A quantity and quality double control irrigation area water resource simulation and regulation method and system

Abstract

This invention discloses a method and system for simulating and regulating irrigation district water resources with dual control of quantity and quality, belonging to the field of water resource management technology. This invention acquires multi-source basic data from the irrigation district, constructs a water quantity-quality coupled simulation model with a hybrid architecture of mechanism and deep learning, establishes a multi-objective optimization regulation model based on the simulation results, dynamically updates model parameters through Bayesian online learning, and distributes the regulation scheme to the execution unit to form a closed-loop regulation. The system includes a data perception layer, a transmission and storage layer, a simulation calculation layer, a decision application layer, and an execution control layer. This invention adopts a dual-driven mechanism-data approach and multi-timescale adaptive correction, significantly improving simulation efficiency and prediction accuracy, achieving total water quantity control, water quality threshold management, and precise water allocation execution. It can effectively reduce irrigation water shortage rate, water quality exceeding standards risk, and operational energy consumption, and is suitable for intelligent water resource scheduling in large irrigation districts, possessing good engineering practicality.

Description

Technical Field

[0001] This invention relates to the field of water resource management and smart irrigation district technology, specifically to a method and system for simulating and regulating irrigation district water resources with both quantitative and qualitative control. Background Technology

[0002] Irrigation districts are crucial infrastructure for ensuring national food security and water ecological security. Traditional irrigation district water resource allocation primarily focuses on ensuring water quantity, with less consideration given to water quality constraints. This has led to situations where some irrigation districts face the dilemma of "having water but being unable to use it" or "water quality exceeding standards." With the increasing severity of agricultural non-point source pollution and the rising water quality requirements for water function zones, irrigation district management urgently needs to shift from "single water quantity control" to "dual control of water quantity and water quality."

[0003] In existing technologies, the China Institute of Water Resources and Hydropower Research and other institutions have disclosed the "Water Resources Quantity, Quality, and Efficiency Synergistic Regulation Technology" (2021 Ministry of Water Resources Advanced and Practical Technology). This technology integrates a quantity, quality, and efficiency synergistic feedback simulation and optimization regulation module, initially realizing the joint scheduling of water quantity and water quality. However, this technology still has the following shortcomings:

[0004] (1) Low computational efficiency of the model: The mechanism model based on the Saint-Venant equations and the convection-diffusion equations takes a long time to calculate, and a single simulation takes several minutes to several hours, which is difficult to meet the requirements of real-time rolling control.

[0005] (2) Lack of adaptive update capability: The model parameters are mostly calibrated offline based on historical data and cannot be dynamically corrected according to real-time monitoring data, resulting in the accumulation of prediction bias, especially during periods of drastic fluctuations in hydrology and water quality, when the model becomes seriously inaccurate.

[0006] (3) Limited coordination between quantity and quality: The optimization targets mostly adopt weighted summation or constraint methods, and a real-time feedback closed loop driven by mechanism and data has not been established. The control scheme is lagging behind the actual operating status, and there is a lack of systematic design for adaptive control of multiple time scales (minutes, hours, days, weeks).

[0007] (4) Low system integration: Existing systems mostly adopt a centralized architecture, with insufficient collaboration between edge computing and cloud computing, making it difficult to meet the needs of real-time response and large-scale training.

[0008] While patent documents such as CN117787658A and CN116805197A involve irrigation district water resource scheduling or water demand forecasting, none of them have solved the problem of efficient calculation and online adaptive correction for water quantity-water quality coupled simulation. Therefore, developing a method and system that can achieve efficient water quantity-water quality coupled simulation, possess online adaptive update capability, and form a closed-loop regulation system for both quantity and quality control is of significant practical importance. Summary of the Invention

[0009] (a) Technical problems to be solved

[0010] To address the shortcomings of existing technologies, this invention provides a method and system for simulating and regulating irrigation district water resources with both quantity and quality control. It has the advantages of being applicable to intelligent decision-making scenarios such as joint scheduling of surface water and groundwater, water quality risk management and control, and coordinated control of gate and pump groups in large irrigation districts. It solves the problems of low efficiency in water quantity-water quality coupled simulation calculation, inability to update model parameters online adaptively, and poor real-time performance of coordinated regulation of water quantity and water quality in existing irrigation district water resource scheduling technologies.

