Multi-farmland cooperative irrigation method and system fusing weather conditions and ai large model

Abstract

The present application belongs to the field of automatic irrigation control, and particularly relates to a multi-farmland collaborative irrigation method and system fusing meteorological conditions and an AI large model. The method accurately calculates the standard water demand deviation of each farmland by constructing a growth cycle water demand model with rainfall prediction, comprehensively considering rainfall intensity, duration and soil layer water content change; adopts a particle swarm algorithm to optimize a BP neural network, takes the maximum probability of water content consistency as the target, combines multi-dimensional parameters such as water flow diffusion mapping weight, farmland area and soil characteristics, and dynamically solves the best opening degree value and opening time stamp of each digital water-saving gate; and introduces an irrigation simulation control platform for closed-loop verification and iterative optimization, realizes accurate control of multi-farmland collaborative irrigation by real-time monitoring of irrigation deviation value and water content consistency probability, effectively improves water resource utilization efficiency and irrigation uniformity, and ensures that the water demand of each farmland is accurately met.

Description

Technical Field

[0001] This invention belongs to the field of automatic irrigation control, and in particular relates to a method and system for collaborative irrigation of multiple farmlands that integrates meteorological conditions and AI large models. Background Technology

[0002] With the large-scale deployment of digital water-saving gates, soil moisture sensors, and weather forecasting modules in irrigation systems via IoT interfaces, multi-farmland precision irrigation systems exhibit characteristics such as multiple water sources, parallel connection of multiple gates, uneven distribution of pipeline impedance, complex coupling of control links, and dynamic migration of crop water demand. In weak pipeline networks or high permeability irrigation scenarios, the central control system usually needs to undertake tasks such as water allocation, pressure / flow stabilization, irrigation timing coordination, and multi-gate collaboration. The system is prone to irrigation deviations caused by the interaction between the controller and pipeline dynamics, and multiple links such as soil, crop, and climate. Moreover, the dominant deviation mode may migrate with changes in pipeline topology, gate equivalent impedance, crop growth stage, and meteorological conditions.

[0003] In existing technologies, a common approach is to plan and regulate irrigation based on static models, such as constructing soil moisture balance models or equivalent hydraulic models, and combining methods such as water demand prediction, timed and quantitative control, zoned rotation irrigation, and empirical adjustments based on historical data to determine irrigation strategies. This type of method has good feasibility when the climate is stable, the soil is homogeneous, and the crop growth stage is known. However, in real-world scenarios with multiple farmlands connected in parallel, diverse crop types, significant spatial variability of soil, and frequent fluctuations in meteorological conditions, the model establishment and parameter calibration costs are high. Furthermore, it is sensitive to real-time dynamics of soil moisture, rainfall uncertainty, sensor noise, and gate response lag, which can easily lead to delayed irrigation decisions or deviations in the tuning results from actual water demand. It is difficult to maintain accurate and balanced irrigation effects when crop water demand changes rapidly.

[0004] Another approach employs data-driven or rule-based reasoning models to assess and make decisions regarding farmland conditions. For instance, fuzzy logic or simple neural networks can be used to extract features from soil moisture, weather forecasts, and other data to determine irrigation needs, triggering gate opening / closing or duration adjustments accordingly. While this approach reduces reliance on precise physical modeling to some extent, existing conventional models still suffer from insufficient adaptability in precision irrigation scenarios. The averaging or partitioning aggregation operations used to reduce computational complexity may weaken the fine-grained identification of specific water requirements for individual farmlands, making it difficult to achieve high-fidelity representation of irrigation deviations coupled across multiple farmlands. Although time-series trends can be modeled, internal decisions are often black-box outputs, making it difficult to identify overall irrigation deviations while providing contribution positioning criteria for coordinated water allocation across multiple gates. Consequently, pressure regulation and water allocation tend to rely on uniform settings or empirical rules, easily leading to problems such as hydraulic interference between parallel gates, insufficient irrigation, or excessive waste.

[0005] Therefore, the technical problem that the existing technology urgently needs to solve is how to extract the water demand deviation characteristics of soil-climate-material coupling with high fidelity in the complex irrigation system environment of multiple farmlands in parallel, and accurately locate the source of deviation contribution while identifying the overall irrigation imbalance mode, so as to carry out targeted pressure regulation and dynamic coordinated allocation of water. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes a multi-farmland collaborative irrigation method and system that integrates meteorological conditions and a large AI model. This method constructs a growth cycle water requirement model with rainfall prediction, comprehensively considering rainfall intensity, duration, and soil moisture content variations to accurately calculate the standard water requirement deviation for each farmland. It employs a particle swarm optimization algorithm to optimize a backpropagation neural network, aiming to maximize the consistency probability of moisture content. Combining multi-dimensional parameters such as water flow diffusion mapping weights, farmland area, and soil characteristics, it dynamically solves for the optimal opening value and opening timestamp of each digital water-saving gate. Furthermore, an irrigation simulation control platform is introduced for closed-loop verification and iterative optimization. By monitoring the irrigation deviation value and the consistency probability of moisture content in real time, precise control of multi-farmland collaborative irrigation is achieved, effectively improving water resource utilization efficiency and irrigation uniformity, and ensuring that the water requirements of each farmland are accurately met.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] Multi-farmland coordinated irrigation methods that integrate meteorological conditions and AI big data models include:

[0009] The real-time water content of each farmland is obtained and combined with a preset growth cycle water requirement model with rainfall prediction to obtain the standard water requirement deviation of each farmland and the standard total water requirement deviation of all farmlands.

[0010] The method obtains the area of ​​each farmland, the distance between the corresponding digital water-saving gate branch and the main gate of each farmland, and aims to maximize the preset water content consistency probability. It uses the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands as constraints. Combining a BP neural network optimized by particle swarm optimization and preset water flow diffusion mapping weights, it obtains the opening value and opening timestamp of each digital water-saving gate branch. Each digital water-saving gate branch corresponds one-to-one with each farmland. The water content consistency probability is used to characterize the probability that the real-time irrigation amount of each farmland meets the corresponding standard water demand deviation under the constraint of the standard total water demand deviation of all farmlands.

