Simulation Method for the Effects of Coordinated Changes in Moisture and Temperature on Crop Growth under Straw Return and No-Till Management

Abstract

This invention discloses a simulation method for the impact of coordinated changes in moisture and temperature on crop growth under a straw-returning no-till model, relating to the field of agricultural crop growth models. By collecting feature data from the target area during the monitoring period, interactive features reflecting coordinated changes in moisture and temperature, time-series features capturing dynamic changes in moisture and temperature, and classification combination features capturing the interaction effects between categorical sub-features are created. A feature system and feature database are constructed, along with a LightGBM model that takes all sub-features and combination features as input and crop growth features as output. The trained LightGBM model is then used to simulate the mechanism by which coordinated changes in moisture and temperature affect crop growth. The simulation method proposed in this scheme combines the efficiency of machine learning with its ability to handle complex relationships, while also introducing feature engineering and model optimization techniques, which can effectively improve the accuracy and generalization ability of the model.

Description

Technical Field

[0001] This invention relates to the field of agricultural crop growth models, and in particular to a method for simulating the effects of coordinated changes in moisture and temperature on crop growth under a straw-returning no-till model. Background Technology

[0002] Soil moisture and temperature are key environmental factors affecting crop growth. The synergistic changes in moisture and temperature have a significant impact on crop growth, development, yield, and quality. Under no-till farming with straw return, the synergistic changes in soil moisture and temperature are more complex. On the one hand, straw mulch can regulate soil temperature and reduce moisture evaporation; on the other hand, it may also affect soil aeration and water infiltration rates. No-till farming reduces soil disturbance and helps retain soil moisture, but may lead to a greater diurnal temperature range. These changes have profound effects on crop growth and soil health.

[0003] Currently, various methods exist for simulating the effects of soil moisture and temperature changes on crop growth. These mainly include physical process-based models (such as RZWQM and APSIM) and data-driven machine learning models (such as artificial neural networks and support vector machines). However, these methods have certain limitations when simulating the coordinated changes in moisture and temperature under no-till farming with straw return:

[0004] While physical process-based models can describe the physical processes of soil moisture and temperature in detail, the calibration and validation of model parameters are difficult when dealing with complex agricultural management practices (such as straw return and no-till farming). Furthermore, these models typically require large amounts of input data, increasing the difficulty of data collection and processing. Data-driven machine learning models, although capable of handling complex nonlinear relationships, lack the ability to interpret physical processes when simulating the coordinated changes in moisture and temperature. In addition, existing machine learning models suffer from slow training speeds when handling large-scale datasets, and their generalization ability needs improvement, further limiting the practical application of these methods. Therefore, this paper proposes a simulation method for the impact of coordinated moisture-temperature changes on crop growth under straw return and no-till farming models. Summary of the Invention

[0005] The main objective of this invention is to provide a simulation method for the impact of coordinated changes in moisture and temperature on crop growth under a straw-returning no-till farming model. This method combines the efficiency and ability to handle complex relationships of machine learning, while introducing feature engineering and model optimization techniques to improve the accuracy and generalization ability of the model, effectively solving the problems in the background technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] Simulation methods for the effects of coordinated changes in water and temperature on crop growth under straw return and no-till farming models include:

[0008] Step 1: Collect characteristic data of the target area during the monitoring period. The characteristics include meteorological characteristics, soil characteristics, crop growth characteristics and management measures characteristics. Each of the characteristics contains at least one sub-characteristic. The numerical sub-characteristic data are standardized with a mean of 0 and a standard deviation of 1. The categorical sub-characteristic data are encoded.

[0009] Step 2: Construct combined features, wherein the combined features include interactive features for reflecting the synergistic changes in moisture and temperature, temporal features for capturing the dynamic changes in moisture and temperature, and classification combined features for capturing the interaction effects between categorical sub-features, wherein the interactive features and the classification combined features are both interaction terms of at least two of the sub-features;

[0010] Step 3: Construct a feature system and a feature database, wherein the feature system contains all sub-features and constructed combined features, and the feature database stores historical data of all sub-features and combined features in the feature system in a categorized manner;

[0011] Step 4: Construct the LightGBM model, wherein the model takes all sub-features and combined features in the feature database as input and all crop growth features as output. The model is trained using collected historical data. By optimizing the LightGBM model parameters and input features, the model's performance indicators reach the set expected values.

[0012] Step 5: Use the trained LightGBM model to simulate the mechanism by which the coordinated changes in water and temperature affect crop growth, and obtain the influence characteristics and prediction results of crop growth characteristics.

