Winter wheat multi-target irrigation dynamic optimization method based on crop model and artificial intelligence

Abstract

The invention discloses a winter wheat multi-target irrigation dynamic optimization method based on a crop model and artificial intelligence, and belongs to the field of agricultural intelligent irrigation optimization and decision, the method comprises the following steps: preprocessing meteorological data, soil data, crop parameter data and irrigation management data to obtain standardized input data; carrying out coupling processing on the AquaCrop model and the PyFAO56 model based on the standardized input data to obtain final output of a coupling model; based on the final output of the coupling model, optimizing decision variables by adopting an NSGA-2 multi-objective optimization algorithm to obtain a Pareto optimal solution set; and sorting the candidate schemes in the Pareto optimal solution set by combining a fusion strategy of an entropy weight method and a random forest algorithm to obtain a final irrigation strategy. According to the invention, through multi-objective optimization, objective weighting and machine learning intelligent evaluation, a novel precise irrigation scheme considering both yield and water efficiency is provided.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural intelligent irrigation optimization and decision-making, and in particular relates to a multi-objective dynamic optimization method for winter wheat irrigation based on crop models and artificial intelligence. Background Art

[0002] Traditional irrigation strategies are often based on field experiments, combined with optimization algorithms such as linear, nonlinear, dynamic programming, and genetic algorithms. However, field experiments are often time-consuming and labor-intensive, and they fail to account for the impact of climate, soil type, and agronomic practices on crop yields across different years and regions. With the advancement of computer technology, crop growth models and machine learning algorithms are playing an increasingly important role in agricultural production and crop irrigation decision-making.

[0003] The AquaCrop model is a widely used tool for simulating crop water responses (Steduto et al., 2009). However, the current AquaCrop model is relatively inflexible, particularly with regard to setting the lower limit for staged irrigation control, making it difficult to meet the needs of refined deficit irrigation. Furthermore, the model's single irrigation strategy makes it impossible to precisely adjust the percentage of irrigation evapotranspiration (ET), limiting it to 100% full irrigation. While AquaCrop's open-source Python version, AquaCrop-OSPy, facilitates secondary development, it does not substantially address these shortcomings. Existing irrigation decision-making techniques suffer from the following technical issues:

[0004] 1. Traditional irrigation strategies based on field trials are time-consuming and labor-intensive, and fail to fully consider the impact of climate, soil, and agronomic conditions in different years and regions on crop yield and water use efficiency;

[0005] 2. Although the existing AquaCrop model is widely used in crop water response simulation, it lacks flexibility in setting the lower limit of staged irrigation control and regulating the irrigation supply percentage, and can only implement a fixed 100% irrigation strategy;

[0006] 3. Although the PyFAO56 model can implement various irrigation strategies (such as fixed-cycle irrigation, ET-based irrigation, and soil moisture measurement irrigation), it still cannot achieve combined optimization of the two key decision variables: irrigation evapotranspiration percentage and soil moisture control limit.

[0007] 4. Single optimization methods (such as traditional genetic algorithms or dynamic programming) are easily affected by the subjectivity of weight setting in the process of multi-objective trade-offs (such as maximizing yield, improving irrigation water use efficiency, and minimizing irrigation water volume), and the generated Pareto frontier solution set lacks a mechanism for further objective evaluation and screening;

[0008] 5. The existing simulation methods have low computational efficiency when evaluating multiple winter wheat irrigation schemes, and are difficult to meet the requirements of rapid response and real-time adjustment in actual decision-making processes.

[0009] Therefore, the present invention provides a multi-objective dynamic optimization method for winter wheat irrigation based on crop models and artificial intelligence. Summary of the Invention

[0010] To solve the above technical problems, the present invention proposes a multi-objective dynamic optimization method for winter wheat irrigation based on crop models and artificial intelligence to solve the problems existing in the above-mentioned prior art.

[0011] To achieve the above objectives, the present invention provides a multi-objective dynamic optimization method for winter wheat irrigation based on crop models and artificial intelligence, comprising:

[0012] Preprocess meteorological data, soil data, crop parameter data and irrigation management data to obtain standardized input data;

[0013] performing coupling processing on the AquaCrop model and the PyFAO56 model based on the standardized input data to obtain a final output of the coupling model;

[0014] Based on the final output of the coupling model, the NSGA-2 multi-objective optimization algorithm is used to optimize the decision variables to obtain the Pareto optimal solution set;

[0015] The fusion strategy of the entropy weight method and the random forest algorithm is combined to sort the candidate solutions in the Pareto optimal solution set to obtain the final irrigation strategy.

[0016] Optionally, the meteorological data includes: daily maximum temperature, daily minimum temperature, rainfall, radiation, wind speed, and humidity;

[0017] The soil data includes: soil profile information, field water holding capacity, wilting point, saturated water content, soil permeability, and soil layer structure;

[0018] The crop parameter data include: planting date, maturity period, root distribution, canopy growth parameters, biomass conversion coefficient, harvest index, and crop sensitivity to water of winter wheat;

[0019] The irrigation management data includes: initial irrigation plan, irrigation time window of each stage, and irrigation amount.

