Intelligent regulation and control decision-making method for agricultural water resources in irrigated area

Abstract

The invention discloses an intelligent regulation and control decision-making method for agricultural water resources in an irrigation area, and relates to the technical field of agricultural water conservancy informatization. According to the method, an irrigation area double-layer dynamic graph model mapping physical topology and hydraulic connection is called, and a water supply capacity scene package containing multiple space-time possibilities is generated; generating a water demand scene based on crop water demand characteristics, decomposing a total risk upper limit of a full irrigation area into a risk budget distribution matrix of space-time unit granularity, and solving and generating a water supply interval contract containing a water supply lower bound, a water supply upper bound and a risk level in combination with a differential water shortage loss function; and establishing an engineering constraint set, solving an optimal engineering regulation and control instruction sequence based on a rolling time domain mechanism on the premise of meeting a contract, and issuing and executing the optimal engineering regulation and control instruction sequence. According to the method, the irrigation area is reconstructed into a manual controllable network, the macroscopic risk is converted into an executable interval contract, and critical period priority protection and closed-loop risk avoidance under the uncertain condition are achieved.

Description

Intelligent Decision-Making Methods for Agricultural Water Resources in Irrigation Districts Technical Field

[0001] This invention relates to the field of agricultural water conservancy information technology, and in particular to a method for intelligent regulation and decision-making of agricultural water resources in irrigation districts. Background Technology

[0002] As the core unit of agricultural water use, the regulation of water resources in irrigation districts is crucial for ensuring food security and improving water resource utilization efficiency. Faced with hydrological fluctuations caused by climate change and the spatial and temporal differences in crop water requirements, how to coordinate water supply and demand, ensuring crop yields during critical periods while mitigating systemic water shortage risks, has become a key issue that urgently needs to be addressed in the modern management of irrigation districts.

[0003] Currently, water resource allocation in irrigation districts mainly relies on distributed hydrological models to simulate natural runoff processes and calculate available water volume based on rainfall and topography. On this basis, conventional scheduling often employs stochastic programming or multi-objective optimization algorithms, typically aiming to minimize the total water shortage or achieve spatial uniformity of water shortage levels when formulating water allocation plans. Commands are then used to schedule water conservancy facilities such as gates and pumping stations based on fixed water allocation quotas.

[0004] However, existing technologies have shortcomings in terms of engineering controllability characterization and risk decision-making feasibility, such as neglecting the dynamic impact of engineering regulation on hydraulic topology, having static configuration schemes and lacking risk decomposition mechanisms, making it difficult to cope with supply and demand disturbances. Therefore, further research and innovation are needed to solve the above-mentioned problems of existing technologies. Summary of the Invention

[0005] Purpose of the invention: This application provides an intelligent regulation and decision-making method for agricultural water resources in irrigation districts to solve the above-mentioned problems existing in the prior art.

[0006] Technical solution: According to one aspect of this application, a method for intelligent regulation and decision-making of agricultural water resources in irrigation districts is provided, comprising:

[0007] A two-layer dynamic graph model of the irrigation district is constructed to map the physical topology and hydraulic relationship of the irrigation district. Based on meteorological ensemble forecast data and parameter uncertainty distribution, the two-layer dynamic graph model of the irrigation district is used to generate water supply capacity scenario packages containing multiple spatiotemporal possibilities.

[0008] Based on the water demand characteristics of crops, a water demand scenario is generated, and a total risk ceiling for the entire irrigation area is set. The total risk ceiling is then decomposed into a risk budget distribution matrix with spatiotemporal unit granularity.

[0009] Based on the risk budget distribution matrix and the water supply capacity scenario package, the water supply interval contract is generated by solving. The water supply interval contract defines the lower bound, upper bound and allowable risk level of water supply for each spatiotemporal unit.

[0010] Establish a set of engineering constraints that includes physical limitations on the operation of engineering equipment. Under the premise of satisfying the water supply section contract, solve for the optimal engineering control command sequence based on the rolling time domain mechanism and issue it for execution.

[0011] According to another aspect of this application, an intelligent regulation and decision-making system for agricultural water resources in irrigation districts is provided, comprising:

[0012] The physical modeling and scene generation module is used to construct a two-layer dynamic graph model of the irrigation area and generate a water supply capacity scene package;

[0013] The risk budget and contract generation module is used to generate water demand scenarios, set the total risk ceiling for the entire irrigation district, decompose the total risk ceiling to obtain the risk budget distribution matrix, and solve the water supply interval contracts.

[0014] The rolling control and execution module is used to establish a set of engineering constraints, solve for the optimal engineering control instruction sequence based on the rolling time domain mechanism, and issue the execution of the optimal engineering control instruction sequence.

[0015] Beneficial effects: This invention reconstructs the irrigation district into an artificially controllable network, transforming macro-level risks into executable interval contracts, and achieving priority protection for critical periods and closed-loop risk avoidance under uncertain conditions. The related technical effects will be described in detail below with reference to specific embodiments. Attached Figure Description

[0016] Figure 1 is a flowchart of the intelligent regulation and decision-making method for agricultural water resources in irrigation districts provided in an embodiment of this application.

[0017] Figure 2 is a flowchart of generating a water supply capacity scenario package containing multiple spatiotemporal possibilities using a two-layer dynamic graph model of an irrigation district, as provided in an embodiment of this application.

[0018] Figure 3 is a flowchart of the construction of spatiotemporal differentiated weights and spatiotemporal differentiated water shortage loss functions provided in the embodiments of this application.

[0019] Figure 4 is a flowchart of the risk budget distribution matrix that decomposes the total risk ceiling into spatiotemporal unit granularity, as provided in an embodiment of this application.

[0020] Figure 5 is a flowchart of the process of generating a water supply interval contract based on the risk budget distribution matrix and water supply capacity scenario package provided in the embodiment of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0023] To address the aforementioned issues, the applicant conducted in-depth searches and analyses, and discovered:

[0024] Correspondingly, traditional hydrological models treat irrigation districts as static natural watersheds, neglecting the dynamic reconfiguration of hydraulic topology caused by the opening and closing of gates and pumps, making it difficult to directly map simulation results into engineering control variables.

[0025] Furthermore, allocation schemes based on uniformity targets often output static point-value water allocation instructions, ignoring the nonlinear differences in crop sensitivity to water shortage at different growth stages, and lacking a mechanism to decompose macro-risk indicators into micro-execution boundaries. As a result, when faced with uncertainties on both the supply and demand sides, the scheduling scheme cannot establish an executable buffer loop between ensuring water use during critical periods and avoiding systemic risks.

[0026] To address these problems, the present invention will be specifically described through the following embodiments, in conjunction with Figures 1 to 5.

[0027] In some embodiments, an exemplary scheme of an intelligent decision-making method for agricultural water resources in irrigation districts is provided. This method addresses the contradiction between the uncontrollability of physical processes and the unexecutability of risk indicators in existing irrigation district scheduling. It constructs a physical digital foundation, introduces a risk budget management mechanism, and implements rolling closed-loop control, achieving a technical closed loop from uncertainty perception to the issuance of deterministic commands.

[0028] In this application, there are two equivalent writing formats, which have the same meaning and can be used interchangeably. For example, the dimension is marked with square brackets, such as Q[i, t]; the dimension is marked with subscripts, such as Q _i_t .

[0029] Accordingly, this method specifically includes:

[0030] Step S101: Construct a two-layer dynamic graph model of the irrigation area that maps the physical topology and hydraulic connections of the irrigation area. Based on meteorological ensemble forecast data and parameter uncertainty distribution, use the two-layer dynamic graph model of the irrigation area to generate a water supply capacity scenario package containing multiple spatiotemporal possibilities.

[0031] Specifically, traditional distributed hydrological models often treat irrigation districts as static natural watersheds, neglecting the decisive influence of human engineering control on water flow paths. Therefore, the two-layer dynamic graph model of the irrigation district in this embodiment is a mathematical model specifically designed for controlled water conservancy systems. Here, "two-layer" refers to logically separating natural hydrological processes (such as field infiltration and evaporation) from artificial hydraulic processes (such as canal water conveyance and gate control) but physically coupling them; "dynamic graph" means that the model's topology (i.e., the connections between nodes) is not fixed but evolves in real time with the engineering control state (such as gate opening or closing).

[0032] For example, when the inlet gate of a branch canal is closed, the branch canal and its downstream fields are disconnected from the main canal in the graph model, and the corresponding elements in the adjacency matrix become zero. Based on this model, the system inputs meteorological ensemble forecast data (e.g., ensemble forecasts containing 50 different rainfall probability distributions) and the uncertainty distribution of model parameters (e.g., confidence intervals for canal roughness and soil saturated hydraulic conductivity) to perform multi-scenario simulations.

[0033] Based on this, the result of the extrapolation is not a single deterministic value, but rather a scenario package of water supply capacity. This scenario package encapsulates the distribution of water inflow capacity at each node of the irrigation district, the distribution of water loss in the channels, and the probability weights of each scenario under different possible scenarios. This process quantifies physical uncertainty into a standard data structure that can be processed by subsequent algorithms.

[0034] It should be understood that this step is used to establish a physical digital base for the irrigation district.

[0035] Step S102: Generate water demand scenario based on crop water demand characteristics, set the total risk ceiling for the entire irrigation area, and decompose the total risk ceiling into a risk budget distribution matrix with spatiotemporal unit granularity.

[0036] Accordingly, based on crop water requirements such as crop type, growth stage, and current soil moisture, and combined with weather forecasts, multiple sets of possible water demand scenarios are generated, reflecting the distribution of water demand in the irrigation district over a future period. Next, managers set a total risk ceiling for the entire irrigation district, typically a value representing the maximum tolerable water shortage loss or default probability for the entire irrigation district. For example, the maximum allowable conditional value of risk (CVaR) for the entire irrigation district is a loss of 1 million yuan.

[0037] Furthermore, based on the varying importance of each spatiotemporal unit (i.e., a specific field area within a specific time period), the system breaks down the total risk ceiling and allocates it to each spatiotemporal unit, forming a risk budget distribution matrix. In this matrix, units in critical growth stages (such as rice heading) or those planting high-value crops will be allocated lower risk tolerance (i.e., a stricter risk budget), or will receive priority in meeting their needs when total risk is limited; while units in non-critical stages or those growing drought-resistant crops may bear a larger share of the risk. This mechanism, in risk scenarios involving water scarcity, ensures that limited safety margins are prioritized for those more in need of protection.