[0011] (II) Technical Solution

[0012] To achieve the above objectives, the present invention provides the following technical solution:

[0013] A method for simulating and regulating irrigation district water resources with both quantitative and qualitative control includes the following steps:

[0014] S1. Obtain multi-source basic data for the irrigation area, including hydrological, meteorological, canal system topology, water source quantity, water quality, crop water requirements and irrigation system data;

[0015] S2. Construct a coupled simulation model of water quantity and quality in the irrigation area. The model adopts a hybrid architecture of mechanistic model and deep learning proxy model:

[0016] The mechanistic model is based on the Saint-Venant equations and convection-diffusion equations to construct a water quantity-water quality coupled physical process, providing physical constraint boundaries;

[0017] Deep learning surrogate models use the historical inputs and outputs of mechanistic models as training samples and employ deep neural networks to construct surrogate models, thereby accelerating computation.

[0018] S3. Based on the calculation results of the simulation model, establish a quantity and quality dual-control optimization and regulation model with multiple objectives of minimizing irrigation water shortage, minimizing water quality exceedance risk, and minimizing energy consumption, and solve for the coordinated operation scheme of multi-water source joint allocation, gate pump scheduling and water quality purification.

[0019] S4. Collect real-time water volume, water quality, and gate pump operating data, calculate the deviation between the simulated predicted value and the measured value, and when the deviation exceeds the preset threshold, start Bayesian online learning to dynamically update the parameters of the hybrid model and synchronize the updated parameters to the optimization and control model.

[0020] S5. The control scheme generated by the optimized control model is sent to the field execution unit to drive the gate valve, pump group and water purification equipment to perform control actions, so as to realize the integrated closed-loop control of total water quantity control, water quality threshold control and precise execution of water distribution in the irrigation area.

[0021] Preferably, the water quantity-water quality coupled simulation model further includes: an inflow forecast module, a canal system hydrodynamic module, a field water balance module, and a pollutant transport and degradation module. Each module adopts a mechanism-data dual-driven architecture to realize the spatiotemporal synchronous simulation of the quantity and quality processes.

[0022] Preferably, the objective function of the quantity and quality dual-control optimization and regulation model is:

[0023]

[0024] in, This represents the amount of water shortage for irrigation (m³). Total water demand (m³). This represents the cumulative concentration of water quality exceeding the standard (mg / L·d). The water quality threshold (mg / L) Actual energy consumption (kWh) The maximum allowable energy consumption is denoted as kWh, where α, β, and γ are weighting coefficients, and α+β+γ=1.

[0025] The constraints include: total water intake, canal flow capacity, water quality control section concentration, groundwater level and extraction volume.

[0026] Preferably, the method further includes: real-time monitoring of water quantity, water quality, and soil moisture data, and rolling calibration of the simulation model and the control model. The rolling calibration works in conjunction with Bayesian online learning: Bayesian online learning is responsible for the global update of model parameters (time scale: daily-weekly), and rolling calibration is responsible for the local correction of short-term deviations (time scale: minutes-hours), thereby realizing dynamic adaptive control at multiple time scales.

[0027] Preferably, the collaborative operation scheme includes: the joint allocation ratio of surface water and groundwater, the timing of water distribution flow in main and branch canals, the gate and pump opening and closing strategy, and the intensity of operation of purification facilities.

[0028] A water resource simulation and regulation system for irrigation districts with both quantitative and qualitative control includes:

[0029] Data sensing layer: used to collect real-time data on irrigation area water quantity, water quality, weather, soil moisture, and canal system conditions;

[0030] Data transmission and storage layer: used for data cleaning, aggregation, and standardized management;

[0031] Simulation layer: Deploy the water quantity-water quality coupled simulation model and the quantity-quality dual control optimization and regulation model as described above, wherein the simulation model adopts a mechanism-data dual-driven hybrid architecture, and the optimization and regulation model integrates a Bayesian online learning module;

[0032] Decision application layer: used for generating scheduling plans, visualization, instruction issuance, and effect evaluation;

[0033] Execution control layer: Used to receive control commands and drive gate valves, pump sets, and water purification equipment to perform control actions.

[0034] Preferably, the data sensing layer includes a water level and flow meter, an online water quality monitoring instrument, a soil moisture station, a weather station, and a gate pump operating condition acquisition terminal.

