Intelligent water-saving control method and system for agricultural irrigation devices

Abstract

This invention provides an intelligent water-saving control method and system for agricultural irrigation devices. The method includes the following steps: generating farmland condition prediction indicators based on comprehensive soil information, meteorological information, and crop information, and generating farmland irrigation strategies; calculating the irrigation urgency level based on the farmland condition prediction indicators; if the irrigation urgency level reaches the highest level, controlling the filter irrigation module to perform alternating irrigation operations; if the irrigation urgency level does not reach the highest level, calculating the backwash urgency level of the filter irrigation module; if the backwash urgency level is greater than or equal to the irrigation urgency level, generating a backwash strategy and executing the backwash strategy and the farmland irrigation strategy sequentially; if the backwash urgency level is less than the irrigation urgency level, controlling the filter irrigation module to perform alternating irrigation operations on the target farmland. This invention can achieve intelligent irrigation and backwash control with excellent water-saving effects.

Description

Technical Field

[0001] This invention belongs to the field of irrigation water-saving control technology, specifically relating to an intelligent water-saving control method and system for agricultural irrigation devices. Background Technology

[0002] In a global context of increasingly scarce water resources and significant challenges to agricultural production, on the one hand, population growth and climate change are exacerbating the imbalance between water supply and demand, placing increasing pressure on agriculture, a major water user, to conserve water. On the other hand, to meet the ever-growing demand for food, agricultural production must improve efficiency and yield. Against this backdrop, achieving the dual goals of water conservation and increased production has become a core issue in the development of agricultural irrigation technology.

[0003] Traditional irrigation systems generally lack intelligent and refined management capabilities. Irrigation devices often rely on fixed irrigation plans or manual experience, making it difficult to adjust flexibly according to actual conditions. This "one-size-fits-all" approach leads to over-irrigation in some areas and under-irrigation in others. Ultimately, this not only affects crop yields but also results in a serious waste of water resources. Summary of the Invention

[0004] This invention provides an intelligent water-saving control method and system for agricultural irrigation devices to solve the problem of serious water waste that easily occurs during the application of traditional irrigation devices.

[0005] In a first aspect, the present invention provides an intelligent water-saving control method for an agricultural irrigation device, which is applied to an agricultural irrigation device including a filtration irrigation module and a backwashing module. The irrigation unit of the filtration irrigation module is connected to the target farmland to be irrigated. The irrigation unit is equipped with a multi-functional irrigation outlet at one end of the target farmland. The backwashing module is used to backwash the filtration irrigation module. The method includes the following steps:

[0006] Comprehensive soil information of the target farmland is collected by a first sensor pre-installed in the target farmland;

[0007] Obtain meteorological information of the location of the target farmland and crop information within the target farmland;

[0008] Based on the comprehensive soil information, the meteorological information, and the crop information, and using a preset farmland condition prediction model, a farmland condition prediction index for the target farmland is generated. Based on the farmland condition prediction index, a farmland irrigation strategy is generated. The farmland condition prediction model is a CNN-LSTM model.

[0009] Based on the farmland condition prediction index, the irrigation urgency level corresponding to the farmland irrigation strategy is calculated using a fuzzy logic system.

[0010] If the irrigation urgency level reaches the highest level of urgency, the filter irrigation module is controlled to perform alternating irrigation operations on the target farmland according to the farmland irrigation strategy.

[0011] If the irrigation emergency level does not reach the highest level of emergency, the filter component data of the filter irrigation module is collected by a second sensor preset inside the filter irrigation module, and the historical backwash data of the backwash module is obtained.

[0012] The urgency of backwashing of the filter irrigation module is calculated by combining the filter component data, the historical backwashing data, and the farmland irrigation strategy, and using a fuzzy logic system.

[0013] If the backwashing urgency is greater than or equal to the irrigation urgency, a backwashing strategy is generated based on the filter component data, in which the backwashing module performs backwashing operations for the filter irrigation module, and the backwashing module and the filter irrigation module are controlled to execute the backwashing strategy and the farmland irrigation strategy in sequence.

[0014] If the urgency of backwashing is less than the urgency of irrigation, then the filter irrigation module is controlled to perform alternating irrigation operations on the target farmland according to the farmland irrigation strategy.

[0015] Optionally, the comprehensive soil information includes soil moisture information and soil spatial distribution information, and the crop information includes crop remote sensing information and crop type information.

[0016] Optionally, the step of generating farmland condition prediction indicators for the target farmland based on the comprehensive soil information, the meteorological information, and the crop information, and using a preset farmland condition prediction model, includes the following steps:

[0017] Preprocess the soil moisture information, the soil spatial distribution information, and the meteorological information;

[0018] The preprocessed soil moisture information and the soil spatial distribution information are fused together in the spatial dimension to form soil moisture distribution data;

[0019] The crop remote sensing information and the crop type information are fused into the soil spatial distribution information in a spatial dimension to obtain crop spatial distribution data;

[0020] The soil moisture distribution data and the crop spatial distribution data are input into a preset farmland condition prediction model, and the soil condition prediction index and crop condition prediction index are output through the convolutional neural network module in the farmland condition prediction model.

[0021] The preprocessed meteorological information is input into the farmland condition prediction model, and meteorological prediction indicators are output through the long short-term memory network module in the farmland condition prediction model.

[0022] The soil condition prediction index, the crop condition prediction index, and the meteorological prediction index are integrated into a farmland condition prediction index.

[0023] Optionally, the step of fusing the preprocessed soil moisture information and the soil spatial distribution information into soil moisture distribution data in the spatial dimension includes the following steps:

[0024] Optionally, spatial registration processing is performed on the soil moisture information and the soil spatial distribution information to place the soil moisture information and the soil spatial distribution information in the same spatial reference system;

[0025] The spatially registered soil moisture information and the spatial distribution information are resampled to a spatial grid of a preset size.

[0026] For each of the spatial grids, the soil moisture information within the spatial grids is aggregated over time to obtain aggregated soil moisture data;

[0027] Traverse all the spatial grids. For null spatial grids where no soil moisture aggregate data exists, estimate the soil moisture aggregate data of the null spatial grids based on all the soil moisture aggregate data of the adjacent spatial grids of the null spatial grids using a spatial interpolation method.

[0028] The soil moisture aggregation data of all the spatial grids are fused with the soil spatial distribution information to form soil moisture distribution data in the form of a multi-layer raster.

[0029] Optionally, the step of spatially fusing the crop remote sensing information and the crop type information into the soil spatial distribution information to obtain crop spatial distribution data includes the following steps:

[0030] The crop remote sensing information is resampled to the spatial grid.

[0031] The crop growth status within each spatial grid is calculated using vegetation indices;

[0032] The crop root distribution data within each spatial grid is estimated by combining the crop growth status and the crop type;

[0033] The crop growth status, crop type, and crop root distribution data are fused with the soil spatial distribution information to form multi-layer raster-style crop spatial distribution data.

[0034] Optionally, the step of calculating the irrigation urgency level corresponding to the farmland irrigation strategy based on the farmland condition prediction index and using a fuzzy logic system includes the following steps:

[0035] The farmland condition prediction index is used as the first fuzzy input variable of the fuzzy logic system.

[0036] Define a fuzzy set and a first membership function for each of the first fuzzy input variables;

[0037] Combining the fuzzy set and the first membership function, and using fuzzy inference methods, all the first fuzzy input variables are mapped to fuzzy output variables according to preset irrigation fuzzy inference rules;

[0038] The centroid method is used to defuzzify the fuzzy output variables to obtain the irrigation urgency level corresponding to the farmland irrigation strategy.

[0039] Optionally, the filter component data includes inlet and outlet water pressure difference, filtered water flow rate, filtered water turbidity, and cumulative filtered water volume.

[0040] Optionally, the step of combining the filter component data, the historical backwash data, and the farmland irrigation strategy, and calculating the backwash urgency of the filter irrigation module using a fuzzy logic system, includes the following steps:

[0041] The Kalman filter method is used to fuse the inlet and outlet water pressure difference, the filtered water flow rate, and the turbidity of the filtered water into real-time filtration data.

[0042] Based on the real-time filtering data, predicted filtering data is obtained by predicting the filtering state using a preset filtering state prediction model, wherein the filtering state prediction model is an LSTM model.

[0043] The predicted filtering data and the historical backwashing data are used as the second fuzzy input variables of the fuzzy logic system;

[0044] Define a second membership function for each of the second fuzzy input variables;

[0045] A backwashing association matrix is ​​established based on the second membership function and according to the preset backwashing fuzzy inference rules;

[0046] The matrix weight vector of the backwashing correlation matrix is ​​obtained by analyzing the second fuzzy input variable using the analytic hierarchy process (AHP).