[0013] Furthermore, the meteorological characteristics include at least one of the following sub-characteristics: daily average temperature, maximum temperature, minimum temperature, daily precipitation, daily evaporation, and daily solar radiation.

[0014] The soil characteristics include at least one of the following sub-characteristics: soil moisture content, soil temperature, soil organic matter content, soil texture, and soil bulk density.

[0015] The crop growth characteristics include at least one of the following sub-characteristics: leaf area index, biomass, plant height, and yield;

[0016] The management measures feature includes at least one of the following sub-features: straw return method, no-tillage, irrigation amount, and fertilizer amount.

[0017] Furthermore, in step four, the LightGBM model parameters include task type, performance metric type, boosting method used by the model, number of leaf nodes, learning rate, number of trees, and random number seed.

[0018] Furthermore, in step four, the LightGBM model parameters are optimized using a Bayesian optimization method. The objective function is the model's performance index; the optimization variables are at least one of the following: the number of leaf nodes, the learning rate, and the number of trees; the optimization acquisition function is any one of the expected improvement, the probability improvement, and the upper confidence bound; and the optimization stopping condition is that the model's performance index reaches the set expected value.

[0019] Furthermore, in step four, the performance indicators include any one of the following: mean absolute error, mean square error, and coefficient of determination.

[0020] Furthermore, in step four, the input feature optimization process for the LightGBM model is as follows:

[0021] Set the importance threshold for the j-th output feature.

[0022] The importance of the i-th input feature to the j-th output feature in the LightGBM model

[0023] Define the feature filtering scenario, and the specific definition is as follows:

[0024] Scenario 1: For any j = 1, 2, ..., m; all are

[0025] Scenario 2: For any j = 1, 2, ..., m; all are

[0026] Scenario 3: There exist j = 1, 2, ..., n; such that and

[0027] Scenario 4: There exist j = 1, 2, ..., n; such that and

[0028] Where m represents the type of output feature;

[0029] Based on the definition and importance of the feature selection scenario The output result is used to perform feature filtering on the i-th input feature.

[0030] The further selection criteria for the i-th input feature are as follows:

[0031] When scenario one is satisfied, retain the i-th input feature;

[0032] When scenario two is met, the i-th input feature is eliminated.

[0033] When scenario three is met, the i-th input feature is retained;

[0034] When scenario four is met, the i-th input feature is eliminated.

[0035] The present invention has the following beneficial effects:

[0036] Compared with existing technologies, this method collects feature data of the target area during the monitoring period, constructs interactive features to reflect the coordinated changes in moisture and temperature, temporal features to capture the dynamic changes in moisture and temperature, and classification combination features to capture the interaction effects between categorical sub-features. It further constructs a feature system containing all sub-features and constructed combination features, and a feature database storing historical data of all sub-features and combination features in the feature system. A LightGBM model is then developed, taking all sub-features and combination features as input and crop growth features as output. The trained LightGBM model simulates the impact mechanism of coordinated changes in moisture and temperature on crop growth. This method combines the efficiency of machine learning with its ability to handle complex relationships, and by introducing feature engineering and model optimization techniques, it can effectively improve the model's accuracy and generalization ability. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the simulation method for the impact of synergistic changes in moisture and temperature on crop growth under the straw return and no-till model of the present invention. Detailed Implementation

[0038] The present invention will be further described below with reference to specific embodiments. The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the present invention. In order to better illustrate the specific embodiments of the present invention, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the actual product size.

[0039] The specific implementation process of the technical solution of this invention includes the following steps:

[0040] Step 1: Collect characteristic data of the target area during the monitoring period.

[0041] The features include meteorological features, soil features, crop growth features, and management features, and each feature contains at least one sub-feature. Specifically, meteorological features include at least one sub-feature from the following: daily average temperature, maximum temperature, minimum temperature, daily precipitation, daily evaporation, and daily solar radiation; soil features include at least one sub-feature from the following: soil moisture content, soil temperature, soil organic matter content, soil texture, and soil bulk density; crop growth features include at least one sub-feature from the following: leaf area index, biomass, plant height, and yield; and management features include at least one sub-feature from the following: straw return method, no-till or no-till, irrigation amount, and fertilizer amount.

[0042] Step 2: Standardize the numerical sub-feature data to have a mean of 0 and a standard deviation of 1, and encode the categorical sub-feature data.

[0043] It should be noted that, for the above-mentioned sub-features, the categorical sub-features include:

[0044] Soil texture, as a characteristic of soil, can be classified into five types according to the "Classification of Soil Particle Size and Its Nomenclature" (GB / T50123-1999): sandy soil, silty soil, loam, clay, and other soil types.

