A rice yield and quality synergistic optimization intelligent cultivation decision method and system

Abstract

The application discloses a kind of rice yield quality synergic optimization intelligent cultivation decision method and system, specifically related to agricultural information technology field, including: zoning modeling and canal system digital twin are carried out to irrigation area, unified time base and data lake are constructed, and canal segment dynamics model and field mapper are established;Fusion multi-source observation data is carried out state estimation and field water level prediction, exports state posteriori and uncertainty, and generates water level forward trajectory;Based on phenology calendar and quality risk identification result, dynamically set water level target band, adopt two-layer rolling prediction control, optimize pump station output and water distribution time sequence in upper layer, refine water outlet opening and closing and drainage arrangement in lower layer, and introduce opportunity constraint or safety margin in control to handle uncertainty.The application can realize the accurate, robust control of field water level under engineering constraint, effectively reduce AWD out-of-bound rate and equipment start-stop frequency, and improve the level of rice yield and quality synergic optimization.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, and more specifically, to a method and system for intelligent cultivation decision-making that coordinates the optimization of rice yield and quality. Background Technology

[0002] Rice irrigation at the field scale generally relies on the coordinated operation of a tiered canal network and pump gate system. Its water supply process is simultaneously affected by multiple factors, including the time variability of incoming water, leakage and evaporation losses along the route, the time lag in hydraulic propagation in the canals, and differences in soil and crop conditions. In actual production, water allocation strategies are often based on experience or static schedules, making it difficult to respond promptly to disturbances such as rainfall, peak water usage, and changes in upstream scheduling. This results in the field surface water level being difficult to maintain stably within the target water level range during key phenological stages, such as heading, flowering, and early grain filling. AWD (intermittent wet-dry alternation) implementation exhibits problems such as frequent boundary violations, frequent start-ups and shutdowns, and high energy consumption.

[0003] Meanwhile, irrigation districts typically employ both SCADA continuous measurements and intermittent field IoT measurements. These two types of data lack uniformity in terms of time and spatial benchmarks and measurement standards, and are generally characterized by late arrivals, missing measurements, and abnormal readings. This makes it difficult to achieve closed-loop state estimation and water use assessment, and hinders the quantification, verification, and traceability of water allocation directives. Existing channel models are mostly offline or black-box representations, not tightly coupled with the engineering object system of zones, canals, nodes, and plots, and their mapping to the fields. This makes it difficult to conduct risk-oriented prediction and optimization while meeting engineering constraints such as pump station output, gate travel, water distribution capacity, and ecological flow, and also makes it difficult to explain the sources and scope of disturbances.

[0004] At the scheduling level, existing solutions often treat upper-level water volume budgeting and lower-level execution timing separately, lacking a rolling update mechanism constrained by uncertainty. Equipment feedback and new measurements fail to form a closed-loop correction. In the presence of propagation time lags and sudden disturbances, control strategies are prone to lag and over-correction, resulting in both waste in discharge and quality risks. In summary, current technologies lack a systematic method and device that integrates multi-source observations under unified spatiotemporal and measurement calibers, interpretably characterizes the hydraulic processes of canals and fields, and enables two-level rolling predictive control within engineering constraints. This would ensure economic efficiency while robustly maintaining the field water level within the target zone, supporting the large-scale, repeatable, and verifiable implementation of AWD. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a smart cultivation decision-making method and system for co-optimizing rice yield and quality to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A smart cultivation decision-making method for synergistic optimization of rice yield and quality includes the following steps:

[0008] The irrigation district is divided into zones and modeled with digital twins of the canal system. A unified time base and data lake are constructed, and a dynamic model of the canal section and a field mapper from the water distribution point to the plot are established.

[0009] Integrating multi-source observations such as SCADA and field IoT, the system performs state estimation and field water level prediction, outputs the posterior state and uncertainty matrix Σ of the covered area / ditch / field, and calculates the prospective trajectory of field water level based on channel propagation time delay and head loss.

[0010] Based on the crop phenological calendar and combined with the plot-level quality risk score, a water level target zone that changes over time is set for each plot. A two-layer rolling predictive control is adopted. In the upper layer, the water level target zone and the uncertainty matrix Σ are used as constraints to optimize the pump station output and water distribution sequence within the constraints of the zonal water volume budget and pump gate engineering. In the lower layer, based on the upper layer water distribution budget and combined with the field mapper and field surface water balance, the opening and closing of the water distribution outlet and drainage instructions are generated. The plot-level quality risk score is obtained by weighted summation of the dry deviation index and the hot night index.

[0011] In a preferred embodiment, the water level target zone is composed of a dynamically changing lower limit function and an upper limit function; the setting of the upper limit function takes into account the field ridge elevation, safe drainage requirements, and equipment access requirements.

[0012] In a preferred embodiment, the lower limit function is set based on a dynamically selected water level control strategy; the water level control strategy includes at least a normal AWD strategy, a lower limit raising strategy, and a shallow wet irrigation strategy.

[0013] In a preferred embodiment, during the critical quality window of heading, flowering, and early to mid-grain filling, a switch is triggered from the normal AWD strategy to either the lower limit enhancement strategy or the shallow wet irrigation strategy, based on the plot-level quality risk score.