[0047] By combining the backwash correlation matrix and the matrix weight vector, and through fuzzy synthesis operation, the backwash inference evaluation result is obtained;

[0048] The centroid method is used to defuzzify the backwash inference evaluation results to obtain the initial backwash urgency level.

[0049] Irrigation influencing factors are constructed based on the irrigation duration and irrigation volume in the farmland irrigation strategy, and the initial backwash urgency is corrected using the irrigation influencing factors to obtain the backwash urgency of the filter irrigation module.

[0050] In a second aspect, the present invention also provides an intelligent water-saving control system for an agricultural irrigation device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent water-saving control method for an agricultural irrigation device as described in the first aspect.

[0051] The beneficial effects of this invention are:

[0052] Compared to existing technologies, this invention achieves more precise and efficient water resource management while significantly improving the overall operational efficiency of irrigation systems. Firstly, by collecting and analyzing multi-dimensional data, including comprehensive soil information, meteorological information, and crop information, this invention provides a comprehensive and accurate basis for irrigation decisions, effectively avoiding the problem of inappropriate irrigation caused by insufficient information in traditional methods. Secondly, this invention employs an advanced CNN-LSTM model as a farmland condition prediction model, which can more accurately predict farmland conditions, thereby formulating optimal irrigation strategies. This deep learning-based approach greatly improves irrigation accuracy and effectively reduces water waste. Furthermore, this invention introduces a fuzzy logic system to assess the urgency of irrigation and backflushing. This flexible decision-making mechanism can better balance the relationship between irrigation demand and equipment maintenance, ensuring the system always operates at its best. This invention also considers the maintenance needs of the irrigation system itself, achieving coordinated optimization of irrigation and equipment maintenance by intelligently judging the necessity and urgency of backflushing. This not only extends the service life of equipment but also improves the reliability and stability of the entire irrigation system. In summary, compared with the prior art, the present invention has achieved significant improvements in water-saving effect, irrigation accuracy, system reliability and operating efficiency, providing strong technical support for modern agricultural production. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating an intelligent water-saving control method for an agricultural irrigation device in one embodiment of this application.

[0054] Figure 2 This is a schematic diagram of the system structure of an intelligent water-saving control system for an agricultural irrigation device in one embodiment of this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0056] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0057] This invention discloses an intelligent water-saving control method for agricultural irrigation devices. This method is applied to agricultural irrigation devices, with reference to... Figure 1 The agricultural irrigation system consists of a filtration irrigation module and a backwashing module. The filtration irrigation module is connected to the target farmland and equipped with a multi-functional irrigation outlet. The backwashing module is used to clean the filtration irrigation module. The system's operating principle is as follows:

[0058] First, the first sensor collects comprehensive soil information of the target farmland, including soil moisture and spatial distribution data. Simultaneously, the smart agriculture management system acquires local meteorological and crop information (including remote sensing and crop type data). The central processing unit uses a CNN-LSTM model as the farmland condition prediction model, generating farmland condition prediction indicators based on the collected information, and formulating farmland irrigation strategies accordingly. Next, the system calculates the irrigation urgency level using fuzzy logic. If the highest urgency level is reached, the system directly controls the filtering irrigation module to perform alternating irrigation operations; if not, the second sensor collects data from the filtering component and combines it with historical backwash data to calculate the backwash urgency level again using fuzzy logic. The system compares the backwash urgency level with the irrigation urgency level; if the former is greater than or equal to the latter, the backwash operation is performed first, followed by irrigation; otherwise, alternating irrigation is performed directly.

[0059] This intelligent control method fully considers soil, weather, crop conditions, and the working status of irrigation equipment. Through multi-sensor data acquisition, deep learning model prediction, and fuzzy logic decision-making, it achieves precision and intelligence in agricultural irrigation. The core of the system lies in its adaptability and predictive capabilities, enabling it to make optimal decisions based on real-time environmental changes and equipment status, ensuring the water needs of crops while maximizing water conservation. The central processing unit, acting as the system's brain, integrates a CNN-LSTM model for farmland condition prediction and an LSTM model for processing time-series data. This deep learning architecture effectively captures the complex dynamic changes in the farmland environment. The introduction of the fuzzy logic system enhances the flexibility and robustness of decision-making, handling the uncertainties and fuzziness in agricultural production. The entire system forms a closed-loop intelligent control process: from environmental perception, data analysis, and condition prediction to decision execution and feedback adjustment, achieving comprehensive intelligent management of agricultural irrigation.

[0060] Figure 2 This is a flowchart illustrating an intelligent water-saving control method for an agricultural irrigation device in one embodiment. It should be understood that, although... Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 2 As shown, the intelligent water-saving control method for agricultural irrigation devices disclosed in this invention specifically includes the following steps:

[0061] S101. Collect comprehensive soil information of the target farmland through a first sensor pre-installed in the target farmland.

[0062] The system employs a pre-deployed first-stage sensor array in the target farmland to collect comprehensive soil information, including soil moisture and spatial distribution. These sensors are high-precision, multi-functional soil monitoring devices capable of simultaneously measuring soil moisture and spatial distribution characteristics. Soil moisture information is acquired using capacitive or time-domain reflectometry sensors, accurately measuring soil water content at different depths. The sensor network covers the entire farmland in a grid pattern, with one sensor installed at each grid point, typically spaced 10-20 meters apart. This dense arrangement ensures spatial representativeness of the data. Soil spatial distribution information is obtained using geoelectric resistivity imaging (GEI), a technique that uses an electrode array to transmit a weak current underground, measuring resistivity changes to infer the spatial distribution of soil type, structure, and moisture content. The sensors collect data every 15 minutes to ensure real-time performance. The collected data is wirelessly transmitted to a central data processing unit, where it undergoes preliminary processing to form a complete comprehensive soil information dataset. These data include parameters such as soil volumetric water content, soil temperature, and electrical conductivity at different depths (e.g., 0-10cm, 10-30cm, 30-60cm), as well as a three-dimensional soil structure model reflecting the spatial heterogeneity of the soil. This step provides crucial foundational data for subsequent precision irrigation decisions, reflecting the dynamic changes and spatial differences in farmland soil moisture conditions, and effectively avoiding the irrigation imbalance problems caused by traditional single-point measurement methods.

[0063] S102. Obtain meteorological information of the target farmland location and crop information within the target farmland.

[0064] Among them, such as Figure 1 As shown, meteorological and crop information can be obtained through the smart agriculture management system. Crop information includes remote sensing data and crop type information. Meteorological information is primarily acquired using automatic weather stations located around farmland. These stations are equipped with various sensors that monitor parameters such as temperature, humidity, wind speed, rainfall, and solar radiation in real time. The weather stations record data every 5 minutes and transmit it in real time to the smart agriculture management system via GPRS or 4G networks. Simultaneously, data from mesoscale numerical weather prediction models provided by regional meteorological departments is also integrated to obtain weather forecasts for the next 3-7 days. After downscaling, this forecast data provides more accurate microclimate predictions for farmland.

[0065] Crop information acquisition combines ground-based observation and remote sensing technologies. Crop remote sensing information is primarily obtained through multispectral satellite imagery and drone aerial photography. Satellite remote sensing data (such as from the Sentinel-2 satellite) is acquired every 5-10 days, with a resolution of up to 10 meters. Drones, equipped with multispectral cameras, conduct low-altitude aerial photography weekly, achieving centimeter-level resolution. After preprocessing including atmospheric correction, geometric correction, and radiometric calibration, this remote sensing data is used to calculate vegetation indices such as the Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI), reflecting crop growth status. Crop type information is obtained through historical planting records.

[0066] By comprehensively utilizing these multi-source data, a comprehensive database of farmland environment and crop growth has been formed. Meteorological information reflects the farmland's water evaporation demand and precipitation replenishment, crop remote sensing information can capture dynamic changes in crop growth in a timely manner, and crop type information provides a basic reference for crop water requirements. The integration of this information provides comprehensive and accurate input data for subsequent farmland condition prediction and irrigation decisions, greatly improving the scientific nature and accuracy of irrigation management.

[0067] S103. Based on comprehensive soil information, meteorological information, and crop information, and using a preset farmland condition prediction model, generate farmland condition prediction indicators for the target farmland, and generate farmland irrigation strategies based on the farmland condition prediction indicators.

[0068] The farmland condition prediction model employs an advanced CNN-LSTM hybrid architecture, fully leveraging the advantages of Convolutional Neural Networks (CNNs) in spatial feature extraction and Long Short-Term Memory Networks (LSTMs) in temporal data processing. The model's input data undergoes preprocessing. Soil moisture and spatial distribution information are converted into a unified spatial grid data format using spatial interpolation and time-series aggregation methods. Crop remote sensing information and crop type information are also resampled into the same spatial grid. Meteorological data is then organized into a time-series format.