[0035] Preferably, the decision application layer supports multi-scenario simulation comparison, anomaly warning, one-click issuance of scheduling instructions and traceability of operation logs; it also supports digital twin visualization to display the spatiotemporal distribution of water volume and water quality in the canal system and the control effect in real time.

[0036] Preferably, the system supports integration with the irrigation district smart management platform and water resource monitoring platform to achieve data sharing and cross-system collaborative management and control; the simulation computing layer supports edge-cloud collaborative computing, with real-time Bayesian online learning tasks executed on edge nodes and large-scale historical data training tasks executed in the cloud.

[0037] (III) Beneficial Effects

[0038] Compared with existing technologies, this invention provides a method and system for simulating and regulating irrigation district water resources with both quantitative and qualitative control, which has the following beneficial effects:

[0039] 1. Significantly improved computational efficiency: This invention constructs a hybrid architecture of a mechanistic model and a deep learning proxy model. The mechanistic model ensures the accuracy of physical constraints, while the deep learning proxy model is trained with historical inputs and outputs to obtain an efficient replacement model. While maintaining simulation accuracy (average relative error <5%), the computational speed can be increased by 1 to 2 orders of magnitude (from minutes to hours to seconds). This solves the technical problems of traditional mechanistic models having long computation time and difficulty in real-time rolling control.

[0040] 2. Strong Adaptive Update Capability: This invention introduces a Bayesian online learning mechanism. When the deviation between real-time monitoring data and simulated prediction values ​​exceeds a threshold, the hybrid model parameters are dynamically updated, and the updated parameters are synchronized to the optimized control model. Compared with traditional offline calibration methods, this invention enables the model to adapt to the time-varying characteristics of irrigation district boundary conditions (such as rainfall, water inflow, and water quality), reducing the root mean square error of prediction by 40% to 60%, significantly improving prediction accuracy and control reliability.

[0041] 3. Multi-timescale collaborative control: This invention combines rolling correction with Bayesian online learning. The former is responsible for local correction of short-term biases (minute-hour scale), while the latter is responsible for global updates of model parameters (day-week scale), forming a multi-timescale adaptive control capability, which effectively solves the defects of existing technologies where model parameters are fixed and cannot respond to environmental changes.

[0042] 4. Clear objectives for quantity and quality control: This invention establishes a quantity and quality dual control optimization and regulation model with multiple objectives, namely minimizing irrigation water shortage, minimizing the risk of water quality exceeding standards, and minimizing energy consumption. It also provides specific mathematical objective functions and constraints, realizing coordinated decision-making for joint allocation of surface water and groundwater, gate and pump opening and closing, and water purification, filling the gap in the limited degree of quantity and quality coordination in existing irrigation district scheduling.

[0043] 5. Advanced System Architecture: This invention proposes an edge-cloud collaborative computing system architecture. Bayesian online learning tasks with high real-time requirements are executed on edge nodes (response time < 1 second), while large-scale historical data training is executed in the cloud (updated every 24 hours). This approach balances real-time response and model generalization capabilities, providing an engineering-deployable solution for intelligent irrigation district management. Detailed Implementation

[0044] The technical solutions of the present invention will be clearly and completely described below with reference to 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 are within the scope of protection of the present invention.

[0045] Example 1: Method Implementation Example (Taking a large well-canal combined irrigation area in North China as an example)

[0046] This embodiment takes a large well-canal combined irrigation area in North China as an example to illustrate the specific implementation process of the quantity and quality dual control method.

[0047] Step S1: Data Acquisition

[0048] Collect basic data from the irrigation district over the past 5 years:

[0049] Hydrological data: daily rainfall, evaporation, surface runoff, and groundwater depth;

[0050] Meteorological data: temperature, wind speed, sunshine duration, relative humidity;

[0051] The canal system consists of 3 main canals (total length 86km), 12 branch canals (total length 210km), and 56 distribution canals (total length 340km), including 78 gate nodes and 12 pumping station nodes.

[0052] Water source data: Permitted surface water diversion volume (maximum 15 million m³ / year), exploitable groundwater volume (maximum 8 million m³ / year).

