An intelligent correction method for numerical wind field simulation

Abstract

The invention discloses an intelligent correction method for a numerically simulated wind field, which belongs to the field of numerical simulation. The method comprises collecting and processing site observed wind field data and multiple characteristic variable data of numerical model simulation, using a random forest model to screen important characteristic variables that have a significant impact on the wind field, dividing the data sets of the observed wind field and the important characteristic variables into a training set, a validation set and a test set, constructing and training a BP (Back Propagation) neural network model, and finding an optimal model by adjusting parameters, evaluating the performance of the optimal BP neural network model, and performing visualization processing. The invention is based on the observed wind field data and the numerical model data, utilizes the random forest model, and combines the BP neural network to construct an intelligent correction method for a numerically simulated wind field. The method can further improve the accuracy of wind field simulation on the basis of a dynamic numerical model simulating a wind field, and provides a new method for correcting a numerically simulated wind field.

Description

Technical Field

[0001] The invention belongs to the field of numerical simulation, and more specifically relates to an intelligent correction method for a numerically simulated wind field. Background Art

[0002] Numerical weather forecast products have become important reference materials in modern weather forecasting business. Numerical simulation wind field data are widely used in weather analysis and forecasting, economic and military activities. Its accuracy has an important impact on weather system forecasting, marine wind energy resource development, navigation safety, marine disaster prevention and mitigation, and maritime military activities. However, the current numerical simulation wind field is limited by many factors and has errors relative to the observed wind field. Establishing a numerical simulation wind field correction model through artificial intelligence methods is an effective means to effectively improve the accuracy of numerical simulation wind fields.

[0003] Traditional wind forecasting methods mainly rely on numerical weather forecasting models. The Weather Research and Forecasting Model (WRF) is a high-resolution atmospheric model developed by the National Center for Atmospheric Research in the United States. It is currently the most widely used numerical wind forecasting model. However, due to factors such as the approximate treatment of the atmospheric motion equations, the cumulative effect of the model's initial field errors, the incompleteness of the physical parameterization scheme, and the uncertainty of the initial field and boundary conditions, there are certain errors between the numerically simulated wind field and the observed wind field. There is currently no mature solution to the above factors that lead to errors in numerically simulated wind fields in the WRF model. Therefore, it is necessary to develop an effective method for correcting errors in numerically simulated wind fields in order to improve the accuracy of numerically simulated wind fields.

[0004] With the development of artificial intelligence and machine learning technology, especially the advantages of random forest algorithm in feature selection and the powerful ability of BP neural network in nonlinear relationship modeling, new possibilities are provided for intelligent correction of wind field data. Random forest can effectively identify important characteristic variables that have a great impact on wind field changes, and BP neural network can learn the complex nonlinear mapping relationship between important characteristic variables and observed wind fields through training, and then optimize and correct the wind field simulated by numerical model. Summary of the invention

[0005] The present invention provides a numerical simulation wind field correction method based on random forest algorithm and BP neural network. Based on observation data and numerical model simulation data, the method uses random forest algorithm to screen out important characteristic variables that affect the accuracy of wind field simulation, and constructs and optimizes BP neural network to construct a correction model for numerical simulation wind field.

[0006] In order to achieve the above object, the present invention is implemented by adopting the following technical scheme: the method comprises:

[0007] Collect site observation wind field data and numerical model simulation data, and process the data;

[0008] The random forest model was used to screen important characteristic variables, and the relative weight of each characteristic variable's influence on the observed wind field was quantified through the random forest model.

[0009] The data sets of important characteristic variables of the observed wind field and numerical simulation are divided into training set, validation set and test set;

[0010] Construct a BP neural network model, use the selected important characteristic variables as the input of the BP neural network model, and use the observed wind field data as the target output of the BP neural network;

[0011] Evaluate the performance of the optimal BP neural network model, input the test set data into the optimal BP neural network model, and evaluate the correction performance of the optimal BP neural network model for the numerical simulation of the wind field.

[0012] In one embodiment, the wind field data observed at the collection site includes: 10m wind speed and wind direction;

[0013] The numerical model simulation variables include 10m latitudinal wind speed, 10m longitudinal wind speed, 2m relative humidity, 2m air temperature and sea level pressure, which are characteristic variables that affect the wind field.

