A Smart Agriculture Soil Improvement and Optimization Method Based on Big Data

Abstract

This invention relates to the field of intelligent agricultural data processing and business forecasting management technology, specifically a smart agricultural soil improvement and optimization method based on big data. The method includes: acquiring real-time monitoring data and fusion data from laboratories to create a soil digital twin model within a unified management framework; simulating the impact of fertilization and tillage and outputting predicted values ​​for the spatiotemporal evolution of soil nutrients; establishing an improvement benefit model, inputting the predicted values ​​and climate factors into a risk-weighted utility function, and outputting an improvement scheme with the optimal service expenditure ratio; creating digital identity cards for agricultural products, and feeding back end-market premium data to the decision-making process to dynamically adjust the benefit model. This invention deeply integrates digital twin simulation with risk-economic utility assessment, utilizing business premiums to form a management closed loop. While effectively mitigating agricultural input risks, it accurately achieves dynamic optimization of agricultural assets, improving the overall efficiency of existing soil improvement schemes.

Description

Technical Field

[0001] This invention relates to the field of intelligent agricultural data processing and business forecasting management technology, specifically to a smart agricultural soil improvement and optimization method based on big data. Background Technology

[0002] With the development of smart agriculture, big data technology is playing a vital role in precision agricultural management. Soil improvement, as a key link in enhancing agricultural product quality, optimizing agricultural asset allocation, and improving overall commercial economic efficiency, is increasingly attracting the attention of agricultural operators.

[0003] Current soil improvement management relies heavily on localized experience or single monitoring data for fertilization planning, lacking in-depth integration of multi-dimensional data and comprehensive spatiotemporal analysis. More importantly, existing improvement strategies often disregard economic factors, failing to incorporate commercial management elements such as climate risk and service expenditure ratios, making it difficult to control the input-output ratio of soil improvement. Furthermore, an information gap exists between upstream soil improvement inputs and downstream agricultural product market sales, making it impossible to use agricultural product market premiums as a core feedback indicator to dynamically guide and review upstream agricultural asset allocation, hindering the formation of a modern commercial closed loop for agricultural management.

[0004] The technical problem to be solved by this invention is that existing soil improvement methods lack accurate spatiotemporal prediction and risk-economic utility assessment, and cannot utilize the premium feedback from the agricultural product market to dynamically optimize resource allocation strategies, resulting in a lack of scientific rigor and economic efficiency in agricultural business decisions.

[0005] Therefore, a smart agriculture soil improvement and optimization method based on big data is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a smart agriculture soil improvement and optimization method based on big data, which achieves dynamic economic optimization of improvement decisions through digital twin extrapolation and market premium feedback.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A smart agriculture soil improvement and optimization method based on big data includes: Acquire real-time monitoring data to obtain soil quality improvement indicators, and bind the real-time monitoring data with geographic information spatial coordinates to form a single-map management architecture; Acquire laboratory sampling data, use co-kriging algorithm to fuse laboratory sampling data with real-time monitoring data in multiple sources and at multiple scales, generate digital soil map in a single map management architecture, and create digital twin model based on digital soil map; By using a digital twin model and combining it with the spatial coordinate attributes in a one-graph management architecture, the soil nutrient diffusion process after fertilization and the impact of tillage intensity on soil bulk density are simulated, and the predicted values ​​of soil nutrient spatiotemporal evolution, including soil nutrient gap and nutrient migration trend, are output. Establish a soil improvement benefit model, inputting the predicted values ​​of soil nutrient spatiotemporal evolution and climate probability factors into a risk-weighted utility function to simulate and output a soil improvement scheme with the optimal service expenditure ratio; execute the soil improvement scheme and generate soil improvement records in real time in a one-map management architecture. Establish digital identity cards for agricultural products produced from the soil, link soil improvement records and soil quality improvement indicators with the digital identity cards, feed agricultural product premium data back to the decision-making process, and dynamically adjust the soil improvement benefit model.

[0008] Preferably, the real-time monitoring data includes: soil moisture collected using a capacitive sensor, conductivity collected using a time-domain reflectometer, pH value collected using an ion-selective electrode, nitrogen, phosphorus, and potassium content collected using near-infrared spectroscopy, and rainfall collected using an automatic weather station; and the real-time monitoring data is cleaned and standardized using a multi-source heterogeneous data interface integration protocol.

[0009] Preferably, the soil quality improvement indicators include: Acquire vegetation physiological data and calculate initial feature values ​​from the vegetation physiological data; the initial feature values ​​include chlorophyll content, vegetation index, and water deficit index; use fuzzy C-means clustering algorithm to perform spatial differential clustering analysis on the initial feature values ​​and the real-time monitoring data to calculate the soil nutrient variation coefficient reflecting the differences in nutrient distribution in the production area, and jointly determine the initial feature values ​​and the soil nutrient variation coefficient as the soil quality improvement index; use a geographic information system to map the real-time monitoring data and the soil quality improvement index to corresponding spatial geographic coordinates through attribute binding, and construct the one-map management architecture.

[0010] Preferably, the laboratory sampling data includes: The process of obtaining a digital soil map through multi-source, multi-scale fusion of soil pH, nitrogen, phosphorus, potassium content, organic matter content, and heavy metal concentration indicators measured in the laboratory includes: acquiring the laboratory sampling data and the real-time monitoring data; using the co-kriging algorithm, with the real-time monitoring data as an auxiliary variable, performing spatial interpolation processing on the laboratory sampling data with discrete spatial coordinate attributes; aligning the interpolated laboratory attribute values ​​with the real-time monitoring data in a single-map management architecture to generate a digital soil map with a preset grid resolution.

