A water and fertilizer integrated decision-making method and device based on multi-source data fusion and random forest, electronic equipment and storage medium
By using multi-source data fusion and random forest algorithms, the problem of inaccurate prediction of water and fertilizer requirements in existing water and fertilizer management is solved, enabling accurate prediction of crop water and fertilizer requirements, improving the precision of irrigation and fertilization, and making it suitable for modern agriculture.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INST OF DRY LAND FARMING SHANXI ACAD OF AGRI SCI
- Filing Date
- 2026-03-06
- Publication Date
- 2026-06-05
AI Technical Summary
Existing water and fertilizer management technologies lack comprehensive analysis and integration of multi-source data, making it impossible to accurately predict crop water and fertilizer requirements. This results in inaccurate irrigation and fertilization, failing to meet the high-precision requirements of modern agriculture.
By fusing multi-source data and using the random forest algorithm, heterogeneous multi-source data on crops are obtained, standardized input features are generated, and water and fertilizer requirement prediction models are trained using the random forest regression algorithm. The model is updated when the error exceeds the threshold, and accurate predictions are made by combining energy balance and aerodynamic principles.
It enables accurate prediction of crop water and fertilizer requirements, improves the precision of irrigation and fertilization, enhances the robustness and adaptability of the model, and is suitable for complex agricultural environments.
Smart Images
Figure CN122155260A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a water and fertilizer integration decision-making method and device, electronic equipment, and storage medium based on multi-source data fusion and random forest, belonging to the field of agricultural planting technology. Background Technology
[0002] In modern agricultural production, water and fertilizer management is a key aspect of improving crop yield and quality. Traditional water and fertilizer management methods mainly rely on farmers' experience and simple meteorological data, lacking precision and scientific rigor.
[0003] While some irrigation systems can automate irrigation, the determination of irrigation volume still relies primarily on fixed irrigation plans or simple meteorological data, failing to reflect the actual water requirements of crops in real time. Although some fertilization technologies can apply fertilizer based on soil nutrient status, they lack comprehensive consideration of crop growth status and environmental factors, thus failing to achieve precision fertilization. Existing integrated water and fertilizer technologies, while improving water and fertilizer use efficiency to some extent, often have relatively simple decision-making processes, relying mainly on a few parameters such as soil moisture or nutrient concentration, lacking comprehensive analysis and integration of multi-source data. These technologies often exhibit significant limitations when facing complex agricultural production environments and diverse crop needs, failing to meet the high-precision requirements of precision agriculture for water and fertilizer management. Summary of the Invention
[0004] To address the technical problems existing in the prior art, this invention provides a water and fertilizer integration decision-making method and device, electronic equipment, and storage medium based on multi-source data fusion and random forest. Through multi-source data fusion and random forest algorithm, it achieves accurate prediction and dynamic adjustment of crop water and fertilizer requirements.
[0005] To achieve the above objectives, the technical solution adopted in this invention is a water and fertilizer integration decision-making method based on multi-source data fusion and random forest, comprising the following steps: Acquire multi-source heterogeneous data of crops, and generate multiple feature matrices after preprocessing; Raw data from different sources and in different formats are transformed into standardized input features that can be effectively learned by the random forest model. Using the principles of energy balance and aerodynamics, the target variables water requirement y1 and fertilizer requirement y2 are calculated. Based on the standardized input features described in the historical data, the training dataset is constructed using the corresponding water requirement y1 and fertilizer requirement y2 as labels. The water requirement prediction model and fertilizer requirement prediction model are trained using the random forest regression algorithm, respectively. The trained water demand prediction model and fertilizer demand prediction model are used to obtain the water demand prediction value and fertilizer demand prediction value for each pixel by processing the latest standardized input features in real time; the raster map of water demand prediction values is compared with the predefined irrigation control zone vector map to perform zone statistics, calculate the average water demand of pixels in each zone, and generate a variable irrigation prescription map containing the total irrigation amount and irrigation duration of each zone; the trigger conditions for irrigation execution are set according to preset operation requirements. When the average relative error between the predicted value and the actual water consumption reported by the sensor exceeds a preset threshold M for N consecutive predictions, a model update is triggered. New data samples are collected and added to the original training dataset. The model is retrained using the same process, and the performance of the new model is verified using an independent validation set. After ensuring that the new model is superior to the old model, the model update is completed.
