Landscaping maintenance system and method based on intelligent water-saving irrigation

Abstract

The invention belongs to the field of garden irrigation and maintenance, and discloses a garden greening and maintenance system and method based on intelligent water-saving irrigation. Extracting initial characteristics and final characteristics of each plant, dividing the growth time of the plant from initial planting to final growth form into a plurality of continuous state updating cycles, and dividing the garden into a plurality of limit circles; at the starting time of each state updating period, for the plants in each limiting circle, extracting characteristic data and carrying out functional region division; dividing according to morphological characteristics, and if the morphological characteristics are not met, dividing according to maintenance requirements; for each functional area, a sensor array is deployed for collecting data, and a deployed central control unit controls a drip irrigation pipe network micro atomizer and a micro sprayer through a regulation and control scheme; a gradient twinborn self-adaptive growth mechanism is predefined, a regulation and control scheme is obtained through the gradient twinborn self-adaptive growth mechanism, and landscaping maintenance based on intelligent water-saving irrigation is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of garden irrigation maintenance, more particularly, the present application relates to a garden maintenance system based on intelligent water-saving irrigation and a method thereof. BACKGROUND

[0002] Due to the dispersed area of roof garden and the unique location, the cost of manual maintenance is extremely high, and intelligent maintenance becomes an extremely suitable means.

[0003] The soil layer of urban building roof garden is thin, and the plants are mainly shallow-rooted flowers and herbs (such as succulent plants and moss). The roof is exposed to sunlight and generally has no shelter, but the light scattering and reflection of the surrounding building surface increase the light intensity, resulting in a soil thermal conductivity much higher than that of normal soil on the ground, and a water evaporation rate increased by 40%. There is a problem of high thermal conductivity of shallow soil layer and rapid water loss, which cannot detect the bottom water gradient, resulting in an overestimation of the demand for upper and lower layers, a large increase in plant root burn rate, and a bottom that is not breathable due to the impermeable water material at the bottom of the building, which induces mold growth, destroys the ecological balance of the garden (such as attracting harmful insects and uncontrolled population), and increases the load on the roof structure.

[0004] Therefore, design and innovation are needed to meet the actual needs when maintaining the garden based on intelligent water-saving irrigation. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purposes, the present application provides the following technical scheme: a garden maintenance method based on intelligent water-saving irrigation, comprising: S1: extracting the initial characteristics and final characteristics of each plant according to the initial planting effect drawing and the final effect drawing, then dividing the growth time of the plant from the initial planting to the final growth form into multiple continuous state update periods, and dividing the garden into multiple limited circles; S2: at the starting time of each state update period, for the plants in each limited circle, extracting the characteristic data and performing function area division; first, according to the morphological characteristics, if it does not meet the requirements, then use the maintenance requirements for division; S3: dividing the soil of the garden into upper, middle and lower layers in proportion, for each function area, deploying a sensor array for data collection, and deploying a central control unit to control the micro-atomizer and micro-sprayer of the drip irrigation pipe network through a control scheme; S4: predefining a gradient twin adaptive evolution mechanism, including a basic layer, a self-organizing microbial membrane layer, a pipe adaptive game layer and an integration layer, and obtaining the control scheme through the gradient twin adaptive evolution mechanism; Among them, the three-dimensional grid is constructed in the base layer to simulate the layered state of the soil, the entire garden is regarded as a dynamic graph structure, and a prediction value is obtained; the self-organizing microbial membrane layer is used to simulate the heat adaptation behavior of the microbial membrane by generating a learning network, and an excretion rate is output; the pipe self-adaptive game layer is based on a multi-agent game model, simulates the dynamic game between the nanometer capillary tubes in different functional areas, regards the drip irrigation pipe network composed of the nanometer capillary tubes in each functional area as an agent, and regards the expansion rate as a strategy.

[0006] Preferably, the S1 comprises: The initial and final characteristics of each plant are extracted, including height, crown width, leaf size, root depth, and the transpiration rate, water demand, thermal conductivity influence, and ecological risk of each plant are recorded; According to the initial and final characteristics, a growth linear expression of the plant from initial planting to final growth form is predicted, and an inflection point with significant changes on the growth linear expression is identified, and a state update period is divided according to the time corresponding to the inflection point; At the end of each state update period, the growth range of each plant is drawn on the garden plan according to the characteristics of the plant at the inflection point, using a circle to represent the crown width, with the plant coordinates as the center and half of the crown width as the radius; If the crown widths of two plants intersect or the distance between them is less than a threshold value, they are considered to be continuous, and the area with continuous crown widths is identified, if the area contains only a single plant, the area is taken as a separate defined circle, otherwise, it is calculated whether its area exceeds a threshold value, if it does, the breaking point in the area is identified, and the area is disconnected at the breaking point to generate a defined circle, if there is no breaking point, the plants in the area are spatially clustered, and each cluster is taken as a defined circle; otherwise, it is marked as a defined circle.

[0007] Preferably, the dividing according to the morphological characteristics comprises: if the average height > a first high value and the average root depth > a first deep value, it is marked as a tall plant area; if the average height < a second high value and the average root depth < a second deep value, it is marked as a low plant area; if the average leaf size > a first leaf value and the tall plant area condition is not met, it is marked as a large leaf plant area; if the average leaf size < a second leaf value and the low plant area condition is not met, it is marked as a small leaf plant area. The dividing according to the maintenance requirements comprises: if the average transpiration rate > a first limit value, it is marked as a high water demand area; if the first limit value > the average transpiration rate > a second limit value, it is marked as a medium water demand area; and if the average transpiration rate < the second limit value, it is marked as a low water demand area.

[0008] Preferably, the collected data comprises the pH / excretion of the microbial membrane, and further comprises: In the tall plant zone, the soil humidity and temperature of the upper, middle and lower layers, the total weight of the soil, the types and density of insects are collected; in the low plant zone, the soil humidity and temperature of the upper and middle layers, the thermal conductivity of the upper soil, the types and coverage of mold are collected; in the medium water requirement zone, the soil humidity and temperature of the upper, middle and lower layers, the thermal conductivity of the upper soil, the types and density of insects, the types and coverage of mold are collected; In the large leaf plant zone and high water requirement zone, the soil humidity and temperature of the upper, middle and lower layers, the transpiration rate of plants, the types and coverage of mold are collected; in the small leaf plant zone and low water requirement zone, the soil humidity and temperature of the upper and middle layers, the types and density of insects are collected.

[0009] Preferably, the drip irrigation pipe network covers the upper, middle and lower layers of soil, and uses nanocapillary drip irrigation pipes and a layered design, each layer being equipped with an independent electromagnetic valve, a micro water pump and a micro voltage controller; at the same time, independent micro atomizers and micro sprayers are deployed, the micro atomizers being used to spray microbial membrane suspensions and the micro sprayers being used to spray ecological regulators, so as to cover the entire defined circle through controllable spraying; In each defined circle, a central control unit is deployed, and the control scheme is issued to the central control unit for coordinating all control components in the zone.

