An intelligent irrigation control system and method for dynamic balance of water and salt in saline-alkali soil

Abstract

The application discloses a kind of saline-alkali soil water salt dynamic balance intelligent irrigation control system and method, belong to wisdom agricultural technology field.It includes the following steps: collecting the dynamic monitoring data of influence water salt dynamics and saving to basic database;Based on the data of basic database, the crop evapotranspiration water requirement is calculated;Combining the calculated crop evapotranspiration water requirement and the data of basic database, realize salt profile estimation, set dynamic target salt threshold, calculate self-adapting leaching water amount, evaluate salt return risk and generate growth season end evaluation report;Based on all calculation and evaluation results, using reinforcement learning algorithm generates and optimizes key strategy control parameters for guiding irrigation;Combined with all the results obtained and fusion meteorological forecast data, generate adaptive irrigation strategy and auxiliary measures suggestion;Irrigation strategy is converted into irrigation and drainage executable instruction, while monitoring the execution state of irrigation and drainage is used to realize double-path feedback.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, and more specifically, to an intelligent irrigation control system and method for dynamic water-salt balance in saline-alkali land. Background Technology

[0002] Saline-alkali soils are widely distributed globally, especially in arid and semi-arid regions, and are a significant factor restricting sustainable agricultural development. The excessively high salt concentration in their soil solution exerts osmotic stress on crops, making it difficult for them to absorb water. Simultaneously, specific ions (such as Na+) can cause salinity. + Cl - and SO4 2- When these substances reach a certain concentration, they can cause direct toxicity. Saline-alkali soils are also accompanied by high pH values ​​and high exchangeable sodium accumulation, which not only reduces the availability of essential nutrients such as phosphorus, iron, and zinc in the soil, but also damages soil aggregate structure, leading to soil compaction and poor water permeability and aeration. These problems severely inhibit the normal growth and development of crops, resulting in decreased land productivity and low water resource utilization efficiency.

[0003] Traditional irrigation management methods, such as experience-based timed and quantitative irrigation, often face a dilemma: insufficient irrigation fails to meet the needs of crop growth and root zone salt leaching, accelerating salt damage; while excessive irrigation (especially under poor drainage conditions) not only wastes precious water resources but may also raise the saline groundwater level, exacerbating the risk of secondary salinization of the topsoil through strong soil evaporation. This contradiction makes it difficult to balance salt control with water conservation goals, and short-term leaching effects with long-term salt return risks.

[0004] Existing technologies attempt to optimize irrigation through crop models or water-salt transport models to improve irrigation efficiency and accuracy. For example, Chinese patent document CN114698535B discloses a method, system, electronic device, and storage medium for precision irrigation of crops. This system determines irrigation water volume and irrigation time nodes based on meteorological data, soil data, and indicators such as historical evapotranspiration, crop coefficient during growth period, and growth period calendar, thereby improving the accuracy of irrigation water volume and irrigation time nodes. Another example is Chinese patent document CN113141940A, which discloses an intelligent water precision irrigation control system and method for fruit and vegetable cultivation in solar greenhouses. Its focus is on determining irrigation time and volume based on the growth status of vegetables and environmental parameters, eliminating the need for manual operation and realizing intelligent irrigation for facility cultivation. However, these technologies are still insufficient when applied to saline-alkali land: (1) One-sided parameter monitoring: There is a lack of comprehensive, real-time and multi-depth monitoring of key parameters such as soil salinity, pH, irrigation water salinity, groundwater level and salinity in saline-alkali land, which leads to irrigation decisions relying solely on water balance and failing to dynamically calculate leaching demand in conjunction with the principle of salinity balance, easily falling into the contradiction of "controlling salt requires large amounts of water, saving water leads to salt damage"; (2) Lack of risk assessment: There is no integrated mechanism for assessing the risk of salt return from groundwater salinity dynamics, soil evaporation and meteorological data, making it difficult to warn of the risk of secondary salinization caused by the rise in groundwater level after irrigation; (3) Disconnection of cross-seasonal management: Focusing on short-term regulation within the growing season, ignoring the long-term cumulative effect of salt, and lacking soil improvement decision support based on monitoring data throughout the season, resulting in unsustainable treatment effects. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention provides an intelligent irrigation control system and method for dynamic water-salt balance in saline-alkali land. It can achieve precise salt control, water conservation and salt suppression, and long-term improvement through a coupling mechanism of dynamic threshold setting and risk assessment.

[0006] The objective of this invention is achieved through the following technical solutions.

[0007] The summary section of this application is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0008] As a first aspect of this application, some embodiments of this application provide an intelligent irrigation regulation method for dynamic water-salt balance in saline-alkali land, comprising the following steps:

[0009] Step 1: Collect dynamic monitoring data that affects water and salt dynamics and save it to the basic database. The dynamic monitoring data includes meteorological parameters, soil parameters, groundwater and irrigation water parameters, and meteorological forecast data.

[0010] Step 2: Calculate crop evapotranspiration water requirement based on data from the basic database. Combine the calculated crop evapotranspiration water requirement with the salt profile estimation, dynamic target salt threshold setting, adaptive leaching water volume calculation, and salt return risk assessment to generate an end-of-growing-season assessment report. Based on the crop evapotranspiration water requirement, salt profile estimation results, dynamic target salt threshold, adaptive leaching water volume calculation results, salt return risk assessment results, and end-of-growing-season assessment report, use reinforcement learning algorithms to generate and optimize key strategy control parameters for guiding irrigation.

[0011] Step 3: Combine the crop evapotranspiration water demand, salinity profile estimation results, dynamic target salinity threshold, adaptive leaching water volume calculation results, salt return risk assessment results and key strategy control parameters obtained in Step 2, and integrate meteorological forecast data to generate an adaptive irrigation strategy.

[0012] Step 4: Transform the irrigation strategy into executable instructions for irrigation and drainage.

[0013] Furthermore, after step 4 is executed, the execution status of irrigation and drainage is monitored in real time, and the irrigation strategy is adjusted according to the execution status of irrigation and drainage. At the same time, execution process data including irrigation amount and duration are obtained, as well as soil salinity change data after irrigation as process effect data. The execution process data and process effect data are fed back to the reinforcement learning optimization unit to optimize the next irrigation strategy.

[0014] Furthermore, the basic database includes dynamic monitoring data as well as pre-set crop basic data and localized parameter library data within the system;

[0015] Meteorological parameters include solar radiation, air temperature, air humidity, wind speed, and rainfall;

[0016] Soil parameters include soil moisture, soil salinity, soil temperature, and soil pH.

[0017] Groundwater and irrigation water parameters include groundwater level, groundwater conductivity, and irrigation water conductivity;

[0018] Weather forecast data includes rainfall, temperature, and wind speed for the next three days;

[0019] Basic crop data includes the criteria for dividing growth stages and the corresponding crop coefficients, crop salt tolerance thresholds, and root depths at different growth stages;

[0020] The localized parameter library includes soil hydraulic function parameters describing soil water holding and water conduction properties, and solute transport parameters describing the diffusion and migration characteristics of salts in the soil.

[0021] Furthermore, basic crop data includes the standards for dividing growth stages and the corresponding crop coefficients, crop salt tolerance thresholds, and root depth at different growth stages.

[0022] The process of setting the dynamic target salinity threshold is as follows: retrieve basic crop data from the basic database, and set the target salinity threshold according to crop type and growth stage.

[0023] Furthermore, the steps for calculating crop evapotranspiration water demand are as follows: based on the collected meteorological parameters, the standard reference crop evapotranspiration ET0 is calculated using the FAO Penman-Monteith formula;

[0024] Based on the crop's growth stage, the corresponding crop coefficient K is obtained from the basic database. c The water requirement ET for basic crops was calculated. c_base The calculation formula is:

[0025] ET c_base = K c × ET0;

[0026] Based on real-time monitoring of root zone soil salinity and crop salt tolerance thresholds in a basic database, the salt stress coefficient K is calculated. s The basic crop water requirement is then corrected to obtain the final crop evapotranspiration water requirement ET. c The calculation formula is:

[0027] ET c = K s × ET c_base .

[0028] Furthermore, the steps for calculating the adaptive rinsing water volume are as follows: Calculate the minimum rinsing fraction LR based on the steady-state salt balance principle. The calculation formula is:

[0029] LR = EC iw / (k × EC e_target );

[0030] In the formula, EC iw EC is the electrical conductivity of irrigation water. e_target The target salinity threshold is given by k, and the leaching efficiency coefficient is given by k.

[0031] Combined with the actual evapotranspiration water requirement (ET) of the crop calculated by the crop model unit after salt stress correction. c The minimum rinsing fraction LR is converted into an adaptive rinsing water volume V. leach The calculation formula is:

[0032] V leach = LR×ET c / (1-LR)。

[0033] Furthermore, the risk of salt return is categorized into low, medium, and high risk; the steps for assessing the risk of salt return include:

[0034] When the groundwater level is greater than 2m and the groundwater conductivity EC gw Less than or equal to 6 dS / m, or groundwater level between 1m and 2m and groundwater conductivity EC gw When the value is less than 3 dS / m, the risk of salt return is low.

[0035] When the groundwater level is greater than 2m and the groundwater conductivity EC gw Greater than 6 dS / m, or groundwater level between 1m and 2m and groundwater conductivity EC gw Between 3 dS / m and 6 dS / m, or when the groundwater level is less than 1m and the groundwater conductivity EC gw When the value is less than 3 dS / m, the risk of salt return is medium.

[0036] When the groundwater level is between 1m and 2m and the groundwater conductivity ECgw is greater than 6 dS / m, or when the groundwater level is less than 1m and the groundwater conductivity ECgw is greater than 6 dS / m, gw When the flow rate is greater than or equal to 3 dS / m, the risk of salt return is high.

[0037] If the daily reference evapotranspiration ET0 is greater than the correction threshold of 6 mm / day and the risk of salt return is low, then the risk of salt return is upgraded to medium risk.

[0038] If the daily reference evapotranspiration ET0 is greater than the correction threshold of 6 mm / day and the risk of salt return is medium, then the risk of salt return is upgraded to high.

[0039] As a second aspect of this application, some embodiments of this application provide an intelligent irrigation control system for the above-mentioned intelligent irrigation control method for dynamic water-salt balance in saline-alkali land, including a data acquisition module, a model and evaluation module, a decision-making module, and an execution module. The modules transmit and process data through a wireless communication network and computing and storage resources deployed on a cloud platform or local server.

[0040] The data acquisition module collects dynamic monitoring data and saves it to the basic database; the model and evaluation module retrieves data from the basic database to calculate crop evapotranspiration water requirements, adaptive leaching water volume, and assess salt return risk, generating strategy optimization parameters which are then output to the decision module; the decision module integrates the output of the model and evaluation module with meteorological forecast data retrieved from the basic database to generate irrigation strategies and executable instructions, which are then sent to the execution module; the execution module coordinates irrigation and drainage to complete intelligent irrigation control and provides dual-path feedback.

[0041] Dual-path feedback includes triggering immediate adjustments or alarms based on the execution status, and feeding back execution process data and process effect data monitored by the data acquisition module to the model and evaluation module for optimization of the next irrigation strategy;

[0042] The basic database includes dynamic monitoring data continuously collected by the data acquisition module, as well as crop basic data and localized parameter libraries pre-set during the system deployment phase.

[0043] Furthermore, the model and evaluation module includes a collaborative crop model unit, a water-salt dynamic balance and risk assessment unit, and a reinforcement learning optimization unit.

[0044] The crop model unit is used to calculate the actual evapotranspiration water requirement of crops considering salt stress correction; the water-salt dynamic balance and risk assessment unit realizes salt profile estimation, dynamic target salt threshold setting, adaptive leaching water volume calculation, salt return risk assessment, and end-of-growing-season assessment report generation by calling the basic database; the reinforcement learning optimization unit generates and optimizes key strategy control parameters for guiding irrigation through interactive learning between the agent and the simulation environment.

[0045] Furthermore, the reinforcement learning optimization unit includes a simulation environment subunit, an agent subunit, a training and optimization subunit, and a policy output and decision feedback subunit.

