A method for optimizing the management of farmland soil organic carbon by improving the pH and phosphorus content

Abstract

The application discloses a management method for improving farmland soil organic carbon by optimizing pH and phosphorus content, and relates to the technical field of farmland precision management, which comprises the following steps: deploying a multi-parameter intelligent sensor network of multi-dimensional sensing nodes covering pH, phosphorus content and temperature and humidity in a target farmland area, and optimizing the layout of the sensing nodes by using a fruit fly algorithm; and continuously collecting multi-dimensional sensing data of soil pH, available phosphorus content, temperature and humidity by using the deployed multi-parameter intelligent sensor network. The application introduces an organic carbon loss risk model based on reinforcement learning, can autonomously learn and establish a complex nonlinear mapping relationship between soil chemical state and organic carbon loss risk from historical and real-time data, has online reasoning capability, can perform real-time loss risk level evaluation on real-time incoming multi-dimensional sensing data, and triggers early warning based on multi-source verification logic, so that early and accurate identification and early warning of the organic carbon loss risk are realized.

Description

Technical Field

[0001] This invention relates to the field of precision farmland management technology, specifically a management method for optimizing pH and phosphorus content to increase organic carbon in farmland soil. Background Technology

[0002] In agricultural production, soil organic carbon is an important indicator affecting soil fertility, structure, and water retention capacity. Optimizing soil pH and phosphorus content is a key strategy to increase soil organic carbon reserves. Soil pH directly affects the activity of microorganisms and the growth of plant roots, while a suitable pH range can promote the decomposition and synthesis of organic carbon. In addition, phosphorus, as an essential nutrient element for plant growth, can promote root development, thereby increasing the absorption and transformation of organic matter by plants. Increasing the phosphorus content in the soil helps plants synthesize more organic matter in photosynthesis, thereby increasing the input of soil organic carbon.

[0003] For example, a method and system for estimating changes in soil organic carbon storage in vegetable fields in the short term, as disclosed in Chinese Patent Publication No. CN120102836A, helps to clarify the carbon emissions and fixation status of farmland ecosystems, thereby better predicting the speed and extent of climate change and taking timely and reasonable vegetable planting and management measures.

[0004] In farmland management involving frequent fertilization and irrigation, soil pH and phosphorus content are in a dynamic state, prone to soil acidification or phosphorus fixation, directly affecting the stability and accumulation efficiency of soil organic carbon. Existing methods lack real-time monitoring and early warning mechanisms for the two key chemical indicators, pH and phosphorus content, making it difficult to identify the risk of organic carbon loss in a timely manner. Furthermore, after discovering an imbalance in the soil chemical environment, it is inconvenient to dynamically formulate and implement precise fertilization and irrigation strategies based on the synergistic relationship between pH and phosphorus content. This makes it difficult to effectively improve and stabilize the organic carbon storage of farmland soil through proactive optimization of the soil chemical environment. Therefore, this paper proposes a management method to optimize pH and phosphorus content to increase farmland soil organic carbon, in order to solve the above-mentioned problems. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a management method for optimizing pH and phosphorus content to increase organic carbon in farmland soil, comprising the following steps:

[0006] S1. Deploy a multi-parameter intelligent sensor network covering multiple sensor nodes for pH, phosphorus content, temperature and humidity in the target farmland area, and optimize the layout of the sensor nodes through the fruit fly algorithm.

[0007] S2. Utilize a deployed multi-parameter intelligent sensor network to continuously collect multi-dimensional sensing data on soil pH, available phosphorus content, temperature, and humidity. Upload the data to the cloud platform in real time via a wireless network to construct a high spatiotemporal resolution farmland soil chemical environment database, providing reliable and timely data support for refined analysis and dynamic early warning.

[0008] S3. Based on historical and real-time multi-dimensional sensor data of the target farmland area, a model for organic carbon loss risk is constructed using reinforcement learning technology. A dynamic mapping relationship between soil chemical state and organic carbon loss risk is established, and the risk of organic carbon loss is assessed and warned in real time. This can identify the risk points of soil organic carbon loss in advance, thus winning valuable time for precise intervention.

[0009] S4. A collaborative decision-making framework is designed by combining fuzzy logic control and genetic algorithm. Fuzzy logic integrates agronomic rules and real-time environmental uncertainties to output a preliminary soil organic carbon regulation strategy. Through iterative optimization using genetic algorithm, a precise quantitative management strategy covering the coordinated regulation of soil pH and phosphorus content is output. This can generate scientific and quantitative management plans and achieve precise management with multi-objective collaborative optimization.

[0010] S5. Based on the early warning results and collaborative decision-making framework, generate a fertilization-irrigation synergy plan that takes into account both pH adjustment and phosphorus activity enhancement, and execute precise operations through intelligent agricultural machinery and irrigation systems, driving the intelligent equipment to execute precisely, and efficiently transforming management strategies into actual field operations.

[0011] S6. Based on the multi-dimensional sensor data of the target farmland after regulation and the feedback on changes in soil organic carbon, update the parameters of the organic carbon loss risk model and collaborative decision-making framework to form a closed loop of monitoring-early warning-regulation-assessment, and achieve long-term stable improvement of soil carbon storage.

[0012] Preferably, S1 specifically includes:

[0013] Deploy a multi-parameter intelligent sensor network, including pH sensors, phosphorus content sensors, and temperature and humidity sensors, within the target farmland area to form a synchronous monitoring capability for key soil chemical and physical parameters, enabling comprehensive and real-time perception of key soil parameters and laying a data foundation for precision management.

[0014] Based on farmland topography, soil type and historical variation data, the fruit fly algorithm is used to spatially optimize the deployment location of each sensor node in the multi-parameter intelligent sensor network, so as to minimize the monitoring blind zone and improve the spatial representativeness of the sampling points. After optimization, the coverage and representativeness of the monitoring network are significantly improved and the data sampling bias is reduced.

[0015] Based on the optimized layout, each sensor node is fixed, and wireless transmission modules and power systems are configured to establish an IoT monitoring infrastructure capable of real-time and stable acquisition of multi-dimensional soil parameter data. This constructs a highly reliable and continuously operating IoT system, ensuring the stability and timeliness of data acquisition.

[0016] Preferably, S2 specifically includes:

[0017] Through a multi-parameter intelligent sensor network, soil pH, available phosphorus content, and soil temperature and humidity are automatically collected at a preset frequency, and the corresponding time and geographical coordinate information are recorded. The data are integrated to form a multi-dimensional sensor data sequence, realizing the continuous and automatic collection of key soil parameters, and providing a complete and time-series raw data foundation for subsequent analysis.

[0018] Low-power wide-area networks are used to upload the data from the collected multi-dimensional sensor data sequence to the cloud data platform in real time, ensuring that the monitoring data can be transmitted to the cloud in a timely and reliable manner, and supporting remote real-time monitoring and centralized management.

[0019] The received multidimensional sensor data is cleaned, calibrated, and fused on the cloud data platform to build a structured farmland soil chemical environment database with high spatiotemporal resolution, forming a high-quality, traceable, standardized database that provides reliable data support for accurate decision-making and model training.

[0020] Preferably, S3 specifically includes:

[0021] Multidimensional sensor data on pH, phosphorus content, temperature, and humidity, including historical and real-time data, are extracted from the farmland soil chemical environment database and used as the input feature set for model training. This ensures that the model training is based on comprehensive, continuous, and spatiotemporally representative real data, laying the foundation for accurate modeling.

[0022] A reinforcement learning algorithm using deep Q-networks is employed, with the loss risk level as the output, to train and establish an organic carbon loss risk model. A nonlinear dynamic mapping relationship is established between the input feature set and the organic carbon loss risk, enabling the model to learn autonomously and capture the complex nonlinear correlation between soil chemical state and carbon loss risk, thereby improving the intelligence and accuracy of risk identification.