[0020] Optionally, the process of coupling the AquaCrop model and the PyFAO56 model based on the standardized input data to obtain the final output of the coupled model includes:

[0021] Inputting the meteorological data, soil data, and crop parameter data in the standardized input data into the AquaCrop model to obtain crop growth indicators and water utilization conditions;

[0022] Inputting meteorological data and soil data in the standardized input data into the PyFAO56 model to generate irrigation events;

[0023] The irrigation events are fed back to the AquaCrop model as input to obtain the final output of the coupled model.

[0024] Optionally, the final output of the coupling model includes several evaluation indicators, including: final yield, total irrigation volume, irrigation water utilization efficiency, and irrigation times.

[0025] Optionally, the decision variables include: soil moisture lower limit and irrigation evapotranspiration recharge percentage.

[0026] Optionally, the process of combining the fusion strategy of the entropy weight method and the random forest algorithm to sort the candidate solutions in the Pareto optimal solution set to obtain the final irrigation strategy includes:

[0027] The entropy weight method is used to calculate the weight of each evaluation index of the candidate solutions in the Pareto optimal solution set to obtain the index weight based on information entropy;

[0028] Determine the contribution of decision variables to each evaluation index based on the random forest algorithm, and obtain the index weight based on the model contribution;

[0029] The index weight based on information entropy and the index weight based on model contribution are integrated to obtain a comprehensive weight;

[0030] Based on the comprehensive weight, the candidate solutions in the Pareto optimal solution set are comprehensively evaluated and ranked to obtain a final irrigation strategy.

[0031] Optionally, the calculation process of the comprehensive weight includes:

[0032] Normalizing the information entropy-based indicator weights;

[0033] Normalizing the weights of the indicators based on the model contribution;

[0034] According to the preset fusion factor, the normalized information entropy-based indicator weight and the normalized model contribution-based indicator weight are weighted averaged to obtain the comprehensive weight.

[0035] Compared with the prior art, the present invention has the following advantages and technical effects:

[0036] The present invention's multi-objective dynamic optimization method for winter wheat irrigation, based on crop models and artificial intelligence, achieves refined management and scientific decision-making for irrigation strategies by coupling the AquaCrop model and the PyFAO56 model, combining the NSGA-2 multi-objective optimization algorithm with a fusion strategy of the entropy weight method and the random forest algorithm. This method dynamically adjusts irrigation strategies for different stages of winter wheat growth, ensuring an appropriate water supply during critical growth periods while avoiding unnecessary irrigation and significantly improving irrigation water use efficiency. Through the multi-objective optimization algorithm, this method achieves an optimal balance between maximizing yield, minimizing irrigation water volume, and maximizing irrigation water use efficiency. Furthermore, the combination of the entropy weight method and the random forest algorithm further enhances the objectivity and accuracy of scenario evaluation, providing a scientific basis for irrigation decision-making. This technology conserves water resources and reduces irrigation costs while ensuring high and stable winter wheat yields. It is of great significance to the sustainable development of agriculture and is particularly suitable for areas with water scarcity. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0038] Figure 1 The Pareto front of the NSGA multi-objective optimization in the embodiment of the present invention;

[0039] Figure 2 The random forest algorithm of the embodiment of the present invention analyzes the contribution of eight decision variables to the decision index;

[0040] Figure 3 This is a dependency graph of the random forest analysis of the irrigation threshold during the seedling emergence period according to an embodiment of the present invention;

[0041] Figure 4 This is a dependency graph of the irrigation threshold value during the slow growth period of the canopy analyzed by random forest analysis according to an embodiment of the present invention;

[0042] Figure 5 This is a dependency graph of the maximum coverage and the irrigation threshold for the beginning of canopy senescence analyzed by random forests according to an embodiment of the present invention;

[0043] Figure 6 This is a dependency graph of the irrigation threshold for the canopy senescence-maturity period analyzed by random forest analysis according to an embodiment of the present invention;

[0044] Figure 7 This is a dependency graph of the percentage of evapotranspiration during the seedling stage of irrigation analyzed by random forests according to an embodiment of the present invention;

[0045] Figure 8 This is a dependency graph of the percentage of irrigation evapotranspiration during the slow growth period of the canopy according to the random forest analysis of an embodiment of the present invention;

[0046] Figure 9 This is a dependency graph of the maximum coverage and the percentage of irrigation evapotranspiration during the canopy senescence period analyzed by random forests according to an embodiment of the present invention;

[0047] Figure 10 This is a dependency graph of the percentage of irrigation evapotranspiration during the canopy senescence-maturity period analyzed by random forest analysis according to an embodiment of the present invention;

[0048] Figure 11 A heat map of the correlation between eight decision variables analyzed by random forest in an embodiment of the present invention;

[0049] Figure 12 This is a comprehensive technical flow chart of an embodiment of the present invention based on AquaCrop, pyfao56, ​​NSGA-II, random forest and entropy weight method;

