Landscaping vegetation sprinkling irrigation system and sprinkling irrigation method

Abstract

The invention relates to the field of landscaping sprinkling irrigation, and discloses a landscaping vegetation sprinkling irrigation system and method, and the system comprises a plant physiology and environment perception module, a data processing and physical coupling modeling module, a deep reinforcement learning decision module, a federated learning optimization and model sharing module, and a sprinkling irrigation execution and feedback control module. The sprinkling irrigation method comprises the following steps: collecting plant moisture physiological parameters and environmental factors, carrying out coupling modeling based on an improved Penman-Monteith model, generating an optimal irrigation strategy by utilizing reinforcement learning, updating a global model through federal learning, regulating and controlling sprinkling irrigation system parameters, and realizing real-time monitoring and feedback closed-loop control. A multi-source sensing technology is adopted, the plant moisture physiological parameters and the environmental factors are collected in real time, the high-quality data flow is generated, and the technical effect of accurately monitoring the plant moisture state and the environmental change is achieved.

Description

Technical Field

[0001] The present invention relates to the field of garden greening sprinkler irrigation, in particular to a garden greening vegetation sprinkler irrigation system and a sprinkler irrigation method. Background Art

[0002] With the development of intelligent agriculture, precision irrigation has gradually become one of the important means of efficient water use, especially in drought-stricken areas and water-scarce areas. How to make irrigation more scientific and water-saving has become a core issue of concern to growers. The traditional experience-based irrigation model can no longer meet the needs of modern agriculture for high and stable yields. The intelligent irrigation control system based on the crop's own status and environmental factors can not only improve the efficiency of water resource utilization, but also promote the healthy growth of crops and reduce labor costs. Therefore, it has extremely high application value and promotion prospects.

[0003] In existing technologies, some systems have achieved automatic irrigation control based on environmental factors such as soil moisture or air temperature and humidity. Such systems generally use simple sensors to collect surrounding environmental parameters, and then perform irrigation tasks through preset logical rules or timed triggering. Some studies have also introduced certain algorithms to optimize irrigation frequency, which to a certain extent reduces water waste and improves irrigation convenience and semi-automation level. Some high-end platforms even support cloud configuration to facilitate remote control and data viewing. These technologies have played a positive role in improving agricultural management efficiency.

[0004] However, these methods still have some shortcomings that cannot be ignored: First, they generally only focus on environmental conditions and fail to consider the plant's own water status, which leads to biased irrigation judgments and easy misjudgment of water needs. Second, the data collection dimension is single, lacking in-depth acquisition of physiological information such as plant leaves and stem flow, making the judgment basis somewhat one-sided. Third, irrigation control strategies are mostly static settings and cannot dynamically learn and adjust with changes in plants and the environment, limiting their ability to cope with complex situations. Another key point is that most systems operate independently, models are not interoperable, cross-regional experience sharing is difficult, and generalization ability and scalability are weak. To this end, those skilled in the art have proposed a landscape vegetation sprinkler irrigation system and sprinkler irrigation method to address the above problems. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a garden greening vegetation sprinkler irrigation system and sprinkler irrigation method, which solves the problems of single irrigation decision-making, insufficient perception information, rigid control strategy and poor system generalization ability in the existing technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A garden greening vegetation sprinkler irrigation system, comprising: Plant physiology and environmental perception module, used to collect plant water physiological parameters and environmental factors in real time and generate multi-source perception data streams; A data processing and physical coupling modeling module is used to process the perception data stream and calculate the coupling characteristics of plant water dynamics and environmental factors using an improved Penman-Monteith transpiration model and a leaf impedance matrix; A deep reinforcement learning decision module receives the feature vector output by the physical coupling modeling module and generates a control strategy based on the state space and action space through reinforcement learning to control the irrigation water pump frequency and sprinkler valve opening; Federated learning optimization and model sharing module for synchronizing multi-region control models and enabling intelligent control strategy migration under different plant communities and soil conditions; The sprinkler irrigation execution and feedback control module is used to adjust the execution actions of the sprinkler irrigation equipment according to the control strategy output by the reinforcement learning decision module, and feed back the status data of the irrigation process to the perception module to achieve closed-loop control of the system.

[0007] Preferably, the plant physiology and environment perception module includes: Microwave Doppler stemflow sensor, used to obtain vegetation stemflow rate data; Leaf surface impedance matrix sensor, used to obtain leaf surface impedance data; Environmental meteorological sensors to obtain environmental parameters such as air temperature, wind speed, radiation intensity, and soil moisture content; Hydraulic network pressure sensor is used to monitor the pressure of each node in the hydraulic network.

[0008] Preferably, the data processing and physical coupling modeling module includes: Data smoothing and differentiation unit, used to smooth the stem flow rate data and calculate its derivative and second-order derivative; a transpiration calculation unit, used to calculate transpiration based on the improved Penman-Monteith model; A leaf surface temperature calculation unit is used to estimate the leaf surface temperature gradient and correct the irrigation factor according to the gradient; The coupled feature extraction unit is used to generate the state space vector, including the rate of change of stem flow rate, transpiration, average value of network pressure and norm of leaf impedance.

[0009] Preferably, the deep reinforcement learning decision module includes: The state space definition unit is used to define the input state vector composed of plant moisture status, environmental factors, and pressure data; the action space definition unit is used to define the action control variables of sprinkler angle, valve opening, and water pump frequency; Reinforcement learning network training unit, which optimizes irrigation control parameters by using graph convolutional networks and TD3 reinforcement learning algorithm training strategies; The reward function unit is used to design appropriate reward functions based on water balance, energy efficiency, and plant leaf response to adjust strategy optimization.