[0014] In a preferred embodiment, the hot night index is represented as the percentage of days within the forecast window where the nighttime minimum temperature exceeds the zoning threshold.

[0015] In a preferred embodiment, the dry deviation index is represented as a normalized index of the time multiplied by the predicted field surface water level being below the lower limit of the target zone within the future rolling time domain.

[0016] In a preferred embodiment, in the upper-level optimization, the uncertainty matrix Σ is written into the constraints of the optimization problem in the form of chance constraints or safety margins to ensure that the probability of the field water level meeting the target constraint is not lower than a preset confidence level.

[0017] In a preferred embodiment, the state estimation employs an unscented Kalman filter or particle filter algorithm to fuse continuous measurement data provided by SCADA with intermittent measurement data provided by the field IoT.

[0018] In a preferred embodiment, continuous measurement data includes one or more of pump station output, gate opening, water level at key cross-sections, and flow rate; intermittent measurement data includes one or more of field surface water level and shallow aquifer.

[0019] In a preferred embodiment, the following modules are included:

[0020] The access twin module is used to perform zonal modeling and digital twinning of irrigation districts and canal systems, build a unified time base and data lake, and establish a canal section dynamic model and a field mapper from the water distribution point to the plot.

[0021] The state prediction module is used to integrate multi-source observations such as SCADA and field IoT to perform state estimation and field water level prediction, output the state posterior and uncertainty matrix Σ of the covered area / ditch / field, and calculate the prospective trajectory of field water level based on channel propagation time delay and head loss.

[0022] The rolling control module is used to set time-varying water level target zones for each plot based on the crop phenological calendar and plot-level quality risk scores. It adopts a two-layer rolling predictive control. In the upper layer, the target water level zone and the uncertainty matrix Σ are used as constraints to optimize the pump station output and water distribution sequence within the constraints of the zoning water budget and pump gate engineering. In the lower layer, based on the upper layer water distribution budget and combined with the field mapper and field surface water balance, it generates water outlet opening and closing and drainage commands. The plot-level quality risk score is obtained by weighted summation of the dry deviation index and the hot night index.

[0023] The technical effects and advantages of this invention are as follows:

[0024] This invention integrates continuous SCADA data with intermittent data from the field IoT system through a unified time base and data lake, utilizing a digital twin of the canal system and a field-to-plot mapper. This forms an observable, interpretable, and traceable dynamic model foundation for the coverage area / canal / field. Based on this, unscented Kalman filtering / particle filtering is used to fuse multi-source observations, outputting the state posterior and uncertainty matrix Σ online, and generating a forward trajectory of field water level considering propagation delay and flow loss. This reliably translates the upstream water diversion intention into the inflow boundary on the plot side, forming a closed loop of estimation, prediction, and execution. This significantly improves the computability and reproducibility of the water use process, solving the pain points of traditional solutions such as inconsistent data caliber, decoupling between the model and the engineering object, and lack of uncertainty management.

[0025] At the control level, this invention uses a crop phenological calendar-driven water level target zone as its core, and introduces a quality risk score composed of the dry deviation index and the hot night index into a two-layer rolling predictive control. The upper layer jointly optimizes the pump station output and water distribution sequence within the constraints of the zoned water budget and pump gate engineering, and writes Σ with opportunity constraints or equivalent safety margins to suppress the risk of exceeding the limit. The lower layer generates water outlet opening and closing and drainage instructions based on field mapping and field surface water balance, and reduces unnecessary adjustments through dead zone tracking and back-cut hysteresis. This mechanism can robustly maintain the water level trajectory within the target zone even under uncertain disturbances and propagation time delays, significantly reducing the AWD (away from water) exceedance rate and equipment start-up and shutdown frequency, reducing waste and energy consumption in external discharge, and stabilizing pollen viability and grain filling environment during the critical windows of heading, flowering, and grain filling, achieving synergistic optimization of water conservation, yield, and rice quality, and providing a verifiable and auditable chain of evidence and operational records for large-scale promotion. Attached Figure Description

[0026] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0027] Figure 1 This is a flowchart illustrating an intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to the present invention.

[0028] Figure 2 This is a schematic diagram of the structure of an intelligent cultivation decision-making system for the collaborative optimization of rice yield and quality according to the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0030] Example 1: The present invention provides an intelligent cultivation decision-making method for synergistic optimization of rice yield and quality, such as... Figure 1 As shown, it includes the following steps:

[0031] Step 1: Zonal Modeling and Canal System Digital Twin

[0032] This step begins with object decomposition and zonal modeling of the target irrigation area to ensure that subsequent estimation and control have a clear feasible domain and computable structure on the engineering side. Specifically, based on water source type, pump station division, canal system topology, administrative / operational boundaries, and soil and planting systems, the operational domain is divided into several regions, denoted as z. Within each region, a graph model of area, canal segment, node, and plot is constructed. The node set represents hydraulic control or measurement points such as pump stations, gates, branch inlets, return inlets, and field inlets and outlets, while the edge set represents open or culvert segments connecting the above nodes. For each canal segment, engineering attributes such as length, cross-sectional shape and material, slope, and historical maintenance status are recorded. For each node, geographical coordinates and elevation benchmarks, equipment specifications, and control range are recorded, thus forming a ternary graph structured base of area, canal, and field that can be parsed and scheduled by a computer.

