A method, system, and apparatus for predicting wheat nitrogen uptake and yield.

Abstract

This invention provides a method, system, and apparatus for predicting wheat nitrogen uptake and yield. The method includes the following steps: S1, acquiring relevant data, including environmental data, nitrogen management data, and output data; S2, preprocessing the relevant data to obtain standardized data; S3, determining the type of prediction model based on nitrogen application conditions, categorized as a nitrogen uptake prediction model without nitrogen application, a yield prediction model without nitrogen application, a nitrogen uptake prediction model with nitrogen application, and a yield prediction model with nitrogen application; S4, using the standardized data to determine the optimal prediction model for the basic model under the AutoGluon framework under different nitrogen application conditions; S5, predicting nitrogen uptake and yield values ​​under different nitrogen application conditions based on the optimal prediction models obtained from the environmental data and nitrogen management data. This invention achieves accurate prediction under different nitrogen environments by introducing multidimensional variable influencing factors.

Description

Technical Field

[0001] This invention belongs to the field of crop prediction technology, specifically relating to a method, system, and device for predicting wheat nitrogen absorption and yield. Background Technology

[0002] Currently, the world faces the dual pressures of green agricultural development and nitrogen fertilizer pollution control. Accurately estimating crop nitrogen uptake capacity and yield performance under different nitrogen management inputs has become a key issue in achieving precision fertilization management, improving nitrogen fertilizer utilization efficiency, and reducing environmental costs. Nitrogen fertilizer application is particularly important for wheat production, affecting its final yield and quality. Traditional fertilization decisions often rely on experience, making it difficult to balance high yield, high efficiency, and environmental protection goals in a complex and ever-changing environmental and management context. Especially at the regional scale, due to significant differences in ecological zoning, climate type, soil properties, and farming systems, the same management strategy may produce completely different results in different locations, severely limiting the improvement of fertilizer utilization efficiency and the control of environmental pollution.

[0003] Traditional statistical regression methods, based on expert experience, often suffer from overfitting, poor generalization ability, or reliance on excessive domain knowledge when modeling high-dimensional nonlinear interaction factors. Furthermore, traditional methods have significant limitations in model interpretability, variable contribution analysis, and analysis of regional combination differences. In recent years, with the accumulation of agricultural data and the development of machine learning technology, researchers have gradually attempted to introduce data-driven modeling into wheat nitrogen fertilizer management optimization. For example, various supervised learning models are used to predict wheat yield and nitrogen uptake under different management combinations, or nitrogen response curves are constructed based on regression, ensemble learning, and deep networks. However, these methods have the following shortcomings: a lack of objective function design oriented towards actual management optimization, making it difficult to comprehensively evaluate the trade-offs between fertilization schemes and yield, quality, and environment; poor model universality and weak regional adaptability, failing to fully consider the heterogeneity of environment and management combinations; and a lack of visualization and interpretation mechanisms, hindering science communication and farmer acceptance. Therefore, there is an urgent need for a methodological system that integrates machine learning prediction, wheat nitrogen response modeling, and scenario simulation to achieve multi-objective, small-scale, and heterogeneous nitrogen application optimization simulation, providing theoretical and data support for green and efficient wheat production. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, system and device for predicting wheat nitrogen absorption and yield, which can achieve accurate prediction under different nitrogen environments by introducing multidimensional variable influencing factors.

[0005] This invention relates to a method for predicting nitrogen uptake and yield in wheat, comprising the following steps:

[0006] S1. Obtain relevant data, including environmental data, nitrogen management measures data, and output data;

[0007] S2. Preprocess the relevant data to obtain standardized data;

[0008] S3. Determine the type of prediction model based on the nitrogen application situation. The types are divided into nitrogen absorption prediction model without nitrogen application, yield prediction model without nitrogen application, nitrogen absorption prediction model with nitrogen application, and yield prediction model with nitrogen application.

[0009] S4. Using the standardized data, determine the optimal prediction model for the basic model under the AutoGluon framework under different nitrogen application conditions;

[0010] S5. Based on the environmental data and nitrogen management measures data, predict the nitrogen uptake and yield values ​​under different nitrogen application conditions using the optimal prediction model.