[0046] The simulation environment subunit is used to construct a simulation environment for intelligent agent interaction to dynamically simulate crop growth and dynamic soil water and salt response under different irrigation strategies.

[0047] The agent subunit includes a state input subunit, an action output subunit, and a reward function calculation subunit; the state input subunit integrates input data; the action output subunit generates policy correction coefficients based on the agent's learning policy; and the reward function calculation subunit guides the agent's policy learning through the reward function.

[0048] The training and optimization subunit executes the training process using reinforcement learning algorithms, updating policy parameters based on the accumulated reward information after repeated interactions between the agent and the environment;

[0049] The strategy output and decision feedback subunit transmits the strategy correction coefficients generated by the agent to the decision module, and receives feedback data on the execution effect of the irrigation strategy to calibrate the simulation environment, thereby realizing strategy iteration and optimization.

[0050] Compared with the prior art, the advantages of this invention are:

[0051] (1) Comprehensive monitoring and dynamic precise control: This invention overcomes the one-sidedness of existing technologies that rely solely on water balance by conducting comprehensive and multi-depth real-time monitoring of meteorology, soil (moisture, salinity, pH), irrigation water quality, and groundwater (water level, salinity); at the same time, it combines dynamically set target salinity thresholds for crop growth stages to accurately meet crop water requirements and salt control needs. Adjustable target salinity thresholds are set based on the salt tolerance characteristics of crop growth stages, and crop water requirements are dynamically corrected by combining real-time salt stress coefficients; adaptive leaching water volume is calculated through the principle of salt balance, and the salt control intensity is dynamically adjusted according to irrigation water quality and changes in salinity in the root zone to ensure that the salinity in the root zone is always within a safe range.

[0052] (2) Integrating salt return risk early warning for proactive defense: This invention integrates the salt return risk assessment process. By comprehensively analyzing factors such as groundwater and evaporation intensity, it can provide early warning and quantify the risk of secondary salinization caused by irrigation or environmental changes. This allows irrigation strategies to not only focus on immediate effects but also avoid long-term risks, solving the problem that existing technologies lack risk control and may exacerbate salinization.

[0053] (3) Balancing seasonal management with long-term soil improvement: This invention breaks through the limitations of traditional irrigation systems that focus on regulation during the growing season. Through the assessment report and soil health diagnosis at the end of the growing season, it proposes targeted cross-seasonal saline-alkali land improvement suggestions (such as the application of soil conditioners, organic materials, etc.). This long-term management mechanism ensures the sustainability of saline-alkali land management. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall structure of an intelligent irrigation control system for dynamic water-salt balance in saline-alkali land according to an embodiment of the present invention;

[0055] Figure 2 This is a flowchart of an intelligent irrigation control method for dynamic water-salt balance in saline-alkali land, as described in one embodiment of the present invention. Detailed Implementation

[0056] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0057] like Figure 1 As shown, the intelligent irrigation control system for dynamic water-salt balance in saline-alkali land according to the present invention includes a data acquisition module, a model and evaluation module, a decision-making module, and an execution module. The various modules of the intelligent irrigation control system achieve data transmission and processing through a wireless communication network and computing and storage resources deployed on a cloud platform or local server.

[0058] The data acquisition module collects dynamic monitoring data and saves it to the basic database. The model and evaluation module includes a crop model unit, a water-salt dynamic balance and risk assessment unit, and a reinforcement learning optimization unit. The model and evaluation module retrieves data from the basic database and works collaboratively through its internal units based on their functional divisions: First, the crop model unit calculates the actual evapotranspiration water requirement (ET) of the crop based on the data in the basic database. c Subsequently, the water-salt dynamic balance and risk assessment unit received ET c In conjunction with other data, the system completes the estimation of salinity profiles, setting of dynamic target salinity thresholds, calculation of adaptive leaching water volume, assessment of salt return risk, and generation of an end-of-growing-season assessment report. Finally, the reinforcement learning optimization unit integrates all the outputs of the first two units to generate and optimize key strategy control parameters for guiding irrigation. The model and evaluation module sends the calculation and evaluation results, along with strategy correction coefficients, to the decision module. The decision module, combining the received data and weather forecast data, generates an irrigation strategy and sends it to the execution module, while also generating supplementary measures suggestions for the irrigation strategy. The execution module performs intelligent irrigation regulation based on the received irrigation strategy and feeds back the execution status for real-time control to the decision module, while feeding back the execution process data for model optimization to the model and evaluation module.

[0059] The base database is a central data repository deployed on a cloud platform or local server, used to manage all the data required for system operation. The base database includes dynamic monitoring data continuously collected by the data acquisition module, as well as pre-configured crop baseline data and localized parameter libraries from the system deployment phase.

[0060] Specifically, the data acquisition module collects dynamic monitoring data in real time through sensors and application programming interfaces (APIs), including meteorological parameters, soil parameters, groundwater and irrigation water parameters, and weather forecast data.

[0061] Meteorological parameters include solar radiation, air temperature, air humidity, wind speed, and rainfall; soil parameters include soil moisture, soil salinity, soil temperature, and soil pH; groundwater and irrigation water parameters include groundwater level and groundwater electrical conductivity (EC). gw ) and irrigation water conductivity (EC) iw Weather forecast data includes rainfall, temperature, and wind speed for the next three days.

[0062] Crop baseline data is a set of structured parameters built and pre-loaded into the baseline database during the system deployment and configuration phases by inputting agronomic manuals (such as FAO publications) or localized experimental data. As the foundation for accurate calculations of crop model units and water-salt dynamic balance and risk assessment units, crop baseline data includes growth stage classification standards and corresponding crop coefficients (K). c), crop salt tolerance threshold, and root parameters at different growth stages.

[0063] The growth stage classification standard defines the various growth stages (such as seedling stage, flowering stage, and maturity stage) of different crops from sowing to maturity, and their corresponding growth degree days (GDD) or calendar days;

[0064] Crop coefficient (K) c () is a coefficient that corresponds one-to-one with each growth stage and is used to calculate the basic water requirement of crops at different growth stages;

[0065] Crop salt tolerance thresholds are core data for saline-alkali land management, used to quantify the degree of salt stress, including salt tolerance thresholds (EC) based on the Maas-Hoffman model. e_threshold ) and the limit salinity value (EC) e_100% Salt tolerance threshold (EC) e_threshold This refers to the critical soil salinity level at which crop yields begin to decline, also known as the limiting salinity value (EC). e_100% This refers to the soil salinity level that causes crop yields to drop to zero.

[0066] Root parameters include the maximum root depth of the crop at different growth stages, which are used to determine the target soil layer range for irrigation and salinity control.

[0067] The localized parameter library is a set of static parameters specifically designed for numerical models of water and salt transport. It is obtained through model calibration and pre-installed in the base database. Specifically, the parameter library includes soil hydraulic function parameters that describe the water holding and water conduction properties of soil, as well as solute transport parameters that describe the diffusion and migration properties of salt in soil.

[0068] In one specific embodiment, an automatic weather station in the monitoring area incorporates a sensor array to collect meteorological parameters at a regular frequency of once per hour. The system automatically switches to a high-frequency acquisition mode based on the following events to accurately capture environmental dynamics during key hydrological events:

[0069] Irrigation event: When the system's decision module issues an irrigation command and the execution module starts the irrigation task, a high-frequency acquisition mode is triggered, increasing the acquisition frequency of meteorological parameters to once every 5 to 15 minutes, and continuing until the irrigation task is completed. Afterward, the system automatically returns to the normal acquisition frequency.

[0070] Rainfall Event: When the rain gauge sensor built into the automatic weather station detects rainfall (e.g., the first tipping bucket meter is detected or the cumulative rainfall exceeds 0.1 mm), the high-frequency acquisition mode is triggered. If the rain gauge does not detect any new rainfall within a preset time period (e.g., 30 minutes), the system determines that the rainfall event has ended and automatically resumes the normal acquisition frequency.

[0071] In one specific embodiment, monitoring points are set up in the monitoring area, and sensors are vertically installed at each monitoring point to collect soil parameters. The collection frequency is once every 30 minutes or 1 hour. The layout of the monitoring points is determined based on the size of the monitoring area, the spatial heterogeneity of the soil, and typical landform features. The points can be evenly distributed using a grid method or targeted at key locations (such as highlands and depressions) to ensure that the collected soil parameters can represent the average condition and spatial variability of the entire area.

[0072] Specifically, a vertical profile multi-depth sensor or sensor array is used to cover the crop root activity layer at a depth of 0 cm to 60 cm in the soil to collect soil parameters at multiple points and depths, thereby accurately grasping the spatial variability and profile distribution characteristics of soil water and salt conditions.

[0073] More specifically, time-domain reflectometry (TDR), frequency-domain reflectometry (FDR), or capacitive sensors are used to simultaneously collect soil moisture, salinity, and temperature. Soil pH is collected using a soil pH electrode probe. Soil moisture at multiple depths reflects soil water storage and transport; soil salinity at multiple depths is used to directly monitor rhizosphere salinity; soil pH at multiple depths is used to monitor the soil acid-base environment; and soil temperature at multiple depths can affect evaporation and chemical reaction rates. In this embodiment, apparent electrical conductivity (EC) is used. a It can be used to characterize soil salinity.

[0074] In one specific embodiment, the sensor's acquisition frequency is increased during irrigation or rainfall, for example, every 5 to 15 minutes, thereby dynamically adjusting the acquisition frequency to ensure that the sensor's acquisition frequency can capture effective changes in soil conditions.

[0075] Online water conductivity sensors are deployed in irrigation water sources or main pipelines to collect irrigation water conductivity during real-time monitoring and irrigation, monitor the salt concentration of irrigation water in real time, and use it to calculate the subsequent rinsing water volume.

[0076] Automatic water level gauges and water conductivity sensors are deployed in groundwater monitoring wells to collect groundwater parameters for assessing the risk of capillary water salinization. Because groundwater parameters change slowly, they are collected at least once every 6 hours.

[0077] In addition, weather forecast data is automatically obtained by backend services deployed on cloud platforms or local servers.

[0078] Specifically, the backend service calls the application programming interface (API) of the third-party weather service provider at a preset frequency, such as once per hour, and periodically sends HTTPS requests carrying the geographic coordinates of the monitoring area to the provider's server. The weather service provider's server returns a response containing weather forecast data for the next few days in a common machine-readable format (such as JSON). After receiving the response, the backend service automatically parses it and stores the required key information such as temperature, rainfall, and wind speed into the basic database for use by other modules of the system.

[0079] Specifically, since farmland environments are typically large and wiring is difficult, this embodiment uses low-power wide-area IoT technology to achieve wireless data transmission of the data acquisition module. The data collected by the data acquisition module is transmitted to the LoRaWAN gateway deployed in or near the detection area via LoRa wireless signal, and then uploaded by the gateway to the basic database of the cloud server or local data center via cellular network, Ethernet or WiFi, for use by the model and evaluation module and the decision module.

[0080] Table 1 shows specific examples of sensor deployment.

[0081] Table 1 Sensor Deployment Examples

[0082]

[0083] The model and evaluation module includes a crop model unit, a water-salt dynamic balance and risk assessment unit, and a reinforcement learning optimization unit. The model and evaluation module takes data from the basic database as input, and through the collaborative work of each unit, generates calculation and evaluation results, as well as policy correction coefficients, which are then transmitted to the decision module. The crop model unit is responsible for estimating the crop's evapotranspiration water requirement (ET). c The data is then passed to the water-salt dynamic balance and risk assessment unit. Based on this water requirement and other input data, the water-salt dynamic balance and risk assessment unit estimates the salinity profile, sets a dynamic target salinity threshold, calculates adaptive leaching water volume, assesses the risk of salt overflow, and generates an end-of-growing-season assessment report. The reinforcement learning optimization unit generates optimization strategies based on the calculation and assessment results from the crop model unit and the water-salt dynamic balance and risk assessment unit. Finally, the calculation and assessment results generated by each unit, along with the strategy correction coefficients, are output to the decision module.