[0011] Preferably, the digital twin model includes a data mapping layer and a mechanism simulation layer: The data mapping layer is used to align and map the historical agricultural activity trajectory data, real-time monitoring data, and laboratory sampling data pre-stored in the single-map management architecture in a spatiotemporal dimension, and use these as input variables for the mechanism simulation layer. A nutrient kinetic prediction model is established in the mechanism simulation layer using a long short-term memory network. Current nutrient content data, fertilizer input data, leaching environment driving factor data determined by rainfall intensity, and crop absorption coefficient data are acquired and input into the nutrient kinetic prediction model for simulation calculation to obtain future nutrient content data after a preset time period. The future nutrient content data is compared with the corresponding standard nutrient threshold data, and the difference between the two is extracted as the soil nutrient gap. The mechanism simulation layer is used to simulate the nutrient migration vector direction in the soil profile to extract the nutrient migration trend, thus forming the predicted value of soil nutrient spatiotemporal evolution.

[0012] Preferably, the soil improvement benefit model includes a decision-making unit and an execution feedback unit: The decision-making unit retrieves the predicted values ​​of soil nutrient spatiotemporal evolution and, in conjunction with a preset climate probability factor, inputs a risk-weighted utility function. The risk-weighted utility function then evaluates the benefit weights between increasing investment in drainage facilities and increasing the application of quick-acting fertilizers. Based on the evaluation results of these benefit weights, the optimal service expenditure ratio is determined, and the recommended fertilization intensity is calculated using the nutrient balance method. The optimal service expenditure ratio is correlated with the recommended fertilization intensity, outputting a soil improvement scheme that includes a variable fertilization prescription map and a service-oriented management strategy. A variable controller and a Beidou navigation terminal are used to execute the variable fertilization prescription map in the soil improvement scheme, capturing response time data, flow accuracy data, and geographic coordinate location data in real time during execution. The response time data, flow accuracy data, and geographic coordinate location data are then structurally integrated to generate a soil improvement record that records the actual fertilization trajectory, fertilizer dosage, and operation time.

[0013] Preferably, the specific process of establishing digital identity cards and dynamically adjusting the soil improvement benefit model includes: A distributed traceability node is established using blockchain technology, and a unique identification code is generated for the agricultural products produced from the current plot. The soil improvement records and soil quality improvement indicators stored in the single-map management architecture are encapsulated as quality metadata. The quality metadata and the unique identification code are hashed and associated using an asymmetric encryption algorithm and stored in the distributed traceability node to construct the digital ID card. The market transaction price of agricultural products with the digital ID card at the sales terminal is collected in real time and compared with the preset benchmark price of conventional agricultural products. The difference between the two is calculated to obtain the agricultural product premium data. The agricultural product premium data is fed back to the decision unit in the soil improvement benefit model to correct the benefit weight parameters in the risk-weighted utility function, thereby completing the dynamic adjustment of the soil improvement benefit model.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention integrates multi-source data from real-time monitoring and laboratory sampling using a co-kriging algorithm, and constructs a digital twin model using a long short-term memory network, breaking through the limitations of traditional static and single monitoring. This model can accurately simulate the migration trends of nutrients after fertilization, such as diffusion and leaching, quantify soil nutrient gaps in advance, improve the accuracy of panoramic spatiotemporal evolution prediction, and provide scientific support for subsequent refined agricultural resource input.

[0015] 2. To address the shortcomings of existing technologies that prioritize agronomic indicators over economic attributes, this invention establishes a soil improvement benefit model. It inputs spatiotemporal predictions and climate probability factors into a risk-weighted utility function to scientifically assess the benefit weight between infrastructure investment and fertilizer consumption. By integrating climate risk and financial cost calculations, it outputs the optimal service expenditure ratio and improvement plan, effectively avoiding the commercial risks of blind investment and achieving precise control over the agricultural input-output ratio.

[0016] 3. This invention utilizes blockchain technology to create digital identity cards for agricultural products, hashing the front-end soil improvement process with the back-end end-market transaction performance. The system calculates agricultural product premium data in real time and transmits it back to the decision-making unit as a core economic feedback indicator, dynamically adjusting the revenue weight parameters. This mechanism eliminates the information gap between soil input and market returns, ensuring that front-end agricultural management and improvement strategies always align with market value laws. Attached Figure Description

[0017] Figure 1 This is a flowchart of a smart agriculture soil improvement and optimization method based on big data proposed in this invention. Figure 2 This is a flowchart illustrating the process structure of a smart agriculture soil improvement and optimization method based on big data proposed in this invention. Figure 3This is a structural diagram of the digital twin model proposed in this invention. Detailed Implementation

[0018] 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.

[0019] Example 1 Please see Figures 1 to 3 This invention provides a smart agriculture soil improvement and optimization method based on big data, the technical solution of which is as follows: A smart agriculture soil improvement and optimization method based on big data, such as... Figures 1-2 As shown, it includes: Acquire real-time monitoring data to obtain soil quality improvement indicators, and bind the real-time monitoring data with geographic information spatial coordinates to form a single-map management architecture; Acquire laboratory sampling data, use co-kriging algorithm to fuse laboratory sampling data with real-time monitoring data in multiple sources and at multiple scales, generate digital soil map in a single map management architecture, and create digital twin model based on digital soil map; By using a digital twin model and combining it with the spatial coordinate attributes in a one-graph management architecture, the soil nutrient diffusion process after fertilization and the impact of tillage intensity on soil bulk density are simulated, and the predicted values ​​of soil nutrient spatiotemporal evolution, including soil nutrient gap and nutrient migration trend, are output. Establish a soil improvement benefit model, inputting the predicted values ​​of soil nutrient spatiotemporal evolution and climate probability factors into a risk-weighted utility function to simulate and output a soil improvement scheme with the optimal service expenditure ratio; execute the soil improvement scheme and generate soil improvement records in real time in a one-map management architecture. Establish digital identity cards for agricultural products produced from the soil, link soil improvement records and soil quality improvement indicators with the digital identity cards, feed agricultural product premium data back to the decision-making process, and dynamically adjust the soil improvement benefit model.

