A smart irrigation monitoring method and system

Abstract

This invention relates to the field of smart landscaping and precision irrigation technology, specifically disclosing an intelligent irrigation monitoring method and system. The method deploys a multi-source sensor array and uses Kalman filtering to fuse environmental data, constructing a three-dimensional state vector input reinforcement learning model to generate irrigation decisions; it combines gradient boosting decision trees to predict vegetation water demand, forming a tiered irrigation strategy and converting it into pump control commands; it collects soil moisture feedback data in real time, dynamically adjusting decision model parameters and filtering rules to achieve closed-loop optimization. This invention significantly improves irrigation accuracy and water resource utilization through multi-source data fusion and adaptive decision-making mechanisms, while enhancing the system's responsiveness to vegetation growth dynamics and environmental changes, exhibiting beneficial effects such as closed-loop optimization, dynamic decision adaptation, and controllable resource consumption.

Description

Technical Field

[0001] This invention relates to the field of smart gardening and precision irrigation technology, and in particular to an intelligent irrigation monitoring method and system. Background Technology

[0002] Precision irrigation is the core of efficient water resource utilization in modern smart landscaping. Soil moisture exhibits significant heterogeneity in spatial distribution and temporal evolution, influenced by the complex coupling of multiple physical factors such as meteorology, soil type, and crop root distribution. Traditional irrigation decisions mainly rely on fixed threshold triggers, empirical rules, or predictive control based on a single soil moisture model, making it difficult to dynamically adapt to this complex spatiotemporal variation. Furthermore, the transport of irrigation water in the soil strictly follows fluid dynamics laws (such as those described by the Navier-Stokes equations). Existing irrigation decision-making methods based on simple rules or non-physically constrained data often fail to effectively couple these physical constraints, leading to decisions that deviate from actual water and fertilizer transport patterns, resulting in uneven irrigation, deep seepage, or localized water shortages.

[0003] While existing intelligent irrigation systems attempt to incorporate data-driven methods (such as reinforcement learning) for decision optimization, they face challenges when dealing with highly nonlinear and multi-constrained systems. On the one hand, the constructed environmental state representation fails to fully integrate the coupling effects of multiple physical fields (water, heat, solutes, etc.), limiting the completeness and accuracy of the state description. On the other hand, the decision model lacks explicit embedding and constraint of the inherent Navier-Stokes physical laws governing the flow of irrigation water in soil pores, leading to learned strategies that may violate fundamental principles of fluid motion and thus lack reliability. Furthermore, the water thresholds used to trigger irrigation decisions are typically static or empirically set, unable to be adaptively and dynamically adjusted in a continuous manifold space based on real-time changes in soil structure, crop water requirements, and climatic conditions, resulting in a delayed response to spatiotemporal heterogeneity.

[0004] In terms of energy efficiency optimization, existing methods often focus on a single objective (such as water saving) or consider energy efficiency objectives and water response objectives separately. They lack a systematic mechanism for unified modeling and collaborative optimization of irrigation system dynamics (such as pipeline hydraulics and pump station energy consumption) and the water response dynamics of the soil-crop-atmosphere system, making it difficult to achieve optimal overall energy efficiency of the irrigation system while ensuring crop water demand. Summary of the Invention

[0005] This invention provides an intelligent irrigation monitoring method and system to address the problem of how to achieve adaptive response to the spatiotemporal heterogeneity of soil moisture and synergistic optimization of irrigation energy efficiency based on the construction of environmental state vectors coupled by multi-physics fields, reinforcement learning decision-making with Navier-Stokes constraints, dynamic threshold correction of Riemannian manifolds, and Hamiltonian optimal control mechanism.

[0006] To address the aforementioned technical problems, this invention provides an intelligent irrigation monitoring method, comprising:

[0007] An array of capacitive humidity sensors, thermocouple temperature sensors, and photoelectric light sensors is deployed. Multi-source data is time-aligned using GPS timing and a double-buffered queue. Kalman filtering is employed to fuse soil moisture, soil temperature, and light intensity data. The capacitive humidity sensor probe is inserted vertically into the soil profile, converting moisture content into a voltage signal output by detecting changes in dielectric constant. The thermocouple temperature sensor, employing a composite structure of thermocouples and thermistors, is deployed in the key temperature-varying layer from the soil surface to a depth of 40 cm, outputting a millivolt-level electrical signal based on a linear resistance-temperature relationship. The photoelectric light sensor, equipped with a photodiode and a spectral filtering module, is fixed one meter above the vegetation canopy, converting the solar radiation spectrum into standard lux or watts per square meter.

[0008] The filtered data is used to construct a three-dimensional environmental state vector, which is then input into a Q-learning decision model to generate a four-dimensional action value evaluation vector. Based on an ε-greedy strategy, dynamic irrigation thresholds and water pump control commands are output. The control commands are encapsulated into hexadecimal data frames via the Modbus-RTU protocol. The frame structure includes five parts: a start character, a device address code, a function code, a data area, and a CRC checksum.

[0009] The expressions for the output dynamic irrigation threshold and water pump control commands include: ;

[0010] in, This is the threshold boundary correction amount; For soil porosity tensor; The Christoffel notation describes the curvature of soil heterogeneity; This represents the spatial gradient of water content. For time step;

[0011] Optimal energy control of water pumps: ;

[0012] in, For the Hamiltonian function of the water pump system; For water flow rate; The coordinates of the water flow position; For water quality; Let gravitational potential energy function be used. For pipeline topology constraint functions; This is the constraint strength coefficient; The norm square operator;

[0013] By associating dynamic irrigation thresholds with historical soil moisture data, and using gradient boosting decision trees to predict vegetation water demand, irrigation scheduling strategies with time windows and graded water quantities are generated.

[0014] The water pump control commands and irrigation scheduling strategies are converted into TLV format commands. PWM control is used to start and stop the water pump in stages. Soil moisture feedback data is collected after the pump stops.

[0015] By comparing the deviation between soil moisture feedback data and the dynamic irrigation threshold calculation area, the weight coefficient of the Q-learning reward function is adjusted, and the four-dimensional state decision model is reconstructed.

[0016] By integrating vegetation health indicators with optimized decision model parameters, the Kalman filter rules are updated through covariance analysis, and a training dataset is generated and fed back to the sensor data fusion module.

[0017] Furthermore, constructing the three-dimensional environment state vector specifically includes:

[0018] Soil moisture data is mapped to a water content coefficient in the range of [0,1], soil temperature data is converted into a temperature variation factor, and light intensity data is derived into a light intensity index.

[0019] The three-dimensional environment state vector is used as the activation value of the input layer of the Q-learning decision model and is transmitted to the hidden layer with 128 ReLU activation function neurons through a fully connected neural network;

[0020] The output layer is configured with four action nodes: maintain the current irrigation status, increase irrigation intensity, decrease irrigation intensity, and emergency water replenishment, generating a four-dimensional action value evaluation vector.

[0021] Furthermore, the process of outputting dynamic irrigation thresholds and pump control commands based on the ε-greedy strategy includes:

[0022] The exploration threshold ε is initially set to 0.7 and decreases by a decay factor of 0.99.

[0023] When the random number is higher than ε, the action corresponding to the maximum value of the action value evaluation vector is selected; when it is lower than ε, an action is randomly selected.

[0024] The dynamic irrigation threshold calculation includes: lowering the lower limit of soil moisture by 3-5% when increasing irrigation intensity, increasing the upper limit by 2-4% when decreasing irrigation intensity, and setting a temporary threshold of 20% above the normal level when making emergency water replenishment.

[0025] Furthermore, the intelligent irrigation monitoring method further includes:

[0026] The execution frequency calculation module analyzes the number of irrigation triggers within 24 hours using a sliding time window. When the frequency per unit time exceeds the warning value, the interval between adjacent actions is extended by 10%-15%.

[0027] Output parameters include: dynamic threshold range of soil moisture content, response level parameters including normal / accelerated / inhibited levels, minimum interval between pump start and stop, and maximum number of operations per day.

[0028] Furthermore, the intelligent irrigation monitoring method further includes:

[0029] The structured threshold control group includes: a lower limit threshold field for soil moisture content, an upper limit threshold field, and a four-bit binary response level identifier; the response mode activation strategy includes: when the last two digits are 1, a 3-5 minute normal mode is activated; when the last two digits are 10, a 6-8 minute accelerated mode is activated; and when the last two digits are 11, a 1-2 minute suppression mode is activated.

[0030] Furthermore, the process of outputting dynamic irrigation thresholds and pump control commands based on the ε-greedy strategy also includes:

[0031] The generation of water pump start / stop control commands includes three levels of decision logic:

[0032] The first level of monitoring involves soil moisture remaining below the lower threshold for 60 consecutive seconds.

[0033] The minimum interval for the second-level test pump is greater than 30 minutes.

[0034] The third level checks that the number of daily operations has not reached the maximum value;

[0035] Once the conditions are met, a quadruple instruction containing the device address, activation duration, water flow intensity level, and emergency flag is generated.

