Citrus water and fertilizer real-time decision-making system and method based on multi-modal sensing and deep learning

Abstract

The invention discloses a citrus water and fertilizer real-time decision-making system and method based on multi-modal sensing and deep learning. The system comprises a real-time data flow monitoring unit, a clear water irrigation prediction unit and a nutrient solution proportioning and control unit. The real-time data flow monitoring unit is used for collecting real-time sensor data and constructing dual-thread task scheduling, so that the whole process is continuously operated; the clear water irrigation prediction unit is used for calculating the tidal irrigation clear water soaking duration in real time by adopting a deep learning model architecture and irrigation decision logic on the basis of optimizing a Penman formula; and the nutrient solution proportioning and control unit is used for dynamically adjusting the target concentration of nitrogen, phosphorus and potassium in real time and calculating the soaking time of the tidal irrigation nutrient solution by utilizing soil conductivity interference correction / hydrogen ion concentration index buffer linkage compensation and a growth stage self-adaptive strategy. The method solves the problems that a traditional irrigation technology depends on artificial experience, cannot achieve full-process closed-loop control, is poor in environmental adaptability and the like, and remarkably improves the survival rate and growth efficiency of plants.

Description

Technical Field

[0001] This invention relates to a real-time decision-making system and method for water and fertilizer management in citrus based on multimodal sensing and deep learning, belonging to the field of agricultural Internet of Things and smart irrigation technology. Background Technology

[0002] China's citrus planting area has exceeded 2 million hectares, with an annual output of over 40 million tons, making the citrus industry the largest pillar of my country's fruit tree economy. Water, as a core element of citrus life activities, directly affects photosynthetic efficiency, nutrient transport, and fruit quality. Especially during critical water-demanding stages such as budding and fruit enlargement, scientific irrigation is crucial for increasing yield. Although the main producing areas receive an average annual rainfall of 1200-1500 mm, the mismatch between water and soil resources is prominent due to topography and climate, manifesting as frequent alternations of drought and flood. For example, high temperatures and drought in summer cause leaf wilting, while heavy autumn rains lead to root rot. Traditional flood irrigation and furrow irrigation rely on manual experience, resulting in water use efficiency of less than 40%, and easily damaging soil structure. Furthermore, the lack of dynamic monitoring of multiple parameters such as soil EC value, pH, and root weight makes it impossible to accurately match the needs of different growth stages. Moreover, traditional water and fertilizer management mostly uses solid fertilizer application or fixed formula nutrient solution irrigation. Although the current mainstream nutrient solution introduces chelated elements to improve utilization, it still relies on manual experience to adjust the concentration, and cannot achieve closed-loop control of multiple parameters of "soil-plant-environment". Furthermore, current irrigation technology still has significant limitations, and existing technologies are difficult to achieve real-time automated closed-loop control of the entire process of "sensing-decision-execution".

[0003] Therefore, it is particularly important to realize the combination of multimodal sensing and deep learning in irrigation methods, as well as a comprehensive real-time decision-making system for water and fertilizer in citrus facility agriculture. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of the prior art by providing a real-time decision-making system for citrus irrigation and fertilization based on multimodal sensing and deep learning. This system automatically and continuously collects and processes multi-source sensor data in real time, and automatically executes a model and method to accurately predict the soaking time and concentration required for tidal irrigation and fertilization of citrus. This enables precise irrigation and fertilization, improves the utilization rate of water and fertilizer resources, reduces agricultural production costs, and ultimately improves the yield and quality of citrus, which has significant practical significance and application value.

[0005] Another objective of this invention is to provide a real-time decision-making method for citrus irrigation and fertilization, which is based on the aforementioned real-time decision-making system for citrus irrigation and fertilization.

[0006] The objective of this invention can be achieved by adopting the following technical solutions:

[0007] A real-time decision-making system for citrus irrigation and fertilization based on multimodal sensing and deep learning includes a real-time data stream monitoring unit, a clean water irrigation prediction unit, and a nutrient solution ratio and control unit.

[0008] The real-time data stream monitoring unit is used to collect real-time sensor data and build a dual-thread task schedule to ensure uninterrupted operation of the entire process.

[0009] The clear water irrigation prediction unit is used to calculate the clear water soaking time of tidal irrigation in real time based on the optimized Penman formula, using a deep learning model architecture and irrigation decision logic.

[0010] The nutrient solution ratio and control unit is used to dynamically adjust the target concentrations of nitrogen, phosphorus, and potassium in real time and calculate the soaking time of nutrient solution for tidal irrigation by utilizing soil conductivity interference correction / hydrogen ion concentration index buffer linkage compensation and growth stage adaptive strategies.

[0011] Furthermore, the real-time data stream monitoring unit includes a sensor module, a wireless data transmission terminal, and a host computer, with the wireless data transmission terminal connected to both the sensor module and the host computer.