[0038] It should be understood that risk management was performed in this step, and the risk control objectives were implemented at the spatiotemporal unit level.

[0039] Step S103: Based on the risk budget distribution matrix and the water supply capacity scenario package, solve to generate the water supply interval contract. The water supply interval contract defines the lower bound, upper bound, and allowable risk level of water supply for each spatiotemporal unit.

[0040] Correspondingly, traditional water allocation schemes typically only provide a single water allocation value (point value), which is difficult to strictly implement under supply and demand fluctuations. Based on this, this embodiment introduces the concept of a water supply interval contract. Under the premise of satisfying the risk budget distribution matrix, the system combines the water supply capacity scenario package and uses an optimization algorithm to solve for the interval value [WL, WU].

[0041] Among them, the lower limit of water supply (WL) is the basic water supply that must be guaranteed under a high confidence level to ensure the survival of crops; the upper limit of water supply (WU) is the maximum water supply that can be provided under the condition of sufficient water supply to strive for high yield; and the permissible risk level corresponds to the probability of default or tail loss limit agreed in the contract for this unit.

[0042] Furthermore, the generated water supply interval contract is a digital agreement that includes flexible boundaries and risk clauses. As middleware connecting the upper-level configuration and the lower-level scheduling, it not only constrains the scheduling behavior of the lower level, but also gives the lower level a certain degree of operational freedom to cope with real-time disturbances.

[0043] In this step, the risk constraints are transformed into an executable contract.

[0044] Step S104: Establish an engineering constraint set that includes physical constraints on the operation of engineering equipment. Under the premise of satisfying the water supply section contract, solve for the optimal engineering control instruction sequence based on the rolling time domain mechanism and issue it for execution.

[0045] Specifically, an engineering constraint set is established, which includes the hard constraints of all physical equipment, such as the maximum and minimum opening of gates, the adjustment speed limit of gates (to prevent water hammer effect), and the start-stop interval of pump stations. When solving the scheduling problem, the system must simultaneously satisfy two types of constraints: one is the water supply interval contract from the upper level, that is, the water supply must fall within the interval [WL, WU] as much as possible; the other is the engineering constraint set from the lower level.

[0046] To cope with real-time changes, a rolling time-domain mechanism (MPC) can be used. For example, at time t, the system predicts the optimal control sequence for the next T time periods (e.g., the next 24 hours) based on the latest state, but only issues the command for the first time period (time t) (e.g., opening gate 1 to 30%) to the actuator. At time t+1, the system collects the latest data again, re-predicts the next T time periods, and solves the problem.

[0047] In this step, this mechanism can promptly correct model prediction biases, ensuring the system always operates within the bounds of contractual obligations and physical safety. Based on this, the generated optimal engineering control command sequence can drive the operation of automated gates and pumping stations on-site.

[0048] In other embodiments, the specific implementation process of the physical modeling method based on a two-layer dynamic graph model is described, along with the dynamic topology update mechanism and the underlying state-space equations, providing a foundation for realizing the digital controllability of irrigation districts. As an example, this embodiment can perform the following steps:

[0049] Step S201: The two-layer dynamic graph model of the irrigation area is constructed as a directed graph structure G=(V, E, A(t)).

[0050] Specifically, this step defines the mathematical framework of the model. The irrigation district is abstracted as a directed graph, where V represents the set of nodes, E represents the set of edges, and A(t) represents the adjacency matrix that varies with time t. This structure discretizes the continuous physical space into a computationally efficient network topology.

[0051] The node set V includes at least field nodes with independent water balance attributes, channel nodes with water transmission and distribution functions, and engineering control nodes representing gates or pumping stations.

[0052] Specifically, the node set V is divided into three categories of objects with different physical properties, including:

[0053] Field node (V) _fEach field node represents an independent irrigation unit (such as a strip field or a grid field), with attributes such as area, soil parameters, and crop type. It involves vertical water exchange, such as rainfall, evaporation, and infiltration.

[0054] Channel Node (V) _c (), representing the canal segments of water conveyance channels at various levels, has attributes such as length, cross-sectional shape, and roughness, and involves longitudinal water flow transportation;

[0055] Engineering control node (V) _g ), representing control structures such as gates and pumping stations located at the head of the canal, the branch point, or the connection point, are the active actors in the system, and their status (opening degree, flow rate) directly determines the distribution and direction of water flow.

[0056] Based on this, the model distinguishes between the water storage unit, the water delivery unit, and the control unit.

[0057] The edge set E includes at least the water supply edge connecting the channel and the field, the overflow edge connecting the adjacent field nodes, and the loss edge representing water conveyance loss.

[0058] Specifically, the edge set E describes the physical path of the water flow. The water supply edge is a directed edge from the channel node to the field node, representing the irrigation water flow from the branch point. The overflow edge is the edge connecting adjacent field nodes; when the water depth on the field surface exceeds the height of the field ridge, the water flows through this edge from the higher field to the lower field. Its flow rate is usually calculated using the broad-crested weir formula, i.e., Q. _drain =C _w ×L _bund ×(d _i -h _bund -d _j ) 1.5 C _w L is the flow coefficient. _bund d is the length of the field ridge. _i d _j The water depths of the upstream and downstream fields are respectively, h. _bund The relative height of the field ridge; or, Q _drain C represents the overflow from field i to field j. _w L is the flow coefficient of a broad-crested weir. _bund d is the effective flow length of the field ridge; _i h is the water depth of upstream field i; _bund The relative height of the field ridge; d _j The water depth of the downstream field j.

[0059] In this context, the loss edge is a virtual edge that points from the channel node to the external environment (sink), used to characterize water loss during the water conveyance process. This edge configuration ensures that the loss is no longer an implicit deduction, but rather a tangible flow that can be monitored and calculated.

[0060] Furthermore, A(t) is a time-varying adjacency matrix used to describe the hydraulic connectivity between nodes in the node set V. The values ​​of the elements of the time-varying adjacency matrix A(t) are not fixed constants, but are dynamically determined by the engineering control state variables of the engineering control nodes.

[0061] In other words, A(t) is a time-varying adjacency matrix, also known as an adjacency matrix, which represents the hydraulic connectivity state between nodes in the node set. Its element values ​​are non-fixed constants and can be dynamically determined by the engineering control state variables of the engineering control nodes.

[0062] Specifically, assuming u _ij (t) represents the control state variable (e.g., opening degree) of the engineering control node (such as a gate) located between node i and node j at time t, with a value range of [0, 1]. The elements a in the adjacency matrix A(t) are... _ij (t) and u _ij (t) has a functional mapping relationship. For example, we can define a _ij (t)=u _ij (t). When u _ij (t)=0, meaning when the gate is closed, a _ij (t)=0, meaning that nodes i and j are hydraulically disconnected at this moment, and the graph's topology breaks; when u _ij (t)>0, meaning when the gate is open, a _ij (t)>0, a connection is established between nodes.

[0063] Furthermore, when the engineering control state variables change, the corresponding elements in the time-varying adjacency matrix A(t) are updated accordingly, mapping the physical engineering control behavior to the mathematical graph topology changes.

[0064] Specifically, at each time step of the simulation or optimization, the system reads the current engineering control commands and updates the time-varying adjacency matrix in real time accordingly. For example, in the rotational irrigation mode, the system sequentially opens and closes the gates of different branch canals. In the model, this is represented by the migration of non-zero values ​​of corresponding sub-blocks in the adjacency matrix over different time periods. This mapping mechanism enables the mathematical model to accurately capture the changes in water flow topology caused by engineering scheduling behaviors such as rotational irrigation and intermittent irrigation, thus achieving the simulation of controlled water conservancy systems.

[0065] Step S202: Establish field water balance equations for field nodes to describe the dynamic changes in field surface water depth with rainfall input, evaporation and transpiration output, and infiltration output.

[0066] Specifically, for each field node i, its state variable is the water depth d on the field surface. _i (t). The water balance equation can be expressed as:

[0067] A _i ×(d(d _i ) / dt)=P _i (t)-ET _i (t)-F _i (t)+Q _in_i (t)-Q _out_i (t);

[0068] Among them, A _i P represents the area of ​​the field. _i (t) represents rainfall input, ET _i (t) represents the evaporative output, F _i (t) represents the infiltration output, Q _in_i and Q _out_i These represent the inflow and outflow of water, respectively. In other words, A _i Let d be the area of ​​the i-th plot; _i ) / dt is the rate of change of the field surface water depth over time; P _i (t) represents the rainfall input intensity; ET _i (t) represents the evaporative transpiration output intensity; F _i (t) represents the soil infiltration rate; Q _in_i and Q _out_i These correspond to the amount of irrigation water flowing into and out of the field, respectively.

[0069] As an alternative implementation method, infiltration output can be calculated using the Green-Ampt model. This model, based on soil physical parameters, can accurately describe the decay of infiltration rate over time. Its infiltration rate f(t) satisfies an implicit equation, specifically:

[0070] K _s ×(tt _p +t _0 )=F(t)-φ×Δθ×ln(1+F(t) / (φ×Δθ));

[0071] Among them, K _s t represents the saturated hydraulic conductivity of the soil; t represents the current time; t _p t is the time when water accumulation begins. _0 φ is the equivalent time corresponding to the initial infiltration; F(t) is the cumulative infiltration; φ is the suction head at the soil wetting front; Δθ is the difference between the soil saturated moisture content and the initial moisture content.

[0072] In some alternative implementations, the infiltration output can also be calculated using the Kostiakov empirical formula, i.e., F _i (t)=k×t -β Where k and β are empirical fitting parameters. Alternatively, k can be described as an empirical coefficient reflecting soil permeability; t as infiltration time; and β as an empirical index reflecting infiltration attenuation characteristics.

[0073] Step S203: Establish open channel hydraulic equations for channel nodes to describe the dynamic changes in channel water level or flow rate with upstream inflow and headwater losses.