[0050] Figure 13 This is a comparison diagram of the effects of the embodiments of the present invention. DETAILED DESCRIPTION

[0051] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0052] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0053] Example 1

[0054] The task of the present invention is to optimize the two key decision variables, the soil moisture lower limit and irrigation evapotranspiration recharge percentage, at each growth stage by combining the AquaCrop model with the pyfao56 model, using the NSGA-2 multi-objective optimization algorithm. At the same time, a random forest algorithm is introduced to evaluate the contribution of each indicator and the entropy weight method is used to objectively weight the scheme, forming a full-process intelligent optimization decision-making method that takes into account yield, irrigation water use efficiency, and irrigation water quantity. This method not only achieves precise control of staged deficit irrigation strategies, but also improves the efficiency and accuracy of irrigation decisions through intelligent, multi-indicator comprehensive evaluation, thereby providing reliable technical support for the sustainable production of winter wheat in water-scarce areas such as the North China Plain.

[0055] This embodiment provides a multi-objective dynamic optimization method for winter wheat irrigation based on a crop model and artificial intelligence, including the following steps:

[0056] 1. Data input. The data required for this technical solution mainly include the following categories: (1) Meteorological data: including daily temperature (maximum and minimum temperature), rainfall, radiation, wind speed, humidity, etc. These data are usually monitored and recorded by meteorological stations and stored in text files or tables. This data is the basis for AquaCrop and PyFAO56 models to simulate crop growth and soil water balance. (2) Soil data: including soil profile information (thickness of each soil layer), field water holding capacity (FC), wilting point (WP), saturated water content (S), soil permeability (Ksat), soil layer structure, etc. These data are obtained through field measurements or existing soil databases, used to construct soil models, and are jointly called by PyFAO56 and AquaCrop. (3) Crop parameter data: including winter wheat planting date, maturity period, root distribution, canopy growth parameters (such as canopy expansion, key days in the decline stage, maximum canopy cover, etc.), biomass conversion coefficient, harvest index (HI) and crop sensitivity parameters to water, etc. These parameters play a decisive role in the AquaCrop model and are usually calibrated using field test data. (4) Irrigation management data: This includes the initial irrigation plan (such as irrigation event records for full or partial irrigation), irrigation time windows for each stage, irrigation volume, etc., providing an initial scheduling basis for subsequent models when simulating irrigation regulation.

[0057] 2. Data preprocessing: Preprocess meteorological data, soil data, crop parameter data, and irrigation management data to obtain standardized input data. All input data are uniformly formatted and preprocessed before operation. The steps include:

[0058] Meteorological data is formatted and missing values ​​are processed using dedicated tool functions (such as prepare_weather);

[0059] The soil and crop parameter files are read through the get_filepath and corresponding loadfile methods and saved as objects used internally in the model;

[0060] Data preprocessing ensures that the data called by different models (AquaCrop and PyFAO56) are consistent, so that the two models can run in the same environment and achieve coupled simulation.

[0061] 3. Model calling and coupling process: Based on the standardized input data, the AquaCrop model and the PyFAO56 model are coupled to obtain the final output of the coupling model.

[0062] Furthermore, the process of coupling the AquaCrop model and the PyFAO56 model based on the standardized input data to obtain the final output of the coupling model includes: inputting the meteorological data, soil data, and crop parameter data in the standardized input data into the AquaCrop model to obtain crop growth indicators and water use; inputting the meteorological data and soil data in the standardized input data into the PyFAO56 model to generate irrigation events; and feeding the irrigation events as input back to the AquaCrop model to obtain the final output of the coupling model.

[0063] The AquaCrop model is used to simulate the growth process of winter wheat under specific meteorological, soil and irrigation conditions, including canopy cover, evapotranspiration, yield generation and biomass accumulation.

[0064] Input preprocessed meteorological data, soil parameters, and crop parameters to construct the AquaCrop model object. Set the crop planting date, maturity, and parameters for each growth stage (such as development index, canopy growth, and root expansion). Load the initial water content (InitialWaterContent). Run the model over a simulation timeframe (e.g., from October 25, 2023, to June 15, 2024), and output crop growth metrics (such as FreshYield) and water use (such as CropTransspiration).

[0065] The PyFAO56 model is primarily used to simulate soil water balance and control irrigation scheduling. This process uses the same meteorological and soil data as AquaCrop to establish the PyFAO56 soil water model. The model calculates reference evapotranspiration (ET0) based on daily meteorological conditions and combines soil parameters to calculate soil moisture changes.

[0066] The model sets an irrigation trigger condition: irrigation begins when the root zone water content falls below a preset threshold (decision variable 1). The amount of water supplied during irrigation is supplemented by a percentage of the crop's actual evapotranspiration requirements at that stage (decision variable 2), generating a specific irrigation amount and irrigation time series for each irrigation event.