[0010] Preferably, in the base layer, the input layer is used to receive the data collected in all functional zones in all defined circles, respectively capturing the spatial scale characteristics of the functional zones, and setting different time windows for capturing the time scale characteristic fluctuations; In the dynamic graph structure, the nodes represent the functional zones, the edges represent the boundary effects between the functional zones, the node features are the fusion results of the collected data, the extracted features and the embedded vectors, and the edge features include the spatial distance between the functional zones and the plant property differences, the features of each functional zone are fused with the features of its adjacent functional zones to generate the boundary effects, and the attention mechanism is used to fuse the multi-scale spatial features and the time features; The predicted values of the soil humidity and temperature of the upper, middle and lower layers in each functional zone in each defined circle, the insect density, the mold coverage and the microbial membrane pH / secretion are outputted. A characteristic embedded vector is designed for each functional zone, and the embedded vector content includes preset target values adjusted according to the type of the functional zone.

[0011] Preferably, in the self-organizing microbial membrane layer, different membrane functions are defined for each functional zone, and the membrane functions are embedded into the embedded vectors corresponding to the functional zones, and then the membrane parameters are optimized.

[0012] In the self-organizing microbial membrane layer, different membrane functions are defined for each functional zone, and the membrane functions are embedded into the embedded vectors corresponding to the functional zones, and then the membrane parameters are optimized.

[0013] Preferably, in the pipe self-adaptive game layer, the reward function and the expansion function are defined according to the type of the functional area, the expansion function is embedded into the embedding vector as the initial strategy of the game, and the target of the reward function is to minimize water waste and root burn rate; The reward of the reward function is defined as the negative value of the sum of two judgment values, including the absolute value of the difference between the predicted humidity and the target humidity of the soil layer and the absolute value of the water waste; and the water supply adjustment of the expansion function is the sum of the absolute value and the expansion rate coefficient; Preferably, the integration layer integrates the outputs of the basic layer, the self-organizing microbial membrane layer and the pipe self-adaptive game layer into parameters in the regulation scheme, including: Extract the target value in the embedding vector, obtain the deviation between the predicted value and the target value output in the basic layer, and record it as the original deviation; adjust the original deviation through the boundary effect between the functional areas, and record it as the adjusted deviation; take the adjusted deviation, the microbial membrane secretion rate prediction value and the nanocapillary expansion rate prediction value of each functional area as the input of the optimization algorithm, minimize the original deviation, and output the irrigation amount, the expansion rate, the microbial membrane dose and the ecological regulator dose of each functional area; and use the pre-defined mapping logic to obtain the final regulation scheme; The regulation scheme includes the irrigation time and flow of each layer of soil, the nanocapillary expansion rate and the corresponding voltage, the spray dose of the microbial membrane suspension, the rotation angle of the micro-atomizer, the required pressure of the spray radius, the spray dose of the ecological regulator, the rotation angle of the micro-sprayer and the required pressure of the spray radius in each functional area in each defined circle.

[0014] A garden greening maintenance system based on intelligent water-saving irrigation, comprising: A period and area division module is used for dividing the growth time of plants from initial planting to final growth form into a plurality of continuous state update periods, dividing the garden into a plurality of defined circles, and dividing functional areas according to form characteristics and maintenance requirements; A component deployment module is used for deploying a central control unit for controlling the micro-atomizer and the micro-sprayer of the drip irrigation pipe network through the regulation scheme; A scheme generation module is used for predefining a gradient twin self-adaptive evolution mechanism and generating a regulation scheme.

[0015] The technical effects and advantages of the garden greening maintenance method based on intelligent water-saving irrigation are as follows: Through the division of the state update period and the weekly retraining, the plant growth and environmental changes can be dynamically adapted. Through the gradient twin self-adaptive evolution mechanism, a closed-loop automatic process from data collection to regulation scheme generation is realized without manual intervention. The central control unit automatically coordinates the drip irrigation pipe network, the micro-atomizer, the micro-sprayer and other execution components according to the regulation scheme issued by the cloud server, thereby improving the maintenance efficiency.

[0016] Through the personalized design of the embedding vector and the film function, the expansion function of the functional area characteristics, it is ensured that the maintenance measures are in line with the needs of the functional area. Through the integration layer optimization of irrigation amount, expansion rate and dosage, the waste of water, microbial membrane suspension and ecological regulator is significantly reduced. Through the precise control of the ecological regulator dosage, the use of chemical antibacterial agents is reduced, the beneficial bacteria is preferentially used, and the ecological balance is maintained.

[0017] The application not only improves the maintenance efficiency and ecological benefits of the roof garden, but also reduces the maintenance cost and environmental burden, and has significant innovation and practicality. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 It is a step schematic diagram of the garden maintenance method based on intelligent water-saving irrigation of the application; Figure 2 It is a structure schematic diagram of the garden maintenance system based on intelligent water-saving irrigation of the application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application. Embodiment 1

[0020] Please refer to Figure 1 and Figure 2 It is shown that the garden maintenance method based on intelligent water-saving irrigation of the application comprises: Before planting, the real garden situation is evaluated and a customized design is made. The diversity of plant species in the roof garden (such as height, leaf size, root depth, etc.), tall plants (such as shrubs or small trees) will form a shading effect, reducing the thermal conductivity of the soil surface, but also increase the local wind resistance, affecting heat diffusion. Shallow-rooted plants (such as succulents, herbs) mainly rely on upper soil moisture, deep-rooted plants (such as shrubs) need middle and lower layer moisture. This will cause spatial heterogeneity of water gradient. Large leaf plants have high transpiration rates, while small leaf or succulent plants have low transpiration rates. The water demand of different plant areas may differ by 2-3 times. Some plants (such as aromatic plants) may naturally inhibit mold or pests, while some plants (such as hygrophytes) may increase the risk of mold. Tall plants may reduce ventilation and increase the risk of mold; low-lying plant areas have good ventilation and more frequent insect activity. The spatial distribution of ecological risks is uneven and needs to be monitored and adjusted by area. Tall plants may increase the local soil weight (may increase 15-20 kg / m²), while low-lying plants have less impact. Deep-rooted plants may have greater pressure on soil structure, increasing the risk of local saturation. The spatial distribution of structural load is uneven and needs to be monitored for weight changes by area.

[0021] Especially, the leaves, roots and other parts of plants are in a dynamic process of change during growth, so the zoning is not fixed and needs to be dynamically updated according to the growth state to follow the plant growth and carry out more precise regulation and maintenance.

[0022] Obtain the initial planting effect diagram and the final effect diagram, and convert them to digital format (such as JPEG, PNG or DWG). Use image processing software (such as OpenCV) to grayscale, denoise and edge detection on the design diagram, and extract the plant area through the garden plan; use target detection algorithm (such as YOLOv5) or manual annotation to identify plants and mark the species and location of each plant in the plant area. Specifically, the location can be represented by coordinates, and the diagram can be divided into grids, each grid is marked with a unique coordinate, and the plants in the grid are matched with the coordinates. It is possible that a plant occupies several grids, then these several grids can be merged into a large grid, and the coordinates are merged into a coordinate.