[0014] In one embodiment, the random forest model is used to screen important feature variables, and after a random forest is formed by constructing multiple decision trees, out-of-bag samples are used to evaluate the importance of the feature variables;

[0015] Based on the prediction error of out-of-bag samples, the importance score of each feature variable is calculated to reflect its contribution in the random forest model;

[0016] According to the importance score, the most representative and simulation-capable important feature variables are screened to achieve the extraction of important feature variables.

[0017] In one scheme, the training set, validation set and test set are divided into the data sets of the screened important characteristic variables and the observed wind field data according to the time series; the first 70% of the data is divided into the training set, the middle 10% of the data is divided into the validation set, and the last 20% of the data is divided into the test set.

[0018] In one embodiment, the construction of the BP neural network model includes:

[0019] Set key model parameters, including the number of network layers, the number of neurons in each layer, the applicable activation function, learning rate, loss function, and optimizer type;

[0020] The model is repeatedly optimized to obtain the optimal model.

[0021] In one scheme, the performance of the optimal BP neural network model is evaluated by first calculating the mean square error between the numerically simulated wind field and the observed wind field, then calculating the mean square error between the numerically simulated wind field and the observed data after correction using the optimal BP neural network model, comparing the reduction in the mean square error before and after correction, and evaluating the correction performance of the BP neural network model on the simulated wind field.

[0022] In one embodiment, the data processing uses inverse distance weighted interpolation to spatially interpolate the model data to the observation data sites because the observation data sites do not coincide with the grid points of the model data.

[0023] Since the time resolution of the observation data and the model data are not uniform, the linear interpolation method is used to unify the time resolution of the model data and the observation data, that is, to align the time and space dimension information of the dynamic numerical model simulation data and the observation data;

[0024] The inversion algorithm formula of the offshore wind speed profile is:

[0025]

[0026] Among them, v h is the wind speed at a height h above the ground; v 0 is the height from the ground h 0 Wind speed at z 0 is the ground condition coefficient, which is generally taken as 0.003 for a flat underlying surface; k 0 is the wind speed height coefficient;

[0027] The inverse distance weighted interpolation formula is:

[0028]

[0029] Among them, d(P,S i ) represents the point P and the known point S i The distance between them; p is the weight exponent (power parameter), which controls the influence of distance on the weight; n is the number of known points;

[0030] The linear interpolation algorithm formula is:

[0031]

[0032] Among them, (x 0 ,y 0 ) and (x 1 ,y 1 ) are two known data points, and x 0 <x 1 , y is the corresponding interpolation value.

[0033] In one solution, the important feature variables are screened as follows: For a feature f, the feature importance score under the out-of-bag sample prediction error in the random forest is approximately expressed as:

[0034]

[0035] Where T is the number of decision trees in the forest; D t Represents the training data set of the t-th decision tree; Impurity t (D t ) is the impurity of the data set before the t-th decision tree is divided; Impurity t (D t |f) is the average impurity of the data subset after being divided at the optimal split point of feature f.

[0036] Beneficial effects of the present invention:

[0037] The present invention combines the random forest algorithm and the BP neural network, aiming to make full use of the information of the numerical model simulation data and the observation data, screen out the important characteristic variables through the random forest algorithm, reduce the characteristic dimension, and improve the efficiency and training speed of the model. On this basis, the BP neural network learns the complex nonlinear relationship, and screens out the numerical simulation wind field correction model with the best performance, further improving the accuracy of the numerical simulation wind field. The present invention has strong universality and can flexibly correct the wind field of any height layer according to the wind field height of the observation data. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention.

[0039] Figure 2 This is a flow chart of the BP neural network model construction, optimization and performance evaluation in the present invention.

[0040] Figure 3 This is a diagram showing the importance of feature variables calculated by random forest in an embodiment of the present invention, where the horizontal axis is the feature importance score and the vertical axis is the feature variable.

