Farmland water-saving irrigation water system design and dynamic water distribution optimization method based on supply and demand multi-source data fusion

Abstract

The application relates to a farmland water-saving irrigation water system design and dynamic water distribution optimization method for supply-demand multi-source data fusion, characterized in that the method comprises the following steps: S1 "crop-soil-weather" multi-source data coupling modeling, S2 water-saving water system design optimization, S3 dynamic water-saving water distribution strategy, and S4 intelligent terminal control system control; the scheme realizes accurate water resource allocation through an artificial intelligence technology, maximally reduces irrigation water consumption, and simultaneously guarantees crop growth requirements.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture and efficient water resource utilization technology, specifically to an artificial intelligence-based method and system for designing and dynamically optimizing farmland water-saving irrigation systems. It achieves efficient planning and dynamic scheduling of irrigation systems through intelligent algorithms, with the core objective of minimizing water resource consumption, while also ensuring crop water needs and ecological sustainability. Background Technology

[0002] Traditional farmland irrigation systems suffer from problems such as extensive design, low water resource utilization, and over-reliance on groundwater. For example, canal leakage, evaporation loss, and inappropriate irrigation timing lead to agricultural water waste rates exceeding 40%. Existing technologies often focus on optimizing single water distribution efficiency or costs, lacking a systematic design for water-saving objectives and struggling to dynamically adjust based on real-time environmental data. Therefore, there is an urgent need for a multi-objective optimization method centered on water conservation, reducing water waste throughout the entire process from source design to dynamic scheduling. Summary of the Invention

[0003] The purpose of this invention is to provide an irrigation system design and dynamic water allocation optimization method with agricultural water conservation as its core, which uses artificial intelligence technology to achieve precise allocation of water resources, minimize irrigation water consumption, and ensure crop growth needs.

[0004] (1) Full-chain water-saving optimization: Water-saving constraints are embedded throughout the entire process, from water system design (reducing water loss), water source allocation (multi-source coordination), to irrigation execution (precise water allocation on demand).

[0005] (2) Dynamic simulation of crop phenology in farmland: predict meteorological conditions, soil moisture, etc., and combine them with the natural growth model of crops to model the phenology of farmland, determine the ideal growth state, and make dynamic adjustments to artificial irrigation accordingly.

[0006] (3) Groundwater protection mechanism: Introduce reclaimed water for agricultural irrigation, increase the utilization rate of reclaimed water in the optimization algorithm, and replace groundwater extraction.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] A method for designing and dynamically optimizing water allocation in farmland water-saving irrigation systems based on multi-source data fusion of supply and demand, the method comprising the following steps:

[0009] S1 “Crop-Soil-Meteorology” Multi-Source Data Coupling Modeling

[0010] S2 water-saving water system design optimization

[0011] S3 dynamic water-saving distribution strategy

[0012] The S4 intelligent terminal control system is used for control.

[0013] Step S1 is as follows:

[0014] Input key water-saving parameters such as regional soil moisture, crop water use efficiency (WUE), root depth, and evapotranspiration coefficient; construct a crop-soil-meteorological coupled water-saving demand model to quantify the minimum ecological water requirement and the maximum allowable water shortage threshold. The implementation process is as follows:

[0015] 1.1 Multi-source data monitoring,

[0016] Multiple sensors (such as soil moisture sensors, weather sensors, and crop monitoring sensors) are deployed throughout the field to monitor data such as soil moisture, temperature, weather conditions, and crop water requirements in real time. Specifically: Intelligent soil moisture sensors: These are used to monitor the moisture content of different soil layers in real time. The sensors can be distributed at key locations in the field and transmit data to the central control system via a wireless network.

[0017] Sensor density calculation: A sensor placement optimization method based on an improved adaptive PSO algorithm. Compared with traditional methods, the numerical solution of the soil moisture transport equation (Richards equation) is used as the particle search direction correction term, and the crop root weight distribution matrix is ​​dynamically adjusted through NDVI time series data. This method includes multi-objective fitness function reconstruction and the construction of a three-dimensional evaluation index by integrating soil property spatial variation coefficient (CV), crop root distribution weight (RW), and meteorological influence factor (MF).

[0018] Fitness=α·CV+β·RW+γ·MF+λ·Cost

[0019] In the formula, Fitness is the fitness value; α, β, and γ are dynamic weighting coefficients; λ is the economic constraint factor; and Cost is a comprehensive economic constraint term used to quantify the implementation cost of the sensor deployment scheme.

[0020] Secondly, farmland-specific improvement strategies are employed, including hierarchical particle coding (separating the encoding of two-dimensional planar coordinates and burial depth, employing a hybrid discrete-continuous search space), the introduction of mutation operators (triggering Gaussian mutation based on the standard deviation of crop ridge width when particles are trapped in local optima), and an energy-constrained model (integrating wireless signal attenuation equations to calculate communication energy costs). All sensor-collected data is uploaded to the data terminal control system via wireless communication technologies (such as LoRa and NB-IoT) to ensure timely data updates and remote monitoring. The transmitted real-time data supports dynamic monitoring of changes in the field environment.

[0021] 1.2 Construct the model architecture, as follows:

[0022] 1.2.1 Soil moisture transport modeling, unsaturated zone flow model (improved Richards equations),

[0023]

[0024] θ: Soil moisture content; t: Time; z: Depth; K(θ): Unsaturated hydraulic conductivity K sat θ represents the saturated hydraulic conductivity, b represents the gain and offset coefficients calculated using a ground calibration plate, and h(θ) represents the soil water head. h e For intake pressure head; S root (z,t): Root water absorption rate, S root (z,t)=a(z)·T p (t), a(z) are the root density distribution functions, T p (t) represents the potential transpiration rate, used to determine the most unfavorable point (defined by the water transport lag index).

[0025]

[0026] WTLI: Time Deficit Water Index; θ crit θ(t): Critical soil moisture content; θ(t): Function of actual soil moisture content over time t; t c The time period for observation or evaluation; Within the time window [0, t] c Within ], the cumulative deficit of soil moisture content below the critical value; θ crit ·t c Theoretical maximum deficit, 1.2.2 Crop Response Module

[0027] A coupling algorithm for the Jarvis stomatal conductance model and the Feddes root uptake function was constructed.

[0028] For the Jarvis porosity conductance model:

[0029] g s =g smax ·f(Q)·f(Dq)·f(ψ s )·f(T)

[0030] In the formula, f(Q) is the photosynthetically active radiation response function; f(Dq) is the water vapor pressure difference suppression function; f(ψ) is the water vapor pressure difference suppression function. s ) is the soil water potential response function; f(T) is the temperature regulation function, and for Feddes, the root water uptake function is:

[0031] S (z) =s(ψ s )·Tp ·La (z)

[0032] In the formula, S (z) s(ψ) represents the amount of water absorbed by the roots at depth z per unit volume of soil. s ) represents the water stress coefficient; T p L represents the potential transpiration rate. (z) Let be the root density distribution function.

[0033]

[0034] In the formula, ψ s ψ1 represents soil water potential; ψ2 represents the anaerobic threshold; ψ3 represents the optimal upper limit of water uptake; and ψ4 represents the starting point of linear descent.

[0035]

[0036] In the formula, La (z) z is the root density distribution function; s The root distribution half-life depth is 20cm to 100cm; z is the depth; m is the distribution shape parameter, where m=1 indicates a slow, linear decrease in root density with decreasing depth, and m=5 indicates a sharp decrease in root density with decreasing depth, with roots concentrated in the surface layer.

[0037] 1.2.3 Meteorological Driven Module

[0038] The Penman-Monteith dual-source model was used to distinguish the contributions of soil / vegetation evapotranspiration.

[0039] Vegetation transpiration:

[0040]

[0041] In the formula, ET veg R represents vegetation transpiration; Δ represents the slope of the saturated vapor pressure-temperature curve; n,v Net radiation of the canopy; ρ a c is the air density. p is the specific heat of air at constant pressure, and Dq is the air vapor pressure deficit. For canopy aerodynamic drag; ∈ represents the total stomatal resistance of the canopy; ∈ represents the wet / dry constant.

[0042] Soil evaporation:

[0043]

[0044] In the formula, ET soil ρ represents soil evaporation; Δ represents the slope of the saturated vapor pressure-temperature curve; a c is the air density.p R is the specific heat at constant pressure of air, Dq is the air vapor pressure deficit; n,s Net radiation at the soil surface; For soil surface aerodynamic drag; r soil ∈ represents soil evaporation resistance; ∈ represents the wet / dry surface constant.

[0045] 1.2.4 Establish the dynamic response function of water stress and photosynthetic product allocation.

[0046] Water stress index:

[0047]

[0048] In the formula, WSI is the water stress index; ET act This refers to the actual evapotranspiration rate; ET p Total photosynthetic products are corrected for potential transpiration.

[0049]

[0050] In the formula, P total Correction for total photosynthetic products; P max For maximum photosynthetic capacity (without stress); g s This represents the actual porosity; g smax This represents the maximum porosity.

[0051] In the above scheme,

[0052] A dynamic feedback mechanism for the SPAC system is constructed by coupling multiple processes of water transport, stomatal response, and dual-source evapotranspiration in the unsaturated zone. The Richards equation is improved by introducing a microbial activity correction term to enhance the accuracy of water critical point identification. The Jarvis-Feddes simultaneous algorithm is used to achieve coordinated simulation of root water uptake and stomatal oscillation. The water stress index and photosynthetic product allocation function are integrated to reveal the threshold response law of carbon-water coupling. Compared with traditional single-process models, the system error is significantly reduced, and minute-level water redistribution processes can be analyzed.

