An iTransformer-based intelligent prediction method for surface turbulent flux

Abstract

The present application belongs to the technical field of deep learning and atmospheric science, and specifically relates to an intelligent prediction method for surface turbulent flux based on iTransformer, which comprises the following steps: obtaining original flux meteorological data; processing flux meteorological data; constructing a prediction model; training the model; and evaluating the model. In the method, an iTransformer model is used. Firstly, the embedding coding mode is changed, and each variable sequence at all time points is independently embedded into a token by transposing the time series, so as to more effectively capture the correlation between variables in the flux meteorological data. Secondly, the multivariate attention mechanism is inverted with the responsibility of the feedforward neural network, and the inverted multivariate attention mechanism is responsible for capturing the correlation between variables or time series, and the feedforward neural network is responsible for encoding and decoding each time series to extract its internal features, so as to achieve better prediction performance.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of deep learning and atmospheric science, and particularly relates to an intelligent prediction method for surface turbulent flux based on iTransformer. BACKGROUND

[0002] Friction velocity, sensible heat flux and latent heat flux are key parameters that depict the complex interaction and energy exchange between the atmosphere and the ground. In the atmospheric boundary layer, friction velocity reflects the turbulent friction, affecting the vertical distribution of wind speed and the development of the boundary layer. Sensible heat flux is the form of heat exchange between the atmosphere and the ground, which plays a role in regulating the ground temperature and local climate. Latent heat flux represents the heat flux released or absorbed by water vapor during phase change, which is very important for climate analysis, resource management and environmental planning. The three interact with each other, affecting the atmospheric system and reflecting the complex energy exchange mechanism between the atmosphere and the ground. The traditional Monin-Obukhov Similarity Theory (MOST) is a commonly used method to describe the relationship between turbulence and temperature, humidity, wind speed and other meteorological elements in the near-surface layer. However, due to the limitations of the MOST method, its prediction performance is not good. Therefore, when studying surface turbulent flux, other methods and means need to be combined for comprehensive analysis and prediction.

[0003] In recent years, the rapid progress of deep learning (DL) and its innovative application in meteorology provide a promising method. Compared with the traditional MOST method, the DL model can handle highly complex and nonlinear problems. In addition, through a large amount of training data, the DL model can perform multi-layer nonlinear operations on the data, learn the complex nonlinear relationship of the data, and more accurately capture the internal rules and patterns of the data. Therefore, the DL model can provide a new way for surface turbulent flux prediction. Among them, the Transformer model can effectively process sequence data and capture long-distance dependencies, and is widely used in time series prediction. However, the Transformer model cannot fully reveal the complex relationship between multiple variables, so it cannot efficiently handle the correlation between multiple variables. SUMMARY

[0004] In view of the technical problem that the above-mentioned Transformer model cannot fully reveal the complex relationship between multiple variables, the application provides an intelligent prediction method for surface turbulent flux based on iTransformer.

[0005] In order to solve the above technical problems, the technical scheme adopted by the application is as follows:

[0006] An iTransformer-based surface turbulent flux intelligent prediction method, comprising the following steps:

[0007] S1, obtaining original flux meteorological data;

[0008] S2, flux meteorological data processing: removing the outliers and missing values in the original flux meteorological data obtained in S1, and calculating the input data using physical formulas, the input data including wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio, and surface turbulent flux observation value, and dividing the input data to obtain a training set, a validation set, and a test set;

[0009] S3, constructing a prediction model: establishing an iTransformer surface turbulent flux intelligent prediction model based on a deep learning method; the basic architecture of the model includes the following layers: an input layer, an embedding coding layer, an iTransformer layer, a projection layer, and an output layer, wherein the iTransformer layer includes a multivariate attention module, a feedforward neural network module, and layer normalization;

[0010] S4, model training: using the mean square error (MSE) between the predicted value and the true value of the surface turbulent flux as the loss function of the model, inputting the training set and the validation set obtained in S2 into the prediction model constructed in S3 for training until the model converges;

[0011] S5, model evaluation: using the test set of S2 to verify and evaluate the prediction model after S4 converges, calculating the root mean square error (RMSE), the mean absolute error (MAE), and the Pearson correlation coefficient (R) between the predicted value and the true value of the surface turbulent flux as evaluation indexes, and comparing the results with the traditional MOST method and the existing prediction model artificial neural network (ANN).