[0023] Extract the initial characteristics (first planting) and final characteristics (extracted from the final effect diagram, or obtained from plant data according to the merchant or category) of each plant, including height, crown width, leaf size, root depth, and record the transpiration rate, water demand, thermal conductivity influence and ecological risk of each plant; can be obtained through plant characteristics database; Since the plant needs to go through a growth period from just planted to the final form effect, it needs to be closely observed and maintained during this growth period, and if the maintenance is not in place, it is easy to cause the plant to die, therefore, according to the initial characteristics and the final characteristics, the growth linear expression of the plant from the initial planting to the final growth form (that is, the state in the final effect diagram) can be predicted using a nonlinear growth model (such as Logistic model), and the growth linear expression formula can be: f_t=A / (1+e -k(t-t_0) ), where f_t represents the characteristic value at time t, A represents the final characteristic value, k is the growth rate, and t_0 is the growth inflection point determined by the plant species and environmental conditions (sunlight, wind, soil depth); identify the inflection point with significant changes in the growth linear expression, and divide the time from the initial planting to the final growth form of the plant into multiple continuous state update periods according to the time corresponding to the inflection point. The input parameters of the nonlinear growth model include the initial characteristic value, the final characteristic value, and the environmental conditions. For each plant, the height, crown width, root depth, transpiration rate, and other characteristics are predicted on the time axis (0 to the final effect time, such as 3 years) with a step of 0.1 year. The growth prediction curve table is output, recording the characteristics of each plant at different time points.

[0024] Use the change detection algorithm (such as CUSUM algorithm) to analyze the growth prediction curve table and identify the significant inflection point of the plant characteristic change (such as the crown width growth rate falling below 10%). According to the inflection point, divide the period, and dynamically adjust the length of each period to ensure that the plant characteristic changes little within the period. Record the start and end time of each period according to the period division table.

[0025] At the end of each state update period, use the circular to represent the crown width according to the corresponding characteristics of the plant at the inflection point, with the center of the circle as the plant coordinates and the radius as half of the crown width, and draw the growth range of each plant on the garden plan; Use the image processing algorithm (such as OpenCV's connected component analysis) to identify the area with continuous crown width. If the crown widths of two plants intersect or the distance between them is less than a threshold value (such as <0.1m), it is considered continuous. If the area only contains a single plant, the area is considered as a separate limited circle. Otherwise, calculate whether the area exceeds a threshold value (such as 30 square meters). If it does, identify the break point in the area and disconnect the area at the break point to generate a limited circle. If there is no break point, perform spatial clustering on the plants in the area, and each cluster is considered as a limited circle. Otherwise, mark it as a limited circle; Assign a unique label to each limited circle and mark it on the garden plan. The label includes the unique number, area, contained plant species, and characteristics of each limited circle; Among them, the "break point" in the region can be identified using an image segmentation algorithm (such as the Watershed algorithm) (crown width distance > 0.5m or plant density < 50%). Spatial clustering refers to using a clustering algorithm (such as K-means clustering) to spatially cluster plants within the region, and the number of clusters can be pre-set by the user according to the area of the region (such as an area of 120m², clustered into 4 sub-zones). The clustering basis can be spatial distance (crown center distance) and plant characteristic difference.

[0026] The boundary of the defined circle can be drawn on the garden plan using different colors or line types, and the boundary of the defined circle can be marked in the garden using an automated marking device (such as a drone or robot), and the marking material is a stealth tag (such as an RFID tag buried in the soil).

[0027] At the start time of each state update cycle, for each plant within each defined circle, extract the characteristic data in the growth prediction curve table and perform functional zoning; first, divide according to morphological characteristics, and if not satisfied, then use maintenance needs for division; Morphological characteristics include: if the average height > the first high value (such as 1m) and the average root depth > the first deep value (such as 15cm), it is marked as a tall plant zone; if the average height < the second high value (such as 0.5m) and the average root depth < the second deep value (such as 10cm), it is marked as a low plant zone; if the average leaf size > the first leaf value (such as 10cm) and does not meet the tall plant zone condition, it is marked as a large leaf plant zone; if the average leaf size < the second leaf value (such as 5cm) and does not meet the low plant zone condition, it is marked as a small leaf plant zone; If the morphological characteristic division is not satisfied, the maintenance needs division is performed: Maintenance needs include: if the average transpiration rate > the first limit value (such as 40%), it is marked as a high water demand zone; if the first limit value > the average transpiration rate > the second limit value (such as between 20%-40%), it is marked as a medium water demand zone; if the average transpiration rate ≤ the second limit value, it is marked as a low water demand zone; If the defined circle contains multiple plants, the average value of the characteristics can be obtained by weighted average calculation, and the weight is the coverage area proportion of the plant.

[0028] The first high value, the first deep value, the second high value, the second deep value, the first leaf value, the second leaf value, the first limit value, and the second limit value can be pre-set by the user according to operating experience or experts according to experimental data.

[0029] On the plan, use different colors or symbols to indicate the type of each defined circle, record the marker, type, and plant species and characteristics of each defined circle. Use an automated marking device to mark the functional zone boundary within the defined circle in the garden.

[0030] The maintenance requirement division is a supplement to the morphological characteristic division, ensuring that each plant can be divided into a certain type of maintenance. If a plant meets both the maintenance requirement division and the morphological characteristic division (such as meeting the tall plant area and meeting the medium water requirement area), the morphological characteristic division is preferred to maintain consistency with the original division method.

[0031] The soil of the garden is divided into upper, middle and lower layers in proportion from top to bottom

such as upper layer (0-5 cm), middle layer (5-15 cm) and lower layer (15-30 cm)

[0032] The method of collecting soil temperature can be to deploy a digital temperature sensor (such as DS18B20) in each layer of soil, with a collection frequency of 1 time per hour.

[0033] The method of collecting soil weight can be to deploy a weighing sensor (such as HX711) at the bottom of the soil, with a collection frequency of 2 times per day (morning and evening).

[0034] The method of collecting thermal conductivity can be to use a heat flow sensor (such as HFP01) to measure the thermal conductivity of the upper layer of soil, with a collection frequency of 1 time per hour.

[0035] The method of collecting transpiration rate can be to use a portable photosynthetic transpiration instrument (such as LI-6400) to measure the transpiration rate of plant leaves, with a collection frequency of 1 time per day (morning).

[0036] The method of collecting the species and density of insects, the species and coverage of mold can be to use integrated OpenCV image recognition scanning, with a collection frequency of 2 times per week.

[0037] Spectrometer embedded in microbial membrane layer for monitoring secretions and pH, once daily in the morning.

[0038] Upload collected data to the cloud (CSV format), remove outliers using the Z-score method (data with Z>3 are removed), and normalize the data to the range of 0-1 (such as humidity 25%→0.25, use Kriging interpolation method to complete the missing data. And store in AWSS3 or local database.

[0039] Within each defined circle, deploy independent drip irrigation pipe networks covering upper, middle, and lower layers of soil. The drip irrigation pipes use a layered design (such as a three-layer composite pipe) to ensure accurate water distribution to the target soil layer. Each layer is equipped with an independent electromagnetic valve and a micro water pump; the layered control irrigation volume (unit: L / m²) and duration of upper, middle, and lower layers of soil, the water source is connected to the rainwater collection system, equipped with a filter.

[0040] The laying of the drip irrigation pipe uses nanocapillary, with a grid spacing (such as 5-10 cm) set according to the pipe diameter and soil level of each layer (such as 50-100 nm). The purpose is to fully cover the area. Nanocapillary is made of shape memory polymer and expands / contracts in response to voltage stimulation. Each layer is equipped with an independent micro voltage controller; control the expansion / contraction rate, the network is buried in the soil 5-10 cm deep, using green or transparent materials.

