Data Processing Method for Winter Wheat Field Soil Fertility Evaluation Based on Comprehensive Weight Calculation

Abstract

This invention discloses a data processing method for evaluating winter wheat field soil fertility based on comprehensive weight calculation, belonging to the field of data processing technology. The method includes: S1, collecting nutrient content and water retention data from field soil using a sensor network, and grouping the collected nutrient content and water retention data using decision node splitting and entropy calculation to obtain soil fertility stratification results; S2, based on the soil fertility stratification results, obtaining root depth and leaf water content indicators of winter wheat in different soil strata from crop monitoring equipment, and analyzing the correlation strength between root depth and leaf water content indicators and fertility stratification using leaf node classification and feature selection criteria to determine the fertility contribution coefficient. This data processing method for evaluating winter wheat field soil fertility based on comprehensive weight calculation achieves efficient resource allocation and stable yield, significantly improving the drought resistance and resource utilization efficiency of winter wheat under drought conditions.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a data processing method for evaluating soil fertility in winter wheat fields based on comprehensive weight calculation. Background Technology

[0002] Winter wheat is one of the world's most important food crops, and its yield stability is directly related to food security and sustainable agricultural development. Against the backdrop of frequent droughts, the drought resistance of winter wheat has become a crucial factor in ensuring yield. Especially in soil environments with varying fertility levels, how to improve the drought resistance of winter wheat through scientific management, while simultaneously optimizing resource allocation to achieve high yields, has become a key research topic in the field of agriculture.

[0003] Current research on drought resistance in winter wheat largely focuses on single environmental or management measures, such as the independent effects of irrigation or fertilizer application. However, these methods often overlook the complex interaction between soil fertility and drought resistance. Existing schemes typically employ fixed patterns for fertility management and irrigation scheduling, lacking dynamic analysis of differences in drought resistance under different soil fertility conditions. This leads to difficulties in accurately matching resource allocation to actual field soil conditions in production, thus affecting the yield stability of winter wheat. Against this backdrop, soil fertility stratification has become a core technical factor influencing drought resistance assessment. Soil fertility stratification refers to classifying field soils into different fertility levels based on differences in soil nutrient status, structure, and water retention capacity. Since fertility levels directly affect crop root development and water use efficiency, soils with different fertility levels exhibit significant differences in their responses to irrigation and fertilizer. However, current research struggles to accurately quantify the specific contribution of fertility levels to drought resistance when assessing fertility stratification. This is because there is a complex relationship between the dynamic changes in fertility stratification and the drought resistance needs of crops at different growth stages, and effective technical means are lacking to capture this dynamic interaction. The core challenge in assessing drought resistance based on fertility stratification lies in accurately quantifying the contribution of fertility levels to drought resistance. Differences in fertility stratification lead to variations in soil moisture retention and nutrient supply capabilities, which in turn affect the physiological responses of winter wheat under drought conditions. For example, in low-fertility soils, winter wheat roots may develop poorly due to insufficient nutrients, resulting in reduced water absorption; while in high-fertility soils, excessive fertilization may lead to water waste or soil compaction, similarly reducing drought resistance. This non-linear relationship between fertility and drought resistance makes quantifying the contribution of fertility a complex technical challenge. Summary of the Invention

[0004] The purpose of this invention is to provide a data processing method for evaluating winter wheat field soil fertility based on comprehensive weight calculation. In field trials, this method uses scientific methods to quantify the differences in drought resistance of winter wheat under different fertility levels, and optimizes fertility management and irrigation strategies based on these differences to improve yield stability.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for processing field soil fertility evaluation data for winter wheat based on comprehensive weight calculation, the method comprising the following steps:

[0006] S1. Nutrient content and water retention data are collected from field soil through a sensor network. The collected nutrient content and water retention data are grouped and processed using decision node splitting and entropy calculation to obtain soil fertility stratification results.

[0007] S2. Based on the soil fertility stratification results, the root depth and leaf moisture content of winter wheat in different soil strata were obtained from the crop monitoring equipment. The correlation strength between root depth and leaf moisture content and fertility stratification was analyzed using leaf node classification and feature selection criteria. The contribution weight of each soil fertility stratum to root depth and leaf moisture content was calculated, and the fertility contribution coefficient was determined.

[0008] S3. If the fertility contribution coefficient exceeds the preset threshold, the change in water use efficiency of winter wheat under drought conditions is calculated by combining the simulation module with the pruning treatment method to obtain the quantitative value of drought resistance performance.

[0009] S4. Based on the quantified drought resistance performance, cross-validation evaluation and classification accuracy are used to classify and optimize fertilizer application rate and irrigation scheduling. The weighted average method is used to calculate the contribution weight of fertilizer application rate and irrigation scheduling to water use efficiency, determine the adjustment direction of resource allocation scheme, and obtain preliminary optimized configuration parameters.

[0010] S5. Extract relevant variables from the initial optimized configuration parameters, integrate root development data and nutrient supply data using initial parameter settings and error function definitions, and determine the dynamic balance model of resource allocation.

[0011] As can be seen from the above technical solution, the present invention has the following beneficial effects:

[0012] This data processing method for evaluating winter wheat field soil fertility based on comprehensive weight calculation addresses the problem of insufficient yield stability in winter wheat under drought conditions. It collects multi-layered soil nutrient and moisture data through a sensor network, employs decision node splitting and entropy calculation to stratify fertility, analyzes the correlation between root depth and leaf water content with fertility stratification, and determines fertility contribution coefficients. When a coefficient exceeds a threshold, the invention calculates water use efficiency through a simulation module combined with pruning treatment to quantify drought resistance. Then, it optimizes fertilizer application and irrigation scheduling using cross-validation and classification accuracy, extracting highly correlated variables to establish a dynamic equilibrium model. If yield stability does not meet the target, the invention uses gradient descent to adjust irrigation intervals and fertilizer ratios, verifies the applicability of the scheme using historical drought data, and updates the contribution coefficients of the fertility stratification database. Through multi-layered data fusion and dynamic optimization, this invention achieves efficient resource allocation and stable yield, significantly improving the drought resistance and resource utilization efficiency of winter wheat under drought conditions. Attached Figure Description

[0013] Figure 1 This is a flowchart of the data processing method for evaluating winter wheat field soil fertility based on comprehensive weight calculation, as described in this invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] like Figure 1 As shown, this invention provides a technical solution: a method for processing field soil fertility evaluation data for winter wheat based on comprehensive weight calculation, the method comprising the following steps:

[0016] S1. Nutrient content and water retention data are collected from field soil through a sensor network. The collected nutrient content and water retention data are grouped and processed by decision node splitting and entropy calculation to obtain soil fertility stratification results.

[0017] S2. Based on the soil fertility stratification results, the root depth and leaf moisture content of winter wheat in different soil strata were obtained from the crop monitoring equipment. The correlation strength between the root depth and leaf moisture content and fertility stratification was analyzed using leaf node classification and feature selection criteria to determine the fertility contribution coefficient.

[0018] S3. If the fertility contribution coefficient exceeds the preset threshold, the change in water use efficiency of winter wheat under drought conditions is calculated by combining the simulation module with the pruning treatment method to obtain the quantitative value of drought resistance performance.

[0019] S4. Based on the quantitative value of drought resistance, cross-validation evaluation and classification accuracy are used to classify and optimize fertilizer application and irrigation scheduling, determine the adjustment direction of resource allocation scheme, and obtain preliminary optimized configuration parameters.

[0020] S5. Extract relevant variables from the initial optimized configuration parameters, integrate root development data and nutrient supply data using initial parameter settings and error function definitions, and determine the dynamic balance model of resource allocation;

[0021] S6. If the dynamic equilibrium model shows that the yield stability index is lower than the target level, the irrigation interval and fertilizer ratio are adjusted by updating the gradient descent and judging the convergence condition to obtain the final resource allocation scheme.

[0022] S7. Based on the final resource allocation plan, obtain validation data on drought environment from the historical database, and combine variable step size adjustment and simulation feedback loop to judge the applicability of the plan and obtain the yield stability assessment results.

[0023] S8. Based on the yield stability assessment results, update the contribution coefficients in the soil fertility stratification database, and determine the input data for the next assessment cycle by combining the proportional balance constraints and the final scheme output.

[0024] This implementation utilizes a distributed sensor network to collect real-time data on soil nutrients and moisture in the field. The collected data is effectively grouped using decision tree splitting nodes and an entropy function to achieve preliminary soil fertility stratification. Regarding crop response data acquisition, growth characteristics such as root depth and leaf moisture content are monitored, and a feature selection algorithm is applied to analyze their correlation with fertility stratification, thereby deriving a fertility contribution coefficient reflecting nutrient utilization and water absorption capacity. When this coefficient exceeds a preset threshold, the system uses a simulation module and pruning strategies to calculate the water use efficiency of winter wheat under drought conditions, thus quantifying drought resistance. Furthermore, cross-validation methods and classification accuracy indicators from machine learning are used to optimize and classify fertilizer application and irrigation strategies, initially generating a reasonable resource allocation scheme. Based on this, a dynamic equilibrium model is constructed by combining crop root development characteristics and soil nutrient supply. If the yield stability shown by this model is unsatisfactory, a gradient descent algorithm is used to adjust the resource allocation parameters to approach the optimal solution. Finally, the system simulates and compares the generated allocation scheme with data from historical drought conditions, and evaluates its feasibility and stability using a feedback mechanism. The evaluation results are then used to update the contribution coefficients and database content, forming a closed-loop optimization path.

[0025] This implementation method significantly improves the scientific rigor and precision of fertilization and irrigation in winter wheat cultivation by constructing a soil fertility evaluation and resource allocation system driven by contribution coefficients. It utilizes a data-driven decision-making process to achieve hierarchical management, enhancing the system's adaptability to field heterogeneity and improving soil resource utilization efficiency. Simultaneously, dynamic adjustment and feedback mechanisms ensure the timeliness and reliability of the plan, thereby optimizing water and fertilizer use structure and reducing resource waste while maintaining yield stability. Especially under drought conditions, simulating and predicting water use efficiency helps improve crop stress resistance. Furthermore, this method has strong adaptability and scalability, facilitating its application in different regions and seasons.

