Farmland soil multi-parameter real-time monitoring system and method based on Internet of Things

Abstract

The invention provides a farmland soil multi-parameter real-time monitoring system and method based on the Internet of Things. The system comprises a sensor module, a sensor collaborative scheduling module, a data transmission module, a data processing and storage module, an intelligent decision module, a user interaction module and a power supply module. Soil monitoring data, meteorological satellite information and historical yield records are fused, deep learning, a decision tree and a nutrient balance algorithm are combined, an accurate fertilization and irrigation scheme is generated, maintenance decision optimization is achieved, the water and fertilizer utilization efficiency is improved, and resource waste caused by extensive management is reduced; through dynamic sampling, data cross validation, sensor fault self-diagnosis and multi-source data fusion, the system can make a decision, effectively solves the defects of an existing system in the aspects of monitoring precision, data reliability, equipment endurance and decision refinement, and provides an efficient and reliable monitoring and management scheme for precision agriculture.

Description

Technical Field

[0001] This invention belongs to the field of agricultural management technology and relates to a real-time monitoring system and method for multiple parameters of farmland soil. Background Technology

[0002] Real-time monitoring systems for farmland soil parameters are an advanced agricultural information technology. Through various sensors deployed in farmland, they continuously and automatically monitor and collect data on key soil parameters. These parameters typically include soil moisture, temperature, electrical conductivity (reflecting soil salinity), pH value, and nutrient content such as nitrogen, phosphorus, and potassium. The system acquires this data in real time and transmits it to a central server or cloud platform via wireless communication networks for analysis and processing. Farmers or agricultural managers can view soil conditions anytime via computer or mobile device, gaining timely insights into crop growth environments and nutrient requirements. This enables precise irrigation, fertilization, and pest and disease control, improving agricultural productivity and crop yield while reducing water and fertilizer waste and environmental pollution. Furthermore, real-time monitoring systems can be integrated with agricultural expert systems and meteorological data to provide more comprehensive and scientific decision support for agricultural production, making them a crucial technological means to promote the intelligent and precise development of modern agriculture.

[0003] However, existing real-time monitoring systems for farmland soil parameters generally suffer from the following problems: fixed sampling frequency makes it difficult to balance monitoring accuracy and energy consumption; the reliability of data from a single sensor is easily affected by environmental interference; the equipment's battery life depends on continuous sunlight and lacks intelligent scheduling; and the decision-making model does not integrate multi-source data, resulting in extensive fertilization and irrigation schemes and abnormal early warnings lagging behind the soil degradation process. These issues restrict the practicality and sustainability of real-time monitoring systems for farmland soil parameters and make it difficult to support the real-time and refined management and control requirements for high-standard farmland construction. Summary of the Invention

[0004] To address the problems described in the background art regarding existing real-time monitoring systems for farmland soil parameters, such as fixed sampling frequency leading to difficulty in balancing monitoring accuracy and energy consumption, susceptibility of single-sensor data reliability to environmental interference, reliance on continuous lighting for equipment endurance lacking intelligent scheduling, and the lack of integration of multi-source data in decision-making models, this invention provides a real-time monitoring system and method for multiple parameters of farmland soil based on the Internet of Things.

[0005] The system of this invention includes a sensor module, a sensor collaborative scheduling module, a data transmission module, a data processing and storage module, an intelligent decision-making module, a user interaction module, and a power supply module; The sensor module includes multiple sensors arranged topologically in farmland soil to acquire soil monitoring data. The sensors include a high-precision capacitive humidity probe, a four-needle conductivity detection unit, and an anti-polarization pH electrode. The sensors are directly connected to the sensor coordination and scheduling module. The sensor collaborative scheduling module includes a sampling frequency controller submodule, an energy consumption optimization scheduler submodule, a multi-sensor cross-validation submodule, and an FPGA signal receiving submodule. The sampling frequency controller submodule analyzes the historical trends of soil parameter changes over a period of time, dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model, and controls the sensor sampling interval. The energy consumption optimization scheduler submodule monitors network signal strength and remaining power in real time, adjusting the signal transmission method and the power supply voltage of the power supply module. The multi-sensor cross-validation submodule constructs a physical correlation model between soil parameters and a data fusion algorithm to perform multiple verifications of soil monitoring data and self-diagnose sensor faults, obtaining cross-validated soil monitoring data. The FPGA signal receiving submodule identifies different sensor data formats, automatically completes timestamp synchronization, unit unification, and data packet reassembly, and obtains fused multi-source heterogeneous data. The data transmission module is used to acquire total monitoring data, including soil monitoring data from the sensor module, soil monitoring data after cross-validation by the sensor collaborative scheduling module, and fused multi-source heterogeneous data, and transmit it to the data processing and storage module. The data processing and storage module is used to preprocess the total monitoring data and then inject it into the intelligent decision-making module; The intelligent decision-making module is used to integrate pre-processed total monitoring data with plot-level soil testing data, meteorological satellite information and historical yield records. Through deep learning algorithms, core decision tree algorithms, nutrient balance methods and crop fertilizer requirements, it generates maintenance recommendation plans that include fertilization timing, type and dosage. The user interaction module is used to display content including soil monitoring data, network signal strength, signal transmission method, remaining power, power supply voltage, cross-validated soil monitoring data, sensor fault status, and maintenance recommendation plan. The power supply module is used to supply power to the entire system.

[0006] Furthermore, all sensors are equipped with standardized magnetic interfaces and waterproof connectors. The sensor module is externally equipped with an IP68 protective shell. The sensor module integrates a self-cleaning structure, which prevents salt crystallization on the electrode surface by periodically releasing microcurrent pulses. The topological arrangement of the sensors in the soil enables multiple sensors to form a complementary detection network within a certain radius area. The topological arrangement is a honeycomb distributed topological structure, with regular hexagonal grid nodes arranged within a certain radius area. Each grid node is equipped with a sensor, and each grid node has a built-in telescopic probe that covers multiple depths of vertical gradient detection covering the surface, root system, and subsurface. Dual-grid nodes are deployed in key areas of irrigation zones and fertility gradient zones. The dual-grid nodes are a main grid node and a backup grid node. The backup grid node automatically switches to working mode when the main grid node fails.

[0007] Furthermore, the sampling frequency controller submodule of the sensor collaborative scheduling module includes an embedded prediction engine unit, a mutation response unit, and a regional coordination unit. The embedded prediction engine analyzes the historical trend of soil parameter changes over a period of time and dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model. When the minute-level change rate of key parameters such as humidity or temperature exceeds a preset threshold, the mutation response unit is activated, switching to an emergency sampling mode and increasing the sampling frequency to ten times that of the normal state. The regional coordination unit obtains real-time data from adjacent monitoring nodes through a LoRa self-organizing network and initiates a collaborative enhanced sampling strategy for areas with sudden changes in soil parameters. The lightweight recurrent neural network model employs an improved gated recurrent unit (GRU). This improved GRU compresses parameters compared to the traditional GRU by removing the reset gate bias term. It also performs dimensionality reduction on features: the input layer retains only three key features—humidity change rate Δθ / Δt, temperature fluctuation variance σ_T², and conductivity gradient ∇EC—and extracts temporal patterns using a sliding window, with the window size set according to actual needs. Furthermore, the improved GRU performs online learning using incremental gradient descent, fine-tuning the model parameters once every certain period using historical data. The collaborative enhancement sampling strategy includes the following mechanisms: Event broadcast: Triggering a grid node to send a broadcast packet via LoRa to neighboring grid nodes within a certain radius, containing the parameter mutation type, location coordinates, and confidence level between 0 and 1; Decision tree response: The receiving grid node decides whether to initiate enhanced sampling based on three conditions: battery level is higher than M%, current sampling period is greater than N minutes, and distance to the event point is less than P meters. If all conditions are met: the sampling frequency is immediately increased to tens of times the original and maintained for a period of time; if some conditions are met: the sampling frequency is increased to several times the original and maintained for a period of time; if not met: the original period is maintained but the event log is recorded. The values ​​of M, N, and P are set according to actual needs. Data aggregation: After ≥X grid nodes in the region confirm the same mutation, an “regional anomaly” flag is automatically sent to the gateway, triggering the cloud data backup mechanism. The value of X is set according to actual needs, and X>1.