[0053] The S2 water-saving water system design optimization is detailed below.

[0054] 2.1 Constructing the System Framework: Topographic Data: Based on the Digital Elevation Model (DEM), topographic slope and runoff network are extracted to drive gradient optimization of gravity water conveyance paths;

[0055] Soil data: Calculate channel seepage rate and regional infiltration loss using the spatial distribution of soil permeability coefficient; Crop data: Divide into differentiated irrigation scheduling units based on the spatial heterogeneity of crop type and water requirement patterns during growth period.

[0056] Meteorological data: By combining the interpolation results of rainfall isohyets, the potential for regional rainwater resource utilization is quantified to guide the spatial layout of rainwater harvesting facilities;

[0057] 2.2 Multi-objective optimization: A multi-objective water-saving irrigation optimization model is established with the objectives of minimizing water conveyance loss, maximizing irrigation uniformity, and maximizing reclaimed water utilization as the main objectives, as detailed below:

[0058] (1) Objective function for water conveyance loss (reducing leakage and evaporation through channel seepage prevention design and pipeline transformation):

[0059]

[0060] In the formula, N c Total number of channel segments; L i Let q be the length of the i-th channel segment; i Let i be the leakage rate per unit length of the i-th channel segment; N represents the channel seepage prevention efficiency coefficient. p P represents the total number of pipeline segments. j E represents the surface area of ​​the j-th open-air pipeline segment. j Evaporation rate per unit area; This is a correction factor for pipeline coverage.

[0061] q i =K s,i ·W i ·H i

[0062] In the formula, q i K represents the leakage rate per unit length of the i-th channel segment; s,i W is the saturated permeability coefficient of the soil below the i-th segment of the channel; i H is the bottom width of the channel; i Design the water depth for the channel.

[0063] P j =π·D j ·L j

[0064] In the formula, P j D is the surface area of ​​the j-th open-air pipeline; j L is the pipe diameter. j This represents the length of the pipe.

[0065]

[0066] In the formula, Δe is the wind speed correction factor. j P represents the air saturation vapor pressure difference; atm Atmospheric pressure

[0067] (2) Objective function for irrigation uniformity (using drip irrigation and sprinkler irrigation zone design to reduce local over-irrigation):

[0068]

[0069] constraint:

[0070]

[0071] In the formula, n is the total number of irrigation units; d k Water supply for the k-th irrigation unit; μ d This represents the average water supply per irrigation unit. ET c Crop water requirements; A total For the total irrigated area,

[0072] (3) Objective function for reclaimed water utilization (integrating rainwater harvesting systems and greywater reuse facilities to replace groundwater extraction):

[0073]

[0074] In the formula, η r For rainwater harvesting efficiency, D total D represents the total irrigation water demand. total =ET c ·A total ·t,V r For the volume of the rainwater collection tank; U w This represents the daily treatment capacity of reclaimed water.

[0075] Ic represents the irrigation cycle; A c Dr is the catchment area; Dr is the design rainfall; Rc is the runoff coefficient.

[0076] In the above scheme,

[0077] For the first time, GIS spatial heterogeneity analysis and multi-objective optimization algorithms are coupled. By driving water system topology optimization through DEM and collaborative modeling of multi-source data of soil, crop and rainfall, Pareto frontier search is achieved for water conveyance loss (Darcy seepage-Penman evaporation joint quantification), irrigation uniformity (Christiansen coefficient constraint) and reclaimed water utilization (rainwater-grey water system integration). This solves the problem of traditional single-objective optimization ignoring explicit spatial constraints and resource coordination, and significantly improves water-saving efficiency.

[0078] The S3 dynamic water-saving distribution strategy is detailed below.

[0079] 3.1 Soil Moisture Prediction (Lag-PhysFusion):

[0080] (1) Historical time series data:

[0081]

[0082] Input feature: X t =[H t ,L t W t ,I t V t ] T

[0083] In the formula, X t S is the input feature vector at time t; t H represents the measured soil moisture value at time t; T is the total length of the time series, i.e., the maximum number of time steps in the historical data; H t To improve the soil moisture calculation (stratified data, dimension Da) using the Richards equation; L t W is the lag index (a scalar quantity reflecting the delayed effect of water transport); t Meteorological data (temperature, precipitation, evaporation, etc.); I t V represents irrigation volume; t S is the vegetation index. t To measure soil moisture, the output value (S) is... t ∈R).

[0084] (2) Physical feature extraction:

[0085] Physical dynamics can be extracted by improving the discrete solutions of the Richards equations.

[0086]

[0087] Discrete form (finite difference method):

[0088]

[0089] In the formula, SW represents soil water potential; t represents time; K(SW) represents unsaturated hydraulic conductivity; zz represents vertical spatial coordinates; YH represents source and sink terms; SW t (d) Let d be the soil moisture at time t; Δt be the time step; Δzz be the soil moisture of the d-th layer. 2 The spatial step size; Let be the hydraulic conductivity of the dd-th layer at time t; For the source and sink terms of the d-th layer at time t, the physical model output and the hysteresis exponent are fused into an enhanced feature:

[0090]

[0091] In the formula, The vector is the encoded physical feature; ReLU is the activation function; [H t L t ] represents the concatenation of input features; W p b is the weight matrix; p For bias vectors,

[0092] (3) Lag perception time encoding (the lag index L) t (Dynamically embedded time position encoding):

[0093]

[0094] In the formula, EB t Standard Transformer position encoding; W L L is a learnable parameter used to adjust the degree of time distortion of the lag effect. t It is a lagging index.

[0095] (4) Spatiotemporal attention mechanism:

[0096]

[0097] In the formula, C X J is the query matrix; Z is the key matrix; M is the value matrix; L This is a lag similarity matrix. τ is a temperature parameter that controls the sensitivity to hysteresis differences. (LZ) i LZ j The lagging index; d k The dimension of the key vector.

[0098] (5) Multi-scale temporal convolution:

[0099]

[0100] In the formula, The output features of the k-th convolutional layer at time t; Let be the s-th weight parameter of the k-th convolutional layer; For time step ts×d (k) The input feature vector; d (k) S is the dilation factor of the k-th layer; S is the time step coverage of the convolution kernel.

[0101] (6) Dynamic fusion prediction:

[0102] Combining physical characteristics P t With time series features C t :

[0103]

[0104] Fusion weights ξ t is dynamically generated by the lagging exponent:

[0105] ξ t =σ(w ξ ZH t +b ξ )

[0106] In the formula, For predicted output; ξ t For dynamic fusion weights; f phy (P t ) represents the physical model branch; f seq (C t ) represents the sequence model branch; ZH t The lag exponent is σ; σ is the Sigmoid function, which controls the contribution ratio between the physical model and the data-driven model; w ξ b is the weight parameter; ξ For bias parameters,

[0107] 3.2 Annual Rainfall Forecast (MPF-Net Model):

[0108] (1) Input data:

[0109] Target variable: Y t ∈R y (Historical annual rainfall sequence, y represents the year);

[0110] Climate factor: X t ∈R (y×m) (m climate indices, such as ENSO, PDO, AMO, etc.);

[0111] Geographical features: G∈R (p) (p static geographic parameters, such as latitude and longitude, elevation, etc.).

[0112] (2) Wavelet multi-scale decomposition: for Y t Perform Mallat wavelet decomposition:

[0113]

[0114] In the formula, Y t The target variable; For the j-th scale detail component (capturing interannual variability); A J (t) represents the approximate component (characterizing the long-term trend); J represents the number of decomposition layers.

[0115] (3) Physical constraint feature engineering:

[0116] Climate system energy constraints (introducing the mass-energy conservation equation as a regularization term):

[0117]

[0118] In the formula, λ represents the climate system energy; v represents the atmospheric vertical velocity field; HD represents the water vapor flux divergence; λ1 and λ2 are learnable constraint coefficients.

[0119] Geographical features embedded:

[0120] Γ=ReLU(W g G+b g )⊙Sigmoid(W t Ω+b t )

[0121] In the formula, W g G, W t T is the learnable weight matrix; Ω is the time-encoded vector.

[0122] (4) Multimodal spatiotemporal attention mechanism:

[0123] Cross-attention to climate factors: In the formula, Q = W q [A J (t); Γ]; K = W k [X t ;G];V=W v [X t ;G],

[0124] In the formula, Attn Q The output attention feature reflects the temporal effect of climate factors on rainfall variability; O is the dynamic query matrix. The wavelet detail components represent the interannual variability signal; d k V is the scaling factor; VZ is the value matrix; A J (t) represents the wavelet approximation component; Γ is the geographic modulation matrix; W q G is the learnable weight matrix; G is the geographic feature vector; W k W v For learnable weight matrix,

[0125] Time-frequency dual-domain gating:

[0126] g t =σ(W g [D j (t); Attn Q ])

[0127]

[0128] In the formula, g t σ is the gate vector; σ is the Sigmoid activation function; W gFor learnable gated weight matrix; To refine the detailed components, data-driven features and physical constraints are integrated.

[0129] (5) Adaptive learning of mutation points:

[0130] Constructing a change probability model for Bayesian change detection:

[0131]

[0132] In the formula, c t ∈{0,1} represents a variable indicating a change point; h t In hidden state; W c h is the learnable weight matrix. t-1 h t Let be the hidden state vectors of adjacent time steps; σ is the Sigmoid function.