[0012] The original flux meteorological data in S1 are air density, wind speed, temperature, pressure, upward longwave radiation, relative humidity, water content, friction velocity, sensible heat flux, and latent heat flux; the original flux meteorological data in S1 are obtained from the Beijing meteorological tower, and the sampling time of the meteorological tower is the whole year of 2023, and the sampling time frequency is 30 min.

[0013] The method for processing the collected original flux meteorological data in S2 is to remove missing values and invalid values, and to obtain wind speed, potential temperature gradient, Richardson number, and water vapor mixing ratio changes as model input variables according to physical formulas in meteorology; to obtain surface turbulent flux model labels: friction velocity, moving sensible heat flux, and moving latent heat flux according to physical formulas for calculating surface turbulent flux in meteorology; the processed input data include wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio, and surface turbulent flux observation value, and the obtained input data are divided into a training set, a validation set, and a test set according to a ratio of 7:2:1.

[0014] The S3 includes constructing a prediction model.

[0015] The input layer transmits data such as wind speed, temperature gradient, Richardson number, water vapor mixing ratio change and surface layer turbulent flux label to the embedding encoding layer.

[0016] The embedding encoding layer transposes and encodes the input data, and embeds the sequence of each variable at all time points into token.

[0017] The iTransformer layer realizes the prediction of the surface layer turbulent flux. First, the token obtained by the embedding encoding layer is transmitted to the multivariate attention module with residual connection. Second, the multivariate attention module can effectively capture the relationship between the input variables wind speed, temperature gradient, Richardson number, water vapor mixing ratio change and the prediction label surface layer turbulent flux, and obtain Q, K, V∈R N×dk by linear projection. The score between wind speed, temperature gradient, Richardson number, water vapor mixing ratio change and surface layer turbulent flux is calculated by using Q and K dot product, and the score matrix A between surface layer turbulent flux and input variables is obtained. Then, the error of the input data is reduced by using the normalization strategy, and then transmitted to the feedforward neural network module for calculation. The calculation result is normalized and projected, and finally the prediction result is transmitted to the output layer to obtain the prediction value of the surface layer turbulent flux.

[0018] The calculation formula of the score matrix A is:

[0019]

[0020] where d k is the projection dimension, q i is the specific representation of the query vector and key of the token corresponding to the i-th input variable, k j is the specific representation of the query vector and key of the token corresponding to the j-th input variable; and the obtained score matrix A∈R N×N represents the correlation between the i-th and j-th input variables.

[0021] The method for training the model in S4 is to calculate the mean square error MSE between the prediction value and the true value of the surface layer turbulent flux as the loss function of the model, and use the training set and the validation set divided by S2 to train and verify the model constructed in S3 until the model converges. The formula of the mean square error MSE is as follows:

[0022]

[0023] Wherein y_pred is the model output friction velocity, moving sensible heat flux, moving latent heat flux prediction value, y_true is the friction velocity, moving sensible heat flux, moving latent heat flux true observation value, n is the sample number.

[0024] The model specifically performs the following process:

[0025] S4.1, the processed flux meteorological data includes wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio and surface layer turbulent flux label, the input layer transmits these data to the embedded coding layer;

[0026] S4.2, the embedded coding layer transposes the flux meteorological data transmitted by the input layer, and encodes the sequence of each variable at all time points into token tokens;

[0027] S4.3, the multivariate attention mechanism is executed on the encoded token tokens, and the correlation score matrix between the input variables wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change and surface layer turbulent flux label is calculated;

[0028] S4.4, layer normalization, feedforward neural network layer operation is performed according to the correlation score matrix, and calculation is performed using residual connection;

[0029] S4.5, layer normalization, projection are performed on the calculation result, and are transmitted to the output layer to obtain the final surface layer turbulent flux prediction result.

