A Crop Irrigation Water Requirement Prediction Method Based on AquaCrop Model and SVR

Abstract

The present invention provides a method for predicting crop irrigation water requirement based on the AquaCrop model and SVR, comprising the following steps: S1, obtaining meteorological station data of the planting area, soil data of the planting area, crop parameter data and field management data; S2, performing local debugging of the AquaCrop model by the trial-and-error method; S3, using the AquaCrop model to select appropriate irrigation strategies and simulate different irrigation regimes; S4, based on the simulation results of the AquaCrop model, combining with SVR to predict the daily crop irrigation water requirement in the next growth period. The present invention utilizes the existing meteorological station data and crop planting management data, combines the crop model with the machine learning method, can select appropriate irrigation strategies according to the water resource supply and demand conditions and crop growth water requirement characteristics in different regions, and predicts the daily crop irrigation water requirement in the next year through the support vector machine combined with the particle swarm optimization algorithm, improves the prediction accuracy, and can provide an effective reference basis for local agricultural planning water use.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural irrigation prediction methods, and particularly relates to a method for predicting crop irrigation water requirement based on the AquaCrop model and SVR. Background Technique

[0002] Water resource shortage is a major problem faced by the world today. The continuous growth of the population and the increasingly severe drought phenomenon have exerted unprecedented pressure on the sustainable utilization of water resources, thus causing a certain degree of food crisis. Agricultural irrigation is the main consumer of fresh water resources, but the effectiveness of irrigation water is very poor and water resources are wasted seriously. Therefore, effective irrigation management strategies and prediction technologies are needed to balance water conservation and crop production.

[0003] The core of crop irrigation decision-making is the prediction of irrigation water requirement. Most of the existing methods for predicting crop irrigation water requirement are based on deploying field sensors to obtain soil moisture, which is expensive and requires huge engineering work.

[0004] In addition, the reduction of agricultural irrigation water and the low water use efficiency make the traditional full irrigation according to the water requirement of crops no longer adapt to the objective reality and do not conform to the connotation of scientific irrigation. The non-full irrigation in different regions should be guided according to the meteorological conditions, soil conditions and water resource conditions in different regions, rationally allocate water volume, predict irrigation water requirement and predict crop yield to minimize the impact of water shortage on crops.

[0005] Most traditional irrigation strategies are based on the combination of field experiments and the use of optimization algorithms, such as linear, non-linear, dynamic programming and genetic algorithms. However, field experiments are often time-consuming and laborious, and it is very difficult to consider the influence of climate types, soil types and agronomic measures in different years and regions on crop yield, which is extremely inconvenient. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for predicting crop irrigation water requirement based on the AquaCrop model and SVR in view of the deficiencies of the above-mentioned prior art, so as to solve the problems put forward in the above background technique.

[0007] To solve the above technical problem, the technical solution adopted by the present invention is: A method for predicting crop irrigation water requirement based on the AquaCrop model and SVR, comprising the following steps:

[0008] S1. Obtain the meteorological station data, soil data, crop parameter data and field management data of the planting area;

[0009] S2. Conduct local debugging of the AquaCrop model by the trial-and-error method;

[0010] S3. Select appropriate irrigation strategies using the AquaCrop model and simulate different irrigation regimes;

[0011] S4. Based on the simulation results of the AquaCrop model, combine with SVR to predict the daily irrigation water requirement of the crop during the next growth period.

[0012] Furthermore, in S1, the meteorological station data includes daily-scale precipitation data, temperature data, relative humidity, wind speed, and sunshine hours, and the reference crop evapotranspiration ET0 is calculated using the meteorological station data according to the Penman-Monteith formula.

[0013] Furthermore, the crop parameter data includes growth period parameter data and crop yield data.

[0014] Furthermore, in S3, there are 6 built-in irrigation strategy formulation methods in the AquaCrop model, namely:

[0015] IrrMethod = 0: Rainfed irrigation;

[0016] IrrMethod = 1: Irrigation is triggered when the soil water content in the root zone is lower than the specified threshold;

[0017] IrrMethod = 2: Irrigation is triggered every N days;

[0018] IrrMethod = 3: Predefined irrigation plan;

[0019] IrrMethod = 4: Net irrigation, filling all soil layer compartments every day to keep the soil moisture at the preset level;

[0020] IrrMethod = 5: Irrigate a fixed depth every day;

[0021] Then, select the optimal irrigation strategy formulation method according to the yield and the average annual irrigation water volume, conduct continuous multi-year growth simulation of the crop, and obtain its daily irrigation water volume data.