[0041] Figure 4 This is a comparison of wind speed in the test set in an embodiment of the present invention: observation data, WRF model data and correction result diagram. DETAILED DESCRIPTION

[0042] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0043] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0044] Embodiment 1:

[0045] like Figure 1 As shown, an intelligent correction method for numerically simulated wind field includes the following steps:

[0046] S1. Collect and process the observed wind field data and numerical model simulation data at the site. Data preprocessing includes quality control of the observed data, eliminating missing values ​​and outliers to ensure the reliability of the data; using the near-sea surface wind speed profile inversion algorithm to convert the height of the numerical model simulated wind field to a height consistent with the observed wind field; using the inverse distance weighted interpolation algorithm to interpolate the numerical model simulation data to the observed wind field site, and using linear interpolation to unify the time resolution of the two sets of data, so that the spatial and temporal dimensions of the two sets of data are consistent.

[0047] The observed wind field includes 10m wind speed and wind direction, and the numerical model simulation variables include but are not limited to 10m zonal wind speed, 10m meridional wind speed, 2m relative humidity, 2m air temperature and sea level pressure, which are characteristic variables that affect the wind field. In order to make the data format of the characteristic variables suitable for random forest and BP neural network models, their time and space dimensions are standardized. The selection of multiple characteristic variables can fully consider the impact of various factors on the wind field, which helps to improve the correction effect.

[0048] In the data processing method of S1, since the sites of the observation data do not coincide with the grid points of the model data, the inverse distance weighted interpolation is used in space to interpolate the model data to the sites of the observation data; since the time resolutions of the observation data and the model data are not uniform, the linear interpolation method is used to unify the time resolutions of the model data and the observation data, that is, to align the spatiotemporal dimension information of the dynamic numerical model simulation data and the observation data.

[0049] The inversion algorithm formula of the offshore wind speed profile is:

[0050]

[0051] Among them, v h is the wind speed at a height h above the ground; v 0 is the height from the ground h 0 Wind speed at z 0 is the ground condition coefficient, which is generally taken as 0.003 for a flat underlying surface; k 0 is the wind speed height coefficient;

[0052] The inverse distance weighted interpolation formula is:

[0053]

[0054] Among them, d(P,S i ) represents the point P and the known point S i The distance between them; p is the weight exponent (power parameter), which controls the influence of distance on the weight; n is the number of known points;

[0055] The linear interpolation algorithm formula is:

[0056]

[0057] Among them, (x 0 ,y 0 ) and (x 1 ,y 1 ) are two known data points, and x 0 <x 1 , y is the corresponding interpolation value.

[0058] S2. Use the random forest model to screen important characteristic variables. The random forest model is used to quantify the relative weight of each characteristic variable's impact on the observed wind field. The characteristic variables simulated by the numerical model are used as the input of the random forest model, and the observed wind field data are used as the target output for the model training process. Finally, the relative weight of each characteristic variable is evaluated, and the characteristic variables with higher weights are determined as important characteristic variables.

[0059] S2 uses the random forest algorithm to select important features. After constructing multiple decision trees to form a random forest, it uses out-of-bag samples to evaluate the importance of feature variables. Based on the prediction error of out-of-bag samples, the importance score of each feature variable is calculated to reflect its contribution in the random forest model. According to these importance scores, the most representative and simulation-capable important feature variables are screened to achieve the extraction of important feature variables.

[0060] For a feature f, its feature importance score under the out-of-bag sample prediction error in the random forest can be approximately expressed as:

[0061]

[0062] Where T is the number of decision trees in the forest; D t Represents the training data set of the t-th decision tree; Impurity t (D t ) is the impurity of the data set before the t-th decision tree is divided; Impurity t (D t|f) is the average impurity of the data subset after being divided at the optimal split point of feature f.

[0063] S3. Divide the data sets of important characteristic variables of the observed wind field and numerical simulation into training set, validation set and test set. Divide the data sets of multiple important characteristic variables selected in step 2 together with the observed wind field data according to the time series, and divide the first 70% of the data into the training set, the middle 10% of the data into the validation set, and the last 20% of the data into the test set.

[0064] S4, build a BP neural network model, use the training set of S3, take the selected important characteristic variables as the input of the BP neural network model, and use the observed wind field data as the target output of the BP neural network. Through training, learn and capture the complex nonlinear relationship between the observed wind field and important characteristic variables. During the training process, continuously monitor the output performance of the validation set to prevent the model from overfitting. Use the method of cyclic adjustment of parameters to continuously optimize the various parameters of the model to find the optimal BP neural network model.