[0036] Furthermore, the intelligent irrigation monitoring method further includes:

[0037] Control commands are encapsulated as Modbus-RTU protocol data frames, which include:

[0038] Start character, device address code, function code, data area, and CRC checksum;

[0039] The first byte of the data area stores the start duration parameter (an integer value in 0.1-second units), the lower 4 bits of the second byte record the water flow intensity level, and the higher 4 bits reserve the emergency flag.

[0040] Furthermore, the process of generating a four-dimensional action value evaluation vector from the input Q-learning decision model includes:

[0041] When the irrigation effect meets expectations, the corresponding action reward function is increased by 0.1 weight coefficient;

[0042] When over-irrigation or insufficient water replenishment occurs, the penalty function is reduced by 0.05 weight coefficients;

[0043] The fully connected weight matrix between the input layer and the hidden layer is updated using a temporal difference algorithm.

[0044] Furthermore, the intelligent irrigation monitoring method further includes:

[0045] When predicting vegetation water demand using gradient boosting decision trees, the time-series features of dynamic irrigation thresholds and historical soil moisture data are integrated to generate an irrigation scheduling strategy that includes time window parameters and graded water volume parameters. The time window parameters are accurate to 15-minute intervals, and the graded water volume parameters are divided into three flow levels according to vegetation type.

[0046] Furthermore, an intelligent irrigation monitoring system, applied to any of the methods described above, includes:

[0047] The multi-source sensor array module includes a capacitive humidity sensor, a thermocouple temperature sensor, and a photoelectric light sensor.

[0048] The reinforcement learning decision module is configured with a Q-learning decision model and an ε-greedy policy selector.

[0049] The water demand prediction module stores historical soil moisture data and runs a gradient boosting decision tree.

[0050] The hierarchical execution module includes a TLV instruction converter and a PWM water pump controller;

[0051] The parameter optimization module configures the region deviation calculator and the reward function weight adjuster.

[0052] The rule iteration module includes a covariance analysis engine and a Kalman filter rule generator.

[0053] The key innovations of this invention include:

[0054] (1) Construct a multi-physics field coupled environmental state vector generation mechanism, integrate the water and heat conduction, root suction and micro-meteorological fluctuation characteristics of the soil-atmosphere-plant system in three-dimensional space, and form a high-dimensional spatiotemporal heterogeneous dynamic characterization system.

[0055] (2) Design a reinforcement learning decision architecture based on the constraints of the Navier-Stokes equations, embed the fluid motion continuity equation into the gradient update process of the policy network, and realize the essential coupling between irrigation action generation and physical field motion law.

[0056] (3) Establish a Riemannian manifold dynamic threshold correction mechanism, and calibrate the spatiotemporal correlation threshold of soil moisture sensing data in real time through manifold curvature analysis to solve the threshold drift problem caused by traditional Euclidean spatial measurement.

[0057] The following are its main beneficial effects:

[0058] (1) This invention breaks through the dependence of traditional irrigation models on static environmental parameters. By integrating multi-physics field dynamic coupling data of atmospheric turbulence, soil water potential gradient and plant transpiration, an environmental state vector with spatiotemporal evolution characteristics is constructed, which significantly improves the perception accuracy of water transport characteristics in complex farmland environments. Especially in areas with strong heterogeneity such as slopes and sandy loam, it has a stronger adaptability to working conditions than traditional homogenization models.

[0059] (2) The proposed reinforcement learning decision architecture transforms the Navier-Stokes fluid dynamics constraints into physical regularization terms of the policy network, enabling the irrigation strategy to automatically satisfy the principles of mass conservation and momentum transfer, fundamentally avoiding the physical law conflict problem common in traditional artificial intelligence decision-making, and greatly improving the executability and stability of irrigation actions in real fluid environments.

[0060] (3) The threshold correction mechanism based on Riemannian manifold establishes a differential geometric correlation model of sensor data to realize curvature-driven dynamic calibration of soil moisture anomalies, effectively suppressing false irrigation behavior caused by sensor drift and spatial interpolation error, and reducing the risk of redundant water consumption in the system. Attached Figure Description

[0061] Figure 1 A flowchart illustrating an intelligent irrigation monitoring method provided in an embodiment of this application;

[0062] Figure 2 This is a structural block diagram of an intelligent irrigation monitoring system provided in an embodiment of this application. Detailed Implementation

[0063] Example 1: Refer to Figure 1 This is a flowchart illustrating an intelligent irrigation monitoring method provided in an embodiment of the present invention. The process may include at least steps S100-S600:

[0064] The S100 system deploys an array of capacitive humidity sensors, thermocouple temperature sensors, and photoelectric light sensors. It uses GPS timing and a double-buffered queue to align the timing of multi-source data and employs Kalman filtering to fuse soil moisture data, soil temperature data, and light intensity data.

[0065] S200: Construct a three-dimensional environmental state vector from the filtered data, input it into the Q-learning decision model to generate a four-dimensional action value evaluation vector, and output dynamic irrigation thresholds and water pump control commands based on the ε-greedy strategy.

[0066] S300, along with dynamic irrigation thresholds and historical soil moisture data, predicts vegetation water demand through gradient boosting decision trees, and generates irrigation scheduling strategies with time windows and graded water quantities.

[0067] The S400 converts water pump control commands and irrigation scheduling strategies into TLV format commands, uses PWM to control the water pump to start and stop in stages, and collects soil moisture feedback data after shutdown.

[0068] S500, by comparing the deviation between soil moisture feedback data and dynamic irrigation threshold calculation area, adjust the weight coefficient of Q-learning reward function, and reconstruct the four-dimensional state decision model.

[0069] The S600 integrates vegetation health indicators with optimized decision model parameters, updates the Kalman filter rules through covariance analysis, and generates a training dataset that is fed back to the sensor data fusion module.

[0070] Step S100 includes at least steps S110-S130:

[0071] S110. Acquire raw monitoring data from soil moisture sensor, temperature sensor and light sensor, and perform data synchronization and alignment.

[0072] In the construction of smart gardens, a distributed multi-source sensor array is deployed within the target vegetation area. This array consists of three types of IoT sensors: a soil moisture sensor using capacitive or frequency domain reflectance technology, with its probe inserted vertically into the soil profile in layers, converting moisture content into a voltage signal output by detecting changes in dielectric constant; a temperature sensor using a thermocouple and thermistor composite structure, deployed in the key temperature-varying layer from the soil surface to a depth of 40 centimeters, outputting a millivolt-level electrical signal through a linear resistance-temperature relationship; and a light sensor equipped with a photodiode and spectral filtering module, fixed one meter above the vegetation canopy, converting the solar radiation spectrum into standard lux or watts per square meter. All sensor nodes establish a star topology connection with the central data acquisition unit via the LoRa wireless networking protocol, triggering a full-network polling acquisition cycle every five minutes.

[0073] During data acquisition, the soil moisture sensor outputs an analog voltage signal representing water content from 0% to 100%, the temperature sensor outputs an impedance value corresponding to the Celsius temperature scale, and the light sensor outputs a quantized digital value of light intensity. Due to electromagnetic interference, multipath effects, and node power consumption fluctuations, the raw data packets experience millisecond-level timing drift and data packet loss during transmission. To address this issue, the central data acquisition unit integrates a GPS timing module, attaching a Coordinated Universal Time (UTC) microsecond-level timestamp to each arriving data packet, and periodically calibrating the internal clock source via the Network Time Protocol (NTP). The timing alignment process employs a dual-buffered queue architecture: raw data packets are first stored in the input buffer. When data packets from all three sensors in the same acquisition cycle arrive (with a time window tolerance set to ±2 seconds), the timestamp comparison engine is activated, discarding timed-out data packets and triggering a sensor retransmission mechanism; valid data packets are rearranged according to their time sequence and transferred to the output buffer. Simultaneously, cyclic redundancy checks are performed, generating retransmission instructions for data packets that fail the check. The final output structured data frame contains the original values ​​of soil moisture, soil temperature, and light intensity after synchronization. This data frame is written into the cache as the input source for step S120 to ensure spatiotemporal consistency and meet the requirements of multi-source fusion.

[0074] S120. Perform Kalman filtering fusion processing on the original monitoring data to reduce environmental noise interference.

[0075] After receiving the synchronization data frame from step S110, the Kalman filter initializes the three-parameter state space: the soil moisture state variable is associated with the physical model of water diffusion, the soil temperature state variable is constructed based on the heat conduction equation, and the light intensity state variable is coupled with the atmospheric attenuation coefficient. The state transition matrix is ​​derived based on the law of thermal inertia and the principle of water conservation. The predicted soil moisture value is corrected by adding the temperature gradient change to the previous humidity value, and the predicted temperature value calculates the energy transfer trend through the thermal capacity-thermal resistance model. The measurement noise covariance matrix is ​​dynamically configured according to the sensor calibration accuracy: the noise variance of the soil moisture sensor is set to 0.5%, the noise variance of the temperature sensor is limited to 0.2 degrees Celsius, and the noise variance of the light intensity sensor is controlled to the order of 5%.