[0074] Specifically, for channel nodes, simplified forms of Saint-Venant's equations, such as the kinematic wave equation or the spreading wave equation, are typically used to describe the evolution of water flow along the channel. Taking the kinematic wave equation as an example, its continuity equation is:

[0075] ;

[0076] Furthermore, the dynamic equation simplifies to S _f =S _0 That is, the friction gradient equals the bottom slope. This is combined with Manning's formula Q=(1 / n)×A×R. 2 / 3 ×S _0 1 / 2 Differential equations can be established to describe the changes in channel water level or flow rate over time and space. Here, A is the cross-sectional area of ​​the channel, i.e., the cross-sectional area where the water flows into contact with the channel wall; t is time; Q is the flow rate of the channel cross-section, i.e., the amount of water passing through that cross-section per unit time; x is the distance along the channel axis, usually taken along the flow direction with the channel origin as the origin; and q... L The lateral inflow or outflow (i.e., loss) per unit length is represented by n, which corresponds to the Manning roughness coefficient, R to the hydraulic radius, and S. _0 Corresponding channel bottom slope.

[0077] In the open channel hydraulic equations, the head loss is not a single variable, but is structured and decomposed into identifiable loss components.

[0078] Specifically, to improve the interpretability of the model and the specificity of parameter correction, this embodiment does not treat the loss term in a general way, but decomposes it into components with clear physical meaning. For example, the total loss Q... _loss It can be decomposed into Q _loss =Q _seep +Q _evap .

[0079] Furthermore, the identifiable loss components include at least the leakage loss term related to permeability and wetted perimeter, and the evaporation loss term related to water surface area and meteorological conditions; when generating the water supply capacity scenario package, the identifiable loss components participate as independent parameters in the uncertainty disturbance.

[0080] Optionally, the leakage loss item Q _seep The following formula can be used for calculation:

[0081] Q _seep =k _soil ×L _channel ×P _wet (H);

[0082] In the formula, Q _seep For the flow rate lost due to seepage in the canal section; k _soil L is the permeability of the canal bed soil, or the seepage loss coefficient. _channel P represents the length of the canal section. _wet (H) represents the wetted perimeter of the channel as the water level H changes.

[0083] Optionally, the evaporation loss term Q _evap The calculation process can be described by the following formula: Q _evap =E _rate ×W _top (H)×L _channel ;

[0084] In the formula, Q _evap E is the flow rate lost due to evaporation from the water surface in the canal section. _rate W represents the evaporation rate per unit area of ​​water surface under current meteorological conditions. _top (H) represents the width of the channel surface as the water level H changes; L _channel This refers to the length of the canal section.

[0085] When generating the water supply capacity scenario package, the model will adjust k. _soil and E _rate Apply uncertainty perturbations respectively (e.g., assume k) _soil Following a normal distribution, multiple scenarios with different loss characteristics are generated. This decomposition allows subsequent feedback corrections to distinguish between excessive leakage and excessive evaporation, thus more accurately adjusting model parameters.

[0086] Step S204: Combine the field water balance equation with the open channel hydraulic equation to construct a state-space equation with engineering control state variables as control inputs and the distribution of water level and flow rate at all nodes in the irrigation area as the system state.

[0087] Specifically, the water depth d of all field nodes and the water level h / flow rate / q of all canal nodes in the entire irrigation district are combined into a high-dimensional state vector X. The opening degree of all gates and the flow rate of all pumping stations are combined into a control vector U. External meteorological conditions such as rainfall and evaporation rate are combined into a disturbance vector W. By simultaneously solving the differential equations of all the above nodes, the state-space equations of the entire irrigation district system can be obtained, in the following form:

[0088] dX / dt=f(X,U,W,Θ);

[0089] Here, Θ is a set of physical parameters including roughness and hydraulic conductivity, expressing how the evolution of the system state X is controlled by U, driven by W, and constrained by Θ. Specifically, for engineering control nodes (gates), their flow rate equations are embedded in this set of equations as boundary conditions connecting the channel nodes, and their specific form is:

[0090] Q_{gate}=u_{gate}×C _d ×A_{gate}×\sqrt{2g|H_{up}-_{down}|}×\text{sign}(H_{up}-H_{down});

[0091] Q _gate =u _gate ×C _d ×A _gate ×(2×g×|H _up -H _down |) 0.5 ×sign(H _up -H _down );

[0092] Among them, Q _gate The flow rate through the gate; u _gate C represents the relative opening of the gate (a control variable, ranging from 0 to 1). _d A is the gate flow coefficient; _gate The flow area is when the gate is fully open; g is the acceleration due to gravity; H_d is the water level upstream of the gate; H _down The value is the water level downstream of the gate; `sign` is a sign function used to determine the flow direction. This formula reflects the physical relationship between flow rate, gate opening, and water level difference.

[0093] Step S205: Based on the state-space equation, the system state trajectory under different meteorological boundary conditions is deduced to generate a water supply capacity scenario package. In other embodiments, based on the state-space equation, the system state trajectory is deduced under given meteorological and parameter disturbances to generate a water supply capacity scenario package.

[0094] Specifically, numerical integration methods (such as the Runge-Kutta method or the finite difference method) are used to solve the aforementioned state-space equations. For each meteorological sample in the meteorological ensemble forecast and each parameter sample in the parameter distribution, the future state trajectory is deduced starting from the initial state. The water supply capacity of each node (i.e., the adjustable water volume of each branch point) and the loss process of each canal section are extracted from this data and summarized to form a water supply capacity scenario package.

[0095] In this embodiment, a two-layer dynamic graph model of the irrigation area is constructed, and control variables such as gate opening are directly mapped to time-varying adjacency matrices. This reconstructs the irrigation area from a passive, static natural watershed into an active and controllable dynamic hydraulic network, realizing real-time linkage between physical topology and control behavior.

[0096] In other embodiments, optional implementations of the supply and demand scenario generation technology are described. Specifically, this addresses how to construct standardized supply and demand scenario packages when facing weather and parameter uncertainties. This embodiment generates multiple scenarios containing probability weights, transforming future uncertainties into a computable data distribution. This solves the problem that using only a single deterministic value for planning leads to an inability to cope with future volatility risks. Specifically, this embodiment includes:

[0097] Step S301: Each scenario in the water supply capacity scenario package shall include at least scenario-based water supply capacity data, scenario-based channel loss data, and corresponding scenario weights.

[0098] Furthermore, the water supply capacity scenario package is a structured dataset used to describe the amount and distribution of water resources available to the irrigation district in future time periods. Specifically, the scenario-based water inflow capacity data corresponds to the sequence of available water volume at each water intake or canal head output by the model. The scenario-based canal segment loss data specifically corresponds to the water loss sequence at each level of the canal during the water conveyance process.

[0099] The loss data is not a fixed percentage, but a dynamic value calculated based on the physical model in the aforementioned embodiments, combined with the water level under different scenarios. For example, in a high water level scenario, the calculated leakage loss will increase accordingly due to the increased wetted perimeter. The corresponding scenario weight is a value between 0 and 1, representing the probability of that scenario occurring, and the sum of the weights of all scenarios is 1.

[0100] Step S302, wherein the scene weights are determined in at least one of the following ways:

[0101] The probability weights are calculated based on the frequency of occurrence of historical meteorological and hydrological data.

[0102] Equal-weighted scenarios are constructed based on preset confidence intervals and equal probability quantiles, and the scenario weights are set to equal values.

[0103] Specifically, this step provides two methods for calculating scenario weights, adaptable to different data bases and computational needs. Optionally, a method based on historical frequency analysis can be used. This involves the system acquiring long-term historical rainfall and inflow data for the irrigation district (e.g., the past 30 years) and performing pattern matching with current short-term weather forecasts. Several historical years most similar to the current forecast trends are then selected, and the frequency of these years in the current season is statistically analyzed.

[0104] For example, if among the 10 selected similar years, 5 years correspond to a low-water scenario, 3 years to a normal-water scenario, and 2 years to a high-water scenario, then the weights for the low-water, normal-water, and high-water scenarios can be set to 0.5, 0.3, and 0.2, respectively. This method can make full use of historical experience data, making the weights statistically significant.

[0105] Optionally, when long-term historical data is lacking or when simplified calculations are required, a method based on equal probability quantiles can be used. That is, the system uses state-space equations to perform Monte Carlo simulations, generating a large number (e.g., 1000) of possible state trajectories. Next, these trajectories are sorted according to the total water supply, and the trajectory at a preset quantile is selected as a representative scenario.

[0106] For example, trajectories at the 10%, 50%, and 90% quantiles are selected to represent extreme drought, normal, and extreme flooding scenarios, respectively. To maintain unbiased calculation, the weights of the three scenarios are all set to one-third, i.e., scenario weight p. _s =1 / 3. This method, through the division of confidence intervals, allows the selected scenarios to cover various possibilities, and the calculation process is standardized and easy to automate.

[0107] Based on this, water requirement scenarios are generated according to the water requirement characteristics of crops, specifically including:

[0108] Furthermore, data on planting structure (including crop type and planting area), crop growth period division, and initial soil moisture values ​​within the irrigation area were obtained.

[0109] Specifically, the planting structure data records the current crop type (such as rice, corn, and wheat) and corresponding planting area in each field within the irrigation district. Crop growth stage data defines the time points for each crop's various growth stages from sowing to maturity; for example, rice can be divided into the greening stage, tillering stage, jointing and booting stage, heading and flowering stage, and milk stage. Initial soil moisture data is collected in real-time by soil moisture sensors buried in the field, or obtained through remote sensing inversion of the current soil moisture content. These types of data collectively constitute the boundary conditions for calculating crop water requirements.

[0110] For example, if a paddy field with a planting area of ​​100 mu is currently in the water-sensitive heading and flowering stage, and soil moisture monitoring shows that the water content is less than 60% of the field water holding capacity, then the initial water demand of the field will be very high.

[0111] Furthermore, by combining meteorological ensemble forecast data, crop water requirement models or data-driven models are used to extrapolate the water requirement process of each field in the future, generating water requirement scenarios with multiple possibilities.