[0008] Furthermore, the energy consumption optimization scheduler submodule of the sensor collaborative scheduling module includes a communication mode decision unit, a dynamic power supply adjustment unit, and a photovoltaic prediction unit; The communication decision unit monitors the network signal strength and remaining power in real time. When the signal is stable and the power is sufficient, it prioritizes high-speed 4G transmission. When the signal weakens or the power is below the safety threshold, it automatically switches to low-power LoRa mode. The dynamic power supply adjustment unit intelligently switches the power supply voltage according to the sensor's working status. During normal monitoring, it uses a low voltage of 1.8V to maintain basic operation, and switches to a full power mode of 3.3V during high-frequency sampling. The photovoltaic prediction unit dynamically adjusts the equipment's working cycle based on the actual sunlight conditions in the farmland, and matches energy supply and consumption through a photovoltaic prediction algorithm; The photovoltaic prediction algorithm adopts a two-stage hybrid prediction model, specifically including: Short-term forecast: forecast time ≤ 1 hour, based on real-time sampling data from a light sensor, sampling time t The future light intensity is predicted using an exponentially weighted moving average algorithm, minute by minute, with the following formula: ; Where α is the weighting coefficient, which is set according to actual needs; For future light intensity, Given the current light intensity, Historical light intensity; Medium-term forecast: 1 hour < forecast time ≤ 24 hours, integrating cloud cover data from meteorological satellites, using a lightweight BP neural network. The lightweight BP neural network input layer: cloud cover, longitude, time period; the lightweight BP neural network output layer: light intensity level, updating the forecast curve daily to predict future light intensity. The device dynamically adjusts its operating mode to match energy supply and consumption based on predicted future light intensity: when there is sufficient light, the device uses full-speed sampling + data transmission + battery charging; when the light is moderate, it maintains full-speed sampling but reduces wireless transmission power; when the light is insufficient, it initiates sleep scheduling, only keeping the humidity sensor awake once every once in a while.

[0009] Furthermore, the multi-sensor cross-validation submodule of the sensor collaborative scheduling module includes: a physical model verification layer that establishes nonlinear constraint relationships between parameters based on the chemical kinetic coupling equation of soil moisture-conductivity-pH value, triggering collaborative verification when single sensor data exceeds the theoretical range; a redundant data fusion layer that dynamically weights and fuses multiple sensors of the same type deployed within a certain radius using an improved Kalman filter algorithm, calculating sensor confidence weights in real time; and a fault diagnosis layer that integrates an electrochemical impedance spectroscopy analysis unit, constructing a Nyquist spectrum to detect electrode polarization state by applying a 10Hz-100kHz scanning signal to the sensor electrodes, predicting sensor lifespan by combining an Arrhenius aging model, and generating a maintenance warning when the characteristic frequency phase angle shift exceeds 15°.

[0010] Furthermore, the chemical kinetic coupling equation for soil moisture-electrical conductivity-pH is as follows: ; Where α represents the soil type correction coefficient; θ represents the volumetric water content; T represents the soil temperature; β represents the ion migration compensation index; NPK represents the total nitrogen, phosphorus, and potassium nutrient content; and γ represents the nutrient coupling coefficient. The formula for the improved Kalman filter algorithm is as follows: ; Where Q0 represents the basic process noise covariance; σ err σ represents the standard deviation of the sensor's recent error. base Indicates the factory calibration error reference value; λ t η represents the noise sensitivity coefficient; η represents the power supply stability weighting factor; ∇P supply This indicates the change in the supply voltage gradient.

[0011] Furthermore, the FPGA signal receiving submodule of the sensor collaborative scheduling module is internally equipped with a built-in high-speed digital interface, a signal preprocessing unit, and an adaptive protocol parsing engine. The hardware layer is equipped with an eight-channel synchronous receiving circuit. The signal preprocessing unit is equipped with a real-time verification algorithm, which verifies data integrity through CRC16 error correction code and Manchester encoding, and initiates a triple redundancy retransmission mechanism for abnormal signals. The protocol parsing engine dynamically identifies different sensor data formats, automatically completes timestamp synchronization, unit unification, and data packet reassembly, and obtains fused multi-source heterogeneous data. The FPGA signal receiving submodule is also equipped with an anti-interference shielding structure, a temperature compensation circuit, and an intelligent buffer manager.

[0012] Furthermore, the intelligent decision-making module integrates preprocessed total monitoring data with plot-level soil testing data, meteorological satellite information, and historical yield records, and establishes a multi-dimensional feature space through deep learning algorithms; The core decision tree algorithm combines the nutrient balance method with crop fertilizer requirements to generate maintenance recommendation plans that include the timing, type, and dosage of fertilization; the configuration visualization engine maps the spatial variation characteristics of soil parameters into heat maps, and, together with aerial imagery, locates areas of soil fertility deficiency. The multidimensional feature space serves as the input engine for the core decision tree algorithm, comprising 42 features across 5 categories: Soil physicochemical features (12 dimensions, including spatiotemporal gradients of humidity, EC, pH, and NPK content at each layer); Meteorological features (8 dimensions, including sliding window statistics of precipitation, accumulated temperature, sunshine duration, and wind speed); Crop physiological features (6 dimensions, including vegetation cover, leaf area index (LAI), and growth stage based on NDVI remote sensing); Historical decision features (9 dimensions, including the type, dosage, and effect feedback coefficient of the past three fertilizations); and Environmental disturbance features (7 dimensions, including soil compaction, organic matter content, and heavy metal background values). These features, after standardization, form a 128×128 feature matrix, providing structured input for the subsequent training of the core decision tree algorithm. The core decision tree algorithm combines the nutrient balance method with crop fertilizer requirements to generate maintenance recommendation plans that include the timing, type, and dosage of fertilization; the configuration visualization engine maps the spatial variation characteristics of soil parameters into heat maps, and, together with aerial imagery, locates areas of soil fertility deficiency. The operation process of the core decision tree algorithm includes: Feature filtering: Redundant features were removed. Root layer humidity data were retained when the correlation between surface layer and root layer humidity was higher than 0.95, and the dimensionality was compressed to within 25 dimensions. The core decision tree algorithm model inference includes: The underlying model uses a random forest regressor to predict the basic fertilizer application rate, with a determination coefficient R² greater than 0.85. Mid-level optimization: Solve multi-objective optimization problems using the improved NSGA-III algorithm. The optimization objectives include maximizing yield, minimizing fertilizer cost, and minimizing environmental risk. Upper-level adjustments: Regular adjustments are made based on farmers' historical acceptance levels; Solution generation: Outputs a three-dimensional decision result containing the following elements: Spatial dimension: Differentiated fertilizer dosage is generated by kriging interpolation to create a grid of a certain area; Time dimension: The optimal fertilization window within the next T days, avoiding periods with a rainy day probability greater than Z%; where the values ​​of T and Z are set according to actual needs; Resource dimension: Recommended fertilizer type combinations and application methods; Execution feedback: The actual fertilization execution data is sent back to the core decision tree algorithm model for parameter iteration, and the decision tree splitting threshold is updated every once in a while.

