Reference crop evapotranspiration prediction method suitable for limited meteorological conditions

Abstract

The invention discloses a reference crop evapotranspiration prediction method suitable for limited meteorological conditions. The reference crop evapotranspiration prediction method comprises the following steps: acquiring historical meteorological data through a meteorological station and preprocessing the historical meteorological data; calculating a reference crop evapotranspiration standard value by adopting a P-M formula; extracting important factors based on an integrated feature selection method, training a LightGBM model by using data after feature selection, automatically determining an optimal hyper-parameter combination of the LightGBM by using a northern eagle optimization algorithm, and further obtaining a trained NGO-LightGBM prediction model; and preprocessing the real-time meteorological data, and inputting the determined important factors into the model to obtain a predicted value. According to the method, the meteorological factors having significant influence on prediction of the evapotranspiration of the reference crop can be effectively extracted, the hyper-parameters in the model can be accurately adjusted, the complex nonlinear relationship between the key meteorological influence factors and the ET0 can be accurately fitted, and the ET0 can be efficiently and accurately predicted under the limited meteorological condition.

Description

Technical Field

[0001] The invention belongs to the technical field of reference crop evapotranspiration prediction, and in particular relates to a reference crop evapotranspiration prediction method suitable for limited meteorological conditions. Background Art

[0002] Precision irrigation is a key measure to improve crop yield and quality, and it is also a core way to optimize water resource utilization. In traditional irrigation management, farmers rely on experience to irrigate, which often leads to excessive water use, which not only wastes water resources but is also not conducive to crop growth. Evapotranspiration is an important indicator for assessing crop water consumption, which involves soil evaporation and crop transpiration processes. Therefore, accurately obtaining crop evapotranspiration is of great significance for calculating crop water requirements, crop irrigation scheduling, and soil and water conservation.

[0003] Although measuring crop evapotranspiration using a weighing lysimeter or eddy covariance system is a direct and effective method, these methods have the disadvantage of high instrument manufacturing and maintenance costs. To address this problem, the Penman-Monteith formula (PM) is regarded by the Food and Agriculture Organization of the United Nations as a standard method for estimating farmland evapotranspiration by calculating reference crop evapotranspiration (ET0). However, the PM formula requires many meteorological factors to be input and the calculation is complicated, which limits the application and promotion of the formula. Therefore, it is necessary to develop an ET0 model that can achieve high prediction accuracy while using as few meteorological factors as possible.

[0004] Lightweight Gradient Boosting Machine (LightGBM) is an improved ensemble learning algorithm based on Gradient Boosting Decision Tree (GBDT). Compared with other machine learning algorithms, LightGBM has excellent computing speed and accurate regression calculation of complex high-dimensional data. It can better fit the complex nonlinear relationship between meteorological factors and ET0 and has good feature extraction capabilities. However, the performance of the LightGBM algorithm model is affected by its own hyperparameters and the distribution of input data. Therefore, how to efficiently adjust the model hyperparameters and determine the model input factors to achieve accurate prediction of ET0 has become a technical problem that urgently needs to be solved. Summary of the invention

[0005] In view of this, in order to solve the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a reference crop evapotranspiration prediction method suitable for limited meteorological conditions. The proposed feature selection method can effectively extract the meteorological combination that has a significant impact on the predicted reference crop evapotranspiration, and use the Northern Goshawk Optimization Algorithm (NGO) to optimize the hyperparameters of the LightGBM model. The improved LightGBM algorithm can accurately fit the complex nonlinear relationship between key meteorological influencing factors and ET0, so as to achieve efficient and accurate prediction of ET0 under limited meteorological conditions.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is: A reference crop evapotranspiration prediction method applicable to limited meteorological conditions comprises the following steps: S1, obtain historical meteorological data from meteorological stations in farmland and perform preprocessing; S2, using the PM formula to calculate the reference crop evapotranspiration standard value as the label for model training; S3, based on the integrated feature selection method, feature selection is performed on the historical meteorological data preprocessed in S1 to obtain key meteorological factors; S4, use the data after feature selection in S3 to train the LightGBM model, and use the Northern Goshawk optimization algorithm to optimize the hyperparameters of the LightGBM model to build the NGO-LightGBM prediction model; S5, collect meteorological data in real time, and process the real-time meteorological data using the same data preprocessing method as S1; S6. According to the key meteorological factors selected in S3, the meteorological data preprocessed in S5 is input into the NGO-LightGBM prediction model to obtain the reference crop evapotranspiration prediction result.

