Artificial Intelligence-Based Method and System for Controlling Liquid Fertilizer Drip Irrigation in Agricultural Planting

Abstract

This invention discloses an artificial intelligence-based method and system for controlling the amount of liquid fertilizer drip irrigation in agricultural planting. The method involves collecting multi-source datasets through a multimodal sensing network; fusing the multi-source datasets using a STAFN spatiotemporal attention fusion network to obtain a fused dataset; establishing a dual-channel demand prediction model, inputting the fused dataset into the model, and outputting prediction results; calculating the fertilizer concentrate mixing ratio based on the prediction results; adjusting the fertilizer absorption rate of the Venturi fertilizer applicator according to the fertilizer concentrate mixing ratio using a fuzzy PID control algorithm; dividing the planting area into several intelligent irrigation zones; and implementing differentiated fertilization for each zone using a pulse drip irrigation strategy. This method adapts to the differences in soil and crops in different regions, improving drip irrigation efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, and in particular to a method and system for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence. Background Technology

[0002] With the acceleration of agricultural modernization, liquid fertilizer drip irrigation has been widely used in the cultivation of various crops due to its advantages of water and fertilizer conservation and precise supply. However, current drip irrigation control technology still has many pain points and cannot meet the needs of precision planting. Traditional drip irrigation fertilization relies heavily on manual experience to set fertilizer dosage and mixing ratio, lacking comprehensive consideration of multiple factors such as soil environment, crop growth, and weather conditions. This easily leads to problems of over-fertilization or under-fertilization. Over-fertilization not only wastes fertilizer and increases planting costs, but also causes soil compaction and water pollution; under-fertilization will affect crop growth and reduce yield and quality. In addition, existing fertilizer control systems mostly use fixed parameter control, which cannot achieve differentiated fertilization based on the spatial differences of soil and crops in the planting area, further reducing the accuracy and efficiency of drip irrigation fertilization. Summary of the Invention

[0003] The purpose of this invention is to solve the above-mentioned problems by designing an artificial intelligence-based method and system for controlling the amount of liquid fertilizer drip irrigation in agricultural planting.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned artificial intelligence-based method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting: Soil environment data, crop growth data, meteorological environment data, fertilizer characteristic data, and root zone microenvironment data are collected in the planting area through a multimodal sensing network. The collected data are preprocessed to obtain a multi-source dataset. The multi-source datasets are fused using the STAFN spatiotemporal attention fusion network to obtain a fused dataset; A dual-channel demand forecasting model is established. The short-term demand forecasting channel uses an LSTM-Attention architecture to predict the crop's fertilizer requirements in the next 36 hours. The long-term growth trend channel uses a Transformer architecture to predict the nutrient demand curve of the crop growth cycle. The fused dataset is input into the dual-channel demand prediction model, and the prediction results are output; the fertilizer concentrate mixing ratio is calculated based on the prediction results. Based on the mixing ratio of the fertilizer concentrate, the fuzzy PID control algorithm is used to adjust the fertilizer absorption of the Venturi fertilizer applicator, and the planting area is divided into several intelligent irrigation zones. A pulse drip irrigation strategy is used to implement differentiated fertilization for each zone.

[0005] Furthermore, in the aforementioned AI-based method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting, the process involves collecting soil environmental data, crop growth data, meteorological environmental data, fertilizer characteristic data, and root zone microenvironment data within the planting area using a multimodal sensing network. The collected data is then preprocessed to obtain a multi-source dataset, including: The 3σ principle is used to identify abnormal data during the acquisition process. After marking the outliers in the abnormal data, the missing data is supplemented by linear interpolation to obtain standard data. The wavelet transform denoising algorithm is used to denoise the random noise in the standard data, remove high-frequency interference signals, and obtain denoised data. After normalizing the denoised data using min-max normalization, the data from different acquisition frequencies are synchronized and aligned based on the timestamp to obtain a structured multi-source dataset.

[0006] Furthermore, in the aforementioned AI-based method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting, the step of using the STAFN spatiotemporal attention fusion network to fuse the multi-source dataset to obtain a fused dataset includes: The multi-source dataset is divided into 5 branches according to data type. Spatiotemporal features are extracted for each type of data. Temporal features are extracted through 1D convolutional layers, and spatial features are extracted through 2D convolutional layers. Each branch outputs the spatiotemporal feature vector of the corresponding data. A spatiotemporal attention mechanism is introduced to assign weights to the feature vectors of the five branches. The temporal attention weight is calculated based on the temporal correlation of the data, and the spatial attention weight is calculated based on the differences in crop growth and soil conditions at different locations within the planting area. The weighted spatiotemporal feature vectors of the five data categories are fused using a fully connected layer. Gradient vanishing is avoided by using residual connections. Batch normalization is introduced during the fusion process to output a fused dataset.

[0007] Furthermore, in the aforementioned AI-based method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting, a dual-channel demand prediction model is established. The short-term demand prediction channel uses an LSTM-Attention architecture to predict the crop's fertilizer requirement over the next 36 hours; the long-term growth trend channel uses a Transformer architecture to predict the nutrient requirement curve for the crop's growth cycle, including: Based on the LSTM long short-term memory network, an attention mechanism is introduced to establish the LSTM-Attention architecture; The LSTM-Attention architecture consists of four layers: an input layer for receiving temporal segments of the fused dataset, an LSTM feature extraction layer for capturing long-term dependencies in the temporal data, an attention layer for reinforcing the weights of key temporal features, and an output layer for outputting short-term fertilizer demand prediction results.