[0053] Water quality data: COD, ammonia nitrogen, and total phosphorus concentrations, monitored twice a week, with historical data for 3 years;

[0054] Crop water requirements: The main crops are winter wheat (planted area of ​​120,000 mu) and summer maize (planted area of ​​100,000 mu). The reference crop evapotranspiration is calculated using the FAO-56 Penman-Monteith formula, and the water requirement is calculated in combination with the crop coefficient.

[0055] Irrigation system data: irrigation quota, irrigation cycle, and number of irrigations for each irrigation group.

[0056] All data is spatiotemporally aligned and stored in a standardized database, and then normalized using the Z-score method.

[0057] Step S2: Construct a hybrid model

[0058] Mechanism model construction:

[0059] Canal hydrodynamic module: Employing the one-dimensional Saint-Venant equations

[0060]

[0061] Where A is the cross-sectional area, Q is the flow rate, h is the water depth, S_f is the friction gradient, and q is the lateral inflow.

[0062] Pollutant transport module: Employs a one-dimensional convection-diffusion equation

[0063]

[0064] Where C is the pollutant concentration, D is the diffusion coefficient, K is the degradation coefficient, and S is the source-sink term.

[0065] The sample was discretized using the Preissmann four-point implicit difference scheme with a time step of Δt = 60s and a spatial step of Δx = 100m.

[0066] Deep learning agent model construction:

[0067] The historical inputs and outputs of the mechanistic model were used as training samples. The input features included 12 features in total, such as upstream boundary flow, upstream boundary concentration, gate opening, pump head, meteorological conditions (rainfall, evaporation), and crop water requirements. The output consisted of 8 variables: flow and water quality concentration at key cross-sections (main canal outlet, branch canal inlet, and drainage outlet).

[0068] Generating the training dataset: The mechanistic model was run 500 times, with each simulation lasting 7 days (60 seconds time step), generating approximately 500 × 7 × 24 × 60 = 5.04 million data sets. Input-output pairs were extracted using the sliding window method, with a window length of 6 hours (360 time steps), yielding approximately 2 million samples.

[0069] Deep Neural Network (DNN) Structure:

[0070] Input layer: 12 nodes (after normalization)

[0071] Hidden layer 1: 128 nodes, ReLU activation function, Dropout=0.2

[0072] Hidden layer 2: 128 nodes, ReLU activation function, Dropout=0.2

[0073] Hidden layer 3: 128 nodes, ReLU activation function, Dropout=0.2

[0074] Output layer: 8 nodes, linear activation

[0075] Training parameters: Optimizer Adam, learning rate 0.001, batch size 256, training epochs 500, loss function is mean squared error (MSE). 80% of the data is used for training, 10% for validation, and 10% for testing.

[0076] Training results: The mean relative error (MAPE) for the test set was 2.8% for flow and 3.2% for concentration. The time taken for a single prediction was 0.08 seconds, which is 80 times faster than the mechanistic model (approximately 6.4 seconds per prediction).

[0077] Step S3: Optimize regulation

[0078] Establish a multi-objective optimization model:

[0079] Objective function:

[0080]

[0081] Where Q_demand = 18 million m³ (total water demand during spring irrigation), C_threshold = 20 mg / L (COD threshold), and E_max = 200,000 kWh.

[0082] Constraints:

[0083] Total water intake of the main canal ≤ 15 million m³

[0084] The flow capacity of each branch canal is ≤3.0 m³ / s

[0085] Groundwater extraction volume ≤ 8 million m³, annual groundwater level drop ≤ 0.5 m.

[0086] COD concentration at each water quality control section ≤ 20 mg / L, ammonia nitrogen ≤ 1.5 mg / L

[0087] The rate of change of gate opening is ≤0.1m / min

[0088] Solution algorithm: NSGA-II (Non-dominated sorting genetic algorithm) is used, with a population size of 200, a crossover probability of 0.9, a mutation probability of 0.1, and 500 iterations. A Pareto front solution set of 32 non-dominated solutions is obtained, from which the solution with the best overall performance is selected as the scheduling scheme.

[0089] The output schemes include:

[0090] Surface water diversion sequence: Days 1-10: 2.5 m³ / s, Days 11-20: 2.2 m³ / s, Days 21-30: 1.8 m³ / s;

[0091] Groundwater extraction rate: 35% of total water supply;

[0092] Water distribution flow of each branch canal: According to the rotation irrigation group arrangement, the 12 branch canals are divided into 4 groups of 3 canals for rotation irrigation, and the water distribution flow of each canal is 0.8-1.2 m³ / s;

[0093] Gate and pump opening and closing strategy: The intake gate of the main canal is opened at 8:00 a.m. and closed at 20:00 p.m. daily; the discharge gate remains closed.