[0067] Coupling and data interaction between the two models: The purpose of the coupling is to ensure that soil moisture changes (calculated by PyFAO56) can be dynamically fed back to AquaCrop, so that the crop growth simulation can be updated according to actual irrigation events, thereby truly reflecting the changes in yield and growth status caused by changes in irrigation strategies.

[0068] The coupling process involves automatically generating irrigation events (including irrigation dates and water volumes) based on pre-set irrigation decision variables during the PyFAO56 run. These irrigation schedules are then passed as input parameters to the AquaCrop model, which uses these irrigation events to update soil moisture during the growth simulation. The model outputs key metrics, including final yield, total irrigation volume, water use efficiency (IWUE), and irrigation frequency, for subsequent optimization evaluation.

[0069] 4. NSGA-2 multi-objective optimization operation process, based on the final output of the coupling model, the NSGA-2 multi-objective optimization algorithm is used to optimize the decision variables to obtain the Pareto optimal solution set.

[0070] The decision variables used in the present invention include:

[0071] Lower limit of soil moisture content control: the percentage value of soil moisture content when irrigation is triggered at each growth stage.

[0072] Irrigation evapotranspiration recharge percentage: the percentage of actual water recharge after irrigation is triggered relative to the reference evapotranspiration of the crop at that stage.

[0073] These two types of decision variables constitute an 8-dimensional parameter space (for the four irrigation stages), and the ranges are set within reasonable physical intervals (setting the moisture content between 0.25 and 0.55 and the water replenishment percentage between 0.5 and 1).

[0074] (1) Population initialization and evaluation

[0075] Population generation: Use FloatRandomSampling to randomly generate an initial population (e.g. 50 individuals) within the defined parameter space.

[0076] Objective function calculation: For each individual, the above-mentioned coupling model is called for simulation to obtain three main target values: total irrigation water volume (the smaller the better); crop yield (the larger the better, and when converted into an objective function, a negative value is usually taken to achieve minimization); and irrigation water use efficiency (IWUE, the larger the better, also converted into a minimization problem).

[0077] During the simulation process, the parameters of each individual are passed to PyFAO56 to generate irrigation events, and then AquaCrop simulates crop growth and outputs various indicators.

[0078] (2) Non-dominated sorting and genetic operations

[0079] Non-dominated sorting: Sort all individuals according to the objective function value and select a set of non-dominated solutions (Pareto frontier), that is, no other individuals are better than this solution in all objectives.

[0080] Crowding distance calculation: To maintain the diversity of the solution set, the crowding distance of each individual in the target space is calculated.

[0081] Selection, crossover and mutation: Based on non-dominated sorting and crowding distance, some excellent individuals are selected for genetic operations.

[0082] Crossover: Use SimulatedBinaryCrossover to mix the decision variables of two parent individuals to generate offspring.

[0083] Mutation: Use PolynomialMutation to slightly perturb the decision variables of offspring individuals to ensure comprehensive exploration of the search space.

[0084] Iterative update: After a certain number of generations (e.g., n_gen generations), the NSGA-2 algorithm converges and outputs a Pareto optimal solution set that covers the trade-offs between different objectives. Each candidate solution consists of 8 decision variables:

[0085] Stage 1 (seedling emergence): Decision variable 1 is the lower limit of soil moisture content (critical value for starting irrigation); decision variable 2 is the percentage of actual irrigation amount to reference evapotranspiration during irrigation.

[0086] Phase 2 (slow canopy growth period): Decision variable 3 is the lower limit of soil moisture content; decision variable 4 is the percentage of irrigation evapotranspiration.

[0087] Stage 3 (maximum coverage - canopy aging): Decision variable 5 is the lower limit of soil moisture content; decision variable 6 is the percentage of irrigation evapotranspiration.

[0088] Stage 4 (canopy begins to age and mature): decision variable 7 is the lower limit of soil moisture content; decision variable 8 is the percentage of irrigation evapotranspiration.

[0089] For each candidate plan, the output indicators obtained through simulation mainly include: yield (Yield), total irrigation amount (Irrigation Amount), irrigation water use efficiency (WUE), and irrigation times.

[0090] These outputs constitute the evaluation matrix, which is the basis for subsequent entropy weight calculation and random forest sensitivity analysis.

[0091] Fifth, the unified computational framework of the entropy weight method and random forest algorithm is used. The combined strategy of the entropy weight method and random forest algorithm is used to rank candidate solutions in the Pareto optimal solution set to obtain the final irrigation strategy. This system integrates the random forest method and entropy weight method into a unified winter wheat irrigation decision support system, rather than using each method independently.

[0092] Furthermore, the process of combining the fusion strategy of the entropy weight method and the random forest algorithm to sort the candidate solutions in the Pareto optimal solution set to obtain the final irrigation strategy includes: using the entropy weight method to calculate the weights of each evaluation indicator of the candidate solutions in the Pareto optimal solution set to obtain the indicator weights based on information entropy; determining the contribution of decision variables to each evaluation indicator based on the random forest algorithm to obtain the indicator weights based on model contribution; fusing the indicator weights based on information entropy with the indicator weights based on model contribution to obtain the comprehensive weights; and comprehensively evaluating and sorting the candidate solutions in the Pareto optimal solution set based on the comprehensive weights to obtain the final irrigation strategy.