[0012] Furthermore, the sensor module includes a soil temperature sensor, a soil moisture sensor, a nitrogen sensor, a phosphorus sensor, a potassium sensor, a hydrogen ion concentration index sensor, a soil conductivity sensor, an air temperature sensor, an air humidity sensor, and a light sensor.

[0013] Furthermore, the wireless data transmission terminal includes a master device and a slave device. The master device is connected to a host computer, and the slave device is connected to a sensor module. The slave device enables each sensor in the sensor module to convert RS485 serial signals into wireless signals, which are then propagated within a preset range and received by the master device. The master device converts the received wireless signals into RS485 signals and uses a MAX485 chip to convert the RS485 differential signals into USB TTL level signals, thereby enabling communication with the host computer's serial port.

[0014] Furthermore, the clear water irrigation prediction unit includes a data preparation and preprocessing module, an optimized Penman formula module, a deep learning model architecture, a first irrigation decision module, and a visualization and output module.

[0015] The data preparation and preprocessing module is used to preprocess the sensor data;

[0016] The optimized Penman formula module is used to match the indoor planting of citrus seedlings;

[0017] The deep learning model architecture is used to integrate multiple deep learning algorithms and train them based on preprocessed sensor data.

[0018] The first irrigation decision module is used to predict water demand and the conversion of water demand to irrigation soaking time. The irrigation soaking time is the water demand volume divided by the water absorption rate per unit of citrus seedling. The water demand volume is the soil area per unit of citrus seedling multiplied by the water demand. The water absorption rate per unit of citrus seedling is the root water absorption rate multiplied by the root weight per unit of citrus seedling.

[0019] The visualization and output module is used to display the convergence status of the deep learning model and, based on the results of the first irrigation decision module, output a clean water irrigation report for tidal irrigation.

[0020] Furthermore, the optimized Penman formula module replaces the solar radiation parameter in the Penman formula with the artificial illumination equivalent.

[0021] Furthermore, the deep learning algorithm in the deep learning model architecture includes a convolutional neural network, a long short-term memory network, and a gated recurrent unit. The convolutional neural network uses one-dimensional convolution to extract local spatiotemporal features of sensor data, captures short-term fluctuations by setting a time step, and compresses the feature dimension within the pooling layer. The long short-term memory network retains long-term memory at the irrigation cycle level to transmit the complete sequence state. The gated recurrent unit captures the trend of evapotranspiration changes and uses L2 regularization to constrain the weights.

[0022] Furthermore, the nutrient solution ratio and control unit includes a sensor data compensation module, a growth stage adaptive strategy module, and a second irrigation decision module;

[0023] The sensor data compensation module is used to combine soil conductivity interference correction and hydrogen ion concentration index buffering effect to perform targeted compensation and improvement on sensor data.

[0024] The growth stage adaptive strategy module is used to automatically select different growth stages based on the number of planting days in order to match the optimal nitrogen, phosphorus and potassium content.

[0025] The second irrigation decision module is used to process the compensated sensor data and output the nutrient solution concentration and nutrient solution irrigation soaking time according to the optimal nitrogen, phosphorus and potassium content. The nutrient solution concentration is obtained by the difference between the actual nitrogen, phosphorus and potassium concentration and the optimal nitrogen, phosphorus and potassium concentration. The nutrient solution irrigation soaking time is the total nutrients required by the citrus seedlings divided by the total absorption rate of the citrus seedlings in the seedbed. The total nutrients required by the citrus seedlings is the nutrient solution concentration multiplied by the volume of nutrient solution in the seedbed. The total absorption rate of the citrus seedlings in the seedbed is the root absorption rate multiplied by the root weight.

[0026] Furthermore, the formula for correcting soil electrical conductivity disturbance is as follows:

[0027] Corrected value for soil electrical conductivity of nitrogen = nitrogen sensor value - current soil electrical conductivity value * 0.18;

[0028] Corrected value for soil electrical conductivity of phosphorus = Sensor value of phosphorus - Current soil electrical conductivity value * 0.12;

[0029] Corrected value for soil electrical conductivity of potassium = Potassium sensor value - Current soil electrical conductivity value * 0.25;

[0030] The formula for the hydrogen ion concentration index buffering effect is as follows:

[0031] Nitrogen coefficient: 1.2 - 0.15 * (ph - 6.5), ph > 6.5; 1.2 + 0.1 * (6.5 - ph), ph < 6.5;

[0032] Nitrogen correction value = Nitrogen soil electrical conductivity correction value * Nitrogen coefficient

[0033] Phosphorus coefficients: 1.1 - 0.12 * (ph - 6.2), ph > 6.2; 1.1 + 0.2 * (6.2 - ph), ph < 6.2;

[0034] The correction value for phosphorus = the correction value for soil electrical conductivity of phosphorus * the coefficient for phosphorus

[0035] Potassium coefficient: 1.0 - 0.1 * (pH - 6.8), pH > 6.8; 1.0 + 0.05 * (6.8 - pH), pH < 6.8;

[0036] The correction value for potassium = the EC correction value for potassium * the coefficient for potassium.