[0020] Furthermore, the real-time monitoring data includes: soil moisture collected using a capacitive sensor, conductivity collected using a time-domain reflectometer, pH value collected using an ion-selective electrode, nitrogen, phosphorus, and potassium content collected using near-infrared spectroscopy, and rainfall collected using an automatic weather station; and the real-time monitoring data is cleaned and standardized using a multi-source heterogeneous data interface integration protocol.

[0021] Specifically, the process of cleaning and standardizing the real-time monitoring data using a multi-source heterogeneous data interface integration protocol includes: Because the sampling frequencies of weather stations and various soil sensors differ, the timestamp features of the real-time monitoring data are extracted. A linear interpolation algorithm is used to align the monitoring data with different sampling frequencies to a preset time resolution (e.g., uniformly aligning them to once per hour). Subsequently, the Z-score algorithm is used to calculate the standard score of the aligned data, outlier data points with absolute values ​​greater than a preset anomaly threshold are removed, and the mean of adjacent normal data points is used for smoothing to obtain cleaned data. The preset anomaly threshold is an empirical value calculated based on the normal fluctuation variance of the plot over 30 consecutive days, and in this embodiment, it is set to 3.

[0022] After obtaining the cleaned data, since characteristic parameters such as soil moisture, electrical conductivity, and rainfall have different physical dimensions and significant numerical differences, a range standardization formula is used to process the cleaned data to generate a standard feature vector that is input into the co-kriging algorithm and digital twin model. The specific range standardization formula is as follows: ; In the formula, For the first Sensor-like sensors Data after cleaning at any time; For the corresponding standardized data; and These are the preset lower and upper physiological safety benchmark values ​​for this type of sensor within the current crop growth cycle.

[0023] This invention effectively overcomes the technical defects of inconsistent sampling frequencies and conflicts in physical dimensions of multi-source sensing devices. By aligning temporal features and cleaning outliers through linear interpolation, it ensures the temporal consistency and high reliability of the underlying data. At the same time, by using range standardization to completely eliminate the dimensional differences of multi-source data, it effectively avoids the computational bias caused by data magnitude imbalance to the co-kriging algorithm and digital twin model, thereby improving the stability and prediction accuracy of subsequent spatiotemporal extrapolation algorithms.

[0024] Furthermore, the soil quality improvement indicators include: Acquire vegetation physiological data and calculate initial feature values ​​from the vegetation physiological data; the initial feature values ​​include chlorophyll content, vegetation index, and water deficit index; use fuzzy C-means clustering algorithm to perform spatial differential clustering analysis on the initial feature values ​​and the real-time monitoring data to calculate the soil nutrient variation coefficient reflecting the differences in nutrient distribution in the production area, and jointly determine the initial feature values ​​and the soil nutrient variation coefficient as the soil quality improvement index; use a geographic information system to map the real-time monitoring data and the soil quality improvement index to corresponding spatial geographic coordinates through attribute binding, and construct the one-map management architecture.

[0025] Specifically, the process of acquiring vegetation physiological data and calculating initial feature values ​​from the vegetation physiological data includes: acquiring multispectral images collected at a preset flight altitude as the vegetation physiological data; extracting the near-infrared and red light band reflectance at the pixel level; and calculating the vegetation index using a normalized difference formula. The formula is: ; In the formula, For near-infrared reflectivity, This refers to the reflectivity in the red light band.

[0026] The initial feature values ​​and the real-time monitoring data were subjected to spatial differential clustering analysis using the fuzzy C-means clustering algorithm. The soil nutrient variation coefficient, reflecting the differences in nutrient distribution within the production area, was calculated, including: The standardized real-time monitoring data is concatenated with the initial feature values ​​to construct a joint feature matrix. The preset number of clusters for the fuzzy C-means clustering algorithm is set to c, and the fuzzy index is set to m (in this embodiment, m=2). Iterative clustering is performed using the joint feature matrix until the objective function converges, outputting the membership matrix corresponding to the c spatial clusters. ,in Indicates the first The spatial sampling point belongs to the first Membership degree of a spatial cluster; For each spatial cluster Extract soil nitrogen, phosphorus, and potassium content data from the real-time monitoring data. Using the membership matrix as statistical weights, the weighted average nutrient content of the cluster is calculated. Compared with weighted nutrient standard deviation The coefficient of variation of soil nutrients was then calculated. The calculation formula is as follows:

[0027]

[0028] ; In the formula, This represents the total number of spatial sampling points within the current plot. For the first Soil nitrogen, phosphorus, and potassium content data from each spatial sampling point; [and the corresponding spatial clusters] The combination constitutes the set of soil nutrient variation coefficients that characterize the differences in distribution across the entire region.

[0029] The soil quality improvement indicators are organized and stored in computer memory in the form of one-dimensional feature vectors. For any j-th spatial grid in the one-map management architecture, its corresponding soil quality improvement indicator vector... Expressed as: ; In the formula, This represents the normalized chlorophyll content. The vegetation index, The water deficit index; This is the soil nutrient variation coefficient for the k-th spatial cluster to which this grid belongs. This feature vector... As a priori attribute of the static environment, it is bound to the spatial coordinates of the geographic information through a key-value pair data structure.