[0133] Dynamic parameter adjustment:

[0134]

[0135] In the formula, MO is the key parameter of the model, enabling abrupt adaptation of the parameter space; MO t These are the key parameters of the model at the current time step; MO0 is the initial parameter value; ΔMO is the parameter adjustment step size; c i The cumulative sum of the indicator variables for historical turning points.

[0136] (6) Physical information neural network architecture:

[0137]

[0138] In the formula, This represents the prediction error; For the KL divergence regularization term in change point detection; Loss due to physical constraints; For adaptive weight parameters,

[0139] (7) Uncertainty Quantification (Monte Carlo Physics Dropout):

[0140] During prediction, dropout is kept on, and N samples are performed:

[0141]

[0142] In the formula, CN represents the model prediction output for the i-th Monte Carlo sampling; CN is the number of samplings. σ represents the network parameters after randomly discarding some neurons during the i-th sampling; 2 To predict variance,

[0143] 3.3 Short-term rainfall forecast (MDSTF model):

[0144] (1) Select key indicators and observe them: air humidity, temperature, wind speed, air pressure, etc.; (2) Decompose the time series data of each indicator:

[0145] 3.4 Simulation of Ideal Crop Growth State

[0146] Based on predictions of future rainfall and soil moisture, an innovative framework integrating crop growth mechanism models, temporal deep learning, and dynamic feedback optimization is proposed to predict the ideal crop growth state under given environmental conditions. This framework is named Hybrid Crop Growth Digital Twin (HCG-DT).

[0147] This innovative method combines deep reinforcement learning-based climate prediction optimization with real-time dynamics of crop growth, achieving closed-loop water irrigation decision-making across multiple time scales. Compared to traditional static water allocation or single-time-scale methods, it significantly improves the overall efficiency of water resource utilization and groundwater sustainability while ensuring crop yield. This method primarily proposes the Lag-PhysFusion model, MPF-Net model, MDSTF model, HCG-DT model, and deep reinforcement learning DRL model to predict soil moisture, weather, and ideal crop growth conditions, demonstrating significant effectiveness. The S4 intelligent terminal control system provides control: UAVs periodically fly to monitor the field, capturing real-time images of crops and uploading the data to the intelligent control terminal for unified decision-making. Based on real-time data, prediction results, and the calculation results of the irrigation model, the terminal controller controls the intelligent sprinkler irrigation equipment, ensuring the accuracy of irrigation operations.

[0148] The algorithm in this control terminal includes using image analysis to determine the crop growth status in each area; using soil moisture transport to determine the soil moisture content in each area; and combining future rainfall information to make an overall judgment on multiple data points and control smart valves in various locations for irrigation.

[0149] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned method for designing and dynamically optimizing farmland water-saving irrigation systems based on the fusion of multi-source supply and demand data.

[0150] A computer-readable storage medium storing computer instructions thereon, which, when executed by a processor, implement the aforementioned method for designing and dynamically optimizing water allocation for farmland water-saving irrigation systems by fusing supply and demand data from multiple sources.

[0151] Compared with existing technologies, the advantages of this invention are as follows: The dynamic water distribution irrigation system proposed in this application achieves theoretical innovation and technological breakthroughs in precision irrigation through the deep integration of multi-physics field coupled modeling and multi-scale intelligent decision-making. The system uses an improved Richards equation combined with a Lag-PhysFusion model to describe non-equilibrium water transport, and corrects the parameterization bias of unsaturated hydraulic conductivity through a hysteresis kernel function, significantly reducing soil moisture prediction errors compared to traditional methods. On this basis, a Jarvis-Feddes joint model is constructed, coupling the dynamic response of stomatal conductance with the heterogeneity of root water uptake, and introducing a dynamic response function for photosynthetic product allocation to quantify the impact of water stress, thus significantly improving transpiration efficiency. By separating the soil / vegetation evapotranspiration contribution through a Penman-Monteith dual-source model, and combining the canopy temperature-leaf area index features (CNN classification accuracy of 92.4%) from UAV multispectral image analysis, a crop growth digital twin (HCG-DT model) is established, achieving accurate characterization of evapotranspiration calculation error ≤0.3mm / d. To address the need for multi-timescale regulation, an MPF-Net model is used to predict annual rainfall types, driving deep reinforcement learning (DRL) to generate long-term water-saving strategies. Multi-objective optimization reduces water transport losses by over 20% and achieves a reclaimed water utilization rate of over 80%. Simultaneously, an MDSTF model is established to achieve high-precision short-term rainfall forecasting (TS score above 0.8). Combined with model predictive control (MPC), a rolling optimization mechanism is constructed, compressing irrigation decision response time based on real-time soil moisture data and a worst-case scenario identification algorithm. The closed-loop control system achieves real-time field-level regulation (delay <1s) through edge computing. Three years of field validation show that the system significantly improves water-saving efficiency, yield increase, water use efficiency, and irrigation uniformity coefficient compared to traditional irrigation methods. This solution, through multi-dimensional collaboration of mechanistic models and data-driven approaches, overcomes the challenge of irrigation decision lag caused by spatiotemporal heterogeneity, providing a verifiable engineering paradigm for smart agriculture. Attached Figure Description

[0152] Figure 1 This is a schematic diagram of the operating mechanism of the intelligent control terminal. Detailed Implementation

[0153] To enhance understanding of the present invention, the embodiments will be described in detail below with reference to the accompanying drawings.

[0154] Example 1: See Figure 1 A method for designing and dynamically optimizing water allocation for farmland water-saving irrigation systems based on multi-source data fusion of supply and demand, the method comprising the following steps:

[0155] S1 “Crop-Soil-Meteorology” Multi-Source Data Coupling Modeling

[0156] Input key water-saving parameters such as regional soil moisture, crop water use efficiency (WUE), root depth, and evapotranspiration coefficient; construct a water-saving demand model coupled with "crop-soil-meteorology" to quantify the minimum ecological water requirement and the maximum allowable water shortage threshold.

[0157] The details are as follows:

[0158] 1.1 Multi-source data monitoring

[0159] Multiple sensors (such as soil moisture sensors, weather sensors, and crop monitoring sensors) are deployed throughout the field to monitor data such as soil moisture, temperature, weather conditions, and crop water requirements in real time. Specifically: Intelligent soil moisture sensor: Used to monitor the moisture content of different soil layers in the field in real time. The sensor can be distributed at key locations in the field and transmits data to the central control system via a wireless network.

[0160] Meteorological data acquisition equipment includes temperature and humidity sensors, wind speed and direction sensors, and air pressure sensors. It collects high spatiotemporal resolution satellite precipitation data, three-dimensional distribution of solar radiation, and canopy microclimate correction coefficients in real time to provide accurate environmental data.

[0161] Crop growth monitoring sensors: Through plant growth monitors, data such as crop growth status and leaf humidity are collected in real time to help determine the crop's water requirements.

[0162] Soil moisture sensors: monitor changes in soil moisture, especially in deep soil, to further improve the accuracy of irrigation decisions.

[0163] Sensor density calculation: A sensor placement optimization method based on an improved adaptive PSO algorithm is proposed. Compared with traditional methods, this method uses the numerical solution of the soil moisture transport equation (Richards equation) as a correction term for the particle search direction and dynamically adjusts the crop root weight distribution matrix using NDVI time series data. This method includes multi-objective fitness function reconstruction and the construction of a three-dimensional evaluation index by fusing soil property spatial variation coefficient (CV), crop root distribution weight (RW), and meteorological influence factors (MF).

[0164] Fitness=α·CV+β·RW+γ·MF+λ·Cost

[0165] In the formula, Fitness is the fitness value; α, β, and γ are dynamic weighting coefficients; λ is the economic constraint factor; and Cost is the comprehensive economic constraint term, used to quantify the implementation cost of the sensor deployment scheme.

[0166] Secondly, there are farmland-specific improvement strategies, including hierarchical particle coding (separating the two-dimensional plane coordinates from the burial depth and using a hybrid discrete-continuous search space), the introduction of mutation operators (when a particle gets stuck in a local optimum, Gaussian mutation is triggered according to the standard deviation of the crop ridge width), and energy consumption constraint models (integrating the wireless signal attenuation equation to calculate the communication energy consumption cost).

[0167] All data collected by the sensors is uploaded to the data terminal control system via wireless communication technologies (such as LoRa, NB-IoT, etc.) to ensure timely data updates and remote monitoring. The transmitted real-time data can support the dynamic monitoring of changes in the field environment.

[0168] 1.2 Model Architecture

[0169] 1.2.1 Soil Moisture Transport Modeling

[0170] Unsaturated zone flow model (improved Richards equations)

[0171]

[0172] θ: Soil moisture content; t: Time; z: Depth; K(θ): Unsaturated hydraulic conductivity K sat θ represents the saturated hydraulic conductivity, b represents the gain and offset coefficients calculated using a ground calibration plate, and h(θ) represents the soil water head. h e For intake pressure head; S root (z,t): Root water absorption rate, S root (z,t)=a(z)·T p (t), a(z) are the root density distribution functions, T p (t) represents the potential transpiration rate. Determining the most unfavorable point (defining the water transport lag index).

[0173]

[0174] WTLI: Time Deficit Water Index; θ crit θ(t): Critical soil moisture content; θ(t): Function of actual soil moisture content over time t; t c The time period for observation or evaluation; Within the time window [0, t] c Within ], the cumulative deficit of soil moisture content below the critical value; θ crit ·t c Theoretical maximum deficit.

[0175] 1.2.2 Crop Response Module

[0176] A coupling algorithm was constructed between the Jarvis stomatal conductance model and the Feddes root water uptake function.