[0023] By using the trained organic carbon loss risk model, online inference is performed on the real-time input multidimensional sensor data to dynamically assess the loss risk level under the current soil chemical state, thereby realizing the instantaneous and automatic assessment of soil organic carbon loss risk and providing real-time decision-making basis for timely early warning and intervention.

[0024] Preferably, S3 further includes:

[0025] When the loss risk level output by the organic carbon loss risk model exceeds the preset safety threshold, an early warning mechanism is automatically triggered to achieve real-time identification and immediate intervention of organic carbon loss risk, preventing further deterioration of soil health.

[0026] The triggered early warning mechanism generates tiered early warning information based on the level of churn risk, including the risk location, dominant imbalance factors, and potential severity. This ensures that the early warning information is targeted, enabling managers to quickly identify key issues and formulate effective measures. The tiered early warning information is then fed back to the management terminal in real time through a visual interface and message push, improving the accessibility and processing efficiency of the early warning information and shortening the decision-making delay from risk discovery to management response.

[0027] Preferably, S4 specifically includes:

[0028] A collaborative decision-making framework combining fuzzy logic control and genetic algorithm design is introduced. A fuzzy logic controller is constructed using fuzzy logic control technology. Its inputs are the real-time monitored soil pH value and available phosphorus content. Its knowledge base integrates empirical rules in agronomy regarding pH, phosphorus availability, and organic carbon stability. Specifically, pH value is divided into three fuzzy subsets: acidic, suitable, and alkaline. Its membership function adopts a combination of trapezoidal and triangular forms, with the intersection points set at pH 5.5, 6.0, 7.0, 7.5, and 8.0, respectively. Available phosphorus content is divided into three fuzzy subsets: deficient, suitable, and excessive, with the critical points set at 15 mg / kg, 20 mg / kg, 50 mg / kg, and 60 mg / kg, respectively. In this way, complex agronomic knowledge is transformed into operable continuous control rules through fuzzy logic, thereby improving the adaptability of the decision-making system.

[0029] The fuzzy logic controller performs fuzzy reasoning based on its inputs and outputs a preliminary soil organic carbon control strategy regarding the amount of organic fertilizer applied, irrigation water volume, and acid-base regulating substances. This makes the preliminary control strategy more closely aligned with actual production needs and improves the pertinence of management measures. Furthermore, the preliminary soil organic carbon control strategy is quantified into decision variables and input into a genetic algorithm for iterative optimization. The genetic algorithm achieves a global search for multi-objective optimization, thereby enhancing the scientific rigor and accuracy of the final control strategy.

[0030] Preferably, S4 further includes:

[0031] A genetic algorithm with multiple objectives—increasing soil organic carbon storage, stabilizing pH within a suitable range, and optimizing phosphorus availability—is used to encode decision variables and simulate selection, crossover, and mutation operations in the biological evolution process. This enables parallel and efficient searching of complex multi-objective decision spaces, taking into account both the globality and convergence speed of the system optimization process.

[0032] Under the constraints of agronomics and environment, the algorithm iteratively searches for the optimal combination of organic fertilizer type, application rate, irrigation strategy, and acid / alkali adjustment measures. This ensures that the generated optimization strategy meets the actual needs of agricultural production while complying with environmental protection and efficient resource utilization standards. When the genetic algorithm converges or reaches the maximum number of iterations, it outputs a set of precise quantitative management strategies that synergistically regulate soil pH and phosphorus content. This provides a specific and quantifiable technical solution for farmland management, directly guiding agricultural operations.

[0033] Preferably, S5 specifically includes:

[0034] The system analyzes precise quantitative management strategies and hierarchical early warning information to generate fertilization-irrigation synergy plans that take into account both pH adjustment and phosphorus activity enhancement. The fertilization-irrigation synergy plans are then broken down into executable agricultural operation instructions, including the amount and location of organic fertilizer application, irrigation duration and flow rate, and whether soil conditioner needs to be applied. This achieves the refinement and operational feasibility of farmland management strategies and enhances the pertinence of measures.

[0035] Agricultural operation instructions are converted into control instructions for smart agricultural machinery and precision irrigation systems. Operation paths and times are planned, and smart agricultural machinery equipped with variable fertilization devices and irrigation systems with controllable valves are driven by IoT instructions to execute fertilization-irrigation coordination schemes in target farmland areas. This ensures that management instructions are executed automatically and accurately, reducing human error and operation delays.

[0036] Preferably, S5 further includes:

[0037] During intelligent agricultural machinery operations, onboard sensors monitor fertilizer application amount, location, and soil disturbance in real time to ensure consistency between operations and strategy requirements, thereby guaranteeing the accuracy of fertilization operations, minimizing human error, and improving operational efficiency and resource utilization.

[0038] During irrigation, the opening of the irrigation valve is dynamically adjusted based on feedback from the flow meter and soil moisture sensor, so as to achieve precise and on-demand water supply, effectively avoid water waste, ensure uniform and reasonable water supply to the crop root zone, and improve water use efficiency.

[0039] After completing a round of adjustment operations, the system automatically records the details of the executed strategy and operation parameters, which serve as the basis for subsequent evaluation and learning. This provides real and reliable feedback data for strategy optimization, supporting the system's continuous self-iteration and improved decision-making accuracy.

[0040] Preferably, S6 specifically includes:

[0041] After the fertilization-irrigation synergistic program is implemented, the dynamic response data of soil pH, phosphorus content and temperature and humidity will continue to be monitored using a multi-parameter intelligent sensor network to obtain detailed dynamic change trajectories of soil chemical and physical indicators, providing continuous and high-precision data support for evaluating the aftereffects of the control measures.

[0042] Soil samples were taken from the target farmland area to measure organic carbon change data. The predictive accuracy of the organic carbon loss risk model was evaluated, and the model parameters were fine-tuned online using the new data. Through model evaluation and optimization, the accuracy of its prediction of organic carbon dynamics was improved, and the reliability of early warning and decision-making was enhanced.

[0043] Based on the implementation effect of the fertilization-irrigation synergy scheme, the rule weights of the fuzzy logic controller and the fitness function parameters of the genetic algorithm are updated to optimize the collaborative decision-making framework, forming an adaptive closed-loop optimization management system of monitoring-early warning-decision-execution-evaluation. This enables the autonomous calibration of system parameters and the continuous evolution of strategies, gradually improving the accuracy and long-term stability of farmland soil organic carbon management.

[0044] This invention provides a management method for optimizing pH and phosphorus content to increase organic carbon in farmland soil. It has the following beneficial effects:

[0045] (i) This management method for optimizing pH and phosphorus content to increase organic carbon in farmland soil introduces an organic carbon loss risk model based on reinforcement learning. It can autonomously learn and establish a complex nonlinear mapping relationship between soil chemical state and organic carbon loss risk from historical and real-time data. It also has online reasoning capabilities, can perform real-time loss risk level assessment on real-time multi-dimensional sensor data, and trigger early warning based on multi-source verification logic, so as to achieve early and accurate identification and early warning of organic carbon loss risk.

[0046] (II) This management method for optimizing pH and phosphorus content to increase organic carbon in farmland soil constructs a collaborative decision-making framework by integrating fuzzy logic control and multi-objective genetic algorithms. It effectively combines and quantifies agronomic experience rules with real-time environmental uncertainties, and can output a precise quantitative management strategy that takes into account multiple objectives such as pH adjustment, phosphorus activity enhancement and organic carbon storage increase. It not only overcomes the one-sidedness of traditional experience-based decision-making, but also performs global optimization search under complex agronomic and environmental constraints, generating a comprehensive management plan with better performance and stronger feasibility. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the workflow of a management method for optimizing pH and phosphorus content to increase organic carbon in farmland soil according to the present invention.