[0008] Furthermore, in the aforementioned AI-based method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting, a dual-channel demand prediction model is established. The short-term demand prediction channel uses an LSTM-Attention architecture to predict the crop's fertilizer requirement over the next 36 hours; the long-term growth trend channel uses a Transformer architecture to predict the nutrient requirement curve for the crop's growth cycle, including: We employ a Transformer encoder-decoder architecture and utilize a self-attention mechanism to capture the correlation patterns of nutrient requirements during the crop growth cycle.

[0009] Furthermore, in the aforementioned AI-based method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting, the step of inputting the fused dataset into a dual-channel demand prediction model and outputting prediction results; and calculating the fertilizer concentrate mixing ratio based on the prediction results, includes: The fused dataset is input into the dual-channel demand forecasting model. The short-term demand forecasting channel outputs the total fertilizer requirement of crops every 6 hours in the next 36 hours, as well as the specific requirements for nitrogen, phosphorus, and potassium nutrients. The long-term growth trend channel outputs the nutrient requirement curve for the remaining stage of the current crop growth cycle, including the key fertilizer requirements, peak fertilizer requirements, and duration of each growth stage. The total fertilizer requirement of crops in the next 36 hours, as well as the demand for nitrogen, phosphorus and potassium nutrients, are calculated. The specific amount of each fertilizer concentrate required is calculated based on the nutrient concentration of each fertilizer concentrate. Using water as a dilution medium, the mixing ratio between each stock solution is calculated based on the specific dosage and the ratio of water dosage, while controlling the total concentration of the fertilizer after mixing, and outputting the fertilizer stock solution mixing ratio.

[0010] Furthermore, in the aforementioned AI-based method for controlling the amount of liquid fertilizer applied during drip irrigation in agricultural planting, the step of adjusting the fertilizer absorption rate of the Venturi fertilizer applicator using a fuzzy PID control algorithm based on the mixing ratio of the fertilizer concentrate, dividing the planting area into several intelligent irrigation zones, and implementing differentiated fertilization for each zone using a pulse drip irrigation strategy includes: The fertilizer absorption amount corresponding to the mixing ratio is used as the target value. The actual fertilizer absorption amount of the Venturi fertilizer applicator is collected in real time by the sensor. The deviation between the target value and the actual value and the rate of change of the deviation are calculated and input into the fuzzy PID controller. The controller dynamically adjusts the PID parameters according to preset fuzzy rules, outputs an adjustment signal, and controls the speed of the fertilizer pump of the fertilizer applicator. Based on the crop fertilizer requirements of each zone, pulse drip irrigation parameters are set separately, including pulse frequency, single drip irrigation duration and drip irrigation pressure. Based on the independent control of each zone, the drip irrigation system synchronously performs pulse drip irrigation fertilization on each zone according to the fixed pulse drip irrigation parameters.

[0011] Furthermore, in the AI-based liquid fertilizer drip irrigation control system for agricultural planting, the system includes the following modules: The multi-source data acquisition module is used to collect soil environmental data, crop growth data, meteorological environmental data, fertilizer characteristic data and root zone microenvironment data in the planting area through a multimodal sensing network. The collected data is preprocessed to obtain a multi-source dataset. The data fusion processing module is used to fuse the multi-source dataset using the STAFN spatiotemporal attention fusion network to obtain a fused dataset; The prediction model building module is used to build a dual-channel demand prediction model. The short-term demand prediction channel is based on the LSTM-Attention architecture to predict the crop fertilizer requirements in the next 36 hours; the long-term growth trend channel is based on the Transformer architecture to predict the nutrient demand curve of the crop growth cycle. The fertilizer concentrate calculation module is used to input the fused dataset into the dual-channel demand prediction model and output the prediction results; and to calculate the fertilizer concentrate mixing ratio based on the prediction results. The fertilizer drip irrigation control module is used to adjust the fertilizer absorption of the Venturi fertilizer applicator according to the mixing ratio of the fertilizer concentrate using a fuzzy PID control algorithm, and to divide the planting area into several intelligent irrigation zones, and to implement zone-differentiated fertilization using a pulse drip irrigation strategy.

[0012] Furthermore, in the AI-based liquid fertilizer drip irrigation control system for agricultural planting, the fertilizer concentrate calculation module includes the following sub-modules: The prediction submodule is used to input the fused dataset into the dual-channel demand prediction model. The short-term demand prediction channel outputs the total fertilizer requirement of crops every 6 hours in the next 36 hours, as well as the specific requirements of nitrogen, phosphorus and potassium nutrients. The long-term growth trend channel outputs the nutrient requirement curve for the remaining stage of the current crop growth cycle, including the key fertilizer requirements, peak fertilizer requirements and duration of each growth stage. The calculation submodule is used to calculate the total fertilizer requirement of crops in the next 36 hours, as well as the demand for nitrogen, phosphorus and potassium nutrients, and to calculate the specific amount of each fertilizer concentrate required based on the nutrient concentration of each fertilizer concentrate. The mixing submodule is used to calculate the mixing ratio between each stock solution based on the ratio of the specific dosage to the amount of water, using water as the dilution carrier, while controlling the total concentration of the fertilizer after mixing, and outputting the mixing ratio of the fertilizer stock solution.