[0177] For the Jarvis porosity conductance model:

[0178] g s =g smax ·f(Q)·f(Dq)·f(ψ s )·f(T)

[0179] In the formula, f(Q) is the photosynthetically active radiation response function; f(Dq) is the water vapor pressure difference suppression function; f(ψ) is the water vapor pressure difference suppression function. s ) is the soil water potential response function; f(T) is the temperature regulation function.

[0180] For Feddes root uptake function:

[0181] S (z) =s(ψ s )·T p ·La (z)

[0182] In the formula, S (z) s(ψ) represents the amount of water absorbed by the roots at depth z per unit volume of soil. s ) represents the water stress coefficient; T p L represents the potential transpiration rate. (z) is the root density distribution function.

[0183]

[0184] In the formula, ψ s ψ1 represents soil water potential; ψ2 represents the anaerobic threshold; ψ3 represents the optimal water absorption limit; and ψ4 represents the starting point of linear descent.

[0185]

[0186] In the formula, La (z) z is the root density distribution function; s denoted as root distribution half-life depth, ranging from 20cm to 100cm; z represents depth; m is the distribution shape parameter, where m=1 indicates a slow, linear decrease in root density with decreasing depth, and m=5 indicates a sharp decrease in root density with decreasing depth, with roots concentrated in the surface layer.

[0187] 1.2.3 Meteorological Driven Module

[0188] The Penman-Monteith dual-source model was used to distinguish the contributions of soil / vegetation evapotranspiration.

[0189] Vegetation transpiration:

[0190]

[0191] In the formula, ET veg R represents vegetation transpiration; Δ represents the slope of the saturated vapor pressure-temperature curve; n,v Net radiation of the canopy; ρ a c is the air density. p is the specific heat of air at constant pressure, and Dq is the air vapor pressure deficit. For canopy aerodynamic drag; denoted as the total stomatal resistance of the canopy; ∈ represents the wet / dry constant.

[0192] Soil evaporation:

[0193]

[0194] In the formula, ET soil ρ represents soil evaporation; Δ represents the slope of the saturated vapor pressure-temperature curve; a c is the air density. p R is the specific heat at constant pressure of air, Dq is the air vapor pressure deficit; n,s Net radiation at the soil surface; For soil surface aerodynamic drag; r soil ∈ represents soil evaporation resistance; ∈ represents the wet / dry constant.

[0195] 1.2.4 Establishing the dynamic response function of water stress and photosynthetic product allocation

[0196] Water stress index:

[0197]

[0198] In the formula, WSI is the water stress index; ET act This refers to the actual evapotranspiration rate; ET p This represents potential transpiration.

[0199] Total photosynthetic products correction:

[0200]

[0201] In the formula, P total Correction for total photosynthetic products; P max For maximum photosynthetic capacity (without stress); g s This represents the actual porosity; g smax This represents the maximum porosity.

[0202] Improvements and advantages:

[0203] A dynamic feedback mechanism for the SPAC system is constructed by coupling multiple processes of water transport, stomatal response, and dual-source evapotranspiration in the unsaturated zone. The Richards equation is improved by introducing a microbial activity correction term to enhance the accuracy of water critical point identification. The Jarvis-Feddes simultaneous algorithm is used to achieve coordinated simulation of root water uptake and stomatal oscillation. The water stress index and photosynthetic product allocation function are integrated to reveal the threshold response law of carbon-water coupling. Compared with traditional single-process models, the system error is significantly reduced, and minute-level water redistribution processes can be analyzed.

[0204] S2 water-saving water system design optimization

[0205] 2.1 System Framework Topographic Data: Based on the digital elevation model (DEM), topographic slope and runoff network are extracted to drive gradient optimization of gravity water conveyance path;

[0206] Soil data: Calculate channel seepage rate and regional infiltration loss using the spatial distribution of soil permeability coefficient; Crop data: Divide into differentiated irrigation scheduling units based on the spatial heterogeneity of crop type and water requirement patterns during growth period.

[0207] Meteorological data: By combining the interpolation results of rainfall isohyets, the potential for regional rainwater resource utilization is quantified, guiding the spatial layout of rainwater harvesting facilities.

[0208] 2.2 Multi-objective optimization

[0209] A multi-objective water-saving irrigation optimization model is established with the objectives of minimizing water conveyance loss, maximizing irrigation uniformity, and maximizing reclaimed water utilization, as detailed below:

[0210] (1) Objective function for water conveyance loss (reducing leakage and evaporation through channel seepage prevention design and pipeline transformation):

[0211]

[0212] In the formula, N c Total number of channel segments; L i Let q be the length of the i-th channel segment; i Let i be the leakage rate per unit length of the i-th channel segment; N represents the channel seepage prevention efficiency coefficient. p P represents the total number of pipeline segments. j E represents the surface area of ​​the j-th open-air pipeline segment. j Evaporation rate per unit area; This is the pipeline coverage correction factor.

[0213] q i =K s,i ·W i ·H i

[0214] In the formula, qi K represents the leakage rate per unit length of the i-th channel segment; s,i W is the saturated permeability coefficient of the soil below the i-th segment of the channel; i H is the bottom width of the channel; i Design the water depth for the channel.

[0215] P j =π·D j ·L j

[0216] In the formula, P j D is the surface area of ​​the j-th open-air pipeline; j L is the pipe diameter. j This represents the length of the pipe.

[0217]

[0218] In the formula, Δe is the wind speed correction factor. j P represents the air saturation vapor pressure difference; atm Atmospheric pressure.

[0219] (2) Objective function for irrigation uniformity (using drip irrigation and sprinkler irrigation zone design to reduce local over-irrigation):

[0220]

[0221] constraint:

[0222]

[0223] In the formula, n is the total number of irrigation units; d k Water supply for the k-th irrigation unit; μ d This represents the average water supply per irrigation unit. ET c Crop water requirements; A total This represents the total irrigated area.

[0224] (3) Objective function for reclaimed water utilization (integrating rainwater harvesting systems and greywater reuse facilities to replace groundwater extraction):

[0225]

[0226] In the formula, η r To improve rainwater harvesting efficiency, D total D represents the total irrigation water demand. total =ET c ·A total ·t,V r For the volume of the rainwater collection tank; U w This represents the daily treatment capacity of reclaimed water.

[0227] Ic represents the irrigation cycle; A c Dr is the catchment area; Dr is the design rainfall; Rc is the runoff coefficient.

[0228] Improvements and advantages:

[0229] For the first time, GIS spatial heterogeneity analysis and multi-objective optimization algorithms are coupled. By driving water system topology optimization through DEM and collaborative modeling of multi-source data of soil, crop and rainfall, Pareto frontier search is achieved for water conveyance loss (Darcy seepage-Penman evaporation joint quantification), irrigation uniformity (Christiansen coefficient constraint) and reclaimed water utilization (rainwater-grey water system integration). This solves the problem of traditional single-objective optimization ignoring explicit spatial constraints and resource coordination, and significantly improves water-saving efficiency.

[0230] S3 Dynamic Water-Saving Distribution Strategy

[0231] Long-term rainfall forecasting is conducted to determine the dry, normal, and wet season conditions for the planned year, and the planning scheme is adjusted accordingly. Combined with soil moisture and meteorological forecasts, simulations are performed to depict the ideal crop growth state under given conditions. Based on real-time soil moisture and short-term rainfall forecasts, irrigation schemes are adjusted in advance, and irrigation duration and water volume are dynamically calculated to achieve "on-demand drip irrigation." Water allocation strategies are optimized, and the reward function includes:

[0232] Minimize water consumption per unit of output;

[0233] Groundwater extraction is below the sustainable threshold;

[0234] Maximize the utilization rate of rainwater resources.

[0235] 3.1 Soil Moisture Prediction (Lag-PhysFusion Model):

[0236] (1) Historical time series data:

[0237]

[0238] Input feature: X t =[H t ,L t W t ,I t V t ] T

[0239] In the formula, X t S is the input feature vector at time t; t H represents the measured soil moisture value at time t; T is the total length of the time series, i.e., the maximum number of time steps in the historical data; H tTo improve the soil moisture calculation (stratified data, dimension Da) using the Richards equation; L t W is the lag index (a scalar quantity reflecting the delayed effect of water transport); t Meteorological data (temperature, precipitation, evaporation, etc.); I t V represents irrigation volume; t S is the vegetation index. t To measure soil moisture, the output value (S) is... t ∈R).

[0240] (2) Physical feature extraction:

[0241] Physical dynamics can be extracted by improving the discrete solutions of the Richards equations.

[0242]

[0243] Discrete form (finite difference method):

[0244]

[0245] In the formula, SW represents soil water potential; t represents time; K(SW) represents unsaturated hydraulic conductivity; zz represents vertical spatial coordinates; YH represents source and sink terms; SW t (d) Let d be the soil moisture at time t; Δt be the time step; Δzz be the soil moisture of the d-th layer. 2 The spatial step size; Let be the hydraulic conductivity of the dd-th layer at time t; Let be the source and sink terms of the d-th layer at time t. The physical model output and the hysteresis exponent are fused into an enhanced feature:

[0246]

[0247] In the formula, The vector is the encoded physical feature; ReLU is the activation function; [H t L t ] represents the concatenation of input features; W p b is the weight matrix; p This is the bias vector.

[0248] (3) Lag perception time encoding (the lag index L) t (Dynamically embedded time position encoding):

[0249]

[0250] In the formula, EB t Standard Transformer position encoding; W L L is a learnable parameter used to adjust the degree of time distortion of the lag effect.t It is a lagging index.