[0030] This invention uses the membership matrix output by a fuzzy clustering algorithm as statistical weights to derive the soil nutrient variation coefficient. This mechanism breaks through the limitations of hard boundaries in traditional spatial regionalization, accurately depicting the continuous and gradually changing heterogeneous distribution characteristics of soil nutrients in natural space. Combined with multispectral vegetation physiological data, it improves the accuracy of spatial difference assessment and the scientific rigor of the algorithm, providing a highly reliable decision-making foundation for the subsequent precise dynamic allocation of agricultural resources.

[0031] Furthermore, the laboratory sampling data includes: The process of obtaining a digital soil map through multi-source, multi-scale fusion of soil pH, nitrogen, phosphorus, potassium content, organic matter content, and heavy metal concentration indicators measured in the laboratory includes: acquiring the laboratory sampling data and the real-time monitoring data; using the co-kriging algorithm, with the real-time monitoring data as an auxiliary variable, performing spatial interpolation processing on the laboratory sampling data with discrete spatial coordinate attributes; aligning the interpolated laboratory attribute values ​​with the real-time monitoring data in a single-map management architecture to generate a digital soil map with a preset grid resolution.

[0032] Specifically, the process of spatial interpolation using the co-kriging algorithm includes: Soil moisture and nutrient characteristics from the real-time monitoring data are obtained as auxiliary variables. Nutrient content in the laboratory sampling data with discrete coordinates is obtained as the main variable. Construct a cross-variance function that reflects the spatial correlation between the main variables and auxiliary variables. The formula is: ; In the formula, For spatial lag distance, The distance is The number of sample pairs; the Kriging matrix equation is constructed using the cross-variation function, and the weight coefficients corresponding to each sampling point are calculated; the interpolated laboratory attribute values. The calculation formula is: ; In the formula, Let i be the weight of the i-th test sampling point. Let be the weight of the j-th real-time monitoring point, and satisfy . and Unbiased constraints.

[0033] The process of determining the preset grid resolution includes: obtaining the geographical span of the current plot, calculating the average proximity distance D of the laboratory sampling points, and setting the preset grid resolution to between 0.1D and 0.2D (for example, when the sampling interval is 50 meters, the resolution is set to 5m). (5m), to ensure that the digital soil map can capture the spatial heterogeneity of nutrients.

[0034] The spatial dimension alignment includes: resampling the real-time monitoring data transmitted at different frequencies to the same level as the digital soil using a bilinear interpolation algorithm. Figure 1 The spatial grid scale is consistent, and pixel-level overlay is performed through a unified spatial coordinate system (such as CGCS2000) in a single map management architecture.

[0035] This invention quantifies the spatial covariance relationship between laboratory sampling data and high-frequency monitoring data by introducing a co-kriging cross-variance function. This mechanism effectively compensates for the lack of accuracy caused by the sparseness of laboratory sampling points, and uses high-density real-time monitoring data as spatial constraints to achieve pixel-level precise alignment and fusion of multi-scale data, thereby improving the resolution and scientific accuracy of digital soil maps in characterizing nutrient heterogeneity.

[0036] Furthermore, such as Figure 3 As shown, the digital twin model includes a data mapping layer and a mechanism simulation layer: The data mapping layer is used to align and map the historical agricultural activity trajectory data, real-time monitoring data, and laboratory sampling data pre-stored in the single-map management architecture in a spatiotemporal dimension, and use these as input variables for the mechanism simulation layer. A nutrient kinetic prediction model is established in the mechanism simulation layer using a long short-term memory network. Current nutrient content data, fertilizer input data, leaching environment driving factor data determined by rainfall intensity, and crop absorption coefficient data are acquired and input into the nutrient kinetic prediction model for simulation calculation to obtain future nutrient content data after a preset time period. The future nutrient content data is compared with the corresponding standard nutrient threshold data, and the difference between the two is extracted as the soil nutrient gap. The mechanism simulation layer is used to simulate the nutrient migration vector direction in the soil profile to extract the nutrient migration trend, thus forming the predicted value of soil nutrient spatiotemporal evolution.

[0037] Specifically, a nutrient kinetics prediction model is established using long short-term memory networks and outputs future nutrient content data, including: Construct a three-dimensional input tensor with a time step T. ,in Batch size, The feature dimension; the feature dimension It is composed of aligned historical agricultural activity trajectories, real-time monitoring data, and laboratory sampling data. The three-dimensional input tensor is then input into the Long Short-Term Memory network for nonlinear feature extraction to obtain the hidden state vector. ; Batch size The value is 32, representing the total feature dimension after concatenation. For example, when D=12, it specifically includes 5 monitoring dimensions (soil moisture, electrical conductivity, pH value, near-infrared spectral estimates of nitrogen, phosphorus and potassium, and rainfall), 4 laboratory test dimensions (high-precision pH, high-precision nitrogen, phosphorus and potassium, organic matter, and heavy metals), and 3 agricultural activity event dimensions (base fertilizer application amount, topdressing amount, and tillage depth).

[0038] The mechanism is used to simulate the hidden state vector. Perform physical constraint corrections and output future nutrient content data after the preset time period. The preset duration is determined based on the crop's growth stage, typically 7 to 15 days into the future.

[0039] Data on environmental driving factors of leaching determined by rainfall intensity The calculation formula is: ; In the formula, For real-time monitoring of rainfall intensity; This is the preset soil infiltration rate threshold; This refers to the initial soil moisture content; and The preset rinsing coefficient is obtained by fitting historical rinsing experimental data.