[0022] Furthermore, S4 includes the following steps:

[0023] S4.1. Combine the meteorological factors, specifically the maximum temperature T max , °C, the minimum temperature T min , °C, precipitation P, mm, reference evapotranspiration ET0, mm / day, sunshine hours SD, h / day with the daily irrigation water volume data output by the AquaCrop model in time series as the input variables of SVR;

[0024] S4.2. Standardize the eigenvalue with standard deviation Z-zero;

[0025] S4.3. Divide the input variables of SVR into a training set and a test set, and determine the highest temperature T max , °C, the lowest temperature T min , °C, precipitation P, mm, reference evapotranspiration ET0, mm / day, sunshine hours SD, h / day, and the best combination of Irrday, mm.

[0026] The present invention has the following advantages compared with the prior art:

[0027] The present invention provides a method for predicting crop irrigation water requirement based on the AquaCrop model and SVR. By using the existing meteorological station data and crop planting management data, combining the crop model with the machine learning method, it can select appropriate irrigation strategies according to the water supply and demand conditions of water resources in different regions and the water requirement characteristics of crop growth. By combining the support vector machine with the particle swarm optimization algorithm, it can predict the daily irrigation water requirement of crops in the next year, improve the prediction accuracy, and provide an effective reference basis for local agricultural planning water use. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is the principle framework diagram of the present invention;

[0029] Figure 2 is the calibration result diagram of the AquaCrop model in the present invention;

[0030] Figure 3 is the principle diagram of SVR;

[0031] Figure 4 is the flowchart of PSO;

[0032] Figure 5 is the prediction result diagram of SVR. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0034] Example 1, as Figure 1 shown, the present invention provides a technical solution: a method for predicting crop irrigation water requirement based on the AquaCrop model and SVR, including the following steps:

[0035] S1. Obtain the meteorological station data, soil data, crop parameter data, and field management data of the planting area; the meteorological station data includes daily-scale precipitation data, temperature data, relative humidity, wind speed, and sunshine hours, and calculate the reference crop evapotranspiration ET0 according to the Penman-Monteith formula using the meteorological station data;

[0036] The crop parameter data includes growth stage parameter data and crop yield data.

[0037] S2. Conduct local debugging of the AquaCrop model through the trial-and-error method;

[0038] S3. Use the AquaCrop model to select appropriate irrigation strategies and conduct simulations of different irrigation regimes;

[0039] Six irrigation strategy formulation methods are built into the AquaCrop model, namely:

[0040] IrrMethod = 0: Rainfed irrigation;

[0041] IrrMethod = 1: Trigger irrigation when the soil water content in the root zone is lower than the specified threshold;

[0042] IrrMethod = 2: Trigger irrigation once every N days;

[0043] IrrMethod = 3: Predefined irrigation plan;

[0044] IrrMethod = 4: Net irrigation, fill all soil layer compartments every day to keep the soil moisture at the preset level;

[0045] IrrMethod = 5: Irrigate a fixed depth every day;

[0046] Then select the optimal irrigation strategy formulation method according to the yield and the average annual irrigation water volume, conduct continuous multi-year growth simulations of the crop, and obtain its daily irrigation water volume data.

[0047] S4. Based on the simulation results of the AquaCrop model, combine with SVR to predict the daily irrigation water requirement of the crop in the next growth stage, specifically including the following steps:

[0048] S4.1. Combine the meteorological factors, specifically the maximum temperature T max , °C, the minimum temperature T min , °C, precipitation P, mm, reference evapotranspiration ET0, mm / day, and sunshine hours SD, h / day with the daily irrigation water volume data output by the AquaCrop model in time series as the input variables of SVR;

[0049] S4.2. Standardize the eigenvalue by standard deviation, Z-zero;

[0050] S4.3. Divide the input variables of SVR into a training set and a test set, and determine the highest temperature T max , °C, the lowest temperature T min , °C, precipitation P, mm, reference evapotranspiration ET0, mm / day, sunshine hours SD, h / day, and the best combination of Irrday, mm.

[0051] Experimental Example 1, as Figures 2 - 5 shown, based on Example 1, conduct growth simulation and irrigation water volume prediction for winter wheat in Luancheng District, Shijiazhuang City, Hebei Province under different irrigation systems.