[0010] Preferably, the federated learning optimization and model sharing module includes: Local model training unit, used to independently train the local reinforcement learning control model in each region; Model parameter transmission unit, used to regularly transmit local model parameters trained in each region to the global aggregation platform; Model aggregation unit, which aggregates model parameters of multiple regions through weighted averaging and updates the global control model; The knowledge distillation unit distills the aggregation model based on regional network topology information to improve the generalization ability of the global model.

[0011] Preferably, the sprinkler irrigation execution and feedback control module includes: The sprinkler control unit receives control signals from the reinforcement learning module, adjusts the pump frequency, valve opening, and sprinkler angle, and dynamically controls the operating status of the sprinkler array. The sprinkler equipment feedback unit is used to monitor the flow rate, pressure, sprinkler status parameters and impedance characteristics of the sprinkler array during the execution process, and feed the impedance data back to the sensing module; The execution status evaluation unit is used to evaluate the execution status of the sprinkler irrigation system in real time, including the impedance change of the sprinkler array, and adjust the control strategy based on the change trend to achieve precise irrigation.

[0012] Preferably, the formula for calculating transpiration using the improved Penman-Monteith model is: Where: ET0(t) is the transpiration rate, which means the amount of water evaporated by plants per unit time; R n is the net radiation, which is the solar radiation received by the plant leaves minus the radiation reflected and dissipated; G is the soil heat flux, which refers to the heat conduction process in the soil; ρ is the air density, which affects the heat capacity of the air; c p is the specific heat capacity, which indicates the heat capacity of air; u2 is the wind speed, and the change of wind speed directly affects the evaporation rate of water; e s is the saturated water vapor pressure, which indicates the maximum water vapor content that air can hold at a given temperature; e a is the actual water vapor pressure, which indicates the actual water vapor content in the air; Δ is the slope of the evaporation curve, which describes the relationship between the evaporation rate and the temperature; γ is the drought coefficient, which reflects the water demand of plants in a drought environment.

[0013] Preferably, the reward function is: Where: θ t is the current policy parameter; θ target is the target strategy parameter; P i (t) is the pressure at the hydraulic network node; P ref is the reference pressure value; S j (t) is the sensor data; S target is the target sensor data; α, β, γ are adjustment coefficients that determine the impact of different items on the total reward; N is the total number of sensors participating in the feedback; ||·|| represents the Euclidean norm.

[0014] Preferably, the impedance calculation formula of the nozzle array is: Where: Z array (f) is the total impedance of the nozzle array; R j is the resistance of the jth nozzle, which affects the resistance of water flow; L j is the inductance of the jth nozzle, which determines the nozzle's response to frequency; C j is the capacitance of the jth nozzle, which determines the capacitive reactance characteristic of the nozzle to frequency; f is the driving frequency, that is, the frequency at which the nozzle works; j is the index of the nozzle, which indicates the different electrical characteristics of each nozzle.

[0015] The present invention also provides a method for spraying irrigation of landscaping vegetation, comprising the following steps: S1. Obtain plant water physiological parameters and environmental factors; S2. Use the improved Penman-Monteith transpiration model to couple plant water dynamics with environmental factors; S3, based on reinforcement learning algorithm, generates optimal irrigation control strategy according to coupling modeling; S4. Aggregate the local models of multiple regions through the federated learning mechanism and update the global model; S5. Adjust the water pump frequency, nozzle valve opening, and nozzle angle in the sprinkler irrigation system according to the generated control strategy; S6. Monitor the operating status of the sprinkler system in real time and feed it back to the sensing module to form a closed-loop control system.

[0016] The present invention provides a garden greening vegetation sprinkler irrigation system and sprinkler irrigation method. It has the following beneficial effects: 1. The present invention adopts multi-source sensing technology to generate high-quality data streams by real-time collection of plant water physiological parameters and environmental factors, thereby achieving the technical effect of accurately monitoring plant water status and environmental changes. Compared with the existing solutions that rely on a single sensor or traditional irrigation system, the present invention solves the problem of traditional solutions' insufficient comprehensive and real-time monitoring of plant physiological changes through the synergistic effect of multiple sensors such as microwave Doppler stem flow sensors and leaf impedance matrix sensors.

[0017] 2. The present invention uses deep reinforcement learning technology to generate the optimal irrigation control strategy, flexibly adjusts the water pump frequency, nozzle angle and valve opening, and achieves the technical effect of automatically adjusting the irrigation amount according to the real-time water demand of the plants. Compared with the manual adjustment or fixed algorithm-based control scheme of the existing irrigation system, the present invention can continuously optimize the control strategy according to the dynamic changes of plants and the environment, effectively avoiding the problems of water waste and insufficient irrigation.

[0018] 3. The present invention introduces a federated learning mechanism, which optimizes the generalization capability of the irrigation system by synchronizing local models in multiple regions and updating the global model, achieving the technical effect of cross-regional intelligent decision-making and sharing. Compared with the control methods in the existing technology that can only optimize a single region, the global optimization and regional adaptation of the present invention combine to solve the shortcomings of traditional technologies that cannot adapt to different soil and climatic conditions.