[0069] The CNN portion processes spatial data. It contains multiple convolutional and pooling layers to extract features of soil moisture distribution and crop spatial distribution. A typical CNN architecture might include three convolutional layers, each using a 3x3 kernel, followed by the ReLU activation function, and a 2x2 max-pooling layer. The output of this part of the network is a series of feature maps representing the spatial features of farmland conditions. The LSTM portion processes temporal data, primarily meteorological information and historical farmland condition data. LSTM networks typically contain 2-3 layers of LSTM units, each containing 128-256 neurons. This part of the network can capture long-term and short-term temporal dependencies, predicting future meteorological changes and their impact on farmland conditions. The outputs of the CNN and LSTM are fused through fully connected layers to ultimately output farmland condition prediction indicators. These indicators include key parameters such as predicted soil moisture content, crop transpiration, and root water uptake. The model is trained using historical data, including all monitoring data from the past few growing seasons and the corresponding actual farmland conditions. The Adam optimizer was used during training, with a learning rate of 0.001, a batch size of 64, and 200 training epochs. Dropout (ratio 0.5) and L2 regularization were employed to prevent overfitting.

[0070] Based on the generated farmland condition prediction indicators, the system further formulates farmland irrigation strategies. This process involves multi-objective optimization, considering factors such as water resource utilization efficiency, crop yield, and economic benefits. Specifically, the irrigation strategy includes irrigation time, irrigation volume, and irrigation method. For example, if the prediction indicators show that soil moisture will fall below the crop's water requirement threshold within the next three days, and there is insufficient rainfall, the system will recommend irrigation within 24 hours. The irrigation volume is calculated based on the predicted soil moisture deficit and crop water requirement, while also considering the efficiency of the irrigation system. The choice of irrigation method (such as drip irrigation, sprinkler irrigation, or flood irrigation) is based on crop type, growth stage, and soil characteristics.

[0071] This method for predicting farmland conditions and generating irrigation strategies based on a CNN-LSTM model can make full use of multi-source data to achieve accurate prediction of farmland conditions and precise assessment of irrigation needs, thereby significantly improving water resource utilization efficiency, reducing unnecessary irrigation, and ensuring that crops receive optimal water supply, ultimately achieving the goal of improving yield and quality.

[0072] S104. Based on farmland condition prediction indicators and through a fuzzy logic system, the irrigation urgency level corresponding to the farmland irrigation strategy is calculated.

[0073] Among them, first, the farmland status prediction indicators are used as the input variables of the fuzzy logic system. These indicators usually include the predicted soil moisture content, crop water requirement, expected rainfall, etc. Corresponding fuzzy sets and membership functions are defined for each input variable. For example, for the soil moisture content, three fuzzy sets of "dry", "moderate", and "wet" can be defined. The membership functions adopt common triangular or trapezoidal functions.

[0074] Taking the soil moisture content as an example, its membership function may be defined as follows:

[0075] "Dry": μ(x) = 1 when x ≤ 15%; μ(x) = (25% - x) / 10% when 15% < x < 25%; μ(x) = 0 when x ≥ 25%

[0076] "Moderate": μ(x) = (x - 15%) / 10% when 15% < x ≤ 25%; μ(x) = (35% - x) / 10% when 25% < x < 35%; μ(x) = 0 otherwise

[0077] "Wet": μ(x) = 0 when x ≤ 25%; μ(x) = (x - 25%) / 10% when 25% < x < 35%; μ(x) = 1 when x ≥ 35%

[0078] Among them, x represents the percentage of the soil moisture content, and μ(x) represents the membership degree.

[0079] Next, a series of fuzzy rules are formulated based on expert knowledge and historical experience. These rules describe the irrigation urgency levels under different combinations of input variables. For example: If the soil moisture content is "dry" and the crop water requirement is "high", then the irrigation urgency level is "very high"; If the soil moisture content is "moderate" and the expected rainfall is "heavy", then the irrigation urgency level is "low". Then, using fuzzy inference methods (such as the Mamdani method), the input variables are mapped to the output variable (irrigation urgency level) through these rules. This process includes four steps: fuzzification, rule evaluation, rule aggregation, and defuzzification. Finally, the centroid method is used for defuzzification to obtain a definite numerical value of the irrigation urgency level.

[0080] S105. If the irrigation urgency level reaches the highest level of urgency, then control the filtration irrigation module to perform alternate irrigation operations on the target farmland according to the farmland irrigation strategy.

[0081] The system determines the current irrigation urgency level based on a pre-set threshold. The highest urgency level typically indicates a severe risk of crop water shortage, requiring immediate irrigation. This threshold is set based on crop type, growth stage, and local climate conditions, usually determined using historical data and experience. Once the highest urgency level is confirmed, the filtration irrigation module is immediately activated. This module is an integrated system combining a water source, filtration system, water supply network, and irrigation equipment. It first checks the water source to ensure sufficient water supply. Then, the water is treated by the filtration system to remove impurities and suspended solids that could clog the irrigation equipment. The filtration system typically includes multi-stage filtration devices such as sand filters, mesh filters, and disc filters, effectively removing impurities of different sizes and types.

[0082] In this embodiment, the farmland irrigation strategy is actually an alternating irrigation strategy. Alternating irrigation is an innovative irrigation method. When implementing the irrigation strategy, the target farmland is first divided into multiple irrigation zones based on soil spatial distribution information and crop distribution. Then, these zones are irrigated sequentially according to a predetermined order. For example, in a farmland divided into four zones, the first zone is irrigated for 15 minutes, then the second zone is irrigated for 15 minutes, and so on. After completing one round of irrigation for all zones, a new round of alternating irrigation begins until the total irrigation time or water volume specified in the farmland irrigation strategy is reached.

[0083] This alternating irrigation method has several significant advantages: First, it allows for a more even distribution of water in the soil. In traditional full irrigation, water may accumulate excessively in some areas while other areas receive insufficient moisture. Alternating irrigation, by giving each area intermittent "rest" time, allows water to penetrate and distribute better in the soil. Second, alternating irrigation can reduce surface runoff and deep seepage, thereby improving water use efficiency. Because each area is irrigated for a shorter period, water is more easily absorbed by the soil rather than runoff or seep into groundwater. This method also improves soil aeration. During the irrigation intervals, the soil has the opportunity to expel excess water, allowing air to enter the soil pores and providing a better growing environment for crop roots.

[0084] S106. If the irrigation emergency level does not reach the highest level of emergency, the filter component data of the filter irrigation module is collected by the second sensor preset inside the filter irrigation module, and the historical backwash data of the backwash module is obtained.

[0085] If the irrigation emergency level does not reach the highest level, a second sensor network pre-installed within the filter irrigation module is activated. These sensors are specifically designed to monitor various parameters of the filter irrigation module, including but not limited to:

[0086] Inlet and outlet water pressure difference: Measured by pressure sensors installed at the filter inlet and outlet. This parameter reflects the degree of filter clogging.

[0087] Filtered water flow rate: Measured using an electromagnetic flow meter or ultrasonic flow meter. This reflects the actual water supply capacity of the irrigation system.

[0088] Turbidity of the filtered water: Measured using an online turbidity meter. This directly reflects the filtration effect.

[0089] Cumulative filtration volume: Calculated using integral flow rate. This data is used to assess the lifespan of the filter media.

[0090] These sensors typically employ high-precision, waterproof, and dustproof industrial-grade equipment, enabling them to operate stably for extended periods in harsh environments. Data acquisition frequency is usually set to once per minute to capture subtle changes in system status.

[0091] Simultaneously, the system retrieves historical backwashing data from the database, including the time of the last backwash, the duration of the backwash, the amount of water used for backwashing, the pressure difference change before and after backwashing, and the backwashing effect (such as the degree of improvement in filtration efficiency after backwashing). This historical data is typically stored for the most recent 3-6 months for long-term trend analysis.