[0033] Based on the ternary map of districts, canals, and fields, resource and capacity boundary parameters are set at the regional scale to limit the engineering feasible region for subsequent optimization and implementation. The upper limit of seasonal or annual water supply is denoted as... This represents the amount of water that region z can access within a specified period through contracts, water rights, or scheduling plans; the transmission and distribution efficiency is denoted as... , a dimensionless coefficient between 0 and 1, is used to characterize the effective proportion of water reaching the field after considering the comprehensive headwind losses from the water inlet to the field inlet. The comprehensive headwind losses include leakage, evaporation, bypass water intake, and backwater backflow. The upper limit of the pumping station capacity is denoted as . , representing the maximum instantaneous allowable output of the main pump or connected pump unit within region z under safe operating conditions; the energy consumption / water price coefficient is denoted as This parameter is used to convert unit water diversion into economic cost or energy consumption weight. The parameters are derived from historical metering and dispatching records, equipment nameplates, energy consumption records, and on-site verification, and can be updated quarterly or monthly.

[0034] To achieve observability, traceability, and controllability of the canal and field systems, this step establishes a unified time base and data access mechanism to form an integrated data lake for the district, canals, and fields.

[0035] Data sources include: pump station output, upstream and downstream water levels, flow rates at key sections, gate openings, and equipment start-up / shutdown feedback provided by the Supervisory Control and Data Center (SCADA) system; and water level data from field IoT monitoring systems. Shallow water aquifer, where i is the plot index and t is the system time; hourly rainfall and reference evapotranspiration provided by meteorological and crop water requirement models. With crop coefficient The operational layer provides water outlet opening and closing logs and anomaly alarms. All data undergoes timestamp standardization, coordinate and elevation benchmark unification, unit conversion, and anomaly and missing measurement labeling before entering the lake. A one-to-one mapping relationship is established with unique identifiers for regions, nodes / channel sections, and plots to ensure that any measurement or command can be located and tracked within the same coordinate system.

[0036] Building upon the aforementioned object model and data foundation, this step constructs a digital twin of the canal system to characterize the dynamic relationships of upstream inflow, propagation and loss along the canal, and downstream response within a feasible level of engineering complexity. Preferably, a gray-boxed one-dimensional equivalent model is used to establish a low-order description of inflow, equivalent storage, and outflow for each canal segment, embedding its static / quasi-static characteristics at the gate and pumping station nodes, respectively. For canal segment e, its dynamics can be expressed using the following recursive form: .

[0037] in, This refers to the equivalent storage capacity or equivalent water level status of the canal section. For observable measurements at downstream sections, such as flow rate or water level; The inflow control quantity is from the upstream pumping station output or the gate flow rate; , , Parameters reflecting propagation and attenuation characteristics, This represents the equivalent term for external disturbances such as unmetered water intake, temporary maintenance, and backwater backflow. For gate nodes, the relationship between pressure difference, opening degree, and flow rate is used, employing the following formula. Characterizing local hydraulic properties, among which The difference in water level between the upstream and downstream sides of the sluice gate. The opening degree refers to the mechanical opening degree of a gate or valve, which is used to characterize the effective opening amount of the cross section that can pass through water. For pump station nodes, the flow rate, head, power characteristic curves and minimum start-stop interval constraints are used to characterize their feasible operating conditions.

[0038] To bridge the gap between canal and field dimensions and provide directly usable input for subsequent modules, this step establishes a field-to-water mapping mechanism from the water distribution point to the plot. This mapping integrates the water distribution point's opening or closing status, the distribution structure and local losses of the canals / irrigation channels, the terrain slope, and the geometric parameters of the plot's inlet, uniformly converting the time-series flow rate of the water distribution point into the unit time water replenishment volume to the field. This mapping is consistent with the node and plot mapping in the data lake. Through this mapping, upstream water diversion and allocation intentions can be translated into plot-side inflow boundaries under the same time base, forming a unified interface for subsequent multi-source state estimation, field water level prediction, and rolling predictive control for AWD.

[0039] The outputs of this step include, but are not limited to: a consistent data lake and real-time data stream, a channel digital twin parameter library with online identification capabilities, and data that can be directly generated. The field mapping device's technical effectiveness is reflected in two aspects: First, by using a ternary map of the area, canal, and field, it unifies engineering objects and field objects within a computable structure, clearly defining regional water volume and capacity boundaries; second, through gray-box twinning and online identification, it enables continuous estimation and correction of key but time-varying hydraulic elements such as propagation delay and friction loss during operation, and... This dimensional output, consistent with the physical processes of the land parcel, provides a stable, interpretable, and traceable model foundation for subsequent state estimation and rolling control.