[0013] Furthermore, the power generation unit of the power supply module is a solar tracking array composed of a curved monocrystalline silicon plate and a dual-degree-of-freedom gimbal, which is controlled in conjunction with a light intensity sensor and an astronomical algorithm. The energy storage unit of the power supply module is equipped with an intelligent charging and discharging router, which dynamically allocates the collaborative working mode of the lithium iron phosphate battery pack and the supercapacitor according to the power load prediction.

[0014] Based on the above system, this invention proposes a real-time monitoring method for multiple parameters of farmland soil based on the Internet of Things, including: The sensor module collects monitoring data of farmland soil through sensors including a high-precision capacitive humidity probe, a four-needle conductivity detection unit, and an anti-polarization pH electrode, and transmits the data to the sensor collaborative scheduling module. The sampling frequency controller submodule of the sensor collaborative scheduling module analyzes the trend of soil parameter changes over a historical period, dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model, and controls the sampling interval of the sensors. The energy consumption optimization scheduler submodule of the sensor collaborative scheduling module monitors the network signal strength and remaining power in real time, and adjusts the signal transmission method and the power supply voltage of the power supply module. The multi-sensor cross-validation submodule of the sensor collaborative scheduling module constructs a physical correlation model between soil parameters and a data fusion algorithm to perform multiple verifications of soil monitoring data and self-diagnose sensor faults, obtaining cross-validated soil monitoring data. The FPGA signal receiving submodule of the sensor collaborative scheduling module is used to identify different sensor data formats, automatically complete timestamp synchronization, unit unification, and data packet reassembly, and obtain fused multi-source heterogeneous data. The data transmission module acquires total monitoring data, including soil monitoring data from the sensor module, soil monitoring data cross-validated by the sensor collaborative scheduling module, and fused multi-source heterogeneous data, and transmits it to the data processing and storage module. The data processing and storage module preprocesses the total monitoring data and then injects it into the intelligent decision-making module; The intelligent decision-making module integrates pre-processed total monitoring data with plot-level soil testing data, meteorological satellite information, and historical yield records. Through deep learning algorithms, core decision tree algorithms, nutrient balance methods, and crop fertilizer requirements, it generates maintenance recommendations that include the timing, type, and dosage of fertilization. The user interaction module displays information including soil monitoring data, network signal strength, signal transmission method, remaining power, power supply voltage, cross-validated soil monitoring data, sensor fault status, and recommended maintenance solutions.

[0015] Compared with the prior art, the present invention has the following advantages: (1) High accuracy of monitoring and sampling: The sampling frequency of the sensor is dynamically adjusted by a lightweight recurrent neural network, which ensures the accuracy of monitoring and sampling; (2) Energy consumption optimization: Through the energy consumption optimization scheduler submodule, the signal transmission mode and power supply voltage are intelligently adjusted according to the network signal strength and remaining power to improve the system's endurance; (3) High data reliability: Through multi-sensor cross-validation and physical correlation model, soil monitoring data is verified multiple times to achieve sensor fault self-diagnosis and improve data reliability; (4) Unified data format: The FPGA signal receiving submodule is used to realize the automatic synchronization, format unification and recombination of multi-source heterogeneous data, thereby enhancing data consistency; (5) Intelligent decision-making and precision agricultural management: By integrating soil monitoring data, meteorological satellite information and historical yield records, and combining deep learning, decision trees and nutrient balance algorithms, precise fertilization and irrigation plans are generated, which optimizes maintenance decisions, improves water and fertilizer utilization efficiency, and reduces resource waste caused by extensive management. (6) User-friendly and real-time interaction: The user interaction module intuitively displays soil parameters, equipment status (such as power and signal strength) and maintenance suggestions, which facilitates real-time monitoring and decision-making by agricultural managers.

[0016] This invention effectively addresses the shortcomings of existing systems in terms of monitoring accuracy, data reliability, equipment endurance, and decision-making precision through dynamic sampling, data cross-validation, sensor fault self-diagnosis, multi-source data fusion, and intelligent decision-making, providing an efficient and reliable monitoring and management solution for precision agriculture. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of the present invention.

[0018] Figure 2 This is an architecture diagram of the sensor collaborative scheduling module of the present invention. Detailed Implementation

[0019] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0020] Example 1 The architecture diagram of the IoT-based real-time multi-parameter monitoring system for farmland soil is as follows: Figure 1As shown, it consists of a sensor module, a sensor collaborative scheduling module, a data transmission module, a data processing and storage module, an intelligent decision-making module, a user interaction module, and a power supply module.

[0021] The sensor module includes multiple sensors arranged topologically in farmland soil to acquire soil monitoring data. The sensors include a high-precision capacitive humidity probe, a four-needle conductivity detection unit, and an anti-polarization pH electrode. The sensors are directly connected to the sensor coordination scheduling module.

[0022] Specifically, all sensors are equipped with standardized magnetic interfaces and waterproof connectors, allowing for easy replacement with a single hand press during field maintenance, without the need for power interruption or tools. The sensor module is externally protected by an IP68 housing, and its internal structure integrates a self-cleaning mechanism. This mechanism prevents salt crystallization on the electrode surface by periodically releasing microcurrent pulses, ensuring stable operation under long-term buried conditions.

[0023] Specifically, the topological arrangement of sensors in the soil enables multiple sensors to form a complementary detection network within a certain radius. More specifically, this topological arrangement is a honeycomb distributed topology, with hexagonal grid nodes deployed within a certain radius, each grid node carrying a sensor. More specifically, each sensor within a grid node incorporates a telescopic probe capable of detecting multiple depths of vertical gradients covering the surface, root zone, and subsurface. Dual-grid nodes are deployed in key areas of the irrigation zone and fertility gradient zone. These dual-grid nodes serve as both primary and backup grid nodes. The backup grid node automatically switches to operating mode in the event of a primary grid node failure, ensuring data continuity.

[0024] In this embodiment, the sensor's spatial layout is as follows: a hexagonal grid of nodes with a radius of 10 meters is used. Each grid node is equipped with a high-precision capacitive humidity probe, a four-pin conductivity detection unit, and an anti-polarization pH electrode. Adjacent nodes overlap to cover 50% of the detection area, forming a monitoring network without blind spots. The sensor's telescopic probe can achieve vertical gradient detection at three depths: 0-20cm, 20-40cm, and 40-60cm, corresponding to the surface layer, root layer, and subsurface layer, respectively. Data synchronization between sensors at different depths within the same grid node is achieved via a CAN bus. This embodiment, through a specially designed topology arrangement, requires only 25-30 nodes to be deployed in 1 hectare of farmland, allowing multiple sensors to form a complementary detection network within a 30cm radius, effectively eliminating single-point measurement errors. Actual measurement data shows that the three-dimensional spatial interpolation accuracy of soil parameters is improved to over 92%.