[0052] Select the winter wheat field experiment data from 2000 to 2006 for AquaCrop model calibration, refer to the water use quota standard in Hebei Province, and use the winter wheat yield data in the Hebei Statistical Yearbook from 1996 to 2018 for verification. The SVR model dataset selects the daily irrigation water volume of winter wheat in Luancheng District simulated by AquaCrop from 1995 to 2018 and the meteorological data of Luancheng Station from 1995 to 2018. Among them, the dataset from 1995 to 2015 is used as the training set, and the dataset from 2015 to 2018 is used as the test set.

[0053] S1. Obtain the meteorological station data in Luancheng District, Shijiazhuang City, Hebei Province from 1995 to 2018. The meteorological station data includes daily-scale precipitation data, temperature data, relative humidity, wind speed, and sunshine hours. Use the meteorological data and calculate ET0 according to the Penman-Monteith formula recommended by the Food and Agriculture Organization of the United Nations. The specific calculation is as follows:

[0054]

[0055] In the formula, Rn is the net radiation at the crop surface [MJ / (m 2 ·d)];

[0056] G is the soil heat flux, MJ / (m 2 ·d);

[0057] γ is the psychrometer constant, kPa / °C;

[0058] T is the average air temperature, °C;

[0059] u2 is the wind speed at 2 m above the ground, m 2 / s;

[0060] e s is the saturated water vapor pressure of the air, kPa;

[0061] e a is the actual water vapor pressure of air, kPa;

[0062] Δ is the slope of the curve of saturated water vapor pressure versus temperature, kPa / °C.

[0063] The organized meteorological data is made into a weather file as the input of the meteorological data for the AquaCrop model.

[0064] Then, the soil data of the planting area, crop parameter data, including growth period parameter data, crop yield per unit area data, and field management data are obtained.

[0065] S2. Localize and debug the AquaCrop model through the trial-and-error method.

[0066] The soil data and crop parameter data are from the research of Zhang Xiying in 2008. The winter wheat yield per unit area data in Luancheng District from 1995 to 2018 is from the Hebei Statistical Yearbook. Other crop parameter references are from the AquaCrop model user manual and the relevant research of other scholars. The field management data, including crop planting density and fertilization level parameters, are obtained according to the experience of local farmers and the research of other scholars. The AquaCrop model provides the typical characteristics of wheat and related crop parameters. In this embodiment, referring to the parameter range given by the model and the research of other scholars, the parameters in the model are debugged through the trial-and-error method to obtain the crop parameter values suitable for the local area.

[0067] Based on the root mean square error (RMSE) and the coefficient of determination (R 2 ), the performance of model calibration is evaluated:

[0068]

[0069]

[0070] where S i and O i are the simulated value and the measured value respectively;

[0071] n is the number of measured values;

[0072] S and O are the means of the simulated value and the measured value respectively.

[0073] The closer the value of RMSE is to 0, the smaller the simulation error. The closer R 2 is to 1, the better the model prediction ability.

[0074] To ensure the accuracy of model simulation, the AquaCrop model is calibrated and verified under the condition of no water stress.

[0075] Figure 2This is the calibration result graph of the AquaCrop model. The results show that there is a strong linear relationship between the simulated and measured yields of winter wheat under different irrigation strategies, indicating that the AquaCrop model can accurately simulate the yield changes of winter wheat.

[0076] Step S3: Simulate the yields and irrigation amounts of winter wheat under different irrigation scenarios.

[0077] After the successful local debugging of the AquaCrop model parameters, the yield responses and daily irrigation demands of winter wheat under different irrigation schemes were simulated for the time series from 1995 to 2018. According to IrrMethod4, three irrigation threshold nodes were set, representing deficit irrigation (15%), normal water use condition (30%), and full water supply (50%) respectively. Each threshold represents the limit of available water. If the root zone soil moisture threshold is less than or equal to the set value, the automatic irrigation cycle is activated. Table 1 shows the yields and total irrigation demands of winter wheat under the three irrigation scenarios from 1995 to 2018.

[0078] Table 1 Yields and irrigation amounts of winter wheat under three irrigation scenarios from 1995 to 2018

[0079]

[0080]

[0081] The results show that the irrigation amounts of winter wheat vary to a certain extent in different years. Under the three irrigation scenarios, the total irrigation amounts vary greatly, but the yield change range is relatively small.

[0082] Step S4: Based on the simulation results of the AquaCrop model, combine with SVR to predict the daily irrigation water demand of the crop in the next growth period.

[0083] Figure 3 This is the schematic diagram of SVR. The model structure creates a transfer function that maps the input to a high-dimensional feature space, called the kernel function. In this space, linear regression can be performed to determine the non-linear distribution of the input data.