[0019] 4. The present invention implements closed-loop control of the system through the sprinkler execution and feedback control module, adjusts irrigation parameters according to real-time data, and achieves the technical effects of refined irrigation and system self-regulation. Compared with the sprinkler irrigation system in the prior art that lacks real-time feedback, the present invention can promptly detect problems that may arise during the sprinkler irrigation process, such as uneven flow or nozzle failure, thereby effectively improving the stability of the irrigation system and the utilization efficiency of water resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of the system architecture of the present invention; Figure 2 This is a schematic diagram of the plant physiology and environment perception module architecture of the present invention; Figure 3 This is a schematic diagram of the data processing and physical coupling modeling module architecture of the present invention; Figure 4 This is a schematic diagram of the deep reinforcement learning decision module architecture of the present invention; Figure 5 This is a schematic diagram of the federated learning optimization and model sharing module architecture of the present invention; Figure 6 This is a schematic diagram of the sprinkler irrigation execution and feedback control module architecture of the present invention; Figure 7 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] Please see the attached Figure 1 -Attached Figure 6 The embodiment of the present invention provides a garden greening vegetation sprinkler irrigation system, comprising: Plant physiology and environmental perception module, used to collect plant water physiological parameters and environmental factors in real time and generate multi-source perception data streams; Specifically, in this embodiment, the plant physiology and environmental perception module is used to collect plant water physiological parameters and environmental factors in real time, generating a multi-source perception data stream. This module is designed to accurately capture plant physiological status and environmental condition data, providing accurate input for subsequent data processing and control decisions.

[0023] The plant physiology and environmental perception module comprises multiple sensor units with clear divisions of labor, working collaboratively to monitor changes in plants and their growing environment from different perspectives. Specifically, the module includes a microwave Doppler stem flow sensor, a leaf impedance matrix sensor, an environmental meteorological sensor, and a hydraulic network pressure sensor.

[0024] The microwave Doppler stemflow sensor is primarily used to obtain data on stemflow rates in vegetation. The sensor measures changes in stemflow by emitting microwaves and receiving reflected signals. Stemflow is the process by which plants transport water absorbed by their roots through their stems to their leaves. It plays a key role in plant water transport and water needs.

[0025] By detecting changes in stem flow, the sensor can reflect plant water needs in real time. Microwave technology is non-invasive and has no direct impact on plants, enabling accurate and long-term monitoring of plant physiological changes. Stem flow rate is a key indicator of plant water status, particularly during irrigation system adjustments, where it can directly influence irrigation volume adjustments.

[0026] The Leaf Impedance Matrix Sensor is used to obtain leaf impedance data. Leaf impedance refers to the resistance of a plant leaf surface to electrical current, reflecting the leaf's moisture content. This sensor infers leaf moisture by measuring changes in leaf surface resistance.

[0027] Measuring leaf impedance is crucial for monitoring water balance. When a plant's water content is high, its leaf impedance is low; when a plant is water-stressed, its leaf resistance increases. By monitoring leaf impedance in real time, the system can promptly identify whether a plant is facing water stress and adjust irrigation strategies accordingly. The leaf impedance matrix sensor provides more detailed and dynamic data on plant water content, better reflecting a plant's actual water needs than traditional soil moisture monitoring.

[0028] Environmental weather sensors are primarily used to measure environmental parameters such as air temperature, wind speed, radiation intensity, and soil moisture. Environmental factors are crucial influencing plant water evaporation and absorption. In particular, temperature, wind speed, and radiation intensity directly influence a plant's transpiration rate, thereby determining its water requirements.

[0029] Soil moisture is also a key indicator of plant water status. When soil moisture is insufficient, plants cannot obtain sufficient water, resulting in limited transpiration and growth. By collecting this environmental data in real time, the system can more accurately assess the relationship between plant water status and environmental conditions, further optimizing irrigation strategies.

[0030] Hydraulic network pressure sensors are used to monitor the hydraulic pressure at each node in an irrigation system. Changes in hydraulic pressure directly reflect changes in water flow, which is crucial for controlling water flow in irrigation systems.

[0031] By deploying pressure sensors at various nodes, the system's operational status can be monitored in real time, ensuring stable water flow and avoiding water waste or insufficient water supply due to insufficient or excessive pressure. Furthermore, feedback on hydraulic network pressure helps the deep reinforcement learning decision-making module more accurately calculate the required irrigation volume.

[0032] The sensors in this module work together to form a complete data acquisition and processing chain. Each sensor generates raw data as it monitors plant and environmental conditions, which is then fed into the Data Processing and Physical Coupling Modeling module. The resulting data stream contains information on multiple dimensions, including plant water physiological parameters (such as stem flow rate and leaf impedance) and environmental factors (such as temperature, humidity, wind speed, and radiation intensity).

[0033] The data processing and physical coupling modeling module is used to process the sensory data stream and calculate the coupling characteristics of plant water dynamics and environmental factors using the improved Penman-Monteith transpiration model and leaf impedance matrix; Specifically, in this embodiment, the Data Processing and Physical Coupling Modeling module processes the data stream from the Plant Physiology and Environmental Perception module and uses the improved Penman-Monteith transpiration model and leaf impedance matrix to calculate the coupling characteristics of plant water dynamics and environmental factors. By accurately processing and modeling this data, this module provides key input information for subsequent control decisions.

[0034] The Data Smoothing and Differentiation Unit smoothes the stem flow rate data collected from the Plant Physiology and Environmental Perception Module and calculates its derivative and second-order derivative. This process removes noise from the collected data and, through differentiation, calculates the trend of stem flow rate to more accurately reflect the dynamic changes in plant water content.

[0035] In practical applications, changes in plant water status exhibit a certain degree of hysteresis and volatility, necessitating smoothing of stem flow rate data to reveal a clearer pattern of change. By calculating the first-order derivative, the rate of change of the stem flow rate can be determined, further analyzing the plant's water absorption or release trends. The second-order derivative reflects the acceleration of water changes, providing more accurate dynamic feedback for the system.

