Plant factory light-temperature coupling energy-saving model with photoperiod dynamic temperature adjustment and cloud control method

Abstract

The invention discloses a plant factory light-temperature coupling energy-saving model with a photoperiod dynamic temperature regulation function and a cloud control method, and the method comprises the following steps: S1, carrying out the multi-mode photoperiod recognition: carrying out the multi-mode photoperiod recognition based on the multi-source data of a fusion natural illumination intensity time sequence, an artificial light supplement time sequence and crop canopy spectral reflectivity, an actual photoperiod type and a dynamic turning point thereof are judged in real time through a deep learning network, and the influence of natural illumination fluctuation and light supplementing time sequence deviation on photoperiod recognition is eliminated. By fusing multi-modal data of natural light, a light supplementing time sequence and canopy spectral reflectivity and utilizing a CNN-LSTM deep learning network, misjudgment caused by natural illumination fluctuation and light supplementing deviation is effectively overcome, high-precision real-time photoperiod judgment is achieved, a universal light-temperature coupling model is constructed based on transfer learning and multi-task regression, and the real-time photoperiod judgment accuracy is improved. The method can quickly adapt to the requirements of different genotype crops and different growth stages, avoids the limitation that a traditional model is only suitable for a single crop, and remarkably improves the application range and stability of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of smart agriculture and protected horticulture, in particular to a light-temperature coupling energy-saving model and cloud control method for photoperiod dynamic temperature regulation in a plant factory. BACKGROUND

[0002] The light-temperature coupling energy-saving model and cloud control method for photoperiod dynamic temperature regulation in a plant factory is a smart agriculture technology system that integrates crop circadian rhythm and light-thermal environment optimization. The model takes plant photoperiod characteristics as the core, analyzes the dynamic demand law of light duration, intensity and temperature at different growth stages, constructs a nonlinear coupling relationship matrix of light-temperature factors, upgrades the traditional static temperature control to a dynamic mode that adapts to the photoperiod rhythm, such as matching higher temperature in the long-day stage to promote photosynthetic product transport, and reducing temperature in the short-day stage to reduce respiration consumption, thus achieving the physiological level of "light promotes temperature demand, temperature assists light efficiency" synergistic energy saving. The cloud control method relies on the Internet of Things sensing layer to collect environmental light intensity, temperature and humidity, and crop phenotype data in real time, pre-processes them through edge computing nodes, uploads them to the cloud platform, uses machine learning algorithms (such as reinforcement learning or digital twin) to optimize the historical growth data and real-time state (considering yield, quality and energy consumption), generates dynamic temperature regulation strategies and sends them to the local execution unit, forming a closed-loop control of "sensing-modeling-decision-feedback". This technology breaks through the limitations of traditional plant factory light control and temperature control, reduces invalid energy consumption (such as redundant heating during non-photoperiod periods) through light-temperature coupling, and realizes cluster optimization scheduling of multiple factories across regions with cloud computing power, ultimately ensuring high yield and quality while reducing energy consumption by more than 30% compared to traditional models, providing key technical support for the low-carbon and intelligent development of protected agriculture.

[0003] With the development of protected agriculture towards intelligence and low carbon, the light-temperature coupling energy-saving model and cloud control method for photoperiod dynamic temperature regulation in a plant factory have gradually become a research hotspot. This method combines plant photoperiod characteristics and dynamic regulation of light-thermal environment to balance crop growth demand and energy consumption, and its core is to build a light-temperature factor coupling model and realize multi-objective optimization control through a cloud platform. However, research has found that existing technologies still have the following significant defects in practical application:

[0004] First, the universality of photoperiod and light-temperature coupling models is insufficient. Existing models are mostly based on the photoperiod response characteristics of specific crop varieties (such as leafy vegetables or fruit vegetables), and have limited coverage of cross-species (such as medicinal plants, flowers, etc.) or different growth stages (such as seedling stage and reproductive growth stage) of the same species. Model parameters often rely on empirical settings or small sample test data training, making it difficult to accurately adapt to the dynamic physiological needs of different genotypes of crops, leading to "over-regulation" or "under-regulation" in actual application. For example, applying mismatched light and temperature combinations to non-model crops can inhibit photosynthetic efficiency or accelerate nutrient consumption, affecting yield and quality stability.

