Search

CN-122018304-A - Intelligent irrigation decision-making method for garden plants

CN122018304ACN 122018304 ACN122018304 ACN 122018304ACN-122018304-A

Abstract

The application belongs to the technical field of artificial intelligence and agricultural intelligence, and particularly relates to an intelligent irrigation decision-making method for garden plants. The method aims to solve the problems of large noise of perceived data, poor adaptability of a decision model and lack of dynamic optimization capability of the traditional irrigation system. The method comprises the steps of realizing multi-source data synchronous acquisition of soil humidity, meteorological parameters and plant canopy temperature by arranging a distributed sensing network, improving data reliability by combining time stamp alignment and self-adaptive Kalman filtering, constructing a water demand prediction model integrating Penman-Monteth models and a soil moisture dynamic equation, generating a dynamic irrigation demand reference, and introducing a lightweight convolutional neural network to perform fusion judgment on multi-mode characteristics. According to the application, through the cooperation of the data driving and the mechanism model, the accuracy and the self-adaptive capacity of irrigation decision making are obviously improved, and multiple targets of water saving, energy saving and healthy growth of plants are realized.

Inventors

  • QI XIAOTANG
  • LEI ZHIHUA
  • Zhu Zhenqiao

Assignees

  • 郁南县林业事务管理中心

Dates

Publication Date
20260512
Application Date
20251208

Claims (10)