[0006] Preferably, the multi-source heterogeneous data includes multispectral UAV remote sensing images, soil moisture and conductivity EC value data collected by ground IoT sensors, and meteorological station data.
[0007] Preferably, the preprocessing of the multi-source heterogeneous data includes: Geometric correction, radiometric correction, and atmospheric correction are performed on the multispectral UAV remote sensing image to generate an orthophoto image, and the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) are extracted to generate the first feature matrix F1. The soil moisture and electrical conductivity EC values collected by the ground IoT sensor are cleaned, outliers are removed using the Laida criterion, and missing values are filled in using time series linear interpolation to generate a second feature matrix F2. Obvious outliers were removed from the meteorological station data to generate a third feature matrix F3; The formulas for calculating the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) are as follows: NDVI = (NIR - Red) / (NIR + Red) EVI=2.5×(NIR-Red) / (NIR + 6×Red-7.5×Blue+1) NIR, Red, and Blue represent the reflectance values of the near-infrared, red, and blue bands of the remote sensing image, respectively.
[0008] Preferably, multi-source data fusion is performed during the process of converting into standardized input features that can be effectively learned by the random forest model. Using ordinary kriging interpolation, the point data in the second feature matrix F2 is interpolated with raster data with the same spatial resolution as the first feature matrix F1 to generate a spatial distribution map of environmental factors. All data in the first feature matrix F1, the interpolated spatial distribution map of environmental factors, and the third feature matrix F3 are uniformly resampled to the same geographic coordinate system and pixel size to complete spatial registration. For each registered pixel, its corresponding NDVI value, EVI value, soil moisture value, soil EC value, daily average temperature, and daily cumulative precipitation (6 features in total) are concatenated into a feature vector. The feature vectors of all pixels together constitute the spatiotemporal comprehensive feature matrix X. The difference between the current value and the value 7 days ago for each feature of each pixel is calculated as the short-term rate of change feature, which, together with the original features, serves as the final input feature of the model.
[0009] Preferably, when calculating the target variable, the reference crop evapotranspiration ET0 is calculated using the FAO-56 Penman-Monteith formula; Establish a linear regression relationship between the crop coefficient Kc and NDVI: Kc = a × NDVI + b. The coefficients a and b were obtained through localized experiments; the soil moisture stress coefficient Ks was obtained by querying a predefined table of the correspondence between the soil moisture stress coefficient Ks and soil volumetric water content, which is constructed based on the crop water production function. Calculate the actual water requirement of the crop y1=ETc=Ks×Kc×ET0; calculate the fertilizer requirement using the nutrient balance formula y2=(U×YS) / E, where U is the nutrient absorption per unit economic yield, Y is the target yield, S is the basic fertilizer supply to the soil, estimated by the soil EC value through a localized calibration model, and E is the fertilizer utilization rate. The table showing the relationship between the soil moisture stress coefficient Ks and soil volumetric water content is as follows: When the soil volumetric water content is higher than 85% of the field capacity, Ks = 1.0; When the soil volumetric water content is between 65% and 85% of the field capacity, Ks decreases linearly from 1.0 to 0.5; When the soil volumetric water content is less than 65% of the field capacity, Ks = 0.5.
[0010] Preferably, the training of the water demand prediction model and the fertilizer demand prediction model includes: using five-fold cross-validation and a grid search method to find the optimal combination of hyperparameters in a preset parameter space, wherein the preset parameter space includes the number of decision trees n_estimators and the maximum depth max_depth; using the coefficient of determination R² and the root mean square error RMSE as model performance evaluation indicators, and selecting the model with the best performance to save. The preset parameter space used by the grid search method is: n_estimators: [50, 100, 200], max_depth: [5, 10, 15].
[0011] Preferably, the preset operating requirements include avoiding irrigation during the midday high temperature period, avoiding irrigation before and after rainfall, and following the principle of water and fertilizer priority during the critical growth period of crops; The irrigation is automatically executed when the predicted water demand exceeds 10 mm and there is no effective rainfall in the next 24 hours. The model update is triggered when the average relative error of three consecutive predictions exceeds 20%.