[0076] The processing is implemented through a prediction-update dual loop. In the prediction phase, the state transition equation is invoked to generate prior estimates of soil parameters at the current moment, and the error covariance matrix is ​​calculated simultaneously to quantify prediction uncertainty. In the update phase, observed values ​​from the S110 input are introduced for correction. The core mechanism is dynamic weight allocation using Kalman gain: when the deviation between the observed and predicted values ​​exceeds the sensor noise threshold, the weight of the observed data is automatically increased to suppress interference such as temperature drift and circuit thermal noise; confidence-weighted fusion is implemented based on the characteristics of multi-source data, for example, capacitive humidity sensors are given higher weight coefficients than thermistor temperature sensors. The multi-sensor data association algorithm uses Mahalanobis distance to measure correlation and performs covariance cross-validation on soil moisture and temperature observations to eliminate data contradictions caused by uneven local solar illumination. Noise suppression employs a fifth-order iterative smoothing filter to effectively filter out transient interference such as light intensity jumps caused by cloud cover. The final output optimized data array includes three fusion parameters: the soil moisture fusion value eliminates the influence of dielectric constant fluctuations from capacitive sensors, the soil temperature fusion value integrates surface and deep temperature difference information, and the light intensity fusion value retains effective spectral characteristics. The array is stored in a shared memory area for use in step S130, improving measurement accuracy by more than 30%.

[0077] S130 outputs soil moisture data, soil temperature data, and light intensity data after noise filtering.

[0078] After inputting the fused data array generated by S120, the central processing unit executes a standardized output process. Soil moisture data is linearly scaled to map the fused values ​​to the zero-to-100% standard range, while embedding data quality tags including GPS timestamps and sensor node numbers. Soil temperature data is converted to a Celsius scale and employs a dual-threshold filtering mechanism: values ​​exceeding the effective range of -10°C to +50°C are marked as invalid data points, and cubic spline interpolation compensation is performed between adjacent data points. Light intensity data is uniformly converted to lux units, and a low-pass filter is applied to eliminate residual high-frequency noise up to 5%.

[0079] All parameters are encapsulated into structured data packets by a format conversion engine, using a JSON key-value pair storage scheme: soil moisture data is written to the "moisture" field with one decimal place precision, temperature data is written to the "temperature" field with a signed identifier, and light intensity data is written to the "illuminance" field in integer format. The data packets are transmitted to non-volatile memory via a serial peripheral interface bus, employing a circular storage strategy to retain data from the most recent 24 hours, ensuring zero data loss during system power outages. The output readiness state is activated through an event-triggered mechanism: after the data packet completes digital signature verification, the system sets the "data ready" flag and releases the interrupt signal, allowing the reinforcement learning decision module in subsequent S200 steps to obtain the data packet in real time via the application programming interface. The final output dataset contains three key parameters: noise-filtered soil moisture data represents the baseline value for irrigable water volume, soil temperature data reflects the thermodynamic state of root activity, and light intensity data quantifies the photosynthetic effective radiation energy. This data packet serves as the core input to the closed-loop irrigation decision chain, directly driving the Q-learning strategy optimization engine in the S210 sub-step.

[0080] Step S200 includes at least steps S210-S230:

[0081] S210. Input the soil moisture data, soil temperature data, and light intensity data into the Q-learning decision model.

[0082] The system receives Kalman-filtered soil moisture, soil temperature, and light intensity data from step S130. Soil moisture is quantified as a percentage volumetric water content, soil temperature is recorded in degrees Celsius, and light intensity is collected in lux. The three types of data are normalized: the original soil moisture value is mapped to a water content coefficient in the [0,1] interval; the original soil temperature value is converted to a temperature variation factor; and the original light intensity value is derived as a light intensity index, collectively constructing a three-dimensional environmental state vector. This environmental state vector serves as the activation value of the input layer of the Q-learning decision model and is transmitted to the hidden layer via a fully connected neural network. The hidden layer has 128 neurons, each using the ReLU activation function to perform a nonlinear transformation on the input value. A fully connected weight matrix is ​​established between the input layer and the hidden layer. This matrix is ​​assigned values ​​using a random normal distribution during model initialization and updated using a temporal difference algorithm after each decision iteration. The output layer sets four sets of action nodes: maintain current irrigation state command node, increase irrigation intensity command node, decrease irrigation intensity command node, and emergency water replenishment command node. The output value of each action node represents the estimated cumulative reward for performing the corresponding action under a specific environmental state. After forward propagation from the input layer to the hidden layer, the environmental state vector generates a four-dimensional action value evaluation vector in the output layer. This vector contains the utility score for each irrigation action. The state data set output in this step fully encodes the current environmental characteristics and generates the action value evaluation vector as the basis for irrigation decisions.

[0083] S220. Irrigation threshold prediction and execution frequency calculation are performed using the Q-learning decision model.

[0084] Based on the four-dimensional action value evaluation vector generated by S210, an ε-greedy strategy is used to select the optimal irrigation action. When the value generated by the random number generator is higher than the exploration threshold ε, the action node corresponding to the maximum value in the action value evaluation vector is selected as the execution instruction; when the random value is lower than the exploration threshold ε, the execution instruction is randomly selected from four types of action nodes. The initial value of the exploration threshold ε is set to 0.7, and it decreases with the number of decision iterations by a decay coefficient of 0.99. Dynamic irrigation thresholds are calculated according to the selected action type: when the maintenance state instruction is selected, the threshold parameters of the previous cycle are inherited; when the increase irrigation intensity instruction is selected, the lower limit of the soil moisture threshold is reduced by 3 to 5 percentage points; when the decrease irrigation intensity instruction is selected, the upper limit of the soil moisture threshold is increased by 2 to 4 percentage points; when the emergency water replenishment instruction is selected, a temporary irrigation threshold exceeding the normal range by 20% is set. The execution frequency calculation module receives the action selection results and historical operation records, and analyzes the number of irrigation action triggers in the last 24 hours through a sliding time window. If the trigger frequency per unit time exceeds the preset warning value, the execution interval between adjacent actions is automatically extended by 10% to 15%. The irrigation threshold prediction outputs two types of parameters: a dynamic threshold range for soil moisture content consisting of a lower and upper critical value, and irrigation response level parameters including three levels: normal response mode, accelerated response mode, and suppressed response mode. The execution frequency calculation generates parameters for the minimum interval between pump start-up and shutdown and the maximum number of operations per day. The Q-learning model weight matrix is ​​updated based on the actual irrigation effect: when the irrigation effect meets expectations, the reward function for the corresponding action increases by 0.1 weight coefficient; when over-irrigation or insufficient water replenishment occurs, the penalty function decreases by 0.05 weight coefficient. The moisture content fluctuation range and response level encoding output in this step constitute the dynamic irrigation threshold, while the time interval threshold and frequency threshold form the execution frequency control parameters.

[0085] S230 generates dynamic irrigation threshold parameters and pump start / stop control commands.

[0086] The dynamic threshold range and response level parameters output by S220 are integrated to construct a structured threshold control group. This control group contains three core fields: a lower limit threshold field recording soil moisture content in the range of 5% to 25%, an upper limit threshold field recording soil moisture content in the range of 15% to 40%, and a response level identifier field storing a four-bit binary code. The corresponding control strategy is activated based on the identifier: when the last bit is 1, a normal response mode is activated, setting the pump's single start-up duration to 3 to 5 minutes; when the last two bits are 10, an accelerated response mode is activated, extending the start-up duration to 6 to 8 minutes; when the last two bits are 11, a suppressed response mode is activated, compressing the start-up duration to 1 to 2 minutes. The pump start-stop control command generation module receives the threshold control group and execution frequency parameters, and executes a three-level decision logic through a state machine: the first level decision monitors real-time soil moisture data, triggering the start-up condition when the soil moisture content remains below the lower limit threshold for 60 seconds; the second level decision verifies the pump's minimum interval time parameter, ensuring that the interval between adjacent operations is greater than 30 minutes; the third level decision checks the daily operation count counter, inserting a forced delay when the maximum value is reached. After meeting the three-level conditions, a four-tuple control instruction is generated: Instruction element one specifies the physical address code of the water pump device; instruction element two defines the start-up duration (with a precision of 0.1 seconds); instruction element three sets the water flow intensity level from 1 to 3; and instruction element four marks the emergency operation flag. The control instruction is encapsulated into a hexadecimal data frame via the Modbus-RTU protocol. The frame structure includes five parts: a start character, device address code, function code, data area, and CRC checksum. The first byte of the data area stores the start-up duration parameter (converted to an integer value in 0.1-second units), the lower 4 bits of the second byte record the water flow intensity level, and the higher 4 bits reserve the emergency operation flag. The final output includes a structured parameter set containing a floating threshold range and a water pump control instruction data packet conforming to industrial control standards.