  1. 1. An intelligent irrigation decision-making method for garden plants comprises the following steps: Carrying out multi-source heterogeneous sensing data acquisition and synchronization processing, arranging a distributed sensing network in a garden area, collecting soil humidity, air temperature and humidity, illumination intensity, wind speed, rainfall and plant canopy temperature data in real time, adopting a timestamp alignment and sliding window interpolation algorithm to carry out sampling synchronization on the multi-source data, carrying out noise suppression on an original observed value through self-adaptive Kalman filtering, and outputting a calibrated environment state vector; Carrying out dynamic modeling on water demand based on a physiological and ecological mechanism, constructing a coupling model fused with a Penman-Montetith evaporation model and a soil moisture dynamics equation, inputting a processed environment state vector, preset plant type parameters, a growth stage coefficient and root system distribution depth, calculating instantaneous potential evaporation quantity and soil effective water deficiency, and generating a dynamic water demand predicted value; Carrying out multi-mode data fusion and irrigation demand discrimination, carrying out multi-dimensional characteristic splicing on a water demand predicted value, a plant canopy temperature gradient change rate and a soil humidity gradient to form a high-dimensional decision input vector, inputting the high-dimensional decision input vector into a lightweight convolutional neural network model subjected to history data training, and outputting an irrigation demand level judging result; Generating and optimizing an irrigation strategy driven by closed-loop feedback, generating an initial irrigation strategy according to an irrigation demand level judging result, wherein the initial irrigation strategy comprises irrigation start-stop time, target irrigation quantity and a control area, continuously collecting soil humidity response curve of the irrigation area and plant physiological state change data after irrigation operation is executed, calculating actual water supplementing efficiency and plant stress response index, and inputting feedback information into an online reinforcement learning module; And carrying out multi-objective collaborative optimization and instruction issuing, comprehensively considering regional water resource quota, power load peak-valley period, weather forecast precipitation probability and plant community competition relationship, carrying out priority sequencing and time peak shifting scheduling on irrigation strategies of a plurality of subareas, generating a global optimal irrigation control instruction sequence, and issuing to an electric valve and water pump controller through a wireless communication module.
  2. 2. The method of claim 1, wherein the distributed sensor network is composed of at least 32 nodes, the deployment interval of each node is 15 meters to 25 meters, the burying depth of the soil humidity sensor is respectively 20 centimeters, 40 centimeters and 60 centimeters, the mounting height of the air temperature and humidity sensor is 1.5 meters, the plant canopy temperature is obtained by scanning at a frequency of 10 minutes through an infrared thermal imaging sensor, the time stamp alignment accuracy is controlled within +/-50 milliseconds, the sliding window length is set to 5 sampling periods, the state transfer matrix of the adaptive Kalman filtering is adjusted in real time according to the dynamic response characteristic of the sensor, the process noise covariance range is 0.01 to 0.08, and the observed noise covariance range is 0.02 to 0.1.
  3. 3. The method according to claim 1, wherein the surface resistance term of Penman-Montetith model is dynamically modified according to plant stomatal conductance empirical formula, the input plant type parameters include broadleaf tree, conifer or turf grass, the growth stage coefficients are divided into seedling stage, growing period, maturing period and dormancy stage, the root system distribution depth is set to 30 cm to 120 cm according to plant type, the soil moisture dynamics equation adopts Richards equation simplified form, the effective moisture reservoir capacity is calculated by combining field water holding capacity and wilting point data, the instantaneous potential evapotranspiration calculation time step is 1 minute, the soil effective moisture deficiency is defined as the difference between the current water holding capacity and the field water holding capacity, and the water replenishing mechanism is triggered when the value is more than 15%.
  4. 4. The method of claim 1, wherein the high-dimensional decision input vector comprises 7 dimensions, namely a filtered soil surface humidity value, a middle humidity change rate, a deep humidity gradient, an air relative humidity, a net radiation intensity, a wind speed square term and a plant canopy temperature first-order difference value, the lightweight convolutional neural network comprises 2 convolutional layers and 1 maximum pooling layer, the convolutional kernel size is 1 x3, the activation function adopts a ReLU, the number of neurons of a full-connection layer is 64, the output layer adopts a Softmax function to realize four classification decisions, the model training uses a historical dataset of the past 12 months, the total sample amount is greater than 8 ten thousand, a cross entropy loss function and Adam optimizer are adopted in the training process, the learning rate initial value is set to be 0.001, the batch size is 32, and the classification accuracy of the model on a verification set after training convergence is greater than 96%, and the F1-score is greater than 0.94.
  5. 5. The method according to claim 1, wherein the actual water replenishment efficiency is defined as a ratio of an increase in water content of an effective layer of soil after irrigation to a theoretical irrigation amount, the plant stress response index is calculated by a weighted sum of a canopy temperature rise rate and a transpiration suppression rate, the weight coefficients are respectively 0.6 and 0.4, a state space of the reinforcement learning module comprises a current soil humidity state, a weather trend and a plant health score, the action space is an irrigation amount adjustment gear, the reward function is designed as a weighted objective function integrating a water saving rate, a plant growth health degree and a system energy consumption, the discount factor is set to 0.9, the learning rate is set to 0.1, and model parameter updating is performed once every complete irrigation period is completed.
  6. 6. The method according to claim 1, wherein the water resource quota is distributed daily, the total daily amount is not more than 50 cubic meters per hectare, the peak-valley period of the power load is divided into a peak, a flat section and a valley, the period with the weather forecast precipitation probability of more than 40% is used for automatically deferring a non-emergency irrigation task, the plant community competition relationship is used for calculating the priority weight through the difference between the adjacent plants and the transpiration, the priority ordering adopts a multi-attribute decision algorithm, the conflict detection and the resource occupation check are carried out after the irrigation control instruction sequence is generated, the same pipe network branch is ensured not to appear concurrent operation, the instruction issuing adopts an encryption communication protocol, and the transmission delay is less than 200 milliseconds.
  7. 7. The method of claim 1, further comprising deploying an edge computing gateway device at a garden management site, wherein the gateway device is used for performing multisource heterogeneous perception data acquisition and synchronization processing, water demand dynamic modeling, multi-mode data fusion and irrigation demand discrimination, closed loop feedback driven irrigation strategy generation and optimized localized data processing and model reasoning, a dual-core processor is built in the gateway device, the main frequency is not lower than 1.2GHz, the memory capacity is not lower than 2GB, the memory space is not lower than 16GB, modbus and MQTT protocol conversion are supported, decision flows are independently operated in an offline state, a local caching strategy is started when a network is interrupted, and synchronous data are synchronized to a cloud management platform after connection is restored.
  8. 8. The method of claim 1, further comprising constructing a three-dimensional geographic information model of the garden area based on a digital twinning visual monitoring module, mapping each sensor position, soil humidity distribution thermodynamic diagram, plant health status identification and irrigation execution track in real time, supporting a time axis backtracking historical decision process, setting plant type labels, adjusting model parameter thresholds or manually intervening irrigation instructions by a manager through an interactive interface, and automatically recording all operation logs for subsequent model retraining.
  9. 9. The method of claim 1, further comprising a anomaly detection and fault tolerance mechanism to triple verify sensor data continuity, numerical rationality, and spatial consistency, and if a node data deviates from an adjacent node mean by more than 3 times standard deviation and lasts longer than 10 minutes, determining as a failed node, automatically enabling an alternative data source based on spatial interpolation and historical trend prediction by the system, and sending alarm information to the operation and maintenance terminal, and simultaneously freezing the weights of the failed node participation decisions until repair confirmation.
  10. 10. The method of claim 1, further comprising a natural resource asset optimization configuration knowledge base, storing historical configuration cases, policy and regulation treatises and expert rules, supporting multidimensional query by a semantic search engine, extracting a multi-objective weight distribution scheme by matching similar historical cases as prior knowledge input, shortening the exploration period of the reinforcement learning module, and further comprising a docking function with a homeland space planning 'one-map' system, acquiring ecological protection red lines, permanent basic farmland and town development boundary data through a geographic information service interface, performing space superposition analysis, checking topological relation between irrigation facility layout and planning boundaries, and automatically adjusting the scheme until compliance requirements are met if infringement or crossing a management and control area is found.