[0025] The architecture diagram of the sensor collaborative scheduling module is as follows: Figure 2As shown, the system consists of a sampling frequency controller submodule, an energy consumption optimization scheduler submodule, a multi-sensor cross-validation submodule, and an FPGA signal receiving submodule. The sampling frequency controller submodule analyzes the historical trend of soil parameter changes over a period of time, dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model, and controls the sampling interval of the sensors. The energy consumption optimization scheduler submodule monitors the network signal strength and remaining power in real time, and adjusts the signal transmission method and the power supply voltage of the power supply module. The multi-sensor cross-validation submodule constructs a physical correlation model between soil parameters and a data fusion algorithm to perform multiple verifications of soil monitoring data and self-diagnose sensor faults, obtaining cross-validated soil monitoring data. The FPGA signal receiving submodule is used to identify different sensor data formats, automatically complete timestamp synchronization, unit unification, and data packet reassembly, and obtain fused multi-source heterogeneous data.

[0026] Specifically, the sampling frequency controller submodule of the sensor collaborative scheduling module includes an embedded prediction engine unit, a mutation response unit, and a regional coordination unit. The embedded prediction engine analyzes the trend of soil parameter changes over a historical period and dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model. When the minute-level change rate of key parameters such as humidity or temperature exceeds a preset threshold, the mutation response unit is activated, switching to an emergency sampling mode and increasing the sampling frequency to ten times that of the normal state. The regional coordination unit obtains real-time data from adjacent monitoring nodes through a LoRa self-organizing network and initiates a collaborative enhanced sampling strategy for areas with sudden changes in soil parameters to ensure the complete capture of high-value data.

[0027] More specifically, the lightweight recurrent neural network model employs an improved gated recurrent unit, which performs parameter compression on the traditional gated recurrent unit by removing the reset gate bias term. The improved gated recurrent unit also performs dimensionality reduction on the features: the input layer retains only three key features—humidity change rate Δθ / Δt, temperature fluctuation variance σ_T², and conductivity gradient ∇EC—and extracts temporal patterns using a sliding window, with the window size set according to actual needs. Furthermore, the improved gated recurrent unit performs online learning by employing incremental gradient descent, fine-tuning the model parameters once every certain period using historical data to avoid the high energy consumption of full retraining.

[0028] In this embodiment, the lightweight recurrent neural network model uses an improved gated recurrent unit, namely GRU-Lite, and the specific optimizations include: Parameter compression: By removing the reset gate bias term of the traditional GRU, the number of hidden layer neurons is reduced from 128 to 32, and the model volume is compressed by 60%, making it suitable for deployment on embedded ARM Cortex-M4 chips; Feature dimensionality reduction: The input layer retains only three key features: humidity change rate Δθ / Δt, temperature fluctuation variance σ_T², and conductivity gradient ∇EC. Temporal patterns are extracted through a sliding window with a window size of 15 minutes. Online learning: Incremental gradient descent is used to fine-tune the model parameters every 24 hours using historical data, avoiding the high energy consumption of full retraining.

[0029] Through the above optimizations, this embodiment achieves a sampling interval prediction error of <8% and a computation time of <20ms in a 512KB memory environment.

[0030] More specifically, the collaborative reinforcement sampling strategy includes the following mechanisms: Event broadcast: Triggering a grid node to send a broadcast packet via LoRa to neighboring grid nodes within a certain radius, containing the parameter mutation type, location coordinates, and confidence level between 0 and 1; Decision tree response: The receiving grid node decides whether to initiate enhanced sampling based on three conditions: battery level is higher than M%, current sampling period is greater than N minutes, and distance to the event point is less than P meters. If all conditions are met: the sampling frequency is immediately increased to tens of times the original and maintained for a period of time; if some conditions are met: the sampling frequency is increased to several times the original and maintained for a period of time; if not met: the original period is maintained but the event log is recorded. The values ​​of M, N, and P are set according to actual needs. Data aggregation: After ≥X grid nodes in the region confirm the same mutation, an “regional anomaly” flag is automatically sent to the gateway, triggering the cloud data backup mechanism. The value of X is set according to actual needs, and X>1.

[0031] In this embodiment, the collaborative enhancement sampling strategy includes the following mechanisms: Event broadcast: The triggering node sends a broadcast packet to neighboring nodes within a 50-meter radius via LoRa, containing parameter mutation types such as sudden increase in humidity, location coordinates, and confidence levels between 0 and 1; Decision tree response: The receiving node decides whether to initiate enhanced sampling based on three conditions: battery level is above 30%, current sampling period is greater than 5 minutes, and distance to the event point is less than 30 meters. If all conditions are met: the sampling frequency is immediately increased to 5 seconds / time for 10 minutes. If some conditions are met: the frequency is increased to 30 seconds / time for 5 minutes. If no conditions are met: the original period is maintained but the event log is recorded. Data aggregation: After ≥3 nodes within a region confirm the same type of mutation, an "regional anomaly" flag is automatically sent to the gateway, triggering the cloud data backup mechanism; This embodiment uses the above strategy to increase the data collection density in abnormal areas by 3-5 times, while reducing ineffective energy consumption by 70% through the neighbor node filtering mechanism.

[0032] Specifically, the energy consumption optimization scheduler submodule of the sensor collaborative scheduling module includes a communication mode decision unit, a dynamic power supply adjustment unit, and a photovoltaic prediction unit. The communication decision unit monitors network signal strength and remaining power in real time. When the signal is stable and the power is sufficient, it prioritizes high-speed 4G transmission. When the signal weakens or the power falls below a safe threshold, it automatically switches to low-power LoRa mode. The dynamic power supply adjustment unit intelligently switches the power supply voltage according to the sensor's operating status. During normal monitoring, it uses a low voltage of 1.8V to maintain basic operation, and switches to a full-power mode of 3.3V during high-frequency sampling. The photovoltaic prediction unit dynamically adjusts the equipment's operating cycle based on the actual sunlight conditions in the farmland, matching energy supply and consumption through a photovoltaic prediction algorithm to reduce ineffective energy consumption.

[0033] More specifically, the photovoltaic prediction algorithm employs a two-stage hybrid prediction model, which includes: Short-term forecast: forecast time ≤ 1 hour, based on real-time sampling data from a light sensor, sampling time t The future light intensity is predicted using an exponentially weighted moving average algorithm, minute by minute, with the following formula: ; Where α is the weighting coefficient, which is set according to actual needs; For future light intensity, Given the current light intensity, Historical light intensity; Medium-term forecast: 1 hour < forecast time ≤ 24 hours, integrating cloud cover data from meteorological satellites, using a lightweight BP neural network. The lightweight BP neural network input layer: cloud cover, longitude, time period; the lightweight BP neural network output layer: light intensity level, updating the forecast curve daily to predict future light intensity. The device dynamically adjusts its operating mode to match energy supply and consumption based on predicted future light intensity: when there is sufficient light, the device uses full-speed sampling + data transmission + battery charging; when the light is moderate, it maintains full-speed sampling but reduces wireless transmission power; when the light is insufficient, it initiates sleep scheduling, only keeping the humidity sensor awake once every once in a while.