[0094] Purification facility commissioning: Aeration is started when the COD of the influent to the constructed wetland exceeds 18 mg / L, with an operating intensity of 60% of the rated power.

[0095] Step S4: Online Update

[0096] Deploy online monitoring equipment at key points in the irrigation area:

[0097] A total of 12 radar water level and flow meters (accuracy ±2mm) are installed at the main canal inlet, branch canal outlet, and outlet.

[0098] Eight sets of multi-parameter online water quality monitoring instruments (COD, ammonia nitrogen, pH, dissolved oxygen) are installed at the water quality control section, with a sampling frequency of 30 minutes / time;

[0099] There are 30 soil moisture monitoring stations, which collect data every 2 hours.

[0100] Set a deviation threshold: trigger Bayesian online learning when the deviation between the predicted and measured COD values ​​exceeds 5 mg / L for three consecutive times.

[0101] Specific steps for online Bayesian learning:

[0102] (1) Determine the parameter to be updated: pollutant degradation coefficient K (the prior distribution is set as a Gaussian distribution N(μ0,σ0²), μ0=0.15 / d, σ0=0.03 / d).

[0103] (2) Construct the likelihood function: Based on the measured concentration data of the most recent 24 hours, assume that the observation error follows a Gaussian distribution N(0,σe²), σe=2mg / L.

[0104] (3) Calculate the posterior distribution: The conjugate prior is used, and the posterior mean μ_n=(σe²·μ0+n·σ0²·ȳ) / (σe²+n·σ0²), where n is the number of samples and ȳ is the measured mean.

[0105] (4) Update model parameters: Substitute the posterior mean K_new=0.18 / d into the mixed model.

[0106] (5) Synchronize to the optimized control model: Rerun the optimized model to generate a new scheduling scheme.

[0107] In this embodiment, on the 7th day of continuous drought, the measured COD was 23 mg / L, while the predicted value was 17 mg / L. The deviation exceeded the threshold, triggering an update. After the update, K=0.18 / d, and a new scheme was obtained through re-optimization: the surface water diversion rate was reduced from 2.5 m³ / s to 2.2 m³ / s, and the artificial wetland was activated ahead of schedule.

[0108] Step S5: Closed-loop control

[0109] The updated scheduling plan will be distributed to the field execution units via the 4G network:

[0110] Gate RTU controller: controls the opening degree of the intake gate of the main canal and the diversion gate of the branch canal;

[0111] Pump station PLC: Regulates pump unit speed and water flow rate;

[0112] Water purification equipment controller: Start the aerator and add chemicals.

[0113] The system collects the execution status every 10 minutes and compares it with the target value. If the deviation exceeds 5%, it automatically makes fine adjustments. At the same time, the execution results are fed back to the decision application layer to update the visualization interface.

[0114] Example 2: System Deployment

[0115] The system of this invention was deployed in a certain irrigation district management sub-center, with the following specific configuration:

[0116] Data perception layer:

[0117] Water level and flow meter: main canal inlet (2 sets), branch canal outlet (12 sets), outlet (2 sets), using radar wave + ultrasonic composite sensor, range 0-5m³ / s, accuracy ±2%;

[0118] Water quality online monitoring instruments: COD (ultraviolet absorption method, range 0-100mg / L, accuracy ±5%), ammonia nitrogen (ion-selective electrode method, range 0-10mg / L), pH (glass electrode method), dissolved oxygen (fluorescence method), a total of 8 sets;

[0119] Weather station: 2 automatic weather stations to monitor rainfall, temperature, humidity, wind speed, and sunshine;

[0120] Soil moisture monitoring station: FDR type, monitoring depth 10cm, 20cm, 40cm, 30 sets in total;

[0121] Gate pump operating condition acquisition terminal: gate opening sensor (accuracy ±0.5%), pump set current / voltage sensor.

[0122] Data transmission and storage layer:

[0123] Communication network: The gate pumps use LoRa (470MHz band, 2km transmission distance) to aggregate to the regional gateway, and then upload to the central server via 4G (APN leased line); water quality, meteorological and soil moisture stations use NB-IoT to upload directly.