[0084] Specifically, the crop model unit estimates the actual evapotranspiration water requirement (ET) of crops under salt-alkali stress through its internally integrated crop growth and water response model (e.g., the AquaCrop model). cThis unit uses real-time meteorological data retrieved from a basic database to calculate the standard reference crop evapotranspiration (ET0) using the Penman-Monteith formula. Then, it automatically determines the crop's current growth stage (e.g., seedling, flowering, maturity) using the cumulative growing days (GDD) method or a preset growth calendar, and queries the basic database for the crop coefficient (K) corresponding to the current growth stage. c ), through the formula ET c_base =K c ×ET0 calculates the evapotranspiration water requirement of the basic crop under ideal conditions (ET). c_base ).

[0085] The crop coefficient reflects the differences between a specific crop and a reference crop at a specific growth stage in terms of canopy coverage, stomatal conductance, and canopy height. For example, the K value at the peak boll-forming stage of cotton. c Approximately 1.10-1.15, mid-stage K in tomatoes c It is approximately 1.10.

[0086] To accurately reflect the conditions of saline-alkali land, the crop model unit compares the real-time monitored root zone soil salinity with the crop salt tolerance threshold in the basic crop data, and calculates the salt stress coefficient (K) using a salt stress model based on the FAO-56 theory. s This is used to correct the baseline water requirement, and the final output is the evapotranspiration water requirement (ET), which represents the actual water demand of the crop at present. c The calculation formula is: ET c =K s ×ET c_base .

[0087] Specifically, the salt stress coefficient (K) s The formula for calculating ) is:

[0088] K s =(EC e_100% EC e_actual ) / (EC 100% EC e_threshold );

[0089] In the formula, EC e_actual EC represents the monitored soil salinity values ​​in the root zone. e_threshold The salt tolerance threshold for the current growth stage of the crop is retrieved from the basic database, under conditions of no salt stress (i.e., EC50). e_actual ≤ EC e_threshold ), K s The value is 1. EC e_100%The crop limit salinity value is a soil salinity level that reflects the minimum soil salinity required to reduce crop yield or transpiration to zero. The crop limit salinity value is a pre-determined, crop-variety-specific empirical parameter, obtained by consulting agronomic handbooks and standard databases. In the implementation of this invention...

[0090] The water-salt dynamic balance and risk assessment unit includes a numerical model of water-salt transport based on physical processes. This unit receives the water requirement ET transmitted from the crop model unit. c Subsequently, the evaluation and calculation are achieved by collaboratively calling the basic crop data in the basic database and the localized parameter library through the water and salt transport numerical model.

[0091] The water-salt dynamic balance and risk assessment unit first sets a dynamic target salinity threshold (EC) from a basic database based on the crop growth stage. e_target Then combined with ET c EC e_target And real-time monitoring of irrigation water conductivity (EC) iw The adaptive rinsing water volume (V) is calculated based on the steady-state salt balance principle. leach Meanwhile, considering the combined groundwater level and groundwater conductivity (EC), gw Based on the reference crop evapotranspiration (ET0), the risk of salt return is assessed using pre-defined quantitative rules, which employ either a risk matrix method or a quantitative scoring method. At the end of the growing season, water and salt balance is calculated based on the cumulative data from the entire growing season, and a growing season end-of-season assessment report is generated.

[0092] Specifically, the functional implementation process of the water-salt dynamic balance and risk assessment unit is as follows:

[0093] (1) Salinity profile estimation: Using soil salinity data and spatial interpolation techniques such as Kriging interpolation or inverse distance weighting, a complete or arbitrary range of salinity spatial distribution map of the monitoring area is generated. At the same time, in order to gain a deeper understanding of the dynamics of salinity in the vertical direction, a water-salt transport numerical model (such as HYDRUS) is used to simulate the collected monitoring point data to more accurately track the migration pattern of salinity peaks after irrigation or rainfall, thereby identifying hotspots and depths of salinity accumulation.

[0094] (2) Dynamic target salinity threshold setting: Based on the crop type and its growth stage (automatically determined by the system), the corresponding target salinity threshold (EC) for the root zone soil is queried from the basic database and set. e_target ).

[0095] In one specific embodiment, the soil salinity tolerance threshold for cotton, a moderately salt-tolerant crop, is approximately 7.7 dS / m. To ensure cotton emergence and robust seedling growth during the sensitive seedling stage, a target salinity threshold of 4.0 dS / m is set. During the flowering and boll-setting stage, cotton's tolerance to soil salinity increases, and the target salinity threshold is set at 8.0 dS / m to conserve leaching water while ensuring yield. The soil salinity tolerance threshold for tomato, a moderately sensitive crop, is approximately 2.5 dS / m. Since moderate salt stress can improve tomato fruit quality (such as sugar content and flavor compound content), the target salinity threshold is set at 6.0 dS / m during the flowering and fruit-setting stage. The target salinity threshold is dynamically adjusted according to production targets and the plant's growth stage to balance yield and quality.

[0096] Table 2 shows examples of dynamically adjusting the target salinity threshold based on crop type and its growth stage.

[0097] Table 2 Examples of dynamically adjusting target salinity thresholds

[0098]

[0099] (3) Adaptive rinsing water volume calculation:

[0100] In order to keep the actual salinity in the root zone below the target salinity threshold, additional leaching water is needed to leach the excess salt accumulated in the root zone below the root layer.

[0101] Specifically, the leaching water volume is calculated based on dynamically set salinity control targets and real-time water quality conditions. The dynamically set salinity control targets are determined based on the crop growth stage (EC). e_target Real-time water quality is measured by the irrigation water conductivity (EC) monitored in real-time by the data acquisition module before or during irrigation. iw Based on the principle of steady-state salinity balance, this unit calculates the adaptive rinsing water volume required to maintain the salinity balance in the root zone. The calculation process is as follows:

[0102] To achieve salt balance, a dimensionless ratio characterizing leaching intensity, namely the minimum leaching fraction (LR), is first determined. This reflects the proportion of irrigation water required to create deep seepage through the root zone to remove excess salt. The calculation formula is as follows:

[0103] LR = EC iw / (k × EC e_target );

[0104] In the formula, LR represents the minimum rinsing fraction, and EC iw The irrigation water conductivity, EC, is monitored in real time by the data acquisition module. e_targetis the target salinity threshold for the current growth stage, and k is the leaching efficiency coefficient, a comprehensive parameter whose value (usually less than 1) characterizes the effectiveness of irrigation water in replacing and removing salts from the soil solution. It is affected by various factors such as soil texture, structure, and irrigation method, and needs to be localized according to specific fields. In this embodiment, k = 0.7 is set.

[0105] Then, the actual evapotranspiration water requirement (ET) of the crop, corrected for salt stress, calculated by the crop model unit, is combined. c The process of converting LR into a specific adaptive rinsing water volume with units is expressed as follows:

[0106] V leach = LR×ET c / (1-LR)

[0107] In the formula, V leach This indicates adaptive rinsing water volume, which ensures the total irrigation water volume (i.e., ET) c +V leach It can not only fully meet the water demand of crop growth, but also provide enough extra water to maintain the salt content in the root zone at a preset safe level, thus precisely achieving the dual goals of irrigation and salt control.

[0108] (4) Risk assessment of salt return:

[0109] To accurately simulate water and salt transport processes under unsteady conditions and consider the influence of complex factors such as solute dispersion, root water uptake, and surface salt accumulation caused by evaporation on salt distribution, the water-salt dynamic balance and risk assessment unit employs a water-salt balance model. By numerically solving the Richards equation describing water flow and the convection-dispersion equation describing solute transport, it simulates and predicts the spatiotemporal dynamic changes of moisture content and salt concentration in soil profiles under specific irrigation management, crop growth, and meteorological conditions. This allows for a more accurate prediction of the irrigation water volume (including leaching water volume) required to reach the target salt threshold and enables quantitative assessment of actual leaching efficiency and salt leaching depth.

[0110] The water-salt balance model is a one-dimensional, two-dimensional, or three-dimensional water-salt transport model based on physical processes. One-dimensional models assume that transport mainly occurs along the vertical direction, while two-dimensional or three-dimensional models consider coupled transport in the horizontal and vertical directions. In this embodiment, the water-salt balance model can be a HYDRUS series model or a SWAP model.

[0111] In a specific embodiment, targeting arid and semi-arid regions, this unit comprehensively monitors the regional groundwater level and groundwater conductivity EC. gw Based on the calculated daily reference evapotranspiration ET0, the risk of salt return is classified into low, medium, and high risk levels. The specific process is as follows:

[0112] First, determine the basic risk level:

[0113] When the groundwater level is greater than 2m and the groundwater conductivity EC gw Not greater than 6 dS / m, or the groundwater level is between 1m and 2m and the groundwater conductivity EC gw When the value is less than 3 dS / m, the basic risk level is low.

[0114] When the groundwater level is greater than 2m and the groundwater conductivity EC gw Greater than 6 dS / m, or groundwater level between 1m and 2m and groundwater conductivity EC gw Between 3 dS / m and 6 dS / m, or when the groundwater level is less than 1m and the groundwater conductivity EC gw When the value is less than 3 dS / m, the basic risk level is medium risk.

[0115] When the groundwater level is between 1m and 2m and the groundwater conductivity ECgw is greater than 6 dS / m, or when the groundwater level is less than 1m and the groundwater conductivity ECgw is greater than 6 dS / m, gw When the value is not less than 3 dS / m, the basic risk level is high risk.

[0116] Secondly, risk adjustments are made based on evaporation:

[0117] If the daily reference evapotranspiration ET0 is greater than the correction threshold of 6 mm / day, the basic risk level will be raised by one level; if the basic risk level is already "high risk", the "high risk" level will remain unchanged.

[0118] In one specific embodiment, the assessment of salt return risk can also employ a quantitative scoring method: obtaining the groundwater level and groundwater conductivity EC in the monitoring area. gw The daily reference evapotranspiration ET0 is calculated based on the collected meteorological parameters. By assigning preset scores to each parameter and calculating the total risk score, the risk of salt return is divided into low-risk, medium-risk, and high-risk levels. The specific process is as follows:

[0119] First, we will conduct a scoring process:

[0120] For groundwater level scores: when the groundwater level is less than 1 m, the groundwater level score is 3 points; when the groundwater level is between 1 m and 2 m, the groundwater level score is 1 point; when the groundwater level is greater than 2 m, the groundwater level score is 0 points.

[0121] For groundwater conductivity EC gw Score: When EC gw When the conductivity is greater than 6 dS / m, the electrical conductivity EC of groundwater gwThe score is 3 points; when EC gw When the electrical conductivity of groundwater is between 3 dS / m and 6 dS / m, the EC value is... gw 1 point is awarded; when EC gw When the conductivity is less than 3 dS / m, the electrical conductivity EC of groundwater is... gw The score is 0 points.

[0122] For the daily reference evapotranspiration ET0 score: when ET0 is greater than 6 mm / day, the daily reference evapotranspiration ET0 score is 2 points; when ET0 is between 3 mm / day and 6 mm / day, the daily reference evapotranspiration ET0 score is 1 point; when ET0 is less than 3 mm / day, the daily reference evapotranspiration ET0 score is 0 points.

[0123] Secondly, calculate the total risk score by adding the scores of the above parameters together. The risk level is then determined based on the total risk score: when the total risk score is less than or equal to 2 points, the risk level is low risk; when the total risk score is between 3 and 5 points, the risk level is medium risk; and when the total risk score is greater than or equal to 6 points, the risk level is high risk.

[0124] (5) Generation of end-of-growing-season assessment report:

[0125] After crop harvest, the water and salt dynamic balance and risk assessment unit collects data from the entire growing season using the data acquisition module to perform water and salt balance calculations and generate a growing season end-of-season assessment report. This report uses a numerical model of water and salt transport to perform a retrospective simulation of the entire growing season and calculate the water and salt balance, generating an assessment report containing detailed information such as total salt input / output, changes in rhizosphere salt storage, and trends in salt profile evolution. The growing season end-of-season assessment report includes the total salt input during the growing season, salt output (crop absorption and leaching), changes in rhizosphere salt storage, trends in salt profile distribution, and the degree of salt accumulation or leaching compared to the beginning of the season (e.g., changes in net salt content in kg / ha).