[0048] Figure 2This is a schematic diagram of the process for a management method of optimizing pH and phosphorus content to increase organic carbon in farmland soil according to the present invention. Detailed Implementation

[0049] 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.

[0050] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: a management method for optimizing pH and phosphorus content to increase organic carbon in farmland soil, comprising the following steps:

[0051] S1. Deploy a multi-parameter intelligent sensor network covering pH, phosphorus content, and temperature and humidity sensors in the target farmland area. Optimize the sensor node layout using the fruit fly algorithm. The optimized network significantly reduces monitoring blind spots and greatly improves the comprehensiveness and efficiency of soil condition data acquisition. The deployment of this multi-parameter intelligent sensor network, including pH, phosphorus, and temperature and humidity sensors, within the target farmland area forms a synchronous monitoring capability for key soil chemical and physical parameters. This enables comprehensive and real-time perception of key soil parameters, laying a data foundation for precision management. Based on farmland topography, soil type, and historical variation data, the fruit fly algorithm is used to spatially optimize the placement of each sensor node in the multi-parameter intelligent sensor network to minimize monitoring blind spots and improve the spatial representativeness of sampling points. The optimized network significantly improves the monitoring network's performance. To improve network coverage and representativeness, reduce data sampling bias, fix each sensor node based on the optimized layout, and configure wireless transmission modules and power systems, an IoT monitoring infrastructure capable of real-time and stable acquisition of multi-dimensional soil parameter data will be established. This will build a highly reliable and continuously operating IoT system, ensuring the stability and timeliness of data acquisition. Specifically, the work involves deploying multi-parameter intelligent sensors in the target farmland area to construct the IoT monitoring infrastructure. The core sensor nodes include a soil pH sensor based on the ion-selective electrode principle (measurement range: pH 3.0-9.0, accuracy ±0.1%), an electrochemically based fast-acting phosphorus content sensor (detection range: 0-100 mg / kg, accuracy ±5%), and an integrated temperature and humidity sensor (temperature measurement range: -20℃~60℃, accuracy ±0.1%).5℃; humidity measurement range 0-100% VWC, accuracy ±2%). Before installation, each sensor node is pre-laid out according to the actual field size (each hectare as the basic planning unit) to ensure that its physical interface is in full contact with the soil. Among them, the pH and phosphorus sensor probes need to be inserted vertically into the cultivated layer (depth set to 0-20cm) and avoid being near fertilizer furrows or field ridges to reduce local interference. Each sensor node is connected to the data acquisition device through a standardized waterproof interface, initially forming a distributed sensor network prototype. On the basis of the initial deployment, intelligent algorithms are used to optimize the spatial structure of the sensor network to improve the representativeness and efficiency of monitoring. In specific operation, based on the detailed geographical information of the farmland (including 1:500 topographic map, 0-20cm soil layer soil type distribution map) and historical soil nutrient variation data (pH and phosphorus spatial distribution maps obtained from previous grid sampling), an optimization function is constructed with the goal of minimizing the monitoring blind area and maximizing the spatial coverage. The fruit fly optimization algorithm is used to solve it: the location of each sensor node is used as the decision variable, and the population size is set to 50-1 The algorithm iterates 200-500 times, simulating the olfactory and visual search behavior of fruit fly colonies to iteratively update node coordinates. In each iteration, the fitness value is analyzed by combining the Euclidean distance between nodes, soil type boundary coverage, and historical data variation coefficients, ultimately outputting a set of optimal spatial coordinates. Based on the optimal spatial coordinates output by the fruit fly algorithm, each sensor node is fixed in the field and the system is integrated. Dedicated mounting brackets are used to fix the sensor nodes, ensuring the sensor probes remain stable at a predetermined depth, and anti-settlement measures are implemented. Each sensor node integrates a low-power wireless transmission module and is equipped with a solar power system (solar panel power ≥10W, battery capacity ≥20Ah) to ensure long-term continuous operation in farmland environments without mains power. All sensor nodes communicate with the field gateway via a self-organizing network protocol. The gateway uploads encrypted real-time monitoring data to the cloud data platform via a 4G / 5G network. This constructs an IoT monitoring infrastructure with spatial representativeness, capable of stably collecting multi-dimensional parameters such as soil pH, available phosphorus content, temperature, and humidity around the clock.

[0052] S2. Utilize a deployed multi-parameter intelligent sensor network to continuously collect multi-dimensional sensor data on soil pH, available phosphorus content, temperature, and humidity. Upload this data in real-time to a cloud platform via a wireless network to construct a high spatiotemporal resolution farmland soil chemical environment database. This provides reliable and timely data support for refined analysis and dynamic early warning. Through the multi-parameter intelligent sensor network, soil pH, available phosphorus content, and soil temperature and humidity are automatically collected at preset frequencies, and the corresponding time and geographic coordinate information is recorded. This data is integrated to form a multi-dimensional sensor data sequence, enabling continuous and automatic collection of key soil parameters. This provides a complete and time-series raw data foundation for subsequent analysis. Low-power wide-area network is used to upload the collected multi-dimensional sensor data sequence to the cloud data platform in real time, ensuring that monitoring data can be transmitted to the cloud in a timely and reliable manner. This supports remote real-time monitoring and centralized management. On the cloud data platform, the received multi-dimensional sensor data is cleaned, calibrated, and fused to construct a structured farmland soil chemical environment database with high spatiotemporal resolution. This forms a high-quality, traceable, standardized database, providing reliable data support for accurate decision-making and model training.

[0053] The specific work content is as follows: In actual operation, the multi-parameter intelligent sensor network automatically starts the measurement cycle according to the preset acquisition frequency (once per hour). Specifically, the pH sensor with ion-selective electrode principle completes the measurement at a depth of 0-20cm in accordance with the relevant technical concepts of "NY / T1121.2-2006 Soil Testing Part 2: Determination of Soil pH". The fast-acting phosphorus sensor based on electrochemical method operates within the range of 0-100mg / kg, with a measurement accuracy of ±5%, referring to the basic principle of soil available phosphorus determination. The integrated temperature and humidity sensor simultaneously records soil temperature and volumetric water content. All data packets are automatically bound to latitude and longitude coordinates (positioning accuracy ≤ 1 meter) and UTC timestamps provided by high-precision GNSS modules, forming raw data packets with spatiotemporal tags. Data collectors within the multi-parameter intelligent sensor network poll and aggregate data from each node within the jurisdiction, performing preliminary format standardization and verification to ensure data integrity and traceability. The pre-processed data packets are then remotely transmitted via low-power wide-area network communication modules integrated into each sensor node. In actual deployment, the NB-IoT or LoRaWAN standard protocol is selected based on the signal coverage of the target farmland area, and data transmission follows a set communication interval. (Aggregated data is reported every 15 minutes), and an application layer protocol with forward error correction and retransmission mechanisms is used to ensure transmission reliability in complex field environments. Data packets are stably uploaded to the cloud data platform via a secure TLS encrypted channel and a dedicated IoT platform access point. The cloud data platform's access service parses the received data stream in real time, verifies its protocol compliance and data integrity, and stores valid data in a temporary buffer queue. The end-to-end latency of the entire transmission link is designed to be within a few minutes to meet the needs of near real-time monitoring. The cloud data platform starts an automated data processing pipeline to clean the data based on the sensor's physical quantities. The system automatically filters and marks abnormal or invalid data based on process, signal rationality, and spatial consistency rules. Then, calibration is performed. The cloud data platform calls the pre-stored calibration curve parameters of each sensor to convert the raw voltage or current readings and calculate accurate pH values, phosphorus content, and temperature and humidity physical quantities. In the data fusion stage, multi-source parameters under the same spatiotemporal label are aligned and correlated, and spatial interpolation algorithms are used to generate regular gridded data products. Finally, all the cleaned, calibrated, and fused data are persistently stored in the time-series database, constructing a farmland soil chemical environment database with a unified time base, accurate geographic reference, and high spatiotemporal resolution.