[0026] S1 includes: acquiring nutrient content and moisture data from multiple depths of field soil using a sensor network, storing it as a structured dataset to obtain raw soil data; preprocessing the raw soil data using data cleaning techniques to remove outliers and missing values ​​to obtain cleaned soil data; extracting feature values ​​from the nutrient content and moisture data based on the cleaned soil data, constructing feature vectors to obtain a soil feature set; splitting the soil feature set using a decision tree algorithm, determining the splitting nodes based on entropy values ​​to obtain preliminary grouping results; merging the groups if the entropy value of the preliminary grouping results is lower than a preset threshold to optimize the grouping structure and obtain optimized grouping results; sorting the optimized grouping results through hierarchical analysis, and determining soil fertility stratification based on nutrient content and moisture data to obtain the final stratification results; generating soil fertility stratification data based on the final stratification results, storing it in a queryable format to obtain a soil fertility stratification dataset.

[0027] The core of this implementation lies in acquiring soil fertility stratification data from different depths of soil using a sensor network, and processing the data according to a rigorous process, including feature extraction, decision tree splitting, group optimization, hierarchical sorting, and structured data storage. First, the sensor network is deployed in field plots, with three depth layers selected for each plot: 10 cm, 30 cm, and 50 cm below the soil surface. Each depth layer is equipped with sensors to collect nitrogen, phosphorus, and potassium concentrations, as well as moisture content. The time interval for each data collection is one hour, and the results are automatically uploaded to a central server. Each record consists of five fields: time, depth, nitrogen concentration, phosphorus concentration, potassium concentration, and moisture content, forming a structured raw soil dataset.

[0028] After data collection, the system performs data cleaning on the raw soil data. The cleaning process includes two steps. The first step is outlier removal. The normal range for nitrogen concentration is set to 0 to 100 mg / kg, phosphorus concentration to 0 to 50 mg / kg, potassium concentration to 0 to 150 mg / kg, and moisture content to 0 to 50%. All values ​​outside these ranges are directly removed. The second step is missing value imputation. If a field in a record is empty and the missing data rate for that field at that depth is less than 5%, it is imputed using the average of two adjacent time points at the same depth. If the missing data rate for that field is greater than 5%, the entire record is deleted and will not be included in subsequent processing.

[0029] The cleaned data was used for feature extraction. Data within each depth layer was aggregated over time, and the average values ​​of nitrogen, phosphorus, potassium, and moisture were calculated for that time period, forming a set of feature values. Each set of feature values ​​constituted a feature vector, including four specific values: average nitrogen, average phosphorus, average potassium, and average moisture. All feature vectors were then aggregated and input into a decision tree model for splitting. The splitting process used information entropy as the splitting criterion. Each split first calculated the distribution ratio of each group in the current sample set under different target classifications, and then calculated the information entropy accordingly. The information entropy was calculated by iterating through each classification, multiplying its proportion by its logarithm, and then taking a weighted sum of the results. Then, the degree to which each feature reduced the information entropy was evaluated, and the feature with the largest information gain was selected as the splitting node for splitting.

[0030] The soil feature set, after data cleaning and feature extraction, is used as input data. This feature set consists of several feature vectors, each containing four values: average nitrogen concentration, average phosphorus concentration, average potassium concentration, and average moisture content. A decision tree classification algorithm is used to classify and split this feature set. The core of this algorithm is to select an optimal feature as the splitting criterion and divide the samples according to the specific value range of that feature, so that the purity of the sub-sample set is significantly improved after each split. The specific splitting process is as follows: all samples are treated as a whole, and their classification information is statistically analyzed. The classification criterion here is a preset soil fertility level label. If there is no initial label, the data is first divided into several initial classes using a clustering method. Then, the total information entropy of the whole sample is calculated. The information entropy is calculated by calculating the proportion of each class in the sample set, multiplying this proportion by its base-2 logarithm, and then summing the weighted values ​​of all classes and taking the negative value. The result is the entropy value of the current sample set, used to measure its degree of contamination. Each separable feature (i.e., nitrogen, phosphorus, potassium, and moisture) is tested separately. The system sorts all values ​​of the feature and selects the nearest intermediate point of each value as a candidate split threshold. For example, if nitrogen concentrations are 10, 15, 20, 25, and 30 mg / kg, the candidate split points are 12.5, 17.5, 22.5, and 27.5. For each candidate split point, the data is divided into two groups: those less than or equal to the value and those greater than the value. The sub-information entropy of each group is calculated, and a weighted average is calculated based on their sample size proportions to obtain the expected entropy value at that split point. The expected entropy value corresponding to the split point is subtracted from the original total entropy to obtain the information gain. For each candidate feature and all its candidate split points, the above process is repeated until the feature with the maximum information gain and its corresponding split point are identified, which are then determined as the node and threshold for this round of decision tree splitting. Using the optimal split feature and split point as the boundary, the soil feature set is divided into two subsets, representing data subsets that are determined to be different preliminary soil categories under the given feature conditions. At this point, the system obtains a valid preliminary grouping result. Repeat steps one through four for these two subsets to continue searching for the optimal splitting feature and executing the next level of splitting until the termination condition is met. The termination condition includes, but is not limited to: the entropy value of the current subset is lower than a preset threshold (e.g., 0.2), the number of samples is less than the minimum number of samples (e.g., 10), or the information gain is lower than the minimum gain threshold (e.g., 0.01), at which point further splitting stops, and the current node becomes a leaf node. The entire process is executed recursively, ultimately constructing a preliminary grouping structure composed of multiple leaf nodes. Each leaf node represents a soil feature group, indicating a preliminary result of soil stratification with similar nitrogen, phosphorus, potassium, and moisture characteristics.

[0031] After the split is complete, the system recalculates the entropy value of each group to determine its classification purity. If the entropy value of a group is lower than a set threshold of 0.2, the group is considered to have relatively concentrated information and can be regarded as a stable category. If there are multiple groups with entropy values ​​less than 0.2, these groups will be merged into a new group to form an optimized grouping result. The threshold of 0.2 is determined as follows: the average entropy value before a significant decrease in information gain in historical data is selected as the standard, 200 historical split nodes are statistically analyzed, and their information entropy distribution ranges from 0.1 to 0.35. The dividing point of 0.2 with relatively strong information stability is selected as the empirical split threshold.

[0032] The optimized grouping structure was then sorted. The sorting process consisted of three steps: First, the four characteristic values ​​of nitrogen, phosphorus, potassium, and moisture in each group were standardized, with a uniform value range of 0 to 1. The standardization method was to subtract the overall average value from the characteristic value and then divide by its range. Second, weights were assigned to the four standardized values, with weights of 30% for nitrogen, 20% for phosphorus, 30% for potassium, and 20% for moisture. Third, the four standardized values ​​for each group were multiplied by their corresponding weights, and the sum was used to obtain the comprehensive score for that group. The groups were then sorted from highest to lowest comprehensive score. The levels corresponding to the sorted groups represent the soil fertility stratification levels, labeled as Level 1, Level 2, and Level 3 from highest to lowest.

[0033] Finally, the results of each stratification are integrated into soil fertility stratification data. Each data point includes the stratification level, depth range, average nitrogen, phosphorus, potassium, and moisture content, sampling time period, and geographical location. All results are stored in a structured data format in a table format in the database system. Each record is indexed by time, location, and stratification level to ensure quick retrieval and reuse for subsequent queries, simulations, or optimization processes.

[0034] S2 includes: acquiring root depth and leaf moisture content data of winter wheat from different soil fertility strata using crop monitoring equipment, storing it as a structured dataset to obtain raw crop data; cleaning the raw crop data using data preprocessing techniques to remove outliers and missing values ​​to obtain cleaned crop data; extracting feature values ​​of root depth and leaf moisture content from the cleaned crop data to construct feature vectors to obtain a crop feature set; classifying the crop feature set using a random forest algorithm, calculating feature importance based on information gain, determining the correlation strength between root depth and leaf moisture content and soil fertility stratification to obtain preliminary correlation results; if the classification accuracy of the preliminary correlation results is lower than a preset threshold, re-selecting feature vectors using feature selection criteria, optimizing the random forest model to obtain optimized correlation results; analyzing the optimized correlation results to calculate the contribution weight of each soil fertility stratum to root depth and leaf moisture content to obtain fertility contribution coefficients; and generating growth adaptability data of winter wheat in different soil fertility strata based on fertility contribution coefficients, storing it in a queryable format to obtain the final growth adaptability dataset.

[0035] Crop monitoring equipment installed in soil fertility stratification zones in the field simultaneously collected data on winter wheat root depth and leaf moisture content across different strata. Root depth was measured in millimeters, and the monitoring equipment recorded the maximum root depth every 24 hours using resistivity response or image recognition technology, with an effective collection range of 10 to 500 millimeters. Leaf moisture content was measured as a percentage, read after conversion using leaf infrared reflectance, with a collection range of 20% to 90%. Each raw data entry included sampling time, corresponding soil stratum number, root depth value, and leaf moisture value, forming a structured table and creating the raw crop dataset. Data preprocessing consisted of two steps. The first step was outlier removal: records with root depths less than 10 millimeters or exceeding 500 millimeters were removed; records with leaf moisture contents below 20% or above 90% were also removed. The second step was missing value handling, checking the data table for null values. If the missing percentage of a field is less than 5%, the average of the two adjacent time points under the same stratum number is used to fill the missing value; if the missing percentage exceeds 5%, the entire record is deleted. After processing, the data is aggregated by stratum number and sampling time interval, and the average root depth and leaf moisture content in each group are calculated to obtain a feature vector for each group containing stratum number, time interval, average root depth, and average leaf moisture content, thus forming a crop feature set.

[0036] The crop feature set is input into a random forest classification model, with the classification target being soil fertility stratification numbering. The model construction process is as follows: First, 100 decision trees are constructed, each with a maximum depth of 10 layers. The number of features considered for each split is the square root of the total number of features; that is, if two features are currently shared, one feature is used each time. During model training, each tree uses bootstrapping to extract a training set with replacement from the original sample set and constructs the tree structure. At each split node, the model calculates the information gain of all possible split points for the currently available feature. The information gain is calculated by first calculating the initial information entropy of the current sample set, then calculating the weighted average entropy of all subsets after the feature split; the difference between the two is the feature gain value. The feature with the largest information gain and its corresponding split point are selected as the splitting criterion for the current node. After all trees are constructed, the system calculates the frequency of each feature used as a split node in all trees and the weighted sum of its corresponding information gain. After normalization, the importance score of the feature is obtained, called the feature contribution value.