[0040] Step 2: Multi-source state estimation and field water level prediction

[0041] The goal of this step is to integrate, under a unified temporal and spatial reference, pump station output, gate opening, key section water level and flow rate provided by the Supervisory Control and Data Center (SCADA) system, water level and shallow water content provided by the field Internet of Things (IoT), and hourly rainfall and reference evapotranspiration provided by meteorological and crop water requirement models. With crop coefficient By integrating these processes and constructing a process prior consistent with the digital twin of the canal system, we can obtain augmented state estimates of the coverage area, canals, and fields, as well as corresponding uncertainty measures. The data includes forward-looking field water level prediction curves required for short-term rolling optimization. Before being entered into the database, the data undergoes timestamp standardization, coordinate and elevation benchmark unification, unit conversion, and anomaly labeling. A unique identification system for the district, canal, and field ternary maps ensures a one-to-one mapping between measurements and objects, guaranteeing the traceability of estimates and predictions. Reference evapotranspiration is the potential evapotranspiration determined solely by meteorological conditions under standard reference crop cover. The standard reference crop is typically a short grass with a height of 12 cm, surface resistance of 70 s·m⁻¹, and albedo of 0.23. The crop coefficient is the proportionality coefficient for converting reference evapotranspiration into target crop evapotranspiration; it reflects crop type, canopy density, and field moisture conditions.

[0042] Next, based on the digital twin of the canal system established in step one, an augmented state vector is defined. The state components include at least: the equivalent storage or equivalent water level of each canal section, the equivalent outflow of key sections, the operating status of pumping stations and gates, and the field surface water level of each plot. At the same time, process disturbance components are set to absorb time-varying factors that are difficult to measure directly, such as unmetered water intake, return water backing, leakage along the process, and evaporation.

[0043] State evolution is driven by low-order gray-box dynamics provided by digital twins. Measurement relationships are composed of continuous SCADA quantities and intermittent IoT quantities. Continuous SCADA quantities include flow rate, water level, and power, while intermittent IoT quantities include surface water level and water content, where water content refers to the amount of water in the soil pores. Discrete information such as equipment start-up / shutdown and gate opening is allowed to participate in the correction as observation constraints. Considering the nonlinearity of the relationship between gate pressure difference, opening, and flow rate, and the asynchronous arrival of multi-source measurements, this embodiment preferably uses unscented Kalman filtering (UKF) as the baseline method to handle weakly nonlinear, multi-rate data and measurement gaps without explicit differentiation. When the measurement distribution deviates significantly from Gaussian or the proportion of missing measurements increases significantly, particle filtering (PF) can be switched to enhance robustness.

[0044] The filter performs prior prediction and measurement correction at a fixed estimation step size, outputting the posterior state and its uncertainty. Σ characterizes the covariance of the estimation errors of each state component and serves as the basis for constructing opportunity constraints or safety margins in subsequent optimization. To ensure the reliability of the uncertainty, the system continuously performs whitening checks and coverage verification on the innovative sequence. If the proportion of the actual trajectory falling within the prediction confidence interval deviates from the preset level, the weights of process noise and measurement noise are adaptively adjusted. The innovative sequence is the difference between the measurement and the prediction.

[0045] The water budget at the plot level is estimated in tandem with the canal status, and the short-term evolution of the field surface water level is calculated using mass conservation principles. The preferred discrete representation is: ;

[0046] In the formula, Let i be the water level on the field surface at time t; To estimate the step size; Effective field area; The amount of water replenished to the field per unit time is calculated in real time from the mapping of the water distribution point to the plot in step one, based on the time-series flow of the water distribution point, the distribution structure of the irrigation canals / farm canals, and local losses. It is the rainfall equivalent; The evapotranspiration deduction item can be derived from the reference evapotranspiration and crop coefficient; It characterizes leakage and infiltration losses as water level and soil condition change. Its parameters can be set according to the soil texture and hydraulic conditions of region z or updated during operation. Overflow or discharge during periods of excess storage in the field is usually related to the elevation of the field embankment or the height of the drainage system. When When it approaches the height of the field ridge, Automatically activated to reflect the physical process of safe drainage; when When below the specified depth of the topsoil surface, The leakage parameters are dominant, characterizing infiltration behavior under alternating wet and dry conditions. The above parameters and function forms can be uniformly set at the regional or plot cluster scale based on measured and historical data to ensure that different plots can be compared and summarized under the same dimensions and calculation caliber.

[0047] To address the asynchronous and missing data from multiple sources, the system implements timestamp-driven sequential absorption of late-arriving measurements based on a unified estimation step size. Measurements exceeding their range, exhibiting jumps, or showing significant inconsistencies with neighboring measurement points are anomaly-labeled and their weights reduced. When critical sensors fail to connect, digital twin predictions replace the corresponding measurements, and the variance of the corresponding state components in Σ is amplified to accurately reflect the increased uncertainty caused by missing information. For the prediction requirements of arrival time and amount at the field, the system maps upstream water diversion and allocation intentions to land parcels via water distribution points. Combined with channel propagation time lag and along-path loss estimation, the system outputs indicators of the estimated time to reach the field and the estimated peak water replenishment at the field, which, together with the field surface water balance, generate indicators for future time windows. The prediction curve and its confidence interval are used to determine whether the upper and lower boundaries of the AWD target zone will be reached and the potential risk of exceeding the boundaries.