[0112] Accordingly, the water requirement is calculated using the crop coefficient method. For each weather forecast scenario s, the reference crop evapotranspiration ET is calculated using the Penman-Monteith formula. _0 [t, s]; t corresponds to the time step, and s corresponds to the scene number. Next, based on the crop type and growth period, the corresponding crop coefficient K is obtained by looking up a table. _c Crop water requirement ET _c Calculated as ET _c =K _c ×ET _0 [t, s]. Next, field water balance calculations are performed, deducting the effective rainfall P under this meteorological scenario. _eff The net irrigation water requirement D[i, t, s] for this field is obtained by combining [t, s] with the current soil moisture contribution. For the entire irrigation district, the water requirement process of all fields is summarized to form the water requirement scenario.

[0113] According to one aspect of this application, the calculation of net irrigation water requirement can be described by the following formula:

[0114] D[i, t, s]=(K _c ×ET _0 [t, s]-P _eff [t, s])×A _i ;

[0115] Where D[i, t, s] represents the net water requirement of field i at time t in scenario s; K _c ET is the crop coefficient. _0 [t, s] represents the reference crop evapotranspiration at time t under scenario s; P _eff [t, s] represents the effective rainfall at time t in scenario s; A _i This refers to the area of ​​the field.

[0116] Because weather forecasts inherently involve uncertainty—for example, rainfall may have multiple predicted outcomes—the resulting water demand scenarios are also diverse. For instance, in scenarios with abundant rainfall, the calculated water demand will show lower irrigation needs; conversely, in scenarios with high temperatures and low rainfall, the water demand will show peak irrigation needs. These scenarios correspond one-to-one with water supply capacity scenario packages and are jointly input into the subsequent risk budgeting module.

[0117] As an example, a demonstrative scheme for constructing a differentiated evaluation system based on agronomic mechanisms is provided. This system integrates agronomic characteristics and socioeconomic factors, breaking away from the extensive model of traditional water allocation based on average area. It quantifies the sensitivity of different crops and different stages to water shortage, establishes spatiotemporally differentiated weights and loss functions, and guides water resources towards high-value, highly sensitive areas. Accordingly, this scheme can be implemented in the following manner:

[0118] Step S401: Based on the crop coefficient at the current growth stage, the water sensitivity coefficient of yield to water shortage, and the preset spatial priority coefficient, calculate the spatiotemporal differential weight of each spatiotemporal unit to characterize the importance of the unit in terms of yield formation and social impact.

[0119] Specifically, the spatiotemporal differential weight w _i (t) can determine the priority of water resource allocation, and its calculation formula is as follows:

[0120] w _i (t)=K _c_i (t)×K _y_i (t)×π _i ;

[0121] Among them, w _i (t) represents the comprehensive weight of the i-th spatiotemporal unit at time t; K _c_i (t) is the crop coefficient, reflecting the physiological water demand intensity; K _y_i (t) is the moisture sensitivity coefficient, reflecting the degree of impact of water shortage on yield; π _i This is a spatial priority coefficient, reflecting its socio-economic importance.

[0122] The above formula integrates three dimensions of features, the first being the crop coefficient K. _c_i (t) reflects the physiological water requirement intensity of crops at different growth stages. For example, for rice, K during the tillering stage... _c The value may be 1.1, while it drops to 0.9 during the yellow ripening stage. Secondly, there is the moisture sensitivity coefficient K. _y_i (t), reflecting the degree of impact of water shortage on final yield at this stage, is usually based on the FAO-recommended water production function 1-(Y). _a / Y _m )=K _y(1-ET _a / ET _m ) Determine; for example, rice is most sensitive to water during the heading and flowering stage, and the K value at this stage is... _y The value is usually set to 1.0 or higher, at which point water shortage will lead to a significant drop in yield; and at the end of the tillering stage, in order to control ineffective tillering, it is often necessary to dry the field, at which point K _y The value is low, even close to 0. Thirdly, there is the spatial priority coefficient π. _i This reflects the socio-economic importance of the area where the field is located.

[0123] Above, Y _a Y refers to the final actual yield of crops under actual field moisture conditions; _m K represents the potential yield that a crop can achieve when all growth conditions, such as water and nutrients, are met; _y It is a yield response factor, dimensionless, used to characterize the sensitivity of different crops or different growth stages of the same crop to water deficit. The higher the sensitivity, the higher the yield response factor. _y The larger the value, the better; _a This represents the actual evapotranspiration consumed by crops during their actual growth in the field, including both crop transpiration and inter-row evaporation; ET _m Potential evapotranspiration of crops under conditions of sufficient water supply can be calculated using the FAO-56 Penman formula in conjunction with meteorological data and crop parameters. This formula quantifies the linear relationship between water deficit and crop yield reduction, reflecting the degree of impact of water stress on crop yield.

[0124] In some alternative implementations, the spatial priority coefficients can also be determined using the Analytic Hierarchy Process (AHP). Specific steps include: constructing an evaluation system comprising four criteria layers: economic benefits, food security, social equity, and water conveyance distance. Next, experts are organized to conduct pairwise comparisons and scoring of the relative importance of the four criteria, constructing a judgment matrix.

[0125] For example, if food security is considered slightly more important than economic benefits, the corresponding element is assigned a value of 2. Next, the largest eigenvalue of the judgment matrix and its corresponding eigenvector are calculated, and after normalization, the weight vectors for each criterion are obtained. Based on this, combined with the specific scores of each field under each criterion, such as whether it is basic farmland, whether it is cultivated by impoverished households, and its distance from the canal head, a weighted calculation is performed to obtain the comprehensive priority of each field. For example, for areas designated as permanent basic farmland, their food security score is higher, resulting in a higher final π. _i A value greater than 1.0 enhances its competitive advantage in water distribution.

[0126] According to another aspect of this application, a formula for calculating the AHP eigenvector of spatial priority coefficients is provided, specifically as follows:

[0127] A×W=λ _max ×W;

[0128] Where A is the judgment matrix constructed based on economy, food, fairness, and distance; W is the normalized eigenvector, i.e., the weight vector of each criterion; λ _max To determine the largest eigenvalue of a matrix. π _i It is a weighted sum of the weights of each criterion and the specific scores of each field.

[0129] Step S402: Based on the spatiotemporal differentiated weights, a spatiotemporal differentiated water shortage loss function is constructed. The function defines a nonlinear mapping relationship between the degree of water shortage and economic loss, so that spatiotemporal units in the critical reproductive period will generate higher marginal loss values ​​under the same water shortage rate.

[0130] Specifically, traditional water shortage loss calculations assume that the loss is directly proportional to the amount of water shortage, which does not align with actual agricultural production patterns. This embodiment constructs a nonlinear, spatiotemporally differentiated water shortage loss function, denoted as δ. _i_t The water shortage rate is calculated using the formula δ. _i_t =(D _i_t -d _i_t ) / D _i_t D _i_t For water demand, d _i_t This represents the actual water supply. The loss function is defined as L. _i_t (δ)=w _i (t)×(δ _i_t ) α Where α is the loss exponent, which is usually taken to be α≥1.

[0131] In another possible implementation, L _i_t (δ)=w _i (t)×[(D _i_t -d _i_t ) / D _i_t ] α Among them, L _i_t (δ) represents the economic loss caused by water shortage; w _i (t) represents the overall weight; D _i_t Water demand; d _i_t This represents the actual water supply; [(D _i_t -d _i_t ) / D _i_t ] represents the water shortage rate; α is the loss index, which can be selected as 2 to reflect the nonlinear marginal effect.

[0132] In another possible implementation, α=2 is set, making the loss function convex. As the water shortage rate increases, the marginal loss increases. For example, when the water shortage rate increases from 0% to 10%, the yield loss may be small; however, when the water shortage rate increases from 40% to 50%, it may lead to crop death, resulting in a huge yield loss. Combined with w _i (t), this function can accurately characterize the risk features of different units.

[0133] For example, for those in the heading stage (high K) _y And located in a basic farmland protection area (high π) _i The weight w of the rice paddy plots is as follows: _i The value of (t) is large, and due to the effect of α=2, the calculated economic loss will rise sharply once water shortage occurs. This will drive the subsequent risk budget allocation algorithm to prioritize meeting the water demand of this field, or to allocate it with more stringent risk control indicators, thus achieving critical protection for the critical period at the mathematical level.

[0134] In this application, a differentiated evaluation system driven by agronomic mechanisms was established. A nonlinear convex loss function was constructed by integrating crop water sensitivity coefficient and spatial priority, which replaced the traditional linear uniform allocation mode, so that water resources are preferentially tilted towards key growth periods and high-value areas under scarce conditions.

[0135] As another example, an optional implementation method for risk budget allocation and water supply interval contract generation mechanism is described. Accordingly, the concept of risk budgeting from the financial field is introduced and improved in combination with the characteristics of water conservancy projects, realizing the quantification, allocation, and contractual management of risk. This solves the problem that risk control only remains at the planning level and cannot be implemented in specific scheduling operations. Exemplarily, this embodiment may include the following steps:

[0136] Step S501: Initially allocate the total risk ceiling of the entire irrigation district using spatiotemporal differential weights to obtain the initial risk budget.

[0137] Specifically, the total risk ceiling for the entire irrigation district is B. _total This is a total target set by the manager based on the water situation and disaster prevention and mitigation goals for the year, for example, setting the expected economic loss from water shortage in the entire irrigation district to no more than 5 million yuan. To implement this target into specific management units, initial allocation is required. This step uses a weighted proportional allocation method. Assume the entire irrigation district has I spatiotemporal units, meaning different zones at different times, and the spatiotemporal difference weight of the i-th unit is w. _i Then the initial risk budget b obtained by this unit _init [i] is calculated as: b _init [i]=B _total ×(w _i / Σw _j);Σw _j This is the sum of the weights of all units.

[0138] In this allocation method, units with larger weights (usually high-value crops or critical growth periods) may have greater potential losses, but in order to ensure their production safety, the risk tolerance index allocated to them in the initial stage is relatively tight or based on a certain benchmark.

[0139] In this context, risk budget typically refers to the maximum allowable tail loss. Units with high weightings are less likely to incur losses, therefore they should be allocated a smaller risk budget (i.e., stricter constraints); or, if it refers to the amount of risky investment that must be undertaken to obtain high returns, the logic is reversed.