[0126] Specifically, the reinforcement learning optimization unit constructs an environmental state vector by integrating the outputs of the crop model unit and the water-salt dynamic balance and risk assessment unit, as well as data from the basic database. The reinforcement learning optimization unit then generates a set of irrigation strategy correction coefficients through interaction between the agent and the simulation environment.

[0127] The reinforcement learning optimization unit comprises an agent trained through interaction with a simulated "virtual experimental field" environment. This simulation environment is constructed by coupling multiple models, including those related to crop growth and water-salt transport. During training, the agent learns the optimal mapping strategy from environmental states to a series of actions, based on a reward function aimed at maximizing crop yield, improving water-salt use efficiency, and reducing costs. Actions are not directly related to irrigation volume, but rather to a set of strategy correction coefficients (such as scaling factors for irrigation quotas and correction coefficients for leachate volume). Upon completion of training, the reinforcement learning optimization unit outputs an agent model capable of generating these optimized strategy correction coefficients. During system operation, actual irrigation performance data is fed back to the reinforcement learning optimization unit and used to continuously calibrate the model parameters of the simulation environment, thereby achieving self-iteration and closed-loop learning of the optimization strategy.

[0128] The reinforcement learning optimization unit is the core of this system to achieve advanced intelligence and adaptive capabilities. It does not replace the basic calculations of the crop model unit and the water-salt dynamic balance and risk assessment unit, but rather, on top of them, it uses deep learning and simulation trial and error to find the optimal strategy among multiple conflicting objectives such as yield, water conservation, salt control, and cost, thereby surpassing the traditional decision-making mode based on fixed rules or thresholds.

[0129] The simulation environment subunit is responsible for providing a high-fidelity, risk-free "virtual test field" for the training of the agent. It simulates the crop and soil responses under different irrigation strategies by coupling crop growth models (such as WOFOST or ORYZA) and water and salt transport models (such as HYDRUS or SWAP).

[0130] Specifically, the coupling process between the crop growth model and the water and salt transport model is achieved using standard co-simulation techniques. For example, the independent models are encapsulated through the Functional Model Interface (FMI) standard to enable cross-platform data exchange and joint simulation.

[0131] The agent subunit is the functional carrier that enables the core decision-making logic of the agent to interact with the simulation environment. It includes a state input subunit, an action output subunit, and a reward function calculation subunit. Each subunit works together to complete the complete interaction process.

[0132] The State input subunit integrates all the information needed for decision-making into a multi-dimensional State Vector, representing the agent's complete perception of the current environment. Specifically, the State Vector includes: soil moisture, soil salinity, meteorological parameters, and irrigation water conductivity (EC) collected by the data acquisition module. iw ), groundwater level and groundwater conductivity (EC) gw), and the current crop growth stage and crop evapotranspiration water requirement (ET) obtained above. c The results included the degree of salt stress, the distribution of salt profile in the root zone, the dynamic target salt threshold, and the risk assessment of salt return.

[0133] The Action subunit is used to output a set of irrigation strategy correction coefficients generated by the agent based on the current state vector, including the adjustment coefficient of the irrigation trigger threshold, the scaling factor of the irrigation quota, and the correction coefficient of the rinsing water volume.

[0134] The adjustment coefficient for the irrigation trigger threshold is used to dynamically fine-tune the soil moisture or salinity threshold for initiating irrigation. When the agent senses an increased risk of salinization (such as rising groundwater levels or increased evaporation), it outputs a positive adjustment coefficient for the irrigation trigger threshold to trigger the irrigation threshold earlier, thus suppressing salinity accumulation on the surface caused by capillary water rise by keeping the topsoil moist. Conversely, when rainfall is forecast, a negative adjustment coefficient for the irrigation trigger threshold is output to delay irrigation and avoid water waste.

[0135] The scaling factor for irrigation quotas is a dynamically adjusted parameter generated by a reinforcement learning agent. It adjusts the total irrigation water volume (i.e., crop evapotranspiration water requirement ET) calculated based on the fundamental formula, according to real-time, multi-dimensional environmental conditions. c The scaling factor is adjusted intelligently and holistically (to be combined with the sum of the rinsing water volume). This scaling factor is not calculated based on a fixed "if-then" rule, but is directly output by the trained agent based on the current real-time state vector.

[0136] Specifically, the decision-making model within the intelligent body (i.e., the optimized neural network strategy) has learned to map complex state vector features to optimal scaling factor values. For example, this decision-making model can identify the state vector representing "deteriorating irrigation water quality" (i.e., EC). iw The system detects patterns where the value increases significantly and outputs a scaling factor greater than 1.0 (e.g., 1.1) to increase the total irrigation volume and enhance the leaching effect. Alternatively, when the system identifies a forecast that the state vector contains information indicating future effective rainfall, it outputs a scaling factor less than 1.0 to proactively reduce the current irrigation volume, thereby saving water. This mechanism allows the adjustment of the total irrigation volume to transcend static formulas and achieve a comprehensive and intelligent response to multi-dimensional real-time information.

[0137] The correction coefficient for leaching water volume is used to correct the calculated theoretical leaching water volume, which is a fine adjustment of the "salt control intensity". Even if the crop is in a growth stage with strong salt tolerance (theoretical leaching demand is low), if the agent senses a continuous upward trend in groundwater salinity from the state vector, it will output a positive correction coefficient for leaching water volume and apply a preventive leaching increment in advance. This kind of "predictive" regulation cannot be achieved by traditional static formulas.

[0138] The reward function subunit is used to guide the agent to learn policies through the reward function. The reward function will give positive rewards to behaviors that can increase expected crop yield, improve water use efficiency, and effectively control root zone salinity within the target threshold, and give negative rewards (penalties) to behaviors that lead to high risk of salt return, deterioration of soil health, or excessive water and electricity costs.

[0139] In a specific embodiment, at each decision time step t, after the agent performs an action, the reward function calculation subunit calculates an immediate reward R. t R t It is a function consisting of a weighted sum of multiple sub-items, expressed as:

[0140] R t = w y ×R yield + w s ×R salt + w r ×R risk + w c ×R cost ;

[0141] In the formula, w y w s w r and w c These are weighting coefficients representing production, salinity control, risk of salt return, and cost, respectively. All these coefficients are positive numbers, and their relative magnitudes determine the degree of importance the agent places on different objectives during decision-making (for example, in areas with extremely scarce water resources, the weight of the cost item can be increased). c ).

[0142] By multiplying each of the above sub-rewards by its corresponding weighting coefficient and summing the results, a comprehensive reward signal R can be obtained that can fully and quantitatively evaluate the merits of the current decision. t The ultimate learning goal of the agent is to find a cumulative total reward ∑R that it can obtain throughout the growing season. t The optimal strategy for maximizing.

[0143] The training and optimization subunit executes a training process using reinforcement learning algorithms (such as PPO, DDPG, or SAC), driving the agent to repeatedly interact and learn in a simulation environment.

[0144] Specifically, interactive learning refers to the process by which an agent observes its current state, performs an action, and receives a reward and a new state from the environment. In each interactive learning session, the agent continuously adjusts its internal neural network parameters based on the accumulated reward signals, using optimization methods such as gradient descent, to find a decision "policy" that maximizes long-term cumulative rewards. The training process terminates when the policy performance converges (e.g., the average cumulative reward no longer shows significant improvement over multiple training rounds).

[0145] The strategy output and decision feedback subunit is used to deploy the optimal strategy learned by the agent and output the optimization results. The optimization results are not direct executable instructions, but rather "key strategy control parameters" (i.e., the aforementioned correction and scaling coefficients) generated by the agent model itself for the current state. In practical applications, the strategy output and decision feedback subunit transmits the key strategy control parameters to the decision module, which combines this with weather forecast information to generate the final irrigation strategy and converts it into executable instructions sent to the execution module. It can also generate suggestions for auxiliary measures. Simultaneously, the strategy output and decision feedback subunit is responsible for receiving feedback data on the effectiveness of the irrigation strategy (such as actual crop growth and real changes in soil salinity monitored by the data acquisition module) and transmitting it to the reinforcement learning optimization unit for calibrating the simulation environment.

[0146] Specifically, the calibration process compares the actual execution results with the simulated output of the simulation environment under the same input. If there is a significant deviation, the key parameters in the simulation model (such as HYDRDUS) are adjusted, including soil hydraulic function parameters and solute transport parameters, as well as crop physiological parameters in the crop growth model, in order to reduce the "reality gap". This improves the fidelity of subsequent training, enables the intelligent agent to adapt to long-term changes in the field, and achieves self-iteration and continuous optimization of the strategy.

[0147] The decision module integrates the complete calculation and evaluation results output by the model and evaluation modules, applies reinforcement learning to optimize the irrigation strategy correction coefficients generated by the optimization unit, and finally generates a system that controls the actual salinity within the target salinity threshold (EC). e_target The following are recommendations for adaptive irrigation strategies and support measures that meet crop water requirements while taking into account risk factors.

[0148] Irrigation strategies include irrigation timing, irrigation quotas, and irrigation frequency. Irrigation timing refers to the conditions that trigger irrigation. It can be based on soil moisture monitoring values, such as initiating irrigation when the root zone average soil moisture content drops below a preset managed allowable deficit (MAD) threshold, or it can be based on water balance predictions, such as based on predicted future ET (Earnings Tolerance). c Based on the current soil moisture content, irrigation should be scheduled in advance to prevent stress. The irrigation quota is the amount of water for a single irrigation, which should be sufficient to supplement the actual evapotranspiration requirement (ET) of the crop calculated by the crop model unit between two irrigations. c It also provides the adaptive rinsing water volume (V) calculated by the water-salt dynamic balance and risk assessment unit. leach The calculation method is expressed as: AW = ET c / (1-LR), where AW is the irrigation quota and LR is the minimum leaching fraction.

[0149] Subsequently, the decision-making module uses this irrigation quota as a benchmark and enters a dynamic correction process guided by a reinforcement learning agent.

[0150] Specifically, the decision-making module inputs the current complete environmental state vector into a trained reinforcement learning agent model, obtaining a set of optimal policy correction coefficients (such as scaling factors for irrigation quotas) output by the model. The decision-making module applies these coefficients to fine-tune the irrigation quotas and, combined with meteorological forecast data (such as deducting future effective rainfall), forms the final net irrigation amount. Simultaneously, the decision-making module also optimizes the irrigation pattern based on the salt return risk level and the agent's suggestions (e.g., favoring small, frequent irrigations when there is a high risk of salt return).

[0151] In one specific embodiment, the decision-making process of the decision-making module is as follows:

[0152] First, based on the ET output of the crop model unit c The irrigation quota is determined to be 7.5 mm based on the adaptive rinsing water volume (e.g., 1.5 mm) output by the water-salt dynamic balance and risk assessment unit (6.0 mm). Subsequently, the decision module inputs the current state, including a "high-risk" warning, into the agent model of the reinforcement learning optimization unit. The agent, based on its learned strategy, outputs an "irrigation quota scaling factor" of 1.20 for this state. The decision module applies this factor to increase the irrigation quota from 7.5 mm to 9.0 mm (7.5 mm × 1.20) to enhance the salt suppression effect. Finally, if the weather forecast indicates 20 mm of rainfall (effective utilization coefficient 0.7), 14 mm of effective rainfall is deducted from the planned irrigation amount. Since the required net irrigation amount after deduction is less than zero, the decision module ultimately decides to cancel this irrigation and update the triggering conditions for the next irrigation.

[0153] In a specific embodiment, when the risk assessment result of salt return output by the model and the evaluation module is "high risk", and the model prediction shows that even with the maximum irrigation and leaching intensity, the surface soil salinity cannot be maintained below the target salinity threshold, the decision module outputs auxiliary measures suggestions, such as taking surface mulching measures during the growing season to reduce soil surface evaporation and inhibit salt accumulation on the surface.