[0013] Furthermore, in the AI-based liquid fertilizer drip irrigation control system for agricultural planting, the fertilizer concentrate calculation module includes the following sub-modules: The acquisition submodule is used to collect the actual fertilizer absorption of the Venturi fertilizer applicator in real time through sensors, using the fertilizer absorption amount corresponding to the mixing ratio as the target value, calculate the deviation between the target value and the actual value and the rate of change of the deviation, and input it into the fuzzy PID controller. The adjustment submodule is used by the controller to dynamically adjust the PID parameters according to preset fuzzy rules, output adjustment signals, and control the speed of the fertilizer pump of the fertilizer applicator. The drip irrigation submodule is used to set pulse drip irrigation parameters according to the crop fertilizer requirements of each zone, including pulse frequency, single drip irrigation duration and drip irrigation pressure. Based on the independent control of each zone, the drip irrigation system synchronously performs pulse drip irrigation fertilization on each zone according to the fixed pulse drip irrigation parameters.

[0014] Its beneficial effects lie in achieving accurate collection and efficient fusion of multi-source data. Through a multimodal sensing network, it comprehensively collects five core data categories: soil, crops, meteorology, fertilizer, and root zone microenvironment. After standardized preprocessing, the STAFN spatiotemporal attention fusion network eliminates data heterogeneity and highlights the weight of key data, providing high-quality data support for fertilizer requirement prediction and solving the problems of single data and inaccurate fusion in existing technologies. It employs a short-term + long-term dual-channel prediction model. The LSTM-Attention architecture accurately predicts the immediate fertilizer requirement for the next 36 hours, while the Transformer architecture predicts the nutrient demand curve throughout the entire growth cycle, balancing immediate supply and long-term planning. The prediction error is controlled within 5%, effectively avoiding over- or under-fertilization. It achieves intelligent and differentiated drip irrigation control. A fuzzy PID control algorithm precisely adjusts the fertilizer uptake, ensuring a stable fertilizer mixing ratio. Combined with K-means clustering for intelligent zoning, a pulsed drip irrigation strategy is used to achieve differentiated fertilization by zone, adapting to the differences in soil and crops in different regions and improving drip irrigation efficiency and accuracy. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 This is a schematic diagram of the first embodiment of the method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence in this invention. Figure 2 This is a schematic diagram of the second embodiment of the method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence in this invention. Figure 3 This is a schematic diagram of the first embodiment of the drip irrigation control system for liquid fertilizer in agricultural planting based on artificial intelligence in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms "one," "an," and "this" used herein may also include the plural forms. It should be further understood that the terminology used in this specification includes the presence of features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, an artificial intelligence-based method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting includes the following steps: Step 101: Collect soil environment data, crop growth data, meteorological environment data, fertilizer characteristic data and root zone microenvironment data in the planting area through a multimodal sensing network, and preprocess the collected data to obtain a multi-source dataset; Specifically, in this embodiment, the 3σ principle is used to identify abnormal data during the acquisition process. After marking the outliers in the abnormal data, the missing data is supplemented by linear interpolation to obtain standard data. Wavelet transform denoising algorithm is used to denoise the random noise in the standard data and remove high-frequency interference signals to obtain denoised data. After normalizing the denoised data using min-max normalization, the data from different acquisition frequencies are synchronized and aligned based on the timestamp to obtain a structured multi-source dataset.

[0020] Soil environmental data, collected via soil sensors, covers soil moisture content, soil pH, soil electrical conductivity, and soil temperature, with a collection frequency of once every 2 hours. Crop growth data is collected using a combination of high-definition cameras and spectral sensors, focusing on crop height, stem diameter, leaf area index, leaf chlorophyll content, and fruit development. Meteorological environmental data, collected via small weather stations, includes air temperature, air humidity, light intensity (unit: lux, measurement range: 0-200,000 lux), precipitation (accuracy: ±0.1 mm), wind speed, and wind direction, with a collection frequency of once every 1 hour. Fertilizer characteristic data, collected via a fertilizer detection module, focuses on core indicators of liquid fertilizers, including nitrogen, phosphorus, potassium, and micronutrient concentrations, fertilizer density, viscosity, and solubility, with a collection frequency of once before each batch of fertilizer is used. Root zone microenvironment data, collected via embedded root zone sensors, includes core data such as root zone soil moisture content, root zone temperature, and root zone nitrogen, phosphorus, and potassium nutrient concentrations, with a collection frequency of once every 2 hours.

