An agricultural field big data intelligent management method based on an internet of things

Abstract

This invention relates to the field of agricultural IoT monitoring technology. Specifically, it relates to an IoT-based intelligent management method for agricultural big data, aiming to solve the problem that the cross-boundary effects of lateral water migration and lateral nutrient diffusion between adjacent plots of contiguous farmland are not included in the decision-making of individual plots, leading to excessive or insufficient water and fertilizer supply. This invention deploys soil moisture and nutrient sensor arrays at the boundaries of adjacent plots to collect cross-boundary characteristic data. It integrates irrigation and fertilization records and internal sensor data of the plots to construct a cross-boundary characteristic dataset, trains a water and fertilizer cross-boundary prediction model, and inputs the water and fertilizer plans of adjacent plots when formulating irrigation and fertilization plans for target plots. It predicts the amount of cross-boundary water migration and nutrient diffusion, and accordingly reverses the target amounts of irrigation and fertilization to generate a precise plan adapted to the cross-boundary impact. This achieves intelligent coordinated regulation of water and fertilizer in contiguous farmland, improves water and fertilizer utilization, and reduces resource waste and non-point source pollution.

Description

Technical Field

[0001] This invention relates to the field of farmland Internet of Things (IoT) monitoring technology, and more specifically, to a smart management method for farmland big data based on the Internet of Things. Background Technology

[0002] Farmland IoT monitoring is an important technology, specifically applied to the intelligent control of irrigation and fertilization in large-scale contiguous farmland. It achieves precise water and fertilizer supply through boundary sensing and cross-boundary prediction. In contiguous farmland, there are cross-boundary interactions involving lateral migration of soil moisture and lateral diffusion of nutrients between adjacent plots. Because irrigation and fertilization decisions for individual plots are based solely on their own soil, crop, and meteorological data, without considering the water and fertilizer plans of adjacent plots and the patterns of water and fertilizer migration at the boundaries, the planned irrigation and fertilization amounts cannot offset the gains or losses from cross-boundary water and nutrients. This leads to problems such as excessive irrigation and fertilization causing resource waste or insufficient fertilization affecting crop growth, making it difficult to achieve precise water and fertilizer supply and efficient utilization in contiguous farmland. To solve this technical problem, we provide an IoT-based intelligent management method for farmland big data. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent management method for farmland big data based on the Internet of Things, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, one of the objectives of this invention is to provide a smart management method for farmland based on Internet of Things (IoT) big data, comprising the following steps: S1. A boundary sensor array is deployed in the boundary area between adjacent first and second fields. The boundary sensor array includes at least one soil moisture sensor and at least one soil nutrient sensor to collect cross-boundary characteristic data reflecting the lateral migration of water and lateral diffusion of nutrients between the first and second fields within the boundary area. S2. Obtain irrigation and fertilization records, standard sensor data within the fields, and cross-boundary feature data containing the first and second fields, and construct a cross-boundary feature dataset. The cross-boundary feature dataset is used to characterize the correlation between the water and fertilizer cross-boundary effect between adjacent fields and the differences in irrigation amount, fertilizer amount, soil texture, and initial moisture content. S3. Train a water and fertilizer cross-border prediction model based on the cross-border feature dataset. The water and fertilizer cross-border prediction model is used to predict the cross-border water migration and cross-border nutrient diffusion in future cycles based on the irrigation and fertilization plan data of adjacent fields. The cross-border water migration and cross-border nutrient diffusion together constitute the predicted cross-border amount. S4. When making irrigation and fertilization decisions for the first field, obtain the current irrigation and fertilization plan data of the second field and input it into the water and fertilizer cross-boundary prediction model to obtain the predicted cross-boundary water migration and the predicted cross-boundary nutrient diffusion. Based on the predicted cross-boundary amounts, perform reverse correction on the irrigation target amount and fertilization target amount of the first field to generate an irrigation and fertilization scheme after cross-boundary influence correction.

[0005] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention effectively solves the problem of supply imbalance caused by the failure to consider cross-boundary water and fertilizer interactions in single-plot decision-making for contiguous farmland by integrating boundary sensing monitoring, cross-boundary water and fertilizer prediction, and reverse correction of irrigation and fertilization plans. It can accurately quantify the lateral migration of water and nutrient diffusion between adjacent plots, enabling dynamic and precise control of irrigation and fertilization amounts. It eliminates the need for manual calculation of cross-boundary losses and forms an intelligent decision-making closed loop based on big data and the Internet of Things, significantly reducing the manpower input for large-scale farmland management and improving the efficiency of water and fertilizer decision-making. It can not only avoid resource waste and agricultural non-point source pollution caused by excessive water and fertilizer, but also make up for the supply shortage caused by cross-boundary losses, stably ensuring the water and fertilizer conditions required for crop growth and helping to improve crop yield and quality. Attached Figure Description

[0006] Figure 1 This is a flowchart illustrating the overall workflow of the present invention. Detailed Implementation

[0007] 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.

[0008] Please see Figure 1 As shown, this embodiment provides a smart management method for farmland based on Internet of Things (IoT) big data, including the following steps: S1. A boundary sensor array is deployed in the boundary area between adjacent first and second fields. The boundary sensor array includes at least one soil moisture sensor and at least one soil nutrient sensor to collect cross-boundary characteristic data reflecting the lateral migration of water and lateral diffusion of nutrients between the first and second fields within the boundary area. S2. Obtain irrigation and fertilization records, standard sensor data within the fields, and cross-boundary feature data containing the first and second fields, and construct a cross-boundary feature dataset. The cross-boundary feature dataset is used to characterize the relationship between the water and fertilizer cross-boundary effect between adjacent fields and the differences in irrigation amount, fertilizer amount, soil texture, and initial moisture content. S3. A water and fertilizer cross-boundary prediction model is trained based on the cross-boundary feature dataset. The water and fertilizer cross-boundary prediction model is used to predict the cross-boundary water migration and cross-boundary nutrient diffusion in the future cycle based on the irrigation and fertilization plan data of adjacent fields. The cross-boundary water migration and cross-boundary nutrient diffusion together constitute the predicted cross-boundary amount. S4. When making irrigation and fertilization decisions for the first field, obtain the current irrigation and fertilization plan data for the second field and input it into the water and fertilizer cross-boundary prediction model to obtain the predicted cross-boundary water migration and predicted cross-boundary nutrient diffusion. Based on the predicted cross-boundary amounts, perform reverse correction on the irrigation target amount and fertilization target amount for the first field to generate an irrigation and fertilization scheme corrected for cross-boundary influence.