[0093] (1) Data preparation and preprocessing: The historical irrigation scheme data of the Pareto optimal solution set were collected, including the parameter settings of different schemes (irrigation threshold at the seedling emergence stage, irrigation threshold at the canopy slow growth stage, irrigation threshold at the maximum coverage-canopy senescence stage, irrigation threshold at the canopy senescence-maturity stage, percentage of irrigation evapotranspiration at the seedling emergence stage, percentage of irrigation evapotranspiration at the canopy slow growth stage, percentage of irrigation evapotranspiration at the maximum coverage-canopy senescence stage, percentage of irrigation evapotranspiration at the canopy senescence-maturity stage) and the corresponding result indicators (yield, total irrigation amount, irrigation water use efficiency, and number of irrigation times).

[0094] (2) Entropy weight method calculates the weights of multi-objective evaluation indicators. The entropy weight method uses the distribution differences of each indicator data in the candidate solutions to determine the weight of each indicator. The steps are as follows:

[0095] Data normalization: For each evaluation indicator (such as yield, irrigation amount, WUE, and irrigation times), normalization is first performed to eliminate the dimension effect.

[0096] Calculate information entropy: Based on the normalized data, calculate the information entropy of each indicator. If a certain indicator is more dispersed across different solutions, its information entropy is lower, indicating that the indicator contributes more to distinguishing the solutions.

[0097] Determine entropy weight: Calculate entropy weight based on the entropy value of each indicator. The result is a set of weight vectors:

[0098] Wentropy=[Wyield,Wirrigation,WWUE,Wfrequency]

[0099] This vector reflects the objective importance of each evaluation indicator from the perspective of the candidate solution. In the code, this part has been implemented by the function get_entropy_weight(data).

[0100] (3) Random forest extraction of the contribution of decision variables to the output. Using the random forest model, the influence of each decision variable (irrigation threshold at the seedling stage, irrigation threshold at the slow growth stage of the canopy, irrigation threshold at the maximum coverage-canopy senescence stage, irrigation threshold at the canopy senescence-maturity stage, percentage of irrigation evapotranspiration at the seedling stage, percentage of irrigation evapotranspiration at the slow growth stage of the canopy, percentage of irrigation evapotranspiration at the maximum coverage-canopy senescence stage, percentage of irrigation evapotranspiration at the canopy senescence-maturity stage) on the simulation output can be evaluated. The basic steps are as follows:

[0101] First, the training data was constructed. Input: candidate solutions (eight decision variables) obtained by NSGA-2. Output: corresponding simulation results (e.g., yield, irrigation amount, WUE, and irrigation times) were used to construct a multi-output regression model.

[0102] Secondly, the random forest model is trained, mainly through cross-validation and parameter tuning (such as adjusting n_estimators, max_features, etc.), to train the random forest regression model to ensure that the model has a good fitting effect.

[0103] Then extract the feature importance. After training is completed, use model.feature_importances_ to extract the importance of each decision variable. This set of importance values ​​can be recorded as:

[0104] RFimp=[rf1,rf2,…,rf8]

[0105] Indicates the extent to which factors such as "lower limit of soil moisture" or "percentage of irrigation evapotranspiration" affect the output at each stage.

[0106] Finally, we map them to output indicators. Since the output of the candidate solution is four evaluation indicators, and decision variables usually appear in pairs, we can further analyze them:

[0107] (4) Detailed steps of the fusion strategy: ① Normalization of entropy weight vector: After obtaining the weight, normalize it to ensure that the sum of each item is 1. ② Normalization of random forest contribution vector: Normalize the contribution of each evaluation index RFoutput obtained by the random forest through sensitivity mapping to obtain:

[0108] wRF=[wRF, Yield, wRF, Irrigation, wRF, WUE, wRF, Frequency].

[0109] ③ Weight fusion: The two weights are fused by weighted average, and the fused indicator weight is calculated: wfsion = α × wentrop + (1-α) × wRF.

[0110] Here, α is the fusion factor, which can be determined based on practical experience or through cross-validation (for example, when α = 0.5, the two parts are equally weighted). The resulting wfusion reflects both the solution's ability to distinguish candidate data (entropy weight) and the sensitivity of each decision variable to the final output (random forest contribution).

[0111] ④ Comprehensive score calculation: Use the fusion weight wfusion to perform weighted summation on the output indicators of each candidate solution to obtain the final comprehensive score. The calculation formula can be written as:

[0112] Where: S i is the comprehensive score of the i-th candidate solution; is the jth normalized output indicator of candidate solution i (e.g., normalized yield, irrigation amount, WUE, and irrigation times).

[0113] In the code, this step can modify the original get_score function to use the fused weights instead of the simple entropy weight method weights.