[0021] Step 102: Use the STAFN spatiotemporal attention fusion network to fuse the multi-source datasets to obtain a fused dataset; Specifically, in this embodiment, the multi-source dataset is divided into 5 branches according to data type. Spatiotemporal features are extracted for each type of data. Temporal features are extracted through 1D convolutional layers, and spatial features are extracted through 2D convolutional layers. Each branch outputs a spatiotemporal feature vector corresponding to the data. A spatiotemporal attention mechanism is introduced to assign weights to the feature vectors of the 5 branches. The temporal attention weight is calculated based on the temporal correlation of the data, and the spatial attention weight is calculated based on the differences in crop growth and soil conditions at different locations within the planting area. The weighted spatiotemporal feature vectors of the 5 types of data are fused through a fully connected layer. The gradient vanishing method is avoided by using residual connections. Batch normalization is introduced during the fusion process to output a fused dataset.

[0022] The preprocessed multi-source datasets were divided into 5 independent branches according to data type. Spatiotemporal features were extracted for each type of data. Temporal features were extracted through 1D convolutional layers to capture the changing patterns of data over time, such as the temporal fluctuations of soil moisture content and light intensity. Spatial features were extracted through 2D convolutional layers to capture the differences in data at different locations within the planting area, such as differences in crop growth and soil fertility in different regions. Each branch ultimately outputs the spatiotemporal feature vector of the corresponding data. A spatiotemporal attention mechanism is introduced to dynamically assign weights to the feature vectors of the five branches. Among them, the temporal attention weight is calculated based on the temporal correlation of the data, giving priority to increasing the weight of recent data, as it has a greater impact on the immediate nutrient requirements of crops; the spatial attention weight is calculated based on the differences in crop growth and soil conditions in different locations within the planting area, giving priority to increasing the weight of soil and crop data in areas with weaker growth and insufficient soil fertility, ensuring that key data features are strengthened. The weighted spatiotemporal feature vectors of the five data categories are deeply fused using a fully connected layer. Residual connections are employed to avoid the vanishing gradient problem during model training. Batch normalization is introduced during the fusion process to improve the stability and consistency of the fused data. The final output is a unified, complementary fused dataset containing three core dimensions: time, space, and features. This dataset can be directly input into subsequent dual-channel demand prediction models.

[0023] Step 103: Establish a dual-channel demand forecasting model. The short-term demand forecasting channel uses an LSTM-Attention architecture to predict the crop's fertilizer requirements in the next 36 hours. The long-term growth trend channel uses a Transformer architecture to predict the nutrient demand curve of the crop growth cycle. Specifically, this embodiment uses an LSTM (Long Short-Term Memory) network as its foundation and introduces an attention mechanism to establish an LSTM-Attention architecture. The LSTM-Attention architecture consists of four layers: an input layer for receiving temporal segments of the fused dataset, an LSTM feature extraction layer for capturing long-term dependencies in the temporal data, an attention layer for strengthening the weights of key temporal features, and an output layer for outputting short-term nutrient requirement prediction results. A Transformer encoder-decoder architecture is adopted, utilizing a self-attention mechanism to capture the correlation patterns of nutrient requirements within the crop growth cycle.

[0024] Based on LSTM (Long Short-Term Memory) networks, an attention mechanism is introduced to address the problem of traditional LSTM networks' insufficient capture of key information in long-sequence data. The model consists of four layers: an input layer that receives time segments from the fused dataset, an LSTM feature extraction layer that captures long-term dependencies in the time-series data, an attention layer that strengthens the weights of key time-series features, and an output layer that outputs short-term demand prediction results. The input is a recent time series segment of the fused dataset, specifically the fused data from the past 72 hours, divided into 12 time series data steps with each time step being 6 hours. The output is the crop fertilizer requirement for the next 36 hours, in kg / acre, with 6 fertilizer requirement values ​​output with each time node being 6 hours. At the same time, the specific fertilizer requirements of nitrogen, phosphorus, potassium and micronutrients at each time node are output to ensure the timeliness and accuracy of short-term fertilizer requirement prediction. Model training utilizes preprocessed historical data from a multi-source dataset. The optimization objective is to minimize the mean squared error (MSE) between the predicted and actual fertilizer requirements. Dropout regularization is employed during training to suppress overfitting, while early stopping (EarlyStopping) is used to prevent overtraining. After training, model accuracy is validated using a validation set to ensure the prediction error is kept below 5%, meeting practical production requirements. It adopts the encoder-decoder architecture of Transformer and uses the self-attention mechanism to capture the correlation of nutrient requirements during the crop growth cycle. Compared with traditional networks, it is more suitable for processing long-sequence, non-linear crop growth data and can effectively capture the correlation of fertilizer requirements at different growth stages. The input is historical growth cycle data from the fused dataset. You can choose the complete data from the previous growth cycle of the crop or all the data collected in the current growth cycle. The output is the nutrient requirement curve of the crop throughout its entire growth cycle. The horizontal axis of the curve represents the crop growth stage, which is divided into seedling stage, tillering stage, jointing stage, grain filling stage, and maturity stage. The vertical axis represents the average daily fertilizer requirement for the corresponding growth stage. The peak fertilizer requirement and duration of each growth stage are also marked to provide a basis for long-term fertilization planning. The model is trained by combining growth cycle data and historical fertilizer requirement data of different crop varieties. A cross-entropy loss function is introduced to optimize model parameters, improving the model's adaptability to different crop varieties and growth environments. After training, the long-term demand curve can be dynamically updated based on the current crop growth stage, ensuring the relevance and accuracy of the prediction results.