[0045] The management measures include straw return methods, such as direct return and indirect return. Direct return includes crushing and plowing, covering and returning, and leaving high stubble. Indirect return includes composting, returning through animal digestion, returning through bioreactors, carbonization and replacement, and returning biogas fertilizer.

[0046] The management measures feature whether or not no-till farming is used, specifically no-till farming;

[0047] The standardization formula is as follows: In the formula, z is the standardized value of the data; x is the data to be processed; μ is the mean of the data; σ is the standard deviation of the data;

[0048] The coding process includes thermal coding and label coding, which labels the soil texture, with different changes representing different soil textures.

[0049] Step 3: Construct combined features.

[0050] Among them, the combined features include interactive features for reflecting the coordinated changes of moisture and temperature, time-series features for capturing the dynamic changes of moisture and temperature, and classification combined features for capturing the interaction effects between categorical sub-features. Both interactive features and classification combined features are interaction terms of at least two sub-features.

[0051] For example:

[0052] For interaction features, they can be:

[0053] 1) The interaction term between soil temperature and soil moisture content: Soil temperature × Soil moisture content

[0054] This feature can capture the synergistic effect between soil temperature and moisture content; for example, the impact of changes in soil temperature on crop growth may differ under wet conditions from that under dry conditions.

[0055] 2) Interaction term between temperature change rate and soil moisture change rate: Temperature change rate × Soil moisture change rate

[0056] This feature can reflect the dynamic synergistic effect of temperature and moisture changes. For example, crops may face greater stress when temperatures rise rapidly and soil moisture decreases rapidly.

[0057] 3) Interaction term between daily average temperature and precipitation: Daily average temperature × Precipitation

[0058] This feature can capture the combined effects of temperature and precipitation on crop growth. For example, under conditions of high temperature and high humidity, crop growth may be affected by pests and diseases.

[0059] For time-series features, they can be:

[0060] 1) Daily rate of change of soil temperature: (current soil temperature) - (previous day's soil temperature)

[0061] This characteristic can reflect short-term trends in soil temperature, which has a direct impact on crop growth and physiological processes.

[0062] 2) Weekly rate of change of soil moisture content: (Current soil moisture content - Soil moisture content one week ago) / Soil moisture content one week ago

[0063] This feature can capture long-term trends in soil moisture, which is of great significance for crop water management.

[0064] For categorical combination features, they can be:

[0065] 1) Combination of straw return method and soil texture: Straw return method × Soil texture

[0066] This feature can reflect the differences in the effectiveness of different straw return methods under different soil textures. For example, the effect of covering straw in sandy soil may be different from that in clay soil.

[0067] 2) Combination of no-tillage and irrigation amount: No-tillage × Irrigation amount

[0068] This feature can capture the impact of irrigation amount on crop growth under no-till farming, for example, that appropriate irrigation may be more beneficial to crop growth under no-till farming.

[0069] Through the aforementioned feature selection and combination process, the combined effects of moisture and temperature variations on crop growth and soil processes can be effectively captured. Feature selection ensures that the features input to the model are significantly correlated with the target variable, while feature combination captures complex nonlinear relationships by creating new features. Ultimately, these optimized features provide more accurate and reliable input for simulations based on the LightGBM model, thereby improving the model's predictive and interpretative capabilities.

[0070] Step 4: Construct the feature system and feature database.

[0071] The feature system contains all sub-features and constructed combined features, while the feature database stores historical data of all sub-features and combined features in the feature system.

[0072] Step 5: Build the LightGBM model.

[0073] The model takes all sub-features and combined features in the feature database as input and all crop growth features as output.

[0074] When simulating the combined effects of moisture and temperature changes on crop growth and soil processes under a straw-returning no-till farming model, the LightGBM model's inputs include meteorological features, soil features, crop growth features, management features, and combinations of features. These input features comprehensively reflect the impacts of environmental conditions, soil conditions, crop growth, and agricultural management practices on crop growth and soil processes.

[0075] The model's outputs include crop growth indicators or management decision support indicators. These outputs not only help agricultural managers understand current crop growth and soil conditions, but also provide a scientific basis for optimizing agricultural management practices, thereby improving crop yields and soil health.

[0076] Step 6: Train the model using the collected historical data.

[0077] The model training process includes the following steps:

[0078] Dataset partitioning

[0079] The dataset is divided into training and testing sets, typically in a ratio of 80:20 or 70:30.