[0036] The transpiration calculation unit calculates plant transpiration based on a modified Penman-Monteith model. Transpiration is an important indicator of plant water consumption and directly influences irrigation decisions. This unit uses the following formula to estimate transpiration: Where: ET0(t) is the transpiration rate, which means the amount of water evaporated by plants per unit time; R n is the net radiation, which is the solar radiation received by the plant leaves minus the radiation reflected and dissipated; G is the soil heat flux, which refers to the heat conduction process in the soil; ρ is the air density, which affects the heat capacity of the air; c p is the specific heat capacity, which indicates the heat capacity of air; u2 is the wind speed, and the change of wind speed directly affects the evaporation rate of water; e s is the saturated water vapor pressure, which indicates the maximum water vapor content that air can hold at a given temperature; e a is the actual water vapor pressure, which indicates the actual water vapor content in the air; Δ is the slope of the evaporation curve, which describes the relationship between the evaporation rate and the temperature; γ is the drought coefficient, which reflects the water demand of plants in a drought environment.

[0037] Using this formula, the system can calculate plant transpiration based on factors such as ambient temperature, humidity, and wind speed, and estimate the plant's water needs accordingly. This calculation provides a theoretical basis for subsequent irrigation decisions.

[0038] The leaf temperature calculation unit estimates the temperature gradient across the plant's leaves and uses this gradient to modify the irrigation factor. Leaf temperature is closely related to water evaporation, so by monitoring changes in leaf temperature, we can further infer the plant's water status.

[0039] Calculating leaf temperature typically relies on factors such as ambient temperature, radiation intensity, and wind speed. Measuring plant leaf temperature provides a more precise understanding of plant transpiration, helping to adjust irrigation strategies. In drought or high-temperature environments, leaf temperature can rise significantly, increasing the plant's water demand. By correcting for leaf temperature gradients, water supply can be more effectively managed to ensure normal plant growth.

[0040] The coupled feature extraction unit combines these physical parameters to generate a state-space vector. This vector includes the rate of change of stemflow rate, transpiration, average network pressure, and leaf impedance norm. These features collectively reflect the plant's water status, the influence of environmental factors, and the operational status of the hydraulic network.

[0041] Stem flow rate change rate: obtained through data smoothing and differential units, reflecting the rate at which plants absorb or release water.

[0042] Transpiration: obtained through the transpiration calculation unit, indicating the rate of water consumption by plants.

[0043] Average network pressure: obtained by the hydraulic network pressure sensor, indicating the overall water flow status in the irrigation system.

[0044] Leaf Impedance Norm: Measured by a leaf impedance matrix sensor, it provides an overall picture of the water status of plant leaves.

[0045] These features are combined to form a high-dimensional state space, which is then fed into a deep reinforcement learning decision module for further analysis and decision-making. Through these coupled features, the system can fully understand the relationship between plant water status and the environment, thereby optimizing irrigation control strategies.

[0046] The deep reinforcement learning decision module receives the feature vector output by the physical coupling modeling module and generates a control strategy based on the state space and action space through reinforcement learning to control the irrigation pump frequency and sprinkler valve opening; Specifically, in the landscape vegetation sprinkler irrigation system of the present invention, the deep reinforcement learning decision-making module plays a crucial role. Based on the feature vectors output by the physical coupling modeling module, this module generates an optimal irrigation control strategy using a reinforcement learning algorithm. Specifically, the deep reinforcement learning decision-making module automatically adjusts control variables such as pump frequency and sprinkler valve opening based on data such as plant moisture status, environmental factors, and hydraulic pressure, achieving efficient and intelligent irrigation control.

[0047] In this embodiment, the deep reinforcement learning decision module includes multiple submodules, specifically: a state space definition unit, an action space definition unit, a reinforcement learning network training unit, and a reward function unit. These submodules are described in detail below.

[0048] In this embodiment, the state space definition unit is responsible for defining the input state vector of the system. This state vector is composed of multiple variables, mainly including the water status of the plant, environmental factors, and hydraulic pressure data. Specifically, the input state space includes: Plant water status: This includes multi-dimensional data on plant water content obtained from the sensing module, such as stem flow rate, leaf resistance, etc. These data reflect the real-time water demand and water status of the plant.

[0049] Environmental factors: External environmental data acquired by environmental meteorological sensors, such as temperature, wind speed, humidity, and radiation intensity. These factors affect the transpiration rate of plants and thus determine their water requirements.

[0050] Hydraulic pressure data: This data is collected by hydraulic network pressure sensors to measure the state of the hydraulic network. Changes in hydraulic pressure are directly related to the flow state of the irrigation system and are a key factor affecting sprinkler irrigation efficiency.

[0051] By combining this data, the state space definition unit provides a comprehensive input data set for the deep reinforcement learning algorithm, enabling subsequent control strategies to be optimized based on multi-dimensional state information.

[0052] In the deep reinforcement learning decision module, the action space definition unit is responsible for defining and constraining the system's action space. The action space refers to the adjustable control variables that determine the specific operation of the irrigation system. Specifically, the action space includes the following variables: Nozzle Angle: The angle of the nozzle determines the coverage of the water flow and the uniformity of water distribution. By adjusting the nozzle angle, you can flexibly adjust the irrigation area according to the needs of the plants.

[0053] Valve opening: The valve opening directly affects the amount of water flow. By adjusting the valve opening, the system can control the flow of water to ensure that the plants in each area receive sufficient water.

[0054] Pump frequency: Pump frequency determines the rate at which water is delivered, impacting the overall water distribution within the sprinkler system. By varying the pump frequency, the water flow rate can be adjusted to further optimize the irrigation process.

[0055] In the present invention, the definition of the action space is not just a simple operational control, it needs to be dynamically adjusted according to the data in the state space to achieve intelligent control.

[0056] The reinforcement learning network training unit is the core component of the deep reinforcement learning decision module. This unit uses a graph convolutional network and the TD3 reinforcement learning algorithm to train policies and optimize irrigation control parameters.