[0037] A real-time decision-making method for citrus irrigation and fertilization, implemented based on the aforementioned real-time decision-making system for citrus irrigation and fertilization, is characterized in that the method includes:

[0038] Collect real-time sensor data and preprocess the sensor data;

[0039] The deep learning model architecture is trained based on the preprocessed sensor data;

[0040] The trained deep learning model architecture is invoked to predict water demand and measure the soil area per unit of citrus seedling and the root weight per unit of citrus seedling.

[0041] Based on water demand, root water absorption rate, soil area per unit citrus seedling, and root weight per unit citrus seedling, calculate irrigation soaking time and output a clear water irrigation report for tidal irrigation.

[0042] By combining soil conductivity interference correction and hydrogen ion concentration index buffering effect, targeted compensation and improvement of sensor data are carried out.

[0043] The system automatically selects different growth stages based on the number of planting days to match the optimal nitrogen, phosphorus, and potassium content.

[0044] The sensor data is processed after compensation, and the nutrient solution concentration and nutrient solution irrigation soaking time are output.

[0045] The present invention has the following advantages over the prior art:

[0046] 1. This invention overcomes the bottlenecks of traditional irrigation systems, such as reliance on manual inspection, delayed decision-making, and inability to update and process data in real time, by constructing a full-process real-time closed-loop system of "real-time acquisition of multi-sensor data - deep learning prediction - automatic and precise control of actuators".

[0047] 2. This invention significantly improves the prediction accuracy of soaking time for clean water irrigation by developing a deep learning model that fuses spatiotemporal features, breaks through the structural limitations of existing deep learning models, and specifically compensates for the shortcomings of the traditional Penman formula in indoor tidal irrigation scenarios by optimizing the Penman formula.

[0048] 3. This invention proposes a soil conductivity interference correction / hydrogen ion concentration index buffering linkage compensation, growth stage adaptation, and a multi-objective decision engine, significantly improving the prediction model for nutrient solution irrigation soaking time and concentration, and specifically addressing the pain points of rigid nutrient solution formulations and low effectiveness in existing tidal irrigation technologies. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0050] Figure 1 This is a flowchart of the real-time decision-making system for citrus irrigation and fertilization according to an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of the real-time data stream monitoring unit according to an embodiment of the present invention.

[0052] Figure 3 This is a structural principle block diagram of the clear water irrigation prediction unit according to an embodiment of the present invention.

[0053] Figure 4 This is a structural principle block diagram of the nutrient solution ratio and control unit according to an embodiment of the present invention. Detailed Implementation

[0054] To better understand the invention, various aspects of the invention will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are merely illustrative of exemplary embodiments of the invention and are not intended to limit the scope of the invention in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0055] In the accompanying drawings, the size, dimensions, and shapes of the elements have been slightly adjusted for ease of illustration. The drawings are for illustrative purposes only and are not strictly to scale. As used herein, the terms “approximately,” “about,” and similar terms are used to indicate approximation, not degree, and are intended to illustrate inherent deviations in measured or calculated values ​​that will be recognized by those skilled in the art. Furthermore, the order in which the steps are described in this invention does not necessarily indicate the order in which these steps occur in actual operation, unless otherwise expressly defined or deduced from the context.

[0056] It should also be understood that expressions such as "comprising," "including," "having," "containing," and / or "comprising" are open-ended rather than closed-ended expressions in this specification, indicating the presence of the stated features, elements, and / or components, but not excluding the presence of one or more other features, elements, components, and / or combinations thereof. Furthermore, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features, not just individual elements in the list. Additionally, when describing embodiments of the invention, the word "may" is used to mean "one or more embodiments of the invention." And the term "exemplary" is intended to refer to examples or illustrations.

[0057] Unless otherwise specified, all terms used herein (including engineering and technical terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that, unless expressly stated herein, terms defined in common dictionaries shall be interpreted as having the meaning consistent with their meaning in the context of the relevant art, and not in an idealized or overly formalized sense.

[0058] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0059] Example:

[0060] like Figure 1As shown in the figure, this embodiment provides a real-time decision-making system for citrus irrigation and fertilization based on multimodal sensing and deep learning. The system includes a real-time data stream monitoring unit, a clean water irrigation prediction unit, and a nutrient solution ratio and control unit. The specific descriptions of each unit are as follows:

[0061] like Figure 2 As shown, the real-time data stream monitoring unit is used to collect real-time sensor data and build a dual-thread task scheduler to ensure uninterrupted operation of the entire process. It includes a sensor module, a wireless data transmission terminal, and a host computer. The wireless data transmission terminal is connected to the sensor module and the host computer, respectively. The host computer is a PC.