[0007] Furthermore, the preprocessing of S1 includes detecting and processing missing values ​​and outliers, and normalizing the data.

[0008] Furthermore, the feature selection in S3 is performed based on an integrated feature selection method, including the following contents: S31, let the meteorological factors collected by the meteorological station be the vector composed of features J = (t1, t2, ..., t n ) T , where n represents the number of features, t i represents the i-th feature; S32, the integrated feature selection method includes M feature selection methods, M = 1, 2, ..., m, feature t i The importance score is taken in each feature selection method, and the feature selection result of the mth method is J m , J m =(t m1 ,t m2 ,...,t mn ) T , combine the results of all the above feature selection methods to construct the feature selection matrix T; S33, normalizing the above importance scores; S34. Sum each row of the feature selection matrix T to obtain n values ​​and arrange them in descending order. The features with the highest ranking are the important features. Finally, the important input combinations are constructed based on the ranking.

[0009] Furthermore, the feature selection matrix T is: Among them, a column of the matrix T represents the results of all meteorological factors under a certain feature selection method.

[0010] Furthermore, the Northern Goshawk Optimization Algorithm (NGO) is used in S4 to optimize the hyperparameters of the LightGBM model, including the following sub-steps: S41. Take the hyperparameters of the LightGBM model as the target to be optimized, and determine the number of iterations (T) of the NGO algorithm, the number of total members (N), the dimension of the problem to be solved, and the hyperparameter optimization limit; S42, NGO algorithm random initialization; S43, Phase 1, in the initial hunting phase, the northern goshawk randomly finds a target and then quickly attacks it; this phase involves a global search of the search space to find the best area; S44, the second stage, when the prey is attacked by the northern goshawk, it will adopt an escape strategy, and the northern goshawk quickly locks on the prey target and continues to chase; this stage includes a local search of the locked area; S45, after updating all population members based on the NGO algorithm of the first stage and the second stage, completing the iteration of the algorithm, determining the new values ​​of the population members, the objective function and the best suggested solution; S46, continue to iterate and update the population members, and when the number of iterations is reached, the position of the northern goshawk located at the final prey position is the optimal solution after optimization; S47. Based on the optimal hyperparameters of the LightGBM model obtained in step S46, an NGO-LightGBM reference crop evapotranspiration prediction model is established.

[0011] The beneficial effects of the present invention are: The present invention proposes a reference crop evapotranspiration prediction method suitable for limited meteorological conditions. By introducing the northern goshawk optimization algorithm, the northern goshawk's predation behavior can be simulated during the LightGBM training process, and the hyperparameter weights of the LightGBM can be intelligently adjusted to determine the optimal hyperparameters, and the complex nonlinear relationship between key meteorological factors and ET0 can be accurately fitted. It can effectively solve the problem of hyperparameter adjustment, improve the prediction accuracy of the LightGBM algorithm, and achieve efficient and accurate prediction of ET0 under limited meteorological conditions. An integrated strategy combining multiple embedded feature selection methods is further proposed, which can automatically evaluate the importance of meteorological factors to the predicted ET0 during the model training process, effectively extract meteorological factors that have a significant impact on the predicted reference crop evapotranspiration, and improve the interpretability of the model; avoid the subjectivity of manual feature selection, and overcome the problem that the results of a single embedded method depend on the specific algorithm, effectively eliminate redundant features, and screen out key meteorological factors that have a significant impact on the predicted ET0, thereby achieving efficient and accurate prediction of reference crop evapotranspiration under limited meteorological conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 It is a framework diagram of the method for predicting reference crop evapotranspiration under limited meteorological conditions of the present invention; Figure 2 It is a flowchart of LightGBM hyperparameter optimization based on the Northern Goshawk algorithm of the present invention; Figure 3 It is a map of the area where the meteorological station is located in Example 2 of the present invention; Figure 4 is a feature importance score map based on the integrated embedded feature selection method in Example 2 of the present invention; Figure 5 It is a scatter density diagram of the NGO-LightGBM prediction model in Example 2 of the present invention. DETAILED DESCRIPTION