[0034] In this embodiment, the photovoltaic prediction algorithm adopts a two-stage hybrid prediction model, specifically including: Short-term forecast: For forecasts with a time horizon of ≤1 hour, based on real-time sampling data from a light sensor (sampling time 1 minute), an exponentially weighted moving average algorithm is used to predict future light intensity. The formula is as follows: ; Where α is the weighting coefficient, with a value of 0.7; For future light intensity, Given the current light intensity, Historical light intensity; Medium-term forecast: 1 hour < forecast time ≤ 24 hours, integrating cloud cover data from meteorological satellites, using a lightweight BP neural network. The lightweight BP neural network input layer: cloud cover, longitude, time period; the lightweight BP neural network output layer: light intensity level. The forecast curve is updated daily at 3:00 AM to predict future light intensity. Based on the predicted future light intensity, energy supply and consumption are matched, and the module's operating mode is dynamically adjusted: When there is sufficient light, the predicted future light intensity is >600 W / m², and the device adopts full-speed sampling + data transmission + battery charging; When the illumination is moderate, the predicted future illumination intensity is 300-600 W / m², so full-speed sampling is maintained but wireless transmission power is reduced; When there is insufficient light, if the predicted future light intensity is <300 W / m², a sleep schedule will be initiated, with only the humidity sensor being activated once every 15 minutes.

[0035] This embodiment uses the photovoltaic prediction algorithm described above to increase the photovoltaic energy utilization rate to 82% and extend the system's battery life to 72 hours under continuous cloudy and rainy days (<200 W / m²).

[0036] Specifically, the multi-sensor cross-validation submodule of the sensor collaborative scheduling module includes: a physical model verification layer based on the chemical kinetic coupling equation of soil moisture-conductivity-pH value, establishing nonlinear constraint relationships between parameters, and triggering collaborative verification when single sensor data exceeds the theoretical range; a redundant data fusion layer that dynamically weights and fuses multiple sensors of the same type deployed within a certain radius using an improved Kalman filter algorithm, calculating sensor confidence weights in real time; and a fault diagnosis layer integrating an electrochemical impedance spectroscopy analysis unit that constructs a Nyquist spectrum to detect electrode polarization state by applying a 10Hz-100kHz scanning signal to the sensor electrodes, and predicts sensor lifespan using an Arrhenius aging model, generating a maintenance warning when the characteristic frequency phase angle shift exceeds 15°. Verification shows that the multi-sensor cross-validation submodule can improve data reliability to 99.7% and reduce the false alarm rate to 1 / 5 of the traditional system.

[0037] More specifically, the chemical kinetic coupling equation for soil moisture-electrical conductivity-pH is: ; Where α represents the soil type correction coefficient; θ represents the volumetric water content; T represents the soil temperature; β represents the ion migration compensation index; NPK represents the total nitrogen, phosphorus, and potassium nutrient content; and γ represents the nutrient coupling coefficient. The formula for the improved Kalman filter algorithm is: ; Where Q0 represents the basic process noise covariance; σ err σ represents the standard deviation of the sensor's recent error. base Indicates the factory calibration error reference value; λ t η represents the noise sensitivity coefficient; η represents the power supply stability weighting factor; ∇P supply This indicates the change in the supply voltage gradient.

[0038] Specifically, the FPGA signal receiving submodule of the sensor collaborative scheduling module is internally equipped with a built-in high-speed digital interface, a signal preprocessing unit, and an adaptive protocol parsing engine. The hardware layer features an eight-channel synchronous receiving circuit, supporting parallel decoding of multiple communication protocols such as LoRa, 4G, and Bluetooth, and utilizing a pipelined architecture to achieve microsecond-level signal parsing. The signal preprocessing unit is equipped with a real-time verification algorithm, using CRC16 error correction codes and Manchester encoding to jointly verify data integrity and initiate a triple redundancy retransmission mechanism for abnormal signals. The protocol parsing engine dynamically identifies different sensor data formats, automatically completing timestamp synchronization, unit unification, and data packet reassembly to obtain fused multi-source heterogeneous data. The FPGA signal receiving submodule also features an anti-interference shielding structure, a temperature compensation circuit, and an intelligent buffer manager. For example, the anti-interference shielding structure and temperature compensation circuit maintain a bit error rate below 0.001% within an environment ranging from -30℃ to 75℃, and the configured intelligent buffer manager can dynamically adjust the data throughput according to network conditions, effectively solving the signal jitter problem in complex farmland environments.

[0039] The data transmission module is used to acquire total monitoring data, including soil monitoring data from the sensor module, soil monitoring data after cross-validation by the sensor collaborative scheduling module, and fused multi-source heterogeneous data, and transmit it to the data processing and storage module.

[0040] Specifically, for example, the data transmission module internally incorporates a LoRa self-organizing network base station, employing adaptive beamforming technology to dynamically adjust the signal transmission direction. Combined with solar-powered relay nodes deployed along field ridges, this forms a stable communication circle with a radius of 2 kilometers. A self-healing routing protocol is embedded in the network layer, enabling the system to automatically reconstruct the data transmission path within 200 milliseconds when a node is damaged due to mechanical operations. For critical agricultural data, the module incorporates a hardware encryption chip for end-to-end protection, combined with time-division multiple access (TDMA) technology to avoid channel congestion. In real-world testing, even with hundreds of concurrent nodes, it can still maintain a processing capacity of 50 data packets per second, with a packet loss rate controlled below 0.3%.

[0041] The data processing and storage module is used to preprocess the total monitoring data and then inject it into the intelligent decision-making module.

[0042] Specifically, for example, the core engine of the data processing and storage module is a distributed time-series database cluster. It uses a hash partitioning algorithm to slice and store 1 billion data points according to both plot coordinates and timestamps, supporting millisecond-level response to complex query requests. The real-time processing layer deploys a streaming computing framework, performing sliding window filtering on the raw data and automatically marking outliers using a soil parameter IoT model. The storage system features a specially designed hot and cold data tiering mechanism. Real-time monitoring data is stored in an in-memory database for the decision-making system to access, while historical data is compressed and transferred to object storage, reducing storage costs by 60% while ensuring that data from any period within the past three years can be retrieved in seconds.

[0043] The intelligent decision-making module integrates pre-processed total monitoring data with plot-level soil testing data, meteorological satellite information, and historical yield records. Through deep learning algorithms, core decision tree algorithms, nutrient balance methods, and crop fertilizer requirements, it generates maintenance recommendations that include the timing, type, and dosage of fertilization.

[0044] Specifically, the intelligent decision-making module integrates preprocessed total monitoring data with plot-level soil testing data, meteorological satellite information, and historical yield records, and establishes a multi-dimensional feature space through deep learning algorithms; the core decision tree algorithm combines the nutrient balance method with crop fertilizer requirements to generate maintenance recommendation plans that include fertilization timing, type, and dosage; and the configuration visualization engine maps the spatial variation characteristics of soil parameters into heat maps, and, together with aerial imagery, locates areas of soil fertility deficiency.