[0062] Furthermore, the sensor module includes a soil temperature sensor, a soil moisture sensor, a nitrogen sensor, a phosphorus sensor, a potassium sensor, a hydrogen ion concentration index (PH) sensor, a soil electrical conductivity (EC) sensor, an air temperature sensor, an air humidity sensor, and a light sensor. These sensors can collect various values ​​of the soil and the growing environment of citrus seedlings in real time. All of these sensors are powered by 24V. The desired results are obtained by analyzing the sensor data. These sensors communicate via RS485 protocol, which has the advantages of high accuracy and low latency.

[0063] Furthermore, the wireless data transmission terminal includes a master device and a slave device. Both the master device and the slave device are developed using LoRa wireless spread spectrum technology. The master device is connected to a host computer, and the slave device is connected to a sensor module. The slave device enables each sensor in the sensor module to convert RS485 serial signals into wireless signals, which are then propagated within a preset range and received by the master device. The master device converts the received wireless signals back into RS485 signals. The master device has an internal adapter that can convert the RS485 differential signals into USB TTL level signals using a MAX485 chip, thereby enabling communication with the host computer's serial port so that the corresponding program on the host computer can process the signals.

[0064] Furthermore, the dual-threaded task scheduling in this embodiment is built on a Python program. The Python program has two parts: the first part is used to collect, process, and save data from various sensors, and the second part has fully automated capabilities for file monitoring, data conversion, secure archiving, and merged storage. They are packaged independently and can run continuously without interruption to ensure the normal operation of the system.

[0065] Specifically, during the execution of the first part of the Python program, a query frame is sent to the host computer's serial port. The 01 and 02 at the beginning of the query frame specify the slave device, the following 03 indicates reading the holding register, and then the starting address, data length, and CRC checksum to ensure data integrity. When the query frame is sent to the host computer's serial port, it is then transmitted to the slave device via the master device of the wireless data transmission terminal. The sensor receives the signal and sends back a response frame. This response frame is transmitted back to the master device via the slave device and then received by the host computer's serial port. The Python program processes and analyzes the response frame to obtain the sensor's values. The program is set to collect sensor data every 10 minutes and to package every 72 data points into an Excel file and store it locally. The program uses BlockingScheduler to create an independent task scheduling thread. By setting the second-level polling interval through the interval parameter, the main function is executed periodically. After scheduler.start is started, APScheduler maintains an event loop internally to continuously detect task triggering conditions. There is no need to manually write a while loop, which ensures that the program runs uninterruptedly and continuously collects and saves the required sensor data.

[0066] Specifically, the Python program in the second part implements an industrial-grade Excel file monitoring and processing system. First, based on the watchdog Observer pattern, it captures newly generated Excel files in real time. Then, it performs Excel to CSV conversion, data merging, exception handling, and file locking, debouncing mechanisms, and integrity checks. The program optimizes the environment and configuration, suppressing deep learning framework logs through environment variables to focus on business logic. Configuration parameters are centrally managed for easy maintenance. It also implements enhanced file monitoring functionality, using four layers of checks—file extension, occupancy status, write stability, and ZIP structure—to avoid processing incomplete files and employs an exponential backoff retry mechanism to address network storage latency issues. Thread-safe processing is implemented with a dual-lock design of thread-level `processing_lock` and file-level `portalocker` to prevent resource contention. A data archiving strategy is also implemented, with internal timestamps including microseconds to avoid filename conflicts in high-concurrency scenarios. `os.rename` is used instead of copy operations to ensure atomic archiving. Similarly, the program runs continuously, constantly monitoring for the generation of new Excel files. When the first part of the Python code collects sensor data and saves it as an Excel file, the second part of the Python program will execute commands to convert the newly generated Excel data into CSV data and add it to the original CSV file. This allows the sensor data in the CSV file to be updated in real time, so that the CSV file can be input into the corresponding model algorithm.

[0067] like Figure 3 As shown, the clear water irrigation prediction unit is used to calculate the clear water soaking time of tidal irrigation in real time based on the optimized Penman formula, using a deep learning model architecture and irrigation decision logic. It includes a data preparation and preprocessing module, an optimized Penman formula module, a deep learning model architecture, a first irrigation decision module, and a visualization and output module.