[0114] ⑤ Alternative Ranking and Decision Output: All candidate options are ranked according to their comprehensive scores (Si). The option with the highest score is the recommended option. The top candidate options are also output for decision-making. You can further demonstrate the trade-offs between options through visualization tools (such as Pareto front diagrams and box plots).

[0115] This fusion framework ensures information sharing and mutual complementation between the two algorithms: random forest provides prediction data and objective importance, and entropy weight method provides multi-indicator trade-offs and objective weighting. Their combination makes the decision-making process automated and scientific.

[0116] Based on the candidate irrigation schemes generated by AquaCrop and PyFAO56 simulation, a series of Pareto optimal solution sets were formed, such as Figure 1 As shown in the figure, each scheme has different performances in terms of performance indicators such as yield, irrigation amount, irrigation water use efficiency and irrigation times, reflecting an obvious trade-off relationship.

[0117] Table 1

[0118]

[0119] On the one hand, the highest-yield scenarios often involve more irrigation inputs (higher irrigation volumes and frequency), leading to lower water use efficiency. On the other hand, extreme water-saving scenarios significantly reduce irrigation volumes and frequency, but also reduce crop yields. Most compromise scenarios fall somewhere in between, reducing water use while moderately sacrificing yield to improve IWUE.

[0120] The entropy weighting method assigned the following weights to each indicator: irrigation frequency 0.4722, irrigation volume 0.3009, yield 0.1207, and water use efficiency 0.1060. This indicates that the comprehensive evaluation places significant emphasis on reduced irrigation frequency and water use (the sum of these weights is close to 0.77), while yield and water use efficiency are given relatively low weights. This weighting distribution directly influences the final ranking: top-ranked solutions exhibit significant advantages in water conservation. First, a common characteristic of highly ranked solutions is the minimal number of irrigations. Many top-ranked solutions reduce irrigation frequency to just one or two times, far lower than other solutions. This results in them receiving the highest evaluation scores for irrigation frequency. Furthermore, reduced irrigation frequency often translates to reduced total water use—a fact that is indeed true: leading solutions typically achieve significant reductions in total irrigation volume, with some using tens of percentage points less than high-yield, high-water-consumption solutions, yet still score highly on irrigation volume. Despite these significant water savings, these winning solutions maintain high yields. The top-ranked strategies generally achieve yields exceeding 80% of what they would achieve under fully irrigated conditions, with some even approaching 90%. Since the yield metric only carries a weight of 0.12, this small difference in yield has a relatively limited impact on the overall score. Furthermore, water use efficiency (WUE) also significantly improves due to water conservation: using less irrigation water achieves near-maximum yield, significantly increasing output per unit of irrigation water. This enables the top-ranked strategies to achieve strong scores in the WUE metric (although WUE carries a relatively low weight, its high WUE further solidifies their overall advantage). Overall, the high scores of the top-ranked strategies stem from "trading a relatively small yield penalty for significant water savings." Since the weight of irrigation frequency is approximately four times that of yield, the score improvement from reducing one irrigation session far outweighs the benefits of a small yield increase. Therefore, these strategies consistently minimize irrigation water volume and frequency within acceptable yield losses, achieving an optimal balance between output and input. This is the primary reason for their superior performance in the overall ranking. For those plans with slightly higher yields but frequent irrigation, the loss is that the excessive number of irrigations and water volume result in serious score deductions. The increased yield cannot make up for the high irrigation costs, and the final comprehensive score is lower than that of the water-saving plan.

[0121] Table 2

[0122]

[0123]

[0124] After a comprehensive evaluation of these multiple indicators using the entropy weight method, the final ranking of the various options was determined. The results showed that the top-ranked options were mostly able to stabilize yields at around 9 t·hm-2, and the change in irrigation water use efficiency also remained high. This demonstrates that a reasonable irrigation strategy (appropriate irrigation quota and appropriate irrigation frequency) can conserve water resources while maintaining stable and high yields. Many of the top-ranked options involved 8-10 irrigations, resulting in higher costs, but not excessively to unacceptable levels. This demonstrates that on-demand irrigation and multiple irrigation cycles have clear advantages in increasing yields and water use efficiency. For areas with limited water resources or funding, a moderate to low irrigation frequency can be chosen, but precise scheduling is crucial to avoid significant yield reductions.

[0125] From the analysis of the contribution of eight variables in four irrigation periods to four evaluation indicators derived from the irrigation scheme derived from the Pareto frontier, it can be seen that the soil water threshold in the key growth stage (peak period of soil water demand) is the most important variable. This threshold determines the extent to which irrigation begins when the soil water content decreases. In the stage when crops have high water demand (usually corresponding to the heading and flowering to filling period), the setting of the threshold has a direct impact on yield and the number of irrigations: if the threshold is set too high ("conservative irrigation", irrigation when the soil is slightly dry) will increase the frequency of irrigation and water consumption to ensure that the crops are not affected by drought, the yield will be guaranteed but the IWUE will decrease; if the threshold is set too low ("loose irrigation", allowing the soil to be drier before irrigation), the number of irrigations and water consumption can be reduced, and the IWUE will be increased, but if it is too low, the water stress of crops in this stage will be aggravated, and the yield will be significantly reduced. In the random forest model, this threshold contributes the most to the fluctuation of the comprehensive score, reflecting its key position. Figure 2 shown.