[0037] The feature contribution values ​​of root depth and leaf moisture content represent their correlation strength with soil stratification classification. The system then evaluates the overall model performance using classification accuracy as the evaluation metric. Accuracy is calculated by using samples outside the training set for prediction and recording the ratio of correctly predicted samples to the total number of samples. If the overall accuracy is below 85%, the system re-enters the feature optimization process. This 85 is a system-preset accuracy threshold, determined based on the inflection point of classification stability in 200 historical model training results; a value below this indicates underfitting or feature redundancy. In the optimization process, the system sorts feature contribution values, retains the top 50 percentile features, removes low-contribution features, and reconstructs a random forest model with unchanged parameters. This training process is repeated until the classification accuracy is greater than or equal to 85.

[0038] Once the model accuracy reaches the target, the system extracts the average information gain value of all samples corresponding to each soil fertility stratum number, obtaining the average contribution values ​​of root depth and leaf moisture content to the stratum classification. The two contribution values ​​are then weighted and averaged to obtain the fertility contribution coefficient for that stratum. This coefficient represents the comprehensive impact of that stratum on crop growth. The weighting method is set according to the actual physiological dependence of the crop, with root depth weighted at 60% and leaf moisture content weighted at 40%. The fertility contribution coefficient value corresponding to each stratum number is uniformly normalized to the range of 0 to 1 to facilitate horizontal comparative analysis. Finally, the stratum number, contribution coefficient, mean root depth, mean leaf moisture content, and sampling time records are integrated into a standardized data structure to construct growth adaptability data records. All records are uniformly stored in a structured format in the database system, indexed by stratum number and time to support on-demand querying and model invocation.

[0039] The previous step yielded the fertility contribution coefficient for each soil fertility stratum. This coefficient is obtained by calculating the average information gain of root depth and leaf moisture content on the soil fertility stratification classification results, then normalizing it to the 0-1 range after weighted averaging with feature weights. Root depth is weighted at 60%, and leaf moisture content at 40%, reflecting their dominant and auxiliary roles in growth response, respectively. Each stratum number corresponds to a unique fertility contribution coefficient. The system determines a growth adaptability score based on this coefficient; the score value is equivalent to the coefficient, with a score closer to 1 indicating stronger crop adaptability to that fertility stratum. Next, the system constructs growth adaptability data records. Each record consists of the following fields: soil stratum number, fertility contribution coefficient, average root depth value, average leaf moisture content value, data collection period, sampling area number, and growth adaptability score. All data comes from previous model outputs or monitoring data processing results, requiring no additional manual input. Each field uses standardized units: root depth is in millimeters, leaf moisture content is expressed as a percentage, the time period is expressed as a date range, and the area number uses the standard plot coding format. To enable subsequent calls and system integration, the system stores the aforementioned growth adaptability dataset in a structured format. The data format is a relational table structure, with each record as a row and field names serving as column names, stored using a standard database management system. The system sets up a joint index on the stratification number field and the time period field to support rapid querying and filtering based on soil layer or time range. Simultaneously, the system provides a unified data interface for other modules to use, including the drought resistance analysis module, resource allocation module, and subsequent evaluation cycle module. The final growth adaptability dataset can be used for internal system integration and can also provide external interface support through standardized export formats, such as CSV, JSON, or SQL structures, serving as a data input source for agricultural intelligent analysis systems, land management systems, or agricultural policy support decision-making systems, achieving efficient sharing and multi-system collaboration.

[0040] Crop monitoring equipment is used to continuously acquire growth status data of winter wheat in different soil fertility strata. It mainly includes two types of equipment: those for monitoring root depth and those for monitoring leaf moisture content. For root depth monitoring, a root scanning imager can be used. This device combines a transparent root box and a high-resolution image acquisition unit, using periodic imaging and image recognition algorithms to accurately measure the maximum root depth with millimeter-level accuracy. Alternatively, a buried resistive root sensor can be selected. This device reflects root distribution in real time based on the principle of soil resistance changes at different depths, making it suitable for long-term field deployment. Segmented root monitoring probes can also be used. By setting electrical contacts at different depths, the time and depth are recorded when the root contacts the sensor area, thus estimating root expansion. For monitoring leaf moisture content, infrared leaf moisture meters can be used. These devices acquire non-contact moisture data by measuring changes in near-infrared reflectance of the leaf surface, making them suitable for rapid measurement over large areas. Portable leaf moisture meters can also be used, which directly read the leaf moisture percentage through a clamp-on capacitive sensing structure, providing stable and reliable measurement results. Additionally, multispectral remote sensing devices can be utilized, using drones or fixed platforms equipped with sensors to collect red, green, blue, and near-infrared images. Algorithm processing can then yield information on crop leaf moisture distribution over a large area. For situations requiring simultaneous monitoring of multiple parameters, a combined crop growth monitoring system can be used. This system integrates root probes, moisture sensors, environmental factor monitoring modules, and data acquisition terminals, transmitting data to a server via wired or wireless means to achieve joint monitoring of root development and leaf condition. All devices can automatically sample at set intervals, and the resulting data is stored in a structured format for easy retrieval in subsequent model calculations and analyses.

[0041] S3 includes the following steps: If the fertility contribution coefficient exceeds a preset threshold, soil fertility data and environmental factor data are acquired from the winter wheat planting area through the data acquisition module to obtain an initial environmental dataset. Based on the initial environmental dataset, principal component analysis is used to extract the main feature vectors of soil fertility and environmental factors to obtain an environmental feature set. The environmental feature set is loaded through the simulation module, and combined with pruning treatment methods, the trend of water use efficiency of winter wheat under drought conditions is calculated to obtain an efficiency change dataset. If the fluctuation range of the efficiency change dataset exceeds a preset range, the efficiency change dataset is classified using the support vector machine algorithm to determine the key influencing factors of water use efficiency and obtain classification results. Based on the classification results, the weighted average method is used to calculate the contribution weight of influencing factors to water use efficiency and obtain a drought resistance performance quantification value. Based on the drought resistance performance quantification value and soil fertility data, adaptive distribution data of winter wheat under different drought conditions is generated and stored in a structured format to obtain a final adaptive dataset. Based on the final adaptive dataset, cluster analysis is used to classify the drought resistance performance of winter wheat to obtain drought resistance performance level data.

[0042] In this scheme, the drought resistance performance analysis process is triggered when the fertility contribution coefficient of a certain soil stratum is greater than or equal to 0.6. This fertility contribution coefficient is calculated by weighting the average information gain of winter wheat root depth and leaf moisture content on the fertility stratification classification results during the analysis process. The value ranges from 0 to 1, with 0.6 set as a preset threshold. The specific value is based on statistical analysis of the drought resistance performance of winter wheat under different fertility conditions in 200 historical samples, selecting the boundary point where yield stability is significantly controlled by soil fertility as the critical value. When a stratification coefficient reaches or exceeds 0.6, the system, through a data acquisition module deployed in that area, collects soil nutrient data, including nitrogen, phosphorus, and potassium concentrations, and environmental factor data from the field at a sampling frequency of once per day for a continuous 30-day period. The environmental factors include temperature, humidity, wind speed, light intensity, and rainfall. Nitrogen concentration was measured in milligrams per kilogram, with a range of 0 to 100; phosphorus in 0 to 50; and potassium in 0 to 150. Temperature was measured in degrees Celsius, with a range of 5 to 40; humidity in percentage, with a range of 20 to 100; wind speed in meters per second, with a range of 0 to 15; light intensity in watts per square meter, with a range of 100 to 1000; and rainfall in millimeters, with a range of 0 to 50. These data constituted the initial environmental dataset.

[0043] After data collection, all data undergoes standardization preprocessing by subtracting the mean from each indicator and dividing by the standard deviation to eliminate the influence of different units of measurement. Principal component analysis (PCA) is then used to reduce the dimensionality of the standardized data, calculating the covariance matrix, extracting eigenvalues ​​and corresponding eigenvectors, and sorting the eigenvalues ​​from largest to smallest. The top principal components with a cumulative contribution rate of 85% or higher are selected as representative variables, typically 3 to 5 principal components, forming an environmental feature set. This environmental feature set is used as input and loaded into the crop water use simulation module. This module employs a daily-scale crop response simulation structure, calculating the impact of each input feature combination on the water use efficiency of winter wheat. Water use efficiency is defined as the reciprocal of the water consumed per unit of dry matter yield, expressed in kilograms per cubic meter. The system calculates net water use efficiency by simulating daily root water uptake and leaf transpiration, simulating continuously for 30 days, and outputting a time series containing daily efficiency values, forming an efficiency change dataset.