[0041] Use AutoCAD to design nanocapillary, materials can use polylactic acid (PLA) + shape memory polymer (SMP), achieve voltage 5V response, control the pipe to expand / contract. According to the absolute difference of soil, dynamically adjust the water supply (such as expand to increase water supply when the upper layer is dry, and contract to reduce water supply when the lower layer is saturated). Nanocapillary achieves precise water distribution through capillary action and shape memory effect, solving the problem of uneven gradient.

[0042] At the same time, deploy independent micro-atomizers and micro-sprayers. Micro-atomizers are used to spray microbial membrane suspension, and micro-sprayers are used to spray ecological regulators (such as antibacterial agents, beneficial bacteria), which cover the entire defined circle through controllable spraying. The atomizer and micro-sprayer are installed at the center of the defined circle. Controllable spraying refers to the use of universal rotation and controllable spray heads. The spray head can rotate 360°, or rotate at a certain angle, or direct spraying. The spray radius can be adjusted by pressure. The atomizer and micro-sprayer are connected to the liquid storage tank, the liquid delivery pipe is buried in the soil or arranged along the plant stem, using green or transparent materials.

[0043] The target of high-heat-resistant microbial membrane is to reduce thermal conductivity and transpiration loss. The microbial membrane acts as a "living barrier" by secreting organic acids and polysaccharides to form a water-retaining layer, serving as a dynamic water-retaining and heat barrier, responding to heat / water stimulation to self-assemble, similar to self-healing materials in biological engineering.

[0044] Buy base strains from suppliers, such as Bacillus subtilis, add aquaporin genes (enhance water retention) and heat-resistant genes (such as Hsp70) using CRISPR technology, place the strain in nutrient agar medium (temperature 30℃, pH7), cultivate for 48 hours to a concentration of 10^6 CFU / ml. Mix the sodium alginate carrier (ratio 1:10) to form a suspension for spraying, control the film thickness of the sprayed film to 1-2mm to ensure air permeability.

[0045] In each defined circle, deploy 1 central control unit (such as Arduino control board), connect all sensors and execution components (such as electromagnetic valve, atomizer, voltage controller, sprayer), issue control schemes to the central control unit for coordination of all control components in the area.

[0046] Receive data collected by sensors, upload data to cloud server through Wi-Fi module. The cloud server runs control algorithms to generate control schemes, parses JSON format, and issues to the Arduino control board.

[0047] Through the division of state update cycles and weekly retraining, it can dynamically adapt to plant growth and environmental changes. For example, automatically increase irrigation during the rapid growth period of plants; automatically increase the amount of microbial film during the summer high temperature. This dynamic adaptability ensures the long-term effectiveness of maintenance measures.

[0048] Gradient twin self-adaptive evolution mechanism, build a digital twin model, virtually replicate the soil gradient (upper, middle, lower), plant characteristics, ecological state, etc. of the real garden, form a "virtual garden", and drive the model evolution through real-time data. The model evolves through data-driven "evolution", dynamically adapting to changes in the garden (such as plant growth, environmental changes). According to the model prediction, generate control schemes for each functional area in each defined circle.

[0049] The gradient twin self-adaptive evolution mechanism includes the basic layer, the self-organizing microbial film layer, the pipe self-adaptive game layer, and the integration layer. In the base layer, a three-dimensional mesh can be constructed using finite element analysis method with a resolution of 5 centimeters, simulating the stratified state of the soil from the upper layer (0-5 centimeters), the middle layer (5-15 centimeters), to the lower layer (15-30 centimeters), the input layer is used to receive all the data collected in all functional areas within the defined circle, including soil humidity, temperature, thermal conductivity, transpiration rate, ecological indicators, microbial membrane pH / secretion, etc. Spatial scale features can be extracted using convolutional neural networks, setting three different sizes of convolutional kernels to capture spatial scale features within the functional area, including soil gradient changes (deviation changes of soil upper, middle, and lower layer humidity and temperature from target values), functional area interaction within the defined circle, and boundary effects between defined circles. Long short-term memory networks are used to extract time scale features, setting different time windows to capture time scale feature fluctuations; for example, setting three different time windows to capture short-term fluctuations (such as humidity changes within 1 hour), daily cycle fluctuations (such as temperature changes within 24 hours), and weekly cycle fluctuations (such as ecological indicator changes within 168 hours).

[0050] "Functional area interaction within the defined circle" refers to the mutual influence between different functional areas within the same defined circle due to spatial proximity, plant characteristic differences, and environmental condition changes. For example, an increase in irrigation volume in the middle and lower layers of the tall plant area (e.g., 6L / m² in the middle layer and 4L / m² in the lower layer) may increase the lower layer humidity of the large leaf plant area through soil lateral conduction; an increase in irrigation volume in the upper layer of the large leaf plant area (e.g., 9L / m² in the upper layer) may increase the upper layer humidity of the tall plant area through soil lateral conduction. An increase in mold coverage in the low plant area (e.g., from 3% to 6%) may increase the mold coverage in the medium water demand area through soil transmission or air flow; an increase in insect density in the medium water demand area (e.g., from 50 to 60 per m²) may affect the insect density in the low plant area through air flow. An increase in microbial membrane secretion rate in the large leaf plant area (e.g., from 35% to 40%) may change the soil pH in the tall plant area through soil conduction; an increase in microbial membrane secretion concentration in the tall plant area (e.g., from 10^6 CFU / ml to 10^7 CFU / ml) may affect the secretion concentration in the large leaf plant area through air flow.

[0051] The entire garden is regarded as a dynamic graph structure, where nodes represent functional areas, and edges represent boundary effects between functional areas (such as the shading effect of tall plant areas reducing the thermal conductivity of adjacent low plant areas, possibly reducing the upper irrigation of large leaf plant areas to avoid water waste. If the thermal conductivity of low plant areas fluctuates, it may increase the upper irrigation of medium water demand areas to compensate for fluctuations.). Node features are the fusion results of collected data, extracted features, and embedded vectors, and edge features include spatial distance and plant property differences between functional areas, such as height difference, transpiration rate difference, etc. Graph convolutional networks are used to model the interaction between functional areas, and through multi-layer graph convolution operations, the features of each functional area are fused with the features of its adjacent functional areas to generate enhanced features, thereby generating boundary effects.

[0052] Extracted features are generated through data processing, including soil humidity gradient (difference between upper, middle, and lower humidity), temperature fluctuation (24-hour standard deviation of upper, middle, and lower temperature), thermal conductivity fluctuation (24-hour standard deviation of upper thermal conductivity), ecological risk (abnormal value of insect density, mold coverage rate (exceeding threshold)), and microbial membrane state (pH / secretion rate of change), providing real-time state information of functional areas.

[0053] Through the embedded vector of the functional area, the type and characteristics of the functional area are encoded (such as tall plant areas paying more attention to lower humidity), providing prior knowledge of the functional area.

[0054] The dynamic part (sensor data and extracted features) and the static part (embedded vector) are spliced to form complete node features for graph convolution operations.

