Plant factory full-wave band artificial light source system and light environment self-adaptive intelligent dynamic adjustment method

Abstract

The invention discloses a plant factory full-band artificial light source system and a light environment self-adaptive intelligent dynamic adjustment method, and the system comprises a multi-band light source array which is composed of a visible light LED module, an ultraviolet light source module and a near-infrared heat source module which are independently divided according to spectral bands, and each module is provided with an independent constant-current drive circuit and a millisecond switching control unit, the controllable spectrum covering 400 nm to 1000 nm and partial ultraviolet and far infrared ranges can be output, and rapid switching and light mixing between wave bands can be achieved in the same plane. Through a modularized multi-band light source and independent driving, on-demand spectrum matching within the range of 400 nm to 1200 nm is achieved, ineffective radiation caused by traditional full-spectrum continuous irradiation is avoided, the actually-measured photosynthetically active radiation utilization rate is increased by about 32%, the overall energy consumption is reduced by 25% to 30% compared with a conventional system, and embedded spectrum monitoring and a TCN-RL mixed model are combined, so that the system has the characteristics of environmental abrupt change, high sensitivity and the like. Or conversion of different growth stages of crops can be realized, spectrum ratio adjustment can be completed within less than or equal to 20ms, and plant population light environment optimization is realized.

Description

Technical Field

[0001] This invention relates to the field of agricultural facilities and intelligent control technology, specifically to a full-band artificial light source system for plant factories and an adaptive intelligent dynamic adjustment method for the light environment. Background Technology

[0002] A full-spectrum artificial light source system for plant factories is an intelligent plant lighting solution integrating multiple wavelengths of light, including visible, ultraviolet, and infrared light. It aims to provide precise light quality, intensity, and photoperiod support for plants at different growth stages. This system typically combines LED, laser, or plasma light sources, covering a wide range from red / green / blue light that promotes photosynthesis to UV-A / B light that regulates secondary metabolism, far-red light that regulates plant growth, development, and physiological and biochemical metabolism, and even infrared light for heating or transpiration regulation, thus achieving full-spectrum on-demand supply. Building upon this, an adaptive intelligent dynamic adjustment method for the light environment relies on multi-sensor fusion (such as PAR sensors, spectrometers, and temperature and humidity probes) and crop growth models. Through AI algorithms, it analyzes plant physiological feedback and environmental parameters in real time, dynamically optimizing spectral ratios, light intensity, and timing. This ensures that the light environment for individual plants and entire plant communities always matches their current needs, thereby increasing yield and quality while reducing energy consumption, and driving the development of plant factories towards high efficiency, precision, economy, and energy sustainability.

[0003] In recent years, with the development of modern agriculture towards intensification and intelligence, plant factories, as a highly efficient agricultural production model under controlled environments, have received widespread attention. To achieve precise cultivation of plants throughout their entire growth cycle, various full-spectrum artificial light source systems and adaptive intelligent dynamic adjustment methods for light environments have been proposed in existing technologies for plant factories. These systems typically combine multi-spectrum light sources (such as visible light, ultraviolet light, and infrared light) with sensors and expert databases to achieve programmed dynamic control of spectrum, light intensity, and photoperiod, thereby improving crop yield and quality. However, research has found that existing technologies still have the following shortcomings in application:

[0004] Insufficient spectral coverage and energy utilization efficiency: Although some systems claim to be able to output full-band spectrum, in actual operation, the coupling and switching of light sources in different bands have problems such as large energy loss, inaccurate spectral combination and poor stability; especially in the ultraviolet and infrared bands, the control precision is not high, which can easily cause ineffective radiation or local heat accumulation, affecting plant growth and increasing energy consumption.

[0005] Dynamic adjustment response lag and adaptive limitations: Existing intelligent adjustment methods mostly rely on offline calibration or fixed growth models, which are not timely in response to sudden environmental changes (such as external light interference, rapid changes in temperature and humidity) and the differentiated needs of different crop varieties and different growth stages. This leads to lag and non-optimal light environment regulation, making it difficult to achieve true real-time optimization and optimal energy saving.

[0006] Multi-source data fusion and decision-making algorithms are limited: Most systems collect limited types of data and lack efficient fusion mechanisms for cross-band spectral information and plant physiological feedback. The adjustment algorithms are often based on empirical rules / single-factor stage experimental parameters or simple threshold control, which makes it difficult to maintain high accuracy and robustness under complex and variable conditions, thus limiting the universality and scalability of the system.

[0007] System integration and reliability issues: The full-band light source, sensing and control modules are often distributed in terms of physical structure and control logic, resulting in complex wiring, high maintenance difficulty, and problems such as inconsistent light source attenuation and sensor drift are prone to occur during long-term operation, affecting the stability and controllability of the overall light environment.