[0092] The data acquisition process employs a distributed architecture, with each sensor node equipped with a local processing unit capable of preliminary data cleaning and anomaly detection. For example, if a sensor reading suddenly fluctuates abnormally, the local processing unit immediately marks this data point as suspicious and increases the sampling frequency to confirm whether it's an equipment malfunction or a genuine system anomaly. All acquired data is transmitted in real-time to the central control system via industrial Ethernet or wireless transmission (such as LoRa, NB-IoT, and other low-power wide-area network technologies). Data transmission uses encryption protocols to ensure data security and integrity. Upon receiving the data, the central control system immediately performs data fusion and preliminary analysis. For example, it compares the current inlet and outlet water pressure difference with historical data to assess the filter's clogging trend; it compares the turbidity of the filtered water with irrigation water quality standards to determine whether the current filtration effect meets requirements. This real-time acquisition of filter component data and historical backwash data provides crucial information for subsequent decision-making. It is used not only to assess the current operating status of the irrigation system but also to predict potential future problems. For example, if the pressure difference between the inlet and outlet water is found to be gradually increasing, but has not yet reached the backwash threshold, the system may schedule a preventative backwash in advance to avoid a decrease in filtration efficiency during critical irrigation periods.

[0093] S107. Combining filter component data, historical backwashing data, and farmland irrigation strategies, the urgency of backwashing for the filter irrigation module is calculated using a fuzzy logic system.

[0094] This involves preprocessing and fusing the collected real-time filtered data. Specifically, this includes the following steps:

[0095] Data cleaning: removing outliers and noise. For example, using median filtering to remove short-term spikes.

[0096] Data standardization: Transforming data of different dimensions (such as pressure, flow rate, turbidity) to a unified standard scale, usually using the min-max standardization method.

[0097] Data fusion: The Kalman filter algorithm is used to fuse parameters such as inlet and outlet water pressure difference, filtered water flow rate, and filtered water turbidity into a comprehensive real-time filtration status indicator. Kalman filtering effectively handles measurement noise and system errors, providing optimal estimates.

[0098] Next, a Long Short-Term Memory (LSTM) network model is used to predict the future filter status. The input to the LSTM model includes a time series of real-time filter status indicators from the past 24 hours, relevant features from historical backwash data (such as runtime since the last backwash, cumulative filtered water volume, etc.), and the projected irrigation demand for the next 24 hours (extracted from farmland irrigation strategies). The output of the LSTM model is the predicted filter status indicators for the next 24 hours. This prediction considers the filter performance degradation trend over time and the potential impact of future irrigation demand on the filter.

[0099] The predicted filter status indicators and historical backwash data are then used as input variables for the fuzzy logic system. A fuzzy set and membership function are defined for each input variable. For example, for the predicted inlet and outlet pressure difference, four fuzzy sets can be defined: "normal," "slightly blocked," "moderately blocked," and "severely blocked." The membership function can be a Gaussian function or a trapezoidal function. Backwash fuzzy inference rules are established based on these fuzzy sets and membership functions. These rules are derived from expert knowledge and historical data analysis. For example, if the pressure difference is "severely blocked" and the time since the last backwash is "long," then the backwash urgency is "very high"; if the pressure difference is "slightly blocked" and the predicted irrigation demand for the next 24 hours is "high," then the backwash urgency is "moderate."

[0100] Next, the Analytic Hierarchy Process (AHP) is used to determine the weights of each input variable in the decision-making process. AHP obtains the weights of each factor by constructing a judgment matrix and calculating eigenvectors. This step considers the relative importance of different parameters to the backwashing decision. Then, combining fuzzy inference rules and parameter weights, fuzzy synthesis operations are used to obtain the backwashing inference evaluation result. Commonly used fuzzy synthesis algorithms include the max-min synthesis method and the weighted average method. The centroid method is used to defuzzify the fuzzy evaluation result, obtaining a clear initial backwashing urgency value. After obtaining the initial backwashing urgency, the impact of irrigation plans in farmland irrigation strategies on the backwashing decision is considered, and an irrigation influence factor is constructed. The final backwashing urgency is obtained by multiplying the initial urgency by the irrigation influence factor.

[0101] S108. If the urgency level of backwashing is greater than or equal to the urgency level of irrigation, a backwashing strategy is generated based on the filter component data to enable the backwashing module to perform backwashing operations for the filter irrigation module, and the backwashing module and the filter irrigation module are controlled to execute the backwashing strategy and the farmland irrigation strategy in sequence.

[0102] First, a detailed backwashing strategy is generated based on the data from the filtering components. This strategy includes the following key parameters:

[0103] Backflushing time: Determined based on the degree of clogging, typically between 30 seconds and 3 minutes.

[0104] Backwash flow rate: Usually set to 2-3 times the normal filtration flow rate to ensure effective removal of accumulated impurities.

[0105] Backwash pressure: typically 20-30% higher than normal filtration pressure to provide sufficient impact force.

[0106] Backwash sequence: For a multi-tank parallel filtration system, determine the backwash sequence for each filter.

[0107] The generation of the backwashing strategy takes into account several factors:

[0108] Current degree of blockage: determined by the pressure difference between the inlet and outlet water.

[0109] Water quality: Adjusted based on raw water turbidity and filtered water quality.

[0110] Historical backwashing effectiveness: Analyzing the relationship between the duration and effectiveness of past backwashing.

[0111] Energy efficiency: Select the optimal combination of backwashing parameters to achieve a balance between cleaning effectiveness and energy consumption.

[0112] After the strategy is generated, the system will control the backwashing module to perform the backwashing operation. The specific steps are as follows:

[0113] Shut down the normal filtration water circuit and open the backwash water circuit. This is usually achieved via an electric or hydraulic valve, ensuring a quick and reliable switchover.

[0114] Start the backwash pump to force clean water through the filter at high pressure. Backwash pumps typically use variable frequency control, allowing for precise adjustment of flow rate and pressure.

[0115] Monitor various parameters during the backwashing process, including pressure, flow rate, and turbidity. Compare these parameters with expected values ​​in real time, and immediately adjust the backwashing strategy or issue an alarm if any abnormality is detected (such as a sudden drop in pressure).

[0116] The backwashing process will automatically end based on the preset backwashing time or water quality indicators (such as the effluent turbidity being below a certain threshold).

[0117] After backwashing, a brief forward flush will be performed to stabilize the filter media and remove residual impurities.

[0118] Once backwashing is complete, the system will immediately switch to the previously established farmland irrigation strategy.

[0119] S109. If the urgency of backwashing is less than that of irrigation, the control filter irrigation module shall perform alternating irrigation operation on the target farmland in accordance with the farmland irrigation strategy.

[0120] In one embodiment, generating farmland condition prediction indicators for the target farmland based on comprehensive soil information, meteorological information, and crop information, and using a preset farmland condition prediction model, includes the following steps:

[0121] Preprocessing soil moisture information, soil spatial distribution information, and meteorological information;

[0122] The preprocessed soil moisture information and soil spatial distribution information are fused into soil moisture distribution data in the spatial dimension.

[0123] By integrating crop remote sensing information and crop type information into soil spatial distribution information in a spatial dimension, crop spatial distribution data is obtained.

[0124] Soil moisture distribution data and crop spatial distribution data are input into a preset farmland condition prediction model, and soil condition prediction indicators and crop condition prediction indicators are output through the convolutional neural network module in the farmland condition prediction model.

[0125] The preprocessed meteorological information is input into the farmland condition prediction model, and the meteorological prediction index is output through the long short-term memory network module in the farmland condition prediction model.

[0126] Soil condition prediction indicators, crop condition prediction indicators, and meteorological prediction indicators are integrated into farmland condition prediction indicators.

[0127] In this implementation, the preprocessing steps aim to clean, standardize, and enhance the raw data, preparing it for subsequent data fusion and model input. For soil moisture information, outlier detection and processing are first performed, using the Median Absolute Deviation (MAD) method to identify and remove outliers. For example, if a sensor reading deviates from the median by more than three times the MAD, it is considered an outlier. Next, time-series interpolation is performed to fill any potential data gaps; common methods are linear interpolation or spline interpolation. Then, a moving average filter is applied to smooth the data and reduce the impact of short-term fluctuations. For soil spatial distribution information, spatial registration is first performed to ensure all data points are in the same coordinate system. Then, Kriging interpolation is used to spatially interpolate discrete sampling points, generating a continuous soil property distribution map. For meteorological information, unit unification and scale transformation are first performed, such as converting temperature to degrees Celsius and wind speed to meters per second. Then, seasonal adjustment and trend decomposition are performed on the time-series data, using the STL (Seasonal and Trend decomposition using Loess) method to separate seasonal, trend, and residual components. Finally, all preprocessed data is standardized, typically using Z-score standardization, to make different types of data comparable. Through these preprocessing steps, the original heterogeneous data is transformed into a clear, consistent, and information-rich dataset, laying a solid foundation for subsequent analysis and modeling.