[0044] To calculate the trend of water use efficiency changes in winter wheat under drought conditions, the specific implementation process is as follows: First, after completing principal component analysis, environmental features with a cumulative contribution rate of 85% or higher are numbered from highest to lowest variance contribution as the 1st principal feature, 2nd principal feature, up to the Nth principal feature, where N ranges from 3 to 5. The determination of N is based on the number of features corresponding to the first time the cumulative contribution rate reaches or exceeds 85% in historical samples. Then, the above environmental features are loaded into the simulation module as the sole input, with a time step of 1 day, a simulation period of 30 days, and a drought scenario of daily series where soil moisture content is below 40% of field capacity and rainfall is less than or equal to 2 mm. The field capacity threshold and rainfall threshold are obtained from statistics of 200 sets of historical samples. The 40% moisture content is the critical point where crop leaves wilt significantly and the root water absorption rate decreases significantly, and the 2 mm rainfall is the upper limit where the effect on effective soil water replenishment is insignificant. Before the simulation starts, all input features are standardized on a daily scale, and the daily feature values ​​are standardized according to the mean and extreme values ​​of the same period. The difference is dimensionless to ensure the comparability of the influence of different dimensions in the model. On each simulation day, the simulation module executes four sub-steps in sequence. Sub-step 1 is the estimation of root water absorption, which updates the effective water absorption rate of the root system daily based on the combined changes of temperature, humidity, light, wind speed, and rainfall corresponding to the first to Nth principal features, and outputs the volume of water absorbed by the root system per unit time. Sub-step 2 is the estimation of evapotranspiration, which updates the combined influence of leaf evaporation and stomatal conductance daily based on the same features, and outputs the volume of evapotranspiration per unit time. Sub-step 3 is to calculate net available water, subtracting evapotranspiration from root water absorption to obtain the net available water for the day. When this value is less than 0, it is recorded as 0 to prevent non-physical situations of negative water supply. Sub-step 4 is to calculate water use efficiency, using the reciprocal of the ratio of the daily increase in dry matter to the daily total water consumption as the numerical measure of water use efficiency, and recording the calculation result as the daily efficiency value in kilograms per cubic meter. The above four sub-steps are executed in a daily cycle until the 30th day, resulting in an efficiency time series of length 30.To reduce redundancy and improve stability, a pruning method was used to filter out environmental features involved in the calculation. The pruning rule was to apply a positive 10% and a negative 10% perturbation to each principal feature, record the average change in water use efficiency, and identify features with an average change of less than 5% as non-critical features and remove them from the next simulation. This 5% threshold was derived from the statistical distribution of perturbation sensitivity in 500 efficiency time series, and was at the upper limit of the acceptable error range. In the subsequent simulation after pruning, only the remaining features were retained, and the above four sub-steps were repeated to recalculate the 30-day efficiency series. Then, a consistency check was performed on the two series. If the daily average difference between the two series was less than 0.05 and the range difference was less than 0.1, the pruned result was adopted; otherwise, the previous unpruned result was retained to avoid information loss due to oversimplification. For the final determined efficiency time series, fluctuation indicators are calculated, including the range, standard deviation, and the daily average of the rates of change between adjacent days. The daily efficiency value, range, standard deviation, daily average rate of change, time index, corresponding principal feature number, and its daily value are written into the efficiency change dataset. The dataset is stored in a structured table format, with fields including date, stratification number, daily efficiency value, range, standard deviation, daily average rate of change, number of retained principal features, and daily values ​​of each principal feature. The stratification number comes from the fertility stratification identifier that triggered this process. The daily efficiency value is used for subsequent classification and quantification calculations. The range and standard deviation are used to determine the fluctuation amplitude. The daily average rate of change is used to identify persistent trends. The number of retained principal features is used to reconstruct the pruning scale. The daily values ​​of each principal feature are used to trace the causes of efficiency changes. This completes the generation of the efficiency change dataset.

[0045] Subsequently, the system performs fluctuation amplitude analysis on the dataset, calculating its range, standard deviation, and rate of change. If the range exceeds 0.4 or the standard deviation exceeds 0.2, it indicates that the water use efficiency response is unstable, and the system needs to further determine the main influencing factors. These two thresholds are derived from the statistical mean of 500 sets of historical efficiency data set by the model, ensuring timely identification of abnormal fluctuations. At this point, the system calls the support vector machine (SVM) classification algorithm, using environmental features as input and efficiency fluctuation category as output, to perform binary classification modeling. The model uses a radial basis function kernel function, setting the kernel width to 0.5 and the penalty factor to 1. The training samples are all sampling points in the efficiency change dataset. After training, the system outputs the classification boundary and identifies the degree of support of each feature for the boundary construction. The support is calculated as the product of the average boundary distance of the feature within the support vector set and the classification accuracy, and after normalization, the feature importance ranking is obtained.

[0046] The top three environmental factors were extracted as key influencing factors, and their weighted impact on changes in water use efficiency was calculated. The weighting method involved summing the contributions of each factor and dividing by the total, ensuring the normalized sum was 1. The resulting weighted total value was the drought resistance performance quantification value, ranging from 0 to 1. A drought resistance performance quantification value closer to 1 indicates a stronger drought adaptability of winter wheat under current soil fertility and environmental characteristics. Subsequently, the system integrated the drought resistance performance quantification value corresponding to each soil stratum with the nutrient content and principal environmental factor values ​​of that stratum, generating a complete record containing the stratum number, nitrogen, phosphorus, and potassium concentrations, key environmental factor values, simulation time period, efficiency curve trend, and drought resistance performance quantification value. This constructed adaptive distribution data and was written into a database table in a structured data format.

[0047] All recorded data underwent cluster analysis using the K-means clustering algorithm. The number of clusters was set to 3, and the initial cluster centers were set to 0.2, 0.5, and 0.8, representing low, medium, and high drought resistance levels, respectively. Euclidean distance was used as the distance metric during algorithm execution, and convergence was determined by a centroid shift distance of less than 0.001 or a maximum of 100 iterations. After clustering, the system assigned values ​​to the levels based on cluster center size, dividing the drought resistance performance quantification into three intervals and outputting the corresponding results of stratified numbers and drought resistance levels. This constructed the final drought resistance performance level dataset, which serves as the input for subsequent fertilization and irrigation optimization and resource scheduling strategies.

[0048] S4 includes classifying fertilizer application rate and irrigation scheduling using cross-validation based on drought resistance performance quantification values ​​to obtain a classification accuracy dataset. If the fluctuation range of the classification accuracy dataset exceeds a preset threshold, a random forest algorithm is used to rank the feature importance of the combination of fertilizer application rate and irrigation scheduling to determine key resource allocation factors, resulting in a factor ranking dataset. Based on the factor ranking dataset, a weighted average method is used to calculate the contribution weights of fertilizer application rate and irrigation scheduling to water use efficiency, resulting in a weight allocation dataset. If the weight of fertilizer application rate in the weight allocation dataset is higher than that of irrigation scheduling, a linear regression method is used to analyze the relationship between fertilizer application rate and water use efficiency, obtaining fertilizer optimization adjustment parameters. Based on the fertilizer optimization adjustment parameters and environmental factor data, soil fertility data is processed using data standardization methods to generate a resource allocation optimization dataset. Based on the resource allocation optimization dataset, a cluster analysis method is used to group irrigation scheduling schemes to obtain irrigation scheduling optimization parameters. Based on the irrigation scheduling optimization parameters and fertilizer optimization adjustment parameters, preliminary optimization configuration parameters are generated, stored in a structured format, and the final optimization configuration dataset is obtained.

[0049] Using drought resistance performance quantification as the classification label source and fertilizer application rate and irrigation scheduling as independent variables to be classified, a sample set for cross-validation was first constructed. Each sample included drought resistance performance quantification, fertilizer application rate, irrigation cycle, single irrigation volume, irrigation start threshold, irrigation stop threshold, sampling time, soil stratification number, and corresponding water use efficiency. Fertilizer application rate was measured in kilograms per hectare, irrigation cycle in days, single irrigation volume in millimeters, and irrigation start and stop thresholds in percentage of field capacity. Drought resistance performance quantification was a quantitative indicator between 0 and 1 obtained from the previous step. The drought resistance performance quantification was then divided into high-adaptability and low-adaptability classes according to a threshold of 0.6, which was consistent with the previous step and determined as the significant dividing point of adaptability by statistical comparison of yield stability of 200 historical samples.

[0050] Cross-validation classification was then performed, with the samples randomly divided into 10 equal parts and rotated sequentially. In each round, one part was used as the validation set, and the remaining nine parts as the training set. The classifier adopted a deterministic threshold rule: the median fertilizer application rate and the median irrigation cycle within each soil layer were used as the dividing line. If the fertilizer application rate of a single record was not lower than the dividing line and the irrigation cycle was not longer than the dividing line, a high-fit class was output; otherwise, a low-fit class was output. After completing one round of training and validation, the accuracy of that round was calculated. The accuracy was calculated by dividing the number of correctly predicted samples in the validation set by the total number of samples in the validation set and converting it to a percentage. The values ​​are recorded in the classification accuracy dataset. After repeating this process for 10 rounds, 10 accuracy values ​​are obtained, and fluctuation indicators are calculated, including the range and standard deviation. The range is the difference between the maximum and minimum values, and the standard deviation is the average dispersion of the accuracy values ​​from the mean in each round. If the range is greater than 0.1 or the standard deviation is greater than 0.05, the fluctuation is considered to exceed the threshold and the process proceeds to the feature ranking stage. Otherwise, this stage is skipped and the process proceeds directly to the subsequent weight calculation. The above two thresholds are determined by statistically analyzing the distribution of the most recent 200 batches of cross-validation results and taking the values ​​that are located at the upper bound of the stable interval to ensure sufficient sensitivity to abnormal fluctuations.

[0051] When feature ranking is triggered, a random forest classification model is trained. The input features are limited to fertilizer application amount, irrigation cycle, single irrigation water volume, irrigation start threshold, and irrigation stop threshold. The target is the aforementioned high-fitness class and low-fitness class. The number of trees in the forest is set to 200 to reduce variance, the maximum depth is set to 12 layers to avoid overfitting, and the minimum number of leaf node samples is set to 10 to ensure statistical stability. The number of features considered in each split is the square root of the total number of features rounded down to introduce randomness. The importance measure is the average contribution based on the reduction in node purity. After training, the importance score of each feature is obtained and sorted from high to low to form a factor-ranked dataset.

[0052] In the weight calculation stage, the scores belonging to the fertilizer category are summed to obtain the total fertilizer score, and the scores belonging to the irrigation category are summed to obtain the total irrigation score. The minimum and maximum values ​​of the two total scores are mapped to fall between 0 and 1, and the fertilizer application weight and irrigation scheduling weight are obtained according to the rule that the sum of the two is equal to 1, forming a weight allocation dataset. At the same time, the feature sorting snapshot on which the weight generation is based is recorded for traceability.