[0008] To address this, we propose a full-band artificial light source system for plant factories and an adaptive intelligent dynamic adjustment method for the light environment. Summary of the Invention

[0009] To achieve the above objectives, the present invention provides the following technical solution: a full-band artificial light source system for plant factories and an adaptive intelligent dynamic adjustment method for the light environment, comprising:

[0010] The multi-band light source array consists of visible light LED modules, ultraviolet light source modules, and near-infrared heat source modules, which are independently divided according to spectral bands. Each module is equipped with an independent constant current drive circuit and a millisecond-level switching control unit, which can output a controllable spectrum covering 400nm~1000nm and part of the ultraviolet (300nm~400nm) and far-infrared (1000nm~1200nm) ranges, and realize rapid switching and light mixing between bands in the same plane;

[0011] A distributed light parameter sensing network, comprising multiple wired or wirelessly connected light sensors (such as miniature spectrometers, PAR sensors, and light intensity distribution sensors), can collect light parameters (such as light intensity distribution, spectral distribution, photosynthetically active radiation (PPFD), and photon flux density (PFD)) from the top of the plant community vertically downwards or from different angles. This data is then linked to a feedback control system to adjust the light intensity and spectral dynamics of the light source in real time based on the dynamic changes in the light parameters. This optimizes the light environment of the entire crop community, maximizing overall photosynthetic efficiency and achieving optimal energy savings.

[0012] A multi-source environmental sensing network, including temperature and humidity sensors, CO2 concentration sensors, leaf surface temperature sensors, stem flow sensors, and crop image acquisition devices, periodically collects environmental parameters and plant physiological state data.

[0013] The intelligent decision-making module, which integrates edge computing and cloud computing, is deployed on a local edge computing node and communicates with the cloud training platform. It integrates spectral data, environmental data, and plant physiological data, uses a temporal convolutional network (TCN) to extract time series features, and combines a reinforcement learning (RL) model to generate and iteratively optimize spectral-light intensity control strategies online under a multi-objective reward function.

[0014] An adaptive closed-loop controller receives the strategy and converts it into driving signals for each module of the light source array, executes independent spectral output for each zone, and simultaneously acquires actual spectral and light intensity data in real time through an embedded spectral monitoring and feedback unit. It uses a dual-loop control algorithm combining proportional-integral-derivative (PID) and model predictive control (MPC) for error correction, achieving millisecond-level dynamic adjustment.

[0015] The thermal management and optical homogenization structure includes a partitioned liquid cooling channel arranged along the back of the light source array, a thermoelectric cooling chip-assisted cooling device, and a variable aperture optical homogenization plate, which are used to suppress local temperature rise caused by near-infrared band and ensure that the irradiation uniformity of the entire cultivation surface is ≥90%.

[0016] The method includes the following steps:

[0017] S1: Simultaneously collect environmental and spectral data;

[0018] S2: Preprocess and extract features from the collected data to identify the current growth stage and physiological state of the plant;

[0019] S3: Call the TCN and RL hybrid model to generate a dynamic spectrum-intensity configuration strategy based on the recognition results;

[0020] S4: Transform the strategy into light source array driving instructions and execute them to achieve independent regional control of spectrum and light intensity;

[0021] S5: Obtain the actual output through the spectral monitoring and feedback unit, calculate the deviation and perform real-time closed-loop correction;

[0022] S6: Send the running data back to the cloud training platform to update the model parameters and improve the accuracy of subsequent decisions.

[0023] Preferably, the multi-band light source array adopts a modular layout, with each module measuring 50mm×50mm and integrating a driving chip and a miniature spectrometer, achieving a band switching time of ≤10ms.

[0024] Preferably, the reinforcement learning model of the intelligent decision-making module is based on photosynthetic efficiency (μmol CO2·m-2·s). -1A multi-objective reward function was constructed using energy consumption ratio (gDW / kWh) and plant morphological indicators (leaf area index, plant height), and the deep deterministic policy gradient (DDPG) algorithm was used to solve the policy.

[0025] Preferably, the adaptive closed-loop controller adopts a dual-loop control structure, with the inner loop being fast spectral tracking based on PWM duty cycle and a response time ≤20ms, and the outer loop being long-term growth target optimization based on MPC and an optimization cycle of 5min.

[0026] Preferably, the liquid cooling channel in the thermal management structure uses deionized water circulation with a flow rate controlled at 0.5–1.0 L / min. Combined with a thermoelectric cooling element, this stabilizes the surface temperature of the near-infrared module at ≤45℃, and the blade temperature fluctuation is ≤±0.5℃.