[0009] Further explanation is needed. After completing the sensor array deployment plan for the boundary area between the adjacent first and second fields in step S1, the specific delineation method of the boundary area must first be clarified. This area is a strip-shaped region formed by extending horizontally to both sides with a specific width, based on the center line of the field ridge shared between the first and second fields. The center line of the field ridge refers to the geometric center axis of the field ridge that separates the two adjacent farmlands. It is the baseline for dividing the water and fertilizer interaction boundary between the two fields. The strip-shaped region is a uniformly wide monitoring area that extends along the direction of the field ridge. To ensure that the lateral migration and diffusion behavior of water and fertilizer within the crop root system can be completely captured, the width of this strip-shaped region is set to be no less than the typical horizontal extension range of the root system of a single field. The typical horizontal extension range of the root system refers to the maximum horizontal distance that the roots of the main crop in the field can reach by growing laterally in the soil tillage layer. This ensures that the monitoring range completely covers the area of ​​water and fertilizer interaction between the roots of the two fields.

[0010] After delineating the boundary area, a boundary sensor array is deployed within the strip-shaped area according to the plan. This array consists of at least one soil moisture sensor and at least one soil nutrient sensor, arranged in a staggered pattern. This staggered arrangement means that the soil moisture and nutrient sensors are alternately arranged horizontally and staggered vertically, avoiding overlap and interference between their detection areas while ensuring comprehensive coverage of the entire boundary area. The soil moisture sensor works by detecting capillary water connectivity between different soil depths in adjacent plots. Capillary water is water adsorbed and moved in soil pores by capillary forces, and is the main carrier of lateral soil moisture migration in farmland. Different soil depths include key soil layers such as the 0-20cm topsoil layer and the 20-40cm root zone. The sensor collects soil moisture content values ​​in real time at different locations on both sides and inside the boundary area. Soil moisture gradient data is obtained by calculating the difference in moisture content between adjacent measuring points. The calculation method is as follows: In the formula This represents the soil moisture content gradient. This represents the soil volumetric water content measured at the lateral boundary of the first field plot. The soil volumetric water content at the measuring point on the lateral boundary of the second field is shown. This gradient data directly reflects the direction and rate of lateral migration of water from fields with high water content to fields with low water content. The soil nutrient sensor works by detecting changes in the ion concentration gradient in the soil solution within the strip-shaped area. The ions in the soil solution mainly include crop nutrient ions such as nitrate nitrogen, ammonium nitrogen, potassium ions, and phosphate ions, which are the main forms of nutrient lateral diffusion. The sensor indirectly reflects the nutrient ion concentration by detecting soil conductivity. It collects soil conductivity values ​​at different measuring points in the boundary area in real time, and obtains the soil conductivity gradient data by calculating the conductivity difference between measuring points. The calculation method is as follows: In the formula This represents the soil electrical conductivity gradient. The soil electrical conductivity at the measuring point on the side boundary of the first field plot. The soil electrical conductivity at the lateral boundary measurement point of the second field directly characterizes the trend and intensity of nutrient lateral diffusion from high-concentration areas to low-concentration areas. Finally, the soil moisture gradient data collected by the soil moisture sensor and the soil electrical conductivity gradient data collected by the soil nutrient sensor are combined to form cross-boundary feature data that can accurately reflect the lateral migration of water and lateral diffusion of nutrients between the first and second fields. This data serves as the basis for constructing the cross-boundary feature dataset and training the water and fertilizer cross-boundary prediction model, providing original data support for the quantitative analysis and accurate prediction of the water and fertilizer cross-boundary effect between adjacent fields.

[0011] The acquisition of irrigation and fertilization records, standard sensor data within the fields, and cross-boundary feature data for the first and second fields is carried out through an IoT data aggregation platform in a unified time series. The irrigation and fertilization records are from the logs reported by the controller of the integrated water and fertilizer machine deployed at the irrigation head of the field. The standard sensor data within the field is the data collected by the standard weather station and soil sensor respectively located at the geometric center of the first and second fields.

[0012] After completing the staggered deployment of the boundary sensor array and the initial collection of cross-boundary feature data, in order to construct a standardized and high-quality model training dataset, it is necessary to uniformly process the heterogeneous data from the fertigation unit controller, standard weather station, soil sensor, and boundary sensor array. The fertigation unit controller is a control device used for precise regulation of irrigation and fertilization amounts; the standard weather station is a standardized monitoring device used to collect meteorological parameters such as ambient temperature, humidity, wind speed, and rainfall; the soil sensor is a standardized monitoring device used to collect soil physicochemical parameters within the field; and the boundary sensor array is a staggered array of sensors used to collect cross-boundary water and fertilizer migration data. Heterogeneous data refers to multi-source data from different devices, with different collection frequencies, and different data formats. First, the above multi-source heterogeneous data undergoes timestamp alignment and cleaning. A timestamp is a unique time identifier generated at the time of collection for each data point. Timestamp alignment involves aligning data from different collection frequencies and formats. Heterogeneous data from different time points are precisely matched according to a unified timeline to ensure one-to-one correspondence between multiple data sources at the same time. During data cleaning, outliers exceeding the normal range of soil physicochemical properties are removed. Missing values ​​generated during data collection are filled using linear interpolation to ensure data continuity and integrity. Subsequently, standard sensor data from within the field within a specific time window before and after an event are extracted as environmental conditions. The event refers to the irrigation and fertilization operations performed by the fertigation machine, and the specific time window is the period from 5 minutes before the event to 30 minutes after. Standard sensor data within the field includes parameters characterizing the basic environmental conditions of the field, such as soil temperature, soil porosity, and soil aeration. Simultaneously, event-related irrigation and fertilization differences are extracted as planning inputs. The irrigation difference refers to the difference between the preset target irrigation amount and the actual irrigation amount performed by the fertigation machine; its calculation formula is... ,in This represents the difference in irrigation volume. The target irrigation amount preset for the water and fertilizer integrated machine. The fertilizer application difference refers to the difference between the preset target fertilizer application amount and the actual fertilizer application amount of the integrated water and fertilizer machine. The calculation formula is as follows: ,in This represents the difference in fertilizer application rate. The target fertilizer application rate preset for the integrated water and fertilizer machine. To determine the actual fertilization amount, soil texture difference quantification index and pre-event initial moisture content difference quantification index were extracted as soil background. The soil texture difference quantification index is the Euclidean distance value calculated from the percentage content of sand, silt, and clay particles in adjacent fields, used to accurately quantify the degree of soil texture difference between adjacent fields. The pre-event initial moisture content difference quantification index refers to the difference in volumetric moisture content of adjacent fields before the irrigation and fertilization event, and its calculation formula is... ,in As a quantitative indicator of the difference in initial moisture content, This represents the initial volumetric water content of the first field before the irrigation and fertilization event occurred. This represents the initial volumetric water content of the second field before the irrigation and fertilization event.