[0014] Specific embodiments are given below to further clearly, completely and in detail illustrate the technical solution of the present invention. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0015] Example 1 A reference crop evapotranspiration prediction method applicable to limited meteorological conditions comprises the following steps: Step 1: Obtain historical meteorological data from the meteorological station in the farmland and perform preprocessing, including detection and processing of missing values ​​and outliers, and normalization of the data. The specific sub-steps include: S11. Historical daily data collected by meteorological stations in farmland, including daily maximum temperature, daily minimum temperature, sunshine hours, relative humidity, wind speed and air pressure; S12. The method for detecting outliers is the quartile criterion: L1=QL-1.5·IQR L2=QU+1.5·IQR Among them, QL is called the lower quartile, which means that one-quarter of all data has a smaller value than it; QU is called the upper quartile, which means that one-quarter of all data has a larger value than it; IQR is called the interquartile range, which is the difference between the upper quartile QU and the lower quartile QL; L1 is the lower limit of the abnormal threshold, and L2 is the upper limit of the abnormal threshold. The abnormal value judgment standard is that the sample value is less than the lower limit L1 or greater than the upper limit L2; S13. The method for processing missing values ​​and outliers is the K nearest neighbor interpolation method; this method identifies k samples in the data set that are close to the missing value space, and uses the Euclidean distance to measure the distance between these samples and the missing values. The missing values ​​and outliers of each sample are interpolated using the average value of the k samples found in the data set; S14, normalize the data after processing missing values ​​and outliers in step S13, the formula is: Among them, X is the normalized value; x is the input value; x max and x min They are the maximum and minimum values ​​in the data sequence corresponding to the input value x.

[0016] Step 2: Use the PM formula to calculate the reference crop evapotranspiration standard value as the label for model training; the specific formula is as follows: Where ET0 is the reference crop evapotranspiration, mm / d; Δ is the slope of the saturated water vapor pressure-temperature curve, kPa / ℃; γ is the hygrometer constant, kPa / ℃; Rn is the net radiation, MJ / (m 2 ·d); e s is the saturated water vapor pressure, kPa; e a is the actual water vapor pressure, kPa; T mean is the average temperature, °C; G is the soil heat flux, MJ / (m 2 ·d); u2 is the wind speed at 2 meters above the ground, m / s.

[0017] Step 3: Use an integrated feature selection method to perform feature selection on the preprocessed historical meteorological data. The integrated feature selection method combines multiple embedded feature selection methods. When performing feature selection, an integrated strategy is used to combine the feature selection results of multiple embedded feature selection methods, specifically including the following contents: S31. Assume that the meteorological factors collected by the meteorological station are used as the feature vector J = (t1, t2, ..., t n ) T , where n represents the number of features, t i represents the i-th feature; if feature t i Importance score t in feature selection method i The value is 2, that is, t i =2; Similarly, other features ti are assigned values ​​in the same way; S32. If there are m feature selection methods, the feature selection result of the mth method is J m , J m =(t m1 ,t m2 ,...,t mn ) T .J m is a column matrix. By combining the results of all feature selection methods, the feature selection matrix T is constructed; Among them, one column of the matrix T represents the results of all meteorological factors under a certain feature selection method; S33. In view of the different importance score ranges assigned by different methods, the scores are normalized to unify the metric standard; then, each row of the feature selection matrix T is summed, such as the sum of the i-th row is t 1i +t 2i +...+t mi Finally, n values ​​are obtained, and these n results are arranged in descending order. The top-ranked features are the important features, and finally the important input combinations are constructed based on the rankings.