[0140] Optionally, in a risk-averse context, the risk budget is defined as the maximum allowable potential loss. Therefore, the initial allocation strategy is adjusted to an inverse proportionate allocation or a fair allocation based on the Gini coefficient, meaning that units with larger weights receive smaller allowable loss amounts. For example, using formula b... _init [i]=B _total ×((1 / w _i ) / Σ(1 / w _j ));b _init [i] represents the initial risk budget allocated to the i-th unit, i.e., the maximum allowable loss, Σ(1 / w _j The formula represents the sum of the reciprocals of the weights of all units. This means that units with higher weights receive a smaller allowable loss. Rice paddies in their critical growth stages will be allocated lower allowable loss amounts, forcing subsequent algorithms to prioritize their water supply.

[0141] Step S502: Based on the water supply capacity scenario package and the water demand scenario, calculate the tail risk contribution of each spatiotemporal unit under the current budget.

[0142] Specifically, after establishing the initial budget, the system needs to assess the actual risk exposure faced by each unit under the current supply and demand uncertainty. Using water supply and demand scenarios, the distribution of water shortage losses for each unit is simulated and calculated. Next, the marginal contribution of each unit to the total risk of the entire irrigation district is calculated. Here, the tail risk contribution ΔCVaR[i] is introduced. If Conditional Value at Risk (CVaR) is used as the metric, the tail risk contribution represents the average loss generated by that unit when the entire system is in the worst-case scenario of Ъ%. In other words, Ъ% is the probability threshold defining the tail extreme scenario, a commonly used parameter in risk analysis, usually taken as a small value, such as 5% or 1%.

[0143] By comparing the tail risk contribution with the initial risk budget, it is possible to identify which units are the hardest hit by risk (i.e., the contribution far exceeds the budget) and which units still have a safety margin.

[0144] Step S503: The initial risk budget is redistributed according to the tail risk contribution, so that the critical period spatiotemporal units with high marginal loss sensitivity can obtain the risk budget amount to be satisfied first, and a risk budget distribution matrix is ​​generated.

[0145] Among them, the critical period spatiotemporal unit can also be called the critical unit. It has a high marginal loss sensitivity, that is, the water shortage loss of the critical period spatiotemporal unit (such as the field during the critical growth period of crops) will increase rapidly with the increase of water shortage. It reacts more violently to marginal changes in water shortage and should be given priority in risk budget allocation.

[0146] Alternatively, the initial risk budget is redistributed based on the tail risk contribution, so that key units with high marginal loss sensitivity receive priority risk budget allocations, generating a risk budget distribution matrix; where key units are spatiotemporal units in the critical growth stage of crop water requirements.

[0147] Step S504, the process of reallocating the initial risk budget based on the tail risk contribution, may be performed in any of the following ways:

[0148] On the one hand, a heuristic iterative strategy can be adopted to identify non-critical units whose tail risk contribution is lower than a preset threshold, and migrate their remaining risk budget to critical units with high risk contribution, until the risk constraint satisfaction rate of all units converges.

[0149] In actual operation, some non-critical units (such as fields in fallow period) may receive budget allocations but not face water shortage risks, resulting in budget waste; while critical units (such as rice in the heading stage) may face significant water shortage pressure, leading to budget overruns. A heuristic iterative strategy simulates an internal risk trading market. The algorithm flow is as follows:

[0150] Calculate the budget utilization rate r for each unit. _i =ΔCVaR[i] / b[i]; where b[i] corresponds to the quantitative index of the risk tolerance limit of unit i, which is the maximum tail risk (measured by CVaR value) that the unit is allowed to bear. Further, r is identified. _i <1 surplus units and r _i A deficit unit greater than 1.

[0151] Next, according to w _i The priority defined by the weights is used to transfer the budget amount Δb from the surplus unit to the deficit unit, updating b[i]. ΔCVaR and r are repeatedly calculated. _i until all critical units r _i The goal is to get as close to or less than 1 as possible, or to reach the maximum number of iterations. This method is intuitive and converges easily, quickly finding an allocation scheme that fits the actual engineering requirements.

[0152] On the other hand, the risk budget constraint can also be transformed into a Lagrange dual problem, treating the risk budget distribution matrix as the original variable to be solved, and automatically adjusting the budget allocation of each spatiotemporal unit by updating the Lagrange multipliers. This is suitable for large-scale optimization problems.

[0153] Accordingly, a system is constructed with the goal of maximizing the total utility of the entire irrigation district, and with a total risk ceiling of B. _total For the optimization model with coupling constraints, construct the Lagrangian function L(x, λ) = f(x) + λ(Σb[i] - B). _total ), where λ is a Lagrange multiplier, which has the economic meaning of risk shadow price. That is to say, x is the decision variable of the optimization model, which can refer to the control parameters such as the risk budget amount of each spatiotemporal unit in the scenario of risk budget allocation in the irrigation area; f(x) is the objective function, which specifically corresponds to the quantitative calculation formula for maximizing the total utility of the entire irrigation area; λ is a Lagrange multiplier, which has the economic meaning of risk shadow price, and its value reflects the degree of scarcity of risk resources in the entire system.

[0154] The algorithm iteratively updates λ using either the subgradient method or the dual ascent method. When λ increases, the risk resources of the entire system become scarce, and the algorithm automatically compresses b[i] of low-efficiency units. When λ decreases, the risk resources become abundant, and the constraints can be appropriately relaxed. The b[i] at convergence is the optimal risk budget distribution matrix.

[0155] Step S505: For each spatiotemporal unit, with the risk budget amount corresponding to the risk budget distribution matrix as a hard constraint, and combined with the spatiotemporally differentiated water shortage loss function, optimization is performed under all scenario sets of the water supply capacity scenario package.

[0156] Specifically, after determining the risk budget b[i, t] for each unit, it is necessary to solve for the water supply contract interval of that unit. A local optimization model is constructed as follows: the objective function is to maximize the width of the water supply interval or minimize the deviation between the lower limit of water supply and the demand. The constraint condition is that the risk metric Risk(L) ≤ b[i, t]. The risk metric is constrained according to one of the following risk metrics.

[0157] Optionally, Conditional Value at Risk (CVaR) can be used, which requires that, at a given confidence level α, the expected value of the tail excess loss of each spatiotemporal unit does not exceed the budget amount allocated to it in the risk budget distribution matrix.

[0158] Specifically, CVaR (Conditional Value at Risk) measures tail-end extreme risk. For a given spatiotemporal unit, let its water shortage loss in scenario s be Z. _s It can be calculated by the loss function, with scene weights being p. _s The discretization formula for CVaR is:

[0159] CVaR _α =min _ξ {ξ+(1 / (1-α))×Σ _s=1 S p _s ×max(0, Z) _s -ξ)};

[0160] In optimization models, auxiliary variables v are typically introduced. _s To linearize the max function by setting ≥0, the constraint condition is transformed into: ξ+(1 / (1-α))×Σp _s ×v _s ≤b[i,t],v _s ≥Z _s -ξv _s The form ≥0 can be conveniently embedded in a linear programming solver, so that even in severe (1-α) probability scenarios (such as when a severe drought occurs), the average loss of the unit is strictly controlled within the budget b[i, t].

[0161] In other words, the discretized formula for Conditional Value at Risk (CVaR) is:

[0162] CVaR _α =ξ+(1 / (1-α))×Σ(p _s ×v _s );

[0163] The constraint is: v _s ≥Z _s -ξ and v _s ≥0;

[0164] The above is about CVaR. _α ξ represents the conditional value at confidence level α; ξ is the value at risk (VaR), which is the optimization variable; p _s The probability weights for the occurrence of scenario s; v _s Z is an auxiliary variable for the tail excess loss in scenario s; _s Σ represents the actual water shortage loss value under scenario s; Σ represents the summation over all scenarios s, and S represents the total number of scenarios to be analyzed.

[0165] Optionally, a chance-constrained satisfaction rate can be adopted, requiring that the probability that the actual water supply of each spatiotemporal unit is not lower than the lower limit of water supply is greater than a preset satisfaction rate threshold η. This is suitable for scenarios with low risk sensitivity.

[0166] Specifically, it can be described as the following expression:

[0167] Prob(Deliver[i,t]≥WL[i,t])≥η;

[0168] Here, Prob represents the probability calculation function, used to measure the likelihood of a water supply event occurring; Deliver[i,t] represents the actual water supply of the i-th field at the t-th time step, reflecting the water supply execution result of this spatiotemporal unit; WL[i,t] represents the lower limit threshold of water demand for the i-th field at the t-th time step, which is usually determined by the minimum water demand or irrigation guarantee rate requirement during the critical growth period of the crop. This formula is used to quantify the probability that the water supply of a field meets the minimum water demand requirement; the closer the value is to 1, the higher the water supply guarantee level of the field.

[0169] In discrete scenarios, a 0-1 variable y is introduced. _s When Deliver _s ≥WL when y _s =1, otherwise y _s =0. The constraint condition is Σp _s ×y _s ≥η. Although integer variables are introduced, the solution efficiency is still acceptable when the number of scenarios is limited. This metric ensures the reliability of water supply compliance, for example, guaranteeing a 95% probability that the water supply will not fall below the lower bound.

[0170] In other embodiments, the opportunity constraint satisfaction rate is calculated using the following formula:

[0171] P _satisfy =Σ(p _s ×y _s )≥η;

[0172] Among them, P _satisfy To satisfy the cumulative probability of the lower limit of water supply; p _s The weights for scene s; y _s η is a 0-1 variable, which is set to 1 when the water supply in scenario s is greater than or equal to the lower limit of water supply, and 0 otherwise; η is a preset satisfaction rate threshold.

[0173] Step S506: Calculate the lower and upper limits of water supply that satisfy the hard constraints, so that the lower limit of water supply can cover the basic water demand requirements under the preset credit level, and the adjustment range formed by the upper and lower limits of water supply is within the executable range of the engineering facilities.

[0174] Specifically, after optimization, two key values ​​are obtained for each unit: the lower limit of water supply WL and the upper limit of water supply WU. Among them, WL usually corresponds to the water supply baseline that must be maintained under severe conditions, such as the minimum water requirement to just keep crops alive; WU corresponds to the ideal water supply under conditions of abundant water.