[0057] Graph Convolutional Network (GCN): A GCN is a deep learning model that can process graph-structured data, particularly complex structures involving nodes and edges. In this example, GCN effectively handles the interactions between different hydraulic network nodes and establishes relationships between nodes within the irrigation system, thereby improving the model's prediction accuracy.

[0058] TD3 reinforcement learning algorithm: The TD3 algorithm is a policy optimization algorithm for reinforcement learning tasks, particularly well-suited for optimizing high-dimensional, continuous action spaces. TD3 effectively avoids overestimation of Q values through double Q-learning, target smoothing, and policy delay updates, thereby stabilizing the learning process and improving the algorithm's convergence.

[0059] During the training process, the reinforcement learning network automatically adjusts the irrigation control strategy based on the real-time feedback reward signal, enabling the system to dynamically optimize the irrigation strategy under different environments and plant needs to ensure the optimal irrigation effect.

[0060] The reward function unit defines the reward function used during reinforcement learning training. It designs appropriate reward functions based on factors such as water balance, energy efficiency, and plant leaf response, adjusting policy optimization. The reward function plays a crucial role in reinforcement learning, directly influencing policy updates and final decisions during the learning process.

[0061] Specifically, the reward function in this embodiment is as follows: Where: θ t is the current policy parameter; θ target is the target strategy parameter; P i (t) is the pressure at the hydraulic network node; P ref is the reference pressure value; Sj (t) is the sensor data; S target is the target sensor data; α, β, γ are adjustment coefficients that determine the impact of different items on the total reward; N is the total number of sensors participating in the feedback; ||·|| represents the Euclidean norm.

[0062] This reward function provides a comprehensive evaluation metric by comprehensively considering factors such as the error between the current policy and the target policy, the difference between the actual hydraulic network pressure and the reference pressure, and the deviation between sensor data and target sensor data. By optimizing this reward function, the reinforcement learning network can generate more reasonable irrigation control strategies.

[0063] Federated learning optimization and model sharing module for synchronizing multi-region control models and enabling intelligent control strategy migration under different plant communities and soil conditions; Specifically, in the landscape vegetation sprinkler irrigation system of the present invention, the federated learning optimization and model sharing module coordinates the control models of multiple regions to achieve intelligent control strategy migration, further improving the system's adaptability and accuracy. The core of this module is to use a distributed approach to model training and optimization, allowing different regions to share and apply optimization experience from other regions while retaining their local characteristics. Through local model training, parameter transmission, global model aggregation, and knowledge distillation, the system can achieve cross-regional intelligent decision-making sharing, ensuring that each region's irrigation system can operate optimally under different environmental conditions.

[0064] In this embodiment, the federated learning optimization and model sharing module consists of four key components: a local model training unit, a model parameter transmission unit, a model aggregation unit, and a knowledge distillation unit. The following describes each component in detail, supplementing the technical details to ensure that those skilled in the art clearly understand the working principles and functions of each module.

[0065] The main function of the local model training unit is to enable each region to independently train its local reinforcement learning model based on local plant communities, soil conditions, climate, and other factors. The training data for each region comes from sensor data such as plant water status, soil moisture, temperature, and radiation intensity, as well as feedback information from the irrigation process.

[0066] Specifically, the model trained for each region is optimized using reinforcement learning algorithms, such as TD3 (Twin Delayed Deep Deterministic Policy Gradient). This process enables each region to adjust irrigation parameters such as sprinkler angle, valve opening, and pump frequency based on its specific needs.

[0067] The model parameter transfer unit uploads the local model parameters trained in each region to the central platform for global model aggregation. The parameters generated by each local model during training (such as neural network weights and biases, reinforcement learning policy parameters, etc.) reflect the optimization results of that region and provide a certain degree of control strategy effectiveness.

[0068] Specifically, each region periodically uploads its trained model parameters to a central server, forming a data transmission link. To ensure data privacy and security, encrypted transmission and differential privacy technologies are typically used to prevent sensitive data leakage.

[0069] Through the model parameter transmission unit, the system can integrate the optimization results of multiple regions and jointly optimize the global model based on these parameters. The contribution of each local model is reflected in the uploaded parameters, and the optimization of the global model represents the aggregation of experience from multiple regions.

[0070] The model aggregation unit aggregates the weighted average of local model parameters from different regions to generate a global control model. Due to regional variations in irrigation conditions, plant types, soil properties, and other factors, models in different regions may exhibit different control strategy effects during training. Therefore, the model aggregation unit calculates the global model based on the weighted average of each regional model's contribution and data quality.

[0071] The aggregation unit assigns different weights to each region's local model based on factors such as model training performance and data quality. More accurate regional models receive larger weights, while less well-trained regional models may receive smaller weights. This way, the global model, while aggregating experiences from various regions, ensures that higher-quality models play a leading role in global decision-making.

[0072] Specifically, assume that the local model parameters of the i-th region are θ i , with weight w i , global model parameters θ global The calculation method can be expressed as: Where: θ i represents the local model parameters of the i-th region; w i is the weight of region i, which is usually determined based on the training effect of the model and the quality of the data; N is the number of regions participating in the training.

[0073] This aggregation process ensures that the global model can take into account the characteristics of each region and improve the overall performance of irrigation control.

[0074] The knowledge distillation unit further optimizes the global model and improves its generalization capabilities under different environmental conditions. Through knowledge distillation, the system can extract effective knowledge from the local models of each region, compress redundant information, and optimize the structure of the global model.

[0075] Distillation process: During the distillation process, regional network topology information (such as the connectivity structure of the irrigation network and plant distribution) is taken into account to ensure the adaptability of the global model across regions. Specifically, the distillation process learns from the local models of each region, allowing the global model to retain the most effective decision-making features, thereby improving its accuracy in multi-region applications.