[0128] Next, both types of data are resampled into a predefined unified spatial grid, typically set to 5m × 5m or 10m × 10m to balance spatial resolution and computational efficiency. For each grid cell, the time-series aggregated values ​​of all soil moisture measurements within it are calculated, such as the mean, median, or weighted average. Weights can be based on the distance from the measurement point to the grid center using an inverse distance weighting method. For grid cells without directly measured data (null grids), geostatistical methods are used for spatial interpolation. Specifically, ordinary kriging is used, which considers spatial autocorrelation and provides the best linear unbiased estimate. The semivariogram model for kriging interpolation is chosen from spherical, exponential, or Gaussian models, determined based on the spatial structure characteristics of the actual data. Finally, the interpolated continuous soil moisture field is fused with the original soil spatial distribution information (such as soil type, texture, organic matter content, etc.) to generate multi-layer raster data. Each layer represents different soil properties or moisture conditions at different time points. This fusion method not only provides high-resolution spatial distribution maps of soil moisture, but also preserves information on the physical properties of the soil itself, providing rich and accurate input data for subsequent farmland condition prediction.

[0129] For remote sensing information, atmospheric and geometric corrections are required to eliminate the influence of atmosphere and topography on the remote sensing imagery. Then, the corrected remote sensing imagery is resampled into a spatial grid with the same soil spatial distribution information, typically using bilinear interpolation or cubic convolution interpolation. For each spatial grid, vegetation indices such as the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) are calculated. These indices effectively reflect crop growth status and biomass. For example, the formula for calculating NDVI is: NDVI = (NIR - RED) / (NIR + RED), where NIR and RED represent the reflectance in the near-infrared and red light bands, respectively. Next, combined with crop type information, specific crop type labels are assigned to each spatial grid. If a grid contains multiple crops, an area-weighted approach is used to determine the dominant crop type. Based on crop type and growth stage, the crop root distribution in each grid is estimated. Root distribution can be described using an exponential decay model, such as R(z) = 1 - βz, where R(z) is the root density at depth z, and β is the decay coefficient, which is related to crop type and growth stage. Finally, crop growth status (such as NDVI value), crop type, and root distribution information are fused with soil spatial distribution information to generate multidimensional raster data. Each raster cell contains multiple attribute layers, including soil properties, crop type, growth status, and root distribution. This fused crop spatial distribution data not only reflects the aboveground condition of the crop but also includes information on the underground parts, providing comprehensive data support for precision irrigation decisions.

[0130] Soil moisture distribution data and crop spatial distribution data are organized into a four-dimensional tensor (batch size, number of channels, height, width), with each channel representing an attribute layer (such as soil moisture, soil type, NDVI, crop type, etc.). The CNN module architecture typically includes multiple convolutional layers, pooling layers, and fully connected layers. For example, a structure similar to the VGG network can be used, containing 5 convolutional blocks, each containing 2-3 3x3 convolutional layers and a 2x2 max-pooling layer. The ReLU activation function is chosen to introduce non-linearity and alleviate the gradient vanishing problem. After the last convolutional block, a global average pooling layer is used to reduce the number of parameters, followed by a fully connected layer. The output of the fully connected layer is divided into two parts: soil state prediction indicators and crop state prediction indicators. Soil state prediction indicators may include soil moisture content and soil temperature for the next few days, while crop state prediction indicators may include leaf area index and biomass. Model training uses mini-batch stochastic gradient descent, with the loss function being a weighted sum of mean squared error (MSE) and mean absolute error (MAE).

[0131] The preprocessed meteorological data is organized into a three-dimensional tensor (batch size, time step, number of features). Each time step contains multiple meteorological features, such as temperature, humidity, rainfall, and wind speed. The LSTM module architecture typically consists of multiple LSTM units and one output layer. For example, a two-layer LSTM can be used, each containing 128 hidden units, followed by a fully connected layer. The core of the LSTM unit is its gating mechanism, including the forget gate, input gate, and output gate. The forget gate determines which information to discard, the input gate determines which information to update, and the output gate determines which information to output. Model training uses the backpropagation through-time (BPTT) algorithm, with the Adam optimizer used for parameter updates. The loss function is the mean squared error (MSE), and an L2 regularization term is introduced to prevent overfitting. The model output is a meteorological forecast for the next few days, including key parameters such as daily average temperature, sunshine hours, and rainfall.

[0132] Integrating soil condition prediction indicators, crop condition prediction indicators, and meteorological prediction indicators into a single farmland condition prediction indicator is a comprehensive data fusion process. First, the three types of prediction indicators are standardized to ensure comparability across different scales and units. Then, Principal Component Analysis (PCA) is used to reduce the dimensionality of each type of indicator and extract key features. Principal components with a cumulative explained variance of 85% or higher are selected as the new feature set. Next, a weighted summation method is used to combine the different categories of principal components into a unified farmland condition prediction indicator. The weights are determined using the Analytic Hierarchy Process (AHP), considering expert opinions and historical data analysis results. For example, a possible weight allocation is: soil condition 40%, crop condition 35%, and meteorological condition 25%. The final formula for calculating the farmland condition prediction indicator is: , where wi is the weight of the i-th principal component and PCi is the score of the i-th principal component. For ease of interpretation and application, the comprehensive index is normalized to a range of 0-100, where 0 represents the worst state and 100 represents the best state. Furthermore, warning thresholds are set based on the index values, such as defining 0-20 as a severe warning, 20-40 as a moderate warning, 40-60 as a minor warning, 60-80 as good, and 80-100 as excellent.

[0133] In one embodiment, fusing the preprocessed soil moisture information and soil spatial distribution information into soil moisture distribution data in the spatial dimension includes the following steps:

[0134] Spatial registration processing is performed on soil moisture information and soil spatial distribution information to place them in the same spatial reference system;

[0135] The spatially registered soil moisture information and soil spatial distribution information are resampled to a spatial grid of a preset size.

[0136] For each spatial grid, all soil moisture information within the spatial grid is aggregated over time to obtain aggregated soil moisture data;

[0137] Traverse all spatial grids. For null spatial grids that do not have soil moisture aggregation data, calculate the soil moisture aggregation data of all adjacent spatial grids of the null spatial grids based on the soil moisture aggregation data of the null spatial grids, and use spatial interpolation methods to estimate the soil moisture aggregation data of the null spatial grids.

[0138] Soil moisture aggregation data from all spatial grids is fused with soil spatial distribution information into a multi-layer raster form of soil moisture distribution data.

[0139] In this embodiment, a unified target coordinate system is determined, typically the Universal Transverse Mercator (UTM) projection or a locally used projection system. For soil moisture information, this data usually comes from sensors distributed throughout the farmland, each with its own specific geographic coordinates. These coordinates are transformed from their original system (e.g., latitude and longitude) to the target system. The transformation process uses coordinate transformation formulas. For soil spatial distribution information, this data may come from soil surveys or remote sensing imagery and may already be in a specific projection system. If this system differs from the target system, reprojection is required. The reprojection process involves coordinate transformation and resampling, using methods such as bilinear interpolation or cubic convolution to ensure spatial continuity of the data. During coordinate transformation, the effects of different ellipsoids and reference surfaces must also be considered, and appropriate transformation parameters (e.g., seven-parameter or Helmert transformation) are used for adjustment. After registration, the two datasets are spatially perfectly aligned, with each data point having the same coordinate system and projection.

[0140] The spatially registered soil moisture and spatial distribution information are resampled to a pre-defined spatial grid. Specifically, the appropriate grid size is first determined, typically based on the area of ​​the study region, the original resolution of the data, and the required precision. For example, for a 100-hectare farmland, a 10m x 10m grid size might be chosen, dividing the entire area into 10,000 grid cells. Different methods are used for resampling soil moisture and spatial distribution information. For soil moisture information, since this data is typically discrete point data, interpolation methods are used to convert it into continuous raster data. For soil spatial distribution information, if the original data is in vector format (e.g., polygons), it needs to be converted to raster format first. During the conversion, each grid cell is assigned the attribute value of the soil type with the largest coverage area. If the original data is already in raster format but has a different resolution, resampling algorithms such as nearest neighbor, bilinear interpolation, or cubic convolution are used for resolution adjustment. Nearest neighbor is suitable for categorical data, while bilinear interpolation and cubic convolution are suitable for continuous data. After resampling, both datasets were converted into grids of the same size and resolution, with each grid cell containing soil moisture values ​​and soil property information.

[0141] For each grid, all measurements from soil moisture sensors within that grid area over a specific period (e.g., one month or one growing season) are collected to form a time-series dataset. Then, statistical analysis and feature extraction are performed on this time series. Common aggregation methods include calculating statistics such as the mean, median, maximum, minimum, and standard deviation. Additionally, time-series features such as autocorrelation coefficients and partial autocorrelation coefficients can be calculated to capture the time dependence of soil moisture. For grids with multiple sensors, the above processing is first performed on the data from each sensor, and then the results are spatially averaged or weighted averaged, with weights based on the distance from the sensor to the grid center. Finally, each grid yields a set of aggregated indicators, such as average humidity, humidity coefficient of variation, and humidity trend.