[0011] Furthermore, step S2 includes the following steps:

[0012] S201. Convert the categorical variable data in the relevant data into string data according to the preset mapping rules;

[0013] S202. Convert the continuous variable data in the relevant data into data with the same unit format according to the preset normalizer.

[0014] Furthermore, step S4 includes the following steps:

[0015] S401. Perform integrity processing on the standardized data according to the feature type of the standardized data to obtain input data;

[0016] S402. Call the preset model candidate set, use the input data to train the basic model, and obtain the initial nitrogen absorption prediction value and the initial yield prediction value.

[0017] S403. Train at least one secondary model using the initial nitrogen uptake prediction value and the initial yield prediction value;

[0018] S404. The secondary models are fused using weighted averaging or adaptive averaging to obtain a prediction model;

[0019] S405. Use the hyperparameter tuning function of the AutoGluon framework to screen the prediction models and select the best prediction model.

[0020] Furthermore, step S401 includes the following steps:

[0021] S40101. Identify the type of the standardized data, including numerical, categorical, textual, and temporal types;

[0022] S40102. Impute missing values ​​for numerical and categorical data; perform One-Hot Encoding on categorical data; and generate embedding vectors for text data using TF-IDF or a pre-trained model.

[0023] Furthermore, in S4, K-fold cross-validation and multi-layer stacking are used to train the model.

[0024] Furthermore, in S4, the optimal nitrogen absorption prediction model without nitrogen application is CatBoost_r177_BAG_L1_FULL, the optimal yield prediction model without nitrogen application is CatBoost_r177_BAG_L1_FULL, the optimal nitrogen absorption prediction model with nitrogen application is WeightedEnsemble_L3, and the optimal yield prediction model with nitrogen application is WeightedEnsemble_L2_FULL.

[0025] Furthermore, it also includes the following steps:

[0026] S6. Perform interpretability analysis on the prediction model to obtain the ranking of variable importance.

[0027] Furthermore, step S6 also includes the following steps:

[0028] S601. The prediction model is interpreted using TreeExplainer, and the SHAP value matrix is ​​extracted;

[0029] S602. Perform full sample aggregation and variable importance ranking based on the SHAP value matrix to obtain the variable importance ranking results.

[0030] This invention also provides a wheat nitrogen uptake and yield prediction system, comprising:

[0031] The acquisition module is used to acquire relevant data, including environmental data, nitrogen management measures data, and output data.

[0032] The preprocessing module is used to preprocess the relevant data to obtain standardized data;

[0033] The model partitioning module is used to determine the type of prediction model based on the nitrogen application situation. The types are divided into nitrogen absorption prediction model without nitrogen application, yield prediction model without nitrogen application, nitrogen absorption prediction model with nitrogen application, and yield prediction model with nitrogen application.

[0034] The model building module is used to determine the optimal prediction model for different nitrogen application conditions based on the standardized data of the basic model under the AutoGluon framework.

[0035] The prediction module is used to predict nitrogen uptake and yield values ​​under different nitrogen application conditions based on the optimal prediction model of the environmental data and nitrogen management measures data.

[0036] The present invention also provides a wheat nitrogen absorption and yield prediction device, including a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the above method.

[0037] The beneficial effects of this invention are:

[0038] 1. This invention achieves joint modeling and intelligent optimization of nitrogen uptake and yield: For the first time, this invention systematically introduces the AutoGluon automated machine learning framework into wheat nitrogen uptake and yield modeling, integrating multiple model structures and automatic feature engineering techniques. It can simultaneously construct nitrogen uptake and yield prediction models, and improves overall prediction accuracy through model stacking and fusion strategies. This invention solves the problems of low prediction accuracy and difficulty in simultaneously considering multiple objectives in traditional agricultural modeling, demonstrating significant technological innovation value.

[0039] 2. This invention is highly versatile and transferable, adaptable to different regions and management scenarios: Through automatic type identification and preprocessing of input data, this invention possesses strong scenario adaptability and can be flexibly applied to farmland datasets under different ecological regions, farming systems, and fertilization strategies. This invention supports rapid deployment and migration expansion, making it particularly suitable for modeling and analyzing large-scale agricultural experimental data, and has excellent application and promotion prospects.