[0013] After completing the above data extraction, the change characteristics of soil moisture content gradient data and soil electrical conductivity gradient data collected by the boundary sensor array within the corresponding time window after the event are extracted as effect output. Soil moisture content gradient data refers to the gradient sequence formed by the difference in soil volumetric water content at different measuring points within the boundary area, which is used to characterize the direction and rate of lateral water migration. Soil electrical conductivity gradient data refers to the gradient sequence formed by the difference in soil electrical conductivity at different measuring points within the boundary area, which is used to characterize the trend and intensity of lateral nutrient diffusion. Finally, the environmental state, planned input, soil background and effect output obtained above are associated and encapsulated according to a unified timestamp to form a single multidimensional data sample. All multidimensional data samples that meet the data specifications within the historical time period are integrated and summarized to finally form a cross-boundary feature dataset for training the water and fertilizer cross-boundary prediction model. The entire processing process uses a unified time axis as a link to standardize and associate multi-source heterogeneous data, ensuring that each multidimensional data sample can completely characterize the correspondence between irrigation and fertilization events and water and fertilizer cross-boundary effects, providing standardized dataset support for the subsequent training of the water and fertilizer cross-boundary prediction model.

[0014] After completing timestamp alignment, outlier cleaning, and missing value interpolation of multi-source heterogeneous data, the cross-boundary feature dataset accurately represents the correlation relationships through a large number of multi-dimensional data samples. These multi-dimensional data samples encapsulate irrigation difference, fertilization difference, soil texture difference quantification indicators, and initial moisture content difference quantification indicators into a set of collaborative input conditions. Simultaneously, the soil moisture content gradient change and soil conductivity gradient change collected by the boundary sensor array are used as a set of coupled response results. The nonlinear mapping relationship between input and output is mined and expressed through the inherent statistical laws of the massive samples in the dataset. Specifically, each set of collaborative input conditions is first vectorized, and the input vector is denoted as... ,in This represents the difference between the target irrigation amount and the actual irrigation amount. This represents the difference in fertilizer application rate, specifically the difference between the target fertilizer application rate and the actual fertilizer application rate. This quantitative index represents the difference in soil texture and is calculated from the Euclidean distance of the sand, silt, and clay content of adjacent fields. Representing the initial moisture content difference, it is the difference in volumetric moisture content of the soil in the two farmlands before the event. The coupled response result is expressed as... , Represents the gradient change in soil moisture content. Representing the gradient change in soil electrical conductivity, after constructing the input and output vectors, an input matrix is ​​built based on all N samples in the cross-boundary feature dataset. With output matrix This method extracts the correlation between input and output by using statistical measures such as the mean, variance, and covariance of the sample population. The sample mean characterizes the central tendency of the variables in the dataset, the sample variance measures the dispersion of the variables, and the covariance quantifies the linear correlation between the input and output variables. The formula for calculating the covariance is... , The mathematical expectation of the input vector, The expected value of the output vector is given by the covariance matrix, which can intuitively reflect the strength and direction of the influence of the cooperative input conditions on the coupled response. Based on this, maximum likelihood estimation is used to fit the joint probability distribution between the input and output. This probability distribution characterizes the likelihood of specific changes in water content gradient and electrical conductivity gradient given differences in irrigation amount, fertilizer amount, soil texture, and initial water content. This enables the accurate characterization of the cross-boundary effect of water and fertilizer by utilizing the overall statistical regularity of the dataset. Finally, the input vector, output vector, statistical features, and probability distribution are stored in a structured manner, so that each multidimensional data sample can fully map the causal relationship of cross-boundary migration of farmland water and fertilizer, providing a standardized and reusable data foundation for subsequent prediction models.

[0015] After establishing the basic framework of the cross-boundary feature dataset with multidimensional data samples, to further accurately quantify the impact of basic soil properties of adjacent fields on cross-boundary water and fertilizer migration, it is necessary to first calculate the quantitative index of soil texture difference and the quantitative index of initial moisture content difference, and then combine these two indices with the differences in irrigation and fertilization amounts to form the complete collaborative input conditions required for model training. The first step is to calculate the quantitative index of soil texture difference, which objectively reflects the degree of difference in the physical structure of the soil between the first and second fields, directly determining the soil pore distribution, capillary water conduction capacity, and nutrient diffusion rate. First, a soil particle composition analysis experiment is needed to determine the mass percentages of sand, silt, and clay particles in the topsoil of the first and second fields. The topsoil layer (0–40 cm) is the 0–40 cm soil layer in farmland where crop roots are concentrated and water and fertilizer interactions are most active. It is also the soil layer where water migrates laterally and nutrients diffuse. Sand, silt, and clay are the three major particle components that classify soil texture. The mass percentage refers to the proportion of each particle component's mass to the total dry weight of the soil, and is a standard parameter characterizing soil texture. Subsequently, the contents of the three components in the two fields are used to construct three-dimensional texture feature vectors. The texture vector of the first field is denoted as... The texture vector of the second field is denoted as ,in These represent the percentage of sand particles by mass in the first and second fields, respectively. These represent the percentage of silt mass in the first and second fields, respectively. The percentages of clay mass in the first and second fields are respectively. The spatial distance between the two texture vectors is then calculated using Euclidean distance, a classic mathematical method for measuring the degree of difference in multidimensional data. A larger Euclidean distance indicates a more significant difference in soil texture between the two fields. The specific calculation formula is as follows: In the formula This is the final quantitative index of soil texture differences.