[0065] The specific steps in S4 are as follows

[0066] S401, construct a BP neural network model and set key parameters of the model, including the number of network layers, the number of neurons in each layer, the applicable activation function, the learning rate, the loss function, and the optimizer type. Through the complex structure and nonlinear activation function inside the neural network, the model can learn the complex nonlinear relationship between the observed wind field and the important characteristic variables, thereby achieving effective mapping from input to target output. The neural network model trained in this process can reflect the intrinsic relationship between the observed wind field and the important characteristic variables.

[0067] The BP neural network consists of an input layer, a hidden layer, and an output layer. The input layer receives important feature variables as input signals. The hidden layer is an intermediate layer used to process the input signal and extract nonlinear features. The output layer gives the final prediction results. The training process of the BP neural network includes two stages: forward propagation and back propagation. During the forward propagation process, the input signal is processed from the input layer through weights and activation functions, and is passed layer by layer to the output layer to obtain the prediction results. During the back propagation process, the error between the prediction results and the observed data is calculated, and the gradient descent algorithm is used to adjust the weights between the layers in the network to minimize the error.

[0068] The forward propagation calculates the prediction result, and then the gradient is calculated through back propagation, and the weights and biases are updated according to a certain learning rate. The formula is as follows:

[0069] Δw=-η×αE / αw

[0070] Among them, Δw represents the weight update amount, η is the learning rate, and αE / αw is the gradient of the loss function with respect to the weight w.

[0071] S402, perform cyclic optimization processing on the model to obtain the optimal model. By adjusting the number of hidden layers and the number of neurons in each layer, repeated training is performed multiple times, and the mean square error of the model with different parameter settings is calculated using the validation set. Based on the principle of minimum mean square error, the numerical simulation wind field correction model with the best performance is selected to improve the accuracy and generalization ability of the model.

[0072] In the above S2, if the performance of the current model fails to meet the expected target, the BP neural network structure parameters will be screened and adjusted repeatedly until the performance of the model on the validation set meets the predetermined accuracy standard. The model obtained through iterative optimization is not only highly accurate, but also has good generalization ability, which can effectively support the intelligent correction of numerical simulation wind fields.

[0073] S5. Evaluate the performance of the optimal BP neural network model. Input the test set data into the optimal BP neural network model, and use the mean square error to evaluate the correction performance of the optimal BP neural network model for the numerical simulation wind field. First, calculate the mean square error between the numerical simulation wind field and the observed wind field, and then calculate the mean square error between the numerical simulation wind field and the observed data after correction using the optimal BP neural network model. Compare the reduction of the mean square error before and after correction, and evaluate the correction performance of the BP neural network model for the simulated wind field.

[0074] Embodiment 2

[0075] like Figure 2 As shown, this embodiment provides an intelligent correction method for numerically simulated wind fields, comprising the following steps:

[0076] S1. Data collection and preprocessing. The data collected in this embodiment mainly include site observation data and model data. The site of the observation data is located at 113.73°E, 21.21°N, the variables include 2m wind speed and wind direction, the time resolution is 10min, and the time range is from September 2021 to March 2022. The model data is WRF model data, the variables include 10m zonal wind speed, 10m meridional wind speed, 2m relative humidity, 2m air temperature and sea level pressure, the time resolution is 1h, the time range is the same as the observation data, the spatial range is 108.85°-119.15°E, 18.40°-25.52°N, and the spatial resolution is 0.1°×0.1°.

[0077] The preprocessing of the observation data mainly eliminates its missing values ​​and eliminates outliers based on the climatological empirical range parameter table of sea surface meteorological elements. The 2m wind speed of the observation data is converted to 10m wind speed using the near-sea surface wind speed profile inversion algorithm to keep consistent with the height of the model simulated wind speed. For the WRF model data, the 10m zonal wind speed and meridional wind speed are converted to 10m wind speed and 10m wind direction. Since the station of the observation data does not coincide with the grid points of the WRF model data, the inverse distance weighted interpolation method is used to interpolate the WRF model data to the observation data station. Since the time resolution of the observation data and the WRF model data is not uniform, the linear interpolation method is used to convert the time resolution of the WRF model data to 10min to be consistent with the time resolution of the observation data;