[0045] More specifically, the multidimensional feature space is the input engine of the core decision tree algorithm, comprising 42 features across 5 categories: Soil physicochemical features (12 dimensions, including spatiotemporal gradients of humidity, EC, pH, and NPK content at each layer); Meteorological features (8 dimensions, including sliding window statistics of precipitation, accumulated temperature, sunshine duration, and wind speed); Crop physiological features (6 dimensions, including vegetation cover, leaf area index (LAI), and growth stage based on NDVI remote sensing); Historical decision features (9 dimensions, including the type, dosage, and effect feedback coefficient of the past three fertilizations); and Environmental disturbance features (7 dimensions, including soil compaction, organic matter content, and heavy metal background values). These features, after standardization, form a 128×128 feature matrix, providing structured input for the subsequent training of the core decision tree algorithm.

[0046] The operation process of the core decision tree algorithm includes: Feature filtering: Redundant features were removed. Root layer humidity data were retained when the correlation between surface layer and root layer humidity was higher than 0.95, and the dimensionality was compressed to within 25 dimensions. The core decision tree algorithm model inference includes: Underlying model: A random forest regressor is used to predict the basal fertilizer application rate, with a coefficient of determination R² greater than 0.85. Mid-level optimization: Solve multi-objective optimization problems using the improved NSGA-III algorithm. The optimization objectives include maximizing yield, minimizing fertilizer cost, and minimizing environmental risk. Upper-level adjustments: Regular adjustments are made based on farmers' historical acceptance levels; Solution generation: Outputs a three-dimensional decision result containing the following elements: Spatial dimension: Differentiated fertilizer dosage is generated by kriging interpolation to create a grid with a certain area; Time dimension: The optimal fertilization window within the next T days, avoiding periods with a rainy day probability greater than Z%; where the values ​​of T and Z are set according to actual needs; Resource dimension: Recommended fertilizer type combinations and application methods; Execution feedback: The actual fertilization execution data is sent back to the core decision tree algorithm model for parameter iteration, and the decision tree splitting threshold is updated every once in a while.

[0047] In this embodiment, the complete operation flow of the intelligent decision-making module is as follows: Feature filtering: Redundant features were removed. Root layer humidity data were retained when the correlation between surface layer and root layer humidity was higher than 0.95, and the dimensionality was compressed to within 25 dimensions. The core decision tree algorithm model inference includes: The underlying model uses a random forest regressor to predict the basic fertilizer application rate, with a determination coefficient R² greater than 0.85. Mid-level optimization: Solving multi-objective optimization problems using an improved NSGA-III algorithm, with optimization objectives including maximizing yield, minimizing fertilizer cost, and minimizing environmental risk. Upper-level adjustments: Rule-based adjustments are made based on farmers' historical acceptance levels, such as the probability of rejecting high-nitrogen programs; Solution generation: Outputs a three-dimensional decision result containing the following elements: Spatial dimension: Differentiated fertilizer dosage is generated using a 10×10 meter grid through Kriging interpolation; Time dimension: The best fertilization window is within the next 7 days, avoiding periods with a rainy day probability greater than 60%; Resource dimension: Recommend fertilizer type combinations and application methods, such as the ratio of organic fertilizer to compound fertilizer and drip irrigation or broadcasting; Execution feedback: The actual fertilization data of farmers is transmitted back via a mobile app, such as the adjusted fertilization amount, for parameter iteration of the core decision tree algorithm model. The decision tree splitting threshold is updated every two weeks.

[0048] Through the above operations, this embodiment has been verified to increase fertilizer utilization to 75% in practical application, reducing over-fertilization by 20% compared to traditional methods.

[0049] The user interaction module displays information including soil monitoring data, network signal strength, signal transmission method, remaining battery power, power supply voltage, cross-validated soil monitoring data, sensor malfunction status, and recommended maintenance plans. For example, the user interaction module's user interface features an augmented reality mobile application, allowing farmers to access real-time soil data by scanning field signs with their mobile phone cameras. Combined with voice control, it enables convenient queries during field operations. The specially designed waterproof and dustproof terminal can withstand the harsh environmental conditions of agricultural settings.

[0050] The power supply module is used to supply power to the entire system. Specifically, the power generation unit of the power supply module is a solar tracking array composed of a curved monocrystalline silicon plate and a two-degree-of-freedom gimbal, which is controlled in conjunction with a light intensity sensor and astronomical algorithms. The energy storage unit of the power supply module is equipped with an intelligent charging and discharging router, which dynamically allocates the collaborative working mode of the lithium iron phosphate battery pack and the supercapacitor according to the power load prediction.

[0051] Example 2 This invention proposes a real-time monitoring method for multiple parameters of farmland soil based on the Internet of Things, including: The sensor module collects monitoring data of farmland soil through sensors including a high-precision capacitive humidity probe, a four-needle conductivity detection unit, and an anti-polarization pH electrode, and transmits the data to the sensor collaborative scheduling module. The sampling frequency controller submodule of the sensor collaborative scheduling module analyzes the trend of soil parameter changes over a historical period, dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model, and controls the sampling interval of the sensors. The energy consumption optimization scheduler submodule of the sensor collaborative scheduling module monitors the network signal strength and remaining power in real time, and adjusts the signal transmission method and the power supply voltage of the power supply module. The multi-sensor cross-validation submodule of the sensor collaborative scheduling module constructs a physical correlation model between soil parameters and a data fusion algorithm to perform multiple verifications of soil monitoring data and self-diagnose sensor faults, obtaining cross-validated soil monitoring data. The FPGA signal receiving submodule of the sensor collaborative scheduling module is used to identify different sensor data formats, automatically complete timestamp synchronization, unit unification, and data packet reassembly, and obtain fused multi-source heterogeneous data. The data transmission module acquires total monitoring data, including soil monitoring data from the sensor module, soil monitoring data cross-validated by the sensor collaborative scheduling module, and fused multi-source heterogeneous data, and transmits it to the data processing and storage module. The data processing and storage module preprocesses the total monitoring data and then injects it into the intelligent decision-making module; The intelligent decision-making module integrates pre-processed total monitoring data with plot-level soil testing data, meteorological satellite information, and historical yield records. Through deep learning algorithms, core decision tree algorithms, nutrient balance methods, and crop fertilizer requirements, it generates maintenance recommendations that include the timing, type, and dosage of fertilization. The user interaction module displays information including soil monitoring data, network signal strength, signal transmission method, remaining power, power supply voltage, cross-validated soil monitoring data, sensor fault status, and recommended maintenance solutions.

[0052] The specific implementation of the method in this embodiment is based on the system implementation of Embodiment 1, and will not be described again here.

[0053] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as object-oriented programming languages ​​like Java, C++, Python, and interpreted scripting languages ​​like JavaScript.