[0030] The method for evaluating the model in S5 is: the root mean square error RMSE, the mean absolute error MAE and the Pearson correlation coefficient R between the surface layer turbulent flux prediction value and the true value are used as evaluation indexes, the test set divided by S2 is used to test the converged model completed by S4, and the results are compared with the traditional MOST method and the existing prediction model artificial neural network (ANN). The evaluation index related formula is as follows:

[0031]

[0032] Wherein y_pred is the model output friction velocity, moving sensible heat flux, moving latent heat flux prediction value, y_true is the friction velocity, moving sensible heat flux, moving latent heat flux true observation value, n is the sample number. cov(y_pred,y_true) indicates the covariance of the prediction value and the true value, σ y_pred Indicates the standard deviation of the prediction value, σ y_true Indicates the standard deviation of the true value.

[0033] Compared with the prior art, the present application has the beneficial effects that:

[0034] The intelligent prediction method for surface turbulent flux based on iTransformer proposed in this invention has higher correlation and smaller error, resulting in superior prediction performance. Furthermore, the method employs the iTransformer model. First, by changing the embedding encoding method, the time series is transposed, and each variable's sequence at all time points is independently embedded into a token, thus more effectively capturing the correlation between variables in flux meteorological data. Second, the responsibilities of the multivariate attention mechanism and the feedforward neural network are reversed. The inverted multivariate attention mechanism is responsible for capturing the correlation between variables or time series, while the feedforward neural network is responsible for encoding and decoding each time series to extract its internal features, thereby achieving better prediction performance. Attached Figure Description

[0035] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0036] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0037] Figure 1 This is a flowchart of the steps of the present invention;

[0038] Figure 2 This is a model diagram of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. These descriptions are only for further illustrating the features and advantages of the present invention, and not for limiting the claims of the present invention. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0040] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0041] The specific implementation process of this invention is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0042] 1. Collect raw flux meteorological data at different time points.

[0043] The data collected included air density, wind speed, temperature, pressure, upward longwave radiation, relative humidity, water content, friction velocity, sensible heat flux, and latent heat flux. The data dimension was 17520×10, where 17520 represents the entire year of 2023, the sampling time interval was 30 minutes, and 10 represents the number of variables collected.

[0044] 2. Process the collected raw flux meteorological data.

[0045] First, remove missing and invalid values ​​from the collected raw flux meteorological data;

[0046] Secondly, the input variables are obtained using the following physical formulas: wind speed (V), potential temperature gradient (ΔQ). surface_high Richardson's number (R) i The water vapor mixing ratio change (Δq) is calculated as follows: wind speed can be directly collected, and the calculation formulas for the other variables are as follows:

[0047] The formula for calculating the potential temperature gradient is:

[0048]

[0049] ΔQ surface_high This is the potential temperature gradient, where T is Kelvin in degrees Celsius, P0 = 1000 hPa represents standard atmospheric pressure, P is the air pressure at altitude h, and L... wout It is upward long-wave radiation, ε = 1 is the emissivity, σ =

[0050] 5.673067×10 -8 It is the Stefan-Boltzmann constant.

[0051] The formula for calculating the Richardson number is:

[0052]

[0053] R i This is the Richardson number, g = 9.81 is the gravitational acceleration, Q is the potential temperature, and u and v are the velocity components.

[0054] The formula for calculating the change in water vapor mixing ratio is:

[0055]

[0056] Δq is the change in water vapor mixing ratio, M avail It refers to soil moisture availability, RH is relative humidity, and e s0

[0057] =6.11 hPa is the saturated vapor pressure at standard temperature (T0 = 273.15 K), L = 2.5 * 10 6 J / kg (0℃, 1 standard atmosphere) is the latent heat of vaporization of water, ε=0.622 is the ratio of the molecular weight of water vapor to dry air, and P is the atmospheric pressure (unit hPa).