[0078] S2. Use the random forest model to screen important feature variables. First, determine the input features and target output of the random forest. The input is the 10m zonal wind speed, 10m meridional wind speed, 2m relative humidity, 2m air temperature, sea level pressure, and the calculated 10m wind speed and 10m wind direction simulated by the WRF numerical model, a total of 7 feature variables, and the target output is the observed 10m wind speed. Secondly, set the parameters of the random forest. In this example, we set the number of decision trees of the random forest model to 100 and fixed the random seed to 42. Finally, the importance of the feature variables is evaluated through out-of-bag samples, and visualization is performed accordingly (such as Figure 3 ), select the top three characteristic variables with the largest contribution rate as important characteristic variables. The important characteristic variables selected in this example are 10m wind speed, sea level pressure and 2m temperature;

[0079] S3. Divide the data set. Normalize the three important feature variables selected in step 2, and construct a data set for training, validating and evaluating model performance according to the time sequence of the processed important feature variables and observation data. In this example, the first 70% of the data is divided into the training set, the middle 10% of the data is divided into the validation set, and the last 20% of the data is divided into the test set. The training set is used to train the model, the validation set is used to monitor whether the training process is overfitting, and the test set is used to evaluate the model performance;

[0080] S4. Construct a BP neural network. First, the important characteristic variables processed in step 3 are used as the input of the neural network, and the 10m wind speed of the observed data is used as the target output for training to learn the complex nonlinear relationship between the observed wind speed and the important characteristic variables. The specific steps for establishing the BP neural network model in this example are as follows:

[0081] S401. We define a neural network structure based on PyTorch's nn.Module class and set the size of the input layer, hidden layer, and output layer. In this example, the input layer contains 3 nodes and the output layer is 1 node. This example uses the Adam optimizer for training, the learning rate is set to 1e-2, and the loss function uses the mean square error. The training process defines the epoch as 3000 times;

[0082] S402. In order to find the optimal model, we performed a cyclic optimization. This example tried a hidden layer structure of 1 to 3 layers. The number of neurons in each layer ranged from 3 to 60, with a step size of 3, and each model was trained 3 times. By comparing the performance of each model on the test set, the optimal model structure was determined based on the principles of maximum correlation coefficient and minimum mean square error. In this example, the optimal model has 3 hidden layers, each layer contains 36 neurons, and the best number of repetitions is the third time;

[0083] S5, put the test set data into the optimal model found in S2, and evaluate the prediction ability of the optimal model by calculating the correlation coefficient and mean square error. In this example, the mean square error in the optimal model dropped significantly from the initial 9.14 to 2.36 after training, and the correlation coefficient also increased slightly, indicating that the intelligent correction method can effectively improve the accuracy of numerical simulation of wind fields. Finally, the training results of the test set are visualized (such as Figure 4 ) to display data information more intuitively. Figure 4 The horizontal axis is time, and the vertical axis is wind speed in m / s; blue scatter points represent observation data, green scatter points represent WRF model data, and red scatter points represent model correction results.

[0084] Through the above steps, intelligent correction of numerical simulation wind field is realized, and the accuracy of numerical simulation wind field is improved. This method can provide reliable data support for fields such as wind energy resource assessment, wind farm optimization design, offshore collection and wind field forecasting.

[0085] In summary, the present invention constructs an intelligent correction method for numerically simulated wind fields based on observed wind field data and numerical model simulation data, using a random forest model and combining it with a BP neural network. This method covers the complete process from data acquisition and processing, feature selection and optimization, model construction and model performance evaluation, which effectively improves the accuracy of wind field simulation.