[0016] After calculating the soil texture difference index, the initial moisture content difference quantification index is calculated. This index characterizes the initial soil moisture gradient between adjacent fields before the irrigation and fertilization event, and is the driving force for lateral water migration. The calculation requires first obtaining the average volumetric moisture content of the topsoil layer of both the first and second fields before the irrigation and fertilization event. The average volumetric moisture content is the arithmetic mean of the soil volumetric moisture content at different depths within the topsoil layer. Volumetric moisture content is the proportion of water volume in a unit volume of soil and is a standardized index characterizing soil moisture content. Then, the relative difference between the two fields is calculated using the following formula: In the formula The average volumetric moisture content of the topsoil layer in the first field. The average volumetric moisture content of the topsoil in the second field. This is the initial moisture content difference quantification index, obtained after obtaining the soil texture difference quantification index. Quantitative indicators of the difference from the initial water content Then, combined with the already determined difference in irrigation volume Difference from fertilizer application The irrigation difference is the difference between the planned irrigation amounts for the first and second fields, and the fertilizer difference is the difference between the planned fertilizer amounts for the first and second fields. These four indicators are standardized and combined according to a unified data dimension to form the collaborative input conditions for the cross-boundary water and fertilizer prediction model. These collaborative input conditions are represented in the form of a four-dimensional feature vector. The four indicators work together to comprehensively cover the three major influencing factors of differences in irrigation and fertilization, soil texture, and initial moisture. They fully characterize all the preconditions that drive the cross-boundary migration of water and fertilizer, providing accurate and complete input features for the subsequent model to learn the mapping law of the cross-boundary effect of water and fertilizer.

[0017] After completing the standardized construction of the cross-boundary feature dataset and accurately defining the collaborative input conditions and effect outputs, the water and fertilizer cross-boundary prediction model can be trained using supervised learning methods based on this dataset. Supervised learning is a machine learning method that uses sample data with clear input features and corresponding output labels to guide the model to automatically learn the nonlinear mapping law between input and output. It can accurately adapt to the quantitative prediction needs of water and fertilizer cross-boundary migration and diffusion. The specific training process first completes the vectorization transformation of the sample data, and then transforms the collaborative input conditions of each multi-dimensional data sample in the cross-boundary feature dataset, namely the difference in irrigation amount, into vectorized data. Fertilizer application rate difference Quantitative indicators of soil texture differences Quantitative indicators of initial moisture content difference These four features are combined and encapsulated into the input feature vector of the water and fertilizer cross-boundary prediction model, denoted as... This vector comprehensively covers all the antecedent factors driving the cross-boundary effect of water and fertilizer, and outputs the effect corresponding to each multidimensional data sample, namely the change in soil moisture content gradient. With changes in soil electrical conductivity gradient The vectors are combined and encapsulated into the target output vector for model training, denoted as... This vector accurately corresponds to the actual response result of cross-boundary migration and diffusion of water and fertilizer. Each set of input feature vectors is matched one-to-one with the target output vector, forming a complete training sample pair required for supervised learning.

[0018] After completing the sample vectorization, the neural network model is first initialized. This model uses a multi-layer fully connected neural network adapted to the nonlinear mapping relationship between water and fertilizer in farmland. The initialization operation involves setting the weight parameters for all neurons in the network. With bias parameters Assigning random small values ​​that conform to a normal distribution avoids excessively large initial parameters that could cause divergence during training. The initialized model can be represented as follows: ,in This is the model's predicted output. For the basic mapping function of the network, This is the network weight matrix. The network bias vector is used, and then historical multidimensional data samples from the cross-boundary feature dataset are used to iteratively train the initialized neural network model. The iterative training process consists of three loops: forward propagation calculation, loss function evaluation, and backpropagation parameter optimization. First, the input feature vector X is input into the model, and the model's predicted output is obtained through forward propagation calculation. Then, the mean squared error loss function is used to calculate the error between the predicted output and the target output. The loss function formula is as follows: ,in This represents the overall loss value. This represents the total number of training samples. For the first The model prediction output for each sample. For the first The model outputs the true target values ​​for each sample. A smaller loss value indicates higher prediction accuracy. Then, the Adam optimization algorithm is used to adaptively adjust the model's internal parameters. This algorithm has adaptive learning rate characteristics, enabling it to quickly converge and accurately fit the nonlinear patterns of farmland water and fertilizer data. The loss function is calculated with respect to the weights using the backpropagation algorithm. and bias gradient , Then update the parameters according to the gradient direction. The parameter update formula is as follows: , ,in A preset learning rate is used to control the step size of parameter updates, preventing model oscillations caused by excessively rapid updates. The entire training process iterates through forward propagation, loss calculation, backpropagation, and parameter updates until the loss function converges to a preset accuracy threshold—meaning the loss value no longer decreases significantly with increasing iterations. At this point, the model has fully learned a stable mapping function between co-input conditions and effect output. This function fully characterizes the intrinsic correlation between differences in irrigation and fertilization, soil texture, initial moisture, and cross-boundary migration and diffusion of water and fertilizer. Finally, a trained and practically applicable cross-boundary water and fertilizer prediction model is obtained. This model can quickly and accurately predict the amount of cross-boundary water migration and cross-boundary nutrient diffusion in future cycles based on the irrigation and fertilization plans of adjacent fields, providing model support for the reverse correction of subsequent irrigation and fertilization schemes.