[0040] 3. This invention facilitates precision agriculture decision-making and has practical application value: This invention not only possesses predictive capabilities but also provides nitrogen fertilizer use efficiency estimation and scenario simulation functions. Users can assess the impact of different fertilization levels on nitrogen absorption and yield through model output, further guiding regional nitrogen fertilizer input decisions to achieve the goals of optimizing resource allocation, improving crop efficiency, and enhancing environmental sustainability. This invention can be directly embedded into agricultural information platforms, providing intelligent support for agricultural policy formulation and precision fertilization services, and possesses significant practical application value. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method in this invention;

[0042] Figure 2The following are the predicted nitrogen uptake results under different models in this invention without nitrogen application (a represents the CatBoost_r177_BAG_L1_FULL model, b represents the LightGBMXT_BAG_L1_FULL model, c represents the LightGBM_BAG_L1_FULL model, d represents the LightGBMXT_BAG_L2_FULL model, e represents the WeightedEnsemble_L2_FULL model, f represents the WeightedEnsemble_L2 model, g represents the LightGBM_r131_BAG_L1 model, h represents the LightGBMXT_BAG_L2 model, and i represents the CatBoost_r177_BAG_L1 model).

[0043] Figure 3 The graph shows the yield prediction results without nitrogen application under different models in this invention (a represents the CatBoost_r177_BAG_L1_FULL model, b represents the LightGBMXT_BAG_L1 model, c represents the NeuralNetTorch_BAG_L2 model, d represents the LightGBMXT_BAG_L1_FULL model, e represents the WeightedEnsemble_L2_FULL model, f represents the NeuralNetTorch_r79_BAG_L2 model, g represents the NeuralNetTorch_BAG_L1_FULL model, h represents the XGBoost_BAG_L1 model, and i represents the LightGBM_r131_BAG_L1_FULL model).

[0044] Figure 4 The figures show the predicted nitrogen uptake under different nitrogen application models in this invention (a represents the LightGBMXT_BAG_L1 model, b represents the LightGBMXT_BAG_L1_FULL model, c represents the WeightedEnsemble_L3 model, d represents the CatBoost_BAG_L2 model, e represents the LightGBMXT_BAG_L2 model, f represents the CatBoost_r177_BAG_L2 model, g represents the LightGBM_r131_BAG_L1 model, h represents the NeuralNetFastAl_BAG_L2 model, and i represents the NeuralNetTorch_BAG_L2 model).

[0045] Figure 5The graph shows the yield prediction results under nitrogen application under different models in this invention (a represents the LightGBMXT_BAG_L1_FULL model, b represents the LightGBMXT_BAG_L1 model, c represents the CatBoost_BAG_L1_FULL model, d represents the WeightedEnsemble_L2_FULL model, e represents the WeightedEnsemble_L2 model, f represents the LightGBM_r131_BAG_L1_FULL model, g represents the LightGBM_r131_BAG_L1 model, h represents the CatBoost_r177_BAG_L1_FULL model, and i represents the CatBoost r177 BAG L1 model).

[0046] Figure 6 This is a distribution diagram of SHAP values, which are important characteristics affecting wheat nitrogen uptake in this invention.

[0047] Figure 7 This is a schematic diagram illustrating the direction and intensity of the influence of key features affecting wheat nitrogen uptake on model predictions in this invention.

[0048] Figure 8 This is a distribution map of SHAP values, which are important characteristics affecting wheat yield in this invention.

[0049] Figure 9 This diagram illustrates the direction and intensity of the influence of key features affecting wheat yield on model predictions in this invention. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0051] like Figures 1 to 9 As shown, a method for predicting wheat nitrogen uptake and yield includes the following steps:

[0052] S1. Data was collected from Web of Science, Google Scholar, and papers related to wheat yield and nitrogen uptake experiments. This data included environmental data, nitrogen management data, and output data. Environmental data included region, climate type, and soil physicochemical properties. Nitrogen management data included fertilization methods, fertilization forms, and whether irrigation / straw return / organic input was used. Output data included nitrogen uptake and yield. A total of 216 papers were consulted, collecting 3473 data points, of which 915 data points were collected without nitrogen application.

[0053] S2. Preprocess the relevant data to obtain standardized data. Specifically, S2 includes the following steps:

[0054] S201. Convert the categorical variable data in the relevant data into string format according to the preset mapping rules. Specifically, convert the categorical variables (including region, climate type variables, main climate zone, precipitation type, temperature type, tillage type, straw return to field, long-term organic fertilizer input, nitrogen application method, nitrogen application form, and irrigation) into string format using the mapping rules. The results are shown in Table 1-2.