[0040] The mechanism described above is used to simulate the nutrient transport vector direction in a soil profile, including: obtaining the coordinates of the current sampling point in the digital soil map. Nutrient predictions for the grid points and their adjacent grid points at future times , and ; Calculate the spatial gradient of nutrient prediction values ​​using the central difference method ; Based on the spatial coordinate attributes in the aforementioned single-map management architecture, calculate the migration vector direction corresponding to the nutrient migration trend. The formula is: ; In the formula, The effective diffusion coefficient is determined by the soil bulk density. The pore water velocity vector is determined by the rinsing environment driving factor.

[0041] This invention addresses the challenge of lacking physical constraints in soil environment extrapolation using purely data-driven models by deeply integrating Long Short-Term Memory (LSTM) networks with physical convection-diffusion mechanisms. It utilizes spatial gradient operators and leaching driving factors to calculate nutrient transport vectors in real time, achieving a leap from "static scalar monitoring" to "dynamic vector evolution." This not only improves the physical accuracy of future nutrient deficit predictions but also enables visualized and precise control of spatiotemporal nutrient changes through a single-map management architecture.

[0042] Furthermore, the soil improvement benefit model includes a decision-making unit and an execution feedback unit: The decision-making unit retrieves the predicted values ​​of soil nutrient spatiotemporal evolution and, in conjunction with a preset climate probability factor, inputs a risk-weighted utility function. The risk-weighted utility function then evaluates the benefit weights between increasing investment in drainage facilities and increasing the application of quick-acting fertilizers. Based on the evaluation results of these benefit weights, the optimal service expenditure ratio is determined, and the recommended fertilization intensity is calculated using the nutrient balance method. The optimal service expenditure ratio is correlated with the recommended fertilization intensity, outputting a soil improvement scheme that includes a variable fertilization prescription map and a service-oriented management strategy. A variable controller and a Beidou navigation terminal are used to execute the variable fertilization prescription map in the soil improvement scheme, capturing response time data, flow accuracy data, and geographic coordinate location data in real time during execution. The response time data, flow accuracy data, and geographic coordinate location data are then structurally integrated to generate a soil improvement record that records the actual fertilization trajectory, fertilizer dosage, and operation time.

[0043] By combining a preset climate probability factor with a risk-weighted utility function, the benefit weights between increasing investment in drainage facilities and increasing the application of quick-acting fertilizers are evaluated using the risk-weighted utility function, and the optimal service expenditure ratio is calculated, including: Obtain the preset climate probability factor The climate probability factor The probability of short-duration heavy rainfall leading to leaching during the crop growing season is calculated using an extreme value distribution model by analyzing the historical meteorological data of the target plot over the past 10 years. Assume the total budget for soil improvement of this plot is The service expenditure ratio is The investment amount used to increase drainage facilities is The investment amount used to increase the application of quick-acting fertilizers is Construct the risk-weighted utility function. Its mathematical expression is: ; In the formula, The yield increase coefficient per unit fertilizer base, determined by the predicted values ​​of soil nutrient spatiotemporal evolution. The sensitivity coefficient for the risk of fertilizer leaching due to rainfall. For the constant parameters of the risk resistance effectiveness of drainage facilities; function terms Characterizes the effective retention rate of fertilizer under current investment intervention in drainage facilities; The quarterly depreciation and amortization rate of drainage facilities is determined based on the pre-set facility life cycle. The risk sensitivity coefficient of fertilizer leaching caused by rainfall This coefficient characterizes the physical leaching rate of a target site under a specific rainfall intensity. It depends on the soil's basic texture and porosity, and is obtained by extracting soil samples from the target site and conducting standard soil column leaching experiments, then fitting a cumulative leaching curve. In this embodiment, for sandy loam soils with poor water and fertilizer retention... The preferred setting range is [1.2, 1.5]; for clay loam soils with strong water and fertilizer retention, The preferred setting range is [0.5, 0.8]. The constant parameter for the risk resistance effectiveness of the drainage facility. This represents the marginal physical utility of a unit investment in flood control projects in reducing the risk of fertilizer runoff. Its value is calculated based on pre-set flood control standards for drainage networks (such as a 5-year or 10-year return period) and local engineering cost quotas. The value quantifies the conversion efficiency of project funds into actual fertilizer retention rate; the higher the flood control level of the project, the higher the corresponding retention rate. The larger the value.

[0044] The risk-weighted utility function Seeking information about Taking the first derivative of the variable and setting it to zero, we find the independent variable that maximizes the expected utility, and use it as the optimal service expenditure ratio. .

[0045] The recommended fertilization intensity is calculated using the nutrient balance method, and the optimal service expenditure ratio is correlated with the recommended fertilization intensity, including: Based on the aforementioned nutrient balance method, a formula for calculating recommended fertilization intensity is constructed: ; In the formula, The recommended fertilization intensity; Fertilizer requirements for the current crop's target yield; The current available nutrient content in the predicted spatiotemporal evolution of soil nutrients; The amount of conventional nutrient loss due to environmental migration; To utilize the optimal service expenditure ratio The revised fertilizer utilization rate, and and They are positively correlated; using the above formula, the recommended fertilization intensity is output, which is dynamically adapted to the optimal service expenditure ratio.

[0046] This invention constructs a risk-weighted utility function that quantifies the probability of extreme weather rainfall and incorporates it into business decisions. It precisely calculates the optimal ratio of investment in drainage facilities and fertilizers to maximize economic benefits in the face of risk. Simultaneously, this economic ratio is integrated into the agronomic nutrient balance method, dynamically adjusting fertilizer utilization and recommended fertilization intensity, and is precisely executed using BeiDou terminals. This mechanism effectively establishes a closed-loop management system from "economic risk control" to "precise implementation of agricultural practices," improving the resource allocation efficiency of agricultural assets.