[0048] The outputs of this step include: a posteriori snapshots of the state of the covered area, canals, and fields, the corresponding uncertainty matrix Σ, and the forward trajectory of field surface water levels and the time stamp of the expected AWD threshold under given short-term weather and water diversion intentions. These outputs are published with the same object identifiers and time coordinates as in step one, providing credible initial values ​​and risk boundaries for subsequent rolling predictive control.

[0049] Step 3: Two-layer rolling predictive control for AWD

[0050] This step aims to ensure that the water level in each plot remains within the target zone at a given confidence level, while comprehensively balancing economics and risks within the feasible domain of available water supply and pumping station engineering. Based on the crop phenological calendar and quality risk identification results, the system generates a time-varying target water level zone for each plot and inputs it as a time function into the controller. Simultaneously, it calls the state posterior and uncertainty matrix Σ output from step two to explicitly constrain out-of-bounds risks through chance constraints or equivalent safety margins, thereby ensuring the feasibility and robustness of the prescription. The crop phenological calendar refers to a timetable that dynamically labels the start and end times and confidence levels of each phenological stage from sowing / transplanting to maturity for a specific variety and plot under a unified time base. This calendar is used to map stage-specific agronomic constraints to control strategies, such as target water level zones and fertilization / pest / disease risk windows.

[0051] The water level target zone is determined by a lower limit function. With upper limit function Composition. The lower limit is used to achieve the water-saving, root-control, and quality stability goals of alternating wet-dry (AWD) systems. The upper limit is jointly defined by field ridge elevation, safe drainage, and machinery access requirements. Preferably, the measured field ridge elevation minus a safety margin is used as the upper limit benchmark, and this upper limit is temporarily lowered during heavy rainfall or backwater warnings to ensure that drainage needs are prioritized. For ease of understanding, it can be equivalently represented as... ,in Let be the equivalent elevation of the field ridge for plot i. For safety margin, This is the upper limit of the operation required for machinery passage. The target zone, composed of the upper and lower limits, is aligned with the phenological calendar in terms of time: the system prioritizes water conservation during the vegetative growth stages such as tillering, jointing, and heading, and prioritizes stable quality during the critical windows of heading and grain filling. It automatically switches the lower limit strategy and adjusts the weight parameters in the controller accordingly, so that the water level deviation penalty in the critical window is significantly higher than in the normal stage.

[0052] To balance water conservation and consistent quality, this embodiment will The generation is divided into three strategies. The controller selects one of them according to the aforementioned criteria in each control cycle and maintains it until the next control cycle.

[0053] Standard AWD strategy: Water conservation and root control, priority use during the vegetative growth period;

[0054] The lower limit of normality is based on the partition parameters. Based on, take The unit is centimeters, which represents the depth of the negative water level from the topsoil. The strategy is adapted to local conditions based on soil texture and water supply. It is recommended to set the level 10 to 12 for areas with heavy clay soil and sufficient water supply, 12 to 15 for conventionally irrigated areas, and 15 to 18 for water-deficient or highly permeable areas. This strategy is effective by default during the vegetative growth stages such as tillering, jointing, and heading, using alternating wet and dry conditions to maintain root oxygen supply and conserve water.

[0055] Raising the lower limit strategy: Quality critical window, activated when there is medium risk or mild heat damage;

[0056] When it enters the heading stage to the early flowering or grain-filling stage, and Reaching a medium-risk level, for example If the value exceeds the threshold TH (TH=0.3), the system will move the lower limit up or down by a certain number of centimeters without changing the upper limit. ,in For risk-driven safety increments, such as two to three centimeters. This upward shift is used to reduce the negative impact of reduced diurnal temperature variation and hot nights on firmness and taste, while avoiding excessive water cut-off that could lead to an imbalance in grouting rate.

[0057] Shallow wetting strategy: Quality critical window, to be activated under high-risk or extreme conditions;

[0058] When in the same key window and Reaching a high-risk level, for example When the minimum nighttime temperature exceeds the threshold TM (TM=0.7), or when a warning is issued indicating that the minimum nighttime temperature is consistently higher than the zone threshold and hot, dry winds occur simultaneously, the system will raise the lower limit to near or reach zero water level on the field surface within a specified time period, while maintaining a shallow water film. in A depth of 0 to 1 cm is recommended. This strategy is equivalent to temporarily switching to shallow wet irrigation within this window, thereby stabilizing pollen viability and the sap-filling environment. The shallow wet irrigation strategy includes a minimum holding time and a shear lag to prevent the water level from oscillating around zero.

[0059] The selection criteria for the three strategies are based on the land parcel-level quality risk score. With phenological windows: It only participates in triggering during critical windows that are highly related to quality, and the critical windows are heading and flowering and the early and middle stages of grain filling; the vegetative growth period is the default normal AWD.

[0060] The risk score is defined as the probability or equivalent risk score of an adverse rice quality event occurring in plot i at time t facing a future window, ranging from zero to one. An adverse rice quality event refers to at least one indicator falling below the target threshold for a given zoning area, such as a decrease in eating quality, deviation of amylose or gel consistency from the target, a decrease in head rice percentage, an increase in chalkiness, an increase in broken rice percentage, or uneven grain shape. The risk assessment is only effective during key growth windows highly correlated with quality, such as heading, flowering, and the early to mid-grain-filling stages. Other time periods serve as a baseline reference and do not trigger the switching of control strategies.