[0018] In this embodiment, the multiple embedded feature selection methods include random forest (RF), adaptive boosting (AdaBoost), CatBoost, gradient boosting decision tree (GBDT) and LightGBM. The principles of the five methods are as follows: RF uses the out-of-bag (OOB) error to calculate the importance of each feature during model training. GBDT is an integrated algorithm based on Boosting. LightGBM is an improved framework based on GBDT. It measures the importance of features by calculating the frequency with which features are used to divide data on all trees and the average gain brought by the features as partitioning attributes. During the iterative training of the learner, the AdaBoost model updates the sample weights based on the performance of the previous round of learners and obtains the importance of features by using the weights of samples in each iteration. The CatBoost model uses a symmetric decision tree as the base learner, which can calculate the importance of features using the loss function or the change in the predicted value.

[0019] Step 4: Use the data after feature selection in S3 to train the LightGBM model, and use the Northern Goshawk Optimization Algorithm (NGO) to optimize the hyperparameters of the LightGBM model to build the NGO-LightGBM prediction model.

[0020] The specific steps include: S41. Take the hyperparameters of the LightGBM model as the target to be optimized, and determine the number of iterations (T) of the NGO algorithm, the number of total members (N), the dimension of the problem to be solved, and the hyperparameter optimization limit; S42, NGO algorithm random initialization process, the formula is: Where X represents the population matrix of northern goshawk, X i represents the initial solution of the i-th goshawk in the population, j represents the specific dimension, x i,j is the value of the jth variable specified by the ith proposed solution, N represents the population size, m represents the solution dimension, F is the objective function vector of the northern goshawk population, and F i is the objective function value of the i-th northern goshawk; S43, the first stage, in the initial hunting stage, the northern goshawk randomly finds a target and then quickly attacks it. This stage includes a global search of the search space to find the best area, the formula is: P i =X k ,i=1,2,…,N,k=1,2,…,i-1,i+1,…,N, Among them, P i is the position of the i-th northern goshawk’s prey, k is a random integer in the range [1, N], F pi is the objective function value of the i-th northern goshawk’s prey position, is the new state of the i-th northern goshawk, F i new,P1 is the objective function value of the i-th northern goshawk in the first stage, r is a random number in the range [0,1], and I is a random number with a value of 1 or 2; S44. When the prey is attacked by the northern goshawk, it will take the escape strategy. However, the northern goshawk has efficient hunting ability and flies very fast. It will quickly lock the prey target and continue to chase. In the end, the prey can be hunted in any situation. The simulation of this hunting behavior improves the utilization ability of local search in the search space, and the formula is: Among them, t is the current iteration number, T is the maximum iteration number of the population, R is the hunting range of the northern goshawk, is the new state of the i-th northern goshawk in the second stage, F i new,P2 is the corresponding objective function value; S45, after updating all population members based on the NGO algorithm of the first stage and the second stage, completing the iteration of the algorithm, determining the new values ​​of the population members, the objective function and the best suggested solution; S46, continue to iterate and update the population members, and when the number of iterations is reached, the position of the northern goshawk located at the final prey position is the optimal solution after optimization; S47. Based on the optimal hyperparameters of the LightGBM model obtained in step S46, an NGO-LightGBM reference crop evapotranspiration prediction model is established.

[0021] Step 5: Use the same data preprocessing method as step 1 to process the real-time collected meteorological data, and input the corresponding processed real-time meteorological data into the NGO-LightGBM prediction model according to the important meteorological factors selected in step 3 to obtain the reference crop evapotranspiration prediction results.

[0022] Example 2 Based on a reference crop evapotranspiration prediction method suitable for limited meteorological conditions provided in Example 1, daily meteorological observation data from 30 meteorological stations in the North China Plain from January 1968 to December 2019 are selected. The locations of the meteorological stations are as follows: Figure 3 As shown, efficient and accurate prediction of ET0 can be achieved under limited meteorological conditions.