[0251] (4) Spatiotemporal attention mechanism:

[0252]

[0253] In the formula, C X J is the query matrix; Z is the key matrix; M is the value matrix; L This is a lag similarity matrix. τ is a temperature parameter that controls the sensitivity to hysteresis differences. (LZ) i LZ j The lagging index; d k The dimension of the key vector.

[0254] (5) Multi-scale temporal convolution:

[0255]

[0256] In the formula, The output features of the k-th convolutional layer at time t; Let be the s-th weight parameter of the k-th convolutional layer; For time step ts×d (k) The input feature vector; d (k) is the expansion factor of the k-th layer; S is the time step coverage of the convolution kernel.

[0257] (6) Dynamic fusion prediction:

[0258] Combining physical characteristics P t With time series features C t :

[0259]

[0260] Fusion weight ξ t Dynamically generated by the lag index:

[0261] ξ t =σ(w ξ ZH t +b ξ )

[0262] In the formula, For predicted output; ξ t For dynamic fusion weights; f phy (P t ) represents the physical model branch; f seq (C t ) represents the sequence model branch; ZH tThe lag exponent is σ; σ is the Sigmoid function, which controls the contribution ratio between the physical model and the data-driven model; w ξ b is the weight parameter; ξ This is the bias parameter.

[0263] 3.2 Annual Rainfall Forecast (MPF-Net Model):

[0264] (1) Input data:

[0265] Target variable: Y t ∈R y (Historical annual rainfall sequence, y represents the year);

[0266] Climate factor: X t ∈R (y×m) (m climate indices, such as ENSO, PDO, AMO, etc.);

[0267] Geographical features: G∈R (p) (p static geographic parameters, such as latitude and longitude, elevation, etc.).

[0268] (2) Wavelet multi-scale decomposition: for Y t Perform Mallat wavelet decomposition:

[0269]

[0270] In the formula, Y t The target variable; For the j-th scale detail component (capturing interannual variability); A J (t) represents the approximate component (characterizing the long-term trend); J represents the number of decomposition layers.

[0271] (3) Physical constraint feature engineering:

[0272] Climate system energy constraints (introducing the mass-energy conservation equation as a regularization term):

[0273]

[0274] In the formula, λ represents the climate system energy; v represents the atmospheric vertical velocity field; HD represents the water vapor flux divergence; λ1 and λ2 are learnable constraint coefficients.

[0275] Geographical features embedded:

[0276] Γ=ReLU(W g G+b g )⊙Sigmoid(W t Ω+b t )

[0277] In the formula, W gG, W t T is the learnable weight matrix; Ω is the time-encoding vector.

[0278] (4) Multimodal spatiotemporal attention mechanism:

[0279] Cross-attention to climate factors: In the formula, Q = W q [A J (t); Γ]; K = W k [X t ;G];V=W v [X t ;G].

[0280] In the formula, Attn Q The output attention feature reflects the temporal effect of climate factors on rainfall variability; O is the dynamic query matrix. The wavelet detail components represent the interannual variability signal; d k V is the scaling factor; VZ is the value matrix; A J (t) represents the wavelet approximation component; Γ is the geographic modulation matrix; W q G is the learnable weight matrix; G is the geographic feature vector; W k W v This is a learnable weight matrix.

[0281] Time-frequency dual-domain gating:

[0282] g t =σ(W g [D j (t); Attn Q ])

[0283]

[0284] In the formula, g t σ is the gate vector; σ is the Sigmoid activation function; W g For learnable gated weight matrix; For the corrected detail components, data-driven features and physical constraints are integrated.

[0285] (5) Adaptive learning of mutation points:

[0286] Constructing a change probability model for Bayesian change detection:

[0287]

[0288] In the formula, c t ∈{0,1} represents a variable indicating a change point; h t In hidden state; W c h is the learnable weight matrix.t-1 h t σ represents the hidden state vectors of adjacent time steps; σ is the Sigmoid function.

[0289] Dynamic parameter adjustment:

[0290]

[0291] In the formula, MO is the key parameter of the model, enabling abrupt adaptation of the parameter space; MO t These are the key parameters of the model at the current time step; MO0 is the initial parameter value; ΔMO is the parameter adjustment step size; c i The cumulative sum of the indicator variables for historical turning points.

[0292] (6) Physical information neural network architecture:

[0293]

[0294] In the formula, This represents the prediction error; For the KL divergence regularization term in change point detection; Loss due to physical constraints; These are adaptive weight parameters.

[0295] (7) Uncertainty Quantification (Monte Carlo Physics Dropout):

[0296] During prediction, dropout is kept on, and N samples are performed:

[0297]

[0298] In the formula, CN represents the model prediction output for the i-th Monte Carlo sampling; CN is the number of samplings. σ represents the network parameters after randomly discarding some neurons during the i-th sampling; 2 To predict variance.

[0299] 3.3 Short-term rainfall forecast (MDSTF model):

[0300] (1) Select key indicators and observe them:

[0301] Air humidity, temperature, wind speed, air pressure, etc.;

[0302] (2) Decompose the time series data of each indicator:

[0303] For one set of time series indicators x i (t):

[0304] Generate a noisy signal:

[0305] x (i) (t)=x(t)+β0w (i) (t)(i=1,2,…,I)

[0306] w (i) (t): Gaussian white noise; β0: initial noise figure; x(t) is the original signal.

[0307] Decompose and average:

[0308]

[0309] In the formula, EMD1 represents the empirical mode decomposition of the signal; is the first-order IMF obtained from the i-th noisy signal decomposition; IMF1(t) is the average of the first-order IMFs of all I-th decomposition results.

[0310] Update residuals:

[0311] r1(t) = x(t) - IMF1(t)

[0312] In the formula, r1(t) is the residual signal after removing the first-order IMF.

[0313] k-th order decomposition:

[0314]

[0315] In the formula, r k-1 (t) represents the residual signal in the (k-1)th stage; E k (w (i) (t) represents the Gaussian white noise w (i) (t) The kth order IMF obtained after EMD decomposition; β k-1 is the noise figure for the (k-1)th stage.

[0316]

[0317] In the formula, The k-th order IMF is obtained from the i-th noisy residual decomposition; IMF k (t) is the average value of the k-th order IMF of all I-th decomposition results.

[0318] r k (t)=r k-1 (t)-IMF k (t)

[0319] In the formula, r k (t) represents the new residual signal after removing the k-th order IMF.

[0320] Final decomposition results:

[0321]

[0322] In the formula, R is a linear combination of all IMFs; R(t) is the residual.

[0323] (3) Short-term forecast:

[0324] Data decomposition and reconstruction (assuming the original rainfall sequence is P(t)):

[0325]

[0326] In the formula, C i R(t) is the i-th intrinsic mode function; R(t) is the residual, reflecting the long-term trend or low-frequency component; P(t) is the original rainfall time series; This represents the total number of IMFs obtained from EMD decomposition.

[0327] Frequency division prediction:

[0328] High-frequency IMF (bidirectional LSTM+TCN model):

[0329]

[0330] In the formula, The length of the time window; Let be the predicted value of the i-th IMF at time t+1; For the i-th IMF in the time window Historical data within, is the window length; BiLSTM is a bidirectional long short-term memory network used to capture bidirectional temporal dependencies; TCN is a temporal convolutional network that extracts local mutation features through dilated convolution.

[0331] Low-frequency IMF (Adaptive Weighted ARIMA Model):

[0332]

[0333] In the formula, For ARIMA coefficients; Let be the predicted value of the j-th low-frequency IMF at time t+1; The order of autoregression (AR); The moving average (MA) order; The nth moving average coefficient; ∈(t) represents the white noise term; This represents the predicted value of the residual term at time t+1; These are the autoregressive coefficients; This is the time trend coefficient; t represents the intercept term; t represents time.

[0334] Dynamic attention fusion mechanism:

[0335]

[0336] In the formula, h i (t) represents the latent state of the i-th IMF prediction model; H(t) represents the global context direction; W i The weight matrix is ​​trainable. Let W be the attention weight for the i-th IMF, representing its importance to the final prediction; i For trainable weight matrix; [h i [H(t)] represents the vector concatenation operation, which concatenates the hidden states h of the IMF. i (t) is merged with the global context H(t); H(t) is the global context vector; This represents the final predicted rainfall amount at time t+1. Let be the predicted value of the i-th IMF; This represents the predicted value of the residual term.

[0337] Residual correction and uncertainty quantification:

[0338]

[0339] In the formula, The adaptive correction coefficients are Δ(t+1), which is the correction term used to adjust the predicted value. MLP stands for Multilayer Perceptron; σ(t) is the absolute value of the recent prediction error; σ(t) is the standard deviation of the recent prediction error, reflecting uncertainty; P final (t+1) is the final predicted value after error correction.

[0340] (4) Substitute into the rainfall prediction model;

[0341]

[0342] P represents the predicted rainfall in mm; ∈ represents the model error term; t represents time; ψ represents the water vapor flux divergence in kg·m³. -2 ·s -1 , q is the specific humidity. Let g be the wind speed vector, g be the acceleration due to gravity, and p be the acceleration due to gravity. s p t These are the surface air pressure and the integral top pressure. For pressure gradient; Temperature gradient, K·km -1 , v is the vertical temperature lapse rate. T For horizontal temperature advection, Cp RH is the specific heat capacity of dry air at constant pressure. Γ : Relative humidity correction term; RH: Relative humidity, %; Γ: Clapeyron correction factor, L v For latent heat of water vapor, R v R is the water vapor gas constant. d The gas constant for dry air. The ambient temperature; Barometric pressure regulation factor; P a Standardized sea-level pressure; ω: vertical velocity. CAPE: Convection Available Potential Energy LFC stands for Free Convection Height, and EL stands for Balance Height. The body is weak and cold. The ambient temperature is virtual; RI: Richardson number. Let u and v be the potential temperature, and v be the horizontal wind speed components. Cloud water content ratio, C w For actual cloud water content, C w0 This is the critical cloud water content.