[0142] Next, all spatial grids are traversed to identify all null grids, which may be missing data due to sensor malfunction, installation location limitations, or data transmission issues. Then, for each null grid, its neighboring grids are determined. Neighboring grids typically include the eight directly adjacent grids (Moore's neighborhood), but in some cases this may be extended to a larger range, such as 16 or 24 nearest grids. A suitable spatial interpolation method is then selected. Commonly used methods include Inverse Distance Weighted (IDW), Ordinary Kriging, and Radial Basis Function (RBF). For each null grid, the selected interpolation method is applied, using aggregated soil moisture data from neighboring grids to estimate its value. During interpolation, considering auxiliary variables such as topography and soil type can improve estimation accuracy. For example, co-kriging can be used to incorporate these auxiliary variables into the interpolation model. After interpolation, the estimation results are checked for reasonableness to ensure that the estimated values ​​are within a reasonable range and do not contain obvious outliers. Cross-validation methods, such as Leave-One-out Cross-Validation (LOOCV), can be used to evaluate interpolation accuracy. This method successfully filled in all the missing grid cells, creating a complete, spatially continuous soil moisture distribution map.

[0143] In the final and crucial data integration step, the spatial resolution and extent of the final raster data must first be determined, typically maintaining consistency with the spatial grid used in previous steps. Then, a multidimensional data structure containing multiple data layers is created for each grid cell. The first layer is aggregated soil moisture data, including indicators such as average moisture, coefficient of variation, and moisture variation trends. Subsequent layers include soil spatial distribution information, such as soil type, soil texture, organic matter content, and pH value. Each attribute serves as an independent data layer. During data fusion, these layers are overlaid using Geographic Information System (GIS) software or custom algorithms. For example, raster algebra operations can be used to combine data from different layers into a multidimensional array, with each grid cell containing a complete set of attribute values. During fusion, data type consistency is crucial. Continuous variables (such as moisture values) and categorical variables (such as soil type) need to be processed separately. For continuous variables, standardization may be necessary. For categorical variables, encoding methods such as one-hot encoding are typically used to convert them into numerical form. The fused multi-layer raster data not only contains spatial information but also temporal information (through the time-series characteristics of soil moisture). This data structure allows for complex spatial analysis and spatiotemporal pattern recognition. For example, it can easily extract moisture variation trends under specific soil types or analyze the impact of different soil textures on moisture distribution.

[0144] In one implementation, the process of spatially fusing crop remote sensing information and crop type information into soil spatial distribution information to obtain crop spatial distribution data includes the following steps:

[0145] Resample crop remote sensing information to a spatial grid;

[0146] The crop growth status within each spatial grid is calculated using vegetation indices.

[0147] Estimate crop root distribution data within each spatial grid by combining crop growth status and crop type;

[0148] The crop growth status, crop type, and crop root distribution data are integrated with soil spatial distribution information to form multi-layer raster-style crop spatial distribution data.

[0149] In this embodiment, resampling crop remote sensing information to a spatial grid is a crucial data preprocessing step, which first involves the acquisition and preliminary processing of remote sensing imagery. Remote sensing imagery typically originates from satellite or UAV platforms and includes multiple spectral bands, such as red, near-infrared, and shortwave infrared. These raw images first require geometric and atmospheric correction. Geometric correction ensures that each pixel of the image corresponds to the correct geographical location, usually achieved using ground control points and an accurate digital elevation model (DEM). Atmospheric correction eliminates the influence of atmospheric scattering and absorption on the remote sensing signal, using atmospheric radiative transfer models such as MODTRAN or 6S. The corrected imagery typically has high spatial resolution, potentially reaching sub-meter levels.

[0150] Next, resampling algorithms are used to convert the corrected high-resolution remote sensing imagery into a spatial grid. Commonly used resampling methods include nearest neighbor, bilinear interpolation, and cubic convolution. Nearest neighbor simply assigns the source pixel value closest to the center of the target pixel to the target pixel, suitable for classification data. Bilinear interpolation uses a distance-weighted average to calculate the new pixel value, suitable for continuous data and maintaining a certain degree of spatial smoothness. Cubic convolution uses a weighted average of the 16 nearest neighbor pixels, providing smooth results while maintaining edge sharpness. After resampling, each spatial grid contains reflectance values ​​for multiple spectral bands. This data is organized into a multidimensional array, where each element corresponds to a grid and contains all the spectral information for that grid.

[0151] Calculating crop growth status within each spatial grid using vegetation indices is a crucial step in transforming remote sensing data into meaningful crop physiological information. Vegetation indices are mathematical combinations of reflectance across different spectral bands, effectively reflecting the biophysical characteristics of vegetation, such as leaf area index, biomass, and chlorophyll content. This process first involves selecting an appropriate vegetation indice. The most commonly used is the Normalized Difference Vegetation Index (NDVI), calculated as: NDVI = (NIR - RED) / (NIR + RED), where NIR is the reflectance in the near-infrared band and RED is the reflectance in the red band. NDVI values ​​range from -1 to 1, with higher positive values ​​indicating healthier, more lush vegetation. For each spatial grid, the selected vegetation index is calculated using resampled multispectral data. This process can be efficiently performed using raster algebra operations. For example, when using NDVI, the following steps are performed for each grid: extract the near-infrared and red band reflectance values ​​for that grid; apply the NDVI formula to calculate the NDVI value for that grid; and store the result as a new attribute for that grid.

[0152] Next, based on crop type and growth status, root distribution is estimated using a root growth model. A commonly used model is the exponential decay model, whose basic form is: Here, R(z) is the relative root density at depth z, and β is the attenuation coefficient. The value of β varies depending on the crop type and growth stage. For example, shallow-rooted crops (such as wheat) have a larger β value, while deep-rooted crops (such as corn) have a smaller β value. To make the model more accurate, the influence of crop growth status is introduced. Thus, crops with better growth status (higher NDVI values) will have deeper root distributions. For each spatial grid, the root density distribution at different depths is calculated according to the above model. Typically, root density values ​​are calculated at 10cm intervals within the depth range of 0-100cm. In this way, each grid will have a 10-dimensional root density vector. Finally, the estimated root distribution data is integrated with the original spatial grid data structure. In addition to containing information about the aboveground parts (such as NDVI and crop type), each grid now also contains a vector describing the vertical distribution of the roots.

[0153] Finally, a multidimensional data structure is created, where each spatial grid corresponds to a data cell and contains multiple attribute layers. These layers include:

[0154] 1. Crop growth status layer: This layer contains multiple vegetation indices, such as NDVI and EVI. Each index is treated as an independent sublayer.

[0155] 2. Crop Type Layer: Different crop types are represented using encoding methods. For example, integer encoding (1=corn, 2=soybean, 3=wheat, etc.) or one-hot encoding (creating multiple binary layers, one for each crop) can be used.

[0156] 3. Crop root distribution layer: contains root density estimates at multiple depths, with each depth as an independent sublayer.

[0157] 4. Soil spatial distribution layer: contains multiple soil properties, such as texture, organic matter content, pH value, etc., with each property as an independent sublayer.

[0158] Data fusion is achieved using raster algebra and Geographic Information System (GIS) tools. During the fusion process, it is crucial to maintain data consistency. Continuous variables (such as NDVI and root density) and categorical variables (such as crop type and soil type) need to be processed separately. For continuous variables, standardization is performed; for categorical variables, encoding methods are used to convert them into numerical forms. For example, for crop types, the following methods can be used:

[0159] 1. Integer encoding: Directly use integers to represent different types (1=corn, 2=soybeans, etc.)

[0160] 2. One-hot encoding: Create a binary layer for each crop type.

[0161] The merged multi-layer raster data structure can be represented as:

[0162]

[0163] Where D(i,j) is the data vector at position (i,j), and Vk(i,j) is the value of the k-th attribute layer.

[0164] In one implementation, the process of calculating the irrigation urgency level corresponding to a farmland irrigation strategy based on farmland condition prediction indicators and using a fuzzy logic system includes the following steps:

[0165] The farmland condition prediction index is used as the first fuzzy input variable of the fuzzy logic system.

[0166] Define a fuzzy set and a first membership function for each first fuzzy input variable;

[0167] By combining fuzzy sets and the first membership function, and using fuzzy inference methods, all first fuzzy input variables are mapped to fuzzy output variables according to preset irrigation fuzzy inference rules;

[0168] The centroid method is used to defuzzify the fuzzy output variables to obtain the irrigation urgency level corresponding to the farmland irrigation strategy.