[0019] After completing supervised learning training and achieving the preset prediction accuracy, the water and fertilizer cross-boundary prediction model can be officially put into practical application. It is used to accurately predict cross-boundary water migration in future periods based on irrigation and fertilization plan data from adjacent fields. The future period refers to the timeframe during which water and fertilizer management will be implemented, typically using 24 or 72 hours as a standard period. The irrigation and fertilization plan data refers to the pre-defined data in the farmland management platform regarding the planned irrigation and fertilization amounts for the first and second fields in the future period. Specifically, during prediction implementation, the expected irrigation amounts for the first and second fields in the future period are first extracted from the irrigation and fertilization plan data, and the difference between the two is calculated to obtain the expected irrigation amount difference. The calculation formula is as follows: In the formula This represents the difference in expected irrigation volume. The planned irrigation amount for the first plot over the future cycle. To calculate the planned irrigation amount for the second field in the future cycle, the estimated fertilizer application amounts for the two fields in the future cycle are extracted simultaneously, and the difference in estimated fertilizer application amounts is calculated using the following formula: In the formula This is the difference in the expected amount of fertilizer applied. This is the planned fertilizer application amount for the first field in the future cycle. The planned fertilizer application amount for the second field in the future cycle is determined by using the fixed quantitative indicators of soil texture differences between the two fields, which have been obtained through soil particle composition analysis. This indicator is a constant characteristic parameter and does not require repeated calculation. Combining current standard sensor data within the field with future weather forecast data, the initial moisture content difference at the start of the future cycle is estimated. Standard sensor data within the field refers to the real-time volumetric moisture content data of the topsoil layer collected by soil moisture sensors located at the center of the field. Future weather forecasts refer to meteorological data released by the meteorological platform regarding rainfall, evaporation, temperature, and other meteorological factors affecting soil moisture during the future cycle. The formula for estimating the initial moisture content difference is as follows: In the formula For the difference in initial water content in future cycles, To estimate the average volumetric moisture content of the first plot of land in the future cycle, based on current moisture content and future weather conditions, To correspond to the estimated average volumetric water content of the topsoil layer at the beginning of the future cycle for the second plot, the above four key parameters were then integrated into the predictive input feature vector required by the model. The feature vector is then input into the trained water and fertilizer cross-boundary prediction model. The model performs forward inference calculations using its internally learned mapping function, and finally outputs the predicted cross-boundary water migration amount. This cross-boundary water migration amount is a quantitative value used to accurately reflect the total net migration of water from one field to another within a future period. A positive value indicates a net migration of water from the second field to the first field, while a negative value indicates a net migration of water from the first field to the second field. This provides a basis for predicting water migration for the reverse correction of subsequent irrigation and fertilization plans for the first field.

[0020] After completing the prediction and inference of cross-boundary water migration between adjacent fields in future cycles, the accurate prediction of cross-boundary nutrient diffusion can be carried out simultaneously based on the same pre-trained water and fertilizer cross-boundary prediction model. This model is a neural network model trained and converged through supervised learning on a cross-boundary feature dataset, and has fully learned the mapping relationship between irrigation and fertilization differences, soil properties, and nutrient cross-boundary diffusion. It can accurately capture the lateral diffusion pattern of nutrients carried by water carriers between adjacent fields. In specific implementation, the planned parameters for the first and second fields in the future cycle are first extracted from the farmland irrigation and fertilization plan data. The difference in the expected fertilization amount is calculated first. This difference is an indicator that quantifies the fertilization gradient between the two fields and drives the lateral diffusion of nutrients. The calculation formula is as follows: ,in This is the difference in the expected amount of fertilizer applied. This is the planned fertilizer application amount for the first field in the future cycle. For the planned fertilizer application amount in the second field over the future cycle, the difference in expected irrigation amount is calculated simultaneously. Irrigation water is the carrier of nutrient molecule diffusion, and this difference directly determines the rate and intensity of nutrient diffusion. The calculation formula is as follows: ,in This represents the difference in expected irrigation volume. The planned irrigation amount for the first plot over the future cycle. For the planned irrigation amount of the second field in the future cycle, the pre-calculated quantitative index of soil texture differences will be used again. This indicator is a constant parameter calculated using the Euclidean distance between the percentages of sand, silt, and clay particles in the topsoil of two fields. It determines soil pore space and nutrient diffusion resistance, eliminating the need for real-time recalculation. Simultaneously, it combines current standard sensor data within the field with future weather forecast data to estimate the initial moisture content difference at the start of a future cycle. The current standard sensor data is the real-time volumetric moisture content of the topsoil collected by the soil sensor at the field center. The future weather forecast includes key meteorological data affecting soil moisture, such as future rainfall, evaporation, and temperature. The formula for estimating the initial moisture content difference is... ,in For the difference in initial water content in future cycles, To determine the average volumetric moisture content of the first tillage layer in the future cycle of the first plot by combining real-time data and meteorological forecasts, To correspond with the estimated values ​​for the second field, the four parameters—difference in expected fertilizer application, difference in expected irrigation, difference in soil texture, and difference in initial moisture content—were then integrated into the model's standard input feature vector. The vector is then input into the trained water and fertilizer cross-boundary prediction model. The model performs forward inference calculations using its internally fixed mapping function, and finally outputs the cross-boundary nutrient diffusion amount. This value is a precise quantitative prediction result, used to intuitively reflect the total net diffusion of major farmland nutrients such as nitrogen, phosphorus, and potassium from one field to another. A positive value represents the net diffusion of nutrients from the second field to the first field, and a negative value represents the net diffusion of nutrients from the first field to the second field. It provides a basis for the cross-boundary loss / gain of nutrients for the reverse correction of subsequent field fertilization target amounts.