[0012] A water and fertilizer integration decision-making device, comprising: The data acquisition and preprocessing module is used to acquire multi-source heterogeneous data and perform preprocessing. The multi-source data fusion module is used to perform feature engineering and fusion on the preprocessed data; The target variable calculation module is used to calculate the target variable according to a preset mechanism. The prediction model training module is used to train the water demand prediction model and the fertilizer demand prediction model. The prediction and prescription map generation module is used to make predictions and generate irrigation and fertilization prescription maps. The model update module is used to update the model when the model's prediction accuracy decreases.
[0013] An electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer program to implement the method as described in any one of claims 1 to 7.
[0014] A computer-readable storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the method as described in any one of claims 1 to 7.
[0015] Compared with existing technologies, the present invention has the following technical advantages: By integrating multiple data sources such as multispectral UAV remote sensing images, ground IoT sensor data, and meteorological station data, the present invention can comprehensively and accurately reflect the growth status of crops and environmental factors, providing richer information support for precise water and fertilizer management. It uses a random forest regression algorithm to predict water and fertilizer requirements. This algorithm has strong classification and regression capabilities, can process large amounts of data, and has good robustness to data noise and outliers, thereby improving the accuracy of prediction.
[0016] In addition, the water and fertilizer integration decision-making device adopts a modular design, including a data acquisition and preprocessing module, a multi-source data fusion module, a target variable calculation module, a prediction model training module, a prediction and prescription map generation module, and a model update module, which facilitates expansion and upgrading. Attached Figure Description
[0017] Figure 1 This is a system architecture diagram of the water and fertilizer integration decision-making method based on multi-source data fusion and random forest of the present invention; Figure 2 This is a flowchart of the water and fertilizer integration decision-making method based on multi-source data fusion and random forest of the present invention; Figure 3 An electronic device provided by the present invention. Detailed Implementation
[0018] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.
[0019] This invention provides a water and fertilizer integration decision-making method based on multi-source data fusion and random forest, as described below. Figures 1 to 3 A detailed description is provided, including the following steps: Multi-source heterogeneous crop data is acquired, and after preprocessing, multiple feature matrices are generated. Raw data from different sources and in different formats are transformed into standardized input features that can be effectively learned by the random forest model. Using the principles of energy balance and aerodynamics, the target variables, water requirement y1 and fertilizer requirement y2, are calculated. Using the historical input features as training features, and the corresponding water requirement y1 and fertilizer requirement y2 as labels, a training dataset is constructed. A random forest regression algorithm is then used to train the water requirement prediction model and the fertilizer requirement prediction model, respectively. When the average relative error between the predicted value and the actual water consumption reported by the sensor exceeds the preset threshold M for N consecutive predictions, the model is updated. New data samples are collected and added to the original training dataset. The model is retrained using the same process, and the performance of the new model is verified using an independent validation set. After ensuring that the new model is superior to the old model, the model update is completed.
[0020] In some embodiments, multi-source heterogeneous data includes multispectral UAV remote sensing images, soil moisture and electrical conductivity EC value data collected by ground IoT sensors, and weather station data.
[0021] Soil moisture sensors and soil EC sensors are deployed at specific field densities to acquire IoT sensor data. A multi-rotor drone and a multispectral camera covering blue, green, red, red-edge, and near-infrared bands are used. GCP ground markers are placed around the field for subsequent image geometric correction. The existing irrigation system is upgraded by installing electric valves and flow meters to ensure it can receive commands from the central control system for opening, closing, and flow regulation. In some embodiments, preprocessing includes performing geometric, radiometric, and atmospheric corrections on remote sensing images to generate orthophotos and extracting the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) to generate a first feature matrix F1; cleaning sensor data by removing outliers using the Laida criterion and filling in missing values using time-series linear interpolation to generate a second feature matrix F2; and removing significant outliers from meteorological data to generate a third feature matrix F3. Pix4Dfields or ENVI software can be used to import UAV imagery and GCP points for automatic stitching, geometric correction, and radiometric correction. Data cleaning scripts can be written on a cloud platform to automatically remove sensor outliers, potentially using the 3σ criterion, and to perform missing value interpolation.