[0027] Preferably, the cloud training platform performs incremental model training once a week, adopts a federated learning mechanism to protect the data privacy of each plant factory, and distributes the optimized network weights to edge nodes.

[0028] Compared with existing technologies, this invention provides a full-band artificial light source system for plant factories and an adaptive intelligent dynamic adjustment method for the light environment, which has the following beneficial effects:

[0029] 1. This plant factory's full-band artificial light source system and adaptive intelligent dynamic adjustment method for light environment achieve on-demand spectral matching in the 400nm-1200nm range through modular multi-band light sources and independent drives. This avoids the ineffective radiation caused by traditional full-spectrum continuous irradiation. The measured photosynthetically effective radiation utilization rate is increased by about 32%, and the overall energy consumption is reduced by 25%-30% compared to conventional systems. The embedded spectral monitoring combined with the TCN-RL hybrid model enables the system to complete spectral adjustment within ≤20ms when there are sudden environmental changes (such as external light interference, rapid changes in temperature and humidity) or crop growth stage transitions. This ensures that the light environment is always in the optimal state, significantly improving crop growth consistency and yield.

[0030] 2. This plant factory's full-band artificial light source system and adaptive intelligent dynamic adjustment method for light environment overcome the shortcomings of single-factor control in meeting the comprehensive needs of crops by jointly modeling spectral, environmental, and physiological three-dimensional data. The multi-objective reward function enables the system to improve photosynthetic efficiency while taking into account energy consumption and quality. Experiments show that the soluble sugar content of lettuce increased by 12% and the vitamin C content increased by 9%. The partitioned liquid cooling + thermoelectric cooling plate effectively suppresses local overheating caused by the near-infrared band, with a leaf surface temperature difference of ≤±0.5℃. The variable aperture homogenization plate ensures that the irradiance uniformity of the cultivation surface is ≥90%, reducing growth differences caused by uneven light spots.

[0031] 3. The plant factory's full-band artificial light source system and adaptive intelligent dynamic adjustment method for light environment, dual-loop closed-loop control and cloud incremental learning mechanism improve the long-term stability of the system. Modular hardware facilitates replacement and expansion, supports rapid adaptation to different crops and plant factories of different sizes, reduces operation and maintenance costs by about 18%, and the federated learning mechanism allows data from each factory to participate in model optimization without leaving the local area, taking into account both intelligent upgrades and commercial privacy and security. Detailed Implementation

[0032] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example

[0034] Examples of a full-spectrum artificial light source system and an adaptive intelligent dynamic adjustment method for the light environment in plant factories.

[0035] A full-spectrum artificial light source system for plant factories and an adaptive intelligent dynamic adjustment method for the light environment, including:

[0036] The multi-band light source array consists of visible light LED modules, ultraviolet light source modules, and near-infrared heat source modules, which are independently divided according to spectral bands. Each module is equipped with an independent constant current drive circuit and a millisecond-level switching control unit, which can output a controllable spectrum covering 400nm~1000nm and part of the ultraviolet (300nm~400nm) and far-infrared (1000nm~1200nm) ranges, and realize rapid switching and light mixing between bands in the same plane;

[0037] A distributed light parameter sensing network, comprising multiple wired or wirelessly connected light sensors (such as miniature spectrometers, PAR sensors, and light intensity distribution sensors), can collect light parameters (such as light intensity distribution, spectral distribution, photosynthetically active radiation (PPFD), and photon flux density (PFD)) from the top of the plant community vertically downwards or from different angles. It is then linked to a feedback control unit to adjust the light intensity and spectral dynamic ratio of the light source in real time based on the dynamic changes in light parameters, thereby optimizing the light environment of the entire crop community, maximizing overall photosynthetic efficiency, and optimizing energy saving.

[0038] A multi-source environmental sensing network, including temperature and humidity sensors, CO2 concentration sensors, leaf surface temperature sensors, stem flow sensors, and crop image acquisition devices, periodically collects environmental parameters and plant physiological state data.

[0039] The intelligent decision-making module, which integrates edge computing and cloud computing, is deployed on a local edge computing node and communicates with the cloud training platform. It integrates spectral data, environmental data, and plant physiological data, uses a temporal convolutional network (TCN) to extract time series features, and combines a reinforcement learning (RL) model to generate and iteratively optimize spectral-light intensity control strategies online under a multi-objective reward function.

[0040] An adaptive closed-loop controller receives the strategy and converts it into driving signals for each module of the light source array, executes independent spectral output for each zone, and simultaneously acquires actual spectral and light intensity data in real time through an embedded spectral monitoring and feedback unit. It uses a dual-loop control algorithm combining proportional-integral-derivative (PID) and model predictive control (MPC) for error correction, achieving millisecond-level dynamic adjustment.