[0154] Specifically, land cover measures include plastic mulch, straw cover / residue mulch, gravel or sand mulch, planting cover crops, and applying soil evaporation inhibitors. Plastic mulch effectively prevents soil moisture evaporation and also has a warming effect; straw mulch uses crop straw or other organic residues to cover the soil surface, significantly reducing evaporation and increasing soil organic matter and improving soil structure; gravel or sand mulch allows for the use of inorganic mulch in specific situations; planting cover crops involves planting specific plants between the rows of the main crop or during fallow periods, reducing evaporation and improving the soil through their canopy shading and root activity; applying soil evaporation inhibitors involves spraying chemical agents to form a film on the soil surface to reduce evaporation.

[0155] As shown in Table 3, the table covering measures and their main role in the management of saline-alkali land are illustrated by example.

[0156] Table 3. Coverage Measures and Their Effects

[0157]

[0158] At the end of the growing season, the decision-making module, based on the growing season end-of-season assessment report generated by the model and evaluation module, conducts a comprehensive diagnosis of the effectiveness of water and salt management, including: the total irrigation water volume and total water consumption (ET) for the entire growing season. c The sum of the total rainfall and the total precipitation are compared to assess water resource utilization efficiency; salt leaching efficiency is calculated, i.e., the relationship between the amount of salt leached and the amount of additional water used for leaching is compared; and the salt stress state (i.e., K) of crops throughout the growing season is statistically analyzed. s The coefficient (less than 1) and the total duration under high salt return risk conditions are used as quantitative indicators of management effectiveness.

[0159] In a specific embodiment, when determining whether soil alkalization or solanization exists, the monitored soil pH value is compared with a standard threshold (e.g., 8.5). If the average value during a key growth stage is higher than this threshold, or if multiple consecutive measurements consistently exceed this threshold, an alkalization risk is considered. Simultaneously, the sodium adsorption ratio (SAR) is calculated; if its value exceeds a standard threshold (e.g., 13), a solanization risk is considered. Based on this comprehensive analysis, the decision-making module generates an assessment result, such as: "Salinity control was basically up to standard this growing season, but there is a slight solanization trend, and water use efficiency needs improvement." Based on this assessment result, the decision-making module generates targeted auxiliary measures recommendations, suggesting soil improvement measures before the start of the next planting season, such as applying gypsum to address the solanization trend, combined with deep loosening to improve soil structure.

[0160] In one specific embodiment, soil improvement measures include engineering leaching, chemical amendments, application of organic materials, physical amendments, vegetative amendments, and combined measures. Engineering leaching refers to the use of large-volume flooding or prolonged soaking in saline / fresh water to forcibly wash away large amounts of salt from the soil profile, provided that good drainage conditions (such as the presence of underground pipe / ditch systems) are available for soils with severe salt accumulation. Chemical amendments are mainly targeted at alkalized or sodium-containing soils (high pH, ​​high SAR). The application of calcium-containing substances such as gypsum (CaSO4·2H2O), phosphogypsum, and desulfurized gypsum utilizes calcium ions to replace sodium ions on soil colloids, which are then removed through leaching, simultaneously improving soil structure.

[0161] For soils with excessively high pH, ​​sulfur or acidic fertilizers can be applied to lower the pH. Adding organic materials, such as organic fertilizers, compost, biochar, green manure, and peat, to the soil can improve its physical structure (increasing aggregate stability and porosity), enhance water and fertilizer retention capacity, buffer the direct impact of salt on crops, and provide a substrate for soil microbial activity.

[0162] Physical soil improvement uses mechanical methods to enhance soil physical properties. For example, deep tillage can break up compacted plow pans or hard crusts, increasing the permeability of deeper soil layers and facilitating water infiltration and salt leaching. Other methods, such as mulching, sand mixing (to improve heavy clay soils), and topsoil improvement (introducing high-quality soils), can also be used under specific conditions.

[0163] Plant improvement involves planting specific salt-tolerant or salt-absorbing plants (halophytes) that absorb and accumulate salt in the soil during their growth process. The plants are then removed, achieving the goal of "biological desalination".

[0164] Combination measures refer to the practice of combining two or more of the above measures, depending on the specific type and severity of the soil problem, to achieve the best improvement effect. For example, applying gypsum to improve alkaline soil and then irrigating and leaching; or combining the application of organic fertilizer with deep loosening.

[0165] Table 4 illustrates, exemplarily, soil improvement measures and their mechanisms of action in addressing saline-alkali soil problems. It can be seen that the long-term soil health management recommendations implemented in this scheme surpass the capabilities of simple seasonal irrigation control.

[0166] Table 4 Soil improvement measures and their mechanisms of action

[0167]

[0168] The execution module translates the executable instructions generated by the decision-making module into specific irrigation and drainage actions, while providing real-time feedback on the execution status. Specifically, this feedback path is bidirectional: the real-time operational status, which ensures accurate execution of instructions, is fed back to the decision-making module, while the process effect data, used for model calibration and system optimization, is fed back to the model and evaluation module.

[0169] The execution module includes an irrigation execution unit, a drainage execution unit, and a feedback monitoring unit.

[0170] After receiving executable instructions from the decision-making module, the execution module, in collaboration with the irrigation and drainage execution units, implements irrigation and drainage through pumps, pipe networks, irrigation devices, or valves to maintain the dynamic balance of soil water and salt. The feedback monitoring unit monitors the execution status of irrigation and drainage in real time and transmits different feedback information to the corresponding upper-level modules.

[0171] The irrigation execution unit includes a water source and power system, a water distribution network, field irrigators, control valves, and an irrigation controller. The irrigation controller controls the irrigation actions. The water source and power system includes a water pump, a filter, and a fertilization device. The water pump is a variable frequency pump, which precisely controls the water flow and pressure by adjusting the frequency. The fertilization device is used to meet the needs of integrated water and fertilizer management. The water distribution network includes main pipes, branch pipes, and valves and fittings connecting each level of pipe. Field irrigators are selected according to crop type and planting pattern, such as drip irrigation tape / driers suitable for row-planted crops, which have high water utilization efficiency and can precisely control the amount of water delivered to the root zone, or micro-sprinklers or small sprinkler heads suitable for crops requiring a larger humidified area or seedlings. Solenoid valves or electric valves are installed at the inlet of the branch pipes or irrigation execution unit as control valves, and their opening and closing are remotely controlled by the irrigation controller to achieve independent and precise control of different irrigation areas. The irrigation controller can be a standalone controller or a control module integrated into a cloud platform. It receives executable instructions from the decision module and precisely controls the start, stop, and speed of the water pump, as well as the opening and closing of valves, to ensure that the predetermined amount of water is applied to the designated area at the correct time. Water volume control can be achieved by controlling the irrigation duration (based on known system flow rate), directly metering the flow rate (using a flow meter), or setting a target volume.

[0172] Based on a collaborative mechanism with the irrigation execution unit, the drainage execution unit promptly collects and discharges high-salinity leaching water generated during irrigation through underground pipes or surface channels. The drainage measures implemented by the drainage execution unit reduce the risk of waterlogging or salt resurgence in the soil, effectively regulate the soil's salinity balance, and ensure the long-term maintenance of the irrigation leaching effect.

[0173] The feedback monitoring unit, by deploying flow meters, pressure sensors, valve status sensors, and pump operation status sensors, monitors the execution status of irrigation and drainage in real time to ensure the accurate execution of irrigation strategies and promptly correct deviations. When the execution status (such as actual flow rate) deviates from the command requirements, the unit feeds back the abnormal information to the decision-making module to trigger automatic adjustments or alarms, ensuring the reliability of irrigation execution. Simultaneously, the unit also summarizes key process data, such as the total water application volume, average flow rate, and irrigation duration of a single irrigation event, and feeds it back to the model and evaluation module along with timestamps, serving as the basis for calibrating the simulation environment and optimizing the model.

[0174] Through clear division of responsibilities and collaborative operation, the various units of the execution module translate the irrigation strategies from the decision-making module into actual field actions. Efficient drainage and real-time monitoring feedback distributed to different upper-level modules form a complete closed loop supporting immediate control and long-term optimization. This ensures the accurate implementation of irrigation strategies, enhances the adaptability and reliability of the intelligent irrigation control system under complex environmental conditions, and provides strong support for precision irrigation and dynamic soil water-salt balance management in saline-alkali land.

[0175] like Figure 2 As shown, the present invention provides an intelligent irrigation regulation method for dynamic water-salt balance in saline-alkali land, comprising the following steps:

[0176] S1, Data Acquisition

[0177] Sensors are used in the monitoring area to collect dynamic monitoring data affecting water and salt dynamics in real time, and the data is saved to a basic database for subsequent steps. The dynamic monitoring data includes meteorological parameters, soil parameters, groundwater and irrigation water parameters, and weather forecast data.

[0178] Meteorological parameters include solar radiation, air temperature, air humidity, wind speed, and rainfall; soil parameters include soil moisture, soil salinity, soil temperature, and soil pH; groundwater and irrigation water parameters include groundwater level and groundwater electrical conductivity (EC). gw ) and irrigation water conductivity (EC) iw Weather forecast data includes rainfall, temperature, and wind speed for the next three days.

[0179] The basic database also includes pre-built crop data and a localized parameter library as the basis for model calculations. The crop data includes the criteria for classifying each growth stage and the corresponding crop coefficients (K). c Crop salt tolerance thresholds, including salt tolerance thresholds (EC50), are also considered. e_threshold ) and limit salinity value (EC) e_100% The localized parameter library includes soil hydraulic function parameters describing soil water holding and conductivity properties, and solute transport parameters describing the diffusion and migration characteristics of salts in the soil.

[0180] In one specific embodiment, an automatic weather station is deployed in the monitoring area, and meteorological parameters are collected at a regular frequency of once per hour using a built-in sensor array. At least three monitoring points are set up in the monitoring area based on soil homogeneity, and at each monitoring point, a multi-depth probe or sensor array covering the crop root activity layer (e.g., 0 cm - 60 cm) is vertically installed to collect soil parameters at a regular frequency of once every 30 minutes. Specifically, sensors such as time domain reflectance (TDR) can be used to simultaneously measure soil moisture, salinity, and temperature, and a soil pH electrode probe can be used to measure pH values.

[0181] The layout of monitoring points is determined based on the size of the monitoring area, the spatial heterogeneity of the soil, and typical landform features. A grid method can be used to distribute the points evenly, or targeted points can be distributed in key locations (such as highlands and depressions) to ensure that the collected soil parameters can represent the average condition and spatial variability of the entire area.

[0182] Online water quality EC sensors are deployed at irrigation water sources or main pipelines to trigger real-time acquisition of irrigation water conductivity during irrigation. At least one groundwater monitoring well is installed within the monitoring area, equipped with an automatic water level gauge and a water conductivity sensor, to collect groundwater parameters at a regular frequency of once every 6 hours. Weather forecast data is automatically and periodically acquired by a backend service deployed on a cloud platform or local server by calling the application programming interface (API) of a third-party weather service provider.

[0183] In one specific embodiment, to ensure effective capture of changes in environmental and soil conditions during critical events, an event-triggered dynamic acquisition frequency adjustment mechanism is employed. When the following events occur, the sensor's acquisition frequency is increased from a normal frequency to a higher frequency mode, for example, from once every 30 minutes to once every 5 minutes:

[0184] Irrigation Events: When the system execution module starts or stops an irrigation task, it will automatically trigger the soil and weather sensors in the associated area to enter high-frequency acquisition mode, and return to normal frequency after a preset time after the irrigation ends.

[0185] Weather events: When the rain gauge of the automatic weather station detects that the rainfall exceeds the micro-threshold (such as 0.1 mm / h), or when the rate of change of meteorological parameters such as temperature and solar radiation exceeds the preset threshold, the system will automatically switch to high-frequency acquisition mode until the weather event ends.