[0076] By combining topological information between regions, distillation can remove redundant information from the model and optimize its structure. For example, if certain regions have similar climate and soil conditions, the system can use distillation to fuse the model features of these regions to reduce the complexity of the global model.

[0077] During the distillation process, knowledge compression is usually achieved by minimizing the following loss function: Where: L distill is the loss function; f i (x) is the output of the local model for the i-th region; f global (x) is the output of the global model; x is the input data (such as plant water requirements, environmental factors, etc.); N is the number of regions participating in the training.

[0078] Through this process, the system is able to optimize the global model so that it can not only process local data but also provide stronger adaptability across multiple regions.

[0079] The sprinkler irrigation execution and feedback control module is used to adjust the execution actions of the sprinkler irrigation equipment according to the control strategy output by the reinforcement learning decision module, and feed back the status data of the irrigation process to the perception module to achieve closed-loop control of the system.

[0080] Specifically, in the landscape irrigation system of the present invention, the irrigation execution and feedback control module plays a crucial role. This module is responsible for precisely adjusting the execution of the irrigation equipment based on the control strategy output by the deep reinforcement learning decision module. It also monitors the irrigation status data in real time and feeds it back to the perception module, thereby achieving closed-loop control of the system. This closed-loop control not only ensures system stability under varying environmental conditions but also effectively prevents over- or under-supply of water, thereby improving overall water resource utilization efficiency.

[0081] In this embodiment, the sprinkler irrigation execution and feedback control module consists of three major units: a sprinkler irrigation control unit, a sprinkler irrigation feedback unit, and an execution status evaluation unit. Each of these units will be described in detail below, along with technical details from actual operations.

[0082] The sprinkler control unit's primary task is to adjust key components of the sprinkler system, including pump frequency, valve opening, and sprinkler angle, based on control signals output by the deep reinforcement learning decision module. By precisely adjusting these control variables, the system can achieve optimized irrigation for different areas and plant types.

[0083] Specifically, the control unit receives the control signal from the decision module and converts it into specific operation instructions: Pump frequency regulation: The pump frequency directly affects the water supply rate. Controlling the pump frequency can adjust the overall flow rate. By adjusting according to plant water needs, environmental changes, and irrigation strategies, the system can ensure that each area receives the appropriate amount of water.

[0084] Valve opening adjustment: The valve opening controls the amount of water flow. Adjusting the valve opening can accurately control the water flow of each nozzle to ensure that water is evenly distributed at the roots of the plants.

[0085] Nozzle Angle Adjustment: The angle of the nozzle determines the coverage and distribution of the water flow. By precisely adjusting the nozzle angle, the system can optimize the spray pattern according to the irrigation needs of each area and avoid waste.

[0086] During this process, the sprinkler equipment control unit can quickly adjust the water pump frequency, valve opening and nozzle angle according to system requirements to ensure that the entire irrigation process is efficient and accurate.

[0087] The sprinkler system feedback unit monitors various parameters during the sprinkler system's operation and transmits real-time feedback data to the perception module. This data includes key parameters such as flow rate, pressure, and sprinkler status. Using this real-time monitoring data, the system can determine whether the irrigation effect is meeting expectations and adjust the equipment status accordingly.

[0088] Flow Monitoring: Flow sensors monitor water flow in sprinkler irrigation systems, ensuring that water distribution to each area is appropriate. Real-time flow data feedback enables the system to identify over- or under-irrigation situations and take appropriate action.

[0089] Pressure Monitoring: Monitoring pressure within the hydraulic network is a critical component of sprinkler irrigation systems. By measuring pressure at each node in the sprinkler network in real time, the system can determine if there is insufficient flow or overpressure. This pressure data helps optimize pump frequency and sprinkler head adjustments to ensure consistent water flow.

[0090] Nozzle status monitoring: Nozzle status monitoring assesses irrigation effectiveness by checking whether sprinklers are functioning properly. This includes checking for blockages, correct angles, and other issues. Real-time feedback on sprinkler status ensures the irrigation system is operating without equipment failure.

[0091] Through the data collected in real time by the feedback unit, the system can understand the status of the sprinkler equipment during execution and further adjust the control strategy based on the feedback information.

[0092] In a sprinkler irrigation system, the electrical characteristics of the sprinkler array also affect the overall irrigation performance. In particular, the impedance characteristics of the sprinkler array are crucial for system regulation. The impedance of the sprinkler array is a complex function influenced by multiple factors, including the resistance, inductance, and capacitance characteristics of each sprinkler.

[0093] Specifically, the impedance calculation formula of the nozzle array takes into account the resistance, inductance, and capacitance of each nozzle, reflecting the response of the entire array to different driving frequencies. The formula can be expressed as: Where: Z array (f) is the total impedance of the nozzle array; R j is the resistance of the jth nozzle, which affects the resistance of water flow; L j is the inductance of the jth nozzle, which determines the nozzle's response to frequency; C j is the capacitance of the jth nozzle, which determines the capacitive reactance characteristic of the nozzle to frequency; f is the driving frequency, that is, the frequency at which the nozzle works; j is the index of the nozzle, which indicates the different electrical characteristics of each nozzle.

[0094] By calculating the impedance of the sprinkler array, the system can adjust the operating state of each sprinkler according to the current drive frequency to ensure a balanced distribution of water flow. This impedance calculation formula is a crucial regulatory mechanism in sprinkler irrigation systems, helping them optimize water flow at different frequencies and improve irrigation efficiency.