[0343] 3.4 Simulation of Ideal Crop Growth State

[0344] Based on predictions of future rainfall and soil moisture, an innovative framework integrating crop growth mechanism model, temporal deep learning, and dynamic feedback optimization is proposed to predict the ideal growth state of crops in fields under given environmental conditions. This framework is named Hybrid Crop Growth Digital Twin (HCG-DT).

[0345] (1) Input data standardization

[0346]

[0347] In the formula, X norm σx represents the standardized data; μx represents the mean of the original data; σx represents the standard deviation of the original data; and X represents the original environmental data.

[0348] (2) Mechanism model construction (photothermal potential baseline)

[0349]

[0350] In the formula, P g This refers to the total photosynthetic rate; Light energy conversion efficiency; This is a temperature correction factor; Photosynthetically active radiation; The half-saturation constant of the light response curve; is the leaf light absorption coefficient; LAI is the leaf area index.

[0351] (3) Deep learning correction (environmental stress modeling)

[0352] δ(t)=f LSTM (soil moisture (t), rainfall (t), T (t), H (t))

[0353] Corrected growth index = Mechanism model output (t) × (1-δ(t))

[0354] In the formula, δ(t) is the stress correction coefficient for the time step (0 to 1, representing the growth attenuation ratio); f LSTM The nonlinear function used to model the LSTM network is: soil moisture (t); rainfall (t) is the predicted rainfall; T(t) is the temperature; and H(t) is the relative humidity.

[0355] (4) Dynamic feedback optimization (field strategy generation)

[0356]

[0357] In the formula, S w (t) represents the water stress index; S T (t) represents the temperature stress index; These are the weighting coefficients; To optimize the time range.

[0358] (5) Visualization of coercion risks

[0359]

[0360] In the formula, R d The probability of risk; τ is an indicator function (1 if the condition is met, 0 otherwise); w , τ T The moisture / temperature stress threshold; This refers to the number of samples or the number of days to be predicted.

[0361] (6) Growth status prediction output

[0362] Y(t) = Y 机理 (t)·(1-δ(t))+η(t)

[0363] In the formula, Y(t) is the predicted growth index; Y 机理 η(t) represents the predicted value from the mechanistic model; η(t) represents the Gaussian noise term. 3.5 Long-term water-saving scheduling (Deep Reinforcement Learning DRL model)

[0364] (1) State space

[0365] s t =[QX t W res W gnd G his ,S t ,R f ]

[0366] In the formula, QX t W represents climate prediction vectors (rainfall, temperature, evaporation, etc.). res W represents the current water storage capacity of the reservoir. gnd G represents the exploitable reserves of groundwater. his Historical groundwater extraction volume; S t For crop growth stages; R f For the capacity of rainwater harvesting facilities.

[0367] (2) Action space

[0368]

[0369] In the formula, Reservoir water allocation ratio; Groundwater distribution ratio; Rainwater utilization ratio.

[0370] (3) Reward function

[0371] J L =v1R WUE +v2R GND +v3R RAIN

[0372] In the formula, v1, v2, and v3 are weighting coefficients used to balance the priorities of different objectives.

[0373] Minimize water consumption per unit of output:

[0374]

[0375] In the formula, YC represents crop yield; R RAIN This represents the amount of rainwater collected during the forecast period; the negative sign indicates a penalty for high water consumption.

[0376] Sustainable constraints on groundwater:

[0377]

[0378] In the formula, G threshold R represents the threshold for sustainable groundwater extraction; R is the over-extraction penalty coefficient.

[0379] Maximizing rainwater resource utilization:

[0380]

[0381] In the formula, R total This refers to the total available rainwater resources.

[0382] (4) State transition equation

[0383]

[0384] In the formula, R inflow The inflow rate to the reservoir; L evap For reservoir evaporation loss; R infil η represents the natural recharge of groundwater; η is the recharge efficiency coefficient.

[0385] 3.6 Short-term precision irrigation (model predictive control, MPC)

[0386] Rolling optimization objective function:

[0387]

[0388] stXW min ≤XW k ≤XW field

[0389] 0≤I k ≤I max

[0390] In the formula, NY represents the prediction time domain; QJ and RJ are weight matrices balancing moisture retention and water conservation; XW field Field water holding capacity (upper limit of crop water requirements); XW min XW is the wilting coefficient (lower limit of water requirement). target Target soil moisture content; I max This represents the maximum water supply rate of the drip irrigation system.

[0391] Improvements and advantages:

[0392] This innovative method combines deep reinforcement learning-based climate prediction optimization with real-time dynamics of crop growth, achieving closed-loop irrigation decision-making across multiple time scales. Compared to traditional static water allocation or single-time-scale methods, it significantly improves the overall efficiency of water resource utilization and groundwater sustainability while ensuring crop yield. This method primarily proposes the Lag-PhysFusion model, MPF-Net model, MDSTF model, HCG-DT model, and deep reinforcement learning DRL model to predict soil moisture, weather conditions, and ideal crop growth states, with remarkable results.

[0393] S4 Intelligent Terminal Control System: Drones periodically fly to monitor the field, capturing real-time images of crops and uploading the data to the intelligent control terminal for unified decision-making. Based on real-time data, forecast results, and irrigation model calculations, the terminal controller controls the intelligent sprinkler irrigation equipment to ensure the precision of irrigation operations.

[0394] The algorithm in this control terminal includes image analysis to determine crop growth in each area; soil moisture transport analysis to determine soil moisture content in each area; and future rainfall forecasts. It performs an overall assessment of this multi-data set to control smart valves at various locations for irrigation.

[0395] like Figure 1 As shown.

[0396] 4.1 Determining whether crops are in an ideal growth state:

[0397] 4.1.1 Image Preprocessing:

[0398] Radiometric correction. The Empirical Linear Method (ELM) is used to eliminate the effects of illumination variations.

[0399]

[0400] In the formula, DN is the original pixel value; as,bs are the gain and offset coefficients calculated by the ground calibration plate.

[0401] 4.1.2 Image stitching:

[0402] SIFT feature matching and fade-in / fade-out fusion algorithms are used to generate orthophotos (error < 0.5 pixels).

[0403] 4.1.3 Image Analysis:

[0404] (1) Crop division:

[0405] Using an improved Mask R-CNN model (backbone network is ResNet-101-FPN):

[0406] Input: RGB image (1024×1024 resolution)

[0407] Output: Pixel-level mask of a single crop (accuracy > 92%)

[0408] (2) Growth stage identification:

[0409] ResNet-50 classification model based on temporal images:

[0410] Input: Three consecutive crop canopy images (time step of 3 days)

[0411] Output: Probability distribution of the growth period (e.g., 85% during emergence, 12% during jointing).

[0412] (3) Key parameter extraction:

[0413] Leaf Area Index (LAI): Inverted through canopy porosity (Beer-Lambert Law)

[0414] Canopy temperature: Radiometric calibration of thermal infrared imagery (accuracy ±0.5℃)

[0415] Biomass estimation: Regression model of NDVI and ground-measured data (R) 2 >0.8)

[0416] 4.2 Irrigation Demand Assessment and Decision-Making:

[0417] By using images scanned and processed by drones, we can determine whether the actual growth trend of crops meets the ideal growth trend, determine whether crops in different regions need irrigation, and optimize the timing of irrigation.

[0418] 4.3 Intelligent Control and Precision Irrigation:

[0419] Based on the decision-making results, the terminal controller controls the intelligent sprinkler system to carry out precise irrigation, ensuring that the irrigation amount meets the actual needs of the crop and avoiding over-irrigation or under-irrigation.

[0420] 4.4 Data Storage and Remote Monitoring:

[0421] All data and system status will be stored on the cloud platform, facilitating long-term monitoring and analysis by users and supporting remote control.

[0422] 4.5 Water-saving effect closed-loop feedback

[0423] Deploy IoT devices (flow meters, smart water meters, etc.) to monitor water consumption and water-saving indicators in real time; use digital twin technology to simulate the water-saving potential of different irrigation strategies and perform dynamic optimization to ensure continuous improvement of water-saving targets.

[0424] It should be noted that the above embodiments are not intended to limit the scope of protection of the present invention. Equivalent transformations or substitutions made based on the above technical solutions all fall within the scope of protection of the claims of the present invention.