[0053] If the weight of fertilizer application rate in the weighted dataset is greater than the weight of irrigation scheduling, then a linear regression analysis is performed to determine the relationship between fertilizer application rate and water use efficiency. Specifically, water use efficiency is used as the dependent variable and fertilizer application rate is used as the independent variable. The samples are divided into training and validation sets in an 8:2 ratio, with a fixed random seed to ensure reproducibility. After fitting a straight line to the training set, the slope, intercept, and coefficient of determination are output. If the slope is positive and the coefficient of determination is not less than 0.6, the minimum application rate increment required to achieve a 5% increase in water use efficiency is calculated and recorded as a positive adjustment parameter. If the slope is negative and the coefficient of determination is not less than 0.6, then the minimum application rate increment is calculated. The maximum allowable reduction in water use efficiency, assuming a risk of decline not exceeding 2%, is recorded as a negative adjustment parameter. If the coefficient of determination is less than 0.6, no adjustment is generated, and a recommendation to maintain the status quo is recorded. The aforementioned 0.6 threshold is derived from the lower limit statistics of historical goodness of fit to ensure that the regression explanatory power meets the standard. Subsequently, a resource allocation optimization dataset is generated. The fertilizer optimization adjustment parameters obtained from the regression are merged with environmental factor data from the same time window, and the soil fertility data is standardized. The standardization method is to calculate the mean and standard deviation of nitrogen, phosphorus, and potassium concentrations respectively, and then subtract the mean from the measured value of each record and divide by the standard deviation. To ensure comparability of different units in subsequent clustering, each generated record includes soil stratification number, standardized nitrogen, phosphorus, and potassium values, environmental factor values, fertilizer optimization adjustment parameters, current irrigation parameters, and corresponding water use efficiency. In the irrigation grouping stage, four fields from the resource allocation optimization dataset—irrigation cycle, single irrigation volume, irrigation start threshold, and irrigation stop threshold—are used as clustering features. Cluster analysis is performed with a maximum of 3 clusters. The initial centroids are taken from the 25th, 50th, and 75th quartiles of each feature. The maximum number of iterations is set to 100, and convergence is determined by the distance the centroid moves between two adjacent iterations. The clustering process stops when the value is less than 0.001. After clustering is completed, the four central values ​​of each cluster are output as irrigation scheduling optimization parameters, and the intra-cluster variance is recorded to measure intra-group consistency. Finally, the irrigation scheduling optimization parameters and fertilizer optimization adjustment parameters are merged according to soil stratification number and time to generate a structured record containing stratification number, fertilizer optimization direction, fertilizer adjustment range, irrigation cycle, single water volume, irrigation start threshold, irrigation stop threshold, generation time, and description fields based on weight. This record is stored in the final optimization configuration dataset. At the same time, a joint index of stratification number and generation time is created on the data table to support efficient query and version tracking.

[0054] S5 includes using principal component analysis to extract relevant variables from the initial optimization parameters and generate a variable screening dataset; if the number of variables in the variable screening dataset exceeds a preset threshold, the correlation coefficient matrix between the variables is calculated using correlation analysis to obtain the key variable combination; based on the key variable combination, root development data and nutrient supply data are integrated to generate a comprehensive feature dataset; using the comprehensive feature dataset, a dynamic equilibrium model for resource allocation is constructed using the support vector machine method to obtain a preliminary dynamic model; if the prediction error of the preliminary dynamic model exceeds a preset threshold, the model parameters are optimized using the gradient descent method to obtain an optimized dynamic model; based on the optimized dynamic model, combined with root development characteristics and nutrient supply characteristics, resource allocation adjustment parameters are generated to determine the final dynamic equilibrium scheme.

[0055] In this step, all numerical fields in the initial optimization parameters are first standardized. The standardization method is to use the mean and standard deviation of the field in the current data window to remove the mean and scale, so as to eliminate the difference in units and provide stable input for subsequent principal component extraction.

[0056] Subsequently, principal component analysis (PCA) is performed to extract relevant variables from the standardized parameters. The specific process involves calculating the covariance matrix and sorting the variables by eigenvalue from largest to smallest, accumulating the variance contribution rate sequentially. When the cumulative variance contribution rate first reaches or exceeds 85%, extraction stops, and the high-loaded original variables of the corresponding principal component are added to the variable screening dataset. The criteria for high loading are that the absolute loading value of the current principal component ranks among the top 3 of all variables of that principal component and the absolute loading value is not less than 0.5. This dual criterion is determined by obtaining a balance between interpretability and redundancy in the interpretability assessment of 300 sets of historical parameters. After variable screening, the number of variables is counted. If the number exceeds a preset threshold of 5, correlation analysis is performed to further reduce redundancy. The preset threshold of 5 is determined based on the cross-validation results of the number of variables and model stability in 300 sets of historical samples. When the number of variables exceeds 5, the generalization performance of the support vector machine begins to decline significantly, so 5 is set as the upper limit.

[0057] Correlation analysis was performed using the Pearson correlation coefficient as a metric. A correlation coefficient matrix was calculated between variables, and the variables were filtered based on their absolute values. A pair of variables with an absolute correlation coefficient of at least 0.7 was considered highly correlated, and only the variable with the larger sum of its overall absolute loadings on the principal component set was retained. The threshold of 0.7 was derived from the critical point for stability and redundancy removal in a structural test of 300 samples. After obtaining the key variable combinations, root development data and nutrient supply data were strictly integrated according to time index and soil stratification number. The root development data fields were fixed as average root length, maximum root depth, and root density, in centimeters, millimeters, and root density per square centimeter, respectively. The nutrient supply data fields were fixed as nitrogen supply rate, phosphorus supply rate, and potassium supply rate, in milliliters. For each kilogram per day, the alignment rule is as follows: when multiple records exist for the same stratum number and on the same day, a time-weighted average is taken. If the proportion of missing records is no higher than 5, the average of the two days before and after the adjacent day is used to fill in the gaps; if it exceeds 5, the record is removed. This process generates a comprehensive feature dataset containing key variable combinations and six physiological and nutritional indicators. Using this comprehensive feature dataset as the sole training input, a support vector machine (SVM) method is employed to construct a dynamic equilibrium model for resource allocation. The model output is a discriminant value indicating whether resource allocation has reached dynamic equilibrium, along with the corresponding equilibrium score. The radial basis function (RBF) kernel is used, with a kernel width of 0.5 and a penalty factor of 1. Both are selected through a grid search using 10-fold cross-validation on a discrete grid with kernel widths ranging from 0.1 to 1 and penalty factors ranging from 0.1 to 10. The minimum validation error point is determined, and the training and validation sets are divided in an 8:2 ratio with a fixed random seed to ensure reproducibility. After training, the mean squared error is calculated using the validation set as the prediction error. If the prediction error exceeds a preset threshold of 0.1, gradient descent is initiated to optimize the model parameters. The threshold of 0.1 is derived from the minimum upper bound separating the stable and unstable performance segments in the error distribution statistics of 200 samples. The learning rate of gradient descent is fixed at 0.01, and the convergence speed and overshoot risk are compared through cross-validation in the range of 0.001 to 0.05 to determine the compromise point. The maximum number of iterations is set to 500 to limit computational cost. The convergence criterion is that the validation error decreases by less than 0.001 in two consecutive iterations, at which point the iterations stop. The direction of parameter updates is determined according to the validation error. The proof error is calculated in the opposite direction of the numerical gradients of the kernel width and the penalty factor. The numerical gradients are calculated using small perturbation differences without changing the model structure. After the optimization is completed and the optimized dynamic model is obtained, the resource allocation adjustment parameter generation step is entered. The optimization dynamic model's balance score on the comprehensive feature dataset under the current stratification number and time index is used as the target. The sensitivity of the score to local changes in three adjustable resource quantities, namely fertilizer application rate, irrigation cycle and single irrigation water volume, is calculated using the numerical perturbation method. The perturbation amplitude is fixed at 5% of their respective historical median values, and the three sensitivities are obtained respectively. Then, three correction values ​​are generated according to the consistency of sensitivity and directionality. The rule for determining the size of the correction values ​​is to raise the balance score to no less than 0 without crossing the physiological safety boundary.The minimum step size is 8. The physiological safety boundaries are: fertilizer application rate not exceeding the 90th percentile and not lower than the 10th percentile of the current stratum's historical range; irrigation cycle not less than 3 days and not longer than 30 days; and single irrigation water volume not less than 5 mm and not more than 40 mm. The scoring target of 0.8 is determined by the minimum compliance score obtained from the joint quantile analysis of yield stability and water use efficiency of the compliant samples. The three correction values, along with the average root length, maximum root depth, root density, and three nutrient supply rates, are recorded as resource allocation adjustment parameters. The model's determination of equilibrium or non-equilibrium is written together with the equilibrium score. Finally, the final dynamic equilibrium scheme is aggregated according to the stratum number and time index. The scheme is saved in a structured format, and a joint index is created for the stratum number and time index to support fast querying and tracing.

[0058] S6 includes obtaining yield stability indicators through a dynamic equilibrium model, calculating the deviation between the indicators and preset thresholds using statistical analysis methods to obtain a deviation dataset; updating irrigation interval parameters using a gradient descent method based on the deviation dataset, and generating an adjusted irrigation parameter set by combining environmental adaptability data; obtaining fertilizer ratio parameters using the adjusted irrigation parameter set, predicting the effect of parameter adjustment using a linear regression method, and obtaining a prediction error value; if the prediction error value exceeds a preset threshold, adjusting the gradient descent step size based on convergence conditions to generate an optimized irrigation parameter set; obtaining agricultural resource allocation data based on the optimized irrigation parameter set, prioritizing resource allocation using cluster analysis to obtain a resource allocation scheme; generating a final resource allocation scheme by combining the resource allocation scheme with environmental adaptability data; and obtaining yield stability indicators based on the final resource allocation scheme, determining whether a preset threshold has been reached, and obtaining verification results.