[0169] In this embodiment, the farmland condition prediction indicators include multiple dimensions such as predicted soil moisture, predicted crop growth status, and meteorological forecasts. Each indicator is treated as an independent fuzzy input variable. For example, predicted soil moisture may range from 0% to 100%, crop growth status is represented by the Normalized Difference Vegetation Index (NDVI), ranging from -1 to 1, and meteorological forecasts include the probability of rainfall and evapotranspiration in the coming days. These precise values ​​need to be converted into linguistic variables, such as "low," "medium," and "high," for processing by the fuzzy logic system. This conversion enables the system to simulate the decision-making process of human experts, handling the inherent uncertainty and fuzziness in agricultural production.

[0170] Next, we need to define the fuzzy sets and the first membership function. Taking soil moisture prediction as an example, we can define three fuzzy sets: "dry," "moderate," and "moist." The membership function determines the degree to which a specific soil moisture value belongs to these fuzzy sets. Commonly used membership functions include the trigonometric function, the trapezoidal function, and the Gaussian function. For example, the membership function for "dry" using the trapezoidal function can be expressed as:

[0171]

[0172] Here, x represents the soil moisture value. Similarly, corresponding fuzzy sets and membership functions are defined for crop growth status and weather forecast values. This definition method allows input variables to smoothly transition between different fuzzy sets, better reflecting the fuzziness in reality. Through carefully designed fuzzy sets and membership functions, the system can more accurately capture subtle changes in farmland conditions, providing richer input information for subsequent fuzzy inference. Combining fuzzy sets and the first membership function, and using fuzzy inference methods to map all first fuzzy input variables to fuzzy output variables according to preset irrigation fuzzy inference rules, is the core inference process of the fuzzy logic system. For example, a rule might be: "If the soil moisture is dry, the crop growth status is good, and there is no future rainfall forecast, then the urgency of irrigation is high." These rules are represented using an IF-THEN structure, covering various possible combinations of inputs.

[0173] Next, these rules are evaluated using fuzzy reasoning methods (such as Mamdani reasoning). For each rule:

[0174] 1. Calculate the truth degree of the antecedent (IF part) using the membership function of the fuzzy set.

[0175] 2. Use fuzzy implication operators (such as the minimum value method) to determine the activation strength of the rule.

[0176] 3. Truncate the fuzzy set of the consequent (THEN part) according to the activation strength.

[0177] The outputs of all rules are aggregated, typically using the maximum value method. This process produces a comprehensive fuzzy output set representing the irrigation urgency. In this way, the system can comprehensively consider multiple factors, simulate the decision-making process of human experts, and derive reasonable irrigation recommendations. Finally, the centroid method is used to defuzzify the fuzzy output variables, obtaining the irrigation urgency corresponding to the farmland irrigation strategy. The centroid method is a commonly used and effective defuzzification method. It calculates the centroid of the fuzzy output set and uses it as the final crisp output value. Through this method, the fuzzy irrigation urgency is converted into a specific numerical value, such as a scale from 0 to 100, where 0 represents no irrigation needed at all, and 100 represents an extremely urgent need for irrigation.

[0178] In one embodiment, the filter component data includes the inlet and outlet water pressure difference, filtered water flow rate, filtered water turbidity, and cumulative filtered water volume. Combining the filter component data, historical backwashing data, and farmland irrigation strategies, and calculating the backwashing urgency of the filtration irrigation module using a fuzzy logic system includes the following steps:

[0179] The Kalman filter method is used to integrate the inlet and outlet water pressure difference, the filtered water flow rate, and the turbidity of the filtered water into real-time filtration data.

[0180] Predicted filter data is obtained based on real-time filtered data and a preset filter state prediction model, which is an LSTM model.

[0181] The predictive filtering data and historical backwashing data are used as the second fuzzy input variables of the fuzzy logic system.

[0182] Define a second membership function for each second fuzzy input variable;

[0183] A backwashing association matrix is ​​established based on the second membership function and according to the preset backwashing fuzzy inference rules;

[0184] The matrix weight vector of the backwashing correlation matrix is ​​obtained by analyzing the second fuzzy input variable using the analytic hierarchy process.

[0185] The backwashing inference and evaluation results are obtained by combining the backwashing correlation matrix and matrix weight vector and performing fuzzy synthesis operations.

[0186] The centroid method is used to defuzzify the backwash inference evaluation results to obtain the initial backwash urgency level.

[0187] Irrigation influencing factors are constructed based on irrigation duration and irrigation volume in farmland irrigation strategies, and the initial backwash urgency is corrected using irrigation influencing factors to obtain the backwash urgency of the filter irrigation module.

[0188] In this embodiment, the Kalman filter is used to fuse the inlet and outlet water pressure difference, filtered water flow rate, and filtered water turbidity into real-time filtration data. The Kalman filter is a recursive estimation algorithm that can effectively handle multi-source, noisy data. In this embodiment, the system state equation and observation equation are first established. The state equation describes the dynamic characteristics of the filtration system, including the variation patterns of pressure difference, flow rate, and turbidity. The observation equation describes the measurement process of these variables. The Kalman filter is performed iteratively through two steps: prediction and update. In the prediction step, the current state is estimated based on the state at the previous time step; in the update step, the prediction is corrected based on the actual observations. The specific calculations involve updating the covariance matrix and calculating the Kalman gain.

[0189] Predicted filtered data is obtained based on real-time filtered data and a pre-defined filtered state prediction model, where the filtered state prediction model is an LSTM model. LSTM (Long Short-Term Memory) is a special type of recurrent neural network, particularly suitable for processing and predicting time series data. In this embodiment, the LSTM model receives real-time filtered data processed by Kalman filtering as input. The model structure includes an input layer, multiple LSTM layers, and an output layer. Each LSTM unit contains an input gate, a forget gate, and an output gate, which can effectively capture long-term dependencies in the data. Model training uses historical filtered data, and the network parameters are optimized through a backpropagation algorithm. During prediction, the model outputs a predicted filtered state for a future period based on the current and historical data sequences.

[0190] Predictive filtering data and historical backwash data are used as the second fuzzy input variables for the fuzzy logic system. Predictive filtering data includes parameters such as pressure differential, flow rate, and turbidity predicted by the LSTM model for a future period. Historical backwash data includes information such as the frequency, duration, and effectiveness of past backwashes. These precise numerical data need to be converted into linguistic variables that the fuzzy logic system can process. For example, the predicted pressure differential might be classified as "low," "medium," or "high," while the historical backwash frequency might be described as "frequent," "moderate," or "sparse." This conversion allows the system to simulate expert decision-making processes and handle the uncertainties and fuzziness in the operation of the filtering system.

[0191] Next, we define a second membership function for each second fuzzy input variable. For the pressure difference in the predicted filtered data, we can define a triangular membership function:

[0192]

[0193] Where 'a' represents the minimum pressure difference and 'm' represents the medium pressure difference. Similarly, define corresponding membership functions for parameters such as flow rate and turbidity. For historical backwash data, such as backwash frequency, a trapezoidal membership function can be used:

[0194]

[0195] [a,b,c,d] defines the four feature points of the fuzzy set "frequent". These carefully designed membership functions allow input variables to transition smoothly between different fuzzy sets, better reflecting the fuzziness and uncertainty in actual operation.

[0196] Based on the second membership function and according to the preset backwashing fuzzy inference rules, a backwashing correlation matrix is ​​established. Specifically, a series of backwashing fuzzy inference rules are first formulated, such as "if the predicted pressure difference is high and the historical backwashing frequency is low, then the backwashing urgency is high." Then, these rules are transformed into a fuzzy relation matrix R. The rows of matrix R correspond to the fuzzy sets of the input variables, and the columns correspond to the fuzzy sets of the output variables (backwashing urgency). The matrix element rij represents the degree of correlation between the i-th input fuzzy set and the j-th output fuzzy set, usually determined using fuzzy implication operations. In this way, all inference rules are encoded into a unified matrix, providing a mathematical foundation for subsequent fuzzy inference.