[0061] The calculation formula is as follows: ; and These are weighting coefficients; for example... =0.4, =0.6;

[0062] in The Hot Night Index represents the percentage of days within a short-term window where the minimum nighttime temperature exceeds the zone threshold, ranging from 0 to 1. K represents the forecast window length in days. The default is 5 days, but it can be 3 to 7 days. Forecast of the daily minimum temperature for plot i; The nighttime temperature threshold is for a given zone, expressed in degrees Celsius. Commonly used values ​​are 26℃ for japonica rice and 27℃ for indica rice, but these can be slightly adjusted according to variety and season. A higher value indicates a higher risk of hot nights during heading, flowering, and grain-filling stages, which can easily lead to decreased grain filling rate, reduced palatability, and increased chalkiness. K represents the prediction window length, for example, 5 days. A discrete-time index with a step size of days; This is an indicator function; it takes the value 1 if the condition is true, and 0 otherwise.

[0063] Dry deviation index , represents a normalized index representing the time multiplier of the predicted water level being below the lower limit of the target band within the short-term rolling time domain, ranging from 0 to 1.

[0064] The calculation formula is as follows: H represents the number of time-domain steps (hourly). For example, if the next 24 hours are in 1-hour increments, then H=24. Conservative water level estimates used for risk calculation , The average predicted water level of the plot is output from step two; Water level fluctuations are normalized to a scale, for example, ten centimeters; For plot i at time... The lower limit of the water level target zone; These are discrete moments within the rolling time domain; As a positive part operator, it only counts the portion below the lower limit; if it is not below the lower limit, it is recorded as 0.

[0065] To maintain the feasibility of the target constrained under the presence of estimation errors and external disturbances, the system writes the uncertainty matrix Σ output from step two into the controller as either a chance constraint or a safety margin. If probabilistic constraints are used, then for each plot and each discrete time point in the rolling time domain... Apply and ,in The confidence level set for operators, for example =0.85. If a deterministic equivalent form is adopted, the constraints are conservatively processed using the predicted mean and standard deviation given in step two. The safety margin derived from Σ is subtracted from the lower constraint, and the corresponding margin is added to the upper constraint to reflect the contribution of estimation uncertainty to the risk of exceeding the limit. The confidence level can take different values ​​under the three operating modes: normal, cautious, and safe. A lower value is used in the normal scenario to improve economy, and a higher value is used when the risk increases to enhance robustness.

[0066] Once the target zone is determined, it is used as a time function in a two-layer rolling predictive control system. The upper layer, within the constraints of the zoned water volume budget and pump station engineering, optimizes pump station output, main and branch canal allocation, and water distribution sequence by combining channel propagation time delay and friction loss estimation. This ensures that the predicted replenishment volume per unit time to the fields can support the water level trajectory of each plot to approach or remain within the target zone within the rolling time domain. The lower layer, within the water distribution budget and timing window given by the upper layer, generates the opening and closing sequence and duration of the water distribution points based on the mapping from the water distribution points to the plots and the conservation of field surface mass. Short-term drainage slots are inserted when necessary to avoid exceeding the upper limit and wasteful discharge. To improve control stability, the system employs a tracking strategy with a dead zone at the target zone boundary. Water replenishment or drainage is only triggered when the water level deviates beyond the set dead zone, thereby suppressing high-frequency starts and stops and unnecessary small adjustments.

[0067] In the above process, the phenological calendar and quality risk identification results are used not only to determine the temporal structure of the target zone, but also to determine the weight configuration and priority order within the controller. Plots located in the critical window and with higher risk levels are assigned a higher water level deviation weight in the objective function and are given priority in obtaining quotas when the available water volume in a zone is limited. For long-distance end plots, the system appropriately advances the water replenishment phase while keeping the target zone unchanged to offset the delay in reaching the field caused by channel propagation time lag. For plots with records of high seepage or damaged field ridges, the system appropriately increases the intensity of a single water replenishment and shortens the water replenishment interval under the same target zone to reduce the amplitude of water fluctuations.

[0068] The controller employs a two-layer rolling prediction structure and operates recursively according to a fixed control cycle. The upper layer focuses on the water distribution network of zone z, jointly optimizing the instantaneous output of pumping stations and the allocation of main and branch canals within a rolling time domain T (preferably 24–72 hours). This ensures that the water demand of downstream areas is met under the combined constraints of available water, equipment capacity, and canal propagation time delays. The main decision variables of the upper layer include pumping station output. Allocation of canal sections , where e is the canal segment index; constraints include zonal water budget constraints and pump station / gate engineering constraints. The zonal water budget constraint limits the total water replenishment to the field within the prediction time domain to no more than the effective portion of the available water, and can be expressed as: ,

