A controllable fog seedling raising method suitable for potato seed production

Abstract

The application relates to the field of agricultural biotechnology, and discloses a controllable mist breeding method suitable for potato seed potato production. The method constructs a four-layer closed-loop regulation framework, collects environmental parameters and plant growth states in real time through a multi-source sensor and a phenotype monitoring unit, generates a joint state vector by using a multi-scale time sequence coding model, dynamically solves an optimal combination of light, temperature and humidity, CO2 concentration and atomization period under the constraint of a multi-objective optimization function, and drives an actuator to be cooperatively regulated. The system automatically switches strategies according to the stages of rooting, stem and leaf growth, tuber induction and enlargement. The application improves the number of tuber proliferation, reduces energy consumption, improves the utilization rate of nutrient solution, and realizes high-yield, high-efficiency and high-quality seed potato production.

Description

Technical Field

[0001] This invention belongs to the field of agricultural biotechnology, specifically relating to a controlled fog cultivation method suitable for potato seed production. Background Technology

[0002] With the development of modern agriculture towards intelligence and intensification, aeroponic technology has shown great application potential in potato seed production due to its advantages such as water and fertilizer conservation, sufficient root oxygen supply, and fewer pests and diseases. The aeroponic system achieves efficient nutrient supply and environmental control by directly spraying atomized nutrient solution onto the suspended root system. Its core lies in the precise and coordinated management of multiple environmental factors, including light, temperature, humidity, CO2 concentration, and atomization cycle. Environmental control not only directly affects plant photosynthetic efficiency, tuber formation, and starch accumulation, but also relates to energy consumption and production costs, making it a crucial factor determining seed potato yield and quality.

[0003] A controlled-air seedling cultivation method for potato seed production focuses on dynamically adjusting environmental parameter combinations at different growth stages to maximize growth potential and ensure the physiological health of seed potatoes. This method requires a control system capable of sensing the real-time physiological state of the plant and adaptively adjusting the operating strategies of each execution unit based on external climate disturbances and internal metabolic demands, thereby maintaining the optimal growth window in a complex, nonlinear environment.

[0004] Existing technologies mostly employ preset thresholds or logic control strategies based on empirical rules, such as turning on the atomizer at fixed times or adjusting the supplemental light intensity according to the diurnal rhythm. These methods struggle to cope with the strong coupling effects between multiple factors and the dynamic evolution of the potato growth cycle, leading to rigid control strategies. Excessive energy supply may waste electricity and water resources, while environmental deviations from the ideal range may inhibit tuber induction or reduce starch synthesis efficiency. In large-scale aeroponic scenarios, frequent environmental disturbances and individual differences mean that traditional control models cannot achieve refined and personalized dynamic optimization, severely restricting the stability and economic efficiency of seed potato production. Therefore, there is an urgent need for an intelligent environmental control mechanism capable of autonomous learning, continuous optimization, and adaptation to changes throughout the entire growth cycle to overcome the current challenges in balancing energy efficiency and biomass output in aeroponic systems. Summary of the Invention

[0005] This invention provides a controllable aeroponic seedling cultivation method suitable for potato seed production, aiming to solve the technical problems of difficulty in long-term optimization of control parameters, low energy efficiency, and insufficient stimulation of plant growth potential caused by the coupling of multiple factors such as light, temperature, humidity, carbon dioxide concentration, and aeroponic interval in the aeroponic environment. In existing technologies, aeroponic seedling cultivation systems generally adopt simple logic control strategies based on fixed setpoints or threshold triggers, such as activating a humidifier when the humidity is below a certain threshold, or turning on the light source within a fixed time period. Such methods do not consider the dynamic interaction between various environmental factors and their nonlinear cumulative effect on the physiological response of potato seed tubers, resulting in an inability to accurately match the actual needs of the plant at different growth stages, leading to ineffective consumption of electrical energy, water, fertilizer, and gas resources, while simultaneously inhibiting tuber formation and proliferation efficiency.

[0006] This invention provides a method for controlled fog cultivation of seed potato seed tubers, comprising:

[0007] A multi-source sensor array deployed in the aeroponic chamber synchronously collects environmental parameter sequences, including light intensity, spectral distribution, air temperature, relative humidity, carbon dioxide concentration, nutrient solution temperature, atomization pressure, droplet size distribution, and atomization start and stop timestamps.