Description

Intelligent irrigation decision-making method for garden plants Technical Field The application belongs to the technical field of artificial intelligence and agricultural intelligence, and particularly relates to an intelligent irrigation decision-making method for garden plants. Background Along with the deep integration of intelligent agriculture and the Internet of things technology, landscaping management is gradually developing to intelligence and refinement. The high-efficiency utilization of water resources occupies a central position in urban garden maintenance, and the traditional irrigation mode is mostly dependent on manual experience or timing control strategies, and lacks dynamic perceptibility of the actual water demand state of plants, so that irrigation decision is delayed and water efficiency is low. Especially in a garden scene of complex climate environment and diversified vegetation configuration, the soil humidity, the meteorological conditions, the plant physiological parameters and other multidimensional factors are affected in an interweaving way, and higher requirements are put on the environmental adaptability and decision scientificity of an irrigation system. The intelligent irrigation system realizes preliminary transformation from 'experience driving' to 'data driving' by deploying environment sensors and automatic control equipment. The system aims at dynamically adjusting an irrigation plan according to environment monitoring data so as to improve the water resource utilization rate and ensure healthy growth of plants. However, in the prior art, the multisource perception data is subjected to simple threshold comparison or linear weighting treatment, so that nonlinear association and space-time evolution rules between the data cannot be fully mined, the generalization capability of a decision model is weak, and personalized irrigation requirements under different plant types, growth stages and microclimate differences are difficult to adapt. In the prior art, multiple technical bottlenecks are still faced in the process of realizing intelligent irrigation decision-making of garden plants. Firstly, the environment perception data has the problems of obvious noise interference and sampling asynchronism, the reliability and timeliness of the input data are difficult to ensure by the traditional filtering and fusion method, and the accuracy of subsequent decisions is influenced. Secondly, most systems adopt a static rule base to carry out decision-making reasoning, lack dynamic modeling capability of physiological and ecological processes such as plant transpiration, soil water holding characteristics and the like, and cannot realize accurate prediction of water demand. In addition, the irrigation strategy generation process ignores execution of feedback closed loops, and the irrigated soil moisture response and plant state change are not brought into a model iterative optimization mechanism, so that the phenomenon of excessive irrigation or insufficient water supply easily occurs in long-term operation of the system. Therefore, a garden plant intelligent irrigation decision-making method capable of integrating multi-source heterogeneous perception information and having dynamic learning and closed-loop optimization capabilities is needed. Disclosure of Invention The invention aims to provide an intelligent irrigation decision-making method for garden plants, which can effectively solve the problems in the background technology. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: A garden plant intelligent irrigation decision-making method comprises the following specific steps of (1) carrying out multi-source heterogeneous perception data acquisition and synchronization processing, namely, arranging a distributed sensing network in a garden area, collecting soil humidity, air temperature and humidity, illumination intensity, wind speed, rainfall and plant canopy temperature data in real time, carrying out sampling synchronization on the multi-source data by adopting a time stamp alignment and sliding window interpolation algorithm by the sensing network, eliminating the data asynchronism problem caused by equipment response delay, carrying out noise suppression on an original observed value through an adaptive Kalman filter, outputting a calibrated environment state vector, and (2) carrying out dynamic modeling on the water demand based on a physiological ecological mechanism, namely, constructing a coupling model of a Penman-Monteth model and a soil moisture dynamics equation, inputting the environment state vector processed in the step (1) and preset plant type parameters, growth stage coefficients and root system distribution depth, calculating the potential evaporation and the soil effective moisture shortage, generating a dynamic water demand prediction value, carrying out multi-mode data fusion and irrigation demand discrimin