[0058] M avail Soil moisture availability is calculated using the following formula:

[0059]

[0060] q fc =0.41 is the maximum amount of water that the soil can retain;

[0061] The surface turbulent flux observations include friction velocity, kinetic heat flux, and kinetic latent heat flux. The kinetic heat flux and kinetic latent heat flux are obtained by dividing the sensible heat flux and latent heat flux by the air density ρ and the air specific heat capacity cp = 1004. The input data are divided into training set, validation set and test set in a ratio of 7:2:1.

[0062] 3. Model Building

[0063] Establish an intelligent prediction model for surface turbulence based on the deep learning technology iTransformer. For example... Figure 2 As shown, the basic architecture of this model includes the following layers: input layer, embedding encoding layer, iTransformer layer, projection layer and output layer, wherein the iTransformer layer contains a multivariate attention module, a feedforward neural network module and layer normalization.

[0064] The input layer transmits data such as wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change, and surface turbulence flux label to the embedded coding layer.

[0065] The embedding encoding layer transposes and encodes the input data, embedding each variable's sequence at all time points into a token; the flux meteorological data dimension is transformed from (32, 48, 5) to (32, 5, 48), where 32 is the batch size, 48 is the length of the selected time series, and 5 is the feature dimension.

[0066] The iTransformer layer realizes the prediction of the surface turbulent flux. First, the token obtained by the embedding encoding layer is transmitted to the multivariate attention module with a residual connection. Second, the multivariate attention module can effectively capture the relationship between the input variables, i.e., wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change, and the predicted label of the surface turbulent flux. Q and K are obtained by linear projection. The score between the wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change, and the surface turbulent flux is calculated by the point multiplication of Q and K, and the score matrix A of the correlation between the surface turbulent flux and the input variables is obtained. Then, the normalized strategy is used to reduce the error of the input data, which is then transmitted to the feedforward neural network module for calculation. The calculation result is subjected to layer normalization and projection operation, and finally the prediction result is transmitted to the output layer to obtain the prediction value of the surface turbulent flux.

[0067] The specific calculation formula of the score matrix A is as follows:

[0068]

[0069] where d k is the projection dimension, q i is the specific representation of the query vector and key of the token corresponding to the i-th input variable, k j is the specific representation of the query vector and key of the token corresponding to the j-th input variable; and the obtained correlation score matrix A ∈ R N×N shows the correlation between the i-th and j-th input variables.

[0070] 4. Model training

[0071] The mean square error (MSE) between the prediction value of the surface turbulent flux and the true value is calculated as the loss function of the model. The training set and the validation set divided by S2 are used to train and validate the model constructed in S3 until the model converges. The mean square error (MSE) formula is as follows:

[0072]

[0073] where y_pred is the prediction value of the friction velocity, motion sensible heat flux, and motion latent heat flux output by the model, y_true is the true observed value of the friction velocity, motion sensible heat flux, and motion latent heat flux, and n is the number of samples.

[0074] The specific execution process of the model is as follows:

[0075] 1) The processed flux meteorological data includes wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio, and surface turbulent flux label. The input layer transmits these data to the embedding encoding layer.

[0076] 2) The embedding layer transposes the flux meteorological data passed from the input layer, encoding the sequence of each variable at all time points into token tokens.

[0077] 3) The multivariate attention mechanism is performed on the encoded token tokens, calculating the correlation score matrix between the input variables wind speed, potential temperature gradient, Richardson number, change in water vapor mixing ratio and the surface turbulent flux label.

[0078] 4) Layer normalization, feedforward neural network layer operations are performed according to the correlation score matrix, and the calculation is performed using residual connection.

[0079] 5) Layer normalization, projection are performed on the calculation results and are passed to the output layer to obtain the final surface turbulent flux prediction results.