[0025] Step 104: Input the fused dataset into the dual-channel demand prediction model and output the prediction results; calculate the fertilizer concentrate mixing ratio based on the prediction results. Specifically, in this embodiment, the fused dataset is input into the dual-channel demand prediction model. The short-term demand prediction channel outputs the total crop fertilizer requirement every 6 hours for the next 36 hours, as well as the specific requirements for nitrogen, phosphorus, and potassium nutrients. The long-term growth trend channel outputs the nutrient requirement curve for the remaining stage of the current crop growth cycle, including the key fertilizer requirements, peak fertilizer requirements, and duration of each growth stage. The total crop fertilizer requirement for the next 36 hours, as well as the requirements for nitrogen, phosphorus, and potassium nutrients, are calculated. The specific dosage of each fertilizer concentrate is calculated based on the nutrient concentration of each fertilizer concentrate. Using water as a dilution carrier, the mixing ratio between each concentrate is calculated based on the ratio of the specific dosage to the amount of water used, while controlling the total concentration of the fertilizer after mixing, and outputting the fertilizer concentrate mixing ratio.

[0026] The prediction results are output by organizing the fused dataset into the format required by the dual-channel model and then inputting it into the model. The short-term demand prediction channel outputs the total fertilizer requirement of the crop every 6 hours in the next 36 hours, as well as the specific requirements of each nutrient such as nitrogen, phosphorus, and potassium. The long-term growth trend channel outputs the nutrient requirement curve for the remaining stage of the current crop growth cycle, clarifying the key fertilizer requirements, peak fertilizer requirements, and duration of each growth stage, providing a reference for the calculation of the mixing ratio. The calculation of the mixing ratio is based on the short-term fertilizer requirement forecast, prioritizing the immediate fertilizer requirements of crops. At the same time, it takes into account the phased requirements of the long-term demand curve to avoid a disconnect between short-term fertilization and long-term growth needs. In addition, it strictly refers to the fertilizer characteristic data collected in step 1, such as the nutrient concentration and density of each fertilizer concentrate, to ensure that the nutrient ratio and concentration of the mixed fertilizer meet the needs of crops, avoiding both excessively high concentrations that could burn seedlings and excessively low concentrations that could not meet growth needs. Summarize the fertilizer requirements, calculate the total fertilizer requirements of crops in the next 36 hours, as well as the specific requirements of each core nutrient such as nitrogen, phosphorus, and potassium, and clarify the proportion of each nutrient; Calculate the dosage of each stock solution. Based on the nutrient concentration of each fertilizer stock solution, use the formula "Stock solution dosage = Target nutrient requirement ÷ Stock solution nutrient concentration" to calculate the specific dosage of each fertilizer stock solution required. Determine the mixing ratio, using water as the dilution medium. Calculate the mixing ratio between each stock solution based on the ratio of the total amount of stock solution used to the amount of water used. At the same time, control the total concentration of the fertilizer after mixing, generally within a safe range suitable for crop absorption. Finally, output a clear fertilizer stock solution mixing ratio, such as stock solution A: stock solution B: water = 1:2:100, to provide a clear basis for subsequent drip irrigation implementation.

[0027] Step 105: Based on the mixing ratio of the fertilizer concentrate, adjust the fertilizer absorption of the Venturi fertilizer applicator using a fuzzy PID control algorithm, divide the planting area into several intelligent irrigation zones, and implement differentiated fertilization for each zone using a pulse drip irrigation strategy.

[0028] Specifically, in this embodiment, the fertilizer absorption amount corresponding to the mixing ratio is used as the target value. The actual fertilizer absorption amount of the Venturi fertilizer applicator is collected in real time by the sensor. The deviation between the target value and the actual value and the rate of change of the deviation are calculated and input into the fuzzy PID controller. The controller dynamically adjusts the PID parameters through preset fuzzy rules, outputs the adjustment signal, and controls the speed of the fertilizer pump of the fertilizer applicator. According to the crop fertilizer requirements of each zone, pulse drip irrigation parameters are set respectively, including pulse frequency, single drip irrigation duration and drip irrigation pressure. The drip irrigation system based on independent control of each zone synchronously performs pulse drip irrigation fertilization on each zone according to the fixed pulse drip irrigation parameters.