[0041] The thermal management and optical homogenization structure includes a partitioned liquid cooling channel arranged along the back of the light source array, a thermoelectric cooling chip-assisted cooling device, and a variable aperture optical homogenization plate, which are used to suppress local temperature rise caused by near-infrared band and ensure that the irradiation uniformity of the entire cultivation surface is ≥90%.

[0042] The method includes the following steps:

[0043] S1: Simultaneously collect environmental and spectral data;

[0044] S2: Preprocess and extract features from the collected data to identify the current growth stage and physiological state of the plant;

[0045] S3: Call the TCN and RL hybrid model to generate a dynamic spectrum-intensity configuration strategy based on the recognition results;

[0046] S4: Transform the strategy into light source array driving instructions and execute them to achieve independent regional control of spectrum and light intensity;

[0047] S5: Obtain the actual output through the spectral monitoring and feedback unit, calculate the deviation and perform real-time closed-loop correction;

[0048] S6: Send the running data back to the cloud training platform to update the model parameters and improve the accuracy of subsequent decisions.

[0049] Specifically, the multi-band light source array adopts a modular layout, with each module measuring 50mm×50mm and integrating a driver chip and a miniature spectrometer, achieving a band switching time of ≤10ms.

[0050] Specifically, the reinforcement learning model of the intelligent decision-making module uses photosynthetic efficiency (μmol CO2·m -2A multi-objective reward function was constructed using s-1), energy consumption ratio (g DW / kWh), and plant morphological indicators (leaf area index, plant height), and the deep deterministic policy gradient (DDPG) algorithm was used to solve the policy.

[0051] Specifically, the adaptive closed-loop controller adopts a dual-loop control structure. The inner loop is a fast spectrum tracking based on PWM duty cycle with a response time of ≤20ms, and the outer loop is a long-term growth target optimization based on MPC with an optimization cycle of 5min.

[0052] Specifically, the liquid cooling channel in the thermal management structure uses deionized water circulation with a flow rate controlled at 0.5–1.0 L / min. Combined with thermoelectric cooling plates, it can stabilize the surface temperature of the near-infrared module at ≤45℃ and the blade temperature fluctuation at ≤±0.5℃.

[0053] Specifically, the cloud-based training platform performs incremental model training once a week, employs a federated learning mechanism to protect the data privacy of each plant factory, and distributes the optimized network weights to edge nodes.

[0054] Through the above technical solution, this invention achieves on-demand spectral matching in the 400nm–1200nm range by using modular multi-band light sources and independent drives, avoiding ineffective radiation caused by traditional full-spectrum continuous irradiation. The measured photosynthetically effective radiation utilization rate is increased by approximately 32%, and the overall energy consumption is reduced by 25%–30% compared to conventional systems. The combination of embedded spectral monitoring and the TCN-RL hybrid model enables the system to complete spectral adjustment within ≤20ms during sudden environmental changes (such as external light interference, rapid temperature and humidity changes) or crop growth stage transitions, ensuring the light environment remains optimal and significantly improving crop growth consistency and yield. By jointly modeling spectral, environmental, and physiological three-dimensional data, it overcomes the limitations of single-factor control in meeting the comprehensive needs of crops. To address the shortcomings of traditional methods, the multi-objective reward function enables the system to improve photosynthetic efficiency while balancing energy consumption and quality. Experiments show that the soluble sugar content of lettuce increased by 12% and the vitamin C content increased by 9%. The partitioned liquid cooling + thermoelectric cooling plate effectively suppresses local overheating caused by the near-infrared band, with a leaf surface temperature difference of ≤±0.5℃. The variable aperture homogenizing plate ensures that the irradiance uniformity of the cultivation surface is ≥90%, reducing growth differences caused by uneven light spots. The dual-loop closed-loop control and cloud-based incremental learning mechanism improve the long-term stability of the system. The modular hardware facilitates replacement and expansion, supports rapid adaptation to different crops and plant factories of different sizes, and reduces operation and maintenance costs by about 18%. The federated learning mechanism allows data from each factory to participate in model optimization without leaving the local machine, balancing intelligent upgrades and commercial privacy security.

[0055] I. System Composition and Hardware Configuration

[0056] The system hardware includes:

[0057] Multi-band light source array: Composed of visible light LED modules (450nm blue light, 660nm red light), ultraviolet light source modules (380-400nm ultraviolet light), near-infrared heat source modules (700-1200nm far-infrared light), and green light (520-560nm). Each module measures 50mm × 50mm and integrates an independent constant current drive circuit and a millisecond-level switching control unit (band switching time ≤ 10ms). It can output a controllable spectrum covering 400nm~1000nm and part of the ultraviolet (300nm~400nm) and far-infrared (1000nm~1200nm), supporting rapid band switching and mixing within the same plane.