[0080] Define model parameters

[0081] The LightGBM model parameters include task type, performance metric type, boosting method used, number of leaf nodes, learning rate, number of trees, and random number seed. The parameters of the LightGBM model are defined according to the nature of the problem (e.g., regression or classification). In this approach, the goal is to explore the relationship between input and output features; therefore, the problem should be a regression problem. For regression problems, the following parameters can be set:

[0082] The task type is: objective: 'regression' (regression task);

[0083] The performance metric type is: metric:'mae' (using mean absolute error as the evaluation metric);

[0084] The boosting method used by the model is: boosting_type:'gbdt' (using gradient boosting decision trees).

[0085] Number of leaf nodes: num_leaves: 31;

[0086] Learning rate: 0.05;

[0087] Number of trees: n_estimators: 100;

[0088] Random number seed: random_state:42;

[0089] Training Model

[0090] The LightGBM model is trained using the training set data. During training, the model learns the relationship between input features and output features.

[0091] Step 7: Optimize the LightGBM model parameters and input features until the model's performance metrics reach the set expected values; the performance metrics include any one of the following: mean absolute error, mean squared error, and coefficient of determination.

[0092] The LightGBM model parameters are optimized using a Bayesian optimization method. For Bayesian optimization: the objective function is the model's performance metric; the optimization variables are at least one of the following: the number of leaf nodes, the learning rate, and the number of trees; the optimization acquisition function is any one of the expected improvement, the probability improvement, and the upper confidence bound; the optimization stopping condition is that the model's performance metric reaches the set expected value. The LightGBM model parameter optimization process is as follows:

[0093] Step 71: Define the optimization problem

[0094] Objective function: Determine the objective function to be optimized, which is usually a performance metric of the model, such as mean absolute error (MAE), mean squared error (MSE), or accuracy.

[0095] Parameter space: Defines the search space for model parameters. For example, for the LightGBM model, parameters that may need to be optimized include num_leaves (number of leaf nodes), learning_rate (learning rate), and n_estimators (number of trees), etc.

[0096] Step 72: Choose the prior distribution

[0097] Prior knowledge: Based on prior knowledge or experience, choose a suitable prior distribution for each parameter. For example, learning_rate can be assumed to be uniformly distributed (e.g., between 0.01 and 0.1), and num_leaves can be assumed to be discretely uniformly distributed (e.g., between 10 and 50).

[0098] Default value: If there is no prior knowledge, the default prior distribution, such as uniform distribution or normal distribution, can be used.

[0099] Step 73: Initialize the optimization process

[0100] Initial points: Select several initial points for evaluation. These initial points can be randomly selected or reasonable values ​​based on experience.

[0101] Evaluate the objective function: Evaluate the objective function at these initial points and record the performance metrics for each parameter combination.

[0102] Step 74: Build the proxy model

[0103] Proxy model: Using a Gaussian process (GP) or other probabilistic model as a proxy model, an approximate model of the objective function is constructed based on the evaluated points.

[0104] Fitting the surrogate model: Input the evaluation results of the initial point into the surrogate model, fit the surrogate model parameters so that it can predict the value of the objective function.

[0105] Step 75: Select the next assessment point

[0106] Acquisition Function: Defines an acquisition function used to select the next evaluation point based on the surrogate model. Common acquisition functions include:

[0107] Expected Improvement (EI): Select the point with the largest expected improvement.

[0108] Probability of Improvement (PI): Select the point with the largest probability of improvement.

[0109] Upper Confidence Bound (UCB): Select the highest point on the upper confidence boundary.

[0110] Optimize the acquisition function: By optimizing the acquisition function, we can find the next most promising combination of parameters.

[0111] Step 76: Evaluate the objective function

[0112] Evaluation at the next selected evaluation point: Evaluate the objective function at the next selected evaluation point and record its performance metrics.

[0113] Update the surrogate model: Add the evaluation results of the new points to the surrogate model and refit the surrogate model.

[0114] Step 77: Iterative Optimization

[0115] Repeat steps 75 and 76: Repeat the process of selecting the next evaluation point, evaluating the objective function, and updating the surrogate model until the stopping condition is met.

[0116] Step 78: Output optimal parameters

[0117] Optimal parameters: Among all evaluation points, the parameter combination with the optimal objective function value is selected as the final optimization result.

[0118] Evaluate the optimal parameters: Retrain the model using the optimal parameters and evaluate its performance on the validation or test set.

[0119] The input feature optimization process for the LightGBM model is as follows:

[0120] Set the importance threshold for the j-th output feature.