[0029] Control the target and adjust the fertilizer absorption of the Venturi fertilizer applicator to ensure that the deviation between the actual fertilizer absorption and the calculated value is controlled within ±3%, maintain the stability of the nutrient concentration of the mixed fertilizer, and meet the fertilizer requirements of the crop. In the control process, the fertilizer absorption amount corresponding to the mixing ratio is taken as the target value. The actual fertilizer absorption amount of the Venturi fertilizer applicator is collected in real time by sensors. The deviation between the target value and the actual value and the rate of change of the deviation are calculated and input into the fuzzy PID controller. The controller is trained by using preset fuzzy rules and combined with historical adjustment data to dynamically adjust the PID parameters, proportional coefficient, integral coefficient and derivative coefficient, and output adjustment signal to control the speed of the fertilizer pump of the fertilizer applicator, thereby achieving precise adjustment of the fertilizer absorption amount. At the same time, the adjusted fertilizer absorption amount data is collected in real time and fed back to the controller to form a closed-loop control and continuously optimize the adjustment effect. In case of anomaly, when the fertilizer absorption deviation exceeds the allowable range of ±3%, the controller will automatically issue an alarm signal and suspend the fertilization operation, prompting staff to check for problems such as sensor malfunction, fertilizer applicator blockage, and fertilizer pump malfunction. After the fault is resolved, the system will automatically resume fertilization operation to avoid crop growth being affected by abnormal fertilizer absorption. The division was based on the soil environmental data collected in step 1, including soil moisture content, soil fertility, crop growth data, plant height, and chlorophyll content. Combined with the topographic differences of the planting area, the K-means clustering algorithm was used to divide the area into zones. The zoning standard is that the difference in soil moisture content, soil fertility, and crop growth within the same zone should not exceed 10%, ensuring that the fertilizer requirements of crops within the same zone are consistent; there should be obvious differences in fertilizer requirements between different zones, such as zones with weaker growth requiring more fertilizer. The number of zones is determined based on the planting area and data differences, generally 5-10 mu per zone. Drip irrigation parameters are set according to the crop nutrient requirements of each zone. The pulse drip irrigation parameters are set separately, including pulse frequency, which is generally once every 10-15 minutes, single drip irrigation duration, 5-10 minutes, and drip irrigation pressure, 0.15-0.25 MPa. For zones with high nutrient requirements, the drip irrigation duration and pulse frequency are appropriately increased to ensure sufficient nutrient supply. During the execution process, a drip irrigation system with independent control of each zone is adopted. According to the set parameters, pulse drip irrigation and fertilization are performed synchronously in each zone. During the drip irrigation process, the soil moisture content and root zone nutrient concentration of each zone are collected in real time. The drip irrigation parameters are dynamically adjusted based on the feedback data to ensure that the nutrient supply is precisely matched with the crop's needs. With guaranteed effectiveness, pulse drip irrigation avoids the soil compaction and nutrient loss problems caused by traditional continuous drip irrigation. At the same time, through differentiated supply to different zones, it ensures that each crop receives precise fertilizer supply, effectively improving fertilizer utilization, reducing fertilizer waste, and lowering planting costs.

[0030] Its beneficial effects lie in achieving accurate collection and efficient fusion of multi-source data. Through a multimodal sensing network, it comprehensively collects five core data categories: soil, crops, meteorology, fertilizer, and root zone microenvironment. After standardized preprocessing, the STAFN spatiotemporal attention fusion network eliminates data heterogeneity and highlights the weight of key data, providing high-quality data support for fertilizer requirement prediction and solving the problems of single data and inaccurate fusion in existing technologies. It employs a short-term + long-term dual-channel prediction model. The LSTM-Attention architecture accurately predicts the immediate fertilizer requirement for the next 36 hours, while the Transformer architecture predicts the nutrient demand curve throughout the entire growth cycle, balancing immediate supply and long-term planning. The prediction error is controlled within 5%, effectively avoiding over- or under-fertilization. It achieves intelligent and differentiated drip irrigation control. A fuzzy PID control algorithm precisely adjusts the fertilizer uptake, ensuring a stable fertilizer mixing ratio. Combined with K-means clustering for intelligent zoning, a pulsed drip irrigation strategy is used to achieve differentiated fertilization by zone, adapting to the differences in soil and crops in different regions and improving drip irrigation efficiency and accuracy.

[0031] Please see Figure 2 In an AI-based method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting, the STAFN spatiotemporal attention fusion network is used to fuse multi-source datasets to obtain a fused dataset. The steps include: Step 201: Divide the multi-source dataset into 5 branches according to data type, and extract spatiotemporal features for each type of data. Temporal features are extracted through 1D convolutional layers, and spatial features are extracted through 2D convolutional layers. Each branch outputs the spatiotemporal feature vector of the corresponding data. Step 202: Introduce a spatiotemporal attention mechanism to assign weights to the feature vectors of the five branches. The temporal attention weight is calculated based on the temporal correlation of the data, and the spatial attention weight is calculated based on the differences in crop growth and soil conditions at different locations within the planting area. Step 203: The weighted spatiotemporal feature vectors of the five data categories are fused through a fully connected layer. The gradient vanishing method is avoided by using residual connection. Batch normalization is introduced during the fusion process, and the fused dataset is output.

[0032] The above describes embodiments of the artificial intelligence-based drip irrigation method for controlling the amount of liquid fertilizer used in agricultural planting according to the present invention. Please refer to [link / reference]. Figure 3 In the AI-based liquid fertilizer drip irrigation control system for agricultural planting, the system includes the following modules: The multi-source data acquisition module is used to collect soil environmental data, crop growth data, meteorological environmental data, fertilizer characteristic data and root zone microenvironment data in the planting area through a multimodal sensing network. The collected data is preprocessed to obtain a multi-source dataset. The data fusion processing module is used to fuse multi-source datasets using the STAFN spatiotemporal attention fusion network to obtain a fused dataset; The prediction model building module is used to build a dual-channel demand prediction model. The short-term demand prediction channel is based on the LSTM-Attention architecture to predict the crop fertilizer requirements in the next 36 hours; the long-term growth trend channel is based on the Transformer architecture to predict the nutrient demand curve of the crop growth cycle. The fertilizer concentrate calculation module is used to input the fused dataset into the dual-channel demand prediction model and output the prediction results; and to calculate the fertilizer concentrate mixing ratio based on the prediction results. The fertilizer drip irrigation control module is used to adjust the fertilizer absorption of the Venturi fertilizer applicator according to the mixing ratio of the fertilizer concentrate using a fuzzy PID control algorithm, and divides the planting area into several intelligent irrigation zones, and uses a pulse drip irrigation strategy to perform differentiated fertilization for each zone.