[0055] Table 1. Types of categorical variables for continents and climate types

[0056]

[0057] Table 2. Types of Classification Variables for Management Measures

[0058]

[0059] S202. Based on the preset standardizer, transform the continuous variable data in the relevant data into data with the same unit format. Specifically, use the standardizer to standardize the continuous variables (including soil variables such as soil pH, soil organic carbon, soil total nitrogen, soil bulk density, sand content, silt content, clay content, nitrogen application frequency, nitrogen application amount, phosphorus input, and potassium input).

[0060] S3. Determine the type of prediction model based on the nitrogen application situation. The types are divided into nitrogen uptake prediction model without nitrogen application, yield prediction model without nitrogen application, nitrogen uptake prediction model with nitrogen application, and yield prediction model with nitrogen application. The model without nitrogen application is used to estimate the contribution of endogenous nitrogen, while the model with nitrogen application is used to model the actual nitrogen uptake or yield after the effect of nitrogen application.

[0061] S4. Using standardized data, determine the optimal prediction model for the base model under the AutoGluon framework under different nitrogen application conditions. This step also employs K-fold cross-validation and multi-layer stacking to train the model. Specifically, S4 includes the following steps:

[0062] S401, the AutoGluon framework performs integrity processing on the standardized data based on its feature types to obtain the input data. Specifically, S401 includes the following steps:

[0063] S40101. Identify the types of standardized data, including numerical, categorical, textual, and temporal types.

[0064] S40102. Impute missing values ​​for both numerical and categorical data. Use "NaN" as the imputed value for categorical features; use the median for numerical features.

[0065] For categorical data, perform One-Hot Encoding and use One-Hot Encoding when necessary.

[0066] For text-based data, use TF-IDF or a pre-trained model to generate embedding vectors.

[0067] The preprocessing results are stored and kept consistent during the training / inference phases to ensure reproducibility for production deployment.

[0068] S402. Call the preset model candidate set and train the base model using the input data to obtain the initial nitrogen uptake prediction value and the initial yield prediction value. The AutoGluon framework enables a set of base model libraries by default (including RandomForest, ExtraTrees, KNeighbors, LightGBM, CatBoost, XGBoost, RealMLP, TabM, Mitra, TabICL, TabPFNv2, NeuralNetTorch, LinearModel, NeuralNetFastAI, TextPredictor, ImagePredictor, MultiModalPredictor, FTTransformer, FastText, WeightedEnsemble). These models are trained in parallel, and each model can use different hyperparameter configurations. Each model has a default configuration and a series of search space configurations as the basis for hyperparameter search.

[0069] S403. Train at least one secondary model using the initial nitrogen uptake prediction and the initial yield prediction.

[0070] S404. The prediction model is obtained by fusing the secondary models through weighted averaging or adaptive averaging.

[0071] S405. Utilize the hyperparameter tuning function of the AutoGluon framework to screen the prediction models and select the best one. After obtaining the initial training results, the framework automatically enters the hyperparameter tuning and model selection stage. This stage uses the trained candidate model set as a basis to further evaluate the performance metrics of each model configuration and select the optimal model:

[0072] Automatic search: Automatically optimizes the hyperparameters and combines features of each model, and compares their performance differences.

[0073] Evaluation metrics calculation: including RMSE, R², etc., forming a leaderboard based on the validation set.

[0074] Optimal model identification rules: The _BAG suffix indicates that the model was trained using the Bagging technique; _L1 (_L2 or _L3) is the level number of the model in the AutoGluon stack structure; the _FULL suffix indicates the model version obtained by refitting using all available training data in AutoGluon, and such models have clear advantages in actual deployment and prediction; models with numeric suffixes (such as _r177) indicate that the model is the 177th candidate model generated by AutoGluon during a hyperparameter search or model initialization process, and is an automatically added unique identifier; no suffix represents the default configuration model.