[0068] Furthermore, the data preparation and preprocessing module is used to preprocess the sensor data, primarily handling the sensor data within CSV files. When the sensor data in the CSV file is updated, data is read with locking, using PortaLocker to implement file-level locking to prevent data corruption caused by simultaneous writes from multiple processes. A 15-second timeout is set to avoid deadlocks affecting system operation. The module also parses the time within the CSV file to ensure the integrity of the time series. Finally, the sensor data is standardized by unifying the units of the data for easier subsequent processing and analysis.

[0069] Furthermore, the Penman formula module was optimized to match the indoor planting of citrus seedlings. Based on the Penman formula, some optimizations were made to make it suitable for indoor planting of citrus seedlings.

[0070] The Penman formula is a method for calculating crop evapotranspiration (ET0), with units of mm / day. The specific formula is as follows:

[0071]

[0072] Where ET0 is potential evapotranspiration (usually expressed in millimeters per day); RS is solar radiation (MJ / m²). 2 • G is the soil heat flux density (usually negligible, especially in atmospheric evapotranspiration calculations); γ is the dry and wet pressure constant (approximately 0.067 kPa / ℃); T is the air temperature (℃); u is the wind speed (m / s); e s The saturated water vapor pressure (kPa) can be calculated from the air temperature, specifically: es = 6.11 * 10^(7.5 * T / (T + 237.3)), where T is the air temperature (°C); ea: the actual water vapor pressure (kPa), which can be calculated from the relative humidity, specifically: e a =e s *RH / 100, where RH is relative humidity (%); Δ is the rate of change of saturated water vapor pressure with temperature (KPa / ℃), which can be calculated from the temperature, specifically: Δ=4098*e s / (T+237.3)^2.

[0073] The above is the traditional Penman formula. This embodiment optimizes the Penman formula for indoor citrus seedling cultivation. The traditional Penman formula relies on natural solar radiation data and cannot be directly applied to indoor citrus seedling cultivation scenarios without sunlight. Therefore, the solar radiation parameter in the Penman formula is replaced with the artificial light equivalent value. Based on experience and expert knowledge, the formula is transformed into 1 Lux = 0.0079 MJ / m 2 Furthermore, the traditional Penman formula is designed based on natural meteorological environments, with wind speed parameters used to characterize aerodynamic effects (such as water vapor exchange between the evaporation surface and the air). However, in enclosed indoor planting environments (such as plant factories and tissue culture seedling rooms), airflow is precisely controlled by artificial ventilation systems, and there is no natural wind disturbance. Directly using traditional wind speed parameters will lead to distortion in evapotranspiration calculations. Under indoor artificial lighting conditions, when the air velocity is below 0.15 m / s, the contribution of the wind speed term in the Penman formula to the evapotranspiration calculation results is relatively small. Based on this, setting the wind speed parameter to 0 simplifies the model complexity while ensuring calculation accuracy, making it more suitable for indoor planting environments.

[0074] Furthermore, the deep learning model architecture is used to fuse multiple deep learning algorithms and train them based on preprocessed sensor data. These deep learning algorithms include Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs). The GRU (Gated Recurrent Unit) is a deep learning model architecture consisting of a convolutional neural network, a long short-term memory network, and a gated recurrent unit cascaded together. The convolutional neural network uses Conv1D one-dimensional convolution to extract local spatiotemporal features of sensor data, and captures short-term fluctuations by setting a time step of 5. It also compresses the feature dimension within the pooling layer to retain significant changes. The long short-term memory network retains long-term memory at the irrigation cycle level through 128 units, and return_sequences=True is used to pass complete sequence states. The gated recurrent unit captures the trend of evapotranspiration changes and uses L2 regularization to constrain the weights to prevent over-reliance on a single sensor. It also employs a multi-stage training optimization strategy. The learning rate is initially set to 0.001 to balance convergence speed and stability. The early stopping mechanism terminates the training if the verification loss does not improve after 10 rounds to avoid ineffective training.

[0075] Furthermore, the first irrigation decision module is used to predict water demand and the conversion of water demand into irrigation soaking time. After sensor data is input into the trained deep learning model architecture, a recursive prediction method is used, based on the latest time window, to simulate the time dependence of a real irrigation system. After obtaining the water demand, the root water absorption rate is obtained through an expert database and experience. Then, the soil area per unit of citrus seedling and the root weight per unit of citrus seedling are measured. The water demand volume is the soil area per unit of citrus seedling multiplied by the water demand, and the water absorption rate per unit of citrus seedling is the root water absorption rate multiplied by the root weight per unit of citrus seedling. Therefore, the irrigation soaking time is the water demand volume divided by the water absorption rate per unit of citrus seedling. After conversion, the irrigation soaking time can be well matched with the irrigation soaking time required for tidal irrigation, which is convenient for planting management.