[0022] In some embodiments, multi-source data fusion is performed during the transformation into standardized input features. Using ordinary kriging interpolation, the point data in the second feature matrix F2 is interpolated into raster data with the same spatial resolution as the first feature matrix F1 to generate an environmental factor spatial distribution map. All data in the first feature matrix F1, the interpolated environmental factor spatial distribution map, and the third feature matrix F3 are uniformly resampled to the same geographic coordinate system and pixel size to complete spatial registration. For each registered pixel, its corresponding NDVI value, EVI value, soil moisture value, soil EC value, daily average temperature, and daily cumulative precipitation (6 features in total) are concatenated into a feature vector. The feature vectors of all pixels together constitute the spatiotemporal comprehensive feature matrix X. The difference between the current value and the value 7 days ago for each feature of each pixel is calculated as the short-term rate of change feature, which, together with the original features, serves as the final input feature of the model.
[0023] Kriging is not merely a simple mathematical interpolation; it is an optimal unbiased estimator. It considers the spatial autocorrelation of data, enabling it to optimally predict values at unknown locations based on the known spatial distribution structure of points, and providing the prediction error. This is more scientific and accurate than simpler methods such as inverse distance weighting, and is particularly suitable for spatially correlated environmental variables.
[0024] The output generates a complete soil moisture distribution map and a soil EC distribution map covering the entire field, with each pixel size consistent with that of the UAV imagery. The core objective of this step is to address the inherent problems of agricultural data—different sources, scales, formats / resolutions—and to inject spatiotemporal understanding into the model. This step involves translating and aligning all this data to a single standard spatial framework. Unified resampling to the same geographic coordinate system and cell size, along with spatial registration, ensures that the data are in the exact same coordinate system and spatial resolution. This is the foundation of pixel-level operations. The QGIS or Python GDAL libraries can be used to unify all spatial data to the same coordinate system and cell size.
[0025] After the first two steps, each pixel location now corresponds to a small plot of land in the field and possesses multiple attribute values from different data sources. This feature vector comprehensively describes the state of a small unit at a specific moment. The random forest model then learns the complex nonlinear relationship between these six features and the target variable.
[0026] In some embodiments, the target variable is calculated using the FAO-56 Penman-Monteith formula to calculate the reference crop evapotranspiration ET0, and a linear regression relationship is established between the crop coefficient Kc and NDVI: Kc = a × NDVI + b, where coefficients a and b are obtained through localized experimental calibration. The soil moisture stress coefficient Ks is obtained by querying a predefined table of correspondence between soil moisture and Ks constructed based on the crop water production function. The actual crop water requirement y1 = ETc = Ks × Kc × ET0 is calculated, and the fertilizer requirement is calculated using the nutrient balance formula y2 = (U × YS) / E, where U is the nutrient uptake per unit of economic yield, Y is the target yield, S is the basic soil fertilizer supply estimated from the soil EC value through a localized calibration model, and E is the fertilizer utilization rate.
[0027] By introducing the time dimension into the model, the practical value and accuracy of the model are greatly improved.
[0028] In some embodiments, predictive model training includes using five-fold cross-validation and grid search to find the optimal combination of hyperparameters in a preset parameter space, the parameter space including the number of decision trees n_estimators and the maximum depth max_depth, using the coefficient of determination R² and root mean square error RMSE as model performance evaluation metrics, and selecting the model with the best performance to save.
[0029] The model was trained using RandomForestRegressor, taking as input the feature vector X of all pixels from the past growing season, corresponding to the calculated water requirement y1 and fertilizer requirement y2. GridSearchCV was used to optimize the model's hyperparameters. The model was trained on 70% of the data and tested on 30% of the data, using the coefficient of determination R² and root mean square error (RMSE) as performance evaluation metrics.
[0030] In some embodiments, the method includes prediction and prescription map generation: inputting the latest feature data generated in real time into the trained prediction model to obtain the predicted water demand and fertilizer demand for each pixel; performing zonal statistics on the raster map of the predicted water demand and the predefined irrigation control zone vector map; calculating the average water demand of pixels in each zone; generating a variable irrigation prescription map containing the total irrigation amount and irrigation duration for each zone; and setting the triggering conditions for irrigation execution according to preset operation requirements.
[0031] In some embodiments, the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) are extracted, and the calculation formulas are as follows: NDVI = (NIR - Red) / (NIR + Red) EVI=2.5×(NIR-Red) / (NIR + 6×Red-7.5×Blue+1) NIR, Red, and Blue represent the reflectance values of the near-infrared, red, and blue bands of the remote sensing image, respectively.