[0058] Green light module: The main wavelength covers 520-560nm (preferably 530nm or 560nm), which is used to enhance light penetration and reflection within the canopy and optimize the light conditions of the lower leaves in the canopy.

[0059] Each module (including the newly added green light module) adopts an independent constant current drive circuit and a millisecond-level switching control unit (band switching time ≤10ms). The module size is 50mm×50mm. It integrates a drive chip and a miniature spectrometer, and can output a controllable spectrum covering 400nm~1000nm and part of the ultraviolet (300nm~400nm) and far-infrared (1000nm~1200nm). It supports rapid switching and mixing of multiple bands (including green light) in the same plane.

[0060] The green light module works in conjunction with the red, blue, ultraviolet, and infrared modules. Its light intensity ratio is dynamically adjusted by the intelligent decision module (adjustment range 0% to 30%, accuracy 1%) to meet the light environment requirements of different crops, growth stages, and population densities.

[0061] The multi-band light source array, while meeting the photosynthetic needs of plants, integrates human-centered lighting control functions: by independently controlling the dynamic ratio and intensity of visible light bands (including blue, green, and red light), it optimizes visual comfort and photoreceptor cell protection while ensuring the optimal light environment for plants. Specific design features include:

[0062] Human eye sensitivity band adjustment: In response to the human eye's high sensitivity to blue light (450-480nm) (which can easily cause retinal damage) and high comfort level to green light (520-560nm) (close to the natural light spectrum), the system dynamically reduces the proportion of high-energy short-wavelength blue light (450-460nm) (default ≤15% during the nutritional / reproductive period, ≤10% during working hours), and increases the proportion of green light (520-560nm) and long-wavelength red light (660-700nm) (≥30% during working hours), simulating the distribution of the natural light spectrum;

[0063] Dynamic Spectral Softening: When personnel enter the plant factory operation area (such as inspection and harvesting), the "human factor mode" is triggered through the edge computing module, which reduces the peak intensity of blue light in the light source array by 20%-30%, while increasing the mixing ratio of green light (520-560nm) and yellow light (570-590nm) (total proportion ≥40%), thereby reducing spectral contrast and glare effect.

[0064] Human eye adaptation transition: The system supports gradual spectral adjustment (response time ≤ 500ms) to avoid visual stimulation from abrupt switching of monochromatic light such as blue light / red light (e.g., when switching from high blue light plant production mode to human factor mode, the proportion of blue light gradually decreases from 40% to 10%, and the proportion of green light gradually increases from 20% to 40%).

[0065] Distributed light parameter sensing network: Miniature spectrometers and PAR sensors (wired / wireless connection) are arranged vertically downwards from the top of the plant community and at different heights (e.g., 20cm, 40cm, 60cm from the ground) to collect light intensity distribution (PPFD), spectral distribution (SPD), and photon flux density (PFD) of different parts of the community in real time, and to monitor whether the bottom leaves are adequately lit and the uniformity of light distribution in the community.

[0066] Embedded Spectral Monitoring and Feedback Unit: A miniature spectrometer and PAR sensor are installed near the light-emitting surface of each light source module to detect the spectral power distribution (SPD) and effective photosynthetic photon flux density (PPFD) of the output spectrum in real time, and the data is transmitted to the controller in real time.

[0067] Multi-source environmental sensing network: including temperature and humidity sensors (accuracy ±0.5℃ / ±2%RH), CO2 concentration sensors (accuracy ±30ppm), leaf surface temperature sensors (accuracy ±0.3℃), stem flow sensors (monitoring water transport status), and crop image acquisition devices (high-definition cameras, resolution ≥5MP), periodically (every 10 minutes) collecting environmental parameters and plant physiological status data (such as leaf area index, plant height, leaf thickness, fruit development stage, color parameters).

[0068] An intelligent decision-making module that integrates edge computing and cloud computing: Deployed on local edge computing nodes (such as industrial-grade ARM servers) and communicating with a cloud-based training platform (based on the TensorFlow / PyTorch framework). The module integrates spectral data, environmental data, and plant physiological data, utilizes a temporal convolutional network (TCN) to extract time-series features (such as the light intensity change trend and leaf growth rate over the past 24 hours), and combines a reinforcement learning (RL) model (using the Deep Deterministic Policy Gradient (DDPG) algorithm) to generate a dynamic spectral-light intensity control strategy under a multi-objective reward function.