[0055] Multi-scale spatial and temporal features are fused using an attention mechanism, and the attention weight is dynamically adjusted according to the type of functional area. For example, for tall plant areas, increase the attention weight of lower soil humidity; for low plant areas, increase the attention weight of upper soil thermal conductivity. Output the predicted values of soil upper, middle, and lower humidity and temperature in each functional area within each defined circle, as well as insect density, mold coverage rate, and microbial membrane pH / secretion rate; A characteristic embedded vector is designed for each functional area to encode the type and characteristics of the functional area, with a vector dimension of 64. The embedded vector is generated through a fully connected layer, with an input of the average characteristics of the functional area (height, leaf size, root depth, transpiration rate), and the embedded vector content includes pre-set target values adjusted according to the type of functional area. Target values include target humidity, target temperature, target total water, target transpiration rate, target insect density (i.e. maximum insect density in healthy state), target mold coverage rate, and target secretion rate for each layer of soil in each functional area, etc.

[0056] The base layer uses historical data to train the global model, aiming to minimize the prediction error. The loss function adopts mean square error to measure the difference between the predicted value and the true value. The training parameters include batch size of 32, training rounds of 100, optimizer of Adam, and learning rate of 0.01.

[0057] After training is completed, the model weights are saved with the file name "global_mstf_model_weights.h5".

[0058] The method of adjusting the embedding vector content according to the functional area type includes: For tall plant areas, increase the weight of lower layer soil humidity (target lower layer humidity > 60%) and reduce the weight of thermal conductivity (due to shading effect); for low plant areas, increase the weight of upper layer soil humidity (target upper layer humidity > 30%) and thermal conductivity (due to long exposure time); for large leaf plant areas, increase the weight of transpiration rate (target transpiration rate < 40%) and upper layer soil humidity; for small leaf plant areas, reduce the weight of water (target total water < 5L / m²) and increase the weight of ecological risk (such as insect repellent); for high water demand areas, increase the weight of upper layer soil humidity (target upper layer humidity > 40%) and transpiration rate; for medium water demand areas, balance the weight of upper and middle layer soil humidity (target > 30%) and increase the weight of thermal conductivity; for low water demand areas, reduce the weight of water (target total water < 5L / m²) and increase the weight of water saving.

[0059] The self-organizing microbial membrane layer simulates the dynamic secretion behavior of microbial membranes under different environmental conditions (temperature, humidity, pH, etc.) by generating a learning network, which is used to simulate the thermal adaptation behavior of microbial membranes and output the secretion rate. Different membrane functions are defined for each functional area, and the membrane functions are embedded in the embedding vector corresponding to the functional area, and then the membrane parameters are optimized through adaptive meta-learning. Among them, different membrane functions are defined for each functional area, and the secretion rate in the membrane function is defined as the product of the difference and the secretion rate coefficient, and the difference is the step difference between the predicted temperature and the target temperature; For example, set the tall plant area to secretion rate = (predicted temperature - 22) * 0.15; the tall plant area has low temperature sensitivity due to shading effect, with a threshold of 22℃, to start secretion at lower temperatures to protect deep root systems; the secretion rate coefficient 0.15 is moderate, meeting the medium water demand.

[0060] Set the low plant area to secretion rate = (predicted temperature - 26) * 0.25; the low plant area has high temperature sensitivity due to long exposure time, with a threshold of 26℃, to start secretion at higher temperatures to control thermal conductivity fluctuations; the secretion rate coefficient 0.25 is higher, meeting the water retention needs of the upper layer soil.

[0061] Set the large leaf plant area to the secretion rate = (predicted temperature - 24) * 0.35; the large leaf plant area has high transpiration rate and medium temperature sensitivity, and the threshold is set to 24°C to start secretion at medium temperature and reduce transpiration rate; the secretion rate coefficient 0.35 is the highest, meeting the high water demand and antibacterial demand.

[0062] Set the small leaf plant area to the secretion rate = (predicted temperature - 28) * 0.10; the small leaf plant area has low water demand and the lowest temperature sensitivity, and the threshold is set to 28°C to start secretion at higher temperature to save resources; the secretion rate coefficient 0.10 is the lowest, meeting the low water demand and insect repellent demand.

[0063] Set the high water demand area to the secretion rate = (predicted temperature - 24) * 0.35; the high water demand area has high transpiration rate and medium temperature sensitivity, and the threshold is set to 24°C to start secretion at medium temperature and reduce transpiration rate; the secretion rate coefficient 0.35 is the highest, meeting the high water demand and antibacterial demand (similar to the large leaf plant area, as the characteristics of the two overlap).

[0064] Set the medium water demand area to the secretion rate = (predicted temperature - 25) * 0.20; the medium water demand area has medium water demand and heat conduction rate fluctuation, and the threshold is set to 25°C to start secretion at medium temperature to balance water retention and heat conduction rate control; the secretion rate coefficient 0.20 is moderate, meeting the water retention demand of the upper and middle soil layers.

[0065] Set the low water demand area to the secretion rate = (predicted temperature - 28) * 0.10; the low water demand area has low water demand and the lowest temperature sensitivity, and the threshold is set to 28°C to start secretion at higher temperature to save resources; the secretion rate coefficient 0.10 is the lowest, meeting the low water demand and water saving demand (similar to the small leaf plant area, as the characteristics of the two overlap).

[0066] In summary, a generative adversarial network module can be designed to simulate the dynamic self-organization behavior of microbial membranes. Each functional area defines a different membrane function as the initial parameters of the generator; the membrane function is embedded in the functional area characteristic embedding vector as the prior knowledge of the generator.

[0067] The generator of the generative adversarial network takes the temperature, humidity, and microbial membrane pH / secretion data of the functional area as input, and outputs the secretion rate of the microbial membrane, with a unit of percentage, simulating the thermal adaptation behavior of the membrane. The network structure consists of 3 layers of fully connected neural networks, each containing 100 neurons, using LeakyReLU activation function.

[0068] The discriminator of the generative adversarial network takes the secretion rate generated by the generator and the real secretion rate data as input, and outputs the probability of distinguishing true and false, ranging from 0 to 1. The network structure consists of 3 layers of fully connected neural network, each layer contains 100 neurons, and uses Sigmoid activation function.

[0069] The training goal is to make the secretion rate generated by the generator as close as possible to the real secretion rate, and the discriminator to distinguish true and false as much as possible. The loss function uses cross entropy loss to calculate the loss of the generator and the discriminator respectively.

[0070] An adaptive meta-learning method is used to optimize the membrane parameters, so that the model can quickly adapt to the microbial membrane behavior in different functional areas. Each functional area is regarded as a task, and the task goal is to minimize the prediction error of secretion rate. A batch of functional areas is randomly selected as tasks, and the membrane parameters are updated for each functional area through 5-step gradient descent. The meta-loss, i.e. the average prediction error of all functional areas, is calculated. The global membrane parameters are updated using the meta-loss. For new functional areas, the updated global membrane parameters are used to quickly adapt through 5-step gradient descent.

[0071] The meta-learning rate is set to 0.001, the task learning rate is set to 0.01, the batch size is set to 32, and the number of training rounds is set to 100. After training, save the meta-learning model weight file named "microbe_maml_weights.h5".

[0072] The pipe adaptive game layer is based on a multi-agent game model, which simulates the dynamic game between different functional areas of nanometer capillary, and optimizes the expansion rate through game theory and reinforcement learning to reduce resource waste.