[0084] SVM provides four types of kernel functions: linear kernel function LINEAR, polynomial kernel function POLY, radial basis kernel function RBF, and sigmoid kernel function, the non-linear activation function of neurons.

[0085] The choice of the kernel function depends on the requirements of data processing. Among them, the RBF kernel function has good properties and shows good performance in practical problems.

[0086] In this experimental example, the RBF kernel function was selected, and the initial parameters were determined as gamma = 0.1, C = 0, epsilon = 100. The specific steps for predicting the daily irrigation water requirement of crops are as follows:

[0087] S4.1. Data integration and preprocessing. The AquaCrop model outputs the daily-scale water flux file model.Outputs.Flux, which includes the daily irrigation water volume of winter wheat under irrigation restrictions. Integrate it with the weather file data and use the complete time series results as the SVR input data.

[0088] S4.2. Data standard normalization. For the same feature, the values in different samples may vary greatly. Some extremely small or large data will mislead the correct training of the model. In addition, the scattered distribution of data will also affect the training results. Therefore, it is necessary to standardize the values in the feature by standard deviation (Z-zero), that is, convert them into a normal distribution with a mean of 0 and a variance of 1. The conversion formula is as follows:

[0089]

[0090] where x * is the value after standardization,

[0091] μ represents the mean of the original data,

[0092] σ represents the standard deviation of the original data.

[0093] S4.3. Divide the training set and the test set, and conduct a sensitivity analysis of the SVR input variables. The data set from 1995 to 2015 is used as the training set, and the data set from 2015 to 2018 is used as the test set. Taking NetIrrSMT50% as an example, evaluate the prediction effect of the SVR model for the daily irrigation water requirement through different variable combinations, and determine the optimal feature combination of the SVR input variables. Table 2 shows the SVR prediction accuracy under different input feature combinations.

[0094] Table 2 SVR prediction accuracy under different input feature combinations

[0095]

[0096] In the table, SD, Tmax, Tmin, P, ET0, and Irrday represent sunshine hours, maximum temperature, minimum temperature, precipitation, reference evapotranspiration, and daily irrigation water volume respectively.

[0097] It can be seen from Table 2 that SD, Tmax, Tmin, and ET0 have a greater impact on the results, and P has the smallest impact on the prediction results. Finally, the SVR input variable combination is determined as shown in Table 3:

[0098] Table 3 SVR input variable combination table

[0099] model combination Scenario 1 <![CDATA[SD+Tmax+Tmin+ET0+P+Irrday15%]]> Scenario 2 <![CDATA[SD+Tmax+Tmin+ET0+P+Irrday30%]]> Scenario 3 <![CDATA[SD+Tmax+Tmin+ET0+P+Irrday50%]]>

[0100] S4.4. Use SVR to predict the daily irrigation water requirement of winter wheat under three irrigation scenarios.

[0101] Select RBF as the kernel function, and use the particle swarm optimization algorithm PSO to optimize the RBF parameters until the best SVR training function is found.

[0102] Figure 4 The process of the PSO algorithm is as follows, and the specific steps are as follows:

[0103] S4.4.1. Initialization processing. Set the maximum number of iterations, the number of independent variables of the objective function, the maximum speed of the particles, and the position information as the entire search space. Randomly initialize the speed and position in the speed interval and search space, set the particle swarm size to M, and randomly initialize a flying speed for each particle. In this embodiment, the number of iteration generations is set to 20, the population size is set to 20, and the number of variables is set to 3.

[0104] S4.4.2. Define the fitness function. The individual extreme value is the optimal solution found by each particle, and a global value is found from these optimal solutions as the global optimal solution for this time. Compare with the historical global optimal solution and update. S4.4.3. Update the speed and position, and the formulas are as follows:

[0105] V id = ωV id + C1random(0,1)(P id - X id ) + C2random(0,1)(P gd - X id )

[0106] X id = X id + V id

[0107] Among them, ω is the inertia factor. By adjusting the size of ω, the global search performance and local search performance can be adjusted;

[0108] C1 and C2 are acceleration constants. The former is the individual learning factor of each particle, and the latter is the social learning factor of each particle;

[0109] P id represents the d-th dimension of the individual extreme value of the i-th variable, and P gd represents the d-th dimension of the global optimal solution.

[0110] S4.4.4. Termination condition:

[0111] 1. The set number of iterations is reached;

[0112] 2. The difference between the algebras satisfies the minimum bound.

[0113] The evaluation metrics for the model prediction results are selected as R 2 and RMSE. Select the model with the minimum RMSE and the maximum R 2 as the trained SVR prediction model.