[0033] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and equivalents.

Claims

1. A method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting based on artificial intelligence, characterized in that, The method for controlling the amount of liquid fertilizer used in drip irrigation for agricultural planting includes the following steps: Soil environment data, crop growth data, meteorological environment data, fertilizer characteristic data, and root zone microenvironment data are collected in the planting area through a multimodal sensing network. The collected data are preprocessed to obtain a multi-source dataset. The multi-source dataset is fused using the STAFN spatiotemporal attention fusion network to obtain a fused dataset; A dual-channel demand forecasting model is established. The short-term demand forecasting channel uses an LSTM-Attention architecture to predict the crop's fertilizer requirements in the next 36 hours. The long-term growth trend channel uses a Transformer architecture to predict the nutrient demand curve of the crop growth cycle. The fused dataset is input into the dual-channel demand prediction model, and the prediction results are output; the fertilizer concentrate mixing ratio is calculated based on the prediction results. Based on the mixing ratio of the fertilizer concentrate, the fuzzy PID control algorithm is used to adjust the fertilizer absorption of the Venturi fertilizer applicator, and the planting area is divided into several intelligent irrigation zones. A pulse drip irrigation strategy is used to implement differentiated fertilization for each zone.

2. The method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence as described in claim 1, characterized in that, The process involves collecting soil environmental data, crop growth data, meteorological environmental data, fertilizer characteristic data, and root zone microenvironment data within the planting area using a multimodal sensing network. The collected data is then preprocessed to obtain a multi-source dataset, including: The 3σ principle is used to identify abnormal data during the acquisition process. After marking the outliers in the abnormal data, the missing data is supplemented by linear interpolation to obtain standard data. The wavelet transform denoising algorithm is used to denoise the random noise in the standard data, remove high-frequency interference signals, and obtain denoised data. After normalizing the denoised data using min-max normalization, the data from different acquisition frequencies are synchronized and aligned based on the timestamp to obtain a structured multi-source dataset.

3. The method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence as described in claim 1, characterized in that, The STAFN spatiotemporal attention fusion network is used to fuse the multi-source dataset to obtain a fused dataset, including: The multi-source dataset is divided into 5 branches according to data type. Spatiotemporal features are extracted for each type of data. Temporal features are extracted through 1D convolutional layers, and spatial features are extracted through 2D convolutional layers. Each branch outputs the spatiotemporal feature vector of the corresponding data. A spatiotemporal attention mechanism is introduced to assign weights to the feature vectors of the five branches. The temporal attention weight is calculated based on the temporal correlation of the data, and the spatial attention weight is calculated based on the differences in crop growth and soil conditions at different locations within the planting area. The weighted spatiotemporal feature vectors of the five data categories are fused using a fully connected layer. Gradient vanishing is avoided by using residual connections. Batch normalization is introduced during the fusion process to output a fused dataset.

4. The method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence as described in claim 1, characterized in that, The aforementioned dual-channel demand forecasting model is established, with the short-term demand forecasting channel based on the LSTM-Attention architecture to predict crop fertilizer requirements within the next 36 hours. The long-term growth trend channel, based on the Transformer architecture, predicts the nutrient requirement curves of the crop growth cycle, including: Based on the LSTM long short-term memory network, an attention mechanism is introduced to establish the LSTM-Attention architecture; The LSTM-Attention architecture consists of four layers: an input layer for receiving temporal segments of the fused dataset, an LSTM feature extraction layer for capturing long-term dependencies in the temporal data, an attention layer for reinforcing the weights of key temporal features, and an output layer for outputting short-term fertilizer demand prediction results.

5. The method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence as described in claim 1, characterized in that, The aforementioned dual-channel demand forecasting model is established, with the short-term demand forecasting channel based on the LSTM-Attention architecture to predict crop fertilizer requirements within the next 36 hours. The long-term growth trend channel, based on the Transformer architecture, predicts the nutrient requirement curves of the crop growth cycle, including: We employ a Transformer encoder-decoder architecture and utilize a self-attention mechanism to capture the correlation patterns of nutrient requirements during the crop growth cycle.

6. The method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence as described in claim 1, characterized in that, The fused dataset is input into the dual-channel demand prediction model, and the prediction result is output. The fertilizer concentrate mixing ratio is calculated based on the prediction results, including: The fused dataset is input into the dual-channel demand forecasting model. The short-term demand forecasting channel outputs the total fertilizer requirement of crops every 6 hours in the next 36 hours, as well as the specific requirements for nitrogen, phosphorus, and potassium nutrients. The long-term growth trend channel outputs the nutrient requirement curve for the remaining stage of the current crop growth cycle, including the key fertilizer requirements, peak fertilizer requirements, and duration of each growth stage. The total fertilizer requirement of crops in the next 36 hours, as well as the demand for nitrogen, phosphorus and potassium nutrients, are calculated. The specific amount of each fertilizer concentrate required is calculated based on the nutrient concentration of each fertilizer concentrate. Using water as a dilution medium, the mixing ratio between each stock solution is calculated based on the specific dosage and the ratio of water dosage, while controlling the total concentration of the fertilizer after mixing, and outputting the fertilizer stock solution mixing ratio.