[0073] The drip irrigation pipe network composed of nanometer capillary in each functional area is regarded as an agent, and the expansion rate is regarded as a strategy, with a percentage ranging from 0 to 25%. The reward function and expansion function are defined according to the type of functional area, and the expansion function is embedded into the embedding vector as the initial strategy of the game. The goal of the reward function is to minimize water waste and root burn rate; The reward of the reward function is defined as the negative value of the sum of two judgment values, including the absolute value of the difference between the predicted humidity and the target humidity of the soil layer, and the absolute value of the water waste; the water supply adjustment of the expansion function is the cumulative value of the absolute value and the expansion rate coefficient; For example, for the tall plant area, the benefit function is defined as benefit = -(|lower predicted humidity - 60%| + |water waste|), the tall plant area pays attention to the lower humidity due to the deep root system, and the target lower humidity is > 60% to protect the deep root system; the water waste is defined as the part of the irrigation amount exceeding the target water amount. The expansion function is defined as water supply adjustment = absolute difference * 0.10 (expansion rate 10-15%); the tall plant area has a medium water demand, and the expansion rate coefficient 0.10 is moderate, and the expansion rate range is 10-15% to avoid saturation of the lower layer.

[0074] For the low plant area, the benefit function is defined as benefit = -(|upper predicted humidity - 30%| + |water waste|), the low plant area pays attention to the upper humidity due to the shallow root system, and the target upper humidity is > 30% to meet the water retention demand of the upper soil; the water waste is defined as the part of the irrigation amount exceeding the target water amount. Water supply adjustment = absolute difference * 0.20 (expansion rate 15-20%), the low plant area has a higher upper water demand, and the expansion rate coefficient 0.20 is higher, and the expansion rate range is 15-20% to increase the upper water supply.

[0075] For the large leaf plant area, benefit = -(|upper predicted humidity - 40%| + |water waste|), the large leaf plant area pays attention to the upper humidity due to the high transpiration rate, and the target upper humidity is > 40% to reduce the transpiration rate; the water waste is defined as the part of the irrigation amount exceeding the target water amount. Water supply adjustment = absolute difference * 0.25 (expansion rate 20-25%), the large leaf plant area has a high water demand, and the expansion rate coefficient 0.25 is the highest, and the expansion rate range is 20-25% to meet the high water demand.

[0076] For the small leaf plant area, benefit = -(|total water amount - 5L / m²| + |water waste|), the small leaf plant area pays attention to the total water amount due to the low water demand, and the target total water amount is < 5L / m² to save resources; the water waste is defined as the part of the irrigation amount exceeding the target water amount. Water supply adjustment = absolute difference * 0.05 (expansion rate 5-10%), the small leaf plant area has a low water demand, and the expansion rate coefficient 0.05 is the lowest, and the expansion rate range is 5-10% to avoid excess water.

[0077] For the high water demand area, benefit = -(|upper predicted humidity - 40%| + |water waste|), the high water demand area pays attention to the upper humidity due to the high transpiration rate, and the target upper humidity is > 40% to reduce the transpiration rate; the water waste is defined as the part of the irrigation amount exceeding the target water amount (similar to the large leaf plant area, as the characteristics of the two overlap). Water supply adjustment = absolute difference * 0.25 (expansion rate 20-25%), the high water demand area has a high water demand, and the expansion rate coefficient 0.25 is the highest, and the expansion rate range is 20-25% to meet the high water demand (similar to the large leaf plant area, as the characteristics of the two overlap).

[0078] For the medium water demand area, the benefit = -(|average predicted humidity of upper and middle layers - 30%| + |water waste|), the medium water demand area has moderate fluctuations in water demand and thermal conductivity, focusing on the humidity of the upper and middle layers, the target average humidity of the upper and middle layers > 30% to balance water retention and thermal conductivity control; water waste is defined as the part of the irrigation amount that exceeds the target water amount. Water supply adjustment = absolute difference * 0.15 (expansion rate 15-20%), the medium water demand area has moderate water demand, the expansion rate coefficient 0.15 is moderate, and the expansion rate range is 15-20% to meet the water retention needs of the upper and middle soil.

[0079] For the low water demand area, the benefit = -(|total water amount - 5L / m²| + |water waste|), the low water demand area has low water demand, focusing on the total water amount, the target total water amount < 5L / m² to save resources; water waste is defined as the part of the irrigation amount that exceeds the target water amount (similar to the small leaf plant area, as the characteristics of the two overlap). Water supply adjustment = absolute difference * 0.05 (expansion rate 5-10%), the low water demand area has low water demand, the expansion rate coefficient 0.05 is the lowest, and the expansion rate range is 5-10% to avoid water excess (similar to the small leaf plant area, as the characteristics of the two overlap).

[0080] The goal of the game is to achieve Nash equilibrium, that is, the expansion rate of each functional area is optimal, while considering the boundary effect of adjacent functional areas.

[0081] The expansion function is embedded in the functional area characteristic embedding vector as prior knowledge of the game model.

[0082] The multi-agent reinforcement learning method is used to optimize the game model, and the algorithm is multi-agent deep deterministic policy gradient (MADDPG). The batch size is set to 32, the training round is set to 500, the optimizer is Adam, the learning rate is 0.01, the discount factor is 0.9, and the exploration rate decays from 1.0 to 0.1. After training is completed, the reinforcement learning model weight is saved, and the file name is "nano_marl_weights.h5".

[0083] The integration layer integrates the outputs of the base layer, self-organizing microbial membrane layer and pipe adaptive game layer into parameters in the regulation scheme, and generates the final regulation scheme parameters through optimization algorithm: Extract the target value in the embedding vector to get the deviation between the predicted value and the target value in the base layer, denoted as the original deviation, for example: Tall plant area: lower layer humidity deviation = |predicted lower layer humidity - 60%|.

[0084] Large leaf plant area: upper layer humidity deviation = |predicted upper layer humidity - 40%|, transpiration rate deviation = |predicted transpiration rate - 40%|.

[0085] Leaflet plant area: total water deviation = |predicted total water - 5 L / m²|.

[0086] Ecological risk deviation: insect density deviation = |predicted insect density - 50 per / m²|, mold coverage deviation = |predicted mold coverage - 5%| (threshold adjusted according to functional area type).

[0087] Microbial membrane state deviation: secretion rate deviation = |predicted secretion rate - target secretion rate| (target secretion rate output by self-organizing microbial membrane layer).

[0088] Adjust the original deviation by the boundary effect between functional areas, called adjusted deviation, as formula: adjusted deviation vector = original deviation vector x (1 + adjustment coefficient), adjustment coefficient is the degree of influence of boundary effect, ranging from -0.2 to 0.2.

[0089] Take the adjusted deviation of each functional area, microbial membrane secretion rate prediction value (output of self-organizing microbial membrane layer) and nanocapillary expansion rate prediction value (output of tube adaptive game layer) as input of optimization algorithm, minimize original deviation, for example, use multi-agent reinforcement learning (MARL) to optimize regulation parameters, the goal is to minimize the adjusted deviation vector, while considering water waste and ecological risk.

[0090] State: adjusted deviation vector, functional area characteristic embedding vector.

[0091] Action: regulation scheme parameters, including: irrigation time and flow (per layer of soil), nanocapillary expansion rate and voltage (per layer of soil), microbial membrane suspension dose, micro-atomizer parameters (rotation angle, pressure), ecological regulator dose, micro-sprayer parameters (rotation angle, pressure).

[0092] Reward: define reward function according to functional area type, goal is to minimize deviation, water waste and ecological risk. For example: Tall plant area: reward = -(|lower layer humidity deviation| + |water waste| + |insect density deviation|).