[0008] Real-time growth status data of potato seed tubers are obtained through a plant phenotypic monitoring unit. The growth status data includes stem and leaf elongation rate, leaf unfolding area, chlorophyll fluorescence parameters, tuber initial swelling time point, tuber number and individual volume growth rate.

[0009] The environmental parameter sequence and growth state data are input into a preset multi-scale temporal coding model to generate a joint state vector representing the characteristics of the current growth stage.

[0010] Based on the joint state vector, under the constraints of the pre-constructed multi-objective optimization function, the optimal setpoints of each environmental factor in the next control cycle are solved. The multi-objective optimization function takes maximizing the tuber proliferation rate, minimizing the energy consumption per unit yield, and maximizing the nutrient solution utilization rate as optimization objectives, and introduces the stage-specific constraints of the physiological response of potato seed tubers.

[0011] Based on the solution results, control commands are generated to drive the light adjustment device, temperature and humidity control unit, carbon dioxide supply system and atomization actuator to work together to complete the closed-loop control of the atomization environment.

[0012] Preferably, environmental parameter sequences are synchronously collected by a multi-source sensor array deployed in the aeroponic chamber, including:

[0013] The full-spectrum light sensor, platinum resistance temperature sensor, capacitive humidity sensor, infrared carbon dioxide concentration detection module, ultrasonic atomizing particle size analyzer, and high-precision pressure transmitter are distributed and installed in the atomization chamber.

[0014] The full-spectrum light sensor captures irradiance changes in the red, far-red, blue, and ultraviolet bands.

[0015] The air temperature was monitored at three different height levels—top, middle, and bottom—of the aeroponic chamber using the platinum resistance temperature sensor.

[0016] The relative humidity is measured inside the plant canopy using the capacitive humidity sensor.

[0017] The infrared carbon dioxide concentration detection module continuously monitors the carbon dioxide concentration in the return air duct.

[0018] The median diameter of the droplet volume is measured in real time at the outlet of the atomizing nozzle using the ultrasonic atomizing particle size analyzer.

[0019] The high-precision pressure transmitter feeds back the working pressure of the atomization system on the main atomization pipeline.

[0020] All sensor data is transmitted synchronously to the central processing unit via industrial Ethernet.

[0021] Preferably, real-time growth status data of potato seed tubers are obtained through a plant phenotypic monitoring unit, including:

[0022] A high-resolution industrial camera mounted on the top takes orthophotos of the plant canopy every 2 hours. The leaf outlines are extracted and the projected area is calculated using an image segmentation algorithm, which in turn calculates the elongation rate of the stems and leaves.

[0023] By using a side-mounted near-infrared imaging module to obliquely irradiate the base area of ​​the plant in the 700 nm to 1000 nm band, the water gradient changes in the initial tuber enlargement area are identified. When the local reflectance decreases by more than 15% of the threshold and lasts for more than 2 hours, it is determined to be the time point of initial tuber enlargement.

[0024] The maximum photochemical efficiency Fv / Fm and the actual photochemical efficiency ΦPSII were measured at 30-minute intervals using an embedded chlorophyll fluorescence detector. Fv represents variable fluorescence, and Fm represents maximum fluorescence.

[0025] Preferably, the environmental parameter sequence and growth state data are input into a preset multi-scale temporal coding model to generate a joint state vector representing the characteristics of the current growth stage, including:

[0026] Slide window slicing is performed on the original environmental parameter sequence to generate local time segments;

[0027] Channel attention weighting is applied to each local time segment to highlight environmental factors that are sensitive to the current growth stage;

[0028] The weighted sequence of segments is input into the global context encoder, which outputs a joint state vector of fixed dimensions.

[0029] Preferably, the original environmental parameter sequence is sliced ​​using a sliding window to generate local time segments, including:

[0030] The environmental parameter sequence over the past 24 hours was divided into 48 local time series segments, each containing environmental data at 360 time steps.

[0031] Preferably, channel attention weighting is applied to each local time segment to highlight environmental factors sensitive to the current growth stage, including:

[0032] According to the weighting calculation formula:

[0033] ;

[0034] For the first Time series of environmental factors Assign attention weights , , and For learnable parameters, For the first Time series of environmental factors This is a transpose.