[0005] Second, the real-time and precision of light-temperature dynamic coupling are limited. In existing technology, photoperiod recognition relies on fixed time procedures or simple light intensity threshold judgments (such as setting daily light duration ≥12h as long-day), without fully considering the interference of natural light fluctuations (such as sudden light intensity drop on rainy days), artificial light timing deviations (such as light supplement delay), and other interference factors on actual photoperiod, resulting in large photoperiod determination errors. At the same time, the modeling of light-temperature coupling relationship mostly uses linear or low-order nonlinear fitting, which is difficult to capture the sudden changes in light intensity (such as from weak light to saturated light) and the hysteresis effect of temperature response (such as the actual microenvironment changes caused by delayed increase in transpiration after temperature rise), making the temperature regulation strategy different from the actual light and heat demand. This may miss the best regulation window, or cause energy waste due to overcompensation.

[0006] Third, the decision-making efficiency and adaptability of cloud control need to be improved. Existing cloud control methods mostly rely on centralized cloud computing architecture, uploading massive environmental data (light intensity, temperature and humidity, CO2 concentration, etc.) and crop phenotype data (such as plant height, leaf area) to the cloud for processing before issuing instructions. The time delay of data transmission and calculation (usually seconds to minutes) cannot meet the high-frequency dynamic regulation needs of plant factories. In addition, cloud model training is mostly based on historical steady-state data, and has weak real-time adaptability to sudden environmental disturbances (such as local high temperature caused by equipment failure, abnormal light supplement caused by power grid fluctuations) or crop sudden stress (such as metabolic rate mutation caused by disease), making the decision-making strategy deviate from the optimal trajectory, and even causing a chain of energy consumption increase.

[0007] Fourthly, the multi-source data fusion and model self-updating capability are weak. In the prior art, the input data of the light-temperature coupling model mainly depends on the environmental parameters collected by preset sensors. The direct representation data of the crop physiological state (such as in-situ monitoring of chlorophyll fluorescence and photosynthetic rate) are difficult to obtain due to high cost and great integration difficulty, so that the model cannot realize the deep fusion of the “environment-crop” state. Meanwhile, the model updating mainly depends on the artificial periodic calibration or offline retraining, and lacks the self-adaptive correction mechanism for the real-time growth feedback (such as yield estimation deviation and quality index change). The model mismatch problem caused by the genetic characteristics drift of crops (such as variety degradation) or the aging of facilities (such as the decrease of heat preservation performance) cannot be solved in time, and the energy saving effect obviously decreases over time.

[0008] Fifthly, the system integration and expansibility are insufficient. The existing light-temperature coupling control system is mostly developed in a customized manner, and the interface protocols of hardware (such as sensors and actuators) and software (such as model algorithms and cloud platforms) are not unified, so that it is difficult to realize the plug-and-play of cross-brand devices. For the multi-factory cluster management scene, the existing cloud platform lacks a lightweight deployment scheme, and when a large number of devices are accessed, the computing power bottleneck or communication congestion may occur. Moreover, the micro-environment differences (such as regional climate and facility structure) of different factories are not included in the global optimization framework, so that the energy saving potential at the cluster level is not fully tapped.

[0009] Therefore, the plant factory light-temperature coupling energy saving model and cloud control method for dynamic temperature adjustment of photoperiod are provided. SUMMARY

[0010] To achieve the above purpose, the technical scheme provided by the present application is as follows: the plant factory light-temperature coupling energy saving model and cloud control method for dynamic temperature adjustment of photoperiod, comprising the following steps:

[0011] S1: multi-modal photoperiod recognition: based on the multi-source data of fused natural light intensity time series, artificial light supplement time sequence and crop canopy spectral reflectance, the actual photoperiod type and its dynamic turning point are determined in real time through a deep learning network, so as to eliminate the influence of natural light fluctuation and light supplement time sequence deviation on photoperiod recognition;

[0012] S2: cross-species light-temperature coupling modeling: a universal light-temperature coupling model is constructed, which takes the crop genotype identifier, growth stage label and environmental light intensity-temperature history data as input. The light-temperature coupling model adopts the transfer learning and multi-task regression algorithm to adaptively map the light-temperature requirements of different crops and different growth stages, and outputs the dynamic target light-temperature curve;

[0013] S3: Cloud-edge collaborative control: Deploy edge computing nodes in the plant factory for real-time collection of environmental parameters and preliminary filtering and event detection, upload key state data and model inference requests to the cloud; the cloud uses a reinforcement learning optimizer combined with a digital twin simulation environment to generate high-precision temperature regulation strategies and issue them to the edge nodes for immediate execution, achieving millisecond-level closed-loop response;

[0014] S4: Online self-updating mechanism: Continuously collect crop physiological state data and yield / quality feedback information, periodically correct parameters and fine-tune the structure of the light-temperature coupling model through incremental learning and Bayesian optimization to maintain the matching of the model and the actual needs of the crops;

[0015] S5: Cluster expansion and heterogeneous compatibility: The cloud platform supports heterogeneous device