[0114] The parameter combinations of the SVR kernel function are determined by the particle swarm optimization algorithm as shown in Table 4:

[0115] Table 4 Parameter combination table of the SVR kernel function

[0116]

[0117] Figure 5 This is the prediction result graph of this embodiment.

[0118] The results show that the prediction performance of the daily irrigation water requirement of winter wheat under the three irrigation scenarios is relatively good, and the prediction accuracy is higher when the irrigation water volume is relatively sufficient. In summary, the daily irrigation water requirement of winter wheat obtained by the prediction method provided by this solution is close to the actual value, and the calibrated model can accurately predict the daily irrigation water requirement of the next-year crops, thus ensuring the accuracy of the prediction of the total water requirement of the crops in the next growth period.

[0119] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0120] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting crop irrigation water requirement based on the AquaCrop model and SVR, characterized in that, Including the following steps: S1. Obtain the meteorological station data, soil data, crop parameter data, and field management data of the planting area. The meteorological station data includes daily precipitation data, temperature data, relative humidity, wind speed, and sunshine hours. Calculate the reference crop evapotranspiration ET0 according to the Penman-Monteith formula using the meteorological station data; S2. Conduct local debugging of the AquaCrop model through the trial-and-error method; calibrate and validate the AquaCrop model under non-water stress conditions to ensure the accuracy of model simulation; S3. Use the AquaCrop model to select appropriate irrigation strategies and conduct simulations of different irrigation regimes. There are 6 built-in irrigation strategy formulation methods in the AquaCrop model, namely: IrrMethod = 0: Rainfed irrigation; IrrMethod = 1: Trigger irrigation when the soil water content in the root zone is lower than the specified threshold; IrrMethod = 2: Trigger irrigation once every N days; IrrMethod = 3: Predefined irrigation plan; IrrMethod = 4: Net irrigation, fill all soil layer compartments every day to keep the soil moisture at the preset level; IrrMethod = 5: Irrigate a fixed depth every day; Then select the optimal irrigation strategy formulation method according to the yield and the average annual irrigation volume, conduct continuous multi-year growth simulations of the crop, and obtain the daily irrigation water volume data; S4. Based on the simulation results of the AquaCrop model, combine with SVR to predict the daily irrigation water requirement of the crop in the next growth period; S4.

1. Combine the meteorological factors, specifically the maximum temperature Tmax, °C, minimum temperature Tmin, °C, precipitation P, mm, reference evapotranspiration ET0, mm / day, and sunshine hours SD, h / day with the daily irrigation water volume data output by the AquaCrop model in a time series as the input variables of SVR; S4.

2. Standardize the eigenvalue by standard deviation Z-zero; S4.

3. Divide the input variables of SVR into a training set and a test set, and determine the best combination of the maximum temperature Tmax, °C, minimum temperature Tmin, °C, precipitation P, mm, reference evapotranspiration ET0, mm / day, sunshine hours SD, h / day, and Irrday, mm required for predicting the crop irrigation water requirement through sensitivity analysis; S4.

4. Use the best combination as the input to train SVR, so as to predict the daily irrigation water requirement of the crop; select RBF as the kernel function, and use the particle swarm optimization algorithm PSO to optimize the RBF parameters until the best SVR training function is found. The specific steps are as follows: S4.4.

1. Initialization processing: Set the maximum number of iterations, the number of independent variables of the objective function, the maximum speed of the particles, and the position information as the entire search space; randomly initialize the speed and position in the speed interval and search space, set the particle swarm size as M, and randomly initialize a flying speed for each particle; S4.4.

2. Define the fitness function. The individual extreme value is the optimal solution found by each particle. From these optimal solutions, find a global value as the global optimal solution for this time. Compare it with the historical global optimum and update it. S4.4.

3. Update the velocity and position. The formulas are as follows: V id = ωV id + C1random(0,1)(P id - X id ) + C2random(0,1)(P gd - X id ) X id = X id + V id where ω is the inertia factor, C1 and C2 are acceleration constants, the former is the individual learning factor of each particle, and the latter is the social learning factor of each particle; P id represents the d-th dimension of the individual extreme value of the i-th variable, P gd represents the d-th dimension of the global optimal solution; S4.4.

4. Termination conditions:

1. Reach the set number of iterations; 2. The difference between generations satisfies the minimum bound.

2. The crop irrigation water requirement prediction method based on the AquaCrop model and SVR according to claim 1, wherein The crop parameter data includes growth period parameter data and crop yield per unit area data.

Images