[0186] In this embodiment, low-power wide-area IoT technology is used for wireless data transmission. The sensor sends the collected data to a LoRaWAN gateway deployed in or near the detection area via LoRa wireless signal. The gateway then uploads the data to the basic database via cellular network (4G / 5G), Ethernet, or WiFi for use by the model and evaluation module and the decision-making module.

[0187] S2, Risk Assessment of Salt Reversion

[0188] Based on data from the basic database, crop model units are invoked to calculate the actual evapotranspiration water requirement (ET) of crops under salt stress. c Secondly, the water-salt dynamic balance and risk assessment unit is invoked, based on ET. cUsing data from the basic database, the system completes the estimation of salinity profiles, setting of dynamic target salinity thresholds, calculation of adaptive leaching water volume, assessment of salt return risk, and generation of end-of-growing-season assessment reports. Finally, it calls the reinforcement learning optimization unit to generate and optimize key strategy control parameters for guiding irrigation.

[0189] Specifically, this involves calculating the actual evapotranspiration water requirement (ET) of crops. c The process is as follows: Based on real-time meteorological data retrieved from the basic database, the standard reference crop evapotranspiration (ET0) is calculated using the FAO Penman-Monteith formula, and the crop coefficient (K) corresponding to the current crop growth stage is queried from the basic database. c The basic crop water requirement was obtained. To accurately reflect the saline-alkali land condition, the real-time monitored root zone soil salinity retrieved from the basic database was compared with the crop salt tolerance threshold in the basic database. A salt stress coefficient (K) was calculated using a salt stress model based on the FAO-56 theory. s This is used to correct the baseline water requirement, ultimately outputting ET, which represents the actual water demand of the crop at present. c The calculation formula is as follows:

[0190] ET c_base =K c ×ET0;

[0191] ET c =K s ×ET c_base ;

[0192] The crop coefficient reflects the differences between a specific crop and a reference crop at a specific growth stage in terms of canopy coverage, stomatal conductance, and canopy height. For example, the K value at the peak boll-forming stage of cotton. c Approximately 1.10-1.15, mid-stage K in tomatoes c It is approximately 1.10.

[0193] Subsequently, the water-salt dynamic balance and risk assessment unit is invoked to estimate salinity profile, set dynamic target salinity threshold, calculate adaptive rinsing water volume, assess water-salt migration and salt return risk, and generate an end-of-growing-season assessment report.

[0194] Specifically, the process of salinity profile estimation is as follows: using soil salinity data retrieved from a basic database, a complete or arbitrary range of salinity spatial distribution map of the monitoring area is generated through spatial interpolation techniques such as Kriging interpolation or inverse distance weighting. Then, a vertical profile function is constructed using soil parameters collected from each monitoring point in the monitoring area to track the vertical distribution and migration patterns of soil salinity (e.g., the downward shift of salinity peaks after irrigation), thereby achieving salinity profile estimation and understanding the hotspots and depth of soil salinity accumulation.

[0195] The process of setting dynamic target salinity thresholds is as follows: Crop baseline data is retrieved from the basic database, and target salinity thresholds are set according to crop type and growth stage. The crop baseline data contains data on the tolerance thresholds of various crops to soil salinity at each growth stage.

[0196] In one specific embodiment, the soil salinity tolerance threshold for cotton, a moderately salt-tolerant crop, is approximately 7.7 dS / m. During the sensitive seedling stage, to ensure cotton emergence and robust seedling growth, a target salinity threshold of 4.0 dS / m is set. During the flowering and boll-setting stage, cotton's tolerance to soil salinity increases, and the target salinity threshold is raised to 8.0 dS / m, achieving water conservation through leaching while maintaining yield. The soil salinity tolerance threshold for tomato, a moderately sensitive crop, is approximately 2.5 dS / m. Since moderate salt stress can improve tomato fruit quality (such as sugar content and flavor compounds), a target salinity threshold of 6.0 dS / m is set during the flowering and fruit-setting stage. The target salinity threshold is dynamically adjusted according to production targets and the plant's growth stage to balance yield and quality.

[0197] Specifically, the process of adaptive rinsing water volume calculation is as follows:

[0198] Determine the minimum leaching fraction (LR). This is a dimensionless ratio characterizing the leaching intensity. The system calculates the LR value using a simplified steady-state salt equilibrium formula:

[0199] LR = EC iw / (k × EC e_target );

[0200] Among them, EC iw It is the electrical conductivity of irrigation water, EC e_target is the current target salinity threshold, while k is the leaching efficiency coefficient, representing the effectiveness of the leaching water in replacing the salt in the soil solution.

[0201] Calculate the adaptive leaching water volume. After obtaining the LR, the system combines it with the crop evapotranspiration water requirement (ET) calculated by the crop model unit. c ), through the conversion formula V leach = LR×ET c / (1-LR) is used to determine the adaptive rinse water volume that needs to be replenished.

[0202] Specifically, the value of k is usually less than 1. Due to the influence of various factors such as soil texture, structure, pore connectivity, irrigation method (such as the wetting front morphology of drip irrigation) and drainage conditions, k = 0.7 in this embodiment.

[0203] The risk assessment of salt return is based on real-time monitoring data of groundwater level depth and groundwater conductivity (EC) retrieved from a basic database.gw The calculated daily reference evapotranspiration (ET0) is used to classify risk levels according to preset quantification and correction rules, ensuring a comprehensive and accurate assessment of risk. The specific process is as follows:

[0204] First, a basic risk level is determined based on groundwater level and salinity, and then based on real-time monitoring of groundwater level and groundwater conductivity (EC). gw The combination of parameters, according to a preset risk matrix (as shown in Table 2) or equivalent logical rules, determines the basic risk level. As shown in Table 5, this matrix covers all parameter combinations, avoiding the logical loopholes of traditional methods.

[0205] Table 5 Preset Risk Matrix

[0206]

[0207] When the groundwater level is greater than 2m and the groundwater conductivity EC gw Not greater than 6 dS / m, or the groundwater level is between 1m and 2m and the groundwater conductivity EC gw When the value is less than 3 dS / m, the basic risk level is low.

[0208] When the groundwater level is greater than 2m and the groundwater conductivity EC gw Greater than 6 dS / m, or groundwater level between 1m and 2m and groundwater conductivity EC gw Between 3 dS / m and 6 dS / m, or when the groundwater level is less than 1m and the groundwater conductivity EC gw When the value is less than 3 dS / m, the basic risk level is medium risk.

[0209] When the groundwater level is between 1m and 2m and the groundwater conductivity ECgw is greater than 6 dS / m, or when the groundwater level is less than 1m and the groundwater conductivity ECgw is greater than 6 dS / m, gw When the value is not less than 3 dS / m, the basic risk level is high risk.

[0210] Secondly, after determining the basic risk level, the evaporation-driven effect, which is the core driving force for the upward migration of salt, is further considered. Risk is adjusted based on evapotranspiration. If the daily reference evapotranspiration ET0 is greater than 6 mm / day, the basic risk level is raised by one level, for example, from "low risk" to "medium risk," or from "medium risk" to "high risk"; if the basic risk level is already "high risk," it remains unchanged.

[0211] In a specific embodiment, in arid and semi-arid regions where the groundwater level is shallow and the mineralization is high, soil moisture evaporation can cause groundwater to rise to the surface or crop root zone through capillary action, resulting in a phenomenon known as "salinization" where surface soil salts accumulate. The water-salt dynamics and risk assessment unit uses the aforementioned assessment methods to comprehensively monitor regional groundwater depth and EC (excessive eosinophilic acid). gw The risk level of salt return is classified by using reference crop evapotranspiration (ET0) calculated from meteorological parameters, thus enabling accurate assessment of salt return risk.

[0212] The process of generating the growing season end-of-season assessment report includes: after the growing season ends, based on the dynamic monitoring data stored in the basic database throughout the growing season and the irrigation execution process data fed back by the execution module, water and salt balance calculations are performed, and a growing season end-of-season assessment report is generated, reflecting the soil salinity accumulation status. The growing season end-of-season assessment report includes the total salt input during the growing season, salt output (crop absorption and leaching), changes in root zone salt storage, the evolution trend of salt profile distribution, and the degree of salt accumulation or leaching compared to the beginning of the season (such as changes in net salt content in kg / ha), etc.

[0213] The reinforcement learning optimization unit combines real-time dynamic monitoring data retrieved from the basic database with calculation and evaluation results output by the crop model unit and the water-salt dynamic balance and risk assessment unit to form an environmental state vector. Then, a pre-trained agent within the reinforcement learning optimization unit receives the current state vector and, guided by a multi-objective reward function balancing crop yield, water use efficiency, salinity control effectiveness, and irrigation and drainage costs, outputs a set of optimal key strategy control parameters. These parameters serve as dynamic correction coefficients for the irrigation strategy and guide downstream decision-making modules. The agent learns the optimal strategy through continuous interactive learning with the simulation environment. The interactive learning process is as follows: the agent observes the current state, selects and executes an action, the simulation environment updates its state based on this action and returns a reward signal, and the agent updates its strategy based on this feedback.

[0214] In one specific embodiment, the simulation environment is used to dynamically simulate crop growth and soil water and salt dynamic responses under different irrigation strategies. An agent-interactive simulation environment is constructed by coupling crop growth models (such as WOFOST or ORYZA) with water and salt transport models (such as HYDRUS or SWAP). Data required for decision-making is integrated as state input, including soil moisture, soil salinity, meteorological parameters, and irrigation water conductivity (EC) collected by the data acquisition module. iw ), groundwater level and groundwater conductivity (EC) gw ), and the current crop growth stage and crop evapotranspiration water requirement (ET) output by the model and evaluation module. cThe results included the degree of salt stress, the distribution of salt profile in the root zone, the dynamic target salt threshold, and the risk assessment of salt return.

[0215] The agent outputs a set of key policy control parameters as actions based on the current state. The key policy control parameters include the adjustment coefficient of the irrigation trigger threshold, the scaling factor of the irrigation quota, and the correction coefficient of the rinsing water volume.

[0216] The adjustment coefficient for the irrigation trigger threshold is used to dynamically fine-tune the soil moisture or salinity threshold for initiating irrigation, with the core objective of optimizing "when to irrigate." Specifically, when the agent senses an increased risk of salinization (such as rising groundwater levels or increased evaporation), it outputs a positive adjustment coefficient to trigger the irrigation threshold earlier, suppressing salinity accumulation on the surface caused by capillary water rise by keeping the topsoil moist. Conversely, when rainfall is forecast, a negative coefficient is output to delay irrigation and avoid water waste.

[0217] The scaling factor of the irrigation quota is used to scale the total water volume of the current irrigation event (i.e., crop evapotranspiration water requirement ET). c The core objective (combined with the amount of rinsing water) is to optimize "how much to irrigate at once". Specifically, when a sudden deterioration in irrigation water quality is detected, the agent outputs a scaling factor greater than 1.0 (e.g., 1.1) based on its learned strategy, adding an extra 10% of irrigation volume on top of the basic calculation to enhance the rinsing effect, as a dynamic safety redundancy.

[0218] The correction coefficient for leaching water volume is used to correct the calculated theoretical leaching water volume, which is a fine adjustment of the "salt control intensity". Specifically, even if the crop is in a growth stage with strong salt tolerance (theoretical leaching demand is low), if the agent senses from the state that the groundwater salinity is on the rise, it may learn to output a positive correction coefficient and apply a preventive leaching increment in advance. This kind of "predictive" regulation cannot be achieved by traditional static formulas.

[0219] Next, the agent's policy learning is guided by a preset reward function. This function gives positive rewards to behaviors that can increase expected crop yield, improve water use efficiency, and effectively control root zone salinity within the target threshold; at the same time, it gives negative rewards to behaviors that lead to high risk of salinization, deterioration of soil health, or excessive water and electricity costs.

[0220] In a specific embodiment, at each decision time step t, after the agent performs an action, the reward function calculation subunit calculates an immediate reward R. t R t It is a function consisting of a weighted sum of multiple sub-terms, and its overall form can be expressed as:

[0221] R t = wy ×R yield + w s ×R salt + w r ×R risk + w c ×R cost ;

[0222] In the formula, w y w s w r and w c These are weighting coefficients representing production, salinity control, risk of salt return, and cost, respectively. All these coefficients are positive numbers, and their relative magnitudes determine the degree of importance the agent places on different objectives during decision-making (for example, in areas with extremely scarce water resources, the weight of the cost item can be increased). c ).