[0054] S3. Based on historical and real-time multi-dimensional sensor data of the target farmland area, a reinforcement learning technique is used to construct an organic carbon loss risk model. This establishes a dynamic mapping relationship between soil chemical state and organic carbon loss risk, enabling real-time assessment and early warning of organic carbon loss risk. It can identify risk points for soil organic carbon loss in advance, gaining valuable time for precise intervention. Multi-dimensional sensor data on pH, phosphorus content, and temperature and humidity, including historical and real-time data, are extracted from the farmland soil chemical environment database and used as the input feature set for model training. This ensures that model training is based on comprehensive, continuous, and spatiotemporally representative real data, laying the foundation for accurate modeling. Deep learning is employed. The reinforcement learning algorithm of Q-degree network, with the loss risk level as the output, is used to train and establish an organic carbon loss risk model. It establishes a nonlinear dynamic mapping relationship between the input feature set and the organic carbon loss risk, enabling the model to learn autonomously and capture the complex nonlinear correlation between soil chemical state and carbon loss risk, thereby improving the intelligence and accuracy of risk identification. Using the trained organic carbon loss risk model, online inference is performed on the real-time input multidimensional sensor data to dynamically assess the loss risk level under the current soil chemical state, thereby realizing the instantaneous and automatic assessment of soil organic carbon loss risk and providing real-time decision-making basis for timely early warning and intervention.

[0055] The specific work involves: extracting the input feature set required for model training and evaluation from the farmland soil chemical environment database. This input feature set includes historically accumulated and real-time updated multi-parameter soil data, specifically covering pH value, available phosphorus content, soil temperature, and soil volumetric water content. During the extraction process, the spatiotemporal alignment principle is followed to ensure that each data point has accurate geographic coordinates and timestamps. In the data preprocessing stage, the raw data is cleaned according to the physical range of each sensor and the established signal rationality rules, removing outliers that significantly exceed the normal fluctuation range. Subsequently, the sensor individual calibration parameters stored in the cloud data platform are called to uniformly convert the raw electrical signal readings into standardized values ​​with physical meaning. For data gaps with a small number of missing data or due to communication interruptions, Kriging interpolation based on spatiotemporal proximity is used for reasonable filling. Finally, a multi-dimensional feature matrix with continuous time series, complete spatial coverage, and standardized values ​​is constructed. A deep Q-network algorithm under the reinforcement learning framework is used to construct and train an organic carbon loss risk model. The input of the model is a pre-defined data set. The processed multidimensional feature matrix outputs discrete loss risk levels, divided into low, medium, and high risk levels. The core of the deep Q-network algorithm is a deep neural network, whose structure includes an input layer, several fully connected hidden layers, and an output layer. The number of nodes in the input layer corresponds to the four feature dimensions of pH, phosphorus content, temperature, and humidity; the number of nodes in the output layer corresponds to the number of loss risk levels. During the training phase, the historical feature dataset is divided into a training set and a validation set. The organic carbon loss risk model learns through interaction with the environment (i.e., historical data state). Its reward function is designed based on the degree of agreement between the loss risk level predicted by the model and the true risk level (as a label) inverted from the measured soil organic carbon values ​​of the same period. The training process uses experience replay and target network technology to stabilize the learning process. The network weight parameters are iteratively updated by minimizing the temporal difference error until the prediction accuracy of the organic carbon loss risk model on the validation set reaches a preset threshold (≥85%) and the loss function converges, establishing a nonlinear dynamic mapping relationship from soil chemical state to organic carbon loss risk.The trained and validated organic carbon loss risk model is deployed in a cloud-based inference service to form an online risk warning module. This module connects to a real-time data stream pipeline via an API interface, continuously receiving the latest, pre-processed multi-dimensional sensor data uploaded by the IoT monitoring infrastructure. When a new data packet arrives, the inference service automatically calls the loaded organic carbon loss risk model and performs real-time calculations on the input feature vector (i.e., the current soil pH, phosphorus content, temperature, humidity values, and their changing trends within a short time window) using a forward propagation method. The model output is the loss risk level assessment result of the organic carbon loss risk under the current soil chemical state at the monitoring point. The assessment result, along with its corresponding spatiotemporal label, is pushed to the visualization interface of the management terminal in real time for dynamic display and early warning. On the other hand, the result, along with the original sensor data, is recorded and fed back to the historical database.

[0056] Furthermore, S3 also includes: when the loss risk level output by the organic carbon loss risk model exceeds the preset safety threshold, an early warning mechanism is automatically triggered to realize real-time identification and immediate intervention of organic carbon loss risk, preventing further deterioration of soil health. The triggered early warning mechanism generates graded early warning information including risk location, dominant imbalance factor and potential severity according to the level of loss risk, ensuring that the early warning information is targeted, making it easy for managers to quickly identify key issues and formulate effective measures. The graded early warning information is fed back to the management terminal in real time through a visual interface and message push, improving the accessibility and processing efficiency of early warning information, and shortening the decision delay from risk discovery to management response.

[0057] The specific work involves: setting a preset safety threshold for organic carbon loss risk as the lower limit of the intermediate loss risk level. When the loss risk level output by the model inference exceeds this safety threshold, a multi-level early warning response mechanism is automatically triggered. The early warning triggering conditions are subject to mandatory verification logic and must simultaneously meet the following conditions: the model output exceeds the safety threshold for three consecutive acquisition cycles; more than 70% of the sensor nodes within the same monitoring unit trigger the early warning synchronously; and the system self-test module confirms that the data acquisition and transmission links are normal. After triggering, the trigger timestamp, geographic coordinate cluster, and unique device code of the triggering node are recorded, generating an encrypted early warning event log, which is stored in the early warning event table of the high-availability database. The log includes event ID, trigger time, risk value, longitude, and latitude fields. Based on the severity of the triggering event (divided into three warning levels according to the risk of loss), it automatically executes a warning information synthesis process. First, spatial clustering analysis is performed on the warning area to identify the core risk area and its impact range. The DBCSAN algorithm is used, with a neighborhood radius ε = 50 meters and a minimum number of points MinPts = 3. Feature importance analysis is used to diagnose the dominant imbalance factor. The deviation of the current monitoring value from the safe range (pH safe range 6.0-7.5, available phosphorus safe range 20-50 mg / kg) is calculated, and the parameter with the highest deviation is identified as the dominant imbalance factor. Quantitative values ​​are a crucial component of early warning information. The early warning information structure includes: early warning level (Level 1 / Level 2 / Level 3), risk area center coordinates (WGS-84 coordinate system), area influence radius (meter-level accuracy), and dominant imbalance factor (specific parameter name and deviation value). The final standardized early warning information package is encapsulated in JSON format and generates a timestamp conforming to the ISO 8601 standard. Early warning information is distributed and fed back through a multi-channel collaborative mechanism. A B / S architecture visual interface is used, presenting the early warning area on a map in the form of a dynamic heatmap. The heatmap color gradient corresponds to the early warning level (green for Level 1, yellow for Level 2, and red for Level 3). A pop-up warning box displays detailed information. The message push system integrates SMS, email, and mobile application push notifications, and formulates differentiated push strategies based on the warning level: Level 2 and above warnings trigger all three push notification methods simultaneously, while Level 1 warnings only trigger mobile application push notifications. The push content is templated and includes summary information and a link to a detailed report. The management terminal is set up with a warning confirmation and handling feedback mechanism. Operators are required to confirm Level 3 warnings within 15 minutes and Level 2 warnings within 1 hour. The system automatically records response time and handling measures, forming a complete closed-loop management file for warning handling. This file is synchronously updated to the warning event log for evaluating warning response efficiency and optimizing warning threshold parameters.