[0175] Furthermore, it is necessary to verify whether the width of [WL, WU] meets the requirements for engineering feasibility. For example, if the calculated WL = 10.1m3 / s, WU=10.2m 3 / s, with a range width of only 0.1m. 3 / s, while the minimum regulating flow rate accuracy of the channel gate is 0.5m³ / s. 3 If the value is / s, then the contract is unexecutable. In this case, the algorithm needs to forcibly widen the interval, for example, adjust it to [10.0, 10.5], and reverse the check to see if it violates the risk budget.

[0176] Step S507 encapsulates the lower and upper limits of water supply and the corresponding risk limits into a water supply interval contract, which serves as the sole basis for performance of subsequent engineering control.

[0177] Accordingly, the generated water supply interval contract is a standardized data package. Its specific data structure can be defined as follows:

[0178] WaterContract={UnitID: Field_01, TimeSlot: 2025-06-01 08:00-12:00, LowerBound_WL: 0.5m 3 / s, UpperBound_WU: 0.8m 3 / s, RiskBudget_b: 2000CNY, LossFunctionType: Quadratic, ConfidenceLevel_α: 0.95};

[0179] In other words, UnitID is the water-using unit number, Field_01 is the unique identifier corresponding to the first field, used to distinguish different water users; TimeSlot is the water supply period, 2025-06-01 08:00-12:00, which specifies the specific time range for this water supply; LowerBound_WL is the lower limit threshold for water supply flow, 0.5m 3 / s is the minimum water supply flow rate to ensure normal crop growth; UpperBound_WU is the upper limit threshold for water supply flow rate, 0.8m. 3 / s is the maximum water supply flow rate to avoid flooding or water waste; RiskBudget_b is the risk budget for the corresponding time period of the water unit, and 2000CNY is the maximum economic loss allowed for the unit due to water supply deviation; LossFunctionType is the loss function type, Quadratic indicates that a quadratic function is used to characterize the non-linear relationship between water supply deviation and loss; ConfidenceLevel_α is the confidence level of risk assessment, and 0.95 indicates that at a 95% confidence level, the risk loss of the unit will not exceed the preset budget.

[0180] The contract defines the rights and obligations of the dispatching layer. Specifically, the dispatching layer has the right to freely adjust water supply within the range of [0.5, 0.8] based on real-time water conditions, but must ensure that the long-term default risk index does not exceed 2000 yuan. This mechanism achieves decoupling between management setting rules and execution layer performing operations.

[0181] According to another aspect of this application, the Lagrange multiplier update formula can be expressed as:

[0182] λ _k+1 =max(0, λ) _k +γ _k ×(Σ(b _i )-B _total ));

[0183] Where, λ _k+1 λ is the Lagrange multiplier for the (k+1)th iteration; _k γ is the multiplier for the k-th iteration; _k Σ(b) is the step size for the k-th iteration; _i B represents the sum of the current risk budgets for all units; _total This represents the upper limit of total risk. This formula is used to automatically adjust the risk shadow price in the dual algorithm.

[0184] In summary, this solution proposes a risk budget allocation and water supply interval contract mechanism, which quantifies and decomposes the total risk of the entire system into risk budgets and water supply interval contracts for each spatiotemporal unit. Combined with a two-stage rolling time-domain control strategy, it successfully transforms risk constraints into specific gate pump scheduling instructions that meet engineering physical limitations, thus achieving a complete closed loop of supply and demand uncertainty, risk controllability, and instruction execution.

[0185] As another example, this paper describes the specific implementation method of rolling time-domain control and two-stage optimization solution strategy. In particular, it describes how to carry out specific engineering scheduling based on water supply interval contracts, solving the problems of multi-objective conflicts and real-time disturbance correction in complex canal systems.

[0186] Furthermore, under the premise of satisfying the water supply interval contract, the optimal engineering control command sequence is solved based on a rolling time-domain mechanism, including:

[0187] Step S601: Construct a comprehensive optimization objective function that includes a contract breach penalty term and an engineering operation cost term. The contract breach penalty term is used to penalize behaviors that deviate from the lower and upper limits of the water supply in the water supply interval contract, and the penalty weight for deviation from the lower limit of the water supply is higher than the penalty weight for deviation from the upper limit of the water supply. The engineering operation cost term is used to characterize the frequency of gate regulation, pump station energy consumption, or water transmission loss.

[0188] Specifically, at each decision point in rolling control (MPC), the system needs to solve the optimal control problem. Its objective function J is defined as:

[0189] J=J _penalty +J _cost ;

[0190] Among them, J _penalty This is a penalty for breach of contract, used to ensure that water supply behavior falls within the contractual interval [WL, WU]. Considering that the harm of insufficient water supply is usually greater than that of excessive water supply, an asymmetric piecewise penalty function is adopted, namely:

[0191] J _penalty =Σ _i，t [λ _1 ×max(0, WL[i, t] - Q[i, t]) 2 +λ _2 ×max(0, Q[i,t]-WU[i,t]) 2 ];

[0192] In the formula, Q[i,t] is the actual water supply flow rate, and λ _1 and λ _2 Let λ be the penalty coefficient. _1 >>λ _2 For example, λ _1 =100, λ _2 =10, falling below the lower bound WL will be severely punished. Σ _i，t This represents the summation calculation over all field units i and all time steps t; WU[i,t] is the upper limit threshold for water supply for the i-th field at the t-th time step.

[0193] Among them, J _cost This is an engineering operation cost item, used to ensure the stability and economy of scheduling, namely:

[0194] J _cost =Σ _m，t [c _1 ×|u[m,t]-u[m,t-1]|+c _2 ×Power[m,t]];

[0195] In the formula, the first term penalizes drastic fluctuations in the control variable u (such as gate opening) to prevent equipment wear and hydraulic oscillations; the second term represents the energy consumption cost of the pumping station. Or, Σ _m，t This indicates a summation calculation over all control device units m and all time steps t; c _1The penalty coefficient for the fluctuation of the control variable is used to quantify the cost of equipment wear and hydraulic oscillation losses caused by sudden changes in parameters such as gate opening; u[m, t] is the value of the control variable of the m-th equipment unit at the t-th time step, u[m, t-1] is the value of the control variable of the equipment unit at the previous time step, and the absolute value of the difference between the two, |u[m, t] - u[m, t-1]|, characterizes the fluctuation amplitude of the control variable; c _2 is the cost coefficient for unit energy consumption of the pumping station, and its value is related to the electricity price; Power[m,t] is the actual output power of the m-th pumping station at the t-th time step.

[0196] Step S602: At each rolling decision moment, with the goal of minimizing the comprehensive optimization objective function, solve the sequence of control variables in the future prediction time domain under the constraints of the engineering constraint set, and issue the first control action of the sequence as the optimal engineering control command sequence.

[0197] Optionally, solving for the sequence of control variables in the future prediction time domain under the constraints of the engineering constraint set can be performed using a two-stage hierarchical optimization strategy, specifically as follows:

[0198] Phase 1 is to find a feasible minimum guarantee solution; that is, to maximize the satisfaction of the lower limit of water supply in the water supply contract as the sole objective, and to find a minimum guarantee control scheme under the constraints of the engineering constraint set, so as to ensure that the basic survival water needs of the irrigation area do not default.

[0199] In practical engineering, directly solving nonlinear programming problems involving complex objective functions may yield no solution due to constraint conflicts. To ensure the robustness of system operation, a relaxation model is proposed. This model temporarily ignores operating costs and the upper bound of water supply WU, focusing only on satisfying the lower bound of water supply WL. The objective function is minΣ. _i，t max(0, WL[i,t]-Q[i,t]); The constraints are the engineering physical constraints (gate opening limits, water level limits, etc.).

[0200] This step yields the baseline control scheme. While this scheme may not be the most economically optimal, it is the safest and can largely prevent catastrophic water shortage events. The system will define a feasible region space based on this scheme, restricting subsequent optimizations to fine-tuning within this framework and preventing excessive deviations.

[0201] Phase two involves risk optimization; that is, locking in the feasible domain space determined by the minimum guarantee control scheme, and performing secondary optimization within this space with the goal of minimizing the sum of contract default penalty items and project operation cost items to obtain the risk optimization control scheme, and using the control actions corresponding to the scheme as the optimal project control instruction sequence.

[0202] Specifically, this stage focuses on improving efficiency. Within the feasible region (e.g., limiting the control variable u to within ±10% of u_base, and ensuring Q ≥ WL),... _a (chieved), and substitute into the objective function J for the search. Utilize the remaining adjustment capacity to approximate WU as much as possible to increase crop yield while satisfying WL, and simultaneously smooth gate action to reduce costs.

[0203] Based on this, the obtained u_opt[t] is the optimal solution that balances safety and economic benefits. Under the rolling time domain framework, the system calculates the sequence u_opt[t], u_opt[t+1], ..., u_opt[t+T-1] for the next T time periods, but only sends u_opt[t] to the SCADA system for execution.

[0204] Alternatively, solving for the sequence of control variables in the future prediction time domain under the constraints of the engineering constraint set can also be performed using a decomposition and coordination optimization strategy. This involves decomposing the overall irrigation district optimization problem into a canal system water distribution subproblem and a zone contract delivery subproblem. The canal system water distribution subproblem handles the engineering constraint set, while the zone contract delivery subproblem handles the contract constraints between water supply zones. This approach is suitable for large-scale systems with large irrigation district areas and multiple canal levels (main canals - branch canals - distribution canals). Directly modeling the entire system would lead to an explosion in the number of variables.

[0205] This step decouples the problem into two sub-problems. The first is the canal system water transport and distribution sub-problem, focusing on the hydraulic transport of main canals and branch canals, and handling physical constraints (Saint-Venant equations, gate characteristics). The goal is to transport water from the source to various distribution points, minimizing water transport costs and losses.

[0206] The second issue is the sub-problem of contract delivery in different zones. This involves focusing on contract execution in downstream fields, using each water diversion point as a boundary, and addressing risk constraints (CVaR, contract interval). The goal is to utilize the water diverted from each water diversion point to meet contract requirements as much as possible.