[0124] Data aggregation platform: EMQX is used for message access. After data cleaning (removing outliers and imputing missing values), it is stored in the time series database InfluxDB with 3-replica redundancy.

[0125] Data standardization: The data is encoded according to the definition in the "Basic Data Dictionary of Irrigation Districts", and the metadata is stored in PostgreSQL.

[0126] Simulation computing layer:

[0127] Edge nodes: 2 edge servers (CPU: Intel Xeon Silver 4214, memory: 64GB, GPU: NVIDIA T4), running real-time Bayesian online learning tasks and rolling correction, with a response time of <1 second.

[0128] Cloud Node: 1 cloud server (32 vCPUs, 128GB memory, 1TB SSD), which performs large-scale historical data training (retraining the DNN agent model) at 2:00 AM every day, with an update cycle of 24 hours.

[0129] Software environment: Python 3.9, TensorFlow 2.10, PyMC5 (Bayesian inference), DEAP (genetic algorithm), Flask (API service).

[0130] Decision application layer:

[0131] Digital Twin Cockpit: Developed using Three.js, it displays canal topology, real-time flow / concentration, gate pump status, and soil moisture distribution. It supports timeline playback (past 7 days) and forecasts for the next 48 hours.

[0132] Scheduling decision module: Provides a "one-click scheduling" function. After clicking, the optimization model will be run automatically, a solution will be generated, and key change items will be highlighted. It supports comparison of multiple scenarios (such as regular scheduling, water conservation priority, and water quality priority).

[0133] Abnormal warning: When it is predicted that the water quality of any section will exceed the limit or the water level will exceed the warning level within the next 2 hours, the system will notify the administrator through pop-up window and SMS, and automatically generate pre-schedule instructions (advanced control).

[0134] Execution control layer:

[0135] Gate valve control: Connect to the existing gate RTU via Modbus TCP protocol, issue opening command (0-100%), and read real-time opening feedback.

[0136] Pump group control: Connect to the pump station PLC via OPC UA protocol to control the frequency converter frequency (0-50Hz) and start / stop status.

[0137] Water purification equipment: The aerator is started and stopped via dry contact, and the dosing pump speed is adjusted via a 4-20mA current signal.

[0138] The system ran continuously for 6 months, and the statistical results are as follows:

[0139] The irrigation guarantee rate increased from 82% to 94%;

[0140] The number of days with water quality exceeding standards decreased by 67% (from 18 days to 6 days);

[0141] Energy consumption for water distribution was reduced by 18% (from 168,000 kWh to 138,000 kWh).

[0142] The time for generating scheduling plans has been reduced from 2 hours manually to 8.5 seconds;

[0143] The model's root mean square error (COD) was 2.3 mg / L, which is better than the 5.7 mg / L of the traditional method.

[0144] Example 3: Comparative Experiment

[0145] In the same irrigation district and during the same period (spring irrigation season from April to June 2024), a comparative experiment was conducted between the method of this invention (Scheme A), the traditional single water quantity scheduling method (Scheme B), and the existing quality-quantity synergy technology (Scheme C, based on the technology of the Ministry of Water Resources in 2021). Scheme B only aims to minimize irrigation water shortage without considering water quality; Scheme C adopts an offline calibrated quality-quantity synergy model without online adaptive updates.

[0146] index Solution A (This invention) Option B Option C Irrigation water shortage (10,000 m³) 86 215 124 Duration of water quality exceeding standards (in hours) 22 168 61 Scheduling plan generation time (seconds) 8.5 320 185 Root mean square error of model prediction (COD, mg / L) 2.3 not applicable 5.7 Energy consumption (10,000 kWh) 12.6 16.8 14.1 System response time (seconds) 0.8 No automation 45

[0147] The results demonstrate that the present invention is significantly superior to existing technologies in terms of water supply assurance, water quality control, computational efficiency, prediction accuracy, and energy consumption.

[0148] Typical Case

[0149] In May 2025, the irrigation district experienced a 30-day drought, and the water quality of the upstream water fluctuated drastically (COD surged from 15 mg / L to 28 mg / L over 5 days). The traditional scheduling plan originally intended to maintain surface water diversion at 3.0 m³ / s, but if implemented, the COD at the downstream control section was predicted to reach 32 mg / L, exceeding the threshold of 20 mg / L, and the irrigation water shortage would reach 1.5 million m³.