[0058] S4. A collaborative decision-making framework is designed by combining fuzzy logic control and genetic algorithm. Fuzzy logic integrates agronomic rules and real-time environmental uncertainties to output a preliminary soil organic carbon regulation strategy. Through iterative optimization using genetic algorithm, a precise quantitative management strategy covering the coordinated regulation of soil pH and phosphorus content is output. This can generate scientific and quantitative management plans and achieve precise management through multi-objective collaborative optimization.

[0059] S5. Based on the early warning results and collaborative decision-making framework, generate a fertilization-irrigation synergy plan that takes into account both pH adjustment and phosphorus activity enhancement, and execute precise operations through intelligent agricultural machinery and irrigation systems, driving the intelligent equipment to execute precisely, and efficiently transforming management strategies into actual field operations.

[0060] S6. Based on the multi-dimensional sensor data of the target farmland after regulation and the feedback of soil organic carbon changes, update the parameters of the organic carbon loss risk model and collaborative decision-making framework to form a closed loop of monitoring-early warning-regulation-assessment, achieve long-term stable improvement of soil carbon storage, and continuously optimize the system through closed-loop feedback, so that soil management decisions have self-learning and long-term adaptability.

[0061] Example 2, as Figure 1 , Figure 2 As shown, based on Example 1, this invention provides a technical solution: S4 specifically includes: introducing a collaborative decision-making framework combining fuzzy logic control and genetic algorithm design; constructing a fuzzy logic controller using fuzzy logic control technology; its inputs being the real-time monitored soil pH value and available phosphorus content; its knowledge base integrating empirical rules from agronomy regarding pH, phosphorus availability, and organic carbon stability; wherein, pH value is divided into three fuzzy subsets: acidic, suitable, and alkaline; its membership function adopts a combination of trapezoidal and triangular forms; the intersection points are set to pH 5.5, 6.0, 7.0, 7.5, and 8.0, respectively; available phosphorus content is divided into three fuzzy subsets: deficient, suitable, and excessive; its critical point... The concentrations were set at 15 mg / kg, 20 mg / kg, 50 mg / kg, and 60 mg / kg, respectively. This allowed fuzzy logic to transform complex agronomical knowledge into operable continuous control rules, improving the adaptability of the decision-making system. The fuzzy logic controller performed fuzzy inference based on its inputs and output preliminary soil organic carbon control strategies regarding organic fertilizer application, irrigation water volume, and pH-regulating substances. This made the preliminary control strategies more closely aligned with actual production needs and improved the pertinence of management measures. Furthermore, the preliminary soil organic carbon control strategies were quantified into decision variables and input into a genetic algorithm for iterative optimization. The genetic algorithm enabled global search for multi-objective optimization, improving the scientific rigor and accuracy of the final control strategy.

[0062] The specific work involves constructing a fuzzy logic controller for outputting preliminary control strategies. The inputs to this controller are two real-time monitored soil chemical indicators: soil pH and available phosphorus content. To achieve intelligent reasoning consistent with agronomic principles, the fuzzy logic controller's knowledge base integrates empirical rules regarding pH, phosphorus availability, and their relationship with soil organic carbon stability, based on textbooks such as *Soil Science*, *Plant Nutrition*, and relevant farmland management technical specifications. Specifically, pH is divided into three fuzzy subsets: acidic, suitable, and alkaline. The membership function adopts a combination of trapezoidal and triangular forms, with the intersection points set at pH 5.5, 6.0, 7.0, 7.5, and 8.0, respectively. Simultaneously, the available phosphorus content is divided into three fuzzy subsets: deficient, adequate, and excessive, with their critical points set at 15 mg / kg, 20 mg / kg, 50 mg / kg, and 60 mg / kg, respectively, referencing soil nutrient grading standards. The fuzzy rules in the knowledge base are encoded in IF-THEN form. The fuzzy logic controller, based on the input precise pH and phosphorus content values, performs fuzzification, knowledge base-based rule reasoning, and defuzzification. Using the center-of-gravity method, this method employs three steps to output preliminary, qualitative to semi-quantitative suggestions for soil organic carbon regulation strategies regarding organic fertilizer application rate, irrigation water volume, and pH-regulating substances. The preliminary regulation strategies output by the fuzzy logic controller are quantified into precise decision variables to serve as input for subsequent optimization algorithms. Fuzzy outputs such as increasing organic fertilizer application, adjusting irrigation, and applying lime are transformed into specific, actionable decision variables. Specifically, three core decision variables are defined: X1 (organic fertilizer application rate, domain 0 to 10000), X2 (increase in irrigation water volume, domain -20 to...). +50 (negative values ​​indicate reduced irrigation), X3 (lime application rate, defined as 0 to 2000; if acid adjustment is required, the corresponding variable is sulfur application rate). The output of the fuzzy logic is mapped to the initial value or initial value range of the decision variables according to the preset transformation function. The quantified decision variables (X1, X2, X3) together constitute a solution vector to be optimized, and are encoded into genes of the population individuals of the genetic algorithm using binary encoding. Each gene position represents a specific value of a decision variable, transforming agronomic experience into mathematical model parameters that can be directly processed and optimized by the genetic algorithm.

[0063] Furthermore, S4 also includes: employing a genetic algorithm with multiple objective functions, namely increasing soil organic carbon storage, stabilizing pH within a suitable range, and optimizing phosphorus availability, to encode decision variables, simulate selection, crossover, and mutation operations in the process of biological evolution, and achieve parallel and efficient search of complex multi-objective decision space, taking into account both the globality and convergence speed of the system optimization process. Under the condition of satisfying agronomic and environmental constraints, iteratively searches for the optimal combination of organic fertilizer type, application rate, irrigation strategy, and acid / alkali adjustment measures, ensuring that the generated optimization strategy meets the actual needs of agricultural production while complying with the norms of environmental protection and efficient resource utilization. When the genetic algorithm converges or reaches the maximum number of iterations, it outputs a set of precise quantitative management strategies for synergistically regulating soil pH and phosphorus content, providing a specific and quantifiable technical solution for farmland management, and directly guiding agricultural operations.