[0087] In another embodiment:

[0088] S210: Fluid dynamics modeling of environmental state vectors and Q-learning input reconstruction.

[0089] Receive Kalman filtered data from S130 output: Soil moisture Soil temperature Light intensity A three-dimensional environment state vector is established through multi-physics coupling and normalization processing:

[0090] ① Calculation of moisture content coefficient:

[0091] enter After linear normalization: ;

[0092] in, : Lower limit of available soil moisture content; : Upper limit of soil saturated moisture content; Normalized moisture content coefficient, dimensionless moisture content index (output to the state vector).

[0093] ② Temperature variation factor modeling:

[0094] enter and its spacetime differential: ;

[0095] in, : Intensity of dynamic temperature change (output to state vector); : Soil temperature change rate over time; : Spatial Laplace operator for soil temperature; Soil thermal diffusivity;

[0096] ③ Light intensity index conversion:

[0097] enter Corrected for canopy attenuation: ;

[0098] in, : Atmospheric top illumination baseline value; Average height of crop canopy; Logarithmic index of canopy transmittance (output to state vector);

[0099] Environment state vector construction: ;

[0100] in, : Environmental state vector; used as input to the Q-learning decision model to achieve multi-source data fusion and dimensional unification.

[0101] Neural network modeling of physical constraints:

[0102] ④ Navier-Stokes embedding state transitions: ;

[0103] in: Nabla operator; : Pressure gradient term; : Viscous force term; Water density (constraint weight initialization); Hydrodynamic viscosity (controlling gradient propagation); : Moisture infiltration velocity vector (mapped to neuron connection weights); : Soil water pressure field (corresponding bias term correction); : Gravitational acceleration vector;

[0104] This equation is added to the loss function as a physical regularization term to ensure that the decision conforms to the laws of fluid dynamics.

[0105] Improved activation function design:

[0106] The 128 neurons in the hidden layer employ compound activation: ;

[0107] in: It is an improved Gaussian decay activation function; The activation function for the rectified linear unit; : Weighted sum of neuron inputs; Gaussian attenuation coefficient (to suppress abnormal pulses);

[0108] ⑤ Q-value update mechanism for ecological constraints: ;

[0109] in: State-action value function; Learning rate (controls the magnitude of model updates); Future earnings discount factor; Vegetation health weight ( (corresponding to different actions) : Soil improvement benefit function after irrigation; Let be the environmental state vector at time t; The irrigation action selected for time t; This is the minimum value operator for the action space;

[0110] Output four-dimensional action value evaluation vector Up to S220.

[0111] Technical Relevance: During the S210 process:

[0112] Input: S130 , , Vegetation health weight After irrigation ;

[0113] ① Moisture content coefficient: Received Combined with preset / Output To the state vector;

[0114] ② Temperature variation factor: receiver and its spatiotemporal differential, combined Output To the state vector;

[0115] ③ Light intensity index: Received light , combined and Output To the state vector;

[0116] Environment state vector: fusion , , Constructing 3D vectors → Input Q-learning network;

[0117] ⑤Q value update: Receives S630's and benefits of soil improvement Output Vector → to S220 decision module;

[0118] Technical Relevance: The environmental state vector of S210 directly supports the decision input of S220, whereby... It participates in the threshold correction calculation of S220.

[0119] S220: Dynamic threshold prediction and execution frequency optimization based on Riemannian manifold.

[0120] Action selection strategy: ;

[0121] in: Initial exploration rate (decay coefficient of 0.99 per step); Uniformly distributed random numbers; Decision action (input from S210) vector);

[0122] ⑥ Correction for non-uniform soil moisture threshold:

[0123] Input S210 and soil structure parameters: ;

[0124] in: Soil porosity tensor ( (corresponding to three-dimensional coordinates) Christoffel notation (describes soil heterogeneity curvature); Spatial gradient of moisture content; Threshold boundary correction amount; For time step;

[0125] Dynamic threshold update rules:

[0126]

[0127] ⑦ Random optimization of execution frequency:

[0128] Enter the number of historical operations and time series: ;

[0129] in: : Secondary potential energy function ( (operating frequency); : Diffusion coefficient (controls the intensity of random fluctuations); Wiener process increment (simulating environmental disturbance);

[0130] Within 24 hours (Preset) Adjustment triggered when: ;

[0131] in: The optimized new operation time interval; This is the current operation time interval; This is the threshold for the number of critical operations.

[0132] Output parameters to S230:

[0133] Dynamic threshold range; This is the lower limit of the dynamic threshold. : Dynamic threshold upper limit;

[0134] : 2-bit response level (high-order bits represent action intensity, low-order bits represent urgency).

[0135] Minimum operation interval (in minutes);

[0136] Maximum number of operations per day (adaptive to seasonality);

[0137] Technical connection: During the S200 process, input: S210 Vector sum Distributed sensors Historical operation data; Action selection: based on S210 Vector, output decision action ⑥ Threshold correction: Receive S210 and soil structure parameters, output →Update / ⑦ Frequency optimization: Receive historical operation count Output and ;

[0138] Output parameters:

[0139] =[ , → To S230 control group;

[0140] Response level → S230 control group;

[0141] Minimum interval → to S230 state machine;

[0142] Daily limit → S230 state machine;

[0143] Technical Relevance: Dynamic Threshold Output of S220 and response level Hamilton control system for direct drive of S230.

[0144] S230: Command generation and protocol encapsulation for Hamiltonian control systems.

[0145] Input control group construction: ;

[0146] in: For the input control group of the Hamiltonian control system; From S220 , From S220 response level;

[0147] ⑧ Optimal control of water pump energy: ;

[0148] in: For the Hamiltonian function of the water pump system; Water flow rate ( ); : Coordinates of the water flow position; Water quality ( (pipe volume); Gravitational potential energy function; Pipeline topology constraint functions; Constraint strength coefficient; The norm square operator;

[0149] Response strategy implementation:

[0150]

[0151] Technical Relevance: During the S230 process:

[0152] Input: S220 and ;real time S220 / ;

[0153] Control group construction: =⟨ , , (All from S220);

[0154] ⑧ Pump control: based on Value determined Scope and energy efficiency targets

[0155] State machine verification:

[0156] Triggering conditions: < (From S220) );

[0157] Interval verification: > (From S220);

[0158] Count verification: < (From S220);

[0159] Instruction encapsulation:

[0160] ← Least significant bit (direct mapping);

[0161] according to The value range is determined;

[0162] Output =⟨ , , , >;

[0163] in: : Duration of operation (s); Flow rate level {1,2,3}; Device address (e.g., 0xD2F3); Emergency sign; : Control instruction quadruple;

[0164] Technology closed loop: The instructions generated by S230 actually change the soil condition, new Feedback is sent to S130 to begin the next round of decision-making.

[0165] S210 technology effect: Multiphysics field fusion: Incorporating water content ( ),temperature( ),illumination( Unified as a dimensionless state vector; Physical constraint embedding: Navier-Stokes equations ensure that decisions conform to fluid dynamics; Ecological optimization: Vegetation weights. Q-learning takes into account the health needs of crops;

[0166] S220 technology benefits: Dynamic threshold adaptation; Riemannian manifold correction. Addressing soil spatial heterogeneity; Intelligent frequency control: Langevin equation optimizes operating frequency (when >8 times in 24 hours). Increased by 34%); Response grading: 2 bits Value encoding indicates the intensity and urgency of an action;

[0167] S230 Technology Performance: Optimal Energy: Hamiltonian System Based on Different Value enables differentiated energy efficiency control; three-level fault tolerance mechanism: threshold / interval / number triple verification reduces false trigger rate to <0.1%; industrial-grade protocol: improved Modbus frame supports 0.1s precision control.

[0168] Step S300 includes at least steps S310-S330:

[0169] S310. Obtain historical soil moisture data and historical irrigation records during the vegetation growth cycle.

[0170] In the intelligent irrigation monitoring system, the historical data acquisition module first accesses the cloud database and retrieves the historical monitoring dataset covering the entire growth cycle of the specified vegetation. This dataset contains a historical soil moisture data matrix indexed by precise timestamps, with its dimension represented by the number of monitoring points multiplied by the number of historical time points. It also associates with a historical irrigation record table, which records the start timestamp, duration, pump power level, and actual water consumption parameters for each irrigation operation. The historical soil moisture data is directly derived from the noise-filtered soil moisture data output from step S130. This data is supported by a time-series database formed through a long-term storage mechanism. Its original acquisition frequency is fixed at one sample every fifteen minutes. Each data record explicitly includes a unique sensor number, corresponding geographic coordinates, a specific soil layer depth identifier, and a calibrated percentage of moisture content.