[0126] The single irrigation amount (irrigation intensity) at the same critical stage follows closely behind and is also a highly important factor. The amount of water applied each time during the critical period irrigation also has a profound impact: if the irrigation intensity is too low, it will not be able to fully alleviate soil drought, and the crops may need to be irrigated more frequently (increase the number of times) or continue to be in mild water stress (affecting yield); while if the irrigation intensity is too high, the single water application amount exceeds the effective capacity of the soil, which will lead to water leakage and waste. Although the irrigation frequency is reduced, the total water consumption increases and the WUE decreases. Random forest analysis shows that within the range considered by the model, the importance of this variable is second only to the critical stage threshold, indicating that the appropriate irrigation dosage is critical to balancing yield and water conservation.

[0127] Furthermore, soil moisture thresholds in the late growth phase (maturity stage) also demonstrate significant importance. Many of the top-ranked options utilize lower soil moisture thresholds in the late crop period (allowing irrigation only after the soil becomes drier, or even eliminating irrigation altogether) to reduce irrigation inputs near maturity. Because crops are less sensitive to water shortages in the late stages, this measure significantly saves water with little yield loss. Therefore, the positive impact of late-stage thresholds on overall performance is also reflected in the model.

[0128] like Figure 3-10 As shown, the partial dependence analysis of random forests further reveals the relationship between important variables and performance indicators. For the most important medium-term soil water threshold variable, there is an optimal range of values: the partial dependence plot shows that when the threshold is too high, the overall score decreases due to over-irrigation; when the threshold is too low, the score decreases due to yield loss. This indicates that a moderate threshold value balances yield and water conservation, maximizing overall performance. Similarly, the dependence relationship of the irrigation intensity variable also has an inflection point: a moderate irrigation amount is conducive to improving the overall score, while too low or too high an intensity will reduce the score (either due to increased frequency or reduced marginal benefits of water). These findings provide a basis for irrigation decision-making: managers should focus on these variables that are highly sensitive to performance and optimize the threshold and irrigation amount settings within the above reasonable range to achieve the optimal balance between yield and water efficiency.

[0129] like Figure 11 As shown in the figure, the correlation between the eight variables shows that there is a strong negative correlation between the slow growth period and the maximum coverage period, which indicates that the water control strategies adopted in these stages are contradictory. When optimizing the irrigation strategy for these stages, it is necessary to balance the water requirements of these two stages.

[0130] The correlation between the emergence stage and other stages is weak, which means that water management during the emergence stage has little impact on subsequent stages, especially the fundamental role of water control in the early stage for later growth.

[0131] There was a positive correlation between the maximum coverage period and the senescence period, indicating that these two stages may have certain commonalities in water demand and the irrigation strategies may be relatively consistent in these two stages.

[0132] Correlation analysis between phases provides clues for optimizing irrigation strategies. For example, the strong negative correlation between the slow growth phase and the maximum coverage phase suggests that different irrigation strategies may be needed during these two phases, as shown in Tables 3 and 4.

[0133] Table 3

[0134]

[0135] Table 4

[0136]

[0137]

[0138] Across all decision-making schemes, irrigation control and evapotranspiration were relatively consistent during the emergence phase. Irrigation rates were generally higher during the maximum cover phase, while evapotranspiration decreased during the senescence phase. Irrigation evapotranspiration fluctuated significantly across the different schemes during the slow growth phase, likely due to differences in soil moisture control. This was particularly true during the slow growth and senescence phases, suggesting that the schemes may have compensated for soil moisture deficits by increasing evapotranspiration.

[0139] Secondly, in order to improve the efficiency of water resource utilization, we can consider the balance between the slow growth period and the aging period, avoid excessive irrigation and lead to water waste, and maintain appropriate soil moisture and evaporation.

[0140] Based on the above detailed analysis, the unified irrigation decision calculation framework of the present invention has the following technical effects:

[0141] Multi-stage water management: By setting the irrigation control lower limit and irrigation evapotranspiration percentage in stages, refined irrigation management is carried out for different key stages of crop growth to ensure sufficient water during the critical period of yield formation (especially from maximum canopy cover to senescence), while avoiding over-irrigation in the early and mature stages.

[0142] Efficient water use: Image analysis shows that adjusting the percentage of irrigation evapotranspiration in the later stages can significantly improve IWUE. This framework can maximize water use efficiency while ensuring yield.

[0143] Reduce irrigation frequency and save costs: By reasonably setting irrigation thresholds, the framework effectively controls irrigation frequency, avoiding over-frequent irrigation and reducing energy consumption and operating costs.