[0035] Preferably, the weighted segment sequence is input into the global context encoder, which outputs a fixed-dimensional joint state vector, including:

[0036] The weighted sequence of segments is input into a global context encoder consisting of two stacked bidirectional LSTM layers, and the output is a joint state vector.

[0037] Preferably, based on the joint state vector, under the constraints of a pre-constructed multi-objective optimization function, the optimal setpoints of each environmental factor in the next control cycle are solved, including:

[0038] Define the optimization objective function as follows: ;

[0039] For tuber proliferation rate, Energy consumption per unit of output To improve nutrient solution utilization, , , These are the weighting coefficients;

[0040] Apply constraints: light intensity maintained between 200 and 800 micromoles per square meter per second, air temperature between 16 and 22 degrees Celsius, relative humidity not less than 70%, carbon dioxide concentration not greater than 1200 ppm, and atomization interval [not specified]. satisfy ,and ; The initial time, Deadline;

[0041] Once the initial tuber enlargement time point is detected, a phased constraint is forcibly applied: the ratio of red light to far-red light is adjusted to 3:1, and the atomization frequency is increased to once every 10 minutes.

[0042] Preferably, control commands are generated based on the solution results to drive the light adjustment device, temperature and humidity control unit, carbon dioxide supply system, and atomizing actuator to work together, including:

[0043] A model predictive control framework is adopted, with the prediction time domain being the next 4 hours and the control time domain being the next 1 hour, with a control step size of 10 minutes.

[0044] At the beginning of each control step, the latest environmental and growth data are collected again, the joint state vector is updated, and the quadratic programming problem is solved online to obtain the optimal control increment for the current step.

[0045] The optimal control increment is added to the current set value to form the final control command.

[0046] Preferably, the objective function of the quadratic programming problem is a second-order Taylor expansion approximation of the multi-objective optimization function at the current operating point, in the form of: ;

[0047] To control the increment vector, For Hessian matrix, The gradient vectors are calculated jointly from the Jacobian matrix of the multi-scale temporal coding model and the second derivative of the objective function.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] 1. By constructing a closed-loop control mechanism of environment-growth dual feedback, dynamic synergistic optimization of multiple factors in aeroponics is achieved, avoiding resource waste and growth inhibition caused by traditional fixed parameter control;

[0050] 2. Utilize a multi-scale time-series coding model to accurately capture the cumulative response of plants to environmental changes, enabling the control strategy to have adaptive capabilities at different growth stages.

[0051] 3. Introduce multi-objective optimization functions and phased constraints to reduce energy and water / fertilizer consumption per unit yield while ensuring high tuber yield;

[0052] 4. Rolling optimization is achieved through a model predictive control framework to address the uncertainties of environmental disturbances and plant physiological states, ensuring that the plant always operates under near-optimal conditions throughout the entire seedling cycle. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0054] Figure 2 This is a schematic diagram of the core principle framework of the closed-loop adaptive control based on multimodal environment perception and growth state feedback driven in this invention.

[0055] Figure 3 This is a flowchart illustrating the data fusion and multi-scale temporal coding logic of the environmental perception layer and the growth state analysis layer in this invention.

[0056] Figure 4 This is a flowchart illustrating the multi-objective optimization and phased constraint control logic of the multi-factor collaborative decision-making layer in this invention.

[0057] Figure 5 This is a flowchart illustrating the logic of model predictive control and rolling optimization instruction generation in the execution control layer of this invention.

[0058] Figure 6 This is a schematic diagram illustrating the control strategy switching and parameter adaptation relationship of the aeroponic system at different growth stages in this invention.

[0059] Detailed implementation method,

[0060] refer to Figures 1 to 6 This invention provides a controllable aeroponic seedling cultivation method suitable for potato seed production. The method constructs a four-layer progressive closed-loop control architecture consisting of an environmental perception layer, a growth state analysis layer, a multi-factor collaborative decision-making layer, and an execution control layer. This method collects multi-dimensional environmental parameters and plant physiological response data in the aeroponic chamber in real time, uses a multi-scale time-series coding model to generate a joint state vector, and dynamically solves for the optimal setpoints of each environmental factor under multi-objective optimization function constraints. Ultimately, it drives the execution mechanism to complete the adaptive control of the aeroponic environment. The following will describe the method step by step according to the order of steps clearly listed in the invention description.