[0064] The specific work involves using a multi-objective genetic algorithm for optimization, simultaneously increasing soil organic carbon storage, stabilizing soil pH within the suitable range of 6.0-7.5, and optimizing available phosphorus content to the ideal range of 20-50 mg / kg. The genetic algorithm encodes three decision variables—organic fertilizer application rate, irrigation water regulation rate, and lime / sulfur application rate—using binary encoding. Each variable uses a 16-bit encoding length, collectively forming a 48-bit individual gene string. The population size is set to 100 individuals. The initial population is uniformly generated within the decision space using the Latin hypercube sampling method to ensure the quality and diversity of the initial solutions. The genetic algorithm is set to a maximum of 500 iterations. The convergence condition is that the Pareto front improvement is less than 1% for 50 consecutive generations or the hypervolume index change rate is less than 0.5%. The fitness function is composed of three normalized objective components weighted together: soil organic carbon increment weight is 0.5, pH deviation weight is 0.3, and phosphorus content optimization weight is 0.2. All components were normalized to the [0,1] interval using Min-Max. During each generation of evolution, the fitness value of all individuals in the population was calculated. Individuals were stratified using a ranking-based non-dominated sorting method, and elites were retained by combining crowding calculation. A binary tournament selection mechanism was used for selection operations. Four individuals were randomly selected from the population each time, and the two individuals with the highest non-dominated sorting level and the highest crowding were retained as parents. A uniform crossover operation was used, with a crossover probability set to 0.85. The exchange of each gene locus was carried out independently. Mutation operations were carried out using position inversion mutation, with a mutation probability of 0.02 for each gene locus to ensure population diversity. The genetic algorithm followed agronomic and environmental constraints, namely, the application of organic fertilizer should not exceed the local standard limit of 10,000 kg / ha; the irrigation regulation should be kept within the range of -20 mm (reduction) to +50 mm (increase); the upper limit of lime application is 2,000 kg / ha, and it is limited when pH > 7.At 5 o'clock, the variable is automatically set to zero; at the same time, the nitrogen, phosphorus and potassium nutrient balance constraints are met to ensure that the proportion of each nutrient element meets the crop's needs. The penalty function method is used for constraint processing, and the fitness of individuals that violate the constraints will decay exponentially according to the degree of violation; when the genetic algorithm meets the convergence condition or reaches the maximum number of iterations, the three optimal solutions with the largest overvolume index are selected from the non-dominated solution set of the final generation as the recommended scheme output. Each scheme includes specific quantitative management strategies: the organic fertilizer application rate is accurate to an integer multiple of 100 kg / ha, and the recommended type (well-rotted farmyard manure or commercial organic fertilizer) is marked; the irrigation scheme clearly gives the adjustment based on the baseline irrigation amount. The savings are indicated by positive and negative values ​​representing increases or decreases in millimeters, respectively. Soil improvement measures specify the exact dosage of lime or sulfur required. The output strategy includes a predicted effect assessment, including key indicators such as the expected increase in soil organic carbon, the expected value after pH adjustment, and the expected change in available phosphorus content. A visualization interface displays the distribution of the Pareto front in the three-dimensional target space and provides an analysis of the trade-offs between each scheme and the three optimization objectives. This assists decision-makers in selecting the most suitable implementation plan based on actual management needs. All outputs are automatically generated into standardized reports, including complete parameter settings, optimization processes, and final recommendations.

[0065] S5 specifically includes: analyzing precise quantitative management strategies and hierarchical early warning information, generating fertilization-irrigation synergy plans, taking into account pH adjustment and phosphorus activity enhancement, and decomposing the fertilization-irrigation synergy plans into executable agricultural operation instructions, including the application amount and location of organic fertilizer, irrigation duration and flow rate, and whether soil conditioner needs to be applied, to achieve the refinement and operability of farmland management strategies, improve the pertinence of measures, convert agricultural operation instructions into control instructions for intelligent agricultural machinery and precision irrigation systems, and plan operation paths and times, and drive intelligent agricultural machinery equipped with variable fertilization devices and controllable valve irrigation systems through IoT instructions to execute fertilization-irrigation synergy plans in target farmland areas, ensuring that management instructions are executed automatically and accurately, reducing human error and operation delays;

[0066] The specific work involves analyzing the output of precise quantitative management strategies and real-time early warning information, and automatically generating a fertilization-irrigation synergy plan that balances soil pH adjustment and phosphorus activity enhancement. The core of this plan is to decompose the macro-strategy into a series of specific, quantifiable, executable agricultural operation instructions with spatiotemporal attributes. These instructions are generated based on field boundaries, soil type zoning, and historical operation records provided by the farmland geographic information system. Specifically, for fertilization instructions, the content includes the precise application rate and type of organic fertilizer (such as commercial organic fertilizer with a specific carbon-nitrogen ratio or fully decomposed farmyard manure). Based on the risk core areas identified by the early warning information, differentiated application rates and application point coordinates (based on 10m × 10m) are assigned to different sub-zones. (Grid-based positioning) For irrigation instructions, based on real-time soil volumetric water content data and target thresholds, the net irrigation water required for each field is calculated and further converted into irrigation duration and flow rate under fixed pipeline pressure. Regarding the application of soil conditioners, their necessity, specific types, application rates, and recommended separate application time windows to avoid fertilizer antagonism are clarified. All agricultural operation instructions are accompanied by execution priority, effective time windows, and associated early warning event IDs, forming a structured digital agricultural work order. The generated agricultural operation instructions are compiled into a standardized control instruction set that can be directly recognized and executed by intelligent agricultural machinery and precision irrigation systems through the instruction conversion module of the central control system. For intelligent agricultural machinery (equipped with Beidou / GNSS high-precision positioning) For tractors with variable displacement fertilization devices, control commands include: an optimized operating path generated based on field boundaries and obstacle information (using Boustrophedon mode coverage, with a minimum turning radius of 4 meters), driving speed, and target fertilization amount at each preset grid point or path point (control signals are sent to the fertilizer dispensing motor in PWM pulse width modulation form). The system automatically plans the operation sequence and timing, avoiding irrigation periods and reserving time for equipment entry and exit. For precision irrigation systems, control commands are converted into on / off control signals for designated field valves, target valve opening duration, and a constant flow rate setpoint maintained through a closed loop of pressure sensors and flow meters (for drip irrigation branches, the flow rate is set to 2 cubic meters per hour, with an error range of ±5%). All control commands... Control commands, path coordinates, and equipment status parameters are encapsulated into data packets conforming to ISO11783 and sent to field operation terminals and irrigation controllers via low-latency 4G / 5G networks. After the commands are sent, the IoT platform drives smart agricultural machinery equipped with variable fertilization devices and irrigation systems equipped with controllable valves to execute the fertilization-irrigation coordination scheme according to plan in the target farmland area. The smart agricultural machinery operates autonomously according to the received path and variable control commands. Its on-board controller compares the GNSS positioning coordinates with the target path in real time and dynamically adjusts the steering and fertilizer discharge. The operation data is transmitted back in real time. After receiving the soil moisture threshold trigger signal, the irrigation controller automatically opens the designated valves in sequence and monitors the pipeline pressure and flow to ensure accurate water supply according to the design.

[0067] S5 also includes: during intelligent agricultural machinery operation, onboard sensors monitor fertilizer application amount, location, and soil disturbance in real time to ensure consistency between operation and strategy requirements, thereby guaranteeing the accuracy of fertilization operations, minimizing human error, and improving operational efficiency and resource utilization. During irrigation, feedback from flow meters and soil moisture sensors dynamically adjusts the opening of irrigation valves to achieve precise on-demand water supply, effectively avoiding water waste, ensuring uniform and reasonable water supply to the crop root zone, and improving water use efficiency. After completing a round of control operations, it automatically records the details of the executed strategy and operation parameters as a basis for subsequent evaluation and learning, providing real and reliable feedback data for strategy optimization, and supporting the system's continuous self-iteration and improved decision-making accuracy.