[0069] Adaptive closed-loop controller: Receives strategy instructions from the intelligent decision-making module and converts them into drive signals for each module of the light source array (adjusting light intensity through PWM duty cycle and controlling spectral ratio through band switching), and executes independent spectral output for each zone; simultaneously, it acquires the actual output through an embedded spectral monitoring and feedback unit and a distributed optical parameter sensing network, and adopts a dual-loop control algorithm combining proportional-integral-derivative (PID) and model predictive control (MPC) (inner loop PID response time ≤ 20ms to track the spectrum, outer loop MPC optimizes long-term growth targets every 5 minutes).

[0070] Thermal management and optical homogenization structure: The back of the light source array is equipped with partitioned liquid cooling channels (deionized water circulation, flow rate 0.5-1.0L / min), combined with thermoelectric cooling plates (auxiliary cooling, near-infrared module surface temperature ≤45℃), and variable aperture optical homogenization plates (large aperture for visible light band to improve uniformity, small aperture for ultraviolet / near-infrared band to prevent local overheating), to ensure that the irradiation uniformity of the cultivation surface is ≥90% and the leaf surface temperature difference is ≤±0.5℃.

[0071] II. Phased Adaptive Regulation Process (Taking Lettuce and Tomatoes as Examples)

[0072] 1. Seedling stage (for lettuce: 0-10 days after sowing; for tomatoes: 0-14 days after sowing)

[0073] Regulation objective: to promote seed germination and early photosynthesis in seedlings, without considering individual or group shading differences.

[0074] Data input: Sensors collect environmental parameters (temperature 22±1℃, humidity 65±5%RH, CO2 400-600ppm) and plant physiological data (seedling height <5cm, number of leaves ≤2, no shading issues).

[0075] Regulation Strategy: The intelligent decision-making module identifies the current stage as seedling stage using the TCN-RL model and generates a spectral configuration dominated by high chlorophyll absorption wavelengths (30% blue light 450-470nm, 50% red light 620-680nm, and 10%-20% green light 520-560nm), while turning off ultraviolet and far-infrared wavelengths (to avoid damage during the sensitive seedling stage). The light intensity is set at 150-200 μmol·m⁻². -2 ·s -1 (PPFD) meets the basic requirements of photosynthesis.

[0076] Results: Actual measurements showed that the chlorophyll a / b ratio of seedlings increased by 15%, the seedling uniformity reached 98%, and energy consumption was reduced by 28% compared to conventional full-spectrum irradiation.

[0077] 2. Nutritional period (for lettuce: 11-25 days after sowing; for tomatoes: 15-45 days after sowing)

[0078] Regulation objective: To balance the light absorption efficiency of individual leaves with the light environment within the plant population, and to solve the shading problem caused by the increase in leaves.

[0079] Data input: Lettuce plant height 15-30cm, 8-12 leaves (leaves mutually shading each other, blocking light to the lower leaves); Tomato plant height 40-60cm, 20-30 leaves (canopy thickness increased, lower leaves PPFD only 30-50μmol·m -2 ·s -1 ).

[0080] Regulation strategies:

[0081] Lettuce: The model dynamically adjusts the spectrum to 40% blue light (450-470nm), 50% red light (620-680nm), and 10% far-infrared light (700-800nm) (far-infrared light promotes leaf spreading and reduces overlapping shading); light intensity is increased to 300-400 μmol·m -2 ·s -1 And by using a distributed optical parameter sensing network to monitor the PPFD (target ≥150 μmol·m⁻¹) of the bottom leaves. -2 ·s -1 ).

[0082] Tomatoes: Add an extra 5% ultraviolet wavelength (380-400nm) (to inhibit excessive vegetative growth and promote thicker stems); adjust the red / blue light ratio to 4:1 (to promote carbohydrate accumulation); set the light intensity to 400-500 μmol·m⁻¹. -2 ·s -1 Furthermore, the light intensity distribution at the top and bottom of the canopy was optimized using an adjustable aperture optical homogenizing plate (bottom PPFD ≥ 200 μmol·m). -2 ·s -1 ).

[0083] Implementation results: PPFD in the bottom leaves of lettuce increased from 80 μmol·m -2 ·s -1 Increased to 180 μmol·m -2 ·s -1 The group's light energy utilization efficiency increased by 40%; tomato stem diameter increased by 12%; and leaf photosynthetic rate increased by 25%.

[0084] 3. Reproductive growth period / maturity and harvest period (for solanaceous vegetables only: tomato as an example, 46 days after sowing to harvest)

[0085] Regulation objective: To simultaneously optimize fruit production capacity (yield) and quality (color, taste, and nutritional components).