[0061] A multi-source sensor array deployed within the aeroponic chamber synchronously collects a sequence of environmental parameters. These parameters include light intensity, spectral distribution, air temperature, relative humidity, carbon dioxide concentration, nutrient solution temperature, atomization pressure, droplet size distribution, and atomization start / stop timestamps. The multi-source sensor array comprises a distributed array of full-spectrum light sensors, platinum resistance temperature sensors, capacitive humidity sensors, an infrared carbon dioxide concentration detection module, an ultrasonic atomization particle size analyzer, and a high-precision pressure transmitter. The full-spectrum light sensor continuously records irradiance changes in the red, far-red, blue, and ultraviolet bands at a sampling frequency of at least once per second, with a spectral resolution covering the 380 nm to 780 nm range, ensuring accurate feedback for light quality control.

[0062] Platinum resistance temperature sensors are deployed at three height levels—top, middle, and bottom—of the aeroponic chamber to monitor air temperature above the canopy, in the stem area, and near the root zone, with a measurement accuracy of ±0.1 degrees Celsius. A capacitive humidity sensor is installed inside the plant canopy, with a measurement range of 5% to 98% and a resolution of 0.1%, used to capture microenvironmental humidity fluctuations. An infrared carbon dioxide concentration detection module is fixed inside the aeroponic chamber's return air duct, with a range of 0 to 2000 ppm and an accuracy of ±10 ppm, achieving stable monitoring of indoor carbon dioxide concentration through continuous airflow sampling. An ultrasonic atomizing particle size analyzer is integrated at the atomizing nozzle outlet, measuring the median volume diameter of droplets in real time based on the multi-frequency resonance principle, with an effective measurement range of 5 to 50 micrometers and a sampling period of 10 seconds. A high-precision pressure transmitter is connected to the main atomizing pipeline, providing real-time feedback on the atomizing system's operating pressure, with a range of 0 to 0.5 MPa and an accuracy of 5 / 1000. All sensor data is synchronously transmitted to the central processing unit via industrial Ethernet, with a timestamp alignment error of no more than 10 milliseconds, ensuring strict consistency of multi-source data in the time dimension.

[0063] Real-time growth status data of potato seed tubers are acquired through a plant phenotyping monitoring unit. This growth status data includes stem and leaf elongation rate, leaf unfolding area, chlorophyll fluorescence parameters, initial tuber enlargement time point, tuber quantity, and individual volume growth rate. The plant phenotyping monitoring unit consists of a top-mounted high-resolution industrial camera, a side-mounted near-infrared imaging module, and an embedded chlorophyll fluorescence detector. The high-resolution industrial camera uses a 5-megapixel global shutter sensor with a 16mm lens focal length, mounted vertically downwards on a support at the top of the aeroponic chamber, capturing orthophotos of the plant canopy every 2 hours. After preprocessing, the image data is input into a semantic segmentation model based on the U-Net architecture to extract leaf contours and calculate projected area. The stem and leaf elongation rate is then calculated using a continuous inter-frame difference method. The near-infrared imaging module operates in the 700-1000 nm wavelength range and is installed on the side wall of the aeroponic tank, illuminating the base of the plant at a 45-degree angle. The moisture gradient changes in the initial tuber enlargement area are identified by the water absorption spectrum characteristics. The initial tuber enlargement time point is defined as a decrease in local reflectance exceeding a threshold of 15% for a duration greater than 2 hours. The embedded chlorophyll fluorescence detector employs pulse modulation technology, with a built-in 660 nm excitation light source and a 740 nm emission filter. The maximum photochemical efficiency Fv / Fm (where Fv is variable fluorescence and Fm is maximum fluorescence) and the actual photochemical efficiency ΦPSII are measured at 30-minute intervals. The measured light intensity is 10 μmol / m² / s, the saturated pulse intensity is 3000 μmol / m² / s, and the pulse width is 0.8 seconds. All phenotypic data are timestamped and aligned with the environmental parameter sequence to form a unified multimodal observation dataset.