[0223] By multiplying each of the above sub-rewards by its corresponding weighting coefficient and summing the results, the system obtains a comprehensive reward signal R that can fully and quantitatively evaluate the merits of the current decision. t The ultimate learning goal of the agent is to find a cumulative total reward ∑R that it can obtain throughout the entire growth season. t Maximize decision-making strategies.

[0224] Finally, reinforcement learning algorithms are used for training and optimization. This method employs reinforcement learning algorithms such as PPO, DDPG, or SAC to drive the agent to repeatedly perform the aforementioned interactive learning in a simulation environment. In each interaction, the agent continuously adjusts its internal neural network parameters based on the accumulated reward signals, using optimization methods such as gradient descent, in order to find a decision "policy" that maximizes long-term cumulative rewards. When the policy performance converges and stabilizes, reaching the preset optimization objective, the training process terminates and the optimization result is output. Simultaneously, the optimization result is transmitted to the decision-making module.

[0225] S3, Irrigation Strategy Generation and Execution

[0226] First, the decision-making module integrates the results output from the model and evaluation module in step S2, including the outputs from the crop model unit and the water-salt dynamic balance and risk assessment unit (crop evapotranspiration water requirement, salt profile estimation results, dynamic target salt threshold, adaptive leaching water calculation results, salt return risk assessment results, and end-of-growing-season assessment report), as well as key strategy control parameters generated by the reinforcement learning optimization unit. The key strategy control parameters include the adjustment coefficient for the irrigation trigger threshold, the scaling factor for the irrigation quota, and the correction coefficient for the leaching water volume.

[0227] Subsequently, the decision-making module executes a multi-level, dynamically revised quantitative decision-making process: First, it determines the irrigation timing: The decision-making module applies an adjustment coefficient to the irrigation trigger threshold to dynamically fine-tune the preset Management Allowable Deficit (MAD) threshold, thereby more intelligently determining the specific time to initiate irrigation. Second, it determines the irrigation quota: This process is based on the basic water demand parameter ET. c and V leach First, apply the correction factor for the rinsing water volume to V. leach Make fine adjustments, and then combine the adjusted rinse water volume with ET. c The initial total irrigation amount is obtained by summing the results. Then, the scaling factor of the irrigation quota is applied to scale the initial total amount to account for some complex and non-linear risk factors. Finally, the future effective rainfall is calculated based on the meteorological forecast data retrieved from the basic database, and this amount is deducted from the revised irrigation quota to obtain the final net irrigation amount.

[0228] After a series of progressively quantitative decisions, a highly adaptive irrigation strategy is generated, including optimized irrigation timing, net irrigation volume, and irrigation frequency. Simultaneously, if the risk level output by S2 is too high, the decision module will also generate recommendations for auxiliary measures during the growing season.

[0229] For example, when the risk assessment of salt return is "high risk," and the model predicts that even with maximum irrigation and leaching intensity, salt control cannot be effectively achieved, the decision-making module will recommend implementing surface mulching measures (such as straw mulching) during the growing season to reduce soil evaporation and inhibit salt accumulation. Furthermore, after the growing season, the decision-making module will generate longer-term soil improvement recommendations based on the end-of-season assessment report and its detailed diagnostic analysis (such as applying gypsum to improve alkaline soils, increasing the application of organic fertilizers to improve soil structure, or combining deep tillage with engineering leaching), to achieve sustainable improvement in land productivity.

[0230] The final irrigation strategy is translated into executable instructions and sent to the execution module. The execution module precisely completes irrigation and drainage tasks by coordinating the control of irrigation units (such as pumps and valves) and drainage units (such as underground pipes). During this process, the feedback monitoring unit monitors the operating status (such as flow rate and pressure) and process effects of the irrigation system in real time, distributing feedback data to the corresponding upper-level modules: it monitors the execution status of irrigation and drainage in real time, feeding back execution status data used for real-time control (such as abnormal valve opening / closing or flow rate deviation from setpoints) to the decision module to trigger immediate adjustments or alarms to irrigation parameters; simultaneously, it acquires execution process data including irrigation volume and duration, and obtains soil salinity change data after irrigation as process effect data. The execution process data characterizing the task's execution, along with the process effect data monitored by the data acquisition module, is fed back to the model and evaluation module to continuously calibrate the model parameters of the simulation environment, and the reinforcement learning optimization unit optimizes the next irrigation strategy. This feedback data will serve as part of the environmental state for the next decision cycle, thus forming a complete intelligent control closed loop that supports both immediate control and long-term optimization.

[0231] In a specific embodiment, taking a saline-alkali land in northern China where cotton is grown in loam soil as the monitoring area, the implementation process of the intelligent irrigation control method for the dynamic balance of water and salt in saline-alkali land based on the intelligent irrigation control system for dynamic balance of water and salt in saline-alkali land of the present invention is as follows:

[0232] Data collection:

[0233] After the intelligent irrigation control system is activated, sensors deployed in the monitoring area continuously collect dynamic monitoring data: automatic weather stations record temperature, humidity, wind speed, radiation, and rainfall data hourly; multi-depth soil sensors (including 0 cm - 20 cm, 20 cm - 40 cm, and 40 cm - 60 cm) distributed at monitoring points measure soil moisture, soil salinity, soil temperature, and soil pH every 30 minutes. Water quality EC sensors at the water source monitor the irrigation water conductivity (EC) in real time before or during each irrigation. iw The water level gauge and EC sensor in the monitoring well record the groundwater level depth and groundwater conductivity (EC) every 6 hours. gw All collected data is transmitted in real time to the cloud platform data server via the LoRaWAN wireless network.

[0234] Status assessment and requirements calculation:

[0235] Before cotton sowing, for example in early April, the intelligent irrigation control system uses sensors to conduct a comprehensive initial soil condition assessment of the monitored area, or retrieves assessment reports and historical data from the end of the previous season. Assuming the assessment results show that the average electrical conductivity (EC) of the saturated soil extract in the 0 cm – 60 cm soil layer is... e , by EC a The converted value (approximately 5.0 dS / m) indicates a moderate salinization level. The system queries the built-in crop database to obtain information on the salt tolerance characteristics of cotton, including its physiological salt tolerance threshold EC. e_threshold (e.g., 7.7 dS / m) and limiting salinity EC e_100% (e.g., 20.1 dS / m). Considering the upcoming cotton seedling stage, which is relatively sensitive to salt, the intelligent irrigation control system dynamically sets the target salt threshold EC in the root zone for the current growth stage (seedling stage) to ensure uniform emergence and robust seedlings. e_target = 4.0 dS / m.

[0236] Taking a day in the cotton seedling stage as an example, we will conduct a status assessment and demand calculation:

[0237] (1) Crop water requirement and stress assessment: Using real-time collected meteorological parameters, the daily reference crop evapotranspiration ET0 = 3.5 mm / day was calculated using the FAO Penman-Monteith formula. Based on the current seedling stage of cotton, the corresponding crop coefficient K was queried. c ≈ 0.5. Calculate the actual water demand ET for cotton evaporation and transpiration on that day. c_base = K c × ET0 = 0.5 × 3.5 = 1.75 mm / day. Simultaneously, the system monitors the actual EC in the current root region (weighted average of 0 cm – 60 cm). e The value is 4.5 dS / m, which does not exceed the physiological salt tolerance threshold of cotton (7.7 dS / m). Therefore, the calculated salt stress coefficient K is... s The value is 1.0. Ultimately, the actual daily crop evapotranspiration water requirement ET is... c (i.e., K) s × ET c_base The actual salinity (4.5 dS / m) remained at 1.75 mm / day. Although no physiological stress occurred, the actual salinity (4.5 dS / m) was already higher than the management target (4.0 dS / m), indicating a risk of salt accumulation that needs to be managed through leaching.

[0238] (2) Adaptive rinsing water volume calculation: The system reads the real-time conductivity EC of the current irrigation water source. iw = 1.8 dS / m. Using the formula LR = EC iw / (k × ECe_target The leaching efficiency was set to k = 0.7, and the minimum leaching fraction was calculated: LR = 1.8 / (0.7 × 4.0) = 1.8 / 2.8 ≈ 0.64. This indicates that theoretically, a leaching fraction LR ≈ 0.64 (i.e., 64%) is needed to maintain the root zone salinity at the target level of 4.0 dS / m under the current irrigation water quality. The minimum leaching fraction is then converted into the specific leaching water volume (V) using the following formula. leach ):V leach = LR×ET c / (1-LR), substitute the value for calculation: V leach = (0.64 × 1.75) / (1 - 0.64) ≈ 3.11 mm. Therefore, in order to maintain the salinity of the root zone at the target level while meeting the water consumption of crop growth, the adaptive leaching water required on that day is approximately 3.11 mm. This also means that the total irrigation water requirement for that day (crop water requirement + leaching water requirement) is 1.75 mm + 3.11 mm = 4.86 mm.

[0239] (3) Risk assessment of salt return:

[0240] The system monitored a groundwater level depth of 1.6 m, between 1 m and 2 m; the groundwater conductivity EC gw The value is 4.0 dS / m, which is between 3 dS / m and 6 dS / m. According to the preset risk assessment scheme, the basic risk level corresponding to the combination of these two ranges is "medium risk".

[0241] Risk correction was performed based on evapotranspiration. The reference crop evapotranspiration ET0 calculated for the day was 3.5 mm / day, which did not exceed the correction threshold of 6 mm / day. Therefore, no correction was needed for the basic risk level, and the current salt return risk level was ultimately assessed as "medium risk".

[0242] Irrigation strategy generation and execution:

[0243] The decision-making module integrates all the above calculation results and real-time monitoring data to form a complete state vector. This state vector is then input into the trained reinforcement learning agent model to obtain the optimal policy correction coefficients.

[0244] Specifically, under the "medium-risk" condition where soil salinity is higher than the management target, to achieve more robust salinity control, the agent outputs an irrigation quota scaling factor of 1.05. The decision module applies this coefficient to adjust the irrigation quota to 4.86 mm × 1.05 ≈ 5.1 mm. Simultaneously, considering that the weather forecast indicates no effective rainfall for the next three days, the system generates an irrigation strategy that includes the irrigation timing (immediate execution), the irrigation quota (5.1 mm), and the estimated irrigation frequency, and converts this into executable instructions that are sent to the execution module.

[0245] The execution module starts the variable frequency water pump and solenoid valve of the corresponding irrigation zone, and precisely controls the irrigation volume through the flow meter. When the cumulative water volume reaches 5.1 mm, the valve and water pump are automatically shut off. During this process, the execution module feeds back the real-time operating status (such as abnormal pressure, flow deviation) to the decision module for real-time control and alarm, and feeds back the process data of this irrigation event (such as the total water volume of 5.1 mm, irrigation duration, etc.) to the model and evaluation module for long-term model calibration and optimization.

[0246] As cotton enters its peak growth period (e.g., flowering and boll-forming stage), crop water consumption increases, and it may also encounter prolonged periods of high temperatures and drought. At this time, groundwater levels were monitored to rise to 0.9 m due to seasonal factors, and EC... gw The evapotranspiration rate rose to 7 dS / m, and the daily reference evapotranspiration rate (ET0) also rose to 7.0 mm / day. The system restarted its risk assessment.

[0247] The groundwater level is 0.9 m < 1 m, and the groundwater conductivity is 7 dS / m > 6 dS / m. According to the risk assessment plan, the basic risk level is determined to be "high risk".

[0248] Risk adjustments were made based on evapotranspiration. Since the base risk level was already at the highest level, "high risk," even though ET0 of 7.0 mm / day exceeded the adjustment threshold, the risk level did not need to be increased further and remained "high risk." Therefore, the current risk assessment level for salt return is "high risk."