[0080] 5. Model evaluation

[0081] The root mean square error RMSE, the mean absolute error MAE and the Pearson correlation coefficient R between the surface turbulent flux prediction value and the true value are used as evaluation indexes, the test set is tested using the converged model, and the results are compared with the traditional MOST method and the existing prediction model artificial neural network (ANN), and the related formulas are as follows:

[0082]

[0083] Where y_pred is the friction velocity, motion sensible heat flux, and motion latent heat flux prediction value output by the model, y_true is the friction velocity, motion sensible heat flux, and motion latent heat flux true observation value, n is the sample number. cov(y_pred, y_true) represents the covariance of the prediction value and the true value, σ y_pred represents the standard deviation of the prediction value, and σ y_true represents the standard deviation of the true value. The experimental results of iTransformer and the traditional MOST method are shown in Table 1, and the experimental results of iTransformer and artificial neural network (ANN) are shown in Table 2. The deep learning method of iTransformer is used to capture the correlation between wind speed, potential temperature gradient, Richardson number, change in water vapor mixing ratio and surface turbulent flux, and better surface turbulent flux prediction effect is achieved.

[0084] Table 1 Comparison of experimental results of iTransformer model and MOST method

[0085]

[0086] Table 2 Comparison of experimental results of iTransformer model and ANN model

[0087]

[0088] The above detailed description merely illustrates preferred embodiments of the application, and is not intended to limit the scope of the application. Various modifications can be made to the embodiments without departing from the spirit of the application, and all modifications are intended to be included within the scope of the application.

Claims

1. An iTransformer-based intelligent prediction method for surface layer turbulent fluxes, characterized in that, Comprise the following steps: S1, original flux meteorological data acquisition; The original flux meteorological data includes air density, wind speed, temperature, pressure, upward long-wave radiation, relative humidity, water content, friction velocity, sensible heat flux and latent heat flux; S2, flux meteorological data processing: remove the outliers and missing values in the original flux meteorological data obtained in S1, and calculate the input data using physical formula, the input data includes wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change and surface turbulent flux observation value, and the input data is divided to obtain training set, validation set and test set; The potential temperature gradient calculation formula is: where ΔQ = Q - Q0 surface_high is the potential temperature gradient, T is in degrees Celsius, P0= 1000 hPa represents the standard atmospheric pressure, P is the pressure corresponding to the height h, L wout is the upward longwave radiation, ε = 1 is the emissivity, σ = 5.673067 x 10 -8 is the Stefan-Boltzmann constant; The Richardson number calculation formula is: where R i is the Richardson number, g = 9.81 is the gravitational acceleration, and u and v are the velocity components; The water vapor mixing ratio change calculation formula is: where Δq is the change in water vapor mixing ratio, M avail is the soil water availability, RH is the relative humidity, e s0 = 6.11 hPa is the saturation vapor pressure at standard temperature, L = 2.5*10 6 J / kg, L is the latent heat of vaporization of water at 0°C and 1 standard atmosphere, γ = 0.622 is the ratio of the molecular weight of water vapor to dry air, and P is the atmospheric pressure; q fc = 0.41 is the maximum amount of water that the soil is able to retain; S3, constructing a prediction model: establishing an iTransformer surface turbulent flux intelligent prediction model based on deep learning method; The basic architecture of the model includes the following layers: input layer, embedding coding layer, iTransformer layer, projection layer and output layer, wherein the iTransformer layer includes multivariate attention module, feedforward neural network module and layer normalization; The S3 includes: The input layer transmits the wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change and surface turbulent flux label data to the embedding coding layer; The embedding coding layer transposes and encodes the input data, and Embedding the sequence of each variable at all time points into token tokens; The iTransformer layer realizes the prediction of the surface turbulent flux. Firstly, the token token obtained by the embedding coding layer is transmitted to a multivariate attention module with a residual connection. Secondly, the multivariate attention module can effectively capture the relationship between the input variables, i.e., the wind speed, the potential temperature gradient, the Richardson number, the change of the water vapor mixing ratio, and the predicted label of the surface turbulent flux. Q, K, and V are obtained by linear projection, wherein Q, K, and V are in R N×dk The score between the wind speed, the potential temperature gradient, the Richardson number, the change of the water vapor mixing ratio, and the surface turbulent flux is calculated by using the Q and K dot product, and a score matrix A of the correlation between the surface turbulent flux and the input variables is obtained. Then, the normalized strategy is used to reduce the error of the input data, and the input data is transmitted to the feedforward neural network module for calculation. The calculation result is subjected to layer normalization and projection operation, and finally the prediction result is transmitted to the output layer to obtain the predicted value of the surface turbulent flux. The calculation formula of the score matrix A is: where d k is the projection dimension, q i is the specific representation of the query vector and key for the token corresponding to the i-th input variable, k j is the specific representation of the query vector and key for the token corresponding to the j-th input variable; the resulting score matrix A e R N×N represents the correlation between the i-th and j-th input variables; S4, model training: using the mean square error MSE between the surface turbulent flux prediction value and the true value as the loss function of the model, inputting the training set and the validation set obtained in S2 into the prediction model constructed in S3 for training until the model converges; S5, model evaluation: using the test set of S2 to verify and evaluate the prediction model after S4 converges, calculating the root mean square error RMSE, mean absolute error MAE and Pearson correlation coefficient R between the surface turbulent flux prediction value and the true value as evaluation indexes, and comparing the results with the traditional MOST method and the existing prediction model artificial neural network ANN respectively.