[0064] The environmental parameter sequence and growth state data are input into a preset multi-scale temporal coding model to generate a joint state vector representing the characteristics of the current growth stage. The multi-scale temporal coding model employs a stacked bidirectional long short-term memory network structure, containing three coding levels. The first level slices the original environmental parameter sequence using a sliding window with a window length of 6 hours and a step size of 30 minutes, generating 48 local temporal segments, each containing environmental data for 360 time steps. The second level applies channel attention weighting to each local temporal segment, with the weights calculated using the following formula:

[0065] ;

[0066] For the first Time series of environmental factors , and These are learnable parameters, obtained through backpropagation training. For the first Time series of environmental factors This is a transpose. The mechanism automatically highlights environmental factors sensitive to the current growth stage, such as assigning a higher weight to red light during tuber induction and strengthening humidity signals during rooting. The third layer inputs the weighted fragment sequence into a global context encoder, which consists of two stacked bidirectional LSTM layers with 128 hidden units, outputting a fixed-dimensional joint state vector with 256 dimensions. This joint state vector integrates the comprehensive dynamic features of environmental disturbances and plant responses over the past 24 hours, serving as the sole input for subsequent decision-making layers.

[0067] Then, based on the joint state vector, and under the constraints of a pre-constructed multi-objective optimization function, the optimal setpoints for each environmental factor in the next control cycle are solved. The multi-objective optimization function aims to maximize tuber proliferation rate, minimize energy consumption per unit yield, and maximize nutrient solution utilization. Let the tuber proliferation rate be... Energy consumption per unit of output is Nutrient solution utilization rate The objective function is then defined as:

[0068] ;

[0069] , , These are weighting coefficients, with values ​​of 0.6, 0.3, and 0.1 respectively. Tuber proliferation rate. The energy consumption per unit yield was calculated from the rate of change in tuber quantity continuously monitored by the near-infrared imaging module. Total energy consumption divided by predicted tuber fresh weight gain; nutrient solution utilization rate The ratio of total nitrogen, phosphorus, and potassium absorbed by the plant to the total supplied nitrogen, phosphorus, and potassium was indirectly estimated using nutrient solution conductivity and ion chromatography analysis. Constraints included: light intensity maintained between 200 and 800 μmol / m² / s, air temperature between 16 and 22 degrees Celsius, relative humidity not less than 70%, carbon dioxide concentration not exceeding 1200 ppm, and a specific atomization interval. satisfy and ,and . The initial time, This is the deadline.

[0070] Once the initial tuber enlargement time point is detected, the system forcibly applies phased constraints: the ratio of red light to far-red light is adjusted to 3:1, and the atomization frequency is increased to once every 10 minutes. , The optimization problem is modeled as a nonlinear programming problem and solved online using a sequential quadratic programming algorithm, with a convergence tolerance set to... .

[0071] Based on the solution results, control commands are generated to drive the coordinated operation of the illumination adjustment device, temperature and humidity control unit, carbon dioxide supply system, and atomizing actuator. The generation of these control commands employs a model predictive control framework, with a prediction time domain of the next 4 hours and a control time domain of the next 1 hour, using a 10-minute control step. At the beginning of each control step, the system re-acquires the latest environmental and growth data, updates the joint state vector, and solves a quadratic programming problem online to obtain the optimal control increment for the current step. The objective function of the quadratic programming problem is a second-order Taylor expansion approximation of a multi-objective optimization function at the current operating point, in the form:

[0072] ;

[0073] To control the increment vector, For Hessian matrix, The gradient vectors are calculated jointly from the Jacobian matrix of the multi-scale temporal coding model and the second derivative of the objective function. Constraints include the aforementioned environmental parameter boundaries and physical limitations of the atomizing device, such as the maximum driving voltage of the piezoelectric ceramic atomizing plate not exceeding 120 volts. The calculated control increments are superimposed on the current setpoint to form the final control command.

[0074] The illumination adjustment device consists of an adjustable spectrum LED array, including a red light chip (peak wavelength 660 nm), a far-red light chip (peak wavelength 730 nm), a blue light chip (peak wavelength 450 nm), and a white light chip. Each chip is independently driven, with a light intensity adjustment accuracy of 1%. The temperature and humidity control unit integrates a heat pump dehumidification module and an ultrasonic humidification module. A proportional-integral-derivative (PID) controller adjusts the compressor speed and humidification power, with a proportional gain of 1.5, an integral time of 300 seconds, and a derivative time of 60 seconds. The carbon dioxide supply system is equipped with an electromagnetic proportional valve, which dynamically adjusts the cylinder output flow rate based on the deviation between the set concentration and the measured value, with a control bandwidth of 0.5 Hz. The atomizing actuator uses a piezoelectric ceramic atomizing plate with an operating frequency of 1.7 MHz. The atomization pressure is maintained at 0.2 MPa by a closed-loop feedback pressure regulating valve, with a pressure fluctuation range of no more than ±5%.