[0086] Data input: When tomatoes enter the flowering and fruit setting period (days 46-60) and the ripening period (day 61 to harvest), the fruit color changes from green to red (chlorophyll degradation, carotenoid / anthocyanin synthesis), and it is necessary to regulate sugar accumulation and color development.

[0087] Regulation strategies:

[0088] During the flowering and fruit setting period (46-60 days): increase the ultraviolet band by 8% (380-400nm) (to promote pollen activity and fruit setting rate) and the far-infrared band by 15% (700-800nm) (to regulate sugar transport in the fruit); adjust the red / blue light ratio to 3:1; maintain light intensity at 400-450 μmol·m -2 ·s -1 .

[0089] During the ripening period (61 days to harvest): The ultraviolet band is turned off, while the far-infrared band is increased to 20% (700-800nm) (to accelerate sugar accumulation) and blue light to 20% (450-470nm) (to inhibit excessive fruit softening). The proportion of red light is reduced to 50%; light intensity is reduced to 350-400 μmol·m⁻¹. -2 ·s -1 (Avoid high-temperature scorching). Monitor fruit color (RGB value R≥180, G≤80) using a crop image acquisition device, and dynamically fine-tune the spectral ratio.

[0090] Results: Tomato yield per plant increased by 18%, soluble sugar content increased by 15% (≥5.2g / 100g), vitamin C content increased by 10% (≥30mg / 100g), fruit color uniformity reached 95% (no obvious shaded areas), and dry matter conversion rate (sugar content ratio) increased by 20%.

[0091] III. Method Execution Flow

[0092] S1 synchronous acquisition: Real-time acquisition of environmental data (temperature, humidity, CO2), spectral data (SPD / PPFD output from each module), plant canopy light parameters (PPFD distribution monitored by distributed sensors), and plant physiological status (plant height, number of leaves, and fruit development stage from image analysis).

[0093] S2 Feature Extraction and Stage Identification: By analyzing time series data (such as leaf growth rate and fruit color changes over the past 3 days) using the TCN model, the current growth stage (seedling stage / nutritional stage / reproductive stage) and physiological state (such as the degree of shading of the bottom leaves and fruit maturity) are identified.

[0094] S3 Strategy Generation: The reinforcement learning model (DDPG) generates dynamic spectrum-light intensity configuration strategies (such as increasing the proportion of far-infrared light during the nutrient period and adjusting the ultraviolet / far-infrared combination during the reproductive period) based on a multi-objective reward function (photosynthetic efficiency, energy consumption ratio, population light energy utilization efficiency, and fruit quality indicators).

[0095] S4 instruction execution: The adaptive closed-loop controller converts the strategy into driving signals, controlling each module of the light source array to independently output the target spectrum and light intensity (e.g., turning off ultraviolet light during the tomato ripening period and increasing the far-infrared ratio).

[0096] S5 closed-loop correction: The actual output (such as whether the underlying PPFD meets the standard) is obtained through embedded spectral monitoring and distributed optical parameter sensing network. The deviation is corrected (error ≤ 5%) by using PID (fast tracking spectrum) and MPC (long-term optimization target) dual-loop control algorithm.

[0097] S6 Model Iteration: Operational data (such as actual yield and quality parameters) are sent back to the cloud training platform for incremental training every week (using a federated learning mechanism to protect the data privacy of each plant factory) to optimize model parameters and improve the accuracy of subsequent decisions.

[0098] IV. Verification of Technical Effects

[0099] Photosynthetic efficiency: The photosynthetic effective radiation utilization rate of lettuce increased by 32% (measured value), and the light energy utilization efficiency of tomato canopy increased by 45% compared with the conventional system.

[0100] Energy consumption: The overall system energy consumption is reduced by 28%-30% compared to traditional LED light sources (under the same production conditions).

[0101] Quality optimization: Soluble sugar content of lettuce increased by 12% and vitamin C content increased by 9%; soluble sugar content of tomatoes increased by 15% and vitamin C content increased by 10%, with fruit color uniformity ≥95%.

[0102] Population uniformity: irradiance uniformity of cultivation surface ≥90%, leaf surface temperature difference ≤±0.5℃, with no localized overheated or under-dark areas.