[0059] In this step, for each stratum number and time index, a 30-day daily-scale yield series is extracted from the dynamic equilibrium model. The daily mean, daily standard deviation, coefficient of variation, and low-yield anomaly ratio of this series are calculated. The coefficient of variation is defined as the ratio of the standard deviation to the mean, and the low-yield anomaly ratio is defined as the percentage of days in which the yield is lower than the mean minus twice the standard deviation. The coefficient of variation is weighted with a 70% inverse weight and the low-yield anomaly ratio is weighted with a 30% inverse weight, and the resulting yield stability index score is calculated to be between 0 and 1. A preset threshold of 0.8 is used, and the value of 0.8 is based on the 20 The minimum achievement score is obtained by combining the joint quantile statistics of yield stability and subsequent achievement rate in group 0 historical samples. Then, the absolute difference between the yield stability index score and the threshold is calculated for each record as the deviation value, generating a deviation dataset containing four fields: stratification number, time index, yield stability index score, and deviation value. Based on the deviation dataset, a gradient descent update process for the irrigation interval parameter is initiated. The initial irrigation interval is taken from the interval days in the current execution plan, with the value boundaries being a minimum of 3 days and a maximum of 30 days. The initial step size is 2 days, and the direction determination uses the finite difference sensitivity method, dividing the interval without exceeding the limits. Do not shorten the irrigation interval by one day and then extend it by one day, and recalculate the new 30-day yield stability index score using the dynamic equilibrium model. Compare the difference between the two scores and the original score, and select the direction that increases the score as the update direction. To reflect the constraint of environmental differences on the adjustment range, introduce the drought resistance performance quantification value from the environmental adaptability data and map the step size to three levels of correction coefficients. The correction coefficient is 1.2 when the drought resistance performance quantification value is less than 0.4, 1.0 when it is between 0.4 and 0.7, and 0.8 when it is greater than or equal to 0.7. These three thresholds are derived from 200 The stability of historical samples under different drought resistance levels is stratified statistically, which can achieve a balance between avoiding excessive fluctuations and effectively improving the situation. The initial step size is multiplied by a correction coefficient to obtain the actual update step size, and the irrigation interval is adjusted along a determined direction. If the value boundary is touched, it is truncated at the boundary. After completing one update, the yield stability index score and deviation are recalculated and written to the adjustment log. The iteration continues until any convergence condition is met. The convergence condition includes three consecutive iterations in which the deviation decreases by less than 0.01 or the cumulative iteration reaches 20. If the deviation increases for two consecutive times, the current step size is multiplied by 0.After step 5, iterations continued. The step size, boundary, and convergence threshold were all determined by comparative experiments on the convergence speed and overshoot risk of historical samples, achieving stable improvement without increasing computational costs excessively. After obtaining the adjusted irrigation parameter set, fertilizer ratio parameters were obtained according to stratum number and production season. The rule for determining fertilizer ratio parameters was to prioritize the median values ​​of nitrogen, phosphorus, and potassium fertilizer ratios from historical execution records of the same stratum and season. If the historical records for a stratum were less than 10, the latest ratio from the previous final optimized configuration dataset was used as a benchmark and weighted and smoothed with the existing execution ratios of the current season at 70% and 30% weights to reduce random fluctuations. Subsequently, a linear regression model was established to predict the effect of parameter adjustment, using water use efficiency as the benchmark. The dependent variable is the irrigation interval and the proportions of nitrogen, phosphorus, and potassium fertilizers as independent variables. An 8:2 training-to-validation ratio is used, with a fixed random seed to ensure reproducibility. The prediction error is calculated on the validation set, defined as the average of the absolute values ​​of the differences between the predicted and observed water use efficiency. A preset error threshold of 0.1 is used, derived from the upper bound of the stable interval of 200 historical validation error distributions. If the prediction error is less than 0.1, the current parameters are accepted and the resource allocation phase begins. If the prediction error exceeds 0.1, the adaptive step size adjustment phase begins, correcting the gradient descent step size based on the convergence condition: the decrease in prediction error between two consecutive regressions is not less than 0.01. If the condition is met three times consecutively, and if it is not met and the error increases, the step size is multiplied by 0.5. If the error neither decreases nor increases, the step size remains unchanged. The step size corresponds to the daily adjustment range of the irrigation interval and is limited to between 1 and 7 days. After adaptive adjustment, the irrigation interval update and regression evaluation are re-executed, with a maximum of 5 rounds to limit computational overhead. The optimized irrigation parameter set is output. Based on the optimized irrigation parameter set, agricultural resource allocation data is obtained. The agricultural resource allocation data includes two main fields: irrigation interval days and the proportion of three fertilizers, as well as two management fields: hierarchical number and time index. After range standardization of the above fields, cluster analysis is performed to classify resource allocation priorities. The number of clusters is fixed at 3, and the initial values ​​are the 25th and 50th quartiles of each field. Using the 75th quantile as the initial center for three groups, standardized Euclidean distance was used as the similarity metric, with a maximum of 100 iterations and a stopping threshold of 0.001 for center movement. After clustering, the three clusters were mapped to priority 1, priority 2, and priority 3 according to the weighted score of the average yield stability index and average water use efficiency of each cluster, from high to low. The weights were 70% for the yield stability index score and 30% for water use efficiency, forming a resource allocation scheme. The resource allocation scheme was combined with environmental adaptability data to generate the final resource allocation scheme. The combination rule was that when the drought resistance performance quantification value of a stratum was less than 0.4, the priority of the corresponding cluster in that stratum was increased by one level but not exceeding priority 1; when the drought resistance performance quantification value of a stratum was between 0.4 and 0, the priority was increased.When the cluster priority is between 7, a 10% resource allocation is given to each of the irrigation interval and nitrogen fertilizer ratio. When the stratified drought resistance performance quantification value is greater than or equal to 0.7, the original scheme is maintained without additional allocation. Finally, this scheme is input back into the dynamic equilibrium model to generate a new 30-day yield sequence. The new yield stability index score is calculated using the same method as in the first paragraph and compared with the threshold of 0.8. If the score is greater than or equal to 0.8, the verification result is recorded as passed; if the score is less than 0.8, it is recorded as failed, and the current parameter set and corresponding deviation value are retained for use in the next iteration. The records of passed and failed are stored in a structured form, and a joint index is created on the stratification number and time index.

[0060] S7 includes: acquiring validation data under drought conditions from historical databases; extracting feature datasets related to the current resource allocation scheme using data filtering methods; calculating the matching degree between the resource allocation scheme and the validation data using the feature dataset and a variable step size adjustment method; obtaining matching degree analysis results if the matching degree analysis results are lower than a preset threshold; adjusting the variable step size through a simulation feedback loop to generate optimized matching degree analysis results; prioritizing resource allocation using cluster analysis based on the optimized matching degree analysis results to obtain a resource allocation priority dataset; generating an adjusted resource allocation scheme by combining the resource allocation priority dataset with the constraints of the drought environment; evaluating yield stability using a simulation feedback loop based on the adjusted resource allocation scheme to obtain yield stability evaluation results; and determining whether the yield stability evaluation results meet the preset threshold to generate the final resource allocation scheme.

[0061] First, validation data under arid conditions was retrieved from historical databases. The arid conditions were defined as daily rainfall less than or equal to 2 mm and soil moisture content below 40% of field capacity. This threshold was standardized using the same criteria from previous systems to ensure consistency. To ensure comparability, only records from the same crop growth stage, soil fertility stratification, and unit of measurement within the most recent three years were retained, consistent with the current resource allocation scheme. The selected fields were fixed as total fertilizer application, nitrogen fertilizer ratio, phosphorus fertilizer ratio, potassium fertilizer ratio, irrigation interval days, single irrigation volume, average root depth, average leaf moisture content, water use efficiency, and time index. Data exceeding the specified agricultural parameters were deleted. Outliers within the reasonable range were identified, specifically: total fertilizer application rate (0-200 kg / ha), nitrogen fertilizer ratio (0-60%), phosphorus fertilizer ratio (0-40%), potassium fertilizer ratio (0-40%), irrigation interval (3-30 days), single irrigation water volume (5-40 mm), root depth (10-500 mm), leaf moisture content (20-90%), and water use efficiency greater than 0. Fields with a missing proportion not exceeding 5 were imputed using the average of the two consecutive days before and after the missing value; records with a missing proportion exceeding 5 were removed. This process generated a feature dataset. Subsequently, the matching degree between the current resource allocation scheme and each record in the feature dataset was calculated. The calculation order was: first, the absolute value of the difference between each pair of parameters was calculated, then... The single-parameter bias value is obtained by normalizing the range of this parameter in the feature dataset. The closer the single-parameter bias value is to 0, the better the match. All single-parameter bias values ​​are weighted and averaged to obtain the comprehensive bias. The matching degree is then obtained by subtracting the comprehensive bias from 1, with the matching degree ranging from 0 to 1. The weights are the values ​​given in the dataset above, with the weights related to fertilizer application and irrigation determined separately according to this dataset. If the fertilizer-related weights need to be allocated among nitrogen, phosphorus, and potassium, they are allocated proportionally to avoid bias. After completing all matching degree calculations, the matching degree analysis results are obtained, and a preset threshold of 0.7 is used for judgment. The threshold is derived from the smallest upper bound separating reliable and unreliable categories in the distribution statistics of 200 historical matching records. If the matching degree of any record is lower than 0.7, a simulation feedback loop is entered to adaptively adjust the step size of the variables to improve the matching degree. The adaptive rule sets the initial step size of each adjustable parameter to 5 times the historical value range of that parameter, such as irrigation interval in days, single water volume in millimeters, and the ratio of three fertilizers in percentages, and limits the boundary to the above agronomic range. If the matching degree does not improve for two consecutive rounds, the step size of the parameter is multiplied by 0.5. If the matching degree improves by more than or equal to 0.05 for two consecutive rounds, the step size is multiplied by 1.2. Each round of updates is executed one parameter at a time, in the following order: irrigation interval, single irrigation water volume, nitrogen fertilizer ratio, phosphorus fertilizer ratio, and potassium fertilizer ratio. The update direction aims to reduce overall deviation, i.e., within the boundary, one step is attempted in both the positive and negative directions, and the new matching degree is calculated. The step with the higher matching degree is selected as the actual update. If the boundary is touched, it is truncated at the boundary. After completing one round, the matching degree of all records is recalculated, and the round number, step size, and matching degree are recorded. This process is repeated until the matching degree of all records is not lower than 0.7 or reaches the upper limit of a maximum of 20 rounds, forming the optimized matching degree analysis result. After obtaining the optimized matching degree analysis result, the resource allocation priority is divided, and the mean iterative clustering method is used to... All schemes are clustered based on two types of features: matching degree and key parameter deviation. The key parameter deviation is a weighted average of the deviations of irrigation interval and single water volume, with equal weights for both. The number of clusters is fixed at 3 to correspond to high, medium, and low priorities. The initial centers are taken from the 25th, 50th, and 75th quantiles of the two types of features. The distance metric is the Euclidean distance after range standardization of the two types of features. The maximum number of iterations is 100. The convergence criterion is that the distance between two adjacent center shifts is less than 0.001. After clustering, each cluster is assigned priority 1, priority 2, and priority 3 in descending order of average matching degree. The resource allocation priority dataset is output. Based on this, an adjusted resource allocation scheme is generated by overlaying drought environmental constraints. The constraints are fixed as follows: single irrigation water volume not exceeding 40 mm, irrigation interval not less than 3 days, nitrogen fertilizer ratio not exceeding 60%, and total fertilizer application not exceeding 200 kg / ha. If any parameter exceeds the limit, it is adjusted to the allowable boundary while the remaining parameters remain unchanged to reduce linkage error. Subsequently, the yield stability of the adjusted resource allocation scheme is assessed. The assessment method is to input the scheme back into the dynamic equilibrium model to generate a 30-day daily-scale yield sequence and calculate the yield stability index. This index is obtained by weighting the coefficient of variation and the proportion of low-yield anomalies with inverse weights of 70% and 30%, respectively. The yield stability assessment results are then obtained. Compared to a threshold of 0.8, which is derived from the joint quantile statistics of stability and achievement rate of 200 historical samples; if the yield stability index of any scheme is lower than 0.8, fine-tuning is performed in the simulation feedback loop with a daily equivalent adjustment range of variable step size of 2. Priority is given to trying one positive and one negative minimum step in irrigation interval and single water volume. If it still does not meet the standard, then the proportions of the three fertilizers are adjusted by one minimum step. The yield stability index is recalculated immediately after each fine-tuning. A maximum of 10 rounds of fine-tuning are allowed. If it still does not meet the standard, the scheme is marked as unsuccessful and its deviation and step size trajectory are retained for the next evaluation cycle; when the yield stability index of all schemes is greater than or equal to 0.At 8:00 AM, if the preset threshold is met, the system outputs the final resource allocation plan in a structured format. Record fields include layer number, time index, nitrogen fertilizer ratio, phosphorus fertilizer ratio, potassium fertilizer ratio, irrigation interval, single water volume, matching degree, priority, yield stability indicators, and the values ​​and sources of all key thresholds and parameters. A composite index is also created on the layer number and time index to support fast querying and audit traceability.