[0095] The execution status assessment unit evaluates the irrigation system's execution status in real time and adjusts the control strategy based on the results. By analyzing real-time monitoring data, the assessment unit can identify potential issues during the irrigation process, such as insufficient water and uneven flow, and optimize system operation accordingly.

[0096] Specifically, the evaluation process includes the following aspects: Water balance assessment: By comparing real-time plant water demand data with irrigation feedback data, the system assesses whether the water content during irrigation is meeting the desired target. If insufficient or excessive irrigation is detected, the system automatically adjusts control parameters such as pump frequency and nozzle angle.

[0097] Irrigation uniformity assessment: This system monitors sprinkler status and water coverage to assess irrigation uniformity. If uneven water flow or insufficient coverage is detected in certain areas, the system adjusts sprinkler angles or valve openings to ensure each area receives the required water.

[0098] Equipment health assessment: Monitors the operating status of pumps and valves to assess their efficiency. If equipment malfunctions or performance degrades, the system triggers an alarm and adjusts operations to ensure continuous and stable operation of the irrigation system.

[0099] By implementing the status assessment unit, the sprinkler system can adjust itself according to real-time feedback during the irrigation process, ensuring that the irrigation strategy always remains at the optimal state.

[0100] The garden greening vegetation sprinkler irrigation method described below and the garden greening vegetation sprinkler irrigation system described above can correspond to each other.

[0101] Please see the attached Figure 7 , a method for spraying water for landscaping vegetation, comprising the following steps: S1. Obtain plant water physiological parameters and environmental factors; S2. Use the improved Penman-Monteith transpiration model to couple plant water dynamics with environmental factors; S3, based on reinforcement learning algorithm, generates optimal irrigation control strategy according to coupling modeling; S4. Aggregate the local models of multiple regions through the federated learning mechanism and update the global model; S5. Adjust the water pump frequency, nozzle valve opening, and nozzle angle in the sprinkler irrigation system according to the generated control strategy; S6. Monitor the operating status of the sprinkler system in real time and feed it back to the sensing module to form a closed-loop control system.

[0102] Specifically, step S1: Real-time acquisition of plant physiological data (such as stem flow rate and leaf surface impedance) and environmental factors (such as temperature, humidity, wind speed, radiation intensity, and soil moisture content) is performed through a sensor network. This data is transmitted to a central processing unit via wireless transmission protocols (such as LoRa and NB-IoT) to ensure real-time and accurate data. The sensor layout should cover the plant growth area to ensure comprehensive perception of various environmental factors.

[0103] Step S2: A modified Penman-Monteith model is used to calculate plant transpiration, taking into account environmental factors such as wind speed and radiation intensity, as well as plant characteristics such as transpiration coefficient and leaf area index. By coupling this data with plant water physiological parameters such as stem flow rate and leaf temperature, a dynamic model is established to accurately predict plant water requirements. This improved model can be customized to suit different regional climate and soil conditions to improve calculation accuracy.

[0104] Step S3: Utilizing reinforcement learning algorithms (such as deep reinforcement learning and Q-learning), the system generates a real-time optimal irrigation control strategy by inputting real-time data (e.g., plant water requirements, environmental factors, soil moisture, etc.). The model continuously learns and optimizes, adjusting irrigation rates and methods based on the plant's different growth stages and environmental conditions, thereby minimizing water waste and maximizing plant growth efficiency. The reinforcement learning algorithm automatically adjusts its strategy based on environmental feedback (e.g., water availability and irrigation effectiveness), achieving dynamic optimization.

[0105] Step S4: Using a federated learning mechanism, local models from different regions are aggregated to ensure data privacy while sharing regional irrigation experience and model advantages. Each region independently trains a local model, and a weighted average algorithm is used to aggregate the parameters of the local models across multiple regions to generate a global model. This global model incorporates cross-regional irrigation strategy optimization experience, thereby improving the generalization capabilities of the models in each region.

[0106] Step S5: Based on the generated optimal control strategy, the sprinkler system's operating parameters are adjusted in real time. By controlling the pump frequency, sprinkler valve opening, and sprinkler angle, water is evenly distributed across the plant root zone. This process not only considers the plant's water needs but also adjusts irrigation intensity and coverage based on real-time environmental conditions (such as wind speed, temperature and humidity) and soil moisture levels, thereby avoiding over-irrigation or water shortages.

[0107] Step S6: During operation, the sprinkler system's operating status is monitored in real time using pressure sensors, flow sensors, and sprinkler status sensors. The collected operating data is fed back to the perception module to assess water distribution uniformity, pump and sprinkler operating conditions, and other factors. Based on the monitoring results, the perception module can identify potential issues (such as sprinkler blockage and pump failure) and automatically adjust the control strategy, forming a closed-loop control system to achieve efficient and self-optimizing operation of the sprinkler system.

[0108] The method of this embodiment can be used to execute the above system embodiment. Its principles and technical effects are similar and will not be described in detail here.

[0109] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A garden greening vegetation sprinkler irrigation system, characterized in that: include: Plant physiology and environmental perception module, used to collect plant water physiological parameters and environmental factors in real time and generate multi-source perception data streams; A data processing and physical coupling modeling module is used to process the perception data stream and calculate the coupling characteristics of plant water dynamics and environmental factors using an improved Penman-Monteith transpiration model and a leaf impedance matrix; A deep reinforcement learning decision module receives the feature vector output by the physical coupling modeling module and generates a control strategy based on the state space and action space through reinforcement learning to control the irrigation water pump frequency and sprinkler valve opening; Federated learning optimization and model sharing module for synchronizing multi-region control models and enabling intelligent control strategy migration under different plant communities and soil conditions; The sprinkler irrigation execution and feedback control module is used to adjust the execution actions of the sprinkler irrigation equipment according to the control strategy output by the reinforcement learning decision module, and feed back the status data of the irrigation process to the perception module to achieve closed-loop control of the system.