[0171] The system preprocesses the raw historical dataset by dividing it into time-window units. The preprocessing process divides the data into independent blocks based on natural calendar days and performs a rigorous data integrity verification process for each block. When the proportion of missing historical soil moisture data within a single calendar day exceeds a system-preset threshold of 5%, an automatic interpolation mechanism for adjacent blocks is activated. This mechanism uses a time-series autoregressive algorithm to reconstruct the missing data points. Simultaneously, an outlier cleaning module calculates the standard deviation within a sliding window with a fixed 24-hour window size and removes outlier data points that deviate from the window mean by more than or equal to three standard deviations. Finally, the system outputs a structured historical dataset, which is categorized and stored according to different physiological growth stages of vegetation, including germination, rapid growth, and maturity. Each category includes a complete time-series curve of soil moisture variation for the corresponding stage, along with strictly matched irrigation operation logs. The output interface of this structured historical dataset establishes a direct physical connection with the input port of the machine learning model in step S320, ensuring smooth data transmission.

[0172] S320. Based on machine learning regression analysis, predict vegetation water demand and associate it with the dynamic irrigation threshold parameter.

[0173] After receiving the structured historical dataset from step S310, the core component of the water demand prediction engine initiates a multivariate regression analysis model. During the model initialization phase, the feature engineering unit first performs the task of constructing the input feature vector. This task extracts key statistical features from historical soil moisture time-series data, specifically covering the average soil moisture over the last 72 hours, the rate of change of the soil moisture decline slope over the past 24 hours, and the maximum amplitude of soil moisture fluctuations during a complete natural day. Simultaneously, the feature engineering unit extracts important correlation features from the associated historical irrigation records, including the time interval between two adjacent independent irrigation events and the cumulative total water consumption of several irrigations prior to this irrigation.

[0174] The target variable generation unit simultaneously performs the task of calculating the theoretical water requirement of vegetation. This calculation is based on the potential evapotranspiration reference value derived from the Penman formula, combined with the specific crop coefficient corresponding to the target vegetation type, to generate an accurate benchmark value of the theoretical water requirement of vegetation on a daily basis. The regression analysis model uses a gradient boosting decision tree as its basic architecture, which contains three interconnected hidden node processing layers. The first processing unit receives the input feature vector generated by the feature engineering processing unit, and simultaneously receives the dynamic irrigation threshold parameter set output from step S230. This parameter set includes the soil moisture trigger threshold, the critical soil temperature threshold allowed for irrigation operation, and the light intensity limit threshold optimized by the reinforcement learning model. The second processing unit performs feature cross-fusion operation. Its core operation is to perform weighted fusion processing on the dynamic irrigation threshold parameters provided by step S230 and the previously extracted historical soil moisture decline slope. The weight factors used in the fusion process are dynamically determined by an adaptive learning algorithm during the model training phase, ultimately generating a composite feature factor representing the correlation between environmental state and historical trend. The third-layer processing unit is responsible for establishing the nonlinear mapping relationship between this composite feature factor and the target variable, namely the theoretical water demand baseline value of vegetation. By minimizing the mean square error function between the model's predicted value and the actual value, it iteratively optimizes key parameters such as the split node parameters, leaf node weight values, and tree structure complexity within the decision tree.

[0175] After the model calculation is completed, the strategy coordinator module associated with its output interface performs a collaborative verification function for the prediction results. This module performs a logical matching verification between the predicted water demand value output by the model and the dynamic irrigation threshold parameters generated in step S230. When the verification finds that the expected value of soil moisture correction calculated based on the predicted water demand is lower than the currently set soil moisture trigger threshold, the threshold recalibration mechanism is automatically activated. This mechanism gradually decreases the soil moisture trigger threshold with a preset step size of 0.5%, until the corrected threshold can meet the irrigation triggering conditions based on the predicted water demand. Finally, the system outputs a set of decision parameters containing the optimized theoretical daily water demand value for vegetation and the recalibrated soil moisture trigger threshold. This set of decision parameters is transmitted in its entirety to the irrigation scheduling strategy generator in step S330 as the core input via the system's internal data bus.

[0176] S330. Generate an irrigation scheduling strategy that includes irrigation time windows and water allocation schemes.

[0177] After receiving the decision parameter set from step S320, the irrigation scheduling strategy generator generates an executable irrigation scheduling scheme according to a specific logical sequence. First, the time window calculation unit determines the precise irrigation triggering time based on the optimized and recalibrated soil moisture trigger threshold. This unit monitors the current soil moisture data stream output from step S130 in real time. When the soil moisture value is continuously monitored to be lower than the recalibrated threshold, the irrigation operation window is immediately activated. Simultaneously, this unit incorporates real-time light intensity data output from step S130 for time constraint control, strictly limiting irrigation operations to only be initiated during periods when the light intensity is lower than the light intensity limit threshold set in step S230. These periods typically correspond to time windows with weaker light at night or in the early morning, thereby minimizing ineffective water evaporation loss caused by strong sunlight.

[0178] The water allocation calculation unit accurately calculates the amount of water to be allocated for a single irrigation operation based on the theoretical daily water requirement of vegetation predicted and output in step S320. The calculation process employs a hierarchical progressive algorithm, scientifically dividing the total water requirement into two parts: a baseline water volume, used to meet the minimum water needs for the vegetation to maintain basic survival functions; and a regulating water volume, used for fine-tuning the water volume according to real-time environmental changes. The baseline water volume is calculated based on the effective root absorption depth of the target vegetation, which is equivalently converted to millimeters of water depth for quantification. The regulating water volume is dynamically correlated with the current soil temperature output in step S130, showing that the regulating water volume increases proportionally with the increase in soil temperature. Specifically, for every one degree Celsius increase in soil temperature, the regulating water volume increases by 0.8% from the baseline water volume.

[0179] The parameter conversion unit is responsible for converting the water allocation scheme into control parameters that can directly drive the water pumps. This unit calls a pre-stored water pump flow characteristic model, which internally stores a mapping table of stable outlet flow rates for different power levels of the water pumps. Based on this mapping, the conversion unit intelligently decomposes the total water demand into multiple sequentially executable irrigation sub-tasks. For example, when the calculated total water demand is fifty liters, this unit may break it down into two independent twenty-five-liter irrigation sub-tasks. The specific execution time of each sub-task is automatically calculated based on the current power level of the water pump configuration, by querying the flow mapping table and using the execution time conversion formula.

[0180] The final generated irrigation scheduling strategy includes clearly defined structured data fields: an irrigation execution time window field accurate to the minute, defining specific start and stop times; a water allocation sequence field broken down by subtasks, clearly listing the planned water supply for each subtask; and a set of associated control parameters, including specified pump power levels, the maximum allowed continuous running time for a single subtask, and the maximum number of retries in case of equipment communication failure or execution failure. The entire irrigation scheduling strategy is encapsulated in a standardized JSON data format, and after encapsulation, it is written to a persistent strategy database for storage and management, synchronously triggering the irrigation control command issuance operation in step S410 to start the actual irrigation process.

[0181] Step S400 includes at least steps S410-S430:

[0182] S410, The pump start / stop control command and irrigation scheduling strategy are sent to the pump execution module.

[0183] The system's central control unit receives the pump start / stop control command from step S230 and the irrigation scheduling strategy from step S330. The pump start / stop control command includes three core parameters: a start threshold voltage signal (range 3.0-5.0V) generated based on a Q-learning decision model, a stop threshold voltage signal (range 4.2-5.5V), and a maximum continuous running duration parameter (in seconds). The irrigation scheduling strategy includes the start time of the time window (accurate to the minute), the window duration (in minutes), the target irrigation water allocation value (in liters), and the regional priority coefficient (range 0.5-1.0).

[0184] The data encapsulation module first encodes the voltage threshold parameter into binary with a precision of 0.1V, mapping it to the ADC input range (0-10V range) of the execution module. The time parameter conversion unit converts the start time in the scheduling strategy into a 32-bit Unix timestamp and the duration into a 16-bit integer in seconds. The water allocation value is converted into the corresponding PWM duty cycle value (range 15%-100%) using a preset water pump flow characteristic curve (stored in the central control unit's FLASH memory), and the priority coefficient is converted into a 4-bit binary code. The encapsulated data packet adopts a TLV (Type-Length-Value) structure: the type field identifies the parameter category (0xA1 represents a start / stop command, 0xB2 represents a scheduling strategy), the length field declares the data length, and the value field stores the specific parameter value.

[0185] The central control unit transmits the encapsulated data packet to the target water pump execution module via the RS-485 physical layer communication interface (19200bps baud rate, parity bit enabled). The execution module's built-in STM32F407 microcontroller receives the data packet via the UART interface and then activates the data parsing engine: verifying the CRC16 redundancy code to confirm data integrity, extracting parameter values ​​according to the TLV structure, storing the voltage threshold in the ADC threshold register, writing the time parameter to the RTC real-time clock module, loading the PWM duty cycle value into the timer compare register, and storing the priority coefficient in the EEPROM non-volatile memory. After configuration is complete, the microcontroller returns an ACK confirmation frame, forming a closed-loop communication verification.

[0186] S420: Control the water pump to perform tiered irrigation operations according to the scheduling strategy.