[0011] In response to the opening value and opening timestamp of each digital water-saving gate, and in conjunction with a preset irrigation simulation control platform, real-time irrigation simulation is performed. Through a configured flow meter and camera, the irrigation deviation value and water content consistency probability of each farmland after the irrigation simulation are monitored. When the water content consistency probability does not meet the corresponding preset threshold, the irrigation deviation value of each farmland after the simulation is fed back to the irrigation simulation control platform for simulation until the water content consistency probability meets the preset threshold.

[0012] Specifically, the construction and training process of a growing cycle water requirement model with rainfall prediction includes:

[0013] The standard growth water requirement of each target crop in each farmland in each growth cycle is obtained, as well as the precipitation parameters of the corresponding farmland area in each growth cycle and the change value of soil moisture content in the stratified soil after rainfall. The timestamps are aligned and preprocessed. The precipitation parameters include rainfall intensity, rainfall duration and historical rainfall prediction accuracy confidence.

[0014] Based on the changes in soil moisture content in different layers of farmland after rainfall and the water retention coefficient corresponding to soil type, a mapping function of rainfall intensity-layer effective infiltration is established; wherein the changes in soil moisture content in different layers of farmland after rainfall are obtained by the moisture content of each layer of soil before rainfall and the moisture content of each layer of soil after rainfall under the corresponding soil type conditions of each farmland, with the corresponding rainfall intensity and duration.

[0015] Specifically, the construction and training process of the growing cycle water requirement model with rainfall prediction also includes:

[0016] Based on the obtained crop growth stage codes, rainfall intensity and duration corresponding to the growth cycle, historical rainfall prediction accuracy confidence, and rainfall intensity-stratified effective infiltration mapping function, a rainfall prediction input sequence is constructed.

[0017] Based on the changes in soil moisture content in different strata after rainfall and the soil moisture content after rainfall compensation, an output layer feature vector is constructed.

[0018] Specifically, the construction and training process of the growing cycle water requirement model with rainfall prediction also includes:

[0019] The rainfall prediction input sequence and the output layer feature vector are input into the growth cycle water requirement model with rainfall prediction constructed by the BP neural network. The model is trained by combining the preset training loss function to obtain the trained growth cycle water requirement model with rainfall prediction. The model outputs the predicted changes in soil moisture content after rainfall and the farmland moisture content after rainfall compensation.

[0020] The preset training loss function is constructed from the difference between the predicted change in soil moisture content after rainfall and the actual measured change in soil moisture content after rainfall, as well as the difference between the predicted rainfall-compensated soil moisture content and the real-time measured rainfall-compensated soil moisture content.

[0021] The standard water requirement deviation for each farmland is obtained by subtracting the predicted rainfall-compensated farmland water content from the standard growth water requirement of the target crop in each growth cycle. The predicted rainfall-compensated farmland water content is obtained by combining the farmland water content before rainfall with the effective infiltration corresponding to the rainfall.

[0022] Specifically, the water flow diffusion mapping weight is used to characterize the effective diffusion area and diffusion uniformity of irrigation water per unit time in different farmlands under the conditions of soil moisture content, topography, and irrigation flow rate. It is used to adjust the opening value and opening time of the digital water-saving gate outlets in different farmlands. The construction process of the water flow diffusion mapping weight includes:

[0023] The soil type, average soil moisture content, slope, irrigation flow rate, and effective irrigation diffusion area and ratio per unit time for each farmland were obtained and preprocessed and aligned. All effective irrigation diffusion area ratios were constructed from the ratio of the effective diffusion area irrigated per unit time to the total area of ​​the corresponding farmland.

[0024] Specifically, the process of constructing the water flow diffusion mapping weights also includes:

[0025] Using the pre-treated effective diffusion area of ​​irrigation as the dependent variable, the irrigation flow rate as the independent variable, and the soil type, average soil moisture content, and slope of each farmland as covariates, a multivariate regression algorithm was used to fit the diffusion fitting function.

[0026] Based on the diffusion fitting function combined with the irrigation simulation control platform and the control variable method, with the goal of maximizing the consistency of the effective diffusion area ratio of irrigation, the irrigation flow velocity is simulated and adjusted to obtain the farmland flow velocity sequence corresponding to the maximum consistency and the regression coefficients corresponding to the diffusion fitting function, and the water flow diffusion mapping weight is constructed.

[0027] Specifically, the opening degree value and opening timestamp of each digital water-saving gate are obtained, including:

[0028] The input sequence for the particle swarm algorithm is constructed by obtaining the area of ​​each farmland, the distance between the branch gate and the main gate of the digital water-saving gate corresponding to each farmland, the soil type, average soil moisture content, farmland slope, the mapping relationship between gate opening and flow velocity, the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands.

[0029] The fitness function is constructed by combining the product of the consistency probability of water content and the consistency of the ratio of effective irrigation diffusion area, the irrigation flow rate and the irrigation start time stamp, and the water flow diffusion mapping weights with a BP neural network. The constraint space is constructed by the standard water demand deviation of each farmland, the standard total water demand deviation of all farmlands, the farmland area, and the farmland slope.

[0030] Specifically, obtaining the opening degree value and opening timestamp of each digital water-saving gate branch also includes:

[0031] The input sequence, fitness function, and constraint space of the particle swarm optimization algorithm are input into the particle swarm optimization algorithm and iteratively trained in combination with a preset training period to obtain the irrigation flow rate and irrigation start timestamp of each iteration.