[0075] The system automatically switches control strategy modes at different growth stages of potato seed tubers. During the rooting stage, high humidity and stable temperature are prioritized, with the atomization interval set to 40 seconds on and 200 seconds off, light intensity maintained at 300 μmol / m² / s, and carbon dioxide concentration set at 600 ppm. During the rapid stem and leaf growth stage, light intensity is increased to over 600 μmol / m² / s, and carbon dioxide concentration is increased to 1000 ppm, while the atomization shutdown time is extended to 240 seconds to promote stomatal opening. During the tuber induction stage, nighttime temperature is lowered to 16 degrees Celsius, while the proportion of red light is increased to 70% and the atomization shutdown time is shortened to 120 seconds to stimulate tuber primordia differentiation. During the tuber enlargement stage, relative humidity is maintained above 85%, light intensity is reduced to 500 μmol / m² / s, the atomization frequency is increased to once every 8 minutes (i.e., 45 seconds on and 435 seconds off), and carbon dioxide supplementation is suspended to avoid excessive respiration consumption. The switching of each stage is jointly triggered by the sudden changes in the initial tuber enlargement time and the stem and leaf elongation rate. The switching logic is implemented using a state machine to ensure a smooth transition of the control strategy.

[0076] The entire process is coordinated and executed by a central processing unit (CPU), which is an industrial-grade embedded computer equipped with a real-time operating system and a task scheduling cycle of 100 milliseconds. Data storage uses a time-series database, retaining original sampled data for at least 90 days. The communication protocol adopts a hybrid architecture of Modbus TCP and CAN bus, ensuring that the transmission delay of control commands is less than 50 milliseconds. Anomaly handling mechanisms include sensor failure detection, actuator jamming alarm, and control divergence suppression. When three consecutive optimization solutions fail, the system automatically switches to a safe mode and restores the preset conservative parameter combination.

[0077] In summary, this embodiment achieves dynamic and coordinated regulation of multiple factors in aeroponic environments through a rigorous four-layer architecture, multimodal data fusion, multi-scale temporal modeling, multi-objective optimization, and model predictive control. This method not only solves the resource waste problem caused by traditional fixed-parameter control but also fully stimulates the growth potential of potato seed tubers through an adaptive growth stage mechanism, providing a complete technical solution for efficient, intelligent, and sustainable aeroponic seedling cultivation.

[0078] It should be noted that, in this document, relational terms such as "first" and "second" are used only 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 process, method, article, or apparatus.