[0207] Furthermore, Lagrange multipliers or coordination variables are introduced to iterate alternately between the two subproblems until the flow rate and head of the coupled node meet the convergence condition, and the sequence of control variables is output.

[0208] Specifically, the flow rate and water level at the water inlet are defined as coordination variables. The iterative process is as follows:

[0209] The canal system subproblem provides a trial flow rate at the branch point and distributes it to each subproblem in the region.

[0210] For each subproblem in a given trial flow rate, optimize the internal water allocation and calculate the marginal benefit at that flow rate, i.e., the Lagrange multiplier λ. _k Feedback to the canal system sub-issue;

[0211] The canal sub-problem is based on the feedback λ _k Adjust the flow allocation trial flow, and direct it to λ. _k Increase traffic in high-traffic areas;

[0212] Repeat the above process until the flow rate and λ are calculated. _k There will be no major changes. This method breaks down the problem into smaller parts and divides it into manageable parts, making accurate and real-time control of large-scale irrigation areas possible.

[0213] In other embodiments, the comprehensive optimization objective function formula can also be calculated using the following formula:

[0214] J=Σ[λ _1 ×max(0, WL) _i_t -Q _i_t ) 2 +λ _2 ×max(0, Q) _i_t -WU _i_t ) 2 ]+

[0215] Σ[c _1 ×|u _m_t -u _m_t-1 |+c _2 ×Power _m_t ];

[0216] Where J is the total objective function value; WL _i_t The lower boundary of the contracted water supply; WU _i_t The upper limit of the contracted water supply; Q _i_t λ represents the actual water supply. _1 λ is the penalty coefficient for deviating from the lower bound; _2 The penalty coefficient for deviating from the upper bound; u _m_t Let u be the control variable for the m-th device at time t; _m_t-1 c is the control variable from the previous time step. _1 c is the cost coefficient for operational stability. _2 Power is the energy cost coefficient. _m_t This refers to the energy consumption of the pumping station.

[0217] According to one aspect of this application, an exemplary scheme for online closed-loop correction based on model residuals is provided. Specifically, it describes how to utilize real-time feedback data from the field to correct water supply scenario packages, enabling the system's adaptive evolution. This can be used to address problems such as seasonal drift of hydrological parameters and changes in characteristics due to equipment aging.

[0218] Furthermore, after collecting and executing on-site observation data, the model residuals are calculated, and online correction of the water supply capacity scenario package is performed based on the model residuals. Specifically, this includes:

[0219] Step S701: Collect the headwater flow, the actual water volume at the dividing point, and the actual operating status of the gate during the actual execution process, and use them as on-site execution observation data.

[0220] Specifically, the system acquires measured values ​​from field sensors via the IoT platform immediately after the command is issued and executed. These values ​​include: readings from the ultrasonic flow meter at the canal head, readings from the electromagnetic flow meters at each branch outlet, and the actual position reported by the gate opening sensor. Furthermore, remote sensing technology can be used to obtain estimated values ​​for evaporation from the canal surface.

[0221] Step S702: Align and compare the field observation data with the predicted output of the irrigation district two-layer dynamic map model, calculate the flow deviation and transmission delay deviation, and generate model residuals.

[0222] Specifically, the system retrieves the current state value Q predicted by the model at the previous time step. _pred The two types of residuals are calculated as follows:

[0223] Flow residual ε _Q =Q _obs -Q _pred If ε _Q A consistently negative ε indicates that the model overestimates water supply capacity or underestimates losses. _Q The flow residual value represents the degree of deviation between the simulated flow rate and the actual observed flow rate; Q _obs The actual observed flow rate at a channel cross-section is the true value obtained through monitoring equipment such as flow meters.

[0224] Delay residual ε _T =T _arrival_obs -T _arrival_pred By comparing wavefront arrival times, the accuracy of the model's roughness parameter n is evaluated. If the water flow arrives slower than predicted (ε... _T A value greater than 0 indicates that the actual roughness may be greater than the model's set value.

[0225] Where, ε _T The time delay residual is used to assess the simulation bias in the arrival time of the water flow wavefront; T _arrival_obs T represents the actual observation time when the water wavefront arrives at the monitoring section; _arrival_pred This represents the wavefront arrival time of the water flow predicted by the model, and its calculation results are closely related to parameters such as channel roughness.

[0226] Step S703: Use the model residuals to perform Bayesian updates or rolling statistical corrections on the scene weights in the water supply capacity scene package, reduce the weights of scenes that do not match the current observations, and simultaneously correct the prior parameters of the channel section loss, so as to obtain the updated water supply capacity scene package for decision-making in the next rolling cycle.

[0227] Specifically, Bayesian inference is used to update the scenario weights. Assume the water supply scenario package contains S preset scenarios, representing different combinations of leakage levels and roughness. For each scenario s, the likelihood probability P(Obs|s) = exp(-(Q)) is calculated between the predicted and observed values. _obs -Q _pred_s ) 2 / (2σ 2 )); above, Q _pred_s The predicted flow value calculated by the model corresponds to the value in scenario s. Different scenarios correspond to different prediction results, σ. 2 The variance of flow observations or prediction errors is a parameter that measures the degree of data dispersion. Its magnitude determines the sensitivity of the likelihood probability to changes in deviation.

[0228] Furthermore, the weights are updated using Bayes' theorem, i.e.:

[0229] p _s (new)=(P(Obs|s)×p _s (old)) / Σ(P(Obs|k)×p _k (old));

[0230] Where, p _s (new) represents the updated weights of scene s; p _s (old) represents the weights before the update; P(Obs|s) represents the likelihood probability of observing the current data under scenario s assumption; Σ is the normalized summation of all possible scenarios k.

[0231] Through this update, scenarios where the predicted results highly match the actual observations, p _s It will automatically increase; conversely, it will decrease. This is equivalent to the system learning, during operation, which scenarios are more similar to the current one.

[0232] Furthermore, the loss parameter is corrected using residuals. For example, for the leakage loss coefficient k... _soil Kalman filtering, such as UKF or EnKF, is used for online parameter estimation.

[0233] k _soil (new)=k _soil (old) + K × (Q) _obs -Q _pred (k _soil ));

[0234] Where K is the Kalman gain. Or, in other words, above, k _soil (new) corresponds to the updated value of the canal bed soil permeability (leakage loss coefficient), which is the optimized parameter after model calibration; k _soil(old) corresponds to the initial value of the canal bed soil permeability before the update or the parameter value of the previous iteration; K corresponds to the parameter update step size coefficient, also known as the learning rate, which is used to control the magnitude of each parameter adjustment to avoid the model not converging due to excessively large update magnitude or converging too slowly due to excessively small update magnitude; Q _pred (k _soil This corresponds to the predicted channel flow rate calculated based on the current soil permeability of the canal bed.

[0235] According to another aspect of this application, the parametric Kalman filter correction formula can also be:

[0236] θ _new =θ _old +K×(Obs-Pred(θ _old ));

[0237] Where, θ _new For corrected physical parameters (such as infiltration coefficient or roughness); θ _old The parameters before correction are: K is the Kalman gain matrix; Obs are the field observations; Pred(θ) _old () represents the model prediction based on the old parameters.

[0238] Based on this, the revised parameters will be immediately used for the next cycle of scenario generation, and subsequent risk budgeting and scheduling instructions will be made based on the latest system characteristics, thus achieving closed-loop adaptation.

[0239] According to another aspect of this application, an optional construction module for the architecture of an irrigation district risk avoidance and control decision system is provided. In this embodiment, complex mathematical models and optimization algorithms are encapsulated into independent computer program modules, which are connected to field equipment through an industrial control network to form an irrigation district risk avoidance and control decision system that can be practically deployed.

[0240] Furthermore, an intelligent agricultural water resource regulation and decision-making system for irrigation districts is provided, which can be used to execute the method of this application, specifically as follows:

[0241] The physical modeling and scene generation module is used to construct a two-layer dynamic graph model of the irrigation area and generate a water supply capacity scene package;

[0242] The risk budget and contract generation module is used to generate water demand scenarios, set the total risk ceiling for the entire irrigation district, decompose the total risk ceiling to obtain the risk budget distribution matrix, and solve the water supply interval contracts.

[0243] The rolling control and execution module is used to establish a set of engineering constraints, solve for the optimal engineering control instruction sequence based on the rolling time domain mechanism, and issue the execution of the optimal engineering control instruction sequence.

[0244] Alternatively, the system may include a physical modeling and scene generation module, which is used to construct a two-layer dynamic graph model of the irrigation area that maps the physical topology and hydraulic connections of the irrigation area. Based on meteorological ensemble forecast data and parameter uncertainty distribution, the system uses the two-layer dynamic graph model of the irrigation area to generate a water supply capacity scene package containing multiple spatiotemporal possibilities.

[0245] Specifically, this module is typically deployed on high-performance computing servers or cloud clusters. From a software architecture perspective, it comprises a GIS (Geographic Information System) interface unit, a hydraulic simulation engine, and an uncertainty generator. The GIS interface unit is responsible for loading static topological data of the irrigation district, such as channel lengths, cross-sectional dimensions, and field boundary coordinates. The hydraulic simulation engine has a built-in state-space equation solver that can reconstruct the adjacency matrix A(t) based on real-time gate opening signals. The uncertainty generator integrates a Monte Carlo simulation algorithm to generate hundreds or thousands of water supply scenarios based on weather forecast data. In terms of physical implementation, this module requires a high-frequency CPU and large-capacity memory to support high-dimensional matrix operations and parallel simulations.

[0246] Optionally, the system may also include a risk budget and contract generation module, which is used to generate water demand scenarios based on crop water demand characteristics, set the total risk ceiling for the entire irrigation area, decompose the total risk ceiling into a risk budget distribution matrix at the spatiotemporal unit granularity, and solve and generate water supply interval contracts based on the risk budget distribution matrix and water supply capacity scenario package. The water supply interval contracts limit the lower limit of water supply, the upper limit of water supply, and the allowable risk level for each spatiotemporal unit.