[0062] S8 includes obtaining contribution coefficients from the soil fertility stratification database based on yield stability assessment results, updating the contribution coefficients using a data fusion method to obtain an updated contribution coefficient dataset; calculating the weight allocation of the resource allocation scheme using a linear regression algorithm based on the updated contribution coefficient dataset and proportional balance constraints to obtain the weight allocation result; if the weight allocation result meets preset constraints, extracting candidate input datasets for the next assessment cycle using a data filtering method to obtain a candidate input dataset; prioritizing the data using a clustering analysis method based on the candidate input dataset to obtain a priority ranking dataset; adjusting the input dataset using an iterative optimization method if the matching degree between the priority ranking dataset and the resource allocation scheme is lower than a preset threshold to obtain an optimized input dataset; and generating the input dataset for the next assessment cycle based on the optimized input dataset and the constraints of the assessment cycle.

[0063] This step first reads the yield stability index score for each stratum number and time index. This score is obtained from previous steps by weighting the coefficient of variation and the proportion of low-yield anomalies, with a value range of 0 to 1. Simultaneously, the existing contribution coefficients for the same stratum are read from the soil fertility stratification database, with values ​​ranging from 0 to 1. Then, data fusion is performed to update the contribution coefficients. Specifically, the yield stability index score is first mapped to an evidence coefficient, and a fusion weight is set accordingly. The mapping rule is that when the yield stability index score is greater than or equal to 0.9, the evidence coefficient is set to 1.0 and the fusion weight to 0.7; when the yield stability index score is between 0.8 and 0.9, the evidence coefficient is set to 0.9 and the fusion weight to 0.7. The fusion weight is set to 0.6. When the production stability index score is between 0.7 and 0.8, the evidence coefficient is set to 0.8 and the fusion weight to 0.5. When the production stability index score is between 0.6 and 0.7, the evidence coefficient is set to 0.6 and the fusion weight to 0.4. When the production stability index score is less than 0.6, the evidence coefficient is set to 0.5 and the fusion weight to 0.3. These five values ​​are based on the reliability stratification statistics of 200 historical samples, ensuring that high-stability records have a higher weight in the fusion and that low-stability records have limited impact on the historical coefficient. For each stratum, a more accurate fusion weight is generated by "weighting the evidence coefficient according to the fusion weight and the old contribution coefficient according to the remaining weight". The new contribution coefficients are used to form an updated contribution coefficient dataset, and all stratified coefficients are truncated to the range of 0 to 1 to avoid exceeding the bounds. Next, the weight allocation of the resource allocation scheme is calculated under the "proportional balance constraint," which stipulates that the sum of the weights of all resources equals 1 and the weight of each item is between 0.2 and 0.6. This range is determined by statistically analyzing the marginal contribution distribution of water and fertilizer to water use efficiency in the implementation schemes of the past three seasons, with lower and upper limits set to avoid extreme resource bias. The weight allocation is calculated using a linear regression algorithm, with water use efficiency as the dependent variable and three types of resource parameters as independent variables: fertilizer application rate, irrigation interval, and single irrigation water volume. To enhance the coupling with soil conditions, the updated contribution coefficients are used as sample weights in the fitting process before regression modeling, meaning that samples with higher updated contribution coefficients are given greater influence. After fitting, the absolute values ​​of the three regression coefficients are taken as the initial importance and mapped from minimum to maximum to 0 to 1. Then, the three importance values ​​are normalized and truncated according to the proportional balance constraint to obtain the weight allocation result consisting of fertilizer weight, irrigation interval weight, and single water volume weight. If the weight allocation result satisfies the proportional balance constraint and the range of the three weights does not exceed 0.4, the next step of data screening is performed; otherwise, the one with the highest importance is set to not exceed 0.6. The upper limit is reduced and the other two items are expanded proportionally until the constraints are met before proceeding to the next step; then, data filtering is performed to form the candidate input dataset for the next evaluation cycle. The filtering rules are to select records from the most recent two seasons that are consistent with the current stratification, have the same units, and have all fields complete, and to remove outliers that exceed the reasonable agronomic range. The range is fixed as follows: total fertilizer application rate 0 to 200 kg per hectare, nitrogen fertilizer ratio 0 to 60%, phosphorus fertilizer ratio 0 to 40%, potassium fertilizer ratio 0 to 40%, irrigation interval 3 to 30 days, single irrigation water volume 5 to 40 mm, and water use efficiency greater than 0; for fields with a missing ratio of no more than 5%, the average of the two consecutive days is used to fill in the missing data, and records with a missing ratio of more than 5% are deleted. After completion, a candidate input dataset is obtained. Based on this dataset, clustering analysis is used to prioritize the data. The clustering features are the updated contribution coefficient, water use efficiency, and the combined value of the parameter differences between the current resource allocation scheme and the updated contribution coefficient. The combined value of the parameter differences is the weighted average of the differences in irrigation intervals and single water volume. The number of clusters is set to 3. The initial centers are taken from the 25th, 50th, and 75th percentiles of the above three features. The distance metric is the Euclidean distance after range standardization. The maximum number of iterations is 100, and the convergence threshold is a center shift of less than 0.001. After clustering, each cluster is assigned a priority of 1 from high to low based on the weighted score of the average updated contribution coefficient and the average water use efficiency. Priority 1, Priority 2, and Priority 3 are assigned weights of 70% for the updated contribution coefficient and 30% for water use efficiency, resulting in a priority-ranked dataset. The matching degree between the priority-ranked dataset and the current resource allocation scheme is then calculated. The matching degree is defined as the weighted sum of the inverted parameter difference composite value and the updated contribution coefficient, ranging from 0 to 1, with a threshold of 0.75. This threshold is derived from a reliable boundary determined by matching the distribution of samples from 200 historical schemes verified through yield stability. If the matching degree is below 0.75, iterative optimization is performed on the candidate input dataset to improve the matching degree. The optimization method involves fine-tuning the input data parameter by parameter, with an initial step size of 5% of the historical range of that parameter. If the matching degree does not improve for two consecutive rounds, the step size is multiplied by 0.5. If the matching degree improves by more than or equal to 0.05 for two consecutive rounds, the step size is multiplied by 1.2. The adjustment order is irrigation interval, single water volume, fertilizer application amount, and the ratio of the three fertilizers. The direction selection is based on improving the matching degree and is truncated within the agronomic boundary. The maximum number of iterations is 20 rounds, and the optimized input dataset is output. Finally, based on the optimized input dataset and combined with the constraints of the evaluation cycle, the input dataset for the next evaluation cycle is generated. The evaluation cycle constraints are: the time window length is fixed at 30 days, the number of records in each layer is not less than 10, the ratio of the maximum to the minimum sample size in each layer does not exceed 3, and the contribution coefficient after any layer update is less than 0.At 4 o'clock, records from neighboring time periods within the same stratum need to be added until the minimum number of records is met. If this is not possible, the stratum will be marked as a priority sampling stratum for the next cycle. Each record in the generated dataset includes a stratum number, a time index, an updated contribution coefficient, total fertilizer application amount and nitrogen, phosphorus, and potassium ratio, irrigation interval and single irrigation volume, water use efficiency, and data source identifier. A joint index is created on the stratum number and time index to support fast retrieval and traceability.