[0121] The importance of the i-th input feature to the j-th output feature in the LightGBM model

[0122] Define the feature filtering scenario, and the specific definition is as follows:

[0123] Scenario 1: For any j = 1, 2, ..., m; all are

[0124] Scenario 2: For any j = 1, 2, ..., m; all are

[0125] Scenario 3: There exist j = 1, 2, ..., n; such that and

[0126] Scenario 4: There exist j = 1, 2, ..., n; such that and

[0127] Where m represents the type of output feature;

[0128] Based on the definition and importance of the feature selection scenario The output result is used to perform feature filtering on the i-th input feature, and the filtering principle is as follows:

[0129] When scenario one is satisfied, retain the i-th input feature;

[0130] When scenario two is met, the i-th input feature is eliminated.

[0131] When scenario three is met, the i-th input feature is retained;

[0132] When scenario four is met, the i-th input feature is eliminated.

[0133] Step 8: Use the trained LightGBM model to simulate the mechanism by which the coordinated changes in water and temperature affect crop growth, and obtain the influence characteristics and prediction results of crop growth characteristics.

[0134] It should be noted that the LightGBM model can output the importance of each input feature. The trained LightGBM model can output input features with different levels of importance for crop growth characteristics, thereby simulating the impact of coordinated changes in water and temperature on crop growth. At the same time, it can predict the future situation of the output features, providing decision support for agricultural management.

[0135] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A simulation method for the effect of the coordinated change of water and temperature on crop growth under the mode of straw returning and no-tillage, characterized in that, Comprise: Step one: collect the characteristic data of the target area in the monitoring period, wherein the characteristics include weather characteristics, soil characteristics, crop growth characteristics and management measures characteristics, and any of the characteristics contains at least one sub-feature, the numerical sub-feature data is standardized to mean 0 and standard deviation 1, and the category sub-feature data is encoded; Step two: construct combined features, wherein the combined features include interaction features for reflecting the synergistic change of water and temperature, time series features for capturing the dynamic change of water and temperature, and classification combined features for capturing the interaction effect between category sub-features, wherein the interaction features and the classification combined features are at least two interaction terms of the sub-features; Step three: build a feature system and a feature database, wherein the feature system contains all sub-features and constructed combined features, and the feature database stores the historical data of all sub-features and combined features in the feature system; Step four: build a LightGBM model, wherein the model takes all sub-features and combined features in the feature database as input, takes all crop growth characteristics as output, trains the model using the collected historical data, and optimizes the LightGBM model parameters and input features until the performance index of the model reaches the set expected value; Step five: use the trained LightGBM model to simulate the influence mechanism of the synergistic change of water-temperature on crop growth, and obtain the influence features and prediction results of crop growth characteristics.

2. The simulation method of the synergistic change of water-temperature on crop growth in the straw returning no-tillage mode according to claim 1, characterized in that: The weather characteristics include at least one sub-feature of daily average temperature, maximum temperature, minimum temperature, daily precipitation, daily evaporation and daily solar radiation; The soil characteristics include at least one sub-feature of soil moisture content, soil temperature, soil organic matter content, soil texture and soil bulk density; The crop growth characteristics include at least one sub-feature of leaf area index, biomass, plant height and yield; The management measures characteristics include at least one sub-feature of straw returning mode, no-tillage or not, irrigation amount and fertilizer amount.

3. The method according to claim 1, wherein the method is characterized by, In step four, the LightGBM model parameters include task type, performance index type, boosting method used by the model, leaf node number, learning rate, tree number and random number seed.

4. The method according to claim 3, wherein the method is characterized by, In step four, the LightGBM model parameters are optimized by Bayesian optimization method, wherein the optimization objective function is the performance index of the model; the optimization variable is at least one of the leaf node number, the learning rate and the tree number; the optimization acquisition function is any one of the expected improvement, the probability improvement and the upper confidence bound; the optimization stop condition is that the performance index of the model reaches the set expected value.

5. The method according to claim 4, wherein the method is characterized by, In step four, the performance index includes any one of the mean absolute error value, the mean square error value and the determination coefficient.

6. The method according to claim 1, wherein the method is characterized by, In step four, the input feature optimization process of the LightGBM model is: Setting an importance level screening threshold for the jth output feature An importance of an i-th input feature to a j-th output feature using a LightGBM model Define a feature screening scenario, and the specific definition is: Scenario 1: For any j = 1, 2,..., m; both are Scenario two: for any j = 1, 2,..., m; both are Scenario three: there exist j = 1, 2,..., n; such that and Scenario four: there exist j = 1, 2,..., n; such that and Wherein, m is the category of the output feature; According to the definition content and importance degree of the feature screening scenario , the i-th input feature is screened.

7. The method according to claim 6, wherein the method is characterized by, The screening principle of the i-th input feature is: When scenario one is met, the i-th input feature is retained; When scenario two is met, the i-th input feature is screened out; When scenario three is met, the i-th input feature is retained; When scenario four is met, the i-th input feature is screened out.

Images