[0069] In the formula Let i be the amount of water supplied to the field per unit time at time t. This is the regional distribution efficiency coefficient. This represents the upper limit of the available water volume for the zone within the stated time window. Pump station capacity and start-up / shutdown engineering constraints are used to ensure safe equipment operation, limiting the amount of water available at any given time. In addition, minimum start-stop intervals and power ramp-up rates are imposed to avoid frequent start-stops and mechanical shocks. Gate and channel capacity constraints are reflected in the rated relationship between target opening, pressure difference, and flow rate, as well as ensuring that the channel cross-sectional flow rate does not exceed the safety upper limit. To explicitly address the uncertainties of estimation and external disturbances, the upper layer incorporates the uncertainty Σ provided in step two through opportunity constraints or equivalent safety margins, ensuring that the probability of the site water level not falling below the lower limit reaches a pre-set confidence level; this confidence level is set by the operator based on quality and risk preferences. The upper-level objective function consists of three weighted components: first, a measure of the cumulative deviation of the site water level from the target zone, with higher weighting for critical options; second, a measure of energy consumption and water intake costs, which are... The calculation is based on the integral of the pump station output curve; the third metric measures the start-stop switching cost to suppress high-frequency start-stop cycles and large output fluctuations. The weights can be preset according to seasonal strategies or business objectives.

[0070] The lower layer focuses on the execution of water distribution from the water inlet to the field plot. Within the water allocation budget and time window given by the upper layer, it considers the channel propagation time lag and the water balance characteristics of the plot to solve for the order and duration of water inlet opening and closing. Short-term drainage slots are arranged when necessary to avoid over-storage and discharge from the field. The lower layer uses the water inlet-to-plot-to-field mapping established in step one to convert the upper layer allocation into the expected water replenishment trajectory for each plot. This trajectory, along with rainfall, evapotranspiration, and current water level, is fed into the field water balance model to obtain the predicted water level. Trajectory; when the trajectory deviates from the target zone, the diversion queue and duration are reordered according to priority until the target zone is met or infeasibility evidence is provided and feedback is given to the upper level for synchronous adjustment; priorities include critical period, plot level, and operation window. Preferably, the upper-level problem is solved using mixed integer quadratic programming or second-order cone programming, and the lower-level problem is solved using linear programming or heuristic fast algorithms. The feasible solution of the previous control cycle is used as the initial value for hot start, and parallel calculation is performed according to branch canals or distribution canals to ensure stable satisfaction of the 15-30 minute online refresh cycle.

[0071] This step proceeds recursively in each control cycle in the order of estimation, prediction, optimization, execution, and re-estimation. At the beginning of the cycle, the posterior state and uncertainty output from step two are used as initial values. Short-term weather and water diversion intentions are combined to advance the prediction and complete two-level optimization. Only the pump station output, target opening degree of key gates, and opening and closing instructions of the water distribution outlet for the current control interval are issued and sent to the field equipment through the upper-level monitoring system. In the next control cycle, the state estimation from step two is incorporated with equipment feedback and new measurements. After updating the initial values, the rolling solution continues.

[0072] Through the above implementation methods, this step, under the uncertainties of propagation time lag, flow losses, and external disturbances, can generate an economically reasonable and risk-controllable water allocation curve at the regional level, and form an executable water delivery sequence at the plot level, enabling... It remains within the target zone for most of the time. Compared to schemes that rely on fixed pump start / stop meters, this step significantly reduces the AWD out-of-bounds rate, the offsetting effect of upstream over-storage and downstream water replenishment lag, and maintains interpretable and measurable optimization effects in terms of energy consumption and equipment start / stop frequency.

[0073] Example 2: The design of the intelligent cultivation decision-making system for synergistic optimization of rice yield and quality according to the present invention is based on the method in Example 1, specifically as follows: Figure 2 The following modules are shown:

[0074] The module integrates a digital twin, which, based on a ternary graph object model of regions, canal sections, nodes, and plots, completes the structured representation of engineering objects and the unified registration of resource capacity boundaries. It also establishes a unified time base, measurement standards, and coordinate / elevation benchmarks, and integrates multi-source data from higher-level monitoring and field IoT to form an integrated data lake. On this basis, a gray-box digital twin of the canal system is constructed to characterize key dynamic relationships such as upstream inflow, propagation and loss along the canal, and downstream response. Simultaneously, it maintains the field-to-plot mapping from the water distribution point, providing an interface and initial values ​​consistent with the physical process for subsequent estimation and control. The module's outputs include: standardized real-time data streams and a historical data lake, an online identifiable digital twin parameter library, and directly callable field-to-plot mappers and object identification systems, ensuring that subsequent modules are traceable, computable, and reproducible within the same spatiotemporal coordinates.

[0075] The state prediction module uses the data lake and digital twin of the connected twin module as priors, integrates asynchronous multi-source observations such as continuous SCADA and intermittent IoT measurements, and advances prior prediction and measurement correction according to a unified estimation step size. It handles operational scenarios such as late measurements, missing measurements, and abnormal weight adjustments, and outputs the posterior state and uncertainty matrix Σ for the coverage area, canal, and field. Simultaneously, it combines the mapping from the water outlet to the plot to the field with the canal propagation time lag and along-path loss estimation to generate a field arrival time / intensity index and a forward trajectory of the field surface water level within a future time window, along with its confidence interval. This is used to determine whether the target AWD zone has been reached or exceeded, and to provide risk boundaries to the control module. The outputs of this module are: a state posterior snapshot, the uncertainty matrix Σ, the predicted water level curve, and the time stamp for reaching the AWD threshold, all published using the same object and time base as module one.