2. A garden greening vegetation sprinkler irrigation system according to claim 1, characterized in that: The plant physiology and environment perception module includes: Microwave Doppler stemflow sensor, used to obtain vegetation stemflow rate data; Leaf surface impedance matrix sensor, used to obtain leaf surface impedance data; Environmental meteorological sensors to obtain environmental parameters such as air temperature, wind speed, radiation intensity, and soil moisture content; Hydraulic network pressure sensor is used to monitor the pressure of each node in the hydraulic network.

3. A garden greening vegetation sprinkler irrigation system according to claim 1, characterized in that: The data processing and physical coupling modeling module includes: Data smoothing and differentiation unit, used to smooth the stem flow rate data and calculate its derivative and second-order derivative; a transpiration calculation unit, used to calculate transpiration based on the improved Penman-Monteith model; A leaf surface temperature calculation unit is used to estimate the leaf surface temperature gradient and correct the irrigation factor according to the gradient; The coupled feature extraction unit is used to generate the state space vector, including the rate of change of stem flow rate, transpiration, average value of network pressure and norm of leaf impedance.

4. A garden greening vegetation sprinkler irrigation system according to claim 1, characterized in that: The deep reinforcement learning decision module includes: A state space definition unit is used to define the input state vector consisting of plant water status, environmental factors, and pressure data; Action space definition unit, used to define the action control variables of nozzle angle, valve opening and water pump frequency; Reinforcement learning network training unit, which optimizes irrigation control parameters by using graph convolutional networks and TD3 reinforcement learning algorithm training strategies; The reward function unit is used to design appropriate reward functions based on water balance, energy efficiency, and plant leaf response to adjust strategy optimization.

5. A garden greening vegetation sprinkler irrigation system according to claim 1, characterized in that: The federated learning optimization and model sharing module includes: Local model training unit, used to independently train the local reinforcement learning control model in each region; Model parameter transmission unit, used to regularly transmit local model parameters trained in each region to the global aggregation platform; Model aggregation unit, which aggregates model parameters of multiple regions through weighted averaging and updates the global control model; The knowledge distillation unit distills the aggregation model based on regional network topology information to improve the generalization ability of the global model.

6. A garden greening vegetation sprinkler irrigation system according to claim 1, characterized in that: The sprinkler irrigation execution and feedback control module includes: The sprinkler control unit receives control signals from the reinforcement learning module, adjusts the pump frequency, valve opening, and sprinkler angle, and dynamically controls the operating status of the sprinkler array. The sprinkler equipment feedback unit is used to monitor the flow rate, pressure, sprinkler status parameters and impedance characteristics of the sprinkler array during the execution process, and feed the impedance data back to the sensing module; The execution status evaluation unit is used to evaluate the execution status of the sprinkler system in real time, including the impedance changes of the sprinkler array, and adjust the control strategy based on the change trend to achieve precise irrigation.

7. A garden greening vegetation sprinkler irrigation system according to claim 3, characterized in that: The formula for calculating transpiration using the improved Penman-Monteith model is: Where: ET0(t) is the transpiration rate, which means the amount of water evaporated by plants per unit time; R n is the net radiation, which is the solar radiation received by the plant leaves minus the radiation reflected and dissipated; G is the soil heat flux, which refers to the heat conduction process in the soil; ρ is the air density, which affects the heat capacity of the air; c p is the specific heat capacity, which indicates the heat capacity of air; u2 is the wind speed, and the change of wind speed directly affects the evaporation rate of water; e s is the saturated water vapor pressure, which indicates the maximum water vapor content that air can hold at a given temperature; e a is the actual water vapor pressure, which indicates the actual water vapor content in the air; Δ is the slope of the evaporation curve, which describes the relationship between the evaporation rate and the temperature; γ is the drought coefficient, which reflects the water demand of plants in a drought environment.

8. A garden greening vegetation sprinkler irrigation system according to claim 4, characterized in that: The reward function is: Where: θ t is the current policy parameter; θ target is the target strategy parameter; P i (t) is the pressure at the hydraulic network node; P ref is the reference pressure value; S j (t) is the sensor data; S target is the target sensor data; α, β, γ are adjustment coefficients that determine the impact of different items on the total reward; N is the total number of sensors participating in the feedback; ||·|| represents the Euclidean norm.

9. A garden greening vegetation sprinkler irrigation system according to claim 6, characterized in that: The impedance calculation formula of the nozzle array is: Where: Z array (f) is the total impedance of the nozzle array; R j is the resistance of the jth nozzle, which affects the resistance of water flow; L j is the inductance of the jth nozzle, which determines the nozzle's response to frequency; C j is the capacitance of the jth nozzle, which determines the capacitive reactance characteristic of the nozzle to frequency; f is the driving frequency, that is, the frequency at which the nozzle works; j is the index of the nozzle, which indicates the different electrical characteristics of each nozzle.

10. A method for sprinkling irrigation for garden greening vegetation, applied to a sprinkling irrigation system for garden greening vegetation according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Obtain plant water physiological parameters and environmental factors; S2. Use the improved Penman-Monteith transpiration model to couple plant water dynamics with environmental factors; S3, based on reinforcement learning algorithm, generates optimal irrigation control strategy according to coupling modeling; S4. Aggregate the local models of multiple regions through the federated learning mechanism and update the global model; S5. Adjust the water pump frequency, nozzle valve opening, and nozzle angle in the sprinkler irrigation system according to the generated control strategy; S6. Monitor the operating status of the sprinkler system in real time and feed it back to the sensing module to form a closed-loop control system.

Images