[0047] Furthermore, the specific process of establishing digital identity cards and dynamically adjusting the soil improvement benefit model includes: A distributed traceability node is established using blockchain technology, and a unique identification code is generated for the agricultural products produced from the current plot. The soil improvement records and soil quality improvement indicators stored in the single-map management architecture are encapsulated as quality metadata. The quality metadata and the unique identification code are hashed and associated using an asymmetric encryption algorithm and stored in the distributed traceability node to construct the digital ID card. The market transaction price of agricultural products with the digital ID card at the sales terminal is collected in real time and compared with the preset benchmark price of conventional agricultural products. The difference between the two is calculated to obtain the agricultural product premium data. The agricultural product premium data is fed back to the decision unit in the soil improvement benefit model to correct the benefit weight parameters in the risk-weighted utility function, thereby completing the dynamic adjustment of the soil improvement benefit model.

[0048] Specifically, the methods for obtaining the preset benchmark prices for conventional agricultural products include: Obtain a market listing dataset of similar conventional agricultural products located in the same provincial-level administrative region as the target plot, within the same listing cycle, and without the aforementioned digital identity card; use an exponentially weighted moving average algorithm to perform time-series smoothing processing on the market listing dataset, and calculate the preset benchmark price of the conventional agricultural products. .

[0049] Calculate the difference between the two to obtain agricultural product premium data. The calculation formula is as follows: ; In the formula, This refers to the real-time market transaction price of agricultural products bearing the aforementioned digital ID at the sales terminal.

[0050] Feeding agricultural product premium data back to the decision-making unit in the soil improvement benefit model, correcting the benefit weight parameters in the risk-weighted utility function, and completing the dynamic adjustment of the soil improvement benefit model, including: Retrieve the unit fertilizer-based yield increase coefficient from the risk-weighted utility function. As the return weight parameter to be corrected; construct a dynamic feedback update equation for the parameter based on the market premium ratio: ; In the formula, This is a dynamically adjusted update cycle number; This is the adjusted unit fertilizer base yield increase coefficient for the current cycle; The unit fertilizer base yield increase coefficient of the previous cycle; The preset premium feedback sensitivity constant is used to control the step size and learning rate of the return weight parameter adjustment across periods. Its specific setting basis and tuning method include: Obtain historical market price time series data of the target agricultural product over the past 3-5 years and calculate its price volatility variance; If the price volatility variance exceeds a preset threshold for severe volatility, it indicates that this type of agricultural product is highly susceptible to short-term speculation or unforeseen events. The value range is limited to 0.05-0.15 to avoid the front-end fertilization improvement decision from fluctuating wildly following short-term distorted market signals (i.e., to prevent model overfitting). If the price fluctuation variance is less than or equal to the drastic fluctuation threshold, it indicates that the market price of the agricultural product is relatively stable, then... The value is set between 0.20 and 0.35 to ensure that the agricultural decision-making model has sufficient agility and responsiveness to normalized market premium feedback, and to avoid lag in strategy adjustments.

[0051] In a preferred embodiment of the present invention, the smoothness of model convergence and the timeliness of response are considered together. The default setting value is 0.25.

[0052] Using the revised By replacing the original coefficients in the risk-weighted utility function, the optimal service expenditure ratio calculation for the next crop cycle is triggered, thereby achieving adaptive closed-loop optimization of the agricultural decision-making model based on sales terminal commercial data.

[0053] This invention utilizes blockchain technology to break down data barriers between production and sales, transforming soil improvement results into a credible market premium. It introduces a dynamic parameter feedback mechanism based on premium ratios, allowing real-world business data from the terminal to directly drive fertilization improvement decisions at the front end. This closed-loop mechanism enables adaptive iteration and optimization of agricultural asset inputs, overcoming the limitations of traditional agricultural blind estimations and improving the overall economic return on investment.

[0054] This invention constructs a digital twin model through multi-source data fusion to accurately predict the spatiotemporal evolution of soil nutrients; it introduces a risk-weighted utility function to scientifically balance climate risks and input costs, outputting the optimal service expenditure ratio; and it utilizes digital identity cards to break down barriers between production and sales, feeding back the final premium of agricultural products to the decision-making system for dynamic optimization. This solution constructs a closed-loop management system of "bottom-level prediction - mid-level decision-making - top-level feedback," effectively avoiding blind investment and improving the allocation efficiency of agricultural assets and overall business returns.

[0055] Example 2 This embodiment uses the soil improvement process of a high-quality, landmark citrus smart planting orchard in southern China as an example to illustrate the specific application scenario of the method of the present invention. This orchard is located in a subtropical monsoon climate zone, and during the citrus fruit enlargement period, it often faces sudden, short-term heavy rainfall, which easily leads to soil leaching and severe nutrient loss.

[0056] The orchard management system first acquires real-time monitoring data, specifically utilizing capacitive sensors deployed within the orchard to collect soil moisture, time-domain reflectometers to collect conductivity, ion-selective electrodes to collect pH values, near-infrared spectroscopy to collect nitrogen, phosphorus, and potassium content, and automatic weather stations to collect rainfall. Due to differences in the hardware sampling frequencies of the weather station and various soil sensors, the system extracts timestamp features and uses a linear interpolation algorithm to unify them to an hourly time resolution. Then, the Z-score algorithm is used to remove outlier data points and perform smoothing. Subsequently, the range standardization formula is used to eliminate the differences in physical dimensions of characteristic parameters such as soil moisture and conductivity, generating standard feature vectors and completing spatial coordinate binding within a single-graph management architecture.