[0075] Automatically select RMSE and R through leaderboard 2 The model with the best cross-validation score was selected. The modeling results based on AutoGluon in this invention are as follows: The top 9 models with the best performance were selected for all model performance diagrams. For the nitrogen uptake prediction model without nitrogen application, CatBoost_r177_BAG_L1_FULL performed best (R²=0.70, RMSE=26.33 kg / ha, see...). Figure 2 For yield prediction models without nitrogen application, CatBoost_r177_BAG_L1_FULL performed best (R²=0.74, RMSE=933.58 kg / ha, see [reference]). Figure 3 For nitrogen uptake prediction models under nitrogen application, WeightedEnsemble_L3 showed the best overall performance, with the best cross-validation score (R²=0.88, RMSE=26.65 kg / ha, see [reference]). Figure 4 For nitrogen application yield prediction models, WeightedEnsemble_L2_FULL performed best, with the best cross-validation score (R²=0.81, RMSE=890.51 kg / ha, see [reference]). Figure 5 All model training was based on the AutoGluon library (version 1.3.0) in Python 3.11.11 and saved as models that can be loaded and run in Python.

[0076] S5. Based on environmental data and nitrogen management data, use the optimal prediction model to predict nitrogen uptake and yield under different nitrogen application conditions. After the model is built, it can be used to predict wheat nitrogen uptake and yield under specific climate and management combinations. The model operation steps are data preprocessing, calling the model ensemble structure, and outputting the prediction results. The data preprocessing steps are completely consistent with the training phase. When the .predict() method is called, AutoGluon will first load the preprocessor saved during training and perform consistent transformation on the input data to ensure that the new data is aligned with the training data in the feature dimension, avoiding feature drift or type inconsistency. In the model ensemble structure calling step, AutoGluon will first predict the input data in parallel using multiple base models. The output result is not the final prediction, but is used as the input feature for the next layer. Then, the outputs of each base model are combined into a feature vector, which is input into the stacked fusion unit as the final predictor output prediction result.

[0077] S6. Perform interpretability analysis on the prediction model to obtain the variable importance ranking results. Specifically, S6 also includes the following steps:

[0078] S601. The prediction model is interpreted using TreeExplainer, and the SHAP value matrix is ​​extracted.

[0079] S602. Based on the SHAP value matrix, perform full-sample aggregation and variable importance ranking to obtain the variable importance ranking results. Summarize and visualize the SHAP values ​​of the top ten important features, and perform cross-variable interpretation. The factors affecting nitrogen uptake in wheat were assessed (see...). Figure 6 ) and production (see Figure 8 The SHAP values ​​of the top ten important features of the model, and the direction and strength of the influence of the variables on the model predictions (see [reference]). Figure 7 and Figure 9 ).

[0080] The analysis revealed that key variables influencing nitrogen uptake included nitrogen application rate, soil organic carbon (SOC), tillage practices, and irrigation. Nitrogen application rate was the dominant linear variable, while the interaction between tillage practices and irrigation measures exhibited complex nonlinear effects in some high-efficiency sample blocks. Plotting these variables allows for precise identification of which variables have a stronger impact under specific conditions, aiding in the development of precision fertilization strategies.

[0081] This invention also provides a wheat nitrogen uptake and yield prediction system, comprising:

[0082] The acquisition module is used to acquire relevant data, including environmental data, nitrogen management measures data, and output data.

[0083] The preprocessing module is used to preprocess the relevant data to obtain standardized data;

[0084] The model partitioning module is used to determine the type of prediction model based on the nitrogen application situation. The types are divided into nitrogen absorption prediction model without nitrogen application, yield prediction model without nitrogen application, nitrogen absorption prediction model with nitrogen application, and yield prediction model with nitrogen application.

[0085] The model building module is used to determine the optimal prediction model for different nitrogen application conditions based on the basic model under the AutoGluon framework using standardized data.

[0086] The prediction module is used to predict nitrogen uptake and yield values ​​under different nitrogen application conditions based on the best prediction model of environmental data and nitrogen management measures data.

[0087] The present invention also provides a wheat nitrogen absorption and yield prediction device, including a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the above method.