[0187] The microcontroller of the water pump execution module acquires three input signals in real time: the first channel obtains the original voltage (0-3.3V) of the local soil moisture sensor through a 12-bit ADC channel. This sensor is the same model as the sensor array deployed in step S110 (FDR frequency domain reflective type, measurement accuracy ±2%); the second channel reads the internal RTC clock (error ±1ppm); and the third channel monitors the water pump operating status feedback signal (overcurrent / overheat flag).

[0188] The control logic unit performs a triple condition check: when the soil moisture value converted by the ADC (linearly converted to volumetric water content) is lower than the start threshold in the start / stop instruction, the RTC time is within the time window defined by the irrigation scheduling strategy (allowing ±2 minutes tolerance), and no fault flag is activated, the irrigation operation is triggered. The microcontroller outputs a high-level signal to the optocoupler (model PC817) via GPIO, driving the MOSFET power transistor (model IRF540N) to conduct, supplying power to the water pump motor (DC 24V / 350W).

[0189] Upon startup, the system immediately enters the graded flow control phase: it reads the target PWM duty cycle corresponding to the current time window from non-volatile memory and outputs a PWM waveform with a frequency of 16kHz through the TIM1 timer channel. The duty cycle adjustment accuracy reaches 1%, corresponding to graded pump speeds (25% duty cycle = 1200rpm, 50% duty cycle = 2400rpm, 75% duty cycle = 3600rpm). Simultaneously, a triple stop monitoring system is activated: a built-in watchdog timer accumulates the running time and forcibly shuts down when the maximum continuous running threshold is reached; the ADC collects soil moisture values ​​every 10 seconds and terminates immediately if the stop threshold is exceeded; and the RTC automatically shuts down the pump when it detects that the time window range has been exceeded.

[0190] During operation, the microcontroller continuously records 10 types of status parameters: actual start time (RTC timestamp), stop time, real-time PWM duty cycle value, motor drive voltage (ADC acquisition), current (ACS712 sensor acquisition), cumulative power consumption (voltage and current integral calculation), fault code, soil moisture change curve (recorded once per minute), ambient temperature (DS18B20 acquisition), and vibration amplitude (ADXL345 triaxial accelerometer acquisition). The data is stored in an external FRAM memory (capacity 256KB) with a ring buffer structure.

[0191] S430: Real-time collection of soil moisture feedback data and water pump operation status data after irrigation.

[0192] After the water pump stops, the module starts a delayed trigger mechanism: the built-in timer begins a countdown (default 15 minutes, configurable range 5-30 minutes). When the countdown ends, the microcontroller sends a measurement command to the soil moisture sensor (model SEN0193) via the I²C bus. The sensor emits a 100MHz electromagnetic wave and receives the reflected signal. The internal ASIC chip calculates the dielectric constant and outputs a 0-3.3V analog signal. Simultaneously, it reads the water pump operating status data records stored in the FRAM.

[0193] Data preprocessing includes three key operations: temperature compensation (compensation coefficient -0.3% / ℃) for the raw soil moisture voltage values, 50Hz power frequency filtering for the water pump current values, and mean filtering for the vibration data. The preprocessed dataset is uploaded to the central control unit via a LoRa wireless module (433MHz band, 20dBm transmit power). The transmission frame includes the device ID, data acquisition timestamp, and a 16-byte checksum.

[0194] After receiving the data, the central control unit starts the Kalman filter fusion engine: processing the current soil moisture value using the same state-space model (state vector [soil moisture, temperature, light intensity]) as step S120. The filtering parameters are completely consistent with S120: process noise covariance Q=diag(0.02,0.01,0.05), observation noise covariance R=diag(0.002,0.005,0.03). During fusion, the previous filtering result is used as a priori estimate, and the optimized value is output through a prediction-update loop. The final soil moisture feedback data accuracy is improved to ±0.8%. The pump operation status data, aligned with time, is written to a MySQL database. The storage structure contains 12 fields: timestamp, equipment coordinates, soil moisture value, temperature value, light intensity value, pump start / stop status, PWM setting value, voltage, current, vibration amplitude, fault code, and data quality flag.

[0195] Step 500 includes at least steps S510-S530:

[0196] S510. Compare the soil moisture feedback data with the dynamic irrigation threshold parameters.

[0197] Upon receiving the soil moisture feedback data packet and water pump operating status dataset transmitted in step S430, the system activates the real-time difference comparison engine. The soil moisture feedback data is transmitted to the input buffer of the central processing module via the LoRaWAN communication interface. This data packet contains the volumetric water content value after Kalman filtering, the corresponding GPS timestamp, and the distributed sensor node number. Simultaneously, the system invokes the dynamic irrigation threshold parameter queue generated in step S230 and stored in non-volatile memory. This queue is sorted by time validity, and its storage structure includes the lower limit threshold, upper limit threshold, and their effective time interval identifiers for soil moisture content.

[0198] The difference comparison calculation unit performs three levels of core operations: The first level activates the time axis synchronization controller to precisely match the collection timestamp of the feedback data with the effective time interval of the dynamic irrigation threshold parameters. The matching mechanism uses a sliding window algorithm with a tolerance window radius of five minutes. When the timestamp of the feedback data falls within ±5 minutes of the effective time interval of a certain set of threshold parameters, that set of parameters is determined to be the current effective comparison benchmark. If no match is found, the threshold backtracking mechanism is activated, automatically selecting the most recent effective threshold parameter set as the comparison benchmark.

[0199] The second-level operation executes the numerical deviation calculation engine, performing a dual deviation analysis on the measured soil moisture values ​​for each sensor node. The absolute difference calculator outputs the arithmetic difference between the measured moisture content and the target threshold, accurate to one decimal place; the relative percentage deviation calculator uses the algorithm (measured value - target threshold) / target threshold × 100% to generate a standardized deviation coefficient. During the calculation process, when the measured value is below the lower threshold, it is marked as a negative deviation; when it is above the upper threshold, it is marked as a positive deviation, and the deviation amount is recorded.

[0200] The third level of operation implements regional anomaly detection, establishing a spatial correlation matrix based on the physical topology of distributed sensor nodes. The detection algorithm first calculates the average deviation of all nodes, and then analyzes the standard deviation of the deviations of the eight neighboring nodes centered on a single node. When the deviation of a node exceeds the average by more than two standard deviations, and more than 30% of its neighboring nodes show the same deviation, the area is determined to be an abnormal irrigation block. The coordinates of the abnormal area are encoded using the UTM coordinate system, recording the longitude and latitude offsets to six decimal places.

[0201] The final output structured variance report contains three core data categories: a threshold deviation matrix storing the monitoring point number, soil layer depth, absolute deviation value, and relative percentage deviation in a two-dimensional array; a trend deviation direction identifier using binary encoding, with a last bit of 0 indicating that the actual humidity change trend is inconsistent with the trend predicted in step S220, and a last bit of 1 indicating consistency; and an anomaly area coordinate set recording the coordinates of the center points of anomaly blocks identified through topological analysis and the anomaly intensity level. This report is written to a shared memory area for use by step S520.

[0202] S520. Adjust the reward function weights of the Q-learning decision model according to the degree of difference.

[0203] Based on the difference report data output from step S510, the reinforcement learning optimization module initiates the dynamic adjustment process of the reward function weights. The system first parses the data in the threshold deviation matrix and performs a deviation grading mapping operation: Absolute differences within 5% are defined as slight deviations, with a base penalty coefficient of 0.1; differences between 5% and 15% are considered moderate deviations, with a base penalty coefficient of 0.3; differences between 15% and 30% are considered severe deviations, with a base penalty coefficient of 0.6; and differences exceeding 30% are considered failure deviations, with a base penalty coefficient of 1.0. The deviation level of each monitoring point is mapped to the corresponding penalty base penalty through a lookup table.

[0204] Perform trend compensation correction: Read the trend deviation direction identifier. When the last digit is 1 (indicating that the predicted trend is consistent with the actual trend), apply a decay factor to the penalty coefficient. The decay amount is dynamically set according to the deviation level: 40% decay for slight deviation, 30% for moderate deviation, 20% for severe deviation, and no decay for failed deviation. The decay calculation uses multiplication. For example, the penalty coefficient correction for moderate deviation with consistent trend is 0.3 × (1 - 30%) = 0.21.

[0205] Perform environmental variable coupling correction: Utilize the current environmental parameters provided in step S610, including soil temperature (degrees Celsius) and light intensity (lux). Establish a light-temperature-humidity coupled correction model. When the temperature sensor return value exceeds 30 degrees Celsius and the light intensity exceeds 80,000 lux, apply a 1.5-fold weighting factor to the penalty coefficient for insufficient humidity (measured value below the threshold). The weighting calculation employs a conditional triggering mechanism, activating only when both temperature and light intensity simultaneously exceed the threshold.