[0023] All meteorological data are from the National Meteorological Information Center of the China Meteorological Administration (CMA) and are subject to strict quality control, including minimum temperature (Tmin), maximum temperature (Tmax), wind speed at 10 m (u10), sunshine hours (n), relative humidity (RH) and air pressure (P), etc. The wind speed is calculated according to the PM calculation guide, and the wind speed u10 is converted into the wind speed u2 at 2 m above the surface according to the wind profile relationship.

[0024] The specific prediction method is as follows: S1. Obtain historical meteorological data from meteorological stations in farmland and perform preprocessing, including detection and processing of missing values ​​and outliers, and normalization of data.

[0025] After the detection and processing of missing values ​​and outliers, the meteorological data of each station is divided into two parts at a ratio of 4:1. The first part (1968-2009) is used to train / calibrate the machine learning model, and the second part (2010-2019) is used to test the model performance. In this embodiment, the training set is used as historical meteorological data, and the test set is used as measured meteorological data.

[0026] In order to eliminate the influence of the difference in dimensions between different meteorological factors on the model fitting effect, the training set was normalized before model training. The formula is: Among them, X is the normalized value, x is the input value, and x max and x min are the maximum and minimum values ​​in the data sequence corresponding to the input value x, respectively. Each meteorological factor in the training set is normalized to [-1,1].

[0027] S2. Based on the meteorological data obtained in S1, the PM formula is used to calculate the reference crop evapotranspiration standard value as the label of the data set. The formula is as follows: Where ET0 is the reference crop evapotranspiration, mm / d; Δ is the slope of the saturated water vapor pressure-temperature curve, kPa / ℃; γ is the hygrometer constant, kPa / ℃; Rn is the net radiation, MJ / (m 2 ·d); e s is the saturated water vapor pressure, kPa; e a is the actual water vapor pressure, kPa; T mean is the average temperature, °C; G is the soil heat flux, MJ / (m 2 ·d); u2 is the wind speed at 2 meters above the ground, m / s.

[0028] S3. Use the integrated feature selection method to perform feature selection on the preprocessed historical meteorological data. This embodiment uses the data of each meteorological station after data preprocessing to perform feature selection, and obtains the importance scores of 6 meteorological factors for predicting ET0 of 30 meteorological stations in the North China Plain under 5 feature selection methods, such as Figure 4 As shown in the figure. The meteorological factors selected by the five embedded methods are integrated with feature selection to obtain the feature selection matrix T. The first, second, third, fourth and fifth columns in the matrix are the results of feature selection by RF, GBDT, LightGBM, AdaBoost and CatBoost methods, respectively. Each column element is the importance score of each feature.

[0029] Each row of data in the matrix T corresponds to each meteorological factor. The data in each row are summed up, and the top-ranked features are the key influencing factors. Based on the ranking of important factors obtained by the integrated embedded feature selection method, the top five meteorological factors with the highest importance for predicting ET0 are Tmax, n, RH, u2, and P.

[0030] As an option, this embodiment selects the top 3, 4, and 5 factors to form 3 important meteorological factor combinations as input items for constructing the ETO model.

[0031] S4. Use the data after feature selection in S3 to train the LightGBM model, and use the Northern Goshawk Optimization Algorithm (NGO) to optimize the hyperparameters of the LightGBM model to build the NGO-LightGBM prediction model.

[0032] This example uses the key hyperparameters of the LightGBM model as the target to be optimized, including: n estimators, max depth, learning rate, num_leaves, max_bin and min_data_in_leaf; Among them, n estimators ranges from 100 to 1000, max depth ranges from 1 to 25, learning rate ranges from 0.001 to 0.1, num_leaves ranges from 5 to 200, max_bin ranges from 50 to 500, min_data_in_leaf ranges from 5 to 100, and other hyperparameters are default values; Initialize the parameters of the NGO algorithm: number of iterations (T) = 50, number of total members (N) = 30, solution dimension = 6.