[0249] At this point, the reinforcement learning agent outputs a stronger irrigation quota scaling factor of 1.20 to execute repressive leaching irrigation. Simultaneously, since it is predicted that simply increasing water volume will not effectively control salt levels, the system generates auxiliary measures recommendations: "High risk of salt return detected... It is recommended to cover the cotton rows with straw..."

[0250] The execution module receives the instruction containing the irrigation quota (12.8 mm) and begins execution, starting the variable frequency water pump in the corresponding irrigation zone and opening the solenoid valve to begin drip irrigation at the set flow rate (achieved by adjusting the pump frequency). A flow meter installed on the pipeline monitors the cumulative irrigation water volume in real time. When the cumulative water volume reaches 12.8 mm, the valve and pump automatically shut off, completing the irrigation cycle. During irrigation, a pressure sensor continuously monitors the pipeline pressure to ensure it remains stable within the normal range.

[0251] The execution module transmits feedback data such as flow rate, pressure, and valve status back to the decision module in real time. When the actual flow rate is significantly lower than expected (possibly due to dripper blockage), the decision module will extend the irrigation time to compensate or issue a maintenance alarm. Leachate generated after irrigation is collected and discharged outside the field through an underground drainage system laid in the field. After irrigation, the data acquisition module continues to monitor soil water and salt dynamics, providing a basis for the next irrigation decision.

[0252] After the cotton harvest, the intelligent irrigation control system automatically compiles all monitoring data and management records from the entire growing season (approximately April to October). By running a water-salt balance model, it generates an annual soil salinity assessment report, indicating that intelligent irrigation control successfully reduced the average EC50 in the root zone (0 cm - 60 cm) during the growing season. e The cotton growth remained largely stable near the dynamic target threshold, indicating good growth. However, due to the use of EC... iw Irrigation was carried out with slightly saline water at a concentration of 1.8 dS / m, and there was some back salinity pressure. The average EC in the root zone at the end of the season was... e Compared to 5.0 dS / m at the beginning of the season, the concentration eventually stabilized at 5.5 dS / m, indicating a slight net salt accumulation, estimated at approximately 150 kg / ha. Meanwhile, monitoring data showed a slight increase in the SAR value of the soil surface layer, but it remained within a safe range.

[0253] Based on this assessment report, the decision-making module generated soil improvement recommendations for the following year: "Water and salt management was effective this year, but there is slight salt accumulation. To ensure long-term sustainable production, it is recommended to apply more organic materials to the soil before sowing next spring, in conjunction with land preparation, such as applying 15 tons / hectare of well-rotted organic fertilizer or 5 tons / hectare of biochar, to further improve soil structure, enhance buffering capacity, and increase leaching efficiency. If crops more sensitive to salt are planned for planting next year, an early spring irrigation leaching can be considered after applying organic fertilizer."

[0254] The invention and its embodiments have been described above illustratively. This description is not restrictive, and the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. The accompanying drawings are only one embodiment of the invention, and the actual structure is not limited thereto. No reference numerals in the claims should limit the scope of the claims. Therefore, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the invention, such design should fall within the scope of protection of this patent. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Multiple elements stated in the product claims may also be implemented by a single element through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A smart irrigation regulation method for dynamic water-salt balance in saline-alkali land, comprising the following steps: Step 1: Collect dynamic monitoring data affecting water and salt dynamics and save it to the basic database. The dynamic monitoring data includes meteorological parameters, soil parameters, groundwater and irrigation water parameters, and meteorological forecast data. Step 2: Calculate the crop evapotranspiration water requirement based on the data in the aforementioned basic database; Based on the data from the aforementioned basic database and crop evapotranspiration water requirements, the salinity profile was estimated. Based on the crop type and its growth stage, the dynamic target salinity threshold of the root zone soil is queried from the basic database and set. Based on the electrical conductivity of irrigation water, the set dynamic target salinity threshold, and the calculated crop evapotranspiration water requirement, the adaptive rinsing water volume is calculated. Based on the data in the aforementioned basic database, a risk assessment of salt return is conducted; A growing season end-of-season assessment report is generated based on water and salt balance accounting throughout the growing season, including changes in salt input and output and root zone salt storage. The data from the basic database, crop evapotranspiration water demand, salt profile estimation results, dynamic target salt threshold, adaptive leaching water volume calculation results, and salt return risk assessment results are integrated as an environmental state vector and input into the reinforcement learning model to generate and output key strategy control parameters for guiding irrigation. Step 3: Combine the crop evapotranspiration water demand, salt profile estimation results, dynamic target salt threshold, adaptive leaching water volume calculation results, salt return risk assessment results, and optimized key strategy control parameters, and integrate meteorological forecast data to generate an adaptive irrigation strategy. Step 4: Convert the irrigation strategy into executable instructions for irrigation and drainage.

2. The intelligent irrigation regulation method for dynamic water-salt balance in saline-alkali land according to claim 1, characterized in that: After step 4 is executed, the execution status of irrigation and drainage is monitored in real time, and the irrigation strategy is adjusted according to the execution status of irrigation and drainage. At the same time, execution process data including irrigation amount and duration are acquired, as well as soil salinity change data after irrigation are acquired as process effect data. The execution process data and process effect data are fed back to the reinforcement learning optimization unit to optimize the next irrigation strategy.

3. The intelligent irrigation regulation method for dynamic water-salt balance in saline-alkali land according to claim 1, characterized in that: The basic database includes dynamic monitoring data, pre-set crop basic data and localized parameter library data within the system; The meteorological parameters include solar radiation, air temperature, air humidity, wind speed, and rainfall. The soil parameters include soil moisture, soil salinity, soil temperature, and soil pH. The groundwater and irrigation water parameters include groundwater level, groundwater conductivity, and irrigation water conductivity; The meteorological forecast data includes rainfall, temperature, and wind speed for the next three days; The basic crop data includes the growth stage classification standards and corresponding crop coefficients, crop salt tolerance thresholds, and root depths at different growth stages. The localized parameter library includes soil hydraulic function parameters describing soil water holding and water conduction properties, and solute transport parameters describing the diffusion and migration properties of salts in the soil.

4. The intelligent irrigation regulation method for dynamic water-salt balance in saline-alkali land according to claim 3, characterized in that: The process of setting the dynamic target salinity threshold is as follows: retrieve basic crop data from the basic database, and set the target salinity threshold according to crop type and growth stage.

5. The intelligent irrigation regulation method for dynamic water-salt balance in saline-alkali land according to claim 1, characterized in that: The steps for calculating crop evapotranspiration water demand are as follows: based on the collected meteorological parameters, the standard reference crop evapotranspiration ET0 is calculated using the FAO Penman-Monteith formula; According to the growth stage of the crop, the corresponding crop coefficient K is obtained from the basic database c The basic crop water requirement ET c_base is calculated, and the calculation formula is: ET c_base = K c × ET0; Based on the real-time monitoring of the root zone soil salinity and the crop salt tolerance threshold in the basic database, the salt stress coefficient K is calculated s , and the basic crop water requirement is corrected to obtain the final crop evapotranspiration water requirement ET c , and the calculation formula is: ET c = K s × ET c_base .

6. The intelligent irrigation regulation method for dynamic water-salt balance in saline-alkali land according to claim 1, characterized in that: The step of calculating the adaptive rinsing water volume includes: calculating the minimum rinsing fraction LR based on the steady-state salinity balance principle, the calculation formula is: LR = EC iw / (k x EC e_target ); where EC is the electrical conductivity of the irrigation water, EC iw where EC is the electrical conductivity of the irrigation water, EC e_target where EC is the electrical conductivity of the irrigation water, EC Combined with the actual evapotranspiration water requirement (ET) of the crop calculated by the crop model unit after salt stress correction. c The minimum rinsing fraction LR is converted into an adaptive rinsing water volume V. leach The calculation formula is: V leach = LR×ET c / (1-LR)。 7. The intelligent irrigation regulation method for dynamic water-salt balance in saline-alkali land according to claim 1, characterized in that: The risk of salt return includes low risk, medium risk, and high risk; The steps for assessing the risk of salt return include: When the groundwater level is greater than 2m and the groundwater conductivity EC gw Less than or equal to 6 dS / m, or groundwater level between 1m and 2m and groundwater conductivity EC gw When the value is less than 3 dS / m, the risk of salt return is low. When the groundwater level is greater than 2m and the groundwater conductivity EC gw Greater than 6 dS / m, or groundwater level between 1m and 2m and groundwater conductivity EC gw Between 3 dS / m and 6 dS / m, or when the groundwater level is less than 1m and the groundwater conductivity EC gw When the value is less than 3 dS / m, the risk of salt return is medium. When the groundwater level is between 1m and 2m and the groundwater conductivity ECgw is greater than 6 dS / m, or when the groundwater level is less than 1m and the groundwater conductivity ECgw is greater than 6 dS / m, gw When the flow rate is greater than or equal to 3 dS / m, the risk of salt return is high. If the daily reference evapotranspiration ET0 is greater than the correction threshold of 6 mm / day and the risk of salt return is low, then the risk of salt return is upgraded to medium risk. If the daily reference evapotranspiration ET0 is greater than the correction threshold of 6 mm / day and the risk of salt return is medium, then the risk of salt return is upgraded to high.

8. An intelligent irrigation control system for implementing the intelligent irrigation control method for dynamic water-salt balance in saline-alkali land as described in any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a model and evaluation module, a decision-making module, and an execution module. The modules transmit and process data through wireless communication networks and computing and storage resources deployed on cloud platforms or local servers. The data acquisition module is used to collect dynamic monitoring data and save it to the basic database; the model and evaluation module calls data from the basic database to calculate crop evapotranspiration water demand, adaptive leaching water volume, and salt return risk assessment, and generates strategy optimization parameters to be output to the decision module. The decision-making module integrates the output of the model and evaluation modules, as well as the meteorological forecast data retrieved from the basic database, to generate irrigation strategies and executable instructions, which are then sent to the execution module. The execution module coordinates and regulates irrigation and drainage to complete intelligent irrigation regulation and performs dual-path feedback. The dual-path feedback includes triggering immediate adjustments or alarms based on the execution status, and feeding back the execution process data and the process effect data monitored by the data acquisition module to the model and evaluation module for optimization of the next irrigation strategy; The basic database includes dynamic monitoring data continuously collected by the data acquisition module, as well as crop basic data and localized parameter libraries pre-set during the system deployment phase.

9. The intelligent irrigation control system for dynamic water-salt balance in saline-alkali land according to claim 8, characterized in that: The model and evaluation module includes a collaborative crop model unit, a water-salt dynamic balance and risk assessment unit, and a reinforcement learning optimization unit. The crop model unit is used to calculate the actual evapotranspiration water demand of crops considering salt stress correction; the water-salt dynamic balance and risk assessment unit realizes salt profile estimation, dynamic target salt threshold setting, adaptive leaching water volume calculation, salt return risk assessment and end-of-growing-season assessment report generation by calling the basic database. The reinforcement learning optimization unit generates and optimizes key strategy control parameters for guiding irrigation through interactive learning between the agent and the simulation environment.

10. The intelligent irrigation control system for dynamic water-salt balance in saline-alkali land according to claim 9, characterized in that: The reinforcement learning optimization unit includes a simulation environment subunit, an agent subunit, a training and optimization subunit, and a policy output and decision feedback subunit. The simulation environment subunit is used to construct a simulation environment for intelligent agent interaction to dynamically simulate crop growth and dynamic soil water and salt response under different irrigation strategies. The intelligent agent subunit includes a state input subunit, an action output subunit, and a reward function calculation subunit; the state input subunit integrates input data; The action output subunit generates the policy correction coefficients based on the agent's learning policy; The reward function calculation subunit guides the agent's policy learning through the reward function; The training and optimization subunit executes the training process using reinforcement learning algorithms, updating policy parameters based on the accumulated reward information after repeated interactions between the agent and the environment; The strategy output and decision feedback subunit transmits the strategy correction coefficients generated by the agent to the decision module, and receives feedback data on the execution effect of the irrigation strategy to calibrate the simulation environment, thereby realizing strategy iteration and optimization.

Images