[0206] Dynamic weight allocation is implemented: the corrected penalty coefficients are substituted into the three-dimensional weight matrix of the reward function. This matrix includes water-saving reward weights (initial value 0.5), vegetation health reward weights (initial value 0.3), and energy consumption control reward weights (initial value 0.2). The allocation process uses a gradient descent algorithm for twenty iterations of optimization. In each iteration, the policy loss function value under the current weight configuration is calculated, and the weight ratios are updated along the negative gradient direction. The optimization objective is to reduce the decision weight bias of significant factors. For example, when the deviation due to insufficient humidity is severe, the vegetation health reward weight is increased to 0.4, and the energy consumption control weight is decreased to 0.1. The adjusted weight matrix is ​​updated in real time to the reward rule base of the Q-learning decision model. The change record includes the old weight value, the new weight value, the change timestamp, and the environmental parameters that triggered the correction, forming a complete weight change log.

[0207] S530, Output the updated irrigation decision model parameters.

[0208] After adjusting the reward function weights, the system performs a decision model parameter reconstruction operation. First, the state space dimension is expanded: based on the new weight matrix, the state variables of the Q-learning decision model are expanded from the original three dimensions (soil moisture, temperature, and light) to four dimensions. The new dimension injects the vegetation growth health index obtained in step S610. This index, collected by a vegetation index sensor, includes three sub-parameters: normalized difference vegetation index, chlorophyll content index, and canopy temperature. These are then compressed into a single health state variable through principal component analysis.

[0209] An incremental training mechanism was then initiated: the most recent twelve sets of valid data records collected in step S430 were retrieved, including six sets of water pump operating status data (start-stop time, runtime, power level) and six sets of soil moisture feedback data (moisture content sequences at five-minute intervals). A mini-training set was constructed using this dataset to perform rapid iterative training of the Q-learning model. The training process employed a temporal difference algorithm for eighty iterations, focusing on optimizing decision paths associated with high deviation regions in the state transition probability table. The optimization strategy was as follows: when a state transition path passed through the coordinates of anomaly regions marked in S510, the learning rate for that path was increased to 1.5 times the normal value; when the decision action associated with the path did not trigger a deviation alarm in the historical records, the learning rate was reduced to 0.8 times the normal value.

[0210] After training, three sets of core output parameters are generated: a decision kernel parameter package encapsulating the updated Q-value table (storing action value evaluation values ​​in the four-dimensional state space), a state transition rule set (containing 3000 state-action transition probabilities), and an action selection strategy tree (a five-layer decision binary tree structure); the threshold control instruction template reconstructs the dynamic irrigation threshold generation logic, and associates the vegetation water demand prediction results of step S320 with the data bus. The template adds a correction coefficient for the soil moisture threshold based on the vegetation health index. For every 0.1 unit decrease in the health index, the lower limit threshold of soil moisture content increases by 2%. The model version identifier is generated using the SHA-256 algorithm. The input parameters include the number of historical updates, the fingerprint of the current environmental parameters (temperature, light, and average health index), and the hash value of the weight change log.

[0211] Output parameters are transmitted via an AES-256 encrypted channel, and the decision kernel parameter package is synchronized to the model loading interface in step S210; the threshold control instruction template is transmitted to the threshold generation module in step S230; the model version identifier is written to the multivariate analysis database index table in step S620, completing the entire closed loop.

[0212] Step S600 includes at least steps S610-S630:

[0213] S610: Obtain current light intensity data, soil temperature data, and vegetation growth health indicators.

[0214] The light intensity sensor deployed in the target irrigation area operates based on the photoelectric conversion principle. Its internal photodiode receives the solar radiation spectrum, filters out non-visible light bands through a spectral filtering module, and converts the effective light radiation energy into a 4-20mA standard current signal output. The signal strength is proportional to the actual light intensity value. Simultaneously, a temperature sensor array buried deep in the soil is activated. This array uses a platinum resistance temperature sensing element (PT1000). Based on the positive correlation between resistance and temperature, when the soil temperature changes, it causes an imbalance in the Wheatstone bridge, outputting a differential voltage signal which is converted into a digital temperature value by a 24-bit ADC. The above light intensity and soil temperature data directly reuse the noise-filtered data output from step S130, eliminating the need for repeated data acquisition.

[0215] The collection of vegetation growth health indicators was achieved through a dual-modal sensing system: the visible-near-infrared spectral analysis module used a multispectral imaging device (spectral range 400-900nm) to scan the vegetation canopy at a 30° tilt angle, generating the NDVI index (range -1 to 1) by calculating the normalized difference in reflectance between the 760nm and 680nm bands. This index quantifies chlorophyll absorption characteristics. The morphological feature extraction module used a 5-megapixel CMOS camera (resolution 2592×1944), identifying leaf contours based on the Canny edge detection algorithm, calculating the leaf curling index through radius of curvature analysis, and simultaneously extracting the saturation channel values ​​(0-255 levels) of the HSV color space. The two health indicators were fused into a single vegetation health parameter with a 7:3 weighting ratio.

[0216] All data is transmitted to the central processing unit via the LoRaWAN protocol. Each transmission frame contains a 16-byte payload (4 bytes for light intensity, 4 bytes for soil temperature, 2 bytes for NDVI value, 2 bytes for curvature index, 2 bytes for saturation, and 2 bytes for checksum). Upon reception, timestamp alignment is performed: linear interpolation compensation is applied to data packets with a delay exceeding 500ms, based on the GPS second pulse signal. The final result is a structured triplet dataset {light intensity (lux), soil temperature (°C), vegetation health}, stored in a DDR4 memory buffer as the input source for the S620.

[0217] S620. Perform multivariate collaborative analysis by associating the updated irrigation decision model parameters.

[0218] Receive the irrigation decision model parameter set output from step S530. This parameter set contains three core elements: the updated Q-learning reward function weight matrix (a three-dimensional vector [w1, w2, w3] where w1 + w2 + w3 = 1), a state-action mapping table (a 256×4 matrix storing action values), and an action selection probability distribution (a four-dimensional vector [p1, p2, p3, p4] representing the probabilities of four irrigation actions). Couple the triplet dataset generated in step S610 with the model parameters for analysis.

[0219] A four-dimensional collaborative analysis space is constructed: An environmental variable space E = {light intensity, soil temperature, vegetation health} is established using triplet data as the basis vector. An action selection probability distribution vector P = [p1, p2, p3, p4] is superimposed as the decision dimension, forming a joint analysis domain [E, P]. The covariance matrix is ​​calculated using principal component analysis; a strong correlation is identified when the calculated value exceeds 0.5. Key coupling coefficients are identified: the Pearson correlation coefficient ρ_(VT) between vegetation health and soil temperature is determined; a temperature compensation mechanism is triggered when |ρ_(VT)| > 0.6.

[0220] The decision parameters are dynamically optimized based on the analysis results: when the vegetation health value is below the 0.65 threshold for three consecutive samplings and the light intensity is consistently higher than the historical average by 15%, the upper limit of the soil moisture control range is reduced by 8%-12% of the current threshold; when the correlation coefficient between soil temperature and light intensity ρ_(LT) < -0.4, the exploration range of the irrigation action selection probability is expanded from ±0.1 to ±0.25. The decision boundary is adjusted by modifying the state-action mapping table: for the state codes 0x3A-0x3F corresponding to the health range of 0.4-0.6, the value Q of the associated "reduce irrigation intensity" action is increased by 30%. A multidimensional variable association rule report is output, including the light-action probability influence factor (0-1 scale), the health-temperature coupling coefficient (signed value), and the decision boundary update coordinates (hexadecimal state code range).

[0221] S630: Generate a new data fusion rule and decision model training dataset, and feed it back to step S100.

[0222] The data fusion mechanism is reconstructed based on the association rules output by S620: When the correlation coefficient ρ_(LT) between light intensity and soil temperature > 0.7, the observation noise weight of the light sensor in the Kalman filter of step S120 is increased, specifically by adjusting the noise variance of the light dimension in the R matrix from 0.05 to 0.035 (a reduction of 30%); when the vegetation health decreases by more than 10% for three consecutive collection cycles (45 minutes), the filtering window for soil temperature data is extended from 5 seconds to 10 seconds by increasing the time constant of the state transition equation. The parameter modifications are encapsulated in an XML format configuration file and written to the FPGA processing unit of step S120 via the PCIe bus to overwrite the original parameters.

[0223] A training dataset for the decision model was constructed: historical data was extracted using a 168-hour sliding window, including soil moisture feedback data from step S430 (sampling interval of 5 minutes), water pump operation status records (start / stop timestamps + power levels), and triplet data from S610. When adding decision labels, the graded irrigation operations performed in step S420 were encoded as four-dimensional one-heat vectors, for example, "increase irrigation intensity" was encoded as [0,1,0,0]. Data augmentation operations were performed based on the association rules of S620: when ρ_(LT) < -0.4, an extreme combination sample of light intensity 120,000 lux + soil temperature 10℃ was synthesized; when the health level was below 0.6, simulated data of temperature 35℃ + health level 0.4 was generated. The final dataset contains 20,000 records, each with 7 fields: timestamp, environmental status triplet, action vector, soil moisture feedback value, and water pump power value.