[0057] A drone cruises the orchard at a preset flight altitude, collecting multispectral images as vegetation physiological data. Pixel-level reflectance in the near-infrared and red bands is extracted and used in a normalized difference formula to calculate the vegetation index. The system concatenates the standardized real-time monitoring data with initial feature values ​​containing the vegetation index to construct a joint feature matrix. Spatial differential clustering analysis is performed using a fuzzy C-means clustering algorithm with a preset fuzziness index of 2, outputting the membership matrix of the corresponding spatial clusters. Using this membership matrix as statistical weights, the weighted average nutrient content and standard deviation of each spatial cluster are calculated. This yields the soil nutrient variation coefficient, reflecting differences in orchard nutrient distribution. This coefficient, along with the initial feature values, is used to determine the soil quality improvement index and mapped to corresponding spatial geographic coordinates through a geographic information system.

[0058] The orchard regularly acquires high-precision soil sampling data from laboratory tests. High-frequency real-time monitoring data is used as an auxiliary variable, while discrete laboratory nutrient content is used as the main variable. A cross-variance function reflecting the spatial correlation between the two is constructed. The system uses a co-kriging algorithm to calculate weight coefficients, performs spatial interpolation on the laboratory sampling data, aligns spatial dimensions within a single-map management architecture, generates a digital soil map with a preset grid resolution, and uses this map to create a digital twin model of the orchard.

[0059] In the mechanism simulation layer of the digital twin model, the system concatenates aligned historical agricultural activity trajectory data, real-time monitoring data, and laboratory sampling data, and inputs them into a long short-term memory network for nonlinear feature extraction to obtain a hidden state vector. Combining the rainfall intensity monitored by the current weather station, the orchard's preset soil infiltration rate threshold, and initial moisture content, the system calculates the leaching environment driving factor data. Furthermore, the spatial gradient of the nutrient prediction value is calculated using the central difference method, and combined with the pore water velocity vector derived from this driving factor, the system simulates the direction of the transport vector that conforms to physical laws, accurately extracting the soil nutrient gap and migration trend for the next two weeks, thus constructing a spatiotemporal evolution prediction value.

[0060] The orchard decision-making unit retrieves the predicted value and, combined with the short-term heavy rainfall probability factor derived from analyzing historical meteorological data over the past ten years, substitutes it into the risk-weighted utility function. For the current orchard's total soil improvement budget for this quarter, the system automatically evaluates the benefit weighting between increasing investment in orchard drainage facilities and increasing the application of quick-acting fertilizers. By differentiating the utility function, it determines the optimal service expenditure ratio that best mitigates leaching risk. Subsequently, the recommended fertilization intensity under the target citrus yield is calculated using the nutrient balance method. The optimal funding ratio is correlated with the fertilization intensity, outputting an improvement plan including a variable fertilization prescription map. The orchard's variable controllers and Beidou-guided fertilization equipment execute this prescription map, and the response time, flow accuracy, and geographic coordinates are structured and integrated to generate actual fertilization trajectory records.

[0061] After harvesting, high-quality citrus fruits are traced using a blockchain-based distributed traceability system, generating unique identification codes. The system encapsulates recorded soil improvement data and soil quality enhancement indicators into quality metadata, which is then hashed with the identification code using an asymmetric encryption algorithm to create a digital identity for each batch of citrus fruits. When sold in supermarkets, the system collects the transaction price of citrus fruits with digital identity codes in real time and compares it with the benchmark price of conventional citrus fruits within the same provincial administrative region that have not undergone digital traceability, calculating the premium data for agricultural products. This premium data is directly fed back to the orchard decision-making unit, incorporated into the dynamic feedback update equation, and used to correct the unit fertilizer-based yield increase coefficient within the risk-weighted utility function. This automatically triggers the optimization of the optimal service expenditure ratio for the next fruit tree growth cycle, completing an intelligent closed-loop management system from agricultural input to commercial feedback.

[0062] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart agriculture soil improvement and optimization method based on big data, characterized in that, include: Acquire real-time monitoring data to obtain soil quality improvement indicators, and bind the real-time monitoring data with geographic information spatial coordinates to form a single-map management architecture; Acquire laboratory sampling data, use co-kriging algorithm to fuse laboratory sampling data with real-time monitoring data in multiple sources and at multiple scales, generate digital soil map in a single map management architecture, and create digital twin model based on digital soil map; By using a digital twin model and combining it with the spatial coordinate attributes in a one-graph management architecture, the soil nutrient diffusion process after fertilization and the impact of tillage intensity on soil bulk density are simulated, and the predicted values ​​of soil nutrient spatiotemporal evolution, including soil nutrient gap and nutrient migration trend, are output. Establish a soil improvement benefit model, input the predicted values ​​of soil nutrient spatiotemporal evolution and climate probability factors into the risk-weighted utility function, simulate and output the soil improvement scheme under the optimal service expenditure ratio; Implement the soil improvement plan and generate soil improvement records in real time within a single-map management architecture; obtain the preset climate probability factor. The climate probability factor The probability of short-duration heavy rainfall leading to leaching during the crop growing season is calculated using an extreme value distribution model by analyzing the historical meteorological data of the target plot over the past 10 years. Assume the total budget for soil improvement of this plot is The service expenditure ratio is The investment amount used to increase drainage facilities is The investment amount used to increase the application of quick-acting fertilizers is Construct the risk-weighted utility function. The mathematical expression is: ； In the formula, The yield increase coefficient per unit fertilizer base, determined by the predicted values ​​of soil nutrient spatiotemporal evolution. The sensitivity coefficient for the risk of fertilizer leaching due to rainfall. For the constant parameters of the drainage facility's resilience to risks; function terms Characterizes the effective retention rate of fertilizer under current investment intervention in drainage facilities; The quarterly depreciation and amortization rate of drainage facilities is determined based on the pre-set facility life cycle. Establish digital identity cards for agricultural products produced from the soil, link soil improvement records and soil quality improvement indicators with the digital identity cards, feed agricultural product premium data back to the decision-making process, and dynamically adjust the soil improvement benefit model.