[0033] S5, using the same data preprocessing method as step S1 to process the meteorological data collected in real time, and using the important meteorological data selected in step S3 to input into the NGO-LightGBM prediction model to obtain the reference crop evapotranspiration prediction result; the details are as follows: According to the method of preprocessing the training set in S1, the same rules are used to preprocess the test set. Subsequently, this embodiment transfers the preprocessed test set data into the NGO-LightGBM reference crop evapotranspiration prediction model based on the important meteorological factors selected in step S3, and outputs the prediction results.

[0034] This embodiment uses root mean square error (RMSE), mean absolute error (MAE), Nash efficiency coefficient (NSE), determination coefficient (R 2 ) evaluates the error and consistency between the prediction results of the model constructed in this embodiment and the standard value of PM calculation. 2 The closer it is to 1, and the closer MAE and RMSE are to 0, the better the performance of the model.

[0035] R 2 The relationship between the true value and the estimated value was evaluated, R 2 The calculation formula is as follows: NSE is used to evaluate the prediction accuracy and fit of the model. The NSE calculation formula is as follows: RMSE is a common indicator used to measure model performance, which represents the root mean square difference between the predicted value and the actual value. The RMSE calculation formula is as follows: MAE represents the actual situation of the predicted value deviation. The MAE calculation formula is as follows: Where, X i represents ET0 calculated by PM equation, Y i represents the ET0 predicted by the machine learning model, and Indicates the corresponding average value.

[0036] Results Analysis As shown in Table 1 below, the statistical accuracy index results of NGO-LightGBM, LightGBM, AdaBoost, Decision Tree and KNN models for predicting daily ET0 at 30 weather stations in the North China Plain under three important combinations of inputs are shown. Among them, LightGBM, AdaBoost, Decision Tree and KNN models all use their own default hyperparameters, while the NGO-LightGBM model uses the Northern Goshawk optimization algorithm in the present invention to optimize the LightGBM hyperparameters.

[0037] The results show that compared with the four original models (LightGBM, AdaBoost, Decision Tree and KNN models), the NGO-LightGBM model has the highest prediction accuracy. The scatter density plots of the NGO-LightGBM model under the three input combinations are shown in Figure 2. Figure 5 shown. Table 1 Statistical accuracy index results of ET0 prediction model at 30 meteorological stations in North China Plain In summary, under the three important combinations, NGO-LightGBM can accurately reflect the complex nonlinear relationship between meteorological factors and ET0, and can obtain satisfactory prediction accuracy (R 2 >0.9). Compared with LightGBM, AdaBoost, Decision Tree and KNN models, the NGO-LightGBM model achieved the best ET0 prediction performance under the three input combinations.

[0038] The above shows and describes the main features, basic principles and advantages of the present invention. The present invention is not limited to the above model, and can be analyzed and calculated for different number of segments and different types of cages; technicians in this industry should understand that the present invention is not limited to the above embodiment, and the above embodiment and description are only to illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will also have various changes and improvements according to actual conditions, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A reference crop evapotranspiration prediction method applicable to limited meteorological conditions, It is characterized in that The following steps are involved: S1, obtain historical meteorological data from meteorological stations in farmland and perform preprocessing; S2, using the PM formula to calculate the reference crop evapotranspiration standard value as the label for model training; S3, based on the integrated feature selection method, feature selection is performed on the historical meteorological data preprocessed in S1 to obtain key meteorological factors; S4, use the data after feature selection in S3 to train the LightGBM model, and use the Northern Goshawk optimization algorithm to optimize the hyperparameters of the LightGBM model to build the NGO-LightGBM prediction model; S5, collect meteorological data in real time, and process the real-time meteorological data using the same data preprocessing method as S1; S6. According to the key meteorological factors selected in S3, the meteorological data preprocessed in S5 is input into the NGO-LightGBM prediction model to obtain the reference crop evapotranspiration prediction result.

2. A reference crop evapotranspiration prediction method applicable to limited meteorological conditions according to claim 1, It is characterized in that The preprocessing of S1 includes detecting and processing missing values ​​and outliers, and normalizing the data.