[0064] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data processing method for evaluating winter wheat field soil fertility based on comprehensive weight calculation, characterized in that, The method includes the following steps: S1. Nutrient content and water retention data are collected from field soil through a sensor network. The collected nutrient content and water retention data are grouped and processed by decision node splitting and entropy calculation to obtain soil fertility stratification results. S2. Based on the soil fertility stratification results, the root depth and leaf moisture content of winter wheat in different soil strata were obtained from the crop monitoring equipment. The correlation strength between root depth and leaf moisture content and fertility stratification was analyzed using leaf node classification and feature selection criteria. The contribution weight of each soil fertility stratum to root depth and leaf moisture content was calculated, and the fertility contribution coefficient was determined. S3. If the fertility contribution coefficient exceeds the preset threshold, the change in water use efficiency of winter wheat under drought conditions is calculated by combining the simulation module with the pruning treatment method to obtain the quantitative value of drought resistance performance. S4. Based on the quantified drought resistance performance, cross-validation evaluation and classification accuracy are used to classify and optimize fertilizer application rate and irrigation scheduling. The weighted average method is used to calculate the contribution weight of fertilizer application rate and irrigation scheduling to water use efficiency, determine the adjustment direction of resource allocation scheme, and obtain preliminary optimized configuration parameters. S5. Extract relevant variables from the initial optimized configuration parameters, integrate root development data and nutrient supply data using initial parameter settings and error function definitions, and determine the dynamic balance model of resource allocation.

2. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 1, characterized in that: S1 includes: Nutrient content and moisture data from multiple depths of soil in the field are acquired through a sensor network and stored as a structured dataset to obtain raw soil data. Data cleaning techniques were used to preprocess the raw soil data, removing outliers and missing values ​​to obtain cleaned soil data. Based on the cleaned soil data, feature values ​​of nutrient content and moisture data are extracted, feature vectors are constructed, and a soil feature set is obtained. The soil feature set was split using a decision tree algorithm, and the splitting nodes were determined based on entropy values ​​to obtain preliminary grouping results. If the entropy value of the initial grouping result is lower than the preset threshold, the groups are merged to optimize the grouping structure and obtain the optimized grouping result. The optimized grouping results were sorted by hierarchical analysis, and soil fertility stratification was determined by combining nutrient content and moisture data to obtain the final stratification results. Based on the final stratification results, soil fertility stratification data is generated and stored in a queryable format to obtain the soil fertility stratification dataset.

3. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 1, characterized in that: S2 includes: Data on root depth and leaf moisture content of winter wheat were obtained from different soil fertility strata using crop monitoring equipment and stored as a structured dataset to obtain raw crop data. Data preprocessing techniques are used to clean the raw crop data, removing outliers and missing values ​​to obtain cleaned crop data. Based on the cleaned crop data, feature values ​​of root depth and leaf moisture content are extracted, feature vectors are constructed, and crop feature set is obtained. The random forest algorithm was used to classify the crop feature set. Based on information gain, the importance of features was calculated to determine the correlation strength between root depth and leaf moisture content and soil fertility stratification, and preliminary correlation results were obtained. If the classification accuracy of the initial association results is lower than the preset threshold, the feature vectors are re-selected using feature selection criteria, the random forest model is optimized, and an optimized association result is obtained. By analyzing and optimizing the correlation results, the contribution weight of each soil fertility layer to root depth and leaf water content was calculated, and the fertility contribution coefficient was obtained. Based on the fertility contribution coefficient, growth adaptability data of winter wheat in different soil fertility strata are generated and stored in a queryable format to obtain the final growth adaptability dataset.

4. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 1, characterized in that: S3 includes: If the fertility contribution coefficient exceeds the preset threshold, the soil fertility data and environmental factor data are obtained from the winter wheat planting area through the data acquisition module to obtain the initial environmental dataset. Based on the initial environmental dataset, principal component analysis was used to extract the main feature vectors of soil fertility and environmental factors, resulting in an environmental feature set. By loading the environmental feature set through the simulation module and combining it with the pruning treatment method, the changing trend of water use efficiency of winter wheat under drought conditions was calculated, and the efficiency change dataset was obtained. If the fluctuation range of the efficiency change dataset exceeds the preset range, the efficiency change dataset is classified by the support vector machine algorithm to determine the key influencing factors of water use efficiency and obtain the classification results. Based on the classification results, the weighted average method was used to calculate the contribution weight of the influencing factors to water use efficiency, and the quantitative value of drought resistance performance was obtained. By quantifying drought resistance performance and combining it with soil fertility data, we generate adaptive distribution data of winter wheat under different drought conditions, store it in a structured format, and obtain the final adaptive dataset. Based on the final adaptive dataset, cluster analysis was used to classify the drought resistance performance of winter wheat, resulting in drought resistance performance level data.

5. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 1, characterized in that: S4 includes: By using the quantified drought resistance performance values, cross-validation was employed to classify fertilizer application rates and irrigation scheduling, resulting in a classification accuracy dataset. If the fluctuation of the classification accuracy dataset exceeds the preset threshold, the combination of fertilizer application and irrigation scheduling is ranked by feature importance using the random forest algorithm to determine the key resource allocation factors and obtain the factor-ranked dataset. Based on the factor-ranked dataset, the weighted average method was used to calculate the contribution weights of fertilizer application rate and irrigation scheduling to water use efficiency, resulting in a weighted dataset. If the fertilizer application rate weight in the weighted dataset is higher than the irrigation scheduling weight, then the relationship between fertilizer application rate and water use efficiency is analyzed by linear regression to obtain fertilizer optimization adjustment parameters. Based on fertilizer optimization adjustment parameters and combined with environmental factor data, soil fertility data is processed using data standardization methods to generate a resource allocation optimization dataset. By optimizing the resource allocation dataset, the irrigation scheduling schemes are grouped using cluster analysis to obtain the irrigation scheduling optimization parameters. Based on the irrigation scheduling optimization parameters and fertilizer optimization adjustment parameters, preliminary optimization configuration parameters are generated, stored in a structured format, and the final optimization configuration dataset is obtained.

6. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 1, characterized in that: S5 includes: Principal component analysis was used to extract relevant variables from the initial optimization parameters to generate a variable screening dataset. If the number of variables in the variable selection dataset exceeds a preset threshold, the correlation coefficient matrix between the variables is calculated using correlation analysis to obtain the key variable combination. Based on the combination of key variables, root development data and nutrient supply data are integrated to generate a comprehensive feature dataset; By integrating the feature dataset, a dynamic equilibrium model for resource allocation is constructed using the support vector machine method, resulting in a preliminary dynamic model. If the prediction error of the initial dynamic model exceeds a preset threshold, the model parameters are optimized using the gradient descent method to obtain an optimized dynamic model. Based on the optimized dynamic model, combined with root development characteristics and nutrient supply characteristics, resource allocation adjustment parameters are generated to determine the final dynamic balance scheme.

7. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 1, characterized in that, It also includes S6. If the dynamic equilibrium model shows that the yield stability index is lower than the target level, the irrigation interval and fertilizer ratio are adjusted by updating the gradient descent and judging the convergence condition to obtain the final resource allocation scheme.

8. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 7, characterized in that, S6 specifically includes: The dynamic equilibrium model is used to obtain the output stability index, and the statistical analysis method is used to calculate the deviation between the index and the preset threshold to obtain the deviation dataset. Based on the deviation dataset, the irrigation interval parameters are updated using the gradient descent method, and combined with environmental adaptability data, an adjusted irrigation parameter set is generated. By obtaining fertilizer ratio parameters through the adjusted irrigation parameter set, the effect of parameter adjustment is predicted using the linear regression method, and the prediction error value is obtained. If the prediction error exceeds the preset threshold, the gradient descent step size is adjusted based on the convergence condition to generate an optimized irrigation parameter set. Based on the optimized irrigation parameter set, agricultural resource allocation data is obtained, and cluster analysis is used to classify resource allocation priorities to obtain a resource allocation scheme. By combining the resource allocation plan with environmental adaptability data, a final resource allocation plan is generated. Based on the final resource allocation plan, obtain the output stability index, determine whether it has reached the preset threshold, and obtain the verification results.

9. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 8, characterized in that, It also includes S7, which involves obtaining validation data on drought conditions from historical databases based on the final resource allocation plan, and using variable step size adjustment and simulation feedback loops to determine the applicability of the plan and obtain yield stability assessment results. Specifically, S7 includes: Validation data under drought conditions were obtained from historical databases, and feature datasets related to the current resource allocation scheme were extracted using data filtering methods to obtain the feature datasets. By using the feature dataset and the variable step size adjustment method, the matching degree between the resource allocation scheme and the validation data is calculated, and the matching degree analysis results are obtained. If the matching degree analysis result is lower than the preset threshold, the variable step size is adjusted through simulation feedback loop to generate an optimized matching degree analysis result; Based on the optimized matching degree analysis results, cluster analysis is used to divide the priority of resource allocation, resulting in a resource allocation priority dataset; By using a resource allocation priority dataset and considering the constraints of an arid environment, an adjusted resource allocation scheme is generated. Based on the adjusted resource allocation plan, a simulated feedback loop is used to evaluate the output stability, and the output stability evaluation results are obtained. Based on the production stability assessment results, it is determined whether the preset threshold is met, and the final resource allocation plan is generated.

10. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 9, characterized in that, It also includes S8, which updates the contribution coefficients in the soil fertility stratification database based on the yield stability assessment results, and determines the input data for the next assessment cycle by combining the proportional balance constraints and the final scheme output. Specifically, S8 includes: Based on the yield stability assessment results, the contribution coefficients in the soil fertility stratification database are obtained, and the contribution coefficients are updated using a data fusion method to obtain the updated contribution coefficient dataset. Based on the updated contribution coefficient dataset and combined with the proportional balance constraint, the weight allocation of the resource allocation scheme is calculated using a linear regression algorithm to obtain the weight allocation result. If the weight allocation result meets the preset constraints, then the candidate input dataset for the next evaluation period is extracted through data filtering methods to obtain the candidate input dataset.

11. The data processing method for winter wheat field soil fertility evaluation based on comprehensive weight calculation according to claim 10, characterized in that, S8 further includes: Based on the candidate input dataset, cluster analysis is used to prioritize the data, resulting in a priority-ranked dataset. If the matching degree between the priority-ranked dataset and the resource allocation scheme is lower than a preset threshold, the input dataset is adjusted through an iterative optimization method to obtain an optimized input dataset. Based on the optimized input dataset and the constraints of the evaluation cycle, the input dataset for the next evaluation cycle is generated.

Images