2. The method for optimizing and improving soil in smart agriculture based on big data according to claim 1, characterized in that, The real-time monitoring data includes: soil moisture collected using a capacitive sensor, electrical conductivity collected using a time-domain reflectometer, pH value collected using an ion-selective electrode, nitrogen, phosphorus, and potassium content collected using near-infrared spectroscopy, and rainfall collected using an automatic weather station; and the real-time monitoring data is cleaned and standardized using a multi-source heterogeneous data interface integration protocol.

3. The method for optimizing and improving soil in smart agriculture based on big data according to claim 1, characterized in that, The soil quality improvement indicators include: Acquire vegetation physiological data and calculate initial feature values ​​from the vegetation physiological data; the initial feature values ​​include chlorophyll content, vegetation index, and water deficit index; use fuzzy C-means clustering algorithm to perform spatial differential clustering analysis on the initial feature values ​​and the real-time monitoring data to calculate the soil nutrient variation coefficient reflecting the differences in nutrient distribution in the production area, and jointly determine the initial feature values ​​and the soil nutrient variation coefficient as the soil quality improvement index; use a geographic information system to map the real-time monitoring data and the soil quality improvement index to corresponding spatial geographic coordinates through attribute binding, and construct the one-map management architecture.

4. The method for optimizing and improving soil in smart agriculture based on big data according to claim 1, characterized in that, The laboratory sampling data includes: The process of obtaining a digital soil map through multi-source, multi-scale fusion of soil pH, nitrogen, phosphorus, potassium content, organic matter content, and heavy metal concentration indicators measured in the laboratory includes: acquiring the laboratory sampling data and the real-time monitoring data; using the co-kriging algorithm, with the real-time monitoring data as an auxiliary variable, performing spatial interpolation processing on the laboratory sampling data with discrete spatial coordinate attributes; aligning the interpolated laboratory attribute values ​​with the real-time monitoring data in a single-map management architecture to generate a digital soil map with a preset grid resolution.

5. The method for optimizing and improving soil in smart agriculture based on big data according to claim 1, characterized in that, The digital twin model includes a data mapping layer and a mechanism simulation layer: The data mapping layer is used to align and map the historical agricultural activity trajectory data, real-time monitoring data, and laboratory sampling data pre-stored in the single-map management architecture in a spatiotemporal dimension, and use these as input variables for the mechanism simulation layer. A nutrient kinetic prediction model is established in the mechanism simulation layer using a long short-term memory network. Current nutrient content data, fertilizer input data, leaching environment driving factor data determined by rainfall intensity, and crop absorption coefficient data are acquired and input into the nutrient kinetic prediction model for simulation calculation to obtain future nutrient content data after a preset time period. The future nutrient content data is compared with the corresponding standard nutrient threshold data, and the difference between the two is extracted as the soil nutrient gap. The mechanism simulation layer is used to simulate the nutrient migration vector direction in the soil profile to extract the nutrient migration trend, thus forming the predicted value of soil nutrient spatiotemporal evolution.

6. The method for optimizing and improving soil in smart agriculture based on big data according to claim 1, characterized in that, The soil improvement benefit model includes a decision-making unit and an execution feedback unit: The decision-making unit retrieves the predicted values ​​of soil nutrient spatiotemporal evolution and, in conjunction with a preset climate probability factor, inputs a risk-weighted utility function. The risk-weighted utility function then evaluates the benefit weights between increasing investment in drainage facilities and increasing the application of quick-acting fertilizers. Based on the evaluation results of these benefit weights, the optimal service expenditure ratio is determined, and the recommended fertilization intensity is calculated using the nutrient balance method. The optimal service expenditure ratio is then correlated with the recommended fertilization intensity, outputting a soil improvement scheme that includes a variable fertilization prescription map and a service-oriented management strategy. The variable fertilizer prescription map in the soil improvement scheme is executed using a variable controller and a Beidou navigation terminal, and the response time data, flow accuracy data and geographic coordinate location data are captured in real time during the execution process. The response time data, the flow rate accuracy data, and the geographic coordinate location data are structured and integrated to generate the soil improvement record, which records the actual fertilization trajectory, fertilization dosage, and operation time.

7. The method for optimizing and improving soil in smart agriculture based on big data according to claim 1, characterized in that, The specific process of establishing digital identity cards and dynamically adjusting soil improvement benefit models includes: A distributed traceability node is established using blockchain technology, and a unique identification code is generated for the agricultural products produced from the current plot. The soil improvement records and soil quality improvement indicators stored in the single-map management architecture are encapsulated as quality metadata. The quality metadata and the unique identification code are hashed and associated using an asymmetric encryption algorithm and stored in the distributed traceability node to construct the digital ID card. The market transaction price of agricultural products with the digital ID card at the sales terminal is collected in real time and compared with the preset benchmark price of conventional agricultural products. The difference between the two is calculated to obtain the agricultural product premium data. The agricultural product premium data is fed back to the decision unit in the soil improvement benefit model to correct the benefit weight parameters in the risk-weighted utility function, thereby completing the dynamic adjustment of the soil improvement benefit model.

Images