[0032] Based on the irrigation flow rate and irrigation start timestamp of each farmland output in each iteration, a fuzzy control algorithm is used to generate coordinated irrigation control instructions. Based on the coordinated irrigation control instructions and the irrigation simulation control platform, irrigation simulation is carried out to obtain the consistency of real-time water demand deviation, water content consistency probability and effective irrigation diffusion area ratio of all farmlands.

[0033] Specifically, obtaining the opening degree value and opening timestamp of each digital water-saving gate branch also includes:

[0034] When at least one of the real-time water demand deviation, water content consistency probability and irrigation effective diffusion area ratio of all farmland does not meet the corresponding threshold, the iteration is repeated until all of them meet the corresponding threshold, and the irrigation flow rate and irrigation start timestamp of each farmland are obtained.

[0035] If at least one farmland still fails to meet the corresponding threshold after the iteration ends, the standard water requirement deviation of the corresponding farmland is fed back into the growth cycle water requirement model with rainfall prediction for correction, until the consistency of real-time water requirement deviation, water content consistency probability and irrigation effective diffusion area ratio of all farmland meets the corresponding threshold.

[0036] A multi-farmland collaborative irrigation system that integrates meteorological conditions and AI big data models includes: a water demand prediction module, an optimization module, and a simulation feedback module;

[0037] The water demand prediction module is used to obtain the real-time water content of each farmland and combine it with the preset growth cycle water demand model with rainfall prediction to obtain the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands.

[0038] The optimization module is used to obtain the area of ​​each farmland, the distance between the corresponding digital water-saving gate branch and the main gate of the digital water-saving gate for each farmland, and to calculate the opening value and opening timestamp of each digital water-saving gate branch with the objective of maximizing the preset water content consistency probability. It uses the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands as constraints, combined with a BP neural network optimized by particle swarm optimization, to achieve this. Each digital water-saving gate branch corresponds one-to-one with each farmland. The water content consistency probability is used to characterize the probability that the real-time irrigation amount of each farmland meets the corresponding standard water demand deviation under the constraint of the standard total water demand deviation of all farmlands.

[0039] The simulation feedback module responds to the opening value and opening timestamp of each digital water-saving gate outlet, and performs real-time irrigation simulation in conjunction with a preset irrigation simulation control platform. It monitors the irrigation deviation value and water content consistency probability of each farmland after the irrigation simulation. When the water content consistency probability does not meet the corresponding preset threshold, the irrigation deviation value of each farmland after the simulation is fed back to the irrigation simulation control platform for simulation until the water content consistency probability meets the preset threshold.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] This invention addresses the shortcomings of existing technologies by integrating real-time moisture content with a growth cycle water requirement model incorporating rainfall prediction. It accurately calculates the standard water requirement for each farmland and the total system water requirement. Furthermore, considering spatial physical constraints such as farmland area and gate distance, and with the core objective of maximizing the probability of consistent moisture content, it uses a BP neural network optimized by particle swarm optimization to dynamically solve for the optimal opening degree and opening / closing sequence of each digital water-saving gate under the constraint of total water requirement. This method transforms the extensive water supply of traditional flood irrigation into precise and coordinated control based on multi-source data and intelligent algorithms. It effectively overcomes irrigation deviations caused by spatial heterogeneity and dynamic water demand, ultimately significantly improving the irrigation uniformity and water use efficiency of multiple farmland groups. While ensuring that each farmland receives irrigation as needed, it reduces water waste and achieves a unity of system-level water conservation and balanced irrigation. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the distribution of water supply pumping stations and farmland according to the present invention;

[0043] Figure 2 This is a flowchart of the multi-farmland collaborative irrigation method that integrates meteorological conditions and AI large-scale models according to the present invention;

[0044] Figure 3 This is a block diagram of a multi-farmland collaborative irrigation system that integrates meteorological conditions and a large AI model, as described in this invention. Detailed Implementation

[0045] Example 1

[0046] This application applies to smart water management scenarios that require multi-source environmental data fusion for precision irrigation decision-making, particularly farmland water demand regulation systems characterized by significant spatiotemporal heterogeneity, multi-objective collaborative optimization, and dynamic prediction and compensation. In such irrigation decision-making processes, due to the multi-dimensional dynamic changes in farmland soil properties, crop growth stages, and meteorological conditions, traditional irrigation systems, lacking precise prediction and collaborative optimization mechanisms, are highly susceptible to problems such as uneven water supply, resource waste, or delayed response. Typical technical bottlenecks in precision irrigation within the field of smart water management include, but are not limited to:

[0047] In the task of water allocation for coordinated irrigation of multiple farmlands, it is necessary to simultaneously meet the differentiated water demand of each farmland and the total water balance of the system. However, the actual water supply process involves complex physical constraints such as differences in water flow diffusion, gate response delays, and pipeline pressure losses.

[0048] In irrigation decisions based on rainfall forecast data, there is an optimization challenge involving the dynamic coupling of forecast uncertainty, effective rainfall conversion bias, and crop growth stage water requirement patterns. Without probabilistic processing and dynamic compensation, it is easy to cause insufficient irrigation or excessive water supply.

[0049] In the multi-gate collaborative control model, the operation of each digital water-saving gate under the same water supply network affects each other. It is necessary to conduct cross-temporal and spatial scale fusion simulation of water flow, soil infiltration and crop water absorption process to accurately generate control commands.

[0050] The method proposed in this application does not rely on fixed irrigation regimes, empirical thresholds, or traditional PID control methods. Instead, it takes water demand prediction, simulation, and closed-loop correction as its main technical line. Through the calculation of effective rainfall infiltration, the construction of water flow diffusion mapping weights, and a multi-objective optimization algorithm mechanism, it completes the generation and verification of dynamic irrigation strategies.

[0051] Please see Figure 1 and Figure 2 The present invention provides an embodiment of a multi-farmland collaborative irrigation method that integrates meteorological conditions and AI large-scale models, comprising the following steps:

[0052] S1. Obtain the real-time water content of each farmland and combine it with the preset growth cycle water requirement model with rainfall prediction to obtain the standard water requirement deviation of each farmland and the standard total water requirement deviation of all farmlands.