[0021] After completing the training and convergence of the cross-boundary water and fertilizer prediction model and verifying the cross-boundary water and nutrient migration prediction logic, when making precise irrigation and fertilization decisions for the first field, it is necessary to simultaneously obtain the irrigation and fertilization plan data of the adjacent second field to construct the collaborative input conditions required for model prediction. The specific implementation process closely connects the preliminary data foundation and model application logic. First, it is clarified that the farmland management IoT platform is an integrated smart agriculture management platform that integrates real-time data from field sensors, irrigation and fertilization plans, and equipment control commands. It is also the data entry point for querying the confirmed water and fertilizer plans of adjacent fields. The future cycle refers to the unified execution period covered by this irrigation and fertilization decision. It is necessary to ensure that the planning cycles of the first and second fields are completely consistent to avoid prediction distortion due to time series deviations. The key time points for determining the irrigation and fertilization plan for the first field are used to query the approved and soon-to-be-implemented irrigation and fertilization plans for the second field within the same future period in real time through the farmland management IoT platform. These plans clearly include the planned irrigation and fertilization targets for the second field. The irrigation target is the total soil moisture requirement to be replenished for the second field, and the fertilization target is the total nitrogen, phosphorus, and potassium requirement to be applied for the second field. Simultaneously, initial irrigation and fertilization targets are first determined for the first field, without considering cross-boundary impacts. These targets are based on the crop's water and fertilizer requirements and the soil's basic fertility, resulting in a basic water and fertilizer supply. The difference between the initial irrigation targets for the first field and the irrigation targets for the second field is calculated using the following formula: ,in The difference between planned irrigation amounts, This is the initial irrigation target amount for the first field. The irrigation target amount for the second field is calculated by subtracting the initial fertilization target amount for the first field from the fertilization target amount for the second field. The calculated difference is given by the following formula: ,in The difference in planned fertilizer application rate, This is the initial fertilizer application target for the first field. The target fertilization amount for the second field was then determined, and this was combined with the quantified soil texture differences between the two fields that had already been measured. This index is a constant attribute parameter calculated using the Euclidean distance of the mass percentages of sand, silt, and clay, as well as a quantitative indicator of the difference in initial moisture content of the topsoil between the two fields at the current moment. Its calculation formula is , The average volumetric moisture content of the topsoil layer in the first field. The average volumetric moisture content of the topsoil layer in the second field is calculated, and the difference in planned irrigation amount is then calculated. Difference in planned fertilizer application Quantitative indicators of soil texture differences Quantitative indicators of initial moisture content difference These four parameters are combined into a complete set of collaborative input conditions according to the model input standard. These collaborative input conditions are then accurately input into the trained water and fertilizer cross-boundary prediction model, providing standardized and high-precision input data for subsequent predictions of cross-boundary water migration and cross-boundary nutrient diffusion, and supporting the accurate reverse correction of the irrigation and fertilization scheme for the first field.

[0022] After inputting the cross-boundary water and fertilizer prediction model with collaborative input conditions and completing inference calculations to obtain the corresponding predicted cross-boundary water migration and predicted cross-boundary nutrient diffusion, the irrigation and fertilization targets for the first field can be reverse-corrected based on these predicted cross-boundary amounts. This reverse correction addresses the water and fertilizer gains or losses caused by cross-boundary migration between adjacent fields, and modifies the initial water and fertilizer plan for the first field in reverse. Ultimately, this generates a precise execution plan that closely matches the actual water and fertilizer distribution in the field. In specific implementation, the predicted cross-boundary water migration output by the model is first directly defined as the amount of water that the first field will gain or lose due to cross-boundary effects under the planned execution conditions. This value is a positive value. The table shows the net water gain from the first field to the second field. A negative value indicates a net water loss from the first field to the second. The predicted transboundary nutrient diffusion output from the model is defined as the amount of nutrients the first field will gain or lose due to transboundary diffusion. A positive value indicates the first field will gain net nitrogen, phosphorus, and potassium from the second field, while a negative value indicates a net nutrient loss. Subsequently, precise correction of the irrigation target is performed. The initial irrigation target is the basic irrigation amount determined for the first field based on crop water requirements and basic soil moisture conditions, without considering transboundary effects. The correction follows the logic of "reducing irrigation when water is gained and increasing irrigation when water is lost," and the specific correction formula is as follows: In the formula This is the target irrigation amount for the first field after cross-boundary impact correction. This is the initial irrigation target amount for the first field. For the model output, the predicted transboundary water migration, when When the value is positive, the water obtained is subtracted from the initial irrigation target to avoid over-irrigation due to cross-boundary water replenishment; when... When the value is negative, subtracting the negative value is equivalent to increasing the irrigation amount to compensate for water loss caused by boundary crossing. Simultaneously, the fertilization target amount is adjusted. The initial fertilization target amount is the basic fertilization amount determined for the first field based on crop nutrient requirements and soil fertility, without considering the impact of boundary crossing. The adjustment follows the logic of "reducing fertilization when fertilizer is available and increasing fertilization when fertilizer is lost." The specific adjustment formula is as follows: In the formula This is the target fertilizer application amount for the first field after cross-boundary impact correction. This is the initial fertilizer application target for the first field. For the predicted transboundary nutrient diffusion output by the model, when When the value is positive, subtract this portion from the initial fertilizer target amount to obtain the nutrient amount, avoiding over-fertilization due to cross-regional fertilization; when When the value is negative, subtracting the negative value is equivalent to increasing the amount of fertilizer, thereby making up for the nutrient loss caused by cross-boundary migration. Finally, the corrected irrigation target amount, fertilizer target amount, and execution information such as irrigation time, water and fertilizer concentration, and equipment operating parameters are integrated to generate a complete irrigation and fertilization plan after cross-boundary impact correction. This plan completely eliminates the supply error caused by the lateral migration and diffusion of water and fertilizer between adjacent fields, and realizes the precise control and efficient utilization of farmland water and fertilizer.

[0023] Next, we will take Experiment 1: Accuracy Validation of Water and Fertilizer Cross-Boundary Prediction Model – Taking a Typical Pair of Adjacent Fields as an Example: To verify the ability of the cross-boundary water and fertilizer prediction model constructed in this invention to quantify the lateral migration of water and lateral diffusion of nutrients between adjacent fields, we selected five representative pairs of adjacent fields (Pair-01 to Pair-05) from large-scale contiguous farmland in the North China Plain, covering winter wheat, summer maize, rice, soybean, and cotton planting systems, respectively. The soil texture difference index for each pair of fields (…) ) range is The difference in planned irrigation volume ( ) between m³ / mu, fertilizer application rate difference ( ) between kgN / mu, initial moisture content difference ( ) between First, follow the steps of this invention. Soil moisture and nutrient sensor arrays were deployed at the field boundaries to continuously collect cross-boundary characteristic data. This data was then combined with records from the integrated water and fertilizer system and internal standard sensor data to construct a cross-boundary characteristic dataset. Subsequently, a supervised learning method was used to train a water and fertilizer cross-boundary prediction model based on this dataset (step S3). To evaluate model performance, the actual monitored cross-boundary migration amounts for each field over three consecutive irrigation cycles were compared with the model predictions. The results are summarized in Table 1. Table 1: Accuracy Validation Results of the Water and Fertilizer Cross-Boundary Prediction Model