[0054] This application is described with reference to flowchart illustrations and / or block diagrams of methods, electronic devices (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing electronic device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing electronic device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0055] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing electronic device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0056] These computer program instructions can also be loaded onto a computer or other programmable data processing electronic device to cause a series of operational steps to be performed on the computer or other programmable electronic device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable electronic device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0057] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0058] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things, characterized in that: It includes a sensor module, a sensor collaborative scheduling module, a data transmission module, a data processing and storage module, an intelligent decision-making module, a user interaction module, and a power supply module; The sensor module includes multiple sensors arranged topologically in farmland soil to acquire soil monitoring data. The sensors include a high-precision capacitive humidity probe, a four-needle conductivity detection unit, and an anti-polarization pH electrode. The sensors are directly connected to the sensor coordination and scheduling module. The sensor collaborative scheduling module includes a sampling frequency controller submodule, an energy consumption optimization scheduler submodule, a multi-sensor cross-validation submodule, and an FPGA signal receiving submodule; The sampling frequency controller submodule analyzes the historical trend of soil parameter changes over a period of time, dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model, and controls the sampling interval of the sensors. The energy consumption optimization scheduler submodule monitors the network signal strength and remaining power in real time, and adjusts the signal transmission method and the power supply voltage of the power supply module. The multi-sensor cross-validation submodule constructs a physical correlation model between soil parameters and a data fusion algorithm to perform multiple verifications on soil monitoring data and self-diagnose sensor faults, obtaining cross-validated soil monitoring data. The FPGA signal receiving submodule is used to identify different sensor data formats, automatically complete timestamp synchronization, unit unification, and data packet reassembly, and obtain fused multi-source heterogeneous data. The data transmission module is used to acquire total monitoring data, including soil monitoring data from the sensor module, soil monitoring data after cross-validation by the sensor collaborative scheduling module, and fused multi-source heterogeneous data, and transmit it to the data processing and storage module. The data processing and storage module is used to preprocess the total monitoring data and then inject it into the intelligent decision-making module; The intelligent decision-making module is used to integrate pre-processed total monitoring data with plot-level soil testing data, meteorological satellite information and historical yield records. Through deep learning algorithms, core decision tree algorithms, nutrient balance methods and crop fertilizer requirements, it generates maintenance recommendation plans that include fertilization timing, type and dosage. The user interaction module is used to display content including soil monitoring data, network signal strength, signal transmission method, remaining power, power supply voltage, cross-validated soil monitoring data, sensor fault status, and maintenance recommendation plan. The power supply module is used to supply power to the entire system.

2. The real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things as described in claim 1, characterized in that: All sensors are equipped with standardized magnetic interfaces and waterproof connectors. The sensor module is externally protected by an IP68 protective shell. The sensor module integrates a self-cleaning structure, which prevents salt crystallization on the electrode surface by periodically releasing microcurrent pulses. The topological arrangement of the sensors in the soil enables multiple sensors to form a complementary detection network within a certain radius area. The topological arrangement is a honeycomb distributed topological structure, with regular hexagonal grid nodes arranged within a certain radius area. Each grid node is equipped with a sensor, and each grid node has a built-in telescopic probe that covers multiple depths of vertical gradient detection covering the surface, root system, and subsurface. Dual-grid nodes are deployed in key areas of irrigation zones and fertility gradient zones. The dual-grid nodes are a main grid node and a backup grid node. The backup grid node automatically switches to working mode when the main grid node fails.

3. The real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things as described in claim 1, characterized in that: The sampling frequency controller submodule of the sensor collaborative scheduling module includes an embedded prediction engine unit, a mutation response unit, and a regional coordination unit. The embedded prediction engine analyzes the historical trend of soil parameter changes over a period of time and dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model. When the minute-level change rate of key parameters such as humidity or temperature exceeds a preset threshold, the mutation response unit is activated, switching to an emergency sampling mode and increasing the sampling frequency to several times that of the normal state. The regional coordination unit obtains real-time data from adjacent monitoring nodes through a LoRa self-organizing network and initiates a collaborative enhanced sampling strategy for areas with sudden changes in soil parameters. The lightweight recurrent neural network model employs an improved gated recurrent unit (GRU). This improved GRU compresses parameters compared to the traditional GRU by removing the reset gate bias term. It also performs dimensionality reduction on features: the input layer retains only three key features—humidity change rate Δθ / Δt, temperature fluctuation variance σ_T², and conductivity gradient ∇EC—and extracts temporal patterns using a sliding window, with the window size set according to actual needs. Furthermore, the improved GRU performs online learning using incremental gradient descent, fine-tuning the model parameters once every certain period using historical data. The collaborative enhancement sampling strategy This includes the following mechanisms: Event broadcast: Triggering a grid node to send a broadcast packet via LoRa to neighboring grid nodes within a certain radius, containing the parameter mutation type, location coordinates, and confidence level between 0 and 1; Decision tree response: The receiving grid node decides whether to initiate enhanced sampling based on three conditions: battery level is higher than M%, current sampling period is greater than N minutes, and distance to the event point is less than P meters. If all conditions are met: the sampling frequency is immediately increased to tens of times the original and maintained for a period of time; if some conditions are met: the sampling frequency is increased to several times the original and maintained for a period of time; if not met: the original period is maintained but the event log is recorded. The values ​​of M, N, and P are set according to actual needs. Data aggregation: After ≥X grid nodes in the region confirm the same mutation, an "regional anomaly" flag is automatically sent to the gateway, triggering the cloud data backup mechanism. The value of X is set according to actual needs, and X>1.

4. The real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things according to claim 3, characterized in that: The energy consumption optimization scheduler submodule of the sensor collaborative scheduling module includes a communication mode decision unit, a dynamic power supply adjustment unit, and a photovoltaic prediction unit. The communication decision unit monitors the network signal strength and remaining power in real time. When the signal is stable and the power is sufficient, it prioritizes high-speed 4G transmission. When the signal weakens or the power is below the safety threshold, it automatically switches to low-power LoRa mode. The dynamic power supply adjustment unit intelligently switches the power supply voltage according to the sensor's working status. During normal monitoring, it uses a low voltage of 1.8V to maintain basic operation, and switches to a full power mode of 3.3V during high-frequency sampling. The photovoltaic prediction unit dynamically adjusts the equipment's working cycle based on the actual sunlight conditions in the farmland, and matches energy supply and consumption through a photovoltaic prediction algorithm; The photovoltaic prediction algorithm adopts a two-stage hybrid prediction model, specifically including: Short-term forecast: forecast time ≤ 1 hour, based on real-time sampling data from a light sensor, sampling time t The future light intensity is predicted using an exponentially weighted moving average algorithm, minute by minute, with the following formula: ； Where α is the weighting coefficient, which is set according to actual needs; For future light intensity, Given the current light intensity, Historical light intensity; Medium-term forecast: 1 hour < forecast time ≤ 24 hours, integrating cloud cover data from meteorological satellites, using a lightweight BP neural network. The lightweight BP neural network input layer: cloud cover, longitude, time period; the lightweight BP neural network output layer: light intensity level, updating the forecast curve daily to predict future light intensity. The device dynamically adjusts its operating mode to match energy supply and consumption based on predicted future light intensity: when there is sufficient light, the device uses full-speed sampling + data transmission + battery charging; when the light is moderate, it maintains full-speed sampling but reduces wireless transmission power; when the light is insufficient, it initiates sleep scheduling, only keeping the humidity sensor awake once every once in a while.