[0076] Furthermore, the visualization and output module is used to display the convergence status of the deep learning model, making it easier for users to understand the model's status. Based on the results of the first irrigation decision module, it outputs a clear water irrigation report for tidal irrigation, helping and guiding users to conduct better planting management.

[0077] like Figure 4 As shown, the nutrient solution ratio and control unit is used to dynamically adjust the target concentrations of nitrogen, phosphorus, and potassium in real time and calculate the soaking time of nutrient solution for tidal irrigation by utilizing soil conductivity interference correction / hydrogen ion concentration index buffer linkage compensation and growth stage adaptive strategy. It includes a sensor data compensation module, a growth stage adaptive strategy module and a second irrigation decision module.

[0078] Furthermore, the sensor data compensation module combines soil electrical conductivity (EC) interference correction with the buffering effect of pH index to provide targeted compensation and improvement for sensor data. Traditional single compensation models only use nitrogen, phosphorus, and potassium concentrations for judgment and analysis, which cannot be well matched with existing citrus irrigation planting systems and has the disadvantage of inaccuracy. Excessively high soil electrical conductivity can cause electrical interference between ions, leading to deviations in concentration readings. Therefore, EC values ​​are used to correct for interference in nitrogen, phosphorus, and potassium readings. Based on expert knowledge and experience, the nutrient compensation coefficients for nitrogen, phosphorus, and potassium are 0.18, 0.12, and 0.25, respectively. The EC correction values ​​for nitrogen, phosphorus, and potassium are equal to the sensor values ​​for nitrogen, phosphorus, and potassium minus the current soil EC multiplied by their respective nutrient compensation coefficients. Simultaneously, pH value also affects nutrient availability. The optimal pH values ​​for nitrogen, phosphorus, and potassium are 6.5, 6.2, and 6.8, respectively. When the pH value is at the optimal value, the availability coefficients of nitrogen, phosphorus, and potassium are 1.2, 1.1, and 1.0, respectively. The coefficients for nitrogen, phosphorus, and potassium that are too high are 0.15, 0.12, and 0.1, respectively, and the coefficients for those that are too low are 0.1, 0.2, and 0.05, respectively.

[0079] The formula for correcting soil electrical conductivity disturbances is as follows:

[0080] Corrected value for soil electrical conductivity of nitrogen = nitrogen sensor value - current soil electrical conductivity value * 0.18;

[0081] Corrected value for soil electrical conductivity of phosphorus = Sensor value of phosphorus - Current soil electrical conductivity value * 0.12;

[0082] Corrected value for soil electrical conductivity of potassium = Potassium sensor value - Current soil electrical conductivity value * 0.25;

[0083] The formula for the hydrogen ion concentration index buffering effect is as follows:

[0084] Nitrogen coefficient: 1.2 - 0.15 * (ph - 6.5), ph > 6.5; 1.2 + 0.1 * (6.5 - ph), ph < 6.5;

[0085] Nitrogen correction value = Nitrogen soil electrical conductivity correction value * Nitrogen coefficient

[0086] Phosphorus coefficients: 1.1 - 0.12 * (ph - 6.2), ph > 6.2; 1.1 + 0.2 * (6.2 - ph), ph < 6.2;

[0087] The correction value for phosphorus = the correction value for soil electrical conductivity of phosphorus * the coefficient for phosphorus

[0088] Potassium coefficient: 1.0 - 0.1 * (pH - 6.8), pH > 6.8; 1.0 + 0.05 * (6.8 - pH), pH < 6.8;

[0089] The correction value for potassium = the EC correction value for potassium * the coefficient for potassium.

[0090] After combining soil conductivity interference correction and hydrogen ion concentration index buffering, the obtained nitrogen, phosphorus and potassium correction values ​​can be closer to the actual effective nitrogen, phosphorus and potassium values.

[0091] Furthermore, the growth stage adaptive strategy module automatically selects different growth stages based on the number of planting days to match the optimal nitrogen, phosphorus, and potassium (NPK) content. This module automatically matches the growth stage based on the number of planting days. Different growth stages of citrus seedlings require different levels of NPK. Specifically, for 30 days or less, the optimal NPK levels are 30 mg / kg, 15 mg / kg, and 25 mg / kg; for 30 days or more but less than or equal to 90 days, the optimal levels are 45 mg / kg, 20 mg / kg, and 35 mg / kg; and for more than 90 days, the optimal levels are 40 mg / kg, 25 mg / kg, and 40 mg / kg. The optimal NPK content is matched by the set planting time and the current actual time.