[0053] like Figure 1As shown, the corresponding scenario includes deployed water supply pumping stations, main water pipelines, and branch water pipelines leading to each farmland. Here, for example, eight farmlands are given. A1 to A8 on each farmland are deployed digital water-saving gate outlets that can regulate the water inflow. Each digital water-saving gate outlet is equipped with a flow meter to monitor the water inflow of each farmland in real time. A is the gate outlet deployed on the main water pipeline.

[0054] S2. Obtain the area of ​​each farmland, the distance between the corresponding branch gate and the main gate of the digital water-saving gate for each farmland, and, with the goal of maximizing the preset water content consistency probability, using the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands as constraints, combine a BP neural network optimized by particle swarm optimization algorithm with preset water flow diffusion mapping weights to obtain the opening value and opening timestamp of each branch gate of the digital water-saving gate; wherein each branch gate of the digital water-saving gate corresponds one-to-one with each farmland; the water content consistency probability is used to characterize that, under the constraint of the standard total water demand deviation of all farmlands, the real-time irrigation amount of each farmland meets the corresponding farmland's requirements. The probability of standard water demand deviation; in this embodiment, the water content consistency probability is used to characterize the proportion or probabilistic measure of the number of farmlands that, under the premise of satisfying the total system water volume constraint of the standard total water demand deviation of all farmlands, achieve the desired irrigation water volume for each farmland through irrigation control. The standard water demand deviation of each farmland is the net water demand that the farmland still needs to be supplemented by the irrigation system within a specific growth cycle after considering predicted rainfall compensation. The total standard water demand deviation of all farmlands represents the sum of the net water demand of all farmlands under the same irrigation system, constituting the total water volume constraint that the irrigation network needs to supply. The standard water demand deviation of each farmland, as the basis for precise control at the field operation level, directly determines the target water supply of the digital water-saving gate outlet corresponding to each independent farmland. It quantifies the net water shortage that farmland still needs for artificial irrigation at a specific growth stage after deducting effective rainfall replenishment, and serves as the fundamental benchmark for generating differentiated irrigation instructions tailored to each field. The standard total water demand deviation for all farmland serves as a global water volume constraint at the system network level, representing the total water volume that the entire network system needs to transport during this irrigation cycle. It is used to ensure that the sum of the planned opening degree and opening time of all branch gates matches the water supply capacity of the pumping stations and the water conveyance capacity of the main pipeline, and is a key constraint for achieving system-wide water balance and avoiding insufficient water supply or resource waste.

[0055] S3. Responding to the opening value and opening timestamp of each digital water-saving gate, and combined with the preset irrigation simulation control platform, real-time irrigation simulation is performed. Through the configured flow meter and camera, the irrigation deviation value and water content consistency probability of each farmland after the irrigation simulation are monitored. When the water content consistency probability does not meet the corresponding preset threshold, the irrigation deviation value of each farmland after the simulation is fed back to the irrigation simulation control platform for simulation until the water content consistency probability meets the preset threshold.

[0056] It should be further explained that the flow meter in this embodiment is used to monitor whether the real-time water intake of each farmland meets the standard water demand deviation, that is, whether the corresponding water content consistency probability meets the corresponding preset threshold. The closing timestamp of the corresponding digital water-saving gate outlet of each farmland is determined by the monitoring results of the flow meter. It should also be explained that the camera in this embodiment is used to monitor and analyze the diffusion velocity of water flow in the corresponding farmland and the uniformity of water content distribution in the farmland through the configured analysis algorithm, that is, the ratio of the effective irrigation diffusion area to the effective irrigation diffusion area of ​​each farmland. The configured analysis algorithm is used by those skilled in the art according to the corresponding analysis requirements. It calls the farmland water content detection model pre-trained with historical irrigation data to detect the ratio of the effective irrigation diffusion area to the effective irrigation diffusion area of ​​the corresponding farmland and to calculate the water diffusion velocity in each farmland. When the water content in the farmland has covered the surface soil layer, the water diffusion velocity is set as the water flow velocity of the corresponding farmland inlet.

[0057] It should be further explained that, in this embodiment, a smoke sensor is also configured at the main gate of the digital water-saving gate to detect potential safety hazards in the devices configured in the entire irrigation system in real time, and to feed the safety monitoring results back to the irrigation simulation control platform to dynamically adjust the opening and closing status of the main gate and the branch gates of the digital water-saving gate in the entire irrigation process.

[0058] It should be further explained that, in addition to being suitable for multi-farm irrigation in this embodiment, the flow meter is also suitable for the application of pesticides and fertilizers to multiple farmlands. Specifically, for each farmland, independent pesticide or fertilizer application requirements parameters are preset at the digital water-saving gate outlet, including the target pesticide concentration or fertilizer application rate per unit area. Secondly, the flow meter monitors and feeds back the unit flow rate value of the corresponding gate outlet to the control unit in real time. Finally, the control unit executes one of the following two precise control modes: In variable concentration mode, based on the preset target concentration and the real-time unit flow rate, the injection rate of the original pesticide or high-concentration fertilizer injected into the main pipeline is dynamically calculated and adjusted so that the concentration of the mixed solution output from the corresponding gate outlet is consistent with the target concentration. In quantitative fertilization mode, based on the preset fertilizer application rate per unit area and the real-time unit flow rate, the total opening time required for the corresponding gate outlet to complete the predetermined total fertilizer application is calculated, and the opening and closing of the gate outlet is controlled quantitatively according to this time, thereby realizing independent and customized variable control of the pesticide concentration or total fertilizer application for each farmland.