[0093] Large leaf plant area: reward = -(|upper layer humidity deviation| + |transpiration rate deviation| + |water waste| + |mold coverage deviation|).

[0094] Other functional area types are similar.

[0095] Output irrigation amount, expansion rate, microbial membrane dose, ecological regulator dose of each functional area, use predefined mapping logic to get final regulation scheme, including each defined circle in each functional area: Watering time and flow rate: the irrigation time (unit: min) and flow rate (unit: L / m²) of each layer of soil (upper, middle, lower). The mapping logic can be irrigation time (unit: min) = irrigation volume / drip pipe flow rate (unit: L / m² / min).

[0096] Nanocapillary expansion rate and voltage: the expansion rate (unit: %) and corresponding voltage (unit: V) of each layer of soil. The mapping logic can be voltage (unit: V) = expansion rate * 0.2 (assuming that 0.2V voltage is needed for every 1% increase in expansion rate).

[0097] Microbial membrane suspension dose: spraying dose (unit: L / m²), marked as 0 if no spraying is needed. The mapping logic can be dose (unit: L / m²) = secretion rate * 0.03 (assuming that 0.03 L / m² dose is needed for every 1% increase in secretion rate).

[0098] Micro-atomizer parameters: rotation angle (unit: degrees), required pressure for spraying radius (unit: Pa). The mapping logic can be rotation angle (unit: degrees) = dose * 180 (assuming that the angle increases by 180 degrees for every 1 L / m² increase in dose).

[0099] Pressure (unit: Pa) = dose * 1000 (assuming that the pressure increases by 1000 Pa for every 1 L / m² increase in dose).

[0100] Ecological regulator dose: spraying dose (unit: L / m²). The mapping logic can be: If the insect density > 50 insects / m², then the beneficial bacteria dose = (insect density - 50) * 0.01.

[0101] If the mold coverage > 5%, then the antibacterial agent dose = (mold coverage - 5) * 0.06.

[0102] If the insect density ≤ 50 insects / m² and the mold coverage ≤ 5%, then the dose is marked as 0.

[0103] Micro-sprayer parameters: rotation angle (unit: degrees), required pressure for spraying radius (unit: Pa). The mapping logic can be: Rotation angle (unit: degrees) = dose * 180 (assuming that the angle increases by 180 degrees for every 1 L / m² increase in dose).

[0104] Pressure (unit: Pa) = dose * 1000 (assuming that the pressure increases by 1000 Pa for every 1 L / m² increase in dose).

[0105] If the dose is 0, then the rotation angle and pressure are marked as 0.

[0106] By the gradient twin adaptive evolution mechanism, the soil gradient (upper, middle, lower humidity, temperature) of each functional area, ecological risk (insect density, mold coverage) and microbial membrane state (pH / secretion) can be accurately predicted, and targeted control schemes can be generated. By integrating the layer to optimize the irrigation amount and the nanometer capillary expansion rate, water waste can be minimized while meeting the water demand of the functional area. In particular, for low water demand areas and small leaf plant areas, by precisely controlling the total water amount, unnecessary irrigation is significantly reduced. By predicting the ecological risk in the basic layer and combining the integrated layer to optimize the dose of ecological regulator, insect density and mold coverage can be effectively controlled to maintain ecological balance. In particular, for high water demand areas and large leaf plant areas, by increasing the dose of antibacterial agent, the risk of mold is reduced; for tall plant areas and small leaf plant areas, by increasing the dose of beneficial bacteria, the risk of insects is reduced.

[0107] By dynamic graph neural networks and multi-scale spatio-temporal fusion models, the boundary effects between functional areas (such as the shading effect of tall plant areas reducing the thermal conductivity of adjacent low plant areas) can be accurately modeled and considered in the control scheme. For example, the increase in middle and lower layer irrigation in tall plant areas may reduce the lower layer irrigation in adjacent large leaf plant areas to avoid water waste. By predicting the soil weight in the basic layer and combining the integrated layer to optimize the irrigation amount and drainage amount, the structural burden of the roof garden can be effectively controlled. For example, when the soil weight increases by 10 kg / m², the pressure relief drainage system is automatically activated to drain excess water. Embodiment 2

[0108] Please refer to Figure 2 The embodiment does not describe some parts in detail, which can be seen from the description of embodiment 1. A garden landscaping maintenance system based on intelligent water-saving irrigation is provided, which comprises: A cycle and area division module is used to divide the growth time of plants from initial planting to final growth form into multiple continuous state update cycles, divide the garden into multiple limited circles, and divide the functional area according to the form characteristics and maintenance requirements; A component deployment module is used to deploy a central control unit for controlling the micro-atomizer and micro-sprayer of the drip irrigation pipe network through the control scheme; A scheme generation module is used to predefine a gradient twin adaptive evolution mechanism and generate a control scheme. Embodiment 3

[0109] The embodiment discloses an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned garden landscaping maintenance method based on intelligent water-saving irrigation is realized.

[0110] Since the electronic device introduced in the embodiment is the electronic device used in the implementation of the method for maintaining garden greening based on intelligent water-saving irrigation in the embodiment, the specific implementation of the electronic device and various changes thereof can be understood by those skilled in the art based on the method for maintaining garden greening based on intelligent water-saving irrigation in the embodiment, and therefore, how the electronic device implements the method in the embodiment will not be introduced in detail. As long as the electronic device used in the implementation of the method for maintaining garden greening based on intelligent water-saving irrigation in the embodiment is implemented by those skilled in the art, it belongs to the scope of protection of the present application.

[0111] The above formulas are dimensionless values calculated, the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and the preset parameters and threshold values in the formula are set by those skilled in the art according to the actual situation.

[0112] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solution falling within the concept of the present application shall be considered as falling within the protection scope of the present application. It should be noted that, for ordinary technical users in the technical field, some improvements and refinements without departing from the principles of the present application shall also be considered as falling within the protection scope of the present application.

Claims

1. A landscaping maintenance method based on intelligent water-saving irrigation, characterized in that, Comprise: S1: Extract the initial and final characteristics of each plant according to the initial planting effect map and the final effect map, then divide the growth time of the plant from the initial planting to the final growth form into multiple continuous state update periods, and divide the garden into multiple defined circles; S2: At the start time of each state update period, for each plant in each defined circle, extract the characteristic data and perform functional zoning; first, divide according to the morphological characteristics, and if not satisfied, then use the maintenance requirements for division; S3: Divide the soil of the garden into upper, middle and lower layers in proportion from top to bottom, and for each functional area, deploy a sensor array for data collection, and a central control unit is deployed to control the drip irrigation pipe network micro atomizer and micro sprayer through the control scheme; S4: Predefine the gradient twin adaptive evolution mechanism, including the basic layer, the self-organizing microbial membrane layer, the pipe adaptive game layer and the integration layer, and obtain the control scheme through the gradient twin adaptive evolution mechanism; Wherein, a three-dimensional grid is constructed in the basic layer for simulating the layered state of the soil, the entire garden is regarded as a dynamic graph structure, and a prediction value is obtained; the self-organizing microbial membrane layer uses a generative learning network to simulate the thermal adaptation behavior of the microbial membrane and outputs the secretion rate; the pipe adaptive game layer is based on a multi-agent game model to simulate the dynamic game between nanometer capillary tubes in different functional areas, and the drip irrigation pipe network composed of nanometer capillary tubes in each functional area is regarded as an intelligent agent, and the expansion rate is regarded as a strategy.