[0092] Furthermore, the second irrigation decision module processes the compensated sensor data and outputs the nutrient solution concentration and nutrient solution soaking time based on the optimal nitrogen, phosphorus, and potassium content. The nutrient solution concentration is calculated by the difference between the actual and optimal nitrogen, phosphorus, and potassium concentrations. Based on expert experience, the root absorption rate is 0.7 ml / h / g. The total nutrients required by the citrus seedlings are calculated by multiplying the nutrient solution concentration by the volume of nutrient solution in the seedbed. The root weight of the citrus seedlings is determined by measurement, and the total absorption rate of the citrus seedlings in the seedbed is calculated by multiplying the root absorption rate by the root weight. Therefore, the nutrient solution soaking time is the total required nutrients divided by the total absorption rate. This method can be well matched with tidal irrigation systems for citrus seedlings, providing users with reference and guidance.

[0093] This embodiment also provides a real-time decision-making method for citrus water and fertilizer management. This method is based on the real-time decision-making system for citrus water and fertilizer management described in the above embodiment, and specifically includes:

[0094] S1. Collect real-time sensor data and preprocess the sensor data;

[0095] S2. Based on the preprocessed sensor data, train the deep learning model architecture;

[0096] S3. Call the trained deep learning model architecture to predict water demand and measure the soil area per unit of citrus seedling and the root weight per unit of citrus seedling.

[0097] S4. Calculate the irrigation soaking time based on water demand, root water absorption rate, soil area per unit citrus seedling, and root weight per unit citrus seedling, and output a report on clear water irrigation for tidal irrigation.

[0098] S5. Combine soil conductivity interference correction and hydrogen ion concentration index buffering effect to perform targeted compensation and improvement on sensor data.

[0099] S6. Automatically select different growth stages based on the number of planting days to match the optimal nitrogen, phosphorus, and potassium content;

[0100] S7. Process the compensated sensor data and output the nutrient solution concentration and nutrient solution irrigation soaking time.

[0101] In summary, this invention provides a model and method for automatically and continuously collecting and processing multi-source sensor data in real time, and for automatically and accurately predicting the soaking time and concentration required for tidal irrigation of citrus. This enables precise irrigation, improves the utilization rate of water and fertilizer resources, reduces agricultural production costs, and ultimately enhances the yield and quality of citrus, thus having significant practical significance and application value.

[0102] It should be noted that, in this invention, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A real-time decision-making system for citrus irrigation and fertilization based on multimodal sensing and deep learning, characterized in that, Includes a real-time data stream monitoring unit, a clean water irrigation prediction unit, and a nutrient solution ratio and control unit; The real-time data stream monitoring unit is used to collect real-time sensor data and build a dual-thread task schedule to ensure uninterrupted operation of the entire process. The clear water irrigation prediction unit is used to calculate the clear water soaking time of tidal irrigation in real time based on the optimized Penman formula, using a deep learning model architecture and irrigation decision logic. The nutrient solution ratio and control unit is used to dynamically adjust the target concentrations of nitrogen, phosphorus, and potassium in real time and calculate the soaking time of nutrient solution for tidal irrigation by utilizing soil conductivity interference correction / hydrogen ion concentration index buffer linkage compensation and growth stage adaptive strategies.

2. The real-time decision-making system for citrus water and fertilizer management according to claim 1, characterized in that, The real-time data stream monitoring unit includes a sensor module, a wireless data transmission terminal, and a host computer. The wireless data transmission terminal is connected to both the sensor module and the host computer.

3. The real-time decision-making system for citrus water and fertilizer management according to claim 2, characterized in that, The sensor module includes a soil temperature sensor, a soil moisture sensor, a nitrogen sensor, a phosphorus sensor, a potassium sensor, a hydrogen ion concentration index sensor, a soil conductivity sensor, an air temperature sensor, an air humidity sensor, and a light sensor.

4. The real-time decision-making system for citrus water and fertilizer management according to claim 2, characterized in that, The wireless data transmission terminal includes a master device and a slave device. The master device is connected to a host computer, and the slave device is connected to a sensor module. The slave device enables each sensor in the sensor module to convert RS485 serial signals into wireless signals, which are then propagated within a preset range and received by the master device. The master device converts the received wireless signals into RS485 signals and uses a MAX485 chip to convert the RS485 differential signals into USB TTL level signals, thereby enabling communication with the host computer's serial port.

5. The real-time decision-making system for citrus water and fertilizer management according to claim 1, characterized in that, The clear water irrigation prediction unit includes a data preparation and preprocessing module, an optimized Penman formula module, a deep learning model architecture, a first irrigation decision module, and a visualization and output module. The data preparation and preprocessing module is used to preprocess the sensor data; The optimized Penman formula module is used to match the indoor planting of citrus seedlings; The deep learning model architecture is used to integrate multiple deep learning algorithms and train them based on preprocessed sensor data. The first irrigation decision module is used to predict water demand and the conversion of water demand to irrigation soaking time. The irrigation soaking time is the water demand volume divided by the water absorption rate per unit of citrus seedling. The water demand volume is the soil area per unit of citrus seedling multiplied by the water demand. The water absorption rate per unit of citrus seedling is the root water absorption rate multiplied by the root weight per unit of citrus seedling. The visualization and output module is used to display the convergence status of the deep learning model and, based on the results of the first irrigation decision module, output a clean water irrigation report for tidal irrigation.