[0059] It should be further explained that the construction and training process of the growing cycle water requirement model with rainfall prediction in this embodiment includes:

[0060] The standard growth water requirement for each target crop in each growth cycle is obtained, along with the precipitation parameters and post-rainfall soil moisture content changes for the corresponding farmland area within each growth cycle. These parameters are then time-stamped and preprocessed. The precipitation parameters include rainfall intensity, rainfall duration, and historical rainfall prediction accuracy confidence level. It should be further noted that in this embodiment, the standard growth water requirement is determined by consulting an agricultural irrigation standard database and combining it with the target crop's growth cycle, serving as the benchmark value for calculating the basic water requirement of farmland. Rainfall intensity and duration are collected in real-time by meteorological monitoring stations deployed in the farmland area to quantify the characteristics of natural rainfall input. The historical rainfall prediction accuracy confidence level is extracted from prediction metadata provided by meteorological forecasting service providers and serves as a weighted basis for assessing prediction reliability. Post-rainfall soil moisture content changes are measured before and after rainfall events using a network of soil moisture sensors buried at depths of 0-10cm, 10-20cm, and 20-30cm, used to construct the response relationship between rainfall infiltration and soil water holding capacity. All parameters are time-labeled using crop growth stages, timestamped using a unified clock source, and preprocessed with anomaly detection based on the 3σ criterion and min-max normalization to form a spatiotemporally consistent input dataset.

[0061] Based on the changes in soil moisture content across different soil layers after rainfall and the water retention coefficients corresponding to soil types, a mapping function between rainfall intensity and effective infiltration of each layer is established. The changes in soil moisture content across different soil layers after rainfall are obtained by comparing the moisture content of each soil layer before rainfall with the corresponding rainfall intensity and duration, and then comparing the moisture content of each soil layer after rainfall under the corresponding soil type conditions for each farmland. In this embodiment, the water retention coefficients corresponding to each soil type are obtained based on matching the soil types of each farmland, the rainfall intensity is obtained through actual measurement or prediction, and the effective infiltration of each layer is obtained based on the analysis of the above parameters. This mapping function is used in this embodiment to quantify the effective replenishment of different soil layers by rainfall, providing relevant parameters for effective rainfall replenishment in a growth cycle water requirement model with rainfall prediction, and supporting the accurate calculation of the standard water requirement deviation of farmland. The motivation for constructing this mapping function is to solve the problem that traditional water requirement estimation does not fully consider the influence of soil stratification characteristics and different soil water retention capacities on rainfall infiltration, resulting in large deviations in the calculation of effective rainfall replenishment, which in turn affects the prediction accuracy of the water requirement model. The specific construction process of the rainfall intensity-layered effective infiltration mapping function is as follows: Based on the obtained changes in soil moisture content of farmland layers after rainfall, the water retention coefficient corresponding to soil type, and rainfall intensity data, first, timestamp alignment, outlier removal, and standardization preprocessing are performed. Then, the coupling characteristics of the changes in soil moisture content, water retention coefficient, and rainfall intensity of each soil layer under different rainfall intensities are extracted. By analyzing the dynamic changes in soil moisture content and water retention coefficient of each layer with rainfall intensity, the correlation weight of each influencing factor on the effective infiltration of the layers is determined. A function model is constructed using a multiple regression algorithm or data fitting method, with rainfall intensity as the independent variable, and the water retention coefficient corresponding to soil type and the changes in soil moisture content of farmland layers after rainfall as intermediate variables. A one-to-one quantitative relationship between rainfall intensity and the effective infiltration of each soil layer is established. The function parameters are iteratively optimized through sample data, and the deviation between the effective infiltration of the layers output by the function and the measured value is verified. Finally, a stable rainfall intensity-layered effective infiltration mapping function is formed, realizing the accurate calculation of the effective infiltration of each soil layer under different rainfall intensities.

[0062] To address the problem that traditional crop water requirement models do not fully integrate crop growth stage specificity, rainfall prediction reliability, and soil stratification and infiltration characteristics, resulting in inaccurate prediction of farmland water content after rainfall compensation and insufficient accuracy in calculating water requirement deviation, this embodiment constructs a rainfall prediction input sequence based on the obtained crop growth stage code, rainfall intensity and duration corresponding to the growth cycle, historical rainfall prediction accuracy confidence, and rainfall intensity-stratified effective infiltration mapping function.

[0063] Based on the changes in soil moisture content in stratified fields after rainfall and the soil moisture content after rainfall compensation, an output layer feature vector is constructed.

[0064] The rainfall prediction input sequence and the output layer feature vector are input into the growth cycle water requirement model with rainfall prediction constructed by the BP neural network. The model is trained by combining the preset training loss function to obtain the trained growth cycle water requirement model with rainfall prediction. The model outputs the predicted changes in soil moisture content after rainfall and the farmland moisture content after rainfall compensation.

[0065] The preset training loss function is constructed from the difference between the predicted change in soil moisture content after rainfall and the actual measured change in soil moisture content after rainfall, as well as the difference between the predicted rainfall-compensated soil moisture content and the real-time measured rainfall-compensated soil moisture content.

[0066] The standard water requirement deviation for each farmland is obtained by subtracting the predicted rainfall-compensated farmland water content from the standard growth water requirement of the target crop in each growth cycle. The predicted rainfall-compensated farmland water content is obtained by combining the farmland water content before rainfall with the effective infiltration corresponding to the rainfall.