2. The garden maintenance method based on intelligent water-saving irrigation according to claim 1, characterized in that, The S1 comprises: Extract the initial and final characteristics of each plant, including height, crown width, leaf size, root depth, and record the transpiration rate, water demand, thermal conductivity influence and ecological risk of each plant; According to the initial and final characteristics, predict the growth linear expression of the plant from the initial planting to the final growth form; identify the inflection points with significant changes on the growth linear expression, and divide the state update period according to the time corresponding to the inflection point; At the end of each state update period, according to the characteristics of the plant at the inflection point, use a circle to represent the crown width, with the plant coordinates as the center and half of the crown width as the radius, and draw the growth range of each plant on the garden plan; If the crown widths of two plants intersect or the distance between them is less than a threshold value, they are considered to be continuous, and the area with continuous crown width is identified, if the area only contains a single plant, the area is regarded as a separate defined circle, otherwise calculate whether its area exceeds a threshold value, if it does, identify the break point in the area, and disconnect the area at the break point to generate a defined circle, if there is no break point, perform spatial clustering on the plants in the area, and each cluster is a defined circle; Otherwise, mark it as a defined circle.

3. The garden maintenance method based on intelligent water-saving irrigation according to claim 2, characterized in that, The dividing according to morphological characteristics includes: if the average height > the first high value and the average root depth > the first deep value, marking as tall plant area; if the average height < the second high value and the average root depth < the second deep value, marking as low plant area; if the average leaf size > the first leaf value and not meeting the tall plant area condition, marking as large leaf plant area; if the average leaf size < the second leaf value and not meeting the low plant area condition, marking as small leaf plant area; The dividing according to maintenance needs includes: if the average transpiration rate > the first limit value, marking as high water demand area; if the first limit value >= the average transpiration rate > the second limit value, marking as medium water demand area; if the average transpiration rate <= the second limit value, marking as low water demand area.

4. The garden maintenance method based on intelligent water-saving irrigation according to claim 3, characterized in that, The collected data includes the pH / secretion of the microbial membrane, and further includes: In the tall plant area, the soil upper, middle and lower layer humidity and temperature, the total weight of the soil, the type and density of insects are collected; in the low plant area, the soil upper and middle layer humidity and temperature, the upper layer soil thermal conductivity, the type and coverage of mold are collected; in the medium water demand area, the soil upper, middle and lower layer humidity and temperature, the upper layer soil thermal conductivity, the type and density of insects, the type and coverage of mold are collected; In the large leaf plant area and the high water demand area, the soil upper, middle and lower layer humidity and temperature, the plant transpiration rate, the type and coverage of mold are collected; in the small leaf plant area and the low water demand area, the soil upper and middle layer humidity and temperature, the type and density of insects are collected.

5. The garden maintenance method based on intelligent water-saving irrigation according to claim 4, characterized in that, The drip irrigation pipe network covers the upper, middle and lower layers of soil, uses nanocapillary tubes and adopts a layered design, each layer is equipped with an independent electromagnetic valve, a micro water pump and a micro voltage controller; at the same time, independent micro atomizers and micro sprayers are deployed, the micro atomizers are used to spray microbial membrane suspensions, and the micro sprayers are used to spray ecological regulators, and the whole defined circle is covered through controllable spraying; In each defined circle, one central control unit is deployed, and the control scheme is issued to the central control unit for coordinating all control components in the area.

6. The garden maintenance method based on intelligent water-saving irrigation according to claim 5, characterized in that, In the base layer, the input layer is used to receive the data collected in all functional areas in all defined circles, respectively capture the spatial scale characteristics inside the functional areas, and set different time windows to capture the time scale characteristic fluctuations; In the dynamic graph structure, the nodes represent the functional areas, the edges represent the boundary effects between the functional areas, the node features are the fusion results of the collected data, the extracted features and the embedded vectors, and the edge features include the spatial distance and the plant property difference between the functional areas, the features of each functional area are fused with the features of its adjacent functional areas to generate the boundary effects, and the attention mechanism is used to fuse the multi-scale spatial features and the time features; The prediction values of the soil upper, middle and lower layer humidity and temperature, the insect density, the mold coverage and the microbial membrane pH / secretion in each functional area in each defined circle are outputted; A characteristic embedded vector is designed for each functional area, and the embedded vector content includes preset target values which are adjusted according to the functional area type.

7. The garden maintenance method based on intelligent water-saving irrigation according to claim 6, characterized in that, The self-organizing microbial membrane layer defines different membrane functions for each functional area, and embeds the membrane functions into the embedding vectors corresponding to the functional areas. Wherein, each functional area defines a different membrane function, and the secretion rate in the membrane function is the product of the difference and the secretion rate coefficient, the difference being the step difference between the predicted temperature and the target temperature.

8. The garden maintenance method based on intelligent water-saving irrigation according to claim 7, characterized in that, In the pipe adaptive game layer, the revenue function and the expansion function are defined according to the type of the functional area, the expansion function is embedded into the embedding vector as the initial strategy of the game, and the target of the revenue function is to minimize water waste and root burn rate; The revenue of the revenue function is the negative value of the sum of two judgment values, the two judgment values including the absolute value of the difference between the predicted humidity and the target humidity of the soil layer and the absolute value of the water waste; the water supply adjustment of the expansion function is the cumulative value of the absolute value and the expansion rate coefficient.

9. The garden maintenance method based on intelligent water-saving irrigation according to claim 8, characterized in that, The integration layer integrates the outputs of the base layer, the self-organizing microbial membrane layer and the pipe adaptive game layer into parameters in the regulation scheme, including: Extracting the target value in the embedding vector, obtaining the deviation between the predicted value and the target value in the base layer, denoted as the original deviation, adjusting the original deviation through the boundary effect between the functional areas, denoted as the adjusted deviation, taking the adjusted deviation, the microbial membrane secretion rate prediction value and the nanocapillary expansion rate prediction value of each functional area as the input of the optimization algorithm, minimizing the original deviation, and outputting the irrigation amount, the expansion rate, the microbial membrane dosage and the ecological regulator dosage of each functional area, using the pre-defined mapping logic to obtain the final regulation scheme; The regulation scheme includes the irrigation time and flow of each layer of soil, the nanocapillary expansion rate and the corresponding voltage, the microbial membrane suspension spraying dosage, the micro-atomizer rotation angle and the spraying radius required pressure, the ecological regulator spraying dosage, the micro-sprayer rotation angle and the spraying radius required pressure in each functional area in each defined circle.

10. A landscaping maintenance system based on intelligent water-saving irrigation, used to implement the landscaping maintenance method based on intelligent water-saving irrigation according to any one of claims 1-9, characterized in that, The garden greening maintenance system based on intelligent water-saving irrigation includes: A period and area division module is used to divide the growth time of plants from initial planting to final growth form into multiple continuous state update periods, divide the garden into multiple defined circles, and divide the functional areas according to the form characteristics and maintenance needs; A component deployment module is used to deploy the central control unit of the drip irrigation pipe network micro-atomizer and micro-sprayer through the regulation scheme; A scheme generation module is used to predefine the gradient twin adaptive evolution mechanism and generate the regulation scheme.

Images