[0144] Fusion of objective and model prediction results: The entropy weight method assigns objective weights to each output indicator (yield, irrigation amount, IWUE, and irrigation times), while random forest analysis reveals the actual contribution of decision variables at each stage to these indicators. The fusion of the two constructs a scientific and reasonable comprehensive scoring system, providing solid data support for decision-making.

[0145] like Figure 12As shown in the figure, the main goal of this technical solution is to improve winter wheat yield and water use efficiency by simulating and optimizing irrigation strategies. Specifically, this technology combines multiple models (AquaCrop, FAO56) and algorithms (NSGA-2, multi-objective optimization, entropy weight method, random forest) to simulate and optimize winter wheat irrigation strategies by comprehensively considering factors such as irrigation strategy, soil characteristics, and crop growth. Then, through the combination of multi-objective optimization and intelligent algorithms, it can find the optimal irrigation plan under various decision variables and constraints to improve crop yield and water use efficiency. At the same time, the entropy weight method and random forest are combined to optimize and make decisions on irrigation plans.

[0146] like Figure 13 As shown, the integrated technical solution based on AquaCrop, pyfao56, ​​NSGA-II, random forest, and entropy weighting can effectively reduce water consumption, improve irrigation water efficiency, and ensure high yields with fewer irrigation times. Compared with traditional agricultural solutions and those in the literature, this technology demonstrates significant advantages across key indicators, particularly in water conservation and efficient water use, demonstrating its high application value in the context of water scarcity.

[0147] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A multi-objective dynamic optimization method for winter wheat irrigation based on crop model and artificial intelligence, characterized in that: The following steps are involved: Preprocess meteorological data, soil data, crop parameter data and irrigation management data to obtain standardized input data; performing coupling processing on the AquaCrop model and the PyFAO56 model based on the standardized input data to obtain a final output of the coupling model; Based on the final output of the coupling model, the NSGA-2 multi-objective optimization algorithm is used to optimize the decision variables to obtain the Pareto optimal solution set; The fusion strategy of the entropy weight method and the random forest algorithm is combined to sort the candidate solutions in the Pareto optimal solution set to obtain the final irrigation strategy.

2. The multi-objective dynamic optimization method for winter wheat irrigation based on crop model and artificial intelligence according to claim 1 is characterized in that: The meteorological data include: daily maximum temperature, daily minimum temperature, rainfall, radiation, wind speed, and humidity; The soil data includes: soil profile information, field water holding capacity, wilting point, saturated water content, soil permeability, and soil layer structure; The crop parameter data include: planting date, maturity period, root distribution, canopy growth parameters, biomass conversion coefficient, harvest index, and crop sensitivity to water of winter wheat; The irrigation management data includes: initial irrigation plan, irrigation time window of each stage, and irrigation amount.

3. The multi-objective dynamic optimization method for winter wheat irrigation based on crop model and artificial intelligence according to claim 1, characterized in that: The process of coupling the AquaCrop model and the PyFAO56 model based on the standardized input data to obtain the final output of the coupled model includes: Inputting the meteorological data, soil data, and crop parameter data in the standardized input data into the AquaCrop model to obtain crop growth indicators and water utilization conditions; Inputting meteorological data and soil data in the standardized input data into the PyFAO56 model to generate irrigation events; The irrigation events are fed back to the AquaCrop model as input to obtain the final output of the coupled model.

4. The multi-objective dynamic optimization method for winter wheat irrigation based on crop model and artificial intelligence according to claim 3 is characterized in that: The final output of the coupling model includes several evaluation indicators, including: final yield, total irrigation volume, irrigation water utilization efficiency, and irrigation times.

5. The multi-objective dynamic optimization method for winter wheat irrigation based on crop model and artificial intelligence according to claim 4 is characterized in that: The decision variables include: soil moisture lower limit and irrigation evapotranspiration recharge percentage.

6. The multi-objective dynamic optimization method for winter wheat irrigation based on crop model and artificial intelligence according to claim 5, characterized in that: The process of combining the fusion strategy of the entropy weight method and the random forest algorithm to sort the candidate solutions in the Pareto optimal solution set to obtain the final irrigation strategy includes: The entropy weight method is used to calculate the weight of each evaluation index of the candidate solutions in the Pareto optimal solution set to obtain the index weight based on information entropy; Determine the contribution of decision variables to each evaluation index based on the random forest algorithm, and obtain the index weight based on the model contribution; The index weight based on information entropy and the index weight based on model contribution are integrated to obtain a comprehensive weight; Based on the comprehensive weight, the candidate solutions in the Pareto optimal solution set are comprehensively evaluated and ranked to obtain a final irrigation strategy.

7. The multi-objective dynamic optimization method for winter wheat irrigation based on crop model and artificial intelligence according to claim 6, characterized in that: The calculation process of the comprehensive weight includes: Normalizing the information entropy-based indicator weights; Normalizing the weights of the indicators based on the model contribution; According to the preset fusion factor, the normalized information entropy-based indicator weight and the normalized model contribution-based indicator weight are weighted averaged to obtain the comprehensive weight.

Images