[0068] The specific work content is as follows: During intelligent agricultural machinery operation, real-time monitoring and precise control of the entire operation process are achieved through multiple types of sensors integrated on the agricultural machinery. The onboard GNSS positioning module continuously acquires the real-time position coordinates of the agricultural machinery, compares them with the preset operation path points at the millisecond level, and sends correction commands to the steering control system via the CAN bus to ensure that the lateral deviation between the driving trajectory and the planned path is always controlled within ±5 cm. An impulse flow sensor installed at the fertilizer discharge outlet monitors the actual fertilizer discharge in real time. The 4-20mA analog signal generated by this sensor is converted by an AD converter and compared by the onboard controller with the target fertilizer application amount corresponding to the current coordinates. A PID algorithm dynamically adjusts the duty cycle of the PWM signal controlling the fertilizer discharge motor to achieve closed-loop control of the fertilizer application amount, ensuring that the fertilizer application amount error per unit area does not exceed ±5% of the set value. Simultaneously, the collaborative work of a triaxial accelerometer and a laser rangefinder installed on the agricultural machinery suspension mechanism monitors the tillage depth and soil disturbance amplitude. The tillage depth control accuracy is ±1 cm. All sensor data is encrypted and transmitted to the cloud data platform via the onboard 4G communication module at 100ms intervals. The platform analyzes the data in real time. The system performs compliance checks. When the fertilizer application deviation exceeds 10% or the path deviation exceeds 15 cm at three consecutive sampling points, it automatically sends a pause command to the agricultural machinery terminal via the MQTT protocol and triggers an audible and visual alarm. During the irrigation operation, a closed-loop control system based on multi-source feedback is constructed to achieve precise on-demand water supply. Using a preset target threshold for soil volumetric moisture content as the control benchmark, soil moisture sensors based on the frequency domain reflectance principle (buried at depths of 10cm, 20cm, and 30cm) deployed in the field collect data once per minute. When a certain monitoring point 1... When the humidity value at a depth of 0cm remains below the target threshold for 15 minutes, the irrigation controller automatically triggers the start-up logic of the irrigation group. During irrigation, the electromagnetic flowmeter installed at the beginning of the branch pipe monitors the actual flow rate in real time and compares it with the preset flow rate value. The controller uses a fuzzy PID algorithm to dynamically calculate and output a 4-20mA control signal to the electric regulating valve based on the flow deviation and the rate of change of the deviation, thereby achieving continuous adjustment of the valve opening and controlling the flow fluctuation within ±5% of the set value. At the same time, the pressure transmitter monitors the pipeline pressure to ensure that it remains stable at 0.2-0.The design working pressure range is 3MPa. The entire control process operates in 30-second control cycles until the soil moisture at all relevant monitoring points reaches the target threshold upper limit. At this point, the valves automatically close and the irrigation cycle ends. Data such as total irrigation volume, duration, flow rate process curve, and soil moisture response curve are all fully recorded. After a complete fertilization-irrigation coordinated control operation, the data archiving and closed-loop management process is automatically initiated. All data related to this operation is extracted, correlated, and persistently stored. The data specifically includes: 1) the warning event ID that triggered the operation and its corresponding original strategy command; 2) the entire process time-series data of the intelligent agricultural machinery operation, including the actual GNSS trajectory point sequence, the target and actual fertilizer application amount at each point, machine speed, hydraulic status, etc., stored in CSV format; 3) the execution log of the irrigation system. This includes the opening and closing times of each valve, instantaneous flow rate, cumulative water volume, pipeline pressure, and response data from key soil moisture sensors; 4) Before and after the operation, soil chemical state snapshots are obtained through a multi-parameter intelligent sensor network, with a focus on changes in pH value and available phosphorus content. Various structured data are injected into a dedicated operation case database through an ETL process. Each record serves as an independent management case, uniquely identified and indexed by the early warning event ID, field code, and operation time window. This database serves as the core data source for subsequent system performance evaluation, agronomic model optimization, and machine learning model retraining. Based on this, operation effect analysis reports are generated regularly to compare the expected goals of the strategy with the actual achievements, quantitatively evaluate the effectiveness of management measures, and provide data-driven decision-making basis for the next round of strategy optimization iteration.

[0069] S6 specifically includes: after the implementation of the fertilization-irrigation synergy program, continuing to use a multi-parameter intelligent sensor network to monitor the dynamic response data of soil pH, phosphorus content, and temperature and humidity, thereby obtaining detailed dynamic change trajectories of soil chemical and physical indicators, providing continuous and high-precision data support for evaluating the aftereffects of control measures; sampling soil in the target farmland area to measure organic carbon change data, evaluating the prediction accuracy of the organic carbon loss risk model, and using new data to fine-tune the model parameters online, improving the prediction accuracy of organic carbon dynamics through model evaluation and optimization, enhancing the reliability of early warning and decision-making; based on the implementation effect of the fertilization-irrigation synergy program, updating the rule weights of the fuzzy logic controller and the fitness function parameters of the genetic algorithm, optimizing the collaborative decision-making framework, forming an adaptive closed-loop optimization management system of monitoring-early warning-decision-execution-evaluation, realizing the autonomous calibration of system parameters and continuous evolution of strategies, and gradually improving the accuracy and long-term stability of farmland soil organic carbon management;

[0070] The specific work involves the following steps: After the fertilization-irrigation synergy program is implemented, the multi-parameter intelligent sensor network immediately switches to a high-frequency post-effect monitoring mode. pH and available phosphorus sensors continuously collect soil data from the 0-20 cm topsoil layer at a frequency of no less than twice daily. Simultaneously, temperature and humidity sensors record environmental parameters. All data includes GNSS coordinates and UTC timestamps, and monitoring continues for at least one complete crop growth cycle. The aim is to comprehensively capture the dynamic response curves of soil chemical indicators (pH, available phosphorus) and physical states (temperature, humidity) caused by fertilization and irrigation measures, providing a high spatiotemporal resolution continuous data sequence for effect evaluation. The data is uploaded in real-time via a low-power wide-area network to [the relevant network / system]. After cleaning and calibration, the cloud-based data platform stores the data, along with baseline data from before the implementation of the fertilization-irrigation synergy program, in a time-series database, forming a before-and-after comparison dataset. To quantitatively assess the management effectiveness and validate the model, standardized soil sampling was conducted in the target area after the monitoring period. Sampling followed the "NY / T1121.1-2006 Soil Testing" standard, focusing on the core risk area and control area from the early warning stage. Soil samples from the 0-20 cm topsoil layer were collected using a grid method, with statistically significant sample numbers (no less than 5 mixed samples per hectare). After air-drying and grinding, the soil organic carbon content was determined using the potassium dichromate external heating method, obtaining high-precision measured organic carbon change data. Subsequently... The measured organic carbon change data were compared with the predicted changes calculated by the organic carbon loss risk model based on sensor data during the same period. The root mean square error and coefficient of determination were calculated to quantitatively assess the model's predictive accuracy. If the assessment results showed systematic bias in the model, an online fine-tuning process was initiated. Newly collected paired data of soil chemical state → organic carbon change were used as the incremental training set. The weight parameters of the deep Q-network model were fine-tuned using mini-batch gradient descent, with a learning rate set to 0.001 and no more than 1000 iterations, to improve the model's predictive adaptability to carbon cycle dynamics under new management measures. This was based on complete closed-loop data of policy implementation, soil response, and organic carbon change. The chain initiates self-optimization of core decision parameters, analyzes the actual implementation effect of the fertilization-irrigation synergy scheme and its conformity with the preset target, and updates the rule base of the fuzzy logic controller based on the analysis results: adjust the confidence level of the relationship between pH, phosphorus content and organic carbon in the IF-THEN rule. If the data shows that the application of organic fertilizer has a significant effect on carbon sequestration within a specific pH range, the output weight of the corresponding rule will be increased. Secondly, optimize the fitness function of the genetic algorithm. Based on the counterfactual analysis of the results of multiple rounds of optimization and the actual effect, adjust the weighting coefficients of the three target components (soil organic carbon increment, pH deviation, and phosphorus content deviation), and dynamically reduce the weight of the target with poor actual effect from 0.3 to 0.25. This approach allows the algorithm's search direction to better align with actual field output, enabling the collaborative decision-making framework to self-calibrate based on historical results. Ultimately, by integrating the entire process of sensor monitoring → risk warning → strategy optimization → precise execution → effect evaluation → model and parameter updates, an adaptive closed-loop optimization management system is formed. This system can continuously improve decision quality and execution accuracy using practical feedback data, providing a self-evolving technological framework for the long-term, dynamic, and precise management of farmland soil health.