[0067] Traditional irrigation systems essentially simplify the irrigation process to a simple water allocation problem, neglecting the hydrodynamic nature of water flow in the field, which is constrained by soil physical properties (such as texture, structure, and initial moisture content) and topographic features (slope, micro-topography). This simplification leads to significant spatial heterogeneity in actual irrigation, even when the total water supply precisely meets demand, due to differences in hydraulic response across different fields. For example, clay soil fields may experience surface runoff due to slow infiltration rates, while sandy soil fields may suffer from insufficient root zone wetting due to rapid infiltration. Simultaneously, gravity in sloping fields exacerbates water flow velocity, further weakening irrigation uniformity. Therefore, this proposed solution stems from a deep understanding that the two core aspects of irrigation—"water allocation" and "spatial distribution of moisture"—must be optimized synergistically. It aims to fundamentally address the systemic defect of "accurate water allocation but uneven irrigation" caused by the heterogeneity of the physical environment by establishing a mapping weight that quantifies water diffusion efficiency and integrating soil hydrodynamic principles into control decisions. This embodiment aims to fundamentally solve the differences in water flow efficiency caused by the spatial heterogeneity of physical properties such as soil infiltration capacity, terrain slope and initial water content by constructing water flow diffusion mapping weights. It upgrades the simple water allocation optimization in traditional irrigation to a collaborative optimization that simultaneously ensures the dual objectives of "accurate total water volume" and "uniform spatial distribution", thereby achieving accurate quantification and dynamic compensation of the actual water transport process in the field.

[0068] It should be further explained that the water flow diffusion mapping weight in this embodiment is used to characterize the effective diffusion area and diffusion uniformity of irrigation water per unit time in different farmlands under the conditions of soil moisture content, topography, and irrigation flow rate. It is used to adjust the opening degree and opening time of the digital water-saving gate outlets in different farmlands. Furthermore, the construction process of the water flow diffusion mapping weight in this embodiment includes:

[0069] The soil type, average soil moisture content, slope, irrigation flow rate, and effective irrigation diffusion area and ratio per unit time for each farmland were obtained and preprocessed and aligned. All effective irrigation diffusion area ratios were constructed from the ratio of the effective diffusion area irrigated per unit time to the total area of ​​the corresponding farmland.

[0070] Using the pre-treated effective diffusion area of ​​irrigation as the dependent variable, the irrigation flow rate as the independent variable, and the soil type, average soil moisture content, and slope of each farmland as covariates, a multivariate regression algorithm was used to fit the diffusion fitting function.

[0071] Based on the diffusion fitting function combined with the irrigation simulation control platform and the control variable method, with the goal of maximizing the consistency of the effective diffusion area ratio of irrigation, the irrigation flow velocity is simulated and adjusted to obtain the farmland flow velocity sequence corresponding to the maximum consistency and the regression coefficients corresponding to the diffusion fitting function, and the water flow diffusion mapping weight is constructed.

[0072] It should be further explained that, in this embodiment, obtaining the opening value and opening timestamp of each digital water-saving gate outlet includes:

[0073] The input sequence for the particle swarm optimization algorithm is constructed by obtaining the area of ​​each farmland, the distance between the branch gate and the main gate of the digital water-saving gate corresponding to each farmland, the soil type, average soil moisture content, and slope of each farmland, as well as the mapping relationship table between gate opening and flow velocity, the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands. In this embodiment, the mapping relationship table between gate opening and flow velocity is constructed by those skilled in the art based on historical measurement data combined with support vector machine.

[0074] The fitness function is constructed using a backpropagation neural network, combining the product of the consistency probability of water content and the consistency of the ratio of effective irrigation diffusion area, along with the functions of irrigation flow rate, irrigation start timestamp, and water flow diffusion mapping weights. The constraint space is constructed using the standard water requirement deviation of each farmland, the standard total water requirement deviation of all farmlands, farmland area, and farmland slope. This embodiment uses the product of the consistency probability of water content and the consistency of the ratio of effective irrigation diffusion area as the core component of the fitness function. Its design motivation stems from a deep understanding of the multi-objective collaborative optimization nature of irrigation systems. The consistency probability of water content ensures the basic accuracy of irrigation, guaranteeing that the amount of water received by each farmland precisely matches its water requirement deviation; while the consistency of the ratio of effective irrigation diffusion area ensures the spatial uniformity of irrigation, solving the problem of uneven water distribution caused by soil and topographic differences. The product relationship between the two has a strict mathematical meaning, requiring that both indicators must be optimized simultaneously. A significant decrease in either indicator will have an amplified negative impact on the fitness value. This effectively avoids the optimization process from falling into a local optimum that prioritizes a single indicator, such as pursuing only the accuracy of total water volume while ignoring the uniformity of field distribution. On this basis, the irrigation flow rate and start time stamp are introduced as optimization variables to take into account the needs of system operating efficiency and energy consumption control. The embedding of water flow diffusion mapping weights ensures that the optimization process can quantify the differences in physical characteristics of different farmlands. Finally, a nonlinear function relationship is constructed through a BP neural network to capture the complex coupling effects and high-order interactions between these multiple physical quantities. This allows the fitness function to truly reflect the overall system performance under the multiple constraints of precise water allocation, uniform wetting, and efficient operation, thereby guiding the particle swarm algorithm to search for a gate control strategy that is truly feasible and efficient in engineering practice. The fitness function forces the simultaneous optimization of water content accuracy and diffusion uniformity through a product structure. It combines flow velocity time parameters and physical property weights to construct a comprehensive evaluation standard that can simultaneously measure water distribution accuracy, wetting uniformity, irrigation efficiency and system physical adaptability. This guides the optimization algorithm to find the optimal gate control scheme that balances multiple objectives.

[0075] The input sequence, fitness function, and constraint space of the particle swarm optimization algorithm are input into the particle swarm optimization algorithm and iteratively trained in combination with a preset training period to obtain the irrigation flow rate and irrigation start timestamp of each iteration.

[0076] Based on the irrigation flow rate and irrigation start timestamp of each farmland output in each iteration, combined with the fuzzy control algorithm, a coordinated irrigation control command is generated. Based on the coordinated irrigation control command and the irrigation simulation control platform, irrigation simulation is carried out to obtain the consistency of real-time water demand deviation, water content consistency probability and effective irrigation diffusion area ratio of all farmlands.