[0079] 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 controlled mist seedling raising method for potato seed production, characterized by, include: A multi-source sensor array deployed in the aeroponic chamber synchronously collects environmental parameter sequences, including light intensity, spectral distribution, air temperature, relative humidity, carbon dioxide concentration, nutrient solution temperature, atomization pressure, droplet size distribution, and atomization start and stop timestamps. Real-time growth status data of potato seed tubers are obtained through a plant phenotypic monitoring unit. The growth status data includes stem and leaf elongation rate, leaf unfolding area, chlorophyll fluorescence parameters, tuber initial swelling time point, tuber number and individual volume growth rate. The environmental parameter sequence and growth state data are input into a preset multi-scale temporal coding model to generate a joint state vector representing the characteristics of the current growth stage, including: Slide the original environmental parameter sequence using a sliding window to generate local time segments; Channel attention weighting is applied to each local time segment to highlight environmental factors sensitive to the current growth stage, including: According to the weighting calculation formula: ； For the first Time series of environmental factors Assign attention weights , , and For learnable parameters, For the first Time series of environmental factors For transpose; The weighted sequence of segments is input into the global context encoder, which outputs a fixed-dimensional joint state vector, including: The weighted sequence of segments is input into a global context encoder consisting of two stacked bidirectional LSTM layers, and the output is a joint state vector. Based on the joint state vector, under the constraints of a pre-constructed multi-objective optimization function, the optimal setpoints of each environmental factor in the next control cycle are solved, including: The optimization objective function is defined as ; for tuber multiplication rate, for unit yield energy consumption, for nutrient solution utilization rate, , , for weight coefficient; Apply constraints: light intensity maintained between 200 and 800 micromoles per square meter per second, air temperature between 16 and 22 degrees Celsius, relative humidity not less than 70%, carbon dioxide concentration not greater than 1200 ppm, and atomization interval [not specified]. satisfy and ,and ; The initial time, Deadline; Once the initial tuber enlargement time point is detected, a phased constraint is forcibly applied: the ratio of red light to far-red light is adjusted to 3:1, and the atomization frequency is increased to once every 10 minutes. The multi-objective optimization function takes maximizing tuber proliferation rate, minimizing energy consumption per unit yield, and maximizing nutrient solution utilization as optimization objectives, and introduces stage-based constraints on the physiological response of potato seed tubers. Based on the solution results, control commands are generated to drive the light adjustment device, temperature and humidity control unit, carbon dioxide supply system, and atomization actuator to work together to complete the closed-loop control of the atomization environment, including: A model predictive control framework is adopted, with the prediction time domain being the next 4 hours and the control time domain being the next 1 hour, with a control step size of 10 minutes. At the beginning of each control step, the latest environmental and growth data are collected again, the joint state vector is updated, and the quadratic programming problem is solved online to obtain the optimal control increment for the current step. The optimal control increment is added to the current set value to form the final control command; The objective function of the quadratic programming problem is a quadratic Taylor expansion approximation of the multi-objective optimization function at the current working point, in the form of ; To control the increment vector, For Hessian matrix, The gradient vectors are calculated jointly from the Jacobian matrix of the multi-scale temporal coding model and the second derivative of the objective function.

2. The controlled mist seedling raising method for potato seed production according to claim 1, characterized by, Environmental parameter sequences are synchronously collected by a multi-source sensor array deployed in the aeroponic chamber, including: The full-spectrum light sensor, platinum resistance temperature sensor, capacitive humidity sensor, infrared carbon dioxide concentration detection module, ultrasonic atomizing particle size analyzer, and high-precision pressure transmitter are distributed and installed in the atomization chamber. The full-spectrum light sensor captures irradiance changes in the red, far-red, blue, and ultraviolet bands. The air temperature was monitored at three different height levels—top, middle, and bottom—of the aeroponic chamber using the platinum resistance temperature sensor. The relative humidity is measured inside the plant canopy using the capacitive humidity sensor. The infrared carbon dioxide concentration detection module continuously monitors the carbon dioxide concentration in the return air duct. The median diameter of the droplet volume is measured in real time at the outlet of the atomizing nozzle using the ultrasonic atomizing particle size analyzer. The high-precision pressure transmitter feeds back the working pressure of the atomization system on the main atomization pipeline. All sensor data is synchronously transmitted to the central processing unit via industrial Ethernet.

3. The controlled mist seedling raising method for potato seed production according to claim 2, characterized by, Real-time growth status data of potato seed tubers were obtained through a plant phenotypic monitoring unit, including: A high-resolution industrial camera mounted on the top takes orthophotos of the plant canopy every 2 hours. The leaf outlines are extracted and the projected area is calculated using an image segmentation algorithm, which in turn calculates the elongation rate of the stems and leaves. By using a side-mounted near-infrared imaging module to obliquely irradiate the base area of ​​the plant in the 700 nm to 1000 nm band, the water gradient changes in the initial tuber enlargement area are identified. When the local reflectance decreases by more than 15% of the threshold and lasts for more than 2 hours, it is determined to be the time point of initial tuber enlargement. The maximum photochemical efficiency Fv / Fm and the actual photochemical efficiency ΦPSII were measured at 30-minute intervals using an embedded chlorophyll fluorescence detector. Fv represents variable fluorescence, and Fm represents maximum fluorescence.

4. The controlled mist seedling raising method for potato seed production according to claim 3, characterized by, The original environmental parameter sequence is sliced ​​using a sliding window to generate local time segments, including: The environmental parameter sequence over the past 24 hours was divided into 48 local time series segments, each containing environmental data at 360 time steps.

Images