2. The iTransformer-based intelligent prediction method for surface layer turbulent fluxes according to claim 1, characterized in that, The method for processing the collected original flux meteorological data in S2 is to remove the missing values and invalid values, obtain wind speed, potential temperature gradient, Richardson number and water vapor mixing ratio change as model input variables according to the physical formula in meteorology; Obtain the surface turbulent flux model label: friction velocity, moving sensible heat flux, moving latent heat flux according to the physical formula for calculating the surface turbulent flux in meteorology; The processed input data includes wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change and surface turbulent flux observation value, and the obtained input data is divided into training set, validation set and test set according to the ratio of 7:2:

1.

3. The iTransformer-based intelligent prediction method for surface layer turbulent fluxes according to claim 1, characterized in that, The method for training the model in S4 is to calculate the mean square error MSE between the surface turbulent flux prediction value and the true value as the loss function of the model, and use the training set and the validation set divided in S2 to train and verify the model constructed in S3 until the model converges, and the mean square error MSE formula is as follows: Wherein y_pred is the model output of the friction velocity, motion sensible heat flux, motion latent heat flux prediction value, y_true is the friction velocity, motion sensible heat flux, motion latent heat flux real observation value, and n is the number of samples.

4. The iTransformer-based intelligent prediction method for surface layer turbulent fluxes according to claim 3, characterized in that, The specific implementation process of the iTransformer surface layer turbulent flux intelligent prediction model based on the deep learning method is as follows: S4.1, the processed flux meteorological data includes wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change and surface layer turbulent flux label, and the input layer transmits these data to the embedding coding layer; S4.2, the embedding coding layer transposes the flux meteorological data transmitted by the input layer, and encodes the sequence of each variable at all time points into token tokens; S4.3, the multivariate attention mechanism is executed on the encoded token tokens, and the correlation score matrix between the input variables wind speed, potential temperature gradient, Richardson number, water vapor mixing ratio change and surface layer turbulent flux label is calculated; S4.4, according to the correlation score matrix, layer normalization, feedforward neural network layer operation is executed, and residual connection is used for calculation; S4.5, layer normalization, projection are executed on the calculation result, and the final surface layer turbulent flux prediction result is obtained by transmitting to the output layer.

5. The iTransformer-based intelligent prediction method for surface layer turbulent fluxes according to claim 1, characterized in that, The method for model evaluation in S5 is as follows: the root mean square error RMSE, the mean absolute error MAE and the Pearson correlation coefficient R between the surface layer turbulent flux prediction value and the true value are used as evaluation indexes, the test set divided in S2 is used to test the converged model completed in S4, and the results are compared with the traditional MOST method and the existing prediction model artificial neural network (ANN); the evaluation index related formula is as follows: where y pred is the model outputted friction velocity, motion sensible heat flux, motion latent heat flux prediction value, y true is the true observed friction velocity, motion sensible heat flux, motion latent heat flux value, n is the number of samples; COV(y pred, y true) represents the covariance of the prediction value and the true value, σ y_pred represents the standard deviation of the prediction value, σ y_true represents the standard deviation of the true value.

Images