[0077] When at least one of the real-time water demand deviation, water content consistency probability and irrigation effective diffusion area ratio of all farmland does not meet the corresponding threshold, the iteration is repeated until all of them meet the corresponding threshold, and the irrigation flow rate and irrigation start timestamp of each farmland are obtained.

[0078] If at least one farmland still fails to meet the corresponding threshold after the iteration ends, the standard water requirement deviation of the corresponding farmland is fed back into the growth cycle water requirement model with rainfall prediction for correction, until the consistency of real-time water requirement deviation, water content consistency probability and irrigation effective diffusion area ratio of all farmland meets the corresponding threshold.

[0079] This embodiment achieves precise control across the entire chain from meteorological forecasting to field irrigation by constructing a water demand model that integrates rainfall forecasting with water flow diffusion mapping weights. By establishing a rainfall intensity-stratified effective infiltration mapping function, it overcomes the limitations of traditional effective rainfall calculation, significantly improving the accuracy of rainfall replenishment prediction and providing a reliable basis for the precise calculation of water demand deviations. Secondly, it innovatively introduces water flow diffusion mapping weights, effectively solving the problem of traditional irrigation systems neglecting the spatial uniformity of water distribution. By quantifying the coupling effects of soil properties, topographic slope, and irrigation flow velocity, it greatly improves the uniformity of field irrigation. Furthermore, it uses the probability of consistent water content and the effective infiltration of irrigation... The product of the dispersion area ratio consistency serves as the core indicator of the fitness function. Through mathematical constraints, it forces the simultaneous optimization of water distribution accuracy and wetting uniformity, ensuring the system's balance in the multi-objective optimization process. Furthermore, by optimizing neural network parameters through particle swarm optimization and combining it with the closed-loop verification mechanism of the irrigation simulation control platform, a complete intelligent decision-making closed loop is formed, significantly improving the system's response speed and adaptive capability. Finally, this scheme constructs a highly intelligent irrigation control system through multi-source data fusion, multi-physical process coupling, and multi-objective collaborative optimization. In practical applications, it demonstrates significant water-saving and energy-saving benefits, providing strong technical support for the development of smart agriculture.

[0080] Example 2

[0081] Please see Figure 3 Another embodiment of the present invention provides a multi-farmland collaborative irrigation system that integrates meteorological conditions and AI large model, including: a water demand prediction module, an optimization module, and a simulation feedback module;

[0082] The water demand prediction module is used to obtain the real-time water content of each farmland and combine it with the preset growth cycle water demand model with rainfall prediction to obtain the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands.

[0083] The optimization module is used to obtain the area of ​​each farmland, the distance between the corresponding digital water-saving gate branch and the main gate of the digital water-saving gate for each farmland, and to calculate the opening value and opening timestamp of each digital water-saving gate branch with the objective of maximizing the preset water content consistency probability. It uses the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands as constraints, combined with a BP neural network optimized by particle swarm optimization, to achieve this. Each digital water-saving gate branch corresponds one-to-one with each farmland. The water content consistency probability is used to characterize the probability that the real-time irrigation amount of each farmland meets the corresponding standard water demand deviation under the constraint of the standard total water demand deviation of all farmlands.

[0084] The simulation feedback module responds to the opening value and opening timestamp of each digital water-saving gate outlet, and performs real-time irrigation simulation in conjunction with a preset irrigation simulation control platform. It monitors the irrigation deviation value and water content consistency probability of each farmland after the irrigation simulation. When the water content consistency probability does not meet the corresponding preset threshold, the irrigation deviation value of each farmland after the simulation is fed back to the irrigation simulation control platform for simulation until the water content consistency probability meets the preset threshold.

[0085] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. A multi-farmland collaborative irrigation method integrating meteorological conditions and AI large-scale models, characterized in that, include: The real-time water content of each farmland is obtained and combined with a pre-set growth cycle water requirement model with rainfall prediction to obtain the standard water requirement deviation of each farmland and the standard total water requirement deviation of all farmlands. The growth cycle water requirement model with rainfall prediction is constructed by a BP neural network, taking the rainfall prediction input sequence as input and outputting the predicted changes in soil moisture content in different strata after rainfall and the farmland moisture content after rainfall compensation. The rainfall prediction input sequence is constructed by encoding the crop growth stage, the rainfall intensity and duration corresponding to the growth cycle, the historical rainfall prediction accuracy confidence, and the rainfall intensity-stratified effective infiltration mapping function. The rainfall intensity-stratified effective infiltration mapping function is established based on the changes in soil moisture content in different strata after rainfall and the water retention coefficient corresponding to the soil type. The standard water requirement deviation of each farmland is obtained by subtracting the predicted farmland moisture content after rainfall compensation from the standard growth water requirement of the target crop in each growth cycle. The standard total water requirement deviation of all farmlands is the sum of the standard water requirement deviations of each farmland. The following data are collected: area of ​​each farmland, distance between each digital water-saving gate branch and the main gate, soil type, average soil moisture content, slope, mapping relationship between gate opening and flow velocity, standard water requirement deviation of each farmland and standard total water requirement deviation of all farmlands. These are used to construct the input sequence for the particle swarm optimization (PSO) algorithm. The fitness function is constructed using the product of the consistency probability of moisture content and the consistency of the ratio of effective irrigation diffusion area, irrigation flow velocity and irrigation start timestamp, and a preset water flow diffusion mapping weight combined with a BP neural network. A constraint space is constructed using the standard water requirement deviation of each farmland, the standard total water requirement deviation of all farmlands, farmland area, and farmland slope. The input sequence, fitness function, and constraint space are then input into the PSO algorithm for iterative training within a preset training period to obtain the opening value and start timestamp of each digital water-saving gate branch. The water flow diffusion mapping weight is used to characterize the effective diffusion area and diffusion uniformity of irrigation water per unit time in different farmlands under the conditions of soil moisture content, topography and irrigation flow rate of each farmland, and is used to adjust the opening value and opening time of the digital water-saving gate outlets in different farmlands. Each of the digital water-saving gate outlets corresponds one-to-one with each farmland; the water content consistency probability is used to characterize the probability that the real-time irrigation amount of each farmland meets the standard water demand deviation of the corresponding farmland under the constraint of the standard total water demand deviation of all farmlands; the irrigation effective diffusion area ratio is constructed by the ratio of the effective diffusion area of ​​irrigation per unit time to the total area of ​​the corresponding farmland. In response to the opening value and opening timestamp of each digital water-saving gate, and in conjunction with a preset irrigation simulation control platform, real-time irrigation simulation is performed. Through a configured flow meter and camera, the irrigation deviation value and water content consistency probability of each farmland after the irrigation simulation are monitored. When the water content consistency probability does not meet the corresponding preset threshold, the irrigation deviation value of each farmland after the simulation is fed back to the irrigation simulation control platform for simulation until the water content consistency probability meets the preset threshold.