[0247] Specifically, this module is used for system decision-making and typically runs as a standalone service process. It integrates an agronomic model library (storing crop coefficients and sensitivity coefficients), a risk analysis algorithm library (including CVaR calculations and heuristic allocation algorithms), and large-scale optimization solvers (such as Gurobi, CPLEX, or the open-source SCIP). This module obtains water supply capacity scenario packages output by the physical modeling and scenario generation module through a database interface and builds a stochastic programming model in memory. Its output water supply interval contracts are typically stored in encrypted JSON or XML data packets and published to downstream modules via an internal message bus. On the hardware side, this module may need to connect to external meteorological service data sources and soil moisture monitoring networks via API interfaces.

[0248] Optionally, the system may also include a rolling control and execution module, which is used to establish a set of engineering constraints that includes physical limitations on the operation of engineering equipment, and, under the premise of satisfying the water supply section contract, solve for the optimal engineering control instruction sequence based on the rolling time domain mechanism and issue it for execution.

[0249] Specifically, this module connects the digital and physical worlds and is typically deployed in a control center or edge computing node close to the site. It employs a rolling time-domain controller (MPCController), which receives contract constraints from the risk budgeting and contract generation module and simultaneously reads the current status of field gates and pump stations in real time via a SCADA (Supervisory Control and Data Acquisition) interface, using this data as the initial conditions for the engineering constraint set. This module incorporates a two-stage hierarchical optimization strategy or a decomposition-coordinated optimization strategy, enabling it to solve for control commands within milliseconds or seconds.

[0250] The optimal engineering control command sequence obtained from the solution is converted into a standard industrial control protocol (such as Modbus, IEC60870-5-104) and directly sent to the local PLC (Programmable Logic Controller) to drive the servo motor to perform gate lifting and lowering operations.

[0251] In addition, the system also includes a sensor network distributed at key nodes in the irrigation area, including ultrasonic flow meters, radar water level gauges and soil moisture sensors, which provide the system with the necessary observation data and form the physical basis for closed-loop control.

[0252] According to another aspect of this application, a different intelligent regulation and decision-making method for agricultural water resources in irrigation districts is provided, which can be implemented by performing the following steps:

[0253] Step S1: Constructing the irrigation district hydrological model. Specifically, the underlying surface of the irrigation district differs from a natural watershed; it consists of individual fields connected by channels. Given this, its runoff generation and confluence characteristics are quite unique. This step is used to construct an irrigation district hydrological model that reflects these characteristics. The model takes meteorological data such as rainfall as input and outputs the runoff process to obtain the available water resources.

[0254] Step S2 involves water resource allocation. The conventional approach is based on optimization theory, inputting both the available water supply and the water demand (estimated using a specific method) for optimal allocation. For example, the objective might be to achieve the most uniform water shortage. This step can incorporate the spatiotemporal distribution characteristics of water demand for crops in the irrigation area, constructing a non-absolutely uniform objective function. Based on this, a model is solved, outputting an allocation scheme that is determined based on the principle of prioritizing key units and moderately regulating non-key units.

[0255] Step S3: Dispatch water conservancy projects in the irrigation area to achieve water resource allocation.

[0256] Currently, research on the optimal scheduling of irrigation district water resources mainly focuses on algorithm optimization, stochastic optimization theory, and the protection of the ecological environment. There is relatively little research on comprehensive optimization scheduling models that consider the risk avoidance of water shortage under uncertain conditions. Therefore, scheduling decisions can be made with the goal of balancing better benefits and lower risks.

[0257] The uncertainties include:

[0258] Conditions that lead to uncertainty in available water supply include, for example, uncertainty in meteorological data, uncertainty in the structure of hydrological models, and uncertainty in hydrological parameters.

[0259] Conditions that lead to uncertainty in water demand forecasting include, for example, more rainfall means less water demand for crops, and uncertainty about the location of rainfall also leads to uncertainty in water demand.

[0260] The optional embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solution of the present invention, and such equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for intelligent regulation and decision-making of agricultural water resources in irrigation districts, characterized in that, include: A two-layer dynamic graph model of the irrigation district is constructed to map the physical topology and hydraulic relationship of the irrigation district. Based on meteorological ensemble forecast data and parameter uncertainty distribution, the two-layer dynamic graph model of the irrigation district is used to generate water supply capacity scenario packages containing multiple spatiotemporal possibilities. Based on the water demand characteristics of crops, a water demand scenario is generated, and a total risk ceiling for the entire irrigation area is set. The total risk ceiling is decomposed into a risk budget distribution matrix with spatiotemporal unit granularity. Based on the risk budget distribution matrix and the water supply capacity scenario package, a water supply interval contract is generated. The water supply interval contract limits the lower and upper limits of water supply and the allowable risk level for each spatiotemporal unit. An engineering constraint set including physical constraints on the operation of engineering equipment is established. Under the premise of satisfying the water supply interval contract, the optimal engineering control command sequence is solved based on the rolling time domain mechanism and issued for execution.

2. The method according to claim 1, characterized in that, The two-layer dynamic graph model of the irrigation district is constructed as a directed graph structure G=(V, E, A(t)), where: the node set V includes at least field nodes with independent water balance attributes, channel nodes with water transmission and distribution functions, and engineering control nodes representing gates or pumping stations; the edge set E includes at least water supply edges connecting channel nodes and field nodes, overflow edges connecting adjacent field nodes, and loss edges representing water transmission losses; A(t) is a time-varying adjacency matrix used to describe the hydraulic connectivity state between nodes in the node set V.

3. The method according to claim 2, characterized in that, The values ​​of the elements in the time-varying adjacency matrix A(t) are not fixed constants, but are dynamically determined by the engineering control state variables of the engineering control nodes. When the engineering control state variables change, the corresponding elements in the time-varying adjacency matrix A(t) are updated accordingly, mapping the physical engineering control behavior to the mathematical graph topology changes.

4. The method according to claim 2, characterized in that, A water supply capacity scenario package containing multiple spatiotemporal possibilities is generated using a two-layer dynamic graph model of the irrigation district. Specifically, this includes: establishing field water balance equations for field nodes to describe the dynamic changes in field surface water depth with rainfall input, evaporation and transpiration output, and infiltration output; establishing open channel hydraulic equations for canal nodes to describe the dynamic changes in canal water level or flow with upstream inflow and headway losses; combining the field water balance equations and the open channel hydraulic equations to construct a state-space equation with engineering control state variables as control inputs and the distribution of water level and flow at all irrigation district nodes as the system state; and based on the state-space equations, the system state trajectory is deduced under given meteorological and parameter disturbances to generate the water supply capacity scenario package.

5. The method according to claim 1, characterized in that, Each scenario in the water supply capacity scenario package includes at least scenario-based water inflow capacity data, scenario-based channel loss data, and corresponding scenario weights. The scenario weights are determined in at least one of the following ways: probability weights are calculated based on the frequency of occurrence of historical meteorological and hydrological data; or equal-weighted scenarios are constructed based on preset confidence intervals and equal probability quantiles, and the scenario weights are set to equal values.

6. The method according to claim 1, characterized in that, Before generating water demand scenarios based on crop water demand characteristics and setting the total risk ceiling for the entire irrigation area, the process also includes constructing spatiotemporally differentiated weights and spatiotemporally differentiated water shortage loss functions. Specifically, based on the crop coefficient at the current growth stage, the water sensitivity coefficient of yield to water shortage, and the preset spatial priority coefficient, the spatiotemporally differentiated weights of each spatiotemporal unit are calculated to characterize the importance of the unit in terms of yield formation and social impact. Based on the spatiotemporally differentiated weights, a spatiotemporally differentiated water shortage loss function is constructed. The function defines a nonlinear mapping relationship between the degree of water shortage and economic loss, so that spatiotemporal units in the critical growth period will generate higher marginal loss values ​​under the same water shortage rate.

7. The method according to claim 6, characterized in that, The total risk ceiling is decomposed into a risk budget distribution matrix at the spatiotemporal unit granularity. Specifically, this includes: initially allocating the total risk ceiling of the entire irrigation area using spatiotemporally differentiated weights to obtain the initial risk budget; calculating the tail risk contribution of each spatiotemporal unit under the current budget based on the water supply capacity scenario package and the water demand scenario; and redistributing the initial risk budget according to the tail risk contribution, so that key units with high marginal loss sensitivity can obtain the risk budget amount to be satisfied first, thereby generating the risk budget distribution matrix.

8. The method according to claim 7, characterized in that, The process of redistributing the initial risk budget based on the tail risk contribution can be executed in any of the following ways: using a heuristic iterative strategy to identify non-critical units with tail risk contributions below a preset threshold, and migrating their remaining risk budgets to critical units with high risk contributions until the risk constraint satisfaction rate of all units converges; or transforming the risk budget constraint into a Lagrange dual problem, treating the risk budget distribution matrix as the original variable to be solved, and automatically adjusting the budget allocation of each spatiotemporal unit by updating the Lagrange multipliers.

9. The method according to claim 7, characterized in that, Based on the risk budget distribution matrix and water supply capacity scenario package, a water supply interval contract is generated. Specifically, for each spatiotemporal unit, the risk budget amount corresponding to the risk budget distribution matrix is ​​used as a hard constraint. Combined with the spatiotemporally differentiated water shortage loss function, optimization is performed under all scenario sets of the water supply capacity scenario package. The lower and upper limits of water supply that satisfy the hard constraints are calculated, so that the lower limit of water supply can cover the basic water demand requirements under the pre-set confidence level, and the adjustment range formed by the upper and lower limits of water supply is within the executable range of the engineering facilities. The lower and upper limits of water supply and the corresponding risk limits are encapsulated into a water supply interval contract, which serves as the sole basis for performance of subsequent engineering regulation.

10. A smart decision-making system for agricultural water resource regulation in irrigation districts, characterized in that, include: The physical modeling and scene generation module is used to construct a two-layer dynamic graph model of the irrigation area and generate a water supply capacity scene package; The risk budget and contract generation module is used to generate water demand scenarios, set the total risk ceiling for the entire irrigation district, and decompose the total risk ceiling to obtain the risk budget distribution matrix; Solve the water supply section contract; The rolling control and execution module is used to establish a set of engineering constraints, solve for the optimal engineering control instruction sequence based on the rolling time domain mechanism, and issue the execution of the optimal engineering control instruction sequence.

Images