[0086] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0087] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent correction method for numerically simulated wind fields, characterized in that: The method includes: Collect site observation wind field data and numerical model simulation data, and process the data; The random forest model was used to screen important characteristic variables, and the relative weight of each characteristic variable's influence on the observed wind field was quantified through the random forest model. The data sets of important characteristic variables of the observed wind field and numerical simulation are divided into training set, validation set and test set; Construct a BP neural network model, use the selected important characteristic variables as the input of the BP neural network model, and use the observed wind field data as the target output of the BP neural network; The specific steps of constructing the BP neural network model are as follows S401, constructing a BP neural network model, setting key parameters of the model, including the number of network layers, the number of neurons in each layer, the applicable activation function, the learning rate, the loss function and the optimizer type; reflecting the intrinsic correlation between the observed wind field and the important characteristic variables; The BP neural network consists of an input layer, a hidden layer, and an output layer; the input layer receives important feature variables as input signals, the hidden layer is the intermediate layer, and the output layer gives the final prediction results; the training process of the BP neural network includes two stages: forward propagation and back propagation; In the forward propagation process, the input signal is processed by weights and activation functions from the input layer, and then transmitted to the output layer layer by layer to obtain the prediction result; In the back propagation process, the weights between the layers in the network are adjusted using the gradient descent algorithm by calculating the error between the predicted results and the observed data; The forward propagation calculates the prediction result, and then the gradient is calculated through back propagation, and the weights and biases are updated according to a certain learning rate. The formula is as follows: ; in, represents the weight update amount, is the learning rate, It's about weight The gradient of the loss function; S402, performing cyclic optimization processing on the model to obtain the optimal model; By adjusting the number of hidden layers and the number of neurons in each layer, repeated training was performed, and the mean square error of the models with different parameter settings was calculated using the validation set. Based on the principle of minimum mean square error, the numerical simulation wind field correction model with the best performance was selected. Evaluate the performance of the optimal BP neural network model, input the test set data into the optimal BP neural network model, and evaluate the correction performance of the optimal BP neural network model for the numerical simulation wind field; The method for evaluating the performance of the optimal BP neural network model comprises inputting the test set data into the optimal BP neural network model, and using the mean square error to evaluate the correction performance of the optimal BP neural network model on the numerical simulation wind field; firstly, the mean square error between the numerical simulation wind field and the observed wind field is calculated, and then the mean square error between the numerical simulation wind field and the observed data after correction using the optimal BP neural network model is calculated, and the reduction degree of the mean square error before and after correction is compared to evaluate the correction performance of the BP neural network model on the simulated wind field.

2. The intelligent correction method for numerically simulated wind field according to claim 1, characterized in that: The wind field data observed at the collection site include: 10 m wind speed and wind direction; The numerical model simulation variables include 10 m zonal wind speed, 10 m meridional wind speed, 2 m relative humidity, 2 m air temperature and sea level pressure, which are characteristic variables that affect the wind field.

3. The intelligent correction method for numerically simulated wind field according to claim 1, characterized in that: The random forest model is used to screen important feature variables. After forming a random forest by constructing multiple decision trees, the importance of feature variables is evaluated using out-of-bag samples. Based on the prediction error of out-of-bag samples, the importance score of each feature variable is calculated to reflect its contribution in the random forest model; According to the importance score, the most representative and simulation-capable important feature variables are screened to achieve the extraction of important feature variables.

4. The intelligent correction method for numerically simulated wind field according to claim 1, characterized in that: The training set, validation set and test set divide the data sets of the screened important characteristic variables together with the observed wind field data according to the time series; the first 70% of the data is divided into the training set, the middle 10% of the data is divided into the validation set, and the last 20% of the data is divided into the test set.

5. The intelligent correction method for numerically simulated wind field according to claim 1, characterized in that: In the data processing, since the sites of the observation data do not coincide with the grid points of the model data, inverse distance weighted interpolation is used in space to interpolate the model data to the sites of the observation data; Since the time resolution of the observation data and the model data are not uniform, the linear interpolation method is used to unify the time resolution of the model data and the observation data, that is, to align the time and space dimension information of the dynamic numerical model simulation data and the observation data; The inversion algorithm formula of the offshore wind speed profile is: ; in, Height from ground Wind speed; Height from ground Wind speed; is the ground condition coefficient, and for a flat underlying surface, its value is 0.003; is the wind speed height coefficient; The inverse distance weighted interpolation formula is: ; in, Indicate point With known points The distance between is the weight exponent, i.e., the power parameter, which controls the influence of distance on the weight; is the number of known points; The linear interpolation algorithm formula is: ; in, and are two known data points, and , The corresponding interpolation value.

6. The intelligent correction method for numerically simulated wind field according to claim 3 is characterized in that: The important feature variables are screened as follows: f , the feature importance score under the out-of-bag sample prediction error in the random forest is approximately expressed as: ; in, is the number of decision trees in the forest; Indicates A training data set for a decision tree; It is The impurity of the data set before the decision tree is divided; It is in the characteristics The average impurity of the data subsets divided at the optimal split point.

Images