[0024] As shown in Table 1, the cross-boundary water and fertilizer prediction model trained in this invention exhibits extremely high prediction accuracy across all five pairs of fields with different soil textures and varying water and fertilizer plans. The maximum prediction error for water migration is only 4.7% (Pair-04), the minimum is only 1.8% (Pair-03), and the average error is approximately 3.1%. The maximum prediction error for nutrient diffusion is 5.9%, with an average of approximately 4.5%. This result demonstrates that the cross-boundary water and fertilizer prediction model can accurately capture the nonlinear mapping relationship between differences in irrigation and fertilization amounts, soil texture differences, and initial moisture content differences, and cross-boundary water and fertilizer migration. Compared to traditional empirical formulas or practices that ignore cross-boundary effects in single fields, this invention achieves, for the first time, dynamic quantitative prediction of lateral water migration and nutrient diffusion between adjacent fields in contiguous farmland. This provides a reliable basis for the reverse correction of subsequent irrigation and fertilization schemes, significantly improving the scientific rigor and operability of cross-boundary water and fertilizer regulation.

[0025] This invention also introduces Experimental Example 2: Comparison of irrigation and fertilization effects after cross-boundary correction—taking a typical field affected by water and fertilizer gain / loss as an example: To verify the actual improvement effect of the reverse correction method described in step S4 of this invention on irrigation and fertilization decisions for single plots of contiguous farmland, we selected 5 typical plots (plots) in the same large-scale planting area. These fields exhibit varying degrees of cross-boundary water gain (field A), water loss (field B), nutrient gain (field C), nutrient loss (field D), and significant differences in sandy loam soil texture (field E). Initial irrigation and fertilization plans were first formulated using a traditional single-field decision-making model (based solely on soil, crop, and meteorological data, without considering water and fertilizer plans for adjacent fields). Subsequently, the method of this invention was used to obtain planning data from adjacent fields, which was then input into a pre-trained cross-boundary water and fertilizer prediction model to obtain the predicted cross-boundary water migration. With cross-boundary nutrient diffusion The corrected scheme was generated according to the reverse correction formula. The actual irrigation amount, crop water requirement satisfaction, fertilizer utilization rate and non-point source pollution risk index were tracked and recorded throughout the growing season. The comparison results are shown in Table 2. Table 2: Comparison of Irrigation and Fertilization Effects Before and After Cross-boundary Correction

[0026] Table 2 clearly shows that for field A, which received water benefits, the corrected irrigation amount decreased from 68.5 m³ / mu to 62.3 m³ / mu, avoiding waste caused by cross-boundary water replenishment. For field B, which suffered water losses, the irrigation amount increased from 68.5 m³ / mu to 75.2 m³ / mu, effectively compensating for water loss. Crop water requirement satisfaction increased from 86.3% and 82.1% to 95.7% and 94.2%, respectively, an average increase of approximately 10.8 percentage points. For field C, which received nutrient benefits, the fertilizer application rate decreased from 22.5 kgN / mu to 19.8 kgN / mu, while for field D, which suffered nutrient losses, it increased from 22.5 kgN / mu to 25.3 kgN / mu. The corrected fertilizer utilization rate increased on average from 61.8% to 74.9%, an increase of over 13%. The water use efficiency improvement rate was... Between then and now, the non-point source pollution risk index has decreased by [percentage missing]. This indicates that the present invention significantly reduces nitrogen and phosphorus loss caused by excessive water and fertilizer, and effectively controls agricultural non-point source pollution. In summary, the present invention truly realizes intelligent water and fertilizer management of contiguous farmland through cross-boundary effect quantification and reverse correction.

[0027] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A smart management method for farmland based on Internet of Things big data, characterized in that: Includes the following steps: S1. A boundary sensor array is deployed in the boundary area between adjacent first and second fields. The boundary sensor array includes at least one soil moisture sensor and at least one soil nutrient sensor to collect cross-boundary characteristic data reflecting the lateral migration of water and lateral diffusion of nutrients between the first and second fields within the boundary area. S2. Obtain irrigation and fertilization records, standard sensor data within the fields, and cross-boundary feature data containing the first and second fields, and construct a cross-boundary feature dataset. The cross-boundary feature dataset is used to characterize the correlation between the water and fertilizer cross-boundary effect between adjacent fields and the differences in irrigation amount, fertilizer amount, soil texture, and initial moisture content. S3. Train a water and fertilizer cross-border prediction model based on the cross-border feature dataset. The water and fertilizer cross-border prediction model is used to predict the cross-border water migration and cross-border nutrient diffusion in future cycles based on the irrigation and fertilization plan data of adjacent fields. The cross-border water migration and cross-border nutrient diffusion together constitute the predicted cross-border amount. S4. When making irrigation and fertilization decisions for the first field, obtain the current irrigation and fertilization plan data of the second field and input it into the water and fertilizer cross-boundary prediction model to obtain the predicted cross-boundary water migration and the predicted cross-boundary nutrient diffusion. Based on the predicted cross-boundary amounts, perform reverse correction on the irrigation target amount and fertilization target amount of the first field to generate an irrigation and fertilization scheme after cross-boundary influence correction.

2. The method for intelligent management of farmland big data based on the Internet of Things according to claim 1, characterized in that: The boundary region is a strip-shaped area extending horizontally to both sides with a specific width, based on the center line of the common field ridge between the first and second fields. The width of the strip-shaped area is not less than the typical horizontal extension range of the root system of a single field. The soil moisture sensor and soil nutrient sensor in the boundary sensor array are arranged in a staggered manner within the strip-shaped area. The soil moisture sensor works by detecting the capillary water connectivity between soil layers at different depths in adjacent fields, and the soil nutrient sensor works by detecting the changes in the ion concentration gradient in the soil solution within the strip-shaped area. Soil moisture content gradient data reflecting the lateral migration of water and soil electrical conductivity gradient data reflecting the lateral diffusion of nutrients are collected. The soil moisture content gradient data and soil electrical conductivity gradient data together constitute the cross-boundary feature data.