[0088] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting nitrogen uptake and yield in wheat, characterized in that, Includes the following steps: S1. Obtain relevant data, including environmental data, nitrogen management measures data, and output data; S2. Preprocess the relevant data to obtain standardized data; S3. Determine the type of prediction model based on the nitrogen application situation. The types are divided into nitrogen absorption prediction model without nitrogen application, yield prediction model without nitrogen application, nitrogen absorption prediction model with nitrogen application, and yield prediction model with nitrogen application. S4. Using the standardized data, determine the optimal prediction model for the basic model under the AutoGluon framework under different nitrogen application conditions; S4 includes the following steps: S401. Perform integrity processing on the standardized data according to the feature type of the standardized data to obtain input data; S402. Call the preset model candidate set, use the input data to train the basic model, and obtain the initial nitrogen absorption prediction value and the initial yield prediction value. S403. Train at least one secondary model using the initial nitrogen uptake prediction value and the initial yield prediction value; S404. The secondary models are fused using weighted averaging or adaptive averaging to obtain a prediction model; S405. Use the hyperparameter tuning function of the AutoGluon framework to screen the prediction models and select the best prediction model. S5. Based on the environmental data and nitrogen management measures data, predict the nitrogen uptake and yield values ​​under different nitrogen application conditions using the optimal prediction model.

2. The method for predicting wheat nitrogen absorption and yield according to claim 1, characterized in that, S2 includes the following steps: S201. Convert the categorical variable data in the relevant data into string data according to the preset mapping rules; S202. Convert the continuous variable data in the relevant data into data with the same unit format according to the preset normalizer.

3. The method for predicting wheat nitrogen absorption and yield according to claim 1, characterized in that, S401 includes the following steps: S40101. Identify the type of the standardized data, including numerical, categorical, textual, and temporal types; S40102. Impute missing values ​​for numerical and categorical data; perform One-Hot Encoding for categorical data; and generate embedding vectors for text data using TF-IDF or a pre-trained model.

4. The method for predicting wheat nitrogen absorption and yield according to claim 1, characterized in that, In S4, K-fold cross-validation and multi-layer stacking are used to train the model.

5. The method for predicting wheat nitrogen absorption and yield according to claim 1, characterized in that, In S4, the optimal nitrogen absorption prediction model without nitrogen application is CatBoost_r177_BAG_L1_FULL, the optimal yield prediction model without nitrogen application is CatBoost_r177_BAG_L1_FULL, the optimal nitrogen absorption prediction model with nitrogen application is WeightedEnsemble_L3, and the optimal yield prediction model with nitrogen application is WeightedEnsemble_L2_FULL.

6. The method for predicting wheat nitrogen absorption and yield according to claim 1, characterized in that, It also includes the following steps: S6. Perform interpretability analysis on the prediction model to obtain the ranking of variable importance.

7. The method for predicting wheat nitrogen absorption and yield according to claim 6, characterized in that, S6 further includes the following steps: S601. The prediction model is interpreted using TreeExplainer, and the SHAP value matrix is ​​extracted; S602. Perform full sample aggregation and variable importance ranking based on the SHAP value matrix to obtain the variable importance ranking results.

8. A wheat nitrogen absorption and yield prediction system, characterized in that, include: The acquisition module is used to acquire relevant data, including environmental data, nitrogen management measures data, and output data. The preprocessing module is used to preprocess the relevant data to obtain standardized data; The model partitioning module is used to determine the type of prediction model based on the nitrogen application situation. The types are divided into nitrogen absorption prediction model without nitrogen application, yield prediction model without nitrogen application, nitrogen absorption prediction model with nitrogen application, and yield prediction model with nitrogen application. The model building module is used to determine the optimal prediction model for different nitrogen application conditions based on the standardized data under the AutoGluon framework's base model; this module performs the following steps: The standardized data is subjected to integrity processing based on its characteristic type to obtain input data. The preset model candidate set is invoked, and the basic model is trained using the input data to obtain the initial nitrogen uptake prediction value and the initial yield prediction value. At least one secondary model is trained using the initial nitrogen uptake predictions and initial yield predictions; The prediction model is obtained by fusing the secondary models through weighted averaging or adaptive averaging. The prediction models were screened using the hyperparameter tuning capabilities of the AutoGluon framework to select the optimal model. The prediction module is used to predict nitrogen uptake and yield values ​​under different nitrogen application conditions based on the optimal prediction model of the environmental data and nitrogen management measures data.

9. A device for predicting nitrogen absorption and yield in wheat, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, and the processor being used to execute the computer program to implement the steps of the method of claim 1.

Images