[0224] Closed-loop feedback is implemented as follows: the training dataset is stored in a MongoDB distributed database and marked as the model training input source for step S210; a hardware restart command is sent via the system control bus to make the updated Kalman filter parameters take effect in the next acquisition cycle, and the new dataset triggers model retraining in step S210. The feedback cycle is set to 24 hours to ensure dynamic iterative updates of parameters throughout the entire process.

[0225] Example 2: Figure 2 A structural block diagram of an intelligent irrigation monitoring system according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0226] The multi-source sensor array module 10 includes a capacitive humidity sensor, a thermocouple temperature sensor, and a photoelectric light sensor. The multi-source sensor array module is used to collect real-time soil moisture, temperature, and light data through the capacitive humidity sensor, thermocouple temperature sensor, and photoelectric light sensor.

[0227] The reinforcement learning decision module 20 is configured with a Q-learning decision model and an ε-greedy policy selector to generate the optimal irrigation strategy;

[0228] The water demand prediction module 30 stores historical soil moisture data and runs a gradient boosting decision tree algorithm to predict future crop water demand.

[0229] The hierarchical execution module 40 includes a TLV instruction converter and a PWM water pump controller, which are used to convert decision instructions into control signals to drive irrigation equipment;

[0230] The parameter optimization module 50 is configured with a regional deviation calculator and a reward function weight adjuster for dynamically optimizing system parameters.

[0231] The rule iteration module 60 includes a covariance analysis engine and a Kalman filter rule generator, used to iteratively generate and update irrigation rules.

[0232] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A smart irrigation monitoring method, characterized in that, include: Deploy arrays of capacitive humidity sensors, thermocouple temperature sensors, and photoelectric light sensors; use GPS timing and double-buffered queues to align multi-source data in time sequence; and employ Kalman filtering to fuse soil moisture data, soil temperature data, and light intensity data. The probe of the capacitive humidity sensor is inserted into the soil profile in a vertically layered manner, and converts the moisture content into a voltage signal output by detecting changes in dielectric constant. The thermocouple temperature sensor adopts a composite structure of thermocouple and thermistor, and is deployed in the key temperature change layer from the soil surface to a depth of 40 cm. It outputs a millivolt-level electrical signal through the linear relationship between resistance and temperature. The photoelectric light sensor is equipped with a photodiode and a spectral filtering module, and is fixed one meter above the vegetation canopy to convert the solar radiation spectrum into a standard lux or watt per square meter value. The filtered data is used to construct a three-dimensional environmental state vector, which is then input into the Q-learning decision model to generate a four-dimensional action value evaluation vector. An ε-greedy strategy is used to select the optimal irrigation action, and the dynamic irrigation threshold is calculated based on the selected action type. Finally, the water pump control command is output. The water pump control command is encapsulated into a hexadecimal data frame via the Modbus-RTU protocol. The frame structure includes five parts: start character, device address code, function code, data area, and CRC check code. The expression for the output water pump control command includes: ;in, This is the threshold boundary correction amount; For soil porosity tensor; The Christoffel notation describes the curvature of soil heterogeneity; This represents the spatial gradient of water content. For time step; Optimal energy control of water pumps: ;in, For the Hamiltonian function of the water pump system; For water flow rate; The coordinates of the water flow position; For water quality; Let gravitational potential energy function be used. For pipeline topology constraint functions; This is the constraint strength coefficient; The norm square operator; By associating dynamic irrigation thresholds with historical soil moisture data, and using gradient boosting decision trees to predict vegetation water demand, irrigation scheduling strategies with time windows and graded water quantities are generated. The water pump control commands and irrigation scheduling strategies are converted into TLV format commands. PWM control is used to start and stop the water pump in stages. Soil moisture feedback data is collected after the pump stops. By comparing the deviation between soil moisture feedback data and the dynamic irrigation threshold calculation area, the weight coefficient of the Q-learning reward function is adjusted, and the four-dimensional state decision model is reconstructed. By integrating vegetation health indicators with optimized decision model parameters, the Kalman filter rules are updated through covariance analysis, and a training dataset is generated and fed back to the sensor data fusion module.

2. The intelligent irrigation monitoring method according to claim 1, characterized in that, Constructing a three-dimensional environment state vector specifically includes: Soil moisture data is mapped to a water content coefficient in the range of [0,1], soil temperature data is converted into a temperature variation factor, and light intensity data is derived into a light intensity index. The three-dimensional environment state vector is used as the activation value of the input layer of the Q-learning decision model and is transmitted to the hidden layer with 128 ReLU activation function neurons through a fully connected neural network; The output layer is configured with four action nodes: maintain the current irrigation status, increase irrigation intensity, decrease irrigation intensity, and emergency water replenishment, generating a four-dimensional action value evaluation vector.

3. The intelligent irrigation monitoring method according to claim 1, characterized in that, The process of selecting the optimal irrigation action using the ε-greedy strategy includes: The exploration threshold ε is initially set to 0.7 and decreases by a decay factor of 0.

99. When the random number is higher than ε, the action corresponding to the maximum value of the action value evaluation vector is selected; when it is lower than ε, an action is randomly selected. The dynamic irrigation threshold calculation includes: lowering the lower limit of soil moisture by 3-5% when increasing irrigation intensity, increasing the upper limit by 2-4% when decreasing irrigation intensity, and setting a temporary threshold of 20% above the normal level when making emergency water replenishment.

4. The intelligent irrigation monitoring method according to claim 1, characterized in that, The process of generating a four-dimensional action value assessment vector from the input Q-learning decision model includes: By analyzing the number of irrigation triggers within 24 hours using a sliding time window, the interval between adjacent actions is extended by 10%-15% when the frequency per unit time exceeds the warning value. The dynamic threshold range of soil moisture content includes response level parameters of three levels: normal / accelerated / inhibited, minimum interval between pump start-up and shutdown, and maximum number of operations per day.

5. The intelligent irrigation monitoring method according to claim 1, characterized in that, The process of calculating dynamic irrigation thresholds based on selected action types includes: The structured threshold control group consists of: a lower limit threshold field for soil moisture content, an upper limit threshold field, and a four-bit binary response level identifier; The response mode activation strategies include: when the last two digits are 1, a 3-5 minute normal mode is activated; when the last two digits are 10, a 6-8 minute acceleration mode is activated; and when the last two digits are 11, a 1-2 minute suppression mode is activated.

6. The intelligent irrigation monitoring method according to claim 1, characterized in that, The process of calculating the dynamic irrigation threshold based on the selected action type and finally outputting the water pump control command also includes: The generation of water pump start / stop control commands includes three levels of decision logic: The first level of monitoring involves soil moisture remaining below the lower threshold for 60 consecutive seconds. The minimum interval for the second-level test pump is greater than 30 minutes. The third level checks that the number of daily operations has not reached the maximum value; Once the conditions are met, a quadruple instruction containing the device address, activation duration, water flow intensity level, and emergency flag is generated.

7. The intelligent irrigation monitoring method according to claim 6, characterized in that, The process of calculating the dynamic irrigation threshold based on the selected action type and finally outputting the water pump control command also includes: Control commands are encapsulated as Modbus-RTU protocol data frames, which include: Start character, device address code, function code, data area, and CRC checksum; The first byte of the data area stores the start time parameter (an integer value in 0.1-second units), the lower 4 bits of the second byte record the water flow intensity level, and the higher 4 bits reserve the emergency flag.

8. The intelligent irrigation monitoring method according to claim 1, characterized in that, The process of generating a four-dimensional action value assessment vector from the input Q-learning decision model includes: When the irrigation effect meets expectations, the corresponding action reward function is increased by 0.1 weight coefficient; When over-irrigation or insufficient water replenishment occurs, the penalty function is reduced by 0.05 weight coefficients; The fully connected weight matrix between the input layer and the hidden layer is updated using a temporal difference algorithm.

9. The intelligent irrigation monitoring method according to claim 1, characterized in that, include: When predicting vegetation water demand using gradient boosting decision trees, the time-series features of dynamic irrigation thresholds and historical soil moisture data are integrated to generate an irrigation scheduling strategy that includes time window parameters and graded water volume parameters. The time window parameters are accurate to 15-minute intervals, and the graded water volume parameters are divided into three flow levels according to vegetation type.

10. An intelligent irrigation monitoring system, applied to the method according to any one of claims 1-9, characterized in that, include: The multi-source sensor array module includes a capacitive humidity sensor, a thermocouple temperature sensor, and a photoelectric light sensor. The reinforcement learning decision module is configured with a Q-learning decision model and an ε-greedy policy selector. The water demand prediction module stores historical soil moisture data and runs a gradient boosting decision tree. The hierarchical execution module includes a TLV instruction converter and a PWM water pump controller; The parameter optimization module configures the region deviation calculator and the reward function weight adjuster. The rule iteration module includes a covariance analysis engine and a Kalman filter rule generator.

Images