[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 full-spectrum artificial light source system for plant factories and an adaptive intelligent dynamic adjustment method for the light environment, characterized by: include: The multi-band light source array consists of visible light LED modules, ultraviolet light source modules, and near-infrared heat source modules, which are independently divided according to spectral bands. Each module is equipped with an independent constant current drive circuit and a millisecond-level switching control unit, which can output a controllable spectrum covering 400nm~1000nm and part of the ultraviolet (300nm~400nm) and far-infrared (1000nm~1200nm) ranges, and realize rapid switching and light mixing between bands in the same plane; A distributed light parameter sensing network, comprising multiple wired or wirelessly connected light sensors (such as miniature spectrometers, PAR sensors, and light intensity distribution sensors), can collect light parameters (such as light intensity distribution, spectral distribution, photosynthetically active radiation (PPFD), and photon flux density (PFD)) from the top of the plant community vertically downwards or from different angles. It is then linked to a feedback control unit to adjust the light intensity and spectral dynamic ratio of the light source in real time based on the dynamic changes in light parameters, thereby optimizing the light environment of the entire crop community, maximizing overall photosynthetic efficiency, and optimizing energy saving. A multi-source environmental sensing network, including temperature and humidity sensors, CO2 concentration sensors, leaf surface temperature sensors, stem flow sensors, and crop image acquisition devices, periodically collects environmental parameters and plant physiological state data. The intelligent decision-making module, which integrates edge computing and cloud computing, is deployed on a local edge computing node and communicates with the cloud training platform. It integrates spectral data, environmental data, and plant physiological data, uses a temporal convolutional network (TCN) to extract time series features, and combines a reinforcement learning (RL) model to generate and iteratively optimize spectral-light intensity control strategies online under a multi-objective reward function. An adaptive closed-loop controller receives the strategy and converts it into driving signals for each module of the light source array, executes independent spectral output for each zone, and simultaneously acquires actual spectral and light intensity data in real time through an embedded spectral monitoring and feedback unit. It uses a dual-loop control algorithm combining proportional-integral-derivative (PID) and model predictive control (MPC) for error correction, achieving millisecond-level dynamic adjustment. The thermal management and optical homogenization structure includes a partitioned liquid cooling channel arranged along the back of the light source array, a thermoelectric cooling chip-assisted cooling device, and a variable aperture optical homogenization plate, which are used to suppress local temperature rise caused by near-infrared band and ensure that the irradiation uniformity of the entire cultivation surface is ≥90%. The method includes the following steps: S1: Simultaneously acquire environmental and spectral data (light environment data at different vertical levels). S2: Preprocess and extract features from the collected data to identify the current growth stage and physiological state of the plant; S3: Call the TCN and RL hybrid model to generate a dynamic spectrum-intensity configuration strategy based on the recognition results; S4: Transform the strategy into light source array driving instructions and execute them to achieve independent regional control of spectrum and light intensity; S5: Acquires actual output through the spectral monitoring and feedback unit, calculates deviations, and performs real-time closed-loop correction; (light environment data within the population) S6: Send the running data back to the cloud training platform to update the model parameters and improve the accuracy of subsequent decisions.

2. The plant factory full-band artificial light source system and the adaptive intelligent dynamic adjustment method for light environment as described in claim 1, characterized in that: The multi-band light source array adopts a modular layout, with each module measuring 50mm×50mm. It integrates a driver chip and a miniature spectrometer, achieving a band switching time of ≤10ms.

3. The plant factory full-band artificial light source system and the adaptive intelligent dynamic adjustment method for light environment as described in claim 1, characterized in that: The reinforcement learning model of the intelligent decision-making module uses photosynthetic efficiency (μmol CO2·m -2 ·s -1 A multi-objective reward function was constructed using energy consumption ratio (gDW / kWh) and plant morphological indicators (leaf area index, leaf thickness, plant height), and the deep deterministic policy gradient (DDPG) algorithm was used to solve the policy.

4. The plant factory full-band artificial light source system and the adaptive intelligent dynamic adjustment method for light environment as described in claim 1, characterized in that: The adaptive closed-loop controller adopts a dual-loop control structure. The inner loop is a fast spectral tracking based on PWM duty cycle with a response time of ≤20ms, and the outer loop is a long-term growth target optimization based on MPC with an optimization cycle of 5min.

5. The plant factory full-band artificial light source system and the adaptive intelligent dynamic adjustment method for light environment as described in claim 1, characterized in that: The liquid cooling channel in the thermal management structure uses deionized water circulation with a flow rate controlled at 0.5–1.0 L / min. Combined with thermoelectric cooling chips / external cryogenic liquid, the surface temperature of the near-infrared module can be stabilized at ≤45℃, and the blade temperature fluctuation is ≤±0.5℃.

6. The plant factory full-band artificial light source system and the adaptive intelligent dynamic adjustment method for light environment as described in claim 1, characterized in that: The cloud-based training platform performs incremental model training once a week, employs a federated learning mechanism to protect the data privacy of each plant factory, and distributes the optimized network weights to edge nodes.