[0032] In some embodiments, the relationship between the soil moisture stress coefficient Ks and the soil volumetric water content is shown in the table below: When the soil volumetric water content is higher than 85% of the field capacity, Ks = 1.0. When the soil volumetric water content is between 65% and 85% of field capacity, Ks decreases linearly from 1.0 to 0.5. When the soil volumetric water content is less than 65% of the field capacity, Ks = 0.5.
[0033] In some embodiments, the preset parameter space used by the grid search method is: n_estimators: [50, 100, 200], max_depth: [5, 10, 15].
[0034] In some embodiments, preset operating requirements include avoiding irrigation during the midday high-temperature period, avoiding irrigation before and after rainfall, and following the water and fertilizer priority principle during the critical growth period of crops. The triggering condition is that irrigation is automatically performed when the predicted cumulative water demand exceeds 10 mm and there is no effective rainfall in the next 24 hours.
[0035] In some embodiments, the model update is triggered when the average relative error of three consecutive predictions exceeds 20%.
[0036] This invention also proposes a water and fertilizer integration decision-making device, comprising: The data acquisition and preprocessing module is used to acquire multi-source heterogeneous data and perform preprocessing. The multi-source data fusion module is used to perform feature engineering and fusion on the preprocessed data; The target variable calculation module is used to calculate the target variable according to a preset mechanism. The prediction model training module is used to train the water demand prediction model and the fertilizer demand prediction model. The prediction and prescription map generation module is used to make predictions and generate irrigation and fertilization prescription maps. The model update module is used to update the model when the model's prediction accuracy decreases.
[0037] The present invention proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described above.
[0038] The present invention proposes a computer-readable storage medium storing computer-readable instructions, which, when executed by one or more processors, cause one or more processors to perform the method described above.
[0039] In the several embodiments provided in this application, it should be understood that the control and memory of the provided electronic device can be implemented in other ways. For example, the division of a certain module is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.
[0040] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0041] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of the present invention.
Claims
1. A water and fertilizer integration decision-making method based on multi-source data fusion and random forest, characterized in that, Includes the following steps: Acquire multi-source heterogeneous data of crops, and generate multiple feature matrices after preprocessing; Raw data from different sources and in different formats are transformed into standardized input features that can be effectively learned by the random forest model. Using the principles of energy balance and aerodynamics, the target variables water requirement y1 and fertilizer requirement y2 are calculated. Based on the standardized input features described in the historical data, the training dataset is constructed using the corresponding water requirement y1 and fertilizer requirement y2 as labels. The water requirement prediction model and fertilizer requirement prediction model are trained using the random forest regression algorithm, respectively. The trained water demand prediction model and fertilizer demand prediction model are used to obtain the water demand prediction value and fertilizer demand prediction value for each pixel by processing the latest standardized input features in real time; the raster map of water demand prediction values is compared with the predefined irrigation control zone vector map to perform zone statistics, calculate the average water demand of pixels in each zone, and generate a variable irrigation prescription map containing the total irrigation amount and irrigation duration of each zone; the trigger conditions for irrigation execution are set according to preset operation requirements. When the average relative error between the predicted value and the actual water consumption reported by the sensor exceeds a preset threshold M for N consecutive predictions, a model update is triggered. New data samples are collected and added to the original training dataset. The model is retrained using the same process, and the performance of the new model is verified using an independent validation set. After ensuring that the new model is superior to the old model, the model update is completed.
2. The water and fertilizer integration decision-making method based on multi-source data fusion and random forest according to claim 1, characterized in that, The multi-source heterogeneous data includes multispectral UAV remote sensing images, soil moisture and electrical conductivity EC value data collected by ground IoT sensors, and meteorological station data.
3. The water and fertilizer integration decision-making method based on multi-source data fusion and random forest according to claim 2, characterized in that, The preprocessing of the multi-source heterogeneous data includes: Geometric correction, radiometric correction, and atmospheric correction are performed on the multispectral UAV remote sensing image to generate an orthophoto image, and the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) are extracted to generate the first feature matrix F1. The soil moisture and electrical conductivity EC values collected by the ground IoT sensor are cleaned, outliers are removed using the Laida criterion, and missing values are filled in using time series linear interpolation to generate a second feature matrix F2. Obvious outliers were removed from the meteorological station data to generate a third feature matrix F3; The formulas for calculating the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) are as follows: NDVI = (NIR - Red) / (NIR + Red) EVI=2.5×(NIR-Red) / (NIR + 6×Red-7.5×Blue+1) NIR, Red, and Blue represent the reflectance values of the near-infrared, red, and blue bands of the remote sensing image, respectively.