[0071] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0072] 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 management method for optimizing pH and phosphorus content to increase organic carbon in farmland soil, characterized in that, Includes the following steps: S1. Deploy a multi-parameter intelligent sensor network covering multiple sensor nodes for pH, phosphorus content, temperature and humidity in the target farmland area, and optimize the layout of the sensor nodes through the fruit fly algorithm. S2. Utilize the deployed multi-parameter intelligent sensor network to continuously collect multi-dimensional sensing data on soil pH, available phosphorus content, temperature and humidity, and upload them to the cloud platform in real time via wireless network to build a farmland soil chemical environment database. S3. Based on historical and real-time multidimensional sensor data of the target farmland area, a model for organic carbon loss risk is constructed using reinforcement learning technology. A dynamic mapping relationship between soil chemical state and organic carbon loss risk is established to assess the organic carbon loss risk in real time and issue early warnings. S4. A collaborative decision-making framework is designed by combining fuzzy logic control and genetic algorithm. Fuzzy logic integrates agronomic rules and real-time environmental uncertainties to output a preliminary soil organic carbon regulation strategy. The genetic algorithm iteratively optimizes the output of a precise quantitative management strategy covering the coordinated regulation of soil pH and phosphorus content. S5. Based on the early warning results and collaborative decision-making framework, generate a fertilization-irrigation synergy plan that takes into account both pH adjustment and phosphorus activity enhancement, and execute precise operations through intelligent agricultural machinery and irrigation systems; S6. Based on the multi-dimensional sensing data of the target farmland after regulation and the feedback on changes in soil organic carbon, update the parameters of the organic carbon loss risk model and collaborative decision-making framework to achieve a stable increase in soil carbon storage. S3 specifically includes: Multidimensional sensor data on pH, phosphorus content, temperature, and humidity, including historical and real-time data, were extracted from the farmland soil chemical environment database and used as the input feature set for model training. A reinforcement learning algorithm using deep Q-networks is employed to train and establish an organic carbon loss risk model with the loss risk level as the output, thereby establishing a nonlinear dynamic mapping relationship between the input feature set and the organic carbon loss risk. Using the trained organic carbon loss risk model, online reasoning is performed on the real-time input multidimensional sensor data to dynamically assess the loss risk level under the current soil chemical state; S4 specifically includes: A collaborative decision-making framework combining fuzzy logic control and genetic algorithm design is introduced. A fuzzy logic controller is constructed using fuzzy logic control technology. Its inputs are the real-time monitored soil pH value and available phosphorus content. Its knowledge base integrates empirical rules in agronomy regarding pH, phosphorus availability, and organic carbon stability. Specifically, pH value is divided into three fuzzy subsets: acidic, suitable, and alkaline. Its membership function adopts a combination of trapezoidal and triangular forms, with the intersection points set at pH 5.5, 6.0, 7.0, 7.5, and 8.0, respectively. Available phosphorus content is divided into three fuzzy subsets: deficient, suitable, and excessive, with the critical points set at 15 mg / kg, 20 mg / kg, 50 mg / kg, and 60 mg / kg, respectively. The fuzzy logic controller performs fuzzy reasoning based on its inputs and outputs a preliminary soil organic carbon control strategy regarding the amount of organic fertilizer applied, irrigation water volume, and acid-base regulating substances. The preliminary soil organic carbon control strategy is then quantified into decision variables and input into a genetic algorithm for iterative optimization.

2. The management method for optimizing pH and phosphorus content to increase farmland soil organic carbon according to claim 1, characterized in that: S1 specifically includes: Deploy a multi-parameter intelligent sensor network, including pH sensors, phosphorus content sensors, and temperature and humidity sensors, within the target farmland area to form a synchronous monitoring capability for key chemical and physical parameters of the soil. Based on farmland topography, soil type and historical variation data, the fruit fly algorithm is used to spatially optimize the deployment location of each sensor node in the multi-parameter intelligent sensor network, so as to minimize the monitoring blind zone and improve the spatial representativeness of the sampling points. Based on the optimized layout, each sensor node is fixed, and a wireless transmission module and power system are configured to establish an IoT monitoring infrastructure capable of collecting multi-dimensional soil parameter data in real time.

3. The management method for optimizing pH and phosphorus content to increase farmland soil organic carbon according to claim 1, characterized in that: S2 specifically includes: Through a multi-parameter intelligent sensor network, soil pH, available phosphorus content and soil temperature and humidity are automatically collected at a preset frequency, and the corresponding time and geographical coordinate information are recorded and integrated to form a multi-dimensional sensor data sequence. Low-power wide-area networks are used to upload data from the collected multidimensional sensor data sequences to the cloud data platform in real time. The received multidimensional sensor data is cleaned, calibrated, and fused on a cloud-based data platform to construct a farmland soil chemical environment database with high spatiotemporal resolution.

4. The management method for optimizing pH and phosphorus content to increase farmland soil organic carbon according to claim 1, characterized in that: S3 further includes: When the loss risk level output by the organic carbon loss risk model exceeds the preset safety threshold, an early warning mechanism is automatically triggered. The triggered early warning mechanism generates graded early warning information based on the level of churn risk, including the risk location, the dominant imbalance factor, and the potential severity. The graded early warning information is then fed back to the management terminal in real time through a visual interface and message push.

5. The management method for optimizing pH and phosphorus content to increase farmland soil organic carbon according to claim 1, characterized in that: S4 further includes: A genetic algorithm with multiple objectives—increasing soil organic carbon storage, stabilizing pH within a suitable range, and optimizing phosphorus availability—was used to encode decision variables and simulate selection, crossover, and mutation operations in the process of biological evolution. Under the constraints of agronomic and environmental factors, the algorithm iteratively searches for the optimal combination of organic fertilizer type, application rate, irrigation strategy, and acid / alkali adjustment measures. When the genetic algorithm converges or reaches the maximum number of iterations, it outputs a set of precise quantitative management strategies for synergistically regulating soil pH and phosphorus content.

6. The management method for optimizing pH and phosphorus content to increase farmland soil organic carbon according to claim 1, characterized in that: S5 specifically includes: The system analyzes precise quantitative management strategies and hierarchical early warning information to generate fertilization-irrigation synergy plans that take into account both pH adjustment and phosphorus activity enhancement. The fertilization-irrigation synergy plans are then broken down into executable agricultural operation instructions, including the amount and location of organic fertilizer application, irrigation duration and flow rate, and whether soil conditioner needs to be applied. Agricultural operation instructions are converted into control instructions for smart agricultural machinery and precision irrigation systems. Operation paths and times are planned, and smart agricultural machinery equipped with variable fertilization devices and irrigation systems with controllable valves are driven by Internet of Things instructions to implement a fertilization-irrigation synergy scheme in the target farmland area.

7. The management method for optimizing pH and phosphorus content to increase farmland soil organic carbon according to claim 6, characterized in that: The S5 also includes: During intelligent agricultural machinery operations, onboard sensors monitor fertilizer application amount, location, and soil disturbance in real time to ensure that operations are consistent with strategy requirements. During irrigation, the opening of the irrigation valve is dynamically adjusted based on feedback from the flow meter and soil moisture sensor to achieve precise on-demand water supply. After completing a round of adjustment operations, the system automatically records the details of the executed strategy and operational parameters, which will serve as a basis for subsequent evaluation and learning.

8. The management method for optimizing pH and phosphorus content to increase farmland soil organic carbon according to claim 1, characterized in that: S6 specifically includes: After the fertilization-irrigation synergy program is implemented, the dynamic response data of soil pH, phosphorus content and temperature and humidity will continue to be monitored using a multi-parameter intelligent sensor network. Soil samples were taken from the target farmland area to measure changes in organic carbon, the accuracy of the organic carbon loss risk model was evaluated, and the model parameters were fine-tuned online using the new data. Based on the execution effect of the fertilization-irrigation synergy scheme, the rule weights of the fuzzy logic controller and the fitness function parameters of the genetic algorithm are updated to optimize the collaborative decision-making framework and form an adaptive closed-loop optimization management system of monitoring-early warning-decision-execution-evaluation.

Images