple aspects such as the technical architecture design of the virtual LiDAR system, the optimization of the data processing flow, and the model optimization algorithm. The dependence on physical LiDAR equipment is reduced, and only a small amount of LiDAR equipment combined with innovative virtual wind measurement technology is needed to achieve effective coverage of the entire wind farm. This transformation not only significantly reduces equipment costs, but also reduces subsequent maintenance costs.

[0060] It is understood that the above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the patent scope of the present invention. It should be pointed out that, for ordinary technicians in this field, without departing from the concept of the present invention, the above technical features can be freely combined, and several deformations and improvements can be made, which all belong to the protection scope of the present invention. Therefore, all equivalent changes and modifications made to the scope of the claims of the present invention should belong to the coverage of the claims of the present invention.

Claims

1. A wind farm virtual laser radar wind measurement method, characterized in that: The method comprises the following steps: Acquire first actual wind condition data of a first unit and second actual wind condition data of a second unit; wherein the first unit is a unit equipped with a laser radar; and the second unit is a unit not equipped with a laser radar; Carry out CFD simulation of the wind farm in all wind directions for the wind farm to obtain CFD simulation results; Based on the CFD simulation result, respectively correcting the first actual wind condition data and the second actual wind condition data to obtain first corrected wind condition data and second corrected wind condition data; constructing a prediction model according to the first actual wind condition data, the first revised wind condition data, the second actual wind condition data and the second revised wind condition data; The second actual wind condition data and the second revised wind condition data are used as input, and based on the prediction model, virtual radar prediction data of the second unit is obtained.

2. The wind farm virtual laser radar wind measurement method according to claim 1, characterized in that: The wind farm CFD simulation is carried out for all wind directions, and the CFD simulation results obtained include: Performing correlation analysis on the first actual wind condition data and the second actual wind condition data according to different wind direction intervals; The first wind turbine that meets the preset correlation threshold is determined, and a CFD simulation of the wind farm in all wind directions is performed on the wind farm to obtain a CFD simulation result.

3. The wind farm virtual laser radar wind measurement method according to claim 2, characterized in that: Determining the first wind turbine generator set that meets the preset correlation threshold and carrying out CFD simulation of the wind farm in all wind directions for the wind farm includes: The SST k-ω model is adopted to perform CFD simulation of all wind directions of the wind farm based on the first wind turbine group that meets the preset correlation threshold.

4. The wind farm virtual laser radar wind measurement method according to claim 3 is characterized in that: The virtual radar prediction data of the second wind turbine generator set includes virtual radar data of the second wind turbine generator set to be tested; Based on the CFD simulation result, the first actual wind condition data and the second actual wind condition data are corrected respectively to obtain the first corrected wind condition data and the second corrected wind condition data, including: A second wind turbine set that meets a preset distance and a preset correlation threshold is determined as the second wind turbine set to be tested, and is corrected according to the CFD simulation result and the first actual wind condition data and the second actual wind condition data to obtain first corrected wind condition data and second corrected wind condition data.

5. The wind farm virtual laser radar wind measurement method according to claim 4, characterized in that: The virtual radar prediction data of the second unit also includes virtual radar data of other second units; The constructing of a prediction model according to the first actual wind condition data, the first revised wind condition data, the second actual wind condition data and the second revised wind condition data comprises: Establishing a first model for predicting virtual radar data of the second wind turbine to be tested by learning a mapping relationship between the first actual wind condition data and the first revised wind condition data; Establishing a second model for predicting virtual radar data of other second wind turbines by learning the virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second revised wind condition data; The step of taking the second actual wind condition data and the second revised wind condition data as input and obtaining the virtual radar prediction data of the second unit based on the prediction model comprises: Using the actual wind condition data of the second wind turbine as input of the first model to obtain virtual radar data of the second wind turbine to be tested; The virtual radar data of the second wind turbine group to be tested, the second actual wind condition data and the second corrected wind condition data are used as inputs of the second model to obtain the virtual radar data of the other second wind turbine group.

6. The wind farm virtual laser radar wind measurement method according to claim 5, characterized in that: The step of using the virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second corrected wind condition data as inputs of the second model to obtain the virtual radar data of the other second turbines comprises: The virtual radar data of the second wind turbine to be tested, the second actual wind condition data and the second corrected wind condition data are used as inputs of the second model and the virtual radar data of the other second turbines are obtained using an inverse proportional distance weighting method.

7. The wind farm virtual laser radar wind measurement method according to claim 6, characterized in that: The constructing of a prediction model according to the first actual wind condition data, the first revised wind condition data, the second actual wind condition data and the second revised wind condition data also includes: Calculating an error evaluation index value of the first model and an error evaluation index value of the second model; Evaluating the prediction accuracy of the first model according to the error evaluation index value of the first model; The prediction accuracy of the second model is evaluated according to the error evaluation index value of the second model.

8. The wind farm virtual laser radar wind measurement method according to claim 7, characterized in that: The performing correlation analysis on the first actual wind condition data and the first actual wind condition data according to different wind direction intervals includes: Calculating the Pearson correlation coefficient between the first actual wind condition data and the second actual wind condition data in different wind direction intervals; The step of determining the first wind turbine generator set that meets the preset correlation threshold comprises: A first wind turbine group that meets a preset correlation threshold is determined according to the Pearson correlation coefficient.

9. The wind farm virtual laser radar wind measurement method according to claim 8, characterized in that: The performing correlation analysis on the first actual wind condition data and the second actual wind condition data according to different wind direction intervals includes: Segmenting the first actual wind condition data and the second actual wind condition data based on a preset wind direction interval; According to different wind direction intervals, correlation analysis is performed on the first actual wind condition data and the second actual wind condition data.

10. A wind farm virtual laser radar wind measurement system, characterized in that: It comprises a processor and a memory storing a computer program, wherein the processor implements the steps of the wind farm virtual lidar wind measurement method according to any one of claims 1 to 9 when executing the computer program.

Images