4. The water and fertilizer integration decision-making method based on multi-source data fusion and random forest according to claim 3, characterized in that, In the process of transforming standardized input features that can be effectively learned by the random forest model, multi-source data fusion is performed. Using ordinary kriging interpolation, the point data in the second feature matrix F2 is interpolated with the raster data with the same spatial resolution as the first feature matrix F1 to generate a spatial distribution map of environmental factors. All data in the first feature matrix F1, the interpolated spatial distribution map of environmental factors, and the third feature matrix F3 are uniformly resampled to the same geographic coordinate system and pixel size to complete spatial registration. For each registered pixel, its corresponding NDVI value, EVI value, soil moisture value, soil EC value, daily average temperature, and daily cumulative precipitation (6 features in total) are concatenated into a feature vector. The feature vectors of all pixels together constitute the spatiotemporal comprehensive feature matrix X. The difference between the current value and the value 7 days ago for each feature of each pixel is calculated as the short-term rate of change feature, which, together with the original features, serves as the final input feature of the model.
5. The water and fertilizer integration decision-making method based on multi-source data fusion and random forest according to claim 1, characterized in that, When calculating the target variable, the reference crop evapotranspiration ET0 was calculated using the FAO-56 Penman-Monteith formula. Establish a linear regression relationship between the crop coefficient Kc and NDVI: Kc = a × NDVI + b. The coefficients a and b were obtained through localized experiments; the soil moisture stress coefficient Ks was obtained by querying a predefined table of the correspondence between the soil moisture stress coefficient Ks and soil volumetric water content, which is constructed based on the crop water production function. Calculate the actual water requirement of the crop y1=ETc=Ks×Kc×ET0; calculate the fertilizer requirement using the nutrient balance formula y2=(U×YS) / E, where U is the nutrient absorption per unit economic yield, Y is the target yield, S is the basic fertilizer supply to the soil, estimated by the soil EC value through a localized calibration model, and E is the fertilizer utilization rate. The table showing the relationship between the soil moisture stress coefficient Ks and soil volumetric water content is as follows: When the soil volumetric water content is higher than 85% of the field capacity, Ks = 1.0; When the soil volumetric water content is between 65% and 85% of the field capacity, Ks decreases linearly from 1.0 to 0.5; When the soil volumetric water content is less than 65% of the field capacity, Ks = 0.
5.
6. The water and fertilizer integration decision-making method based on multi-source data fusion and random forest according to claim 1, characterized in that, The training of the water demand prediction model and the fertilizer demand prediction model includes using five-fold cross-validation and a grid search method to find the optimal combination of hyperparameters in a preset parameter space, which includes the number of decision trees n_estimators and the maximum depth max_depth; using the coefficient of determination R² and the root mean square error RMSE as model performance evaluation indicators, and selecting the model with the best performance to save. The preset parameter space used by the grid search method is: n_estimators: [50, 100, 200], max_depth: [5, 10, 15].
7. The water and fertilizer integration decision-making method based on multi-source data fusion and random forest according to claim 1, characterized in that, The preset operating requirements include avoiding irrigation during the midday heat, avoiding irrigation before and after rainfall, and following the principle of prioritizing water and fertilizer during the critical growth stages of crops. The irrigation is automatically executed when the predicted water demand exceeds 10 mm and there is no effective rainfall in the next 24 hours. The model update is triggered when the average relative error of three consecutive predictions exceeds 20%.
8. A water and fertilizer integration decision-making device, characterized in that, include: The data acquisition and preprocessing module is used to acquire multi-source heterogeneous data and perform preprocessing. The multi-source data fusion module is used to perform feature engineering and fusion on the preprocessed data; The target variable calculation module is used to calculate the target variable according to a preset mechanism. The prediction model training module is used to train the water demand prediction model and the fertilizer demand prediction model. The prediction and prescription map generation module is used to make predictions and generate irrigation and fertilization prescription maps. The model update module is used to update the model when the model's prediction accuracy decreases.
9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The storage medium stores computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the method as described in any one of claims 1 to 7.