3. A reference crop evapotranspiration prediction method applicable to limited meteorological conditions according to claim 2, It is characterized in that The S1 specifically includes the following sub-steps: S11, obtaining historical daily data collected by the meteorological station in the farmland, including daily maximum temperature, daily minimum temperature, sunshine hours, relative humidity, wind speed and air pressure; S12, use the quartile criterion to detect outliers; S13, processing missing values ​​and outliers by K nearest neighbor interpolation method; S14, normalizing the data after processing missing values ​​and outliers in step S13.

4. A reference crop evapotranspiration prediction method applicable to limited meteorological conditions according to claim 1, It is characterized in that In S2, the PM formula is used to calculate the standard value of reference crop evapotranspiration, and the formula is as follows: Among them, ET 0 is the reference crop evapotranspiration, mm / d; Δ is the slope of the saturated water vapor pressure-temperature curve, kPa / ℃; γ is the hygrometer constant, kPa / ℃; Rn is the net radiation, MJ / (m 2 ·d); e s is the saturated water vapor pressure, kPa; e a is the actual water vapor pressure, kPa; T mean is the average temperature, °C; G is the soil heat flux, MJ / (m 2 ·d);u 2 Wind speed at 2 meters above the ground, m / s.

5. A reference crop evapotranspiration prediction method applicable to limited meteorological conditions according to claim 1, It is characterized in that In S3, feature selection is performed based on an integrated feature selection method. Includes the following: S31, assuming that the meteorological factors collected by the meteorological station are used as the vector of features J = (t 1 ,t 2 ,...,t n ) T , where n represents the number of features, t i represents the i-th feature; S32, the integrated feature selection method includes M feature selection methods, M = 1, 2, ..., m, feature t i The importance score is taken in each feature selection method, and the feature selection result of the mth method is J m , J m =(t m1 ,t m2 ,...,t mn ) T , combine the results of all the above feature selection methods to construct the feature selection matrix T; S33, normalizing the above importance scores; S34. Sum each row of the feature selection matrix T to obtain n values ​​and arrange them in descending order. The features with the highest ranking are the important features. Finally, the important input combinations are constructed based on the ranking.

6. A reference crop evapotranspiration prediction method applicable to limited meteorological conditions according to claim 5, It is characterized in that The feature selection matrix T is: Among them, a column of the matrix T represents the results of all meteorological factors under a certain feature selection method.

7. A reference crop evapotranspiration prediction method applicable to limited meteorological conditions according to claim 5, It is characterized in that The M is 5, and the feature selection methods are random forest RF, adaptive boosting AdaBoost, CatBoost, gradient boosting decision tree GBDT and lightweight gradient boosting machine LightGBM.

8. A reference crop evapotranspiration prediction method applicable to limited meteorological conditions according to claim 1, It is characterized in that In S4, the hyperparameters of the LightGBM model are optimized using the Northern Goshawk optimization algorithm, including the following sub-steps: S41. Take the hyperparameters of the LightGBM model as the target to be optimized, and determine the number of iterations of the NGO algorithm, the number of overall members, the dimension of the problem to be solved, and the hyperparameter optimization limit; S42, NGO algorithm random initialization; S43, Phase 1, in the initial hunting phase, the northern goshawk randomly finds a target and then quickly attacks it; this phase involves a global search of the search space to find the best area; S44, the second stage, when the prey is attacked by the northern goshawk, it will adopt an escape strategy, and the northern goshawk quickly locks on the prey target and continues to chase; this stage includes a local search of the locked area; S45, after updating all population members based on the NGO algorithm of the first stage and the second stage, completing the iteration of the algorithm, determining the new values ​​of the population members, the objective function and the best suggested solution; S46, continue to iterate and update the population members, and when the number of iterations is reached, the position of the northern goshawk located at the final prey position is the optimal solution after optimization; S47. Based on the optimal hyperparameters of the LightGBM model obtained in step S46, an NGO-LightGBM reference crop evapotranspiration prediction model is established.

Images