[0150] After the system detects three consecutive deviations of the measured COD value exceeding 5 mg / L in step S4, it automatically initiates Bayesian online learning to update the pollutant degradation coefficient K from 0.15 / d to 0.22 / d. A new scheme is then generated after re-optimization.

[0151] Surface water flow was reduced to 2.2 m³ / s;

[0152] Increase groundwater extraction by 0.6 m³ / s (groundwater quality is good, COD < 10 mg / L);

[0153] Artificial wetlands were activated ahead of schedule, increasing operational intensity to 80%.

[0154] Adjust the irrigation rotation sequence to prioritize supplying vegetable growing areas with lower water quality tolerance.

[0155] After actual implementation, the peak COD level at the downstream control section was 19.5 mg / L (below the threshold of 20 mg / L), and the irrigation water demand was met at 97%, avoiding water quality exceeding standards and irrigation failures. Meanwhile, adjacent irrigation districts using traditional methods suffered three days of irrigation failure due to excessive COD, resulting in direct economic losses of approximately 2 million yuan.

[0156] Experimental Example: Model Performance Evaluation

[0157] Experimental conditions: The irrigated area used actual measurement data for the whole year of 2024 (365 days in total, collected every 30 minutes). The first 200 days were used for training, and the last 165 days were used for testing.

[0158] Evaluation indicators:

[0159] Root Mean Square Error (RMSE):

[0160] Mean Absolute Percentage Error (MAPE):

[0161] Calculation time (seconds / time)

[0162] Rate of decrease in prediction error after update

[0163] result:

[0164] Model RMSE (COD, mg / L) MAPE (flow, %) Calculation time (seconds / time) Error reduction rate after update Pure mechanistic model 4.8 8.2 6.4 - Pure data-driven (LSTM) 6.2 10.5 0.05 - Hybrid model of this invention (not updated) 3.6 4.5 0.08 - This invention utilizes a hybrid model (Bayesian online update). 2.1 2.8 0.08 41.7%

[0165] Conclusion: The hybrid model of this invention outperforms pure mechanism or pure data-driven models in both accuracy and speed, and the Bayesian online update further reduces the error by more than 40%.

[0166] Supplementary explanation of specific implementation methods

[0167] Those skilled in the art will understand that the method of the present invention is not only applicable to the aforementioned North China irrigation area, but also to other well-canal combined irrigation areas or gravity-fed irrigation areas. The deep neural network structure in step S2 can be adjusted according to the complexity of the actual problem and the amount of data; for example, a convolutional neural network (CNN) can be used to process spatial topological features, or a long short-term memory network (LSTM) can be used to process time series. The weight coefficients α, β, and γ in step S3 can be dynamically adjusted by irrigation area managers according to actual preferences (e.g., increasing α when drought is severe). The bias threshold and Bayesian prior distribution in step S4 can be set according to the quality of the monitoring data.

[0168] Industrial applicability

[0169] This invention can be applied to irrigation district management units at all levels, water conservancy information companies, and smart agriculture service providers, serving as the core algorithm and system for intelligent water resource scheduling and decision-making in irrigation districts. By integrating with existing automatic gate pump control systems, a closed-loop "perception-simulation-decision-control" mechanism can be achieved, significantly improving water resource utilization efficiency and environmental safety in irrigation districts.

[0170] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for simulating and regulating water resources in an irrigation area under the control of quantity and quality, characterized in that, include: S1. Obtain multi-source basic data for the irrigation area, including hydrological, meteorological, canal system topology, water source quantity, water quality, crop water requirements and irrigation system data; S2. Construct a coupled simulation model of water quantity and quality in the irrigation area. The model adopts a hybrid architecture of mechanistic model and deep learning proxy model: The mechanistic model is based on the Saint-Venant equations and convection-diffusion equations to construct a water quantity-water quality coupled physical process, providing physical constraint boundaries; Deep learning surrogate models use the historical inputs and outputs of mechanistic models as training samples and employ deep neural networks to construct surrogate models, thereby accelerating computation. S3. Based on the calculation results of the simulation model, establish a quantity and quality dual-control optimization and regulation model with multiple objectives of minimizing irrigation water shortage, minimizing water quality exceedance risk, and minimizing energy consumption, and solve for the coordinated operation scheme of multi-source joint allocation, gate pump scheduling and water purification. S4. Collect real-time water volume, water quality, and gate pump operating data, calculate the deviation between the simulated predicted value and the measured value, and when the deviation exceeds the preset threshold, start Bayesian online learning to dynamically update the parameters of the hybrid model and synchronize the updated parameters to the optimization and control model. S5. The control scheme generated by the optimized control model is sent to the field execution unit to drive the gate valve, pump group and water purification equipment to perform control actions, so as to realize the integrated closed-loop control of total water quantity control, water quality threshold control and precise execution of water distribution in the irrigation area.