6. The real-time decision-making system for citrus water and fertilizer management according to claim 5, characterized in that, The optimized Penman formula module replaces the solar radiation parameter in the Penman formula with the artificial illumination equivalent.

7. The real-time decision-making system for citrus water and fertilizer management according to claim 5, characterized in that, The deep learning algorithm in the deep learning model architecture includes a convolutional neural network, a long short-term memory network, and a gated recurrent unit. The convolutional neural network uses one-dimensional convolution to extract local spatiotemporal features of sensor data, captures short-term fluctuations by setting a time step, and compresses the feature dimension within the pooling layer. The long short-term memory network retains long-term memory at the irrigation cycle level to transmit the complete sequence state. The gated recurrent unit captures the trend of evapotranspiration changes and uses L2 regularization to constrain the weights.

8. The real-time decision-making system for citrus water and fertilizer management according to claim 1, characterized in that, The nutrient solution ratio and control unit includes a sensor data compensation module, a growth stage adaptive strategy module, and a second irrigation decision module. The sensor data compensation module is used to combine soil conductivity interference correction and hydrogen ion concentration index buffering effect to perform targeted compensation and improvement on sensor data. The growth stage adaptive strategy module is used to automatically select different growth stages based on the number of planting days in order to match the optimal nitrogen, phosphorus and potassium content. The second irrigation decision module is used to process the compensated sensor data and output the nutrient solution concentration and nutrient solution irrigation soaking time according to the optimal nitrogen, phosphorus and potassium content. The nutrient solution concentration is obtained by the difference between the actual nitrogen, phosphorus and potassium concentration and the optimal nitrogen, phosphorus and potassium concentration. The nutrient solution irrigation soaking time is the total nutrients required by the citrus seedlings divided by the total absorption rate of the citrus seedlings in the seedbed. The total nutrients required by the citrus seedlings is the nutrient solution concentration multiplied by the volume of nutrient solution in the seedbed. The total absorption rate of the citrus seedlings in the seedbed is the root absorption rate multiplied by the root weight.

9. The real-time decision-making system for citrus water and fertilizer management according to claim 8, characterized in that, The formula for correcting soil electrical conductivity disturbance is as follows: Corrected value for soil electrical conductivity of nitrogen = nitrogen sensor value - current soil electrical conductivity value * 0.18; Corrected value for soil electrical conductivity of phosphorus = Sensor value of phosphorus - Current soil electrical conductivity value * 0.12; Corrected value for soil electrical conductivity of potassium = Potassium sensor value - Current soil electrical conductivity value * 0.25; The formula for the hydrogen ion concentration index buffering effect is as follows: Nitrogen coefficient: 1.2 - 0.15 * (ph - 6.5), ph > 6.5; 1.2 + 0.1 * (6.5 - ph), ph < 6.5; Nitrogen correction value = Nitrogen soil electrical conductivity correction value * Nitrogen coefficient Phosphorus coefficients: 1.1 - 0.12 * (ph - 6.2), ph > 6.2; 1.1 + 0.2 * (6.2 - ph), ph < 6.2; The correction value for phosphorus = the correction value for soil electrical conductivity of phosphorus * the coefficient for phosphorus Potassium coefficient: 1.0 - 0.1 * (pH - 6.8), pH > 6.8; 1.0 + 0.05 * (6.8 - pH), pH < 6.8; The correction value for potassium = the EC correction value for potassium * the coefficient for potassium.

10. A real-time decision-making method for citrus water and fertilizer, implemented based on the real-time decision-making system for citrus water and fertilizer as described in any one of claims 1-9, characterized in that, The method includes: Collect real-time sensor data and preprocess the sensor data; The deep learning model architecture is trained based on the preprocessed sensor data; The trained deep learning model architecture is invoked to predict water demand and measure the soil area per unit of citrus seedling and the root weight per unit of citrus seedling. Based on water demand, root water absorption rate, soil area per unit citrus seedling, and root weight per unit citrus seedling, calculate irrigation soaking time and output a clear water irrigation report for tidal irrigation. By combining soil conductivity interference correction and hydrogen ion concentration index buffering effect, targeted compensation and improvement of sensor data are carried out. The system automatically selects different growth stages based on the number of planting days to match the optimal nitrogen, phosphorus, and potassium content. The compensated sensor data is processed, and the nutrient solution concentration and nutrient solution irrigation soaking time are output according to the optimal nitrogen, phosphorus and potassium content.

Images