[0076] The rolling control module aims to maintain the land parcel water level within the target zone at a predetermined confidence level. It generates a time-varying target water level zone by combining phenological calendar data and quality risk identification results. The Σ output from module two is then written into opportunity constraints or equivalent safety margins to ensure that boundary exceedance risks are controlled. At the upper level, considering zonal water volume budgets and pumping station engineering constraints, it jointly optimizes pump station output, main and branch canal allocation, and water distribution sequence. At the lower level, based on field mapping and field surface water balance, it refines the upper-level water distribution budget into the opening and closing sequence and duration of water outlets, inserting drainage slots when necessary to prevent upper limit exceedances and wasteful discharge. This module operates on a fixed cycle of estimation, prediction, optimization, execution, and re-estimation, issuing only instructions for the current control interval and updating with equipment acknowledgments and new measurements in a closed loop. This maintains economy and robustness even under propagation delays and disturbances, significantly reducing AWD boundary exceedance rates and unnecessary start-ups and shutdowns.

[0077] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0078] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0079] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0080] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0081] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A smart cultivation decision-making method for synergistic optimization of rice yield and quality, characterized in that, Includes the following steps: The irrigation area is divided into zones and modeled with digital twins of the canal system. A unified time base and data lake are constructed, and a dynamic model of the canal section and a field mapper from the water distribution point to the plot are established. The system integrates SCADA and field IoT multi-source observations to perform state estimation and field water level prediction, outputs the state posterior and uncertainty matrix Σ of the covered area / ditch / field, and calculates the prospective trajectory of field water level based on channel propagation time delay and head loss. Based on the crop phenological calendar and combined with the plot-level quality risk score, a water level target zone that changes over time is set for each plot. A two-layer rolling predictive control is adopted. In the upper layer, the water level target zone and the uncertainty matrix Σ are used as constraints to optimize the pump station output and water distribution sequence within the constraints of the zonal water volume budget and pump gate engineering. In the lower layer, based on the upper layer water distribution budget and combined with the field mapper and field surface water balance, the opening and closing of the water distribution outlet and drainage instructions are generated. The plot-level quality risk score is obtained by weighted summation of the dry deviation index and the hot night index.

2. The intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to claim 1, characterized in that: The water level target zone consists of a dynamically changing lower limit function and an upper limit function; the setting of the upper limit function takes into account the field ridge elevation, the need for safe drainage, and the need for machinery passage.

3. The intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to claim 2, characterized in that: The lower limit function is set based on the dynamically selected water level control strategy; the water level control strategy includes at least the normal AWD strategy, the lower limit raising strategy, and the shallow wet irrigation strategy.

4. The intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to claim 3, characterized in that: During the critical quality window of heading, flowering, and early to mid-grain filling, the strategy is to switch from the normal AWD strategy to the strategy of raising the lower limit or the shallow wet irrigation strategy, depending on the plot-level quality risk score.

5. The intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to claim 1, characterized in that: The hot night index is represented by the percentage of days within the forecast window where the nighttime minimum temperature exceeds the zone threshold.

6. The intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to claim 1, characterized in that: The dry deviation index is a normalized index representing the magnitude of the predicted field surface water level falling below the lower limit of the target zone over the future rolling time domain.

7. The intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to claim 1, characterized in that: In the upper-level optimization, the uncertainty matrix Σ is written into the constraints of the optimization problem in the form of chance constraints or safety margins to ensure that the probability of the field water level meeting the target constraint is not lower than the preset confidence level.

8. The intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to claim 1, characterized in that: State estimation employs unscented Kalman filtering or particle filtering algorithms to fuse continuous measurement data provided by SCADA with intermittent measurement data provided by field IoT.

9. The intelligent cultivation decision-making method for synergistic optimization of rice yield and quality according to claim 8, characterized in that: Continuous measurement data includes one or more of the following: pump station output, gate opening, water level at key cross-sections, and flow rate; intermittent measurement data includes one or more of the following: surface water level and shallow aquifer.

10. A smart cultivation decision-making system for synergistic optimization of rice yield and quality, characterized in that, The decision system is used to implement the method according to any one of claims 1-9, and includes the following modules: The access twin module is used to perform zonal modeling and digital twinning of irrigation districts and canal systems, build a unified time base and data lake, and establish a canal section dynamic model and a field mapper from the water distribution point to the plot. The state prediction module is used to integrate SCADA and field IoT multi-source observations for state estimation and field water level prediction, outputting the state posterior and uncertainty matrix Σ of the covered area / ditch / field, and calculating the prospective trajectory of field water level based on channel propagation time delay and head loss. The rolling control module is used to set time-varying water level target zones for each plot based on the crop phenological calendar and plot-level quality risk scores. It adopts a two-layer rolling predictive control. In the upper layer, the target water level zone and the uncertainty matrix Σ are used as constraints to optimize the pump station output and water distribution sequence within the constraints of the zoning water budget and pump gate engineering. In the lower layer, based on the upper layer water distribution budget and combined with the field mapper and field surface water balance, it generates water outlet opening and closing and drainage commands. The plot-level quality risk score is obtained by weighted summation of the dry deviation index and the hot night index.

Images