2. The quantity and quality controlled irrigation district water resource simulation and regulation method according to claim 1, characterized in that, The water quantity-water quality coupled simulation model also includes: water inflow forecast module, canal system hydrodynamic module, field water balance module, and pollutant transport and degradation module. Each module adopts a mechanism-data dual-driven architecture to realize the spatiotemporal synchronous simulation of the quantity and quality process.

3. The method for simulating and regulating irrigation district water resources with both quantitative and qualitative control as described in claim 1, characterized in that, The objective function of the quantitative and qualitative dual-control optimization and regulation model is: in, For irrigation water shortage, Total water demand This represents the cumulative value of water quality exceeding standards. Water quality threshold Actual energy consumption The maximum allowable energy consumption is represented by α, β, and γ, which are weighting coefficients. The constraints include: total water intake, canal flow capacity, water quality control section concentration, groundwater level and extraction volume.

4. The method for simulating and regulating irrigation district water resources with both quantitative and qualitative control as described in claim 1, characterized in that, include: Real-time monitoring of water quantity, water quality, and soil moisture data is used to perform rolling corrections on the simulation model and the control model. The rolling correction works in conjunction with Bayesian online learning: Bayesian online learning is responsible for the global update of model parameters, while rolling correction is responsible for the local correction of short-term deviations, thereby achieving dynamic adaptive control across multiple time scales.

5. The method for simulating and regulating irrigation district water resources with both quantitative and qualitative control as described in claim 1, characterized in that, The coordinated operation plan includes: the joint allocation ratio of surface water and groundwater, the timing of water distribution flow in main and branch canals, the gate and pump opening and closing strategy, and the intensity of operation of purification facilities.

6. A water resource simulation and regulation system for irrigation districts with both quantitative and qualitative control, characterized in that, include: The data sensing layer is used to collect real-time data on irrigation area water quantity, water quality, weather, soil moisture, and canal system conditions; The data transmission and storage layer is used for data cleaning, aggregation, and standardized management. The simulation computing layer deploys the water quantity-water quality coupled simulation model and the quantity-quality dual-control optimization and regulation model as described in any one of claims 1-5, wherein the simulation model adopts a mechanism-data dual-driven hybrid architecture, and the optimization and regulation model integrates a Bayesian online learning module. The decision application layer is used to generate scheduling plans, visualize the results, issue instructions, and evaluate the effects. The execution control layer is used to receive control commands and drive gate valves, pump sets, and water purification equipment to perform control actions.

7. The irrigation district water resources simulation and regulation system with both quantity and quality control as described in claim 6, characterized in that, The data sensing layer includes a water level and flow meter, an online water quality monitoring instrument, a soil moisture station, a weather station, and a gate pump operating condition acquisition terminal.

8. The irrigation district water resources simulation and regulation system with both quantity and quality control as described in claim 6, characterized in that, The decision application layer supports multi-scenario simulation comparison, anomaly warning, one-click issuance of scheduling instructions and traceability of operation logs; it also supports digital twin visualization to display the spatiotemporal distribution of water volume and water quality in the canal system and the control effect in real time.

9. A water resource simulation and regulation system for irrigation districts with both quantitative and qualitative control as described in claim 6, characterized in that, The system supports integration with irrigation district smart management platforms and water resource monitoring platforms to achieve data sharing and cross-system collaborative management and control. The simulation computing layer supports edge-cloud collaborative computing, with real-time Bayesian online learning tasks executed on edge nodes and large-scale historical data training tasks executed in the cloud.

Images