5. The real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things according to claim 1, characterized in that: The multi-sensor cross-validation submodule of the sensor collaborative scheduling module includes: a physical model validation layer that establishes nonlinear constraint relationships between parameters based on the chemical kinetic coupling equation of soil moisture-conductivity-pH value, triggering collaborative validation when single sensor data exceeds the theoretical range; a redundant data fusion layer that dynamically weights and fuses multiple sensors of the same type deployed within a certain radius using an improved Kalman filter algorithm, calculating sensor confidence weights in real time; and a fault diagnosis layer that integrates an electrochemical impedance spectroscopy analysis unit, constructing a Nyquist spectrum to detect electrode polarization state by applying a 10Hz-100kHz scanning signal to the sensor electrodes, predicting sensor lifespan by combining an Arrhenius aging model, and generating a maintenance warning when the characteristic frequency phase angle shift exceeds 15°.

6. The real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things according to claim 5, characterized in that: The chemical kinetic coupling equation for soil moisture-electrical conductivity-pH value is as follows: ； Where α represents the soil type correction coefficient; θ represents the volumetric water content; T represents the soil temperature; β represents the ion migration compensation index; NPK represents the total nitrogen, phosphorus, and potassium nutrient content; and γ represents the nutrient coupling coefficient. The formula for the improved Kalman filter algorithm is as follows: ； Where Q0 represents the basic process noise covariance; σ err σ represents the standard deviation of the sensor's recent error. base Indicates the factory calibration error reference value; λ t η represents the noise sensitivity coefficient; η represents the power supply stability weighting factor; ∇P supply This indicates the change in the supply voltage gradient.

7. The real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things according to claim 1, characterized in that: The FPGA signal receiving submodule of the sensor collaborative scheduling module is internally equipped with a built-in high-speed digital interface, a signal preprocessing unit, and an adaptive protocol parsing engine. The hardware layer is equipped with an eight-channel synchronous receiving circuit. The signal preprocessing unit is equipped with a real-time verification algorithm, which verifies data integrity through CRC16 error correction code and Manchester encoding, and initiates a triple redundancy retransmission mechanism for abnormal signals. The protocol parsing engine dynamically identifies different sensor data formats, automatically completes timestamp synchronization, unit unification, and data packet reassembly, and obtains fused multi-source heterogeneous data. The FPGA signal receiving submodule is also equipped with an anti-interference shielding structure, a temperature compensation circuit, and an intelligent buffer manager.

8. The real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things according to claim 1, characterized in that: The intelligent decision-making module integrates preprocessed total monitoring data with plot-level soil testing data, meteorological satellite information, and historical yield records, and establishes a multi-dimensional feature space through deep learning algorithms; The multidimensional feature space serves as the input engine for the core decision tree algorithm, comprising 42 features across 5 categories: Soil physicochemical features (12 dimensions, including spatiotemporal gradients of humidity, EC, pH, and NPK content at each layer); Meteorological features (8 dimensions, including sliding window statistics of precipitation, accumulated temperature, sunshine duration, and wind speed); Crop physiological features (6 dimensions, including vegetation cover, leaf area index (LAI), and growth stage based on NDVI remote sensing); Historical decision features (9 dimensions, including the type, dosage, and effect feedback coefficient of the past three fertilizations); and Environmental disturbance features (7 dimensions, including soil compaction, organic matter content, and heavy metal background values). These features, after standardization, form a 128×128 feature matrix, providing structured input for the subsequent training of the core decision tree algorithm. The core decision tree algorithm combines the nutrient balance method with crop fertilizer requirements to generate maintenance recommendation plans that include the timing, type, and dosage of fertilization; the configuration visualization engine maps the spatial variation characteristics of soil parameters into heat maps, and, together with aerial imagery, locates areas of soil fertility deficiency. The operation process of the core decision tree algorithm includes: Feature filtering: Redundant features were removed. Root layer humidity data were retained when the correlation between surface layer and root layer humidity was higher than 0.95, and the dimensionality was compressed to within 25 dimensions. The core decision tree algorithm model inference includes: The underlying model uses a random forest regressor to predict the basic fertilizer application rate, with a determination coefficient R² greater than 0.

85. Mid-level optimization: Solve multi-objective optimization problems using the improved NSGA-III algorithm. The optimization objectives include maximizing yield, minimizing fertilizer cost, and minimizing environmental risk. Upper-level adjustments: Regular adjustments are made based on farmers' historical acceptance levels; Solution generation: Outputs a three-dimensional decision result containing the following elements; Spatial dimension: Differentiated fertilizer dosage is generated by kriging interpolation to create a grid with a certain area; Time dimension: The optimal fertilization window within the next T days, avoiding periods with a rainy day probability greater than Z%; where the values ​​of T and Z are set according to actual needs; Resource dimension: Recommended fertilizer type combinations and application methods; Execution feedback: The actual fertilization execution data is sent back to the core decision tree algorithm model for parameter iteration, and the decision tree splitting threshold is updated every once in a while.

9. The real-time monitoring system for multiple parameters of farmland soil based on the Internet of Things according to claim 1, characterized in that: The power generation unit of the power supply module is a solar tracking array composed of a curved monocrystalline silicon plate and a two-degree-of-freedom gimbal. It is controlled by a light intensity sensor and an astronomical algorithm. The energy storage unit of the power supply module is equipped with an intelligent charging and discharging router, which dynamically allocates the collaborative working mode of the lithium iron phosphate battery pack and the supercapacitor according to the power load prediction.

10. The method for real-time monitoring of multiple parameters of farmland soil based on the Internet of Things according to any one of claims 1-9, characterized in that, include: The sensor module collects monitoring data of farmland soil through sensors including a high-precision capacitive humidity probe, a four-needle conductivity detection unit, and an anti-polarization pH electrode, and transmits the data to the sensor collaborative scheduling module. The sampling frequency controller submodule of the sensor collaborative scheduling module analyzes the trend of soil parameter changes over a historical period, dynamically calculates the optimal sampling interval using a lightweight recurrent neural network model, and controls the sampling interval of the sensors. The energy consumption optimization scheduler submodule of the sensor collaborative scheduling module monitors the network signal strength and remaining power in real time, and adjusts the signal transmission method and the power supply voltage of the power supply module. The multi-sensor cross-validation submodule of the sensor collaborative scheduling module constructs a physical correlation model between soil parameters and a data fusion algorithm to perform multiple verifications of soil monitoring data and self-diagnose sensor faults, obtaining cross-validated soil monitoring data. The FPGA signal receiving submodule of the sensor collaborative scheduling module is used to identify different sensor data formats, automatically complete timestamp synchronization, unit unification, and data packet reassembly, and obtain fused multi-source heterogeneous data. The data transmission module acquires total monitoring data, including soil monitoring data from the sensor module, soil monitoring data cross-validated by the sensor collaborative scheduling module, and fused multi-source heterogeneous data, and transmits it to the data processing and storage module. The data processing and storage module preprocesses the total monitoring data and then injects it into the intelligent decision-making module; The intelligent decision-making module integrates preprocessed total monitoring data with plot-level soil testing data, meteorological satellite information, and historical yield records. Through deep learning algorithms, core decision tree algorithms, nutrient balance methods, and crop fertilizer requirements, it generates maintenance recommendations that include the timing, type, and dosage of fertilization. The user interaction module displays information including soil monitoring data, network signal strength, signal transmission method, remaining power, power supply voltage, cross-validated soil monitoring data, sensor fault status, and recommended maintenance solutions.

Images