[0197] Next, the weight vector of the backwashing correlation matrix is ​​obtained by analyzing the second fuzzy input variable using the Analytic Hierarchy Process (AHP). The AHP first establishes a hierarchical decision structure, decomposing the backwashing decision problem into three levels: objective, criteria, and alternatives. Then, a judgment matrix A is constructed by comparing the importance of indicators pairwise. For example, if the predicted pressure difference is considered twice as important as the historical backwashing frequency, then the corresponding matrix element aij = 2. Next, the maximum eigenvalue λmax and the corresponding eigenvector W of the judgment matrix are calculated. The eigenvector W, after normalization, becomes the weight vector. Simultaneously, the consistency of the judgments is checked by calculating the consistency ratio. Then, the backwashing correlation matrix and the matrix weight vector are combined, and the backwashing inference evaluation result is obtained through fuzzy synthesis. Specifically, the weight vector W and the correlation matrix R are first subjected to fuzzy synthesis using the maximum-minimum synthesis method. The symbol "∘" represents the fuzzy composition operator. The result B is a fuzzy set representing the membership distribution of the backwash urgency level. For example, B could be [0.2, 0.5, 0.3], corresponding to the membership levels of low, medium, and high backwash urgency, respectively. The centroid method is used to defuzzify the backwash inference evaluation result, obtaining the initial backwash urgency level. The centroid method calculates the centroid of the fuzzy set B, using it as the final output value. The calculated initial backwash urgency level is a specific value between 0 and 100.

[0198] The present invention also discloses an intelligent water-saving control system for agricultural irrigation devices, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the intelligent water-saving control method for agricultural irrigation devices described in any of the above embodiments.

[0199] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it.

[0200] The memory can be an internal storage unit of a computer device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) provided on the computer device. Furthermore, the memory can be a combination of internal storage units and external storage devices of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0201] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0202] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A smart water-saving control method for agricultural irrigation devices, characterized in that, An agricultural irrigation device is applied, comprising a filtration irrigation module and a backwashing module. The irrigation unit of the filtration irrigation module is connected to the target farmland to be irrigated. A multi-functional irrigation outlet is configured at one end of the irrigation unit located in the target farmland. The backwashing module is used to backwash the filtration irrigation module. The method includes the following steps: Comprehensive soil information of the target farmland is collected by a first sensor pre-installed in the target farmland; Obtain meteorological information of the location of the target farmland and crop information within the target farmland; Based on the comprehensive soil information, the meteorological information, and the crop information, and using a preset farmland condition prediction model, a farmland condition prediction index for the target farmland is generated. Based on the farmland condition prediction index, a farmland irrigation strategy is generated. The farmland condition prediction model is a CNN-LSTM model. Based on the farmland condition prediction index, the irrigation urgency level corresponding to the farmland irrigation strategy is calculated using a fuzzy logic system. If the irrigation urgency level reaches the highest level of urgency, the filter irrigation module is controlled to perform alternating irrigation operations on the target farmland according to the farmland irrigation strategy. If the irrigation emergency level does not reach the highest level of emergency, the filter component data of the filter irrigation module is collected by a second sensor preset inside the filter irrigation module, and the historical backwash data of the backwash module is obtained. The urgency of backwashing of the filter irrigation module is calculated by combining the filter component data, the historical backwashing data, and the farmland irrigation strategy, and using a fuzzy logic system. If the backwashing urgency is greater than or equal to the irrigation urgency, a backwashing strategy is generated based on the filter component data, in which the backwashing module performs backwashing operations for the filter irrigation module, and the backwashing module and the filter irrigation module are controlled to execute the backwashing strategy and the farmland irrigation strategy in sequence. If the urgency of backwashing is less than the urgency of irrigation, then the filter irrigation module is controlled to perform alternating irrigation operations on the target farmland according to the farmland irrigation strategy.

2. The intelligent water-saving control method for agricultural irrigation devices according to claim 1, characterized in that, The comprehensive soil information includes soil moisture information and soil spatial distribution information, and the crop information includes crop remote sensing information and crop type information.

3. The intelligent water-saving control method for agricultural irrigation devices according to claim 2, characterized in that, The process of generating farmland condition prediction indicators for the target farmland based on the comprehensive soil information, meteorological information, and crop information, and using a preset farmland condition prediction model, includes the following steps: Preprocess the soil moisture information, the soil spatial distribution information, and the meteorological information; The preprocessed soil moisture information and the soil spatial distribution information are fused together in the spatial dimension to form soil moisture distribution data; The crop remote sensing information and the crop type information are fused into the soil spatial distribution information in a spatial dimension to obtain crop spatial distribution data; The soil moisture distribution data and the crop spatial distribution data are input into a preset farmland condition prediction model, and the soil condition prediction index and crop condition prediction index are output through the convolutional neural network module in the farmland condition prediction model. The preprocessed meteorological information is input into the farmland condition prediction model, and meteorological prediction indicators are output through the long short-term memory network module in the farmland condition prediction model. The soil condition prediction index, the crop condition prediction index, and the meteorological prediction index are integrated into a farmland condition prediction index.

4. The intelligent water-saving control method for agricultural irrigation devices according to claim 3, characterized in that, The step of fusing the preprocessed soil moisture information and the soil spatial distribution information into soil moisture distribution data in the spatial dimension includes the following steps: Spatial registration processing is performed on the soil moisture information and the soil spatial distribution information to place the soil moisture information and the soil spatial distribution information in the same spatial reference system; The spatially registered soil moisture information and the spatial distribution information are resampled to a spatial grid of a preset size. For each of the spatial grids, the soil moisture information within the spatial grids is aggregated over time to obtain aggregated soil moisture data; Traverse all the spatial grids. For null spatial grids where no soil moisture aggregate data exists, estimate the soil moisture aggregate data of the null spatial grids based on all the soil moisture aggregate data of the adjacent spatial grids of the null spatial grids using a spatial interpolation method. The soil moisture aggregation data of all the spatial grids are fused with the soil spatial distribution information to form soil moisture distribution data in the form of a multi-layer raster.

5. The intelligent water-saving control method for agricultural irrigation devices according to claim 4, characterized in that, The step of fusing the crop remote sensing information and the crop type information into the soil spatial distribution information in a spatial dimension to obtain crop spatial distribution data includes the following steps: The crop remote sensing information is resampled to the spatial grid. The crop growth status within each spatial grid is calculated using vegetation indices; The crop root distribution data within each spatial grid is estimated by combining the crop growth status and the crop type; The crop growth status, crop type, and crop root distribution data are fused with the soil spatial distribution information to form multi-layer raster-style crop spatial distribution data.

6. The intelligent water-saving control method for agricultural irrigation devices according to claim 3, characterized in that, The step of calculating the irrigation urgency level corresponding to the farmland irrigation strategy based on the farmland condition prediction index and using a fuzzy logic system includes the following steps: The farmland condition prediction index is used as the first fuzzy input variable of the fuzzy logic system. Define a fuzzy set and a first membership function for each of the first fuzzy input variables; Combining the fuzzy set and the first membership function, and using fuzzy inference methods, all the first fuzzy input variables are mapped to fuzzy output variables according to preset irrigation fuzzy inference rules; The centroid method is used to defuzzify the fuzzy output variables to obtain the irrigation urgency level corresponding to the farmland irrigation strategy.

7. The intelligent water-saving control method for agricultural irrigation devices according to claim 1, characterized in that, The data for the filtration components include the inlet and outlet water pressure difference, the filtered water flow rate, the turbidity of the filtered water, and the cumulative filtered water volume.

8. The intelligent water-saving control method for agricultural irrigation devices according to claim 7, characterized in that, The step of combining the filter component data, the historical backwash data, and the farmland irrigation strategy, and calculating the backwash urgency of the filter irrigation module using a fuzzy logic system, includes the following steps: The Kalman filter method is used to fuse the inlet and outlet water pressure difference, the filtered water flow rate, and the turbidity of the filtered water into real-time filtration data. Based on the real-time filtering data, predicted filtering data is obtained by predicting the filtering state using a preset filtering state prediction model, wherein the filtering state prediction model is an LSTM model. The predicted filtering data and the historical backwashing data are used as the second fuzzy input variables of the fuzzy logic system; Define a second membership function for each of the second fuzzy input variables; A backwashing association matrix is ​​established based on the second membership function and according to the preset backwashing fuzzy inference rules; The matrix weight vector of the backwashing correlation matrix is ​​obtained by analyzing the second fuzzy input variable using the analytic hierarchy process (AHP). By combining the backwash correlation matrix and the matrix weight vector, and through fuzzy synthesis operation, the backwash inference evaluation result is obtained; The centroid method is used to defuzzify the backwash inference evaluation results to obtain the initial backwash urgency level. Irrigation influencing factors are constructed based on the irrigation duration and irrigation volume in the farmland irrigation strategy, and the initial backwash urgency is corrected using the irrigation influencing factors to obtain the backwash urgency of the filter irrigation module.

9. An intelligent water-saving control system for agricultural irrigation devices, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent water-saving control method for agricultural irrigation devices as described in any one of claims 1 to 8.

Images