7. The method for controlling the amount of liquid fertilizer drip irrigation in agricultural planting based on artificial intelligence as described in claim 1, characterized in that, The process involves adjusting the fertilizer absorption rate of the Venturi fertilizer applicator using a fuzzy PID control algorithm based on the fertilizer concentrate mixing ratio, dividing the planting area into several intelligent irrigation zones, and implementing differentiated fertilization for each zone using a pulse drip irrigation strategy. The fertilizer absorption amount corresponding to the mixing ratio is used as the target value. The actual fertilizer absorption amount of the Venturi fertilizer applicator is collected in real time by the sensor. The deviation between the target value and the actual value and the rate of change of the deviation are calculated and input into the fuzzy PID controller. The controller dynamically adjusts the PID parameters according to preset fuzzy rules, outputs an adjustment signal, and controls the speed of the fertilizer pump of the fertilizer applicator. Based on the crop fertilizer requirements of each zone, pulse drip irrigation parameters are set separately, including pulse frequency, single drip irrigation duration and drip irrigation pressure. Based on the independent control of each zone, the drip irrigation system synchronously performs pulse drip irrigation fertilization on each zone according to the fixed pulse drip irrigation parameters.

8. An artificial intelligence-based drip irrigation system for controlling the amount of liquid fertilizer used in agricultural planting, characterized in that, The agricultural liquid fertilizer drip irrigation control system includes the following modules: The multi-source data acquisition module is used to collect soil environmental data, crop growth data, meteorological environmental data, fertilizer characteristic data and root zone microenvironment data in the planting area through a multimodal sensing network. The collected data is preprocessed to obtain a multi-source dataset. The data fusion processing module is used to fuse the multi-source dataset using the STAFN spatiotemporal attention fusion network to obtain a fused dataset; The prediction model building module is used to build a dual-channel demand prediction model. The short-term demand prediction channel is based on the LSTM-Attention architecture to predict the crop fertilizer requirements in the next 36 hours; the long-term growth trend channel is based on the Transformer architecture to predict the nutrient demand curve of the crop growth cycle. The fertilizer concentrate calculation module is used to input the fused dataset into the dual-channel demand prediction model and output the prediction results; and to calculate the fertilizer concentrate mixing ratio based on the prediction results. The fertilizer drip irrigation control module is used to adjust the fertilizer absorption of the Venturi fertilizer applicator according to the mixing ratio of the fertilizer concentrate using a fuzzy PID control algorithm, and to divide the planting area into several intelligent irrigation zones, and to implement zone-differentiated fertilization using a pulse drip irrigation strategy.

9. The artificial intelligence-based liquid fertilizer drip irrigation control system for agricultural planting as described in claim 8, characterized in that, The fertilizer concentrate calculation module includes the following sub-modules: The prediction submodule is used to input the fused dataset into the dual-channel demand prediction model. The short-term demand prediction channel outputs the total fertilizer requirement of crops every 6 hours in the next 36 hours, as well as the specific requirements of nitrogen, phosphorus and potassium nutrients. The long-term growth trend channel outputs the nutrient requirement curve for the remaining stage of the current crop growth cycle, including the key fertilizer requirements, peak fertilizer requirements and duration of each growth stage. The calculation submodule is used to calculate the total fertilizer requirement of crops in the next 36 hours, as well as the demand for nitrogen, phosphorus and potassium nutrients, and to calculate the specific amount of each fertilizer concentrate required based on the nutrient concentration of each fertilizer concentrate. The mixing submodule is used to calculate the mixing ratio between each stock solution based on the ratio of the specific dosage to the amount of water, using water as the dilution carrier, while controlling the total concentration of the fertilizer after mixing, and outputting the mixing ratio of the fertilizer stock solution.

10. The artificial intelligence-based liquid fertilizer drip irrigation control system for agricultural planting as described in claim 8, characterized in that, The fertilizer concentrate calculation module includes the following sub-modules: The acquisition submodule is used to collect the actual fertilizer absorption of the Venturi fertilizer applicator in real time through sensors, using the fertilizer absorption amount corresponding to the mixing ratio as the target value, calculate the deviation between the target value and the actual value and the rate of change of the deviation, and input it into the fuzzy PID controller. The adjustment submodule is used by the controller to dynamically adjust the PID parameters according to preset fuzzy rules, output adjustment signals, and control the speed of the fertilizer pump of the fertilizer applicator. The drip irrigation submodule is used to set pulse drip irrigation parameters according to the crop fertilizer requirements of each zone, including pulse frequency, single drip irrigation duration and drip irrigation pressure. Based on the independent control of each zone, the drip irrigation system synchronously performs pulse drip irrigation fertilization on each zone according to the fixed pulse drip irrigation parameters.

Images