Claims

1. A method for designing and dynamically optimizing water allocation for farmland water-saving irrigation systems based on multi-source data fusion of supply and demand, characterized in that the method includes the following steps: S1 “Crop-Soil-Meteorology” Multi-Source Data Coupling Modeling S2 water-saving water system design optimization S3 dynamic water-saving distribution strategy The S4 intelligent terminal control system is used for control; Step S1 is as follows: 1.1 Multi-source data monitoring, Multiple sensors are deployed throughout the field to monitor soil moisture, temperature, weather, and crop water requirements in real time, as shown below: Sensor density calculation: Based on the improved adaptive PSO algorithm, the sensor density optimization method uses the numerical solution of the soil moisture transport equation, namely the Richards equation, as the particle search direction correction term, and dynamically adjusts the crop root weight distribution matrix through NDVI time series data. This includes multi-objective fitness function reconstruction and the construction of a three-dimensional evaluation index by integrating soil property spatial variation coefficient CV, crop root distribution weight RW, and meteorological influence factor MF. In the formula, This is the fitness value; , , These are dynamic weighting coefficients; Economic constraint factor; The comprehensive economic constraint term is used to quantify the implementation cost of the sensor deployment scheme; CV is the spatial variability coefficient of soil properties; RW is the weight of crop root distribution; MF is the meteorological influence factor. Secondly, there are farmland-specific improvement strategies, including hierarchical particle coding, the introduction of mutation operators, and energy-constrained models. All data collected by the sensors is uploaded to the data terminal control system via wireless communication technology, ensuring timely data updates and remote monitoring. The transmitted real-time data can support dynamic monitoring of environmental changes in Datian. 1.2 Construct the model architecture, as follows: 1.2.1 Soil moisture transport modeling, Unsaturated zone flow model : Soil moisture content; t: time; z: depth; Unsaturated hydraulic conductivity , denoted as saturated hydraulic conductivity, and b represents the gain and offset coefficients calculated using a ground calibration plate. Soil water pressure head , For intake pressure head; Root water absorption rate , Let be the root density distribution function. For potential transpiration rate, Determination of the most unfavorable point, WTLI: Time Deficit Water Index; Critical soil moisture content; : The function of actual soil moisture content changing with time t; The time period for observation or evaluation; : within the time window The cumulative deficit of soil moisture content below the critical value; Theoretical maximum deficit 1.2.2 Crop Response Module A coupling algorithm for the Jarvis stomatal conductance model and the Feddes root uptake function was constructed. For the Jarvis porosity conductance model: In the formula, The photosynthetically active radiation response function; This is the water vapor pressure difference suppression function; This is the soil water potential response function; This is a temperature regulation function. For Feddes root uptake function: In the formula, This represents the amount of water absorbed by the roots at depth z per unit volume of soil. This is the water stress coefficient; Potential transpiration rate; Let be the root density distribution function. In the formula, For soil water potential; The anaerobic threshold; This represents the optimal upper limit of water absorption. This is the starting point of a linear descent; For permanent wilting points, In the formula, Root density distribution function; The root distribution half-life depth is 20cm to 100cm; z is the depth; m is the distribution shape parameter, where m=1 indicates a slow, linear decrease in root density with decreasing depth, and m=5 indicates a sharp decrease in root density with decreasing depth, with roots concentrated in the surface layer. 1.2.3 Meteorological Driven Module The Penman-Monteith dual-source model was used to distinguish the contributions of soil / vegetation evapotranspiration. Vegetation transpiration: In the formula, This refers to vegetation transpiration. The slope of the saturated water vapor pressure-temperature curve; Net radiation of the canopy; air density; The specific heat of air at constant pressure. This is due to a deficiency in air vapor pressure. For canopy aerodynamic drag; This represents the total stomatal resistance of the canopy. This is the constant of the wet and dry meter. Soil evaporation: In the formula, This refers to soil evaporation. The slope of the saturated water vapor pressure-temperature curve; air density; The specific heat of air at constant pressure. This is due to a deficiency in air vapor pressure. Net radiation at the soil surface; For soil surface aerodynamic drag; For soil evaporation resistance; This is the constant of the wet and dry meter. 1.2.4 Establish the dynamic response function of water stress and photosynthetic product allocation. Water stress index: In the formula, The water stress index; This represents the actual evapotranspiration rate. Potential transpiration Total photosynthetic products correction: In the formula, Correction for total photosynthetic products; To maximize photosynthetic capacity; This represents the actual porosity. This represents the maximum porosity.

2. The method for designing and dynamically optimizing farmland water-saving irrigation systems based on multi-source data fusion of supply and demand as described in claim 1, characterized in that, The S2 water-saving water system design optimization is as follows: 2.1 Construct the system framework, including terrain data, soil data, crop data, and meteorological data; 2.2 Multi-objective optimization A multi-objective water-saving irrigation optimization model is established with the objectives of minimizing water conveyance loss, maximizing irrigation uniformity, and maximizing reclaimed water utilization, as detailed below: (1) Objective function of water conveyance loss: reduce leakage and evaporation through channel seepage prevention design and pipeline transformation. In the formula, Total number of channel segments; Let be the length of the i-th channel segment; Let i be the leakage rate per unit length of the i-th channel segment; The efficiency coefficient for channel seepage prevention; This represents the total number of pipeline segments; Let J be the surface area of ​​the j-th open-air pipeline segment; Evaporation rate per unit area; This is a correction factor for pipeline coverage. In the formula, Let i be the leakage rate per unit length of the i-th channel segment; Let be the saturated permeability coefficient of the soil below the i-th segment of the channel; For the bottom width of the channel; Design the water depth for the channel. In the formula, Let J be the surface area of ​​the j-th open-air pipeline segment; The diameter of the pipe; For the length of the pipe, In the formula, This is the wind speed correction factor; This represents the pressure difference of saturated water vapor in the air. Atmospheric pressure (2) Objective function for irrigation uniformity: constraint: In the formula, n is the total number of irrigation units; Water supply for the k-th irrigation unit; This represents the average water supply per irrigation unit. ; Water requirements for crops; For the total irrigated area, (3) Objective function for reclaimed water utilization: In the formula, For rainwater harvesting efficiency, ; This represents the total irrigation water demand. , The volume of the rainwater collection tank; This represents the daily treatment capacity of reclaimed water. Ic represents the irrigation cycle; Dr is the catchment area; Dr is the design rainfall; Rc is the runoff coefficient.