2. The multi-farmland collaborative irrigation method integrating meteorological conditions and AI large-scale models as described in claim 1, characterized in that, The construction and training process of the growth cycle water requirement model with rainfall prediction includes: The standard growth water requirement of each target crop in each farmland in each growth cycle is obtained, as well as the precipitation parameters of the corresponding farmland area in each growth cycle and the change value of soil moisture content in the stratified soil after rainfall. The timestamps are aligned and preprocessed. The precipitation parameters include rainfall intensity, rainfall duration and historical rainfall prediction accuracy confidence. The changes in soil moisture content in farmland after rainfall are obtained by taking the moisture content of each soil layer before rainfall and the corresponding rainfall intensity and duration, and then taking the moisture content of each soil layer after rainfall for each farmland under the corresponding soil type conditions.

3. The multi-farmland collaborative irrigation method integrating meteorological conditions and AI large-scale models as described in claim 2, characterized in that, The construction and training process of the growth cycle water requirement model with rainfall prediction also includes: Based on the changes in soil moisture content in different strata after rainfall and the soil moisture content after rainfall compensation, an output layer feature vector is constructed.

4. The multi-farmland collaborative irrigation method integrating meteorological conditions and AI large-scale models as described in claim 3, characterized in that, The construction and training process of the growth cycle water requirement model with rainfall prediction also includes: The rainfall prediction input sequence and the output layer feature vector are input into the growth cycle water requirement model with rainfall prediction constructed by the BP neural network. The model is trained by combining the preset training loss function to obtain the trained growth cycle water requirement model with rainfall prediction. The model outputs the predicted changes in soil moisture content after rainfall and the farmland moisture content after rainfall compensation. The preset training loss function is constructed from the difference between the predicted change in soil moisture content after rainfall and the actual measured change in soil moisture content after rainfall, as well as the difference between the predicted rainfall-compensated soil moisture content and the real-time measured rainfall-compensated soil moisture content. The predicted farmland moisture content after rainfall compensation is obtained by combining the farmland moisture content before rainfall with the effective infiltration corresponding to the rainfall.

5. The multi-farmland collaborative irrigation method integrating meteorological conditions and AI large-scale models as described in claim 4, characterized in that, The process of constructing the water flow diffusion mapping weights includes: The soil type, average soil moisture content, slope, irrigation flow rate, and effective irrigation diffusion area and ratio per unit time for each farmland were obtained and pre-processed and aligned.

6. The multi-farmland collaborative irrigation method integrating meteorological conditions and AI large-scale models as described in claim 5, characterized in that, The process of constructing the water flow diffusion mapping weights also includes: Using the pre-treated effective diffusion area of ​​irrigation as the dependent variable, the irrigation flow rate as the independent variable, and the soil type, average soil moisture content, and slope of each farmland as covariates, a multivariate regression algorithm was used to fit the diffusion fitting function. Based on the diffusion fitting function combined with the irrigation simulation control platform and the control variable method, with the goal of maximizing the consistency of the effective diffusion area ratio of irrigation, the irrigation flow velocity is simulated and adjusted to obtain the farmland flow velocity sequence corresponding to the maximum consistency and the regression coefficients corresponding to the diffusion fitting function, and the water flow diffusion mapping weight is constructed.

7. A multi-farmland collaborative irrigation system integrating meteorological conditions and AI large-scale models, used to implement the multi-farmland collaborative irrigation method integrating meteorological conditions and AI large-scale models as described in any one of claims 1-6, characterized in that, include: Water demand prediction module, optimization module, simulation feedback module; The water demand prediction module is used to obtain the real-time water content of each farmland and combine it with a preset growth cycle water demand model with rainfall prediction to obtain the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands. The optimization module is used to obtain the area of ​​each farmland, the distance between the corresponding digital water-saving gate branch and the total digital water-saving gate of each farmland, and with the preset goal of maximizing the consistency probability of water content, and with the standard water demand deviation of each farmland and the standard total water demand deviation of all farmlands as constraints, it calculates the opening value and opening timestamp of each digital water-saving gate branch by combining a BP neural network optimized by particle swarm optimization algorithm and preset water flow diffusion mapping weights; wherein each digital water-saving gate branch corresponds one-to-one with each farmland; the consistency probability of water content is used to characterize the probability that the real-time irrigation amount of each farmland meets the standard water demand deviation of the corresponding farmland under the constraint of the standard total water demand deviation of all farmlands. The simulation feedback module responds to the opening value and opening timestamp of each digital water-saving gate outlet, and performs real-time irrigation simulation in conjunction with a preset irrigation simulation control platform. Through a configured flow meter and camera, it monitors the irrigation deviation value and water content consistency probability of each farmland after the irrigation simulation. When the water content consistency probability does not meet the corresponding preset threshold, the irrigation deviation value of each farmland after the simulation is fed back to the irrigation simulation control platform for simulation until the water content consistency probability meets the preset threshold.

Images