3. The method for intelligent management of farmland big data based on the Internet of Things according to claim 1, characterized in that: The acquisition of irrigation and fertilization records, standard sensor data within the fields, and cross-boundary feature data for the first and second fields is carried out through an Internet of Things data aggregation platform in a unified time series. The irrigation and fertilization records are from the logs reported by the controller of the water and fertilizer integrated machine deployed at the irrigation head of the field. The standard sensor data within the fields are data collected by the standard weather station and soil sensor respectively located at the geometric center of the first and second fields. The specific process for constructing the cross-boundary feature dataset is as follows: Heterogeneous data from the fertigation unit controller, standard weather station, soil sensor, and boundary sensor array are timestamped and cleaned. Standard sensor data within the field within a specific time window before and after the event are extracted as environmental status. The difference in irrigation amount and fertilizer amount related to the event are extracted as planning input. The soil texture difference quantification index and the initial moisture content difference quantification index before the event are extracted as soil background. Simultaneously, the change characteristics of soil moisture content gradient data and soil electrical conductivity gradient data collected by the boundary sensor array within the corresponding time window after the event are extracted as effect output. Finally, the environmental state, planned input, soil background, and effect output are associated and encapsulated to form a multidimensional data sample, which is composed of all historical multidimensional data samples to form the cross-boundary feature dataset.

4. The method for intelligent management of farmland big data based on the Internet of Things according to claim 3, characterized in that: The cross-boundary feature dataset represents the correlation through multidimensional data samples. Each multidimensional data sample takes the differences in irrigation amount, fertilizer amount, soil texture, and initial moisture content as a set of collaborative input conditions and the effect output as a set of coupled response results. The representation is completed through the statistical regularities contained in a large number of samples in the multidimensional dataset.

5. The method for intelligent management of farmland big data based on the Internet of Things according to claim 4, characterized in that: The soil texture difference quantification index is obtained by comparing the mass percentages of sand, silt, and clay particles in the topsoil of the first and second plots and calculating their vector distance. The initial moisture content difference quantification index is obtained by comparing the relative difference in the average volumetric moisture content of the topsoil of the first and second plots before the irrigation and fertilization events. The soil texture difference quantification index and the initial moisture content difference quantification index are combined with the differences in irrigation amount and fertilization amount to form the collaborative input conditions.

6. The method for intelligent management of farmland big data based on the Internet of Things according to claim 5, characterized in that: The training of the water and fertilizer cross-boundary prediction model based on the cross-boundary feature dataset is carried out using a supervised learning method. The specific process is as follows: The collaborative input conditions of each multidimensional data sample in the cross-boundary feature dataset are used as the input feature vector of the water and fertilizer cross-boundary prediction model, and the effect output of each multidimensional data sample is used as the target output vector for training the water and fertilizer cross-boundary prediction model. The initial neural network model is iteratively trained using historical multidimensional data samples. The internal parameters of the water and fertilizer cross-boundary prediction model are adjusted through optimization algorithms to learn the mapping function between the cooperative input conditions and the effect output, and finally a well-trained water and fertilizer cross-boundary prediction model is obtained.

7. The method for intelligent management of farmland big data based on the Internet of Things according to claim 1, characterized in that: The water and fertilizer cross-boundary prediction model is used to predict the cross-boundary water migration in future cycles based on irrigation and fertilization plan data of adjacent fields. Specifically, it uses the trained water and fertilizer cross-boundary prediction model and inputs the difference in expected irrigation amount, difference in expected fertilization amount, difference in soil texture, and difference in initial water content at the beginning of the future cycle estimated based on current standard sensor data and future weather forecasts between the first and second fields in the irrigation and fertilization plan data. The water and fertilizer cross-boundary prediction model outputs the predicted cross-boundary water migration from one field to another. The cross-boundary water migration reflects the total net migration of water from one field to another.

8. The method for intelligent management of farmland big data based on the Internet of Things according to claim 1, characterized in that: The water and fertilizer cross-boundary prediction model is used to predict cross-boundary nutrient diffusion in future cycles based on irrigation and fertilization plan data from adjacent fields. Specifically, it includes: Using a trained water and fertilizer cross-boundary prediction model, the model takes into account the difference in expected fertilization amount, the difference in expected irrigation amount, the difference in soil texture, and the difference in initial water content at the start of the future cycle between the first and second plots in the irrigation and fertilization plan data. The model outputs the predicted cross-boundary nutrient diffusion amount from one plot to another. The cross-boundary nutrient diffusion amount is used to reflect the total net diffusion of major nutrients from one plot to another.

9. The method for intelligent management of farmland big data based on the Internet of Things according to claim 1, characterized in that: When making irrigation and fertilization decisions for the first field, obtaining the current irrigation and fertilization plan data for the second field specifically includes: At the time point when the irrigation and fertilization plan for the first field is formulated, the irrigation and fertilization plan for the second field within the same future period is queried through the farmland management IoT platform. The irrigation and fertilization plan includes the planned irrigation target amount and fertilizer target amount for the second field. The initial irrigation target amount and fertilizer target amount proposed for the first field are subtracted from the corresponding planned values ​​for the second field to obtain the planned irrigation amount difference and planned fertilization amount difference. Combined with the current soil texture difference quantification index and the initial moisture content difference quantification index of the two fields, a set of collaborative input conditions for prediction is formed, and the collaborative input conditions are input into the water and fertilizer cross-boundary prediction model.

10. The method for intelligent management of farmland big data based on the Internet of Things according to claim 9, characterized in that: The reverse correction of the irrigation and fertilization targets for the first field based on the predicted cross-boundary amount specifically includes: The predicted cross-boundary water migration output by the water and fertilizer cross-boundary prediction model is regarded as the amount of water that the first field will gain or lose under the planned conditions, and the predicted cross-boundary nutrient diffusion is regarded as the amount of nutrients that the first field will gain or lose. The irrigation and fertilization plan, corrected for cross-boundary effects, is generated by subtracting the additional water gained or adding the additional water lost from the initial irrigation target for the first field, or by subtracting the additional nutrients gained or adding the additional nutrients lost from the initial fertilization target for the first field.

Images