3. The method for designing and dynamically optimizing farmland water-saving irrigation systems based on multi-source data fusion of supply and demand as described in claim 1, characterized in that... The S3 dynamic water-saving distribution strategy is as follows: 3.1 Soil moisture prediction using Lag-PhysFusion: (1) Historical time series data: Input features: In the formula, Let be the input feature vector at time t; Let be the measured soil moisture value at time t; T is the total length of the time series, i.e., the maximum number of time steps in the historical data. To improve the soil moisture calculation using the Richards equation; It is a lagging index; For meteorological data; This refers to irrigation volume; Vegetation index; To measure soil moisture, the output value is... , (2) Physical feature extraction: Physical dynamics can be extracted by improving the discrete solutions of the Richards equations. Discrete form: In the formula, SW represents soil water potential; t represents time. represents unsaturated hydraulic conductivity; zz represents vertical spatial coordinates; YH represents source and sink terms; Let be the moisture content of the d-th soil layer at time t; For time step; The spatial step size; Let be the hydraulic conductivity of the dd-th layer at time t; Let d be the source and sink terms of the d-th layer at time t. The physical model output is fused with the hysteresis exponent to form an enhanced feature: In the formula, A vector encoded from physical features; For activation functions; This involves concatenating the input features. This is the weight matrix; For bias vectors, (3) Lag perception time coding: In the formula, Standard Transformer position encoding; This is a learnable parameter used to adjust the degree of time distortion of the lag effect; It is a lagging index. (4) Spatiotemporal attention mechanism: In the formula, J is the query matrix; J is the key matrix; Z is the value matrix; This is a lag similarity matrix. , For temperature parameters, the sensitivity to hysteresis differences is controlled. It is a lagging index; The dimension of the key vector. (5) Multi-scale temporal convolution: In the formula, The output features of the k-th convolutional layer at time t; Let be the s-th weight parameter of the k-th convolutional layer; For time step The input feature vector; Let k be the expansion factor of the k-th layer; The number of timesteps covered by the convolution kernel. (6) Dynamic fusion prediction: Combining physical characteristics P t With time series features C t : ， Fusion weights Dynamically generated by the lag index. ， In the formula, For predicting output; For dynamic fusion weights; This is a branch of the physical model; For the sequence model branch; It is a lagging index; The Sigmoid function controls the contribution ratio between the physics model and the data-driven model. These are weight parameters; For bias parameters, 3.2 Annual Rainfall Forecast: (1) Input data: Target variable: Historical annual rainfall sequence, where y represents the year; Climate factors: m climate indices Geographical features: p static geographic parameters (2) Wavelet multi-scale decomposition: for Perform Mallat wavelet decomposition: In the formula, The target variable; For the j-th scale detail component; The approximate component is J; J is the number of decomposition layers. (3) Physical constraint feature engineering: Climate system energy constraints: In the formula, Energy for the climate system; HD represents the atmospheric vertical velocity field; HD represents the water vapor flux divergence. , These are the learnable constraint coefficients. Geographical features embedded: In the formula, , The weight matrix is ​​a learnable weight matrix; For time-coded vectors, (4) Multimodal spatiotemporal attention mechanism: Cross-attention to climate factors: In the formula, ; ; , In the formula, The output attention feature reflects the temporal effect of climate factors on rainfall variability; O is the dynamic query matrix. These are wavelet detail components, representing the interannual variability signal; V is the scaling factor; VZ is the value matrix; These are wavelet approximation components; Geographic modulation matrix; G is the learnable weight matrix; G is the geographic feature vector. For learnable weight matrix, Time-frequency dual-domain gating: In the formula, This is the gate vector; Use the Sigmoid activation function; For learnable gated weight matrix; To refine the detailed components, data-driven features and physical constraints are integrated. (5) Adaptive learning at mutation points: Constructing a change probability model for Bayesian change detection: In the formula, Indicates a variable as a change point indicator; It is in a hidden state; The weight matrix is ​​a learnable weight matrix; These are the hidden state vectors of adjacent time steps; For the Sigmoid function, Dynamic parameter adjustment: In the formula, These are key parameters of the model, enabling abrupt adaptation of the parameter space. These are the key parameters of the model at the current moment; These are the initial parameter values; Adjust the step size for the parameters; The cumulative sum of the indicator variables for historical turning points. (6) Physical information neural network architecture: In the formula, This represents the prediction error; For the KL divergence regularization term in change point detection; Loss due to physical constraints; , , For adaptive weight parameters, (7) Quantification of uncertainty: During prediction, dropout is kept on, and N samples are performed: In the formula, CN represents the model prediction output for the i-th Monte Carlo sampling; CN is the number of samplings. These are the network parameters after randomly discarding some neurons during the i-th sampling. To predict variance; 3.3 Short-term rainfall forecast: (1) Select key indicators and observe them: air humidity, temperature, wind speed, and air pressure; (2) Decompose the time series data of each indicator: For one set of indicator time series : Generate a noisy signal: Gaussian white noise; Initial noise figure; The original signal, Decompose and average: In the formula, To perform empirical mode decomposition on the signal; The first-order IMF is obtained by decomposing the i-th noisy signal; The average of the first-order IMF of all I-order decomposition results. Update residuals: In the formula, To remove the residual signal after the first-order IMF, k-th order decomposition: In the formula, For the first The residual signal of the stage; For Gaussian white noise The k-th order IMF obtained after EMD decomposition; For the first noise figure of the stage In the formula, The k-th order IMF is obtained from the i-th noisy residual decomposition. The average of the k-th order IMF of all I-th decomposition results. In the formula, To remove the new residual signal after removing the k-th order IMF, Final decomposition results: In the formula, For a linear combination of all IMFs; For residuals, (3) Short-term forecast: Decompose the data and reconstruct it. Let the original rainfall sequence be P(t): In the formula, Let i be the i-th intrinsic mode function; The residual reflects long-term trends or low-frequency components; This is the original time series of rainfall; The total number of IMFs obtained from EMD decomposition. Frequency division prediction: High-frequency IMF, bidirectional LSTM+TCN model: In the formula, The length of the time window; Let be the predicted value of the i-th IMF at time t+1; For the i-th IMF in the time window Historical data within, The length of the window; It is a bidirectional long short-term memory network used to capture bidirectional temporal dependencies; This is a temporal convolutional network that extracts local abrupt change features through dilated convolution. Low-frequency IMF, adaptive weighted ARIMA model: In the formula, , For ARIMA coefficients; Let be the predicted value of the j-th low-frequency IMF at time t+1; The order of the autoregressive AR; The moving average (MA) order; The coefficient of the nth moving average; This is the white noise term; This represents the predicted value of the residual term at time t+1; These are the autoregressive coefficients; This is the time trend coefficient; The term represents the intercept; t represents time. Dynamic attention fusion mechanism: In the formula, Let be the implicit state of the i-th IMF prediction model; For global context; The weight matrix is ​​trainable. Let be the attention weight for the i-th IMF; The weight matrix is ​​trainable. For vector concatenation operations, the hidden states of the IMF are concatenated. With global context merge; For global context vectors; This represents the final predicted rainfall amount at time t+1. Let be the predicted value of the i-th IMF; The predicted value of the residual term. Residual correction and uncertainty quantification: In the formula, For adaptive correction coefficients; This is a correction term used to adjust the predicted values. ; It is a multilayer perceptron; This represents the absolute value of the recent forecast error; This represents the standard deviation of recent forecast errors, reflecting uncertainty. This is the final predicted value after error correction. (4) Substitute into the rainfall prediction model; P represents the predicted rainfall, in mm; This is the model error term; For water vapor flux divergence, , q represents the specific moisture content. Let g be the wind speed vector, g be the acceleration due to gravity, and p be the acceleration due to gravity. s p t These are the surface air pressure and the integral top pressure. For pressure gradient; Temperature gradient , , This is the vertical temperature lapse rate. For horizontal temperature advection, The specific heat capacity of dry air at constant pressure; : Relative humidity correction term; RH: relative humidity, % : Clapeyron correction factor , Latent heat of water vapor The constant of water vapor. The gas constant for dry air. The ambient temperature; Barometric pressure regulation factor; Standardized sea level pressure; Vertical velocity CAPE: Convective Available Potential Energy LFC stands for Free Convection Height, and EL stands for Equilibrium Height. The body is weak and cold. The ambient temperature is artificial; :Richardson number, , Let u and v be the potential temperature, and v be the horizontal wind speed components. Cloud water content ratio This represents the actual cloud water content. The critical cloud water content; 3.4 Simulation of Ideal Crop Growth State (1) Input data standardization In the formula, The data is standardized. This represents the mean of the original data. The standard deviation of the original data; This is the original environmental data. (2) Mechanism model construction In the formula, This refers to the total photosynthetic rate; Light energy conversion efficiency; This is a temperature correction factor; Photosynthetically active radiation; The half-saturation constant of the light response curve; The light absorption coefficient of the leaf; LAI is the leaf area index. (3) Deep learning correction In the formula, The stress correction factor for each time step (0~1, representing the growth attenuation ratio); Nonlinear functions used to model LSTM networks; Soil moisture content; To predict rainfall; For temperature; This refers to the relative humidity of the air. (4) Dynamic feedback optimization In the formula, The water stress index; The temperature stress index; These are the weighting coefficients; To optimize the time range, (5) Visualization of coercion risk In the formula, The probability of risk; This is an indicator function; it returns 1 if the condition is met, and 0 otherwise. , The moisture / temperature stress threshold; For sample size or number of days to predict, (6) Growth status prediction output In the formula, For predicted growth indicators; These are predictions from the mechanistic model; This is a Gaussian noise term. 3.5 Long-term water-saving scheduling, (1) State space In the formula, The climate prediction vectors are rainfall, temperature, and evaporation. This represents the current water storage capacity of the reservoir. This refers to exploitable groundwater reserves. This represents historical groundwater extraction volume. This refers to the crop growth stage; For the capacity of rainwater harvesting facilities, (2) Action space In the formula, The water allocation ratio of the reservoir; Groundwater distribution ratio; Rainwater utilization ratio (3) Reward function In the formula, , , These are weighting coefficients used to balance the priorities of different objectives. Minimize water consumption per unit of output: In the formula, YC represents crop yield; This represents the predicted rainwater collection volume; the negative sign indicates a penalty for high water consumption. Sustainable constraints on groundwater: In the formula, The threshold for sustainable groundwater extraction; This is the over-extraction penalty coefficient. Maximizing rainwater resource utilization: In the formula, For the total available rainwater resources, (4) State transition equation In the formula, The inflow rate to the reservoir; Losses due to evaporation from the reservoir; This refers to the natural recharge of groundwater. This is the supply efficiency coefficient. 3.6 Short-term precision irrigation, Rolling optimization objective function: In the formula, NY represents the prediction time domain; QJ and RJ are weight matrices that balance moisture retention and water conservation. Field water holding capacity; The wilting coefficient; The target soil moisture content; This represents the maximum water supply rate of the drip irrigation system.

4. The method for designing and dynamically optimizing farmland water-saving irrigation systems based on multi-source data fusion of supply and demand as described in claim 1, characterized in that, The S4 intelligent terminal control system performs the control, as detailed below. 4.1 Determining whether crops are in an ideal growth state: 4.1.1 Image Preprocessing: Radiometric correction, using the empirical linear method ELM to eliminate the effects of illumination variations: In the formula, DN is the original pixel value; as,bs are the gain and offset coefficients calculated using the ground calibration board. 4.1.2 Image stitching: An orthophoto was generated using SIFT feature matching and fade-in / fade-out fusion algorithms, with an error of <0.5 pixels. 4.1.3 Image Analysis: (1) Crop division: Using an improved Mask R-CNN model: Input: RGB image Output: Pixel-level mask for a single crop, with an accuracy >92%. (2) Growth stage identification: ResNet-50 classification model based on temporal images: Input: Crop canopy images from three consecutive periods Output: Probability distribution of reproductive period (3) Extraction of key parameters: Leaf Area Index (LAI): Inversion via Canopy Porosity Canopy temperature: Radiometric calibration of thermal infrared images Biomass estimation: The regression model of NDVI and ground-measured data showed R² > 0.

8. 4.2 Irrigation Demand Assessment and Decision-Making By using images scanned and processed by drones, it can be determined whether the actual growth trend of crops meets the ideal growth trend, whether crops in different areas need irrigation, and to optimize the timing of irrigation. 4.3 Intelligent control and precision irrigation Based on the decision-making results, the terminal controller controls the intelligent sprinkler system to provide precise irrigation, ensuring that the irrigation amount meets the actual needs of the crop and avoiding over-irrigation or under-irrigation. 4.4 Data storage and remote monitoring All data and system status will be stored on the cloud platform, facilitating long-term monitoring and analysis by users, and supporting remote control. 4.5 Water-saving effect closed-loop feedback Deploy IoT devices to monitor water consumption and water-saving indicators in real time; use digital twin technology to simulate the water-saving potential of different irrigation strategies and perform dynamic optimization to ensure that water-saving targets are continuously improved.

5. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for designing and dynamically optimizing farmland water-saving irrigation systems by fusing supply and demand multi-source data as described in any one of claims 1 to 4.

6. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instruction is executed by the processor, it implements the method for designing and dynamically optimizing farmland water-saving irrigation systems by fusing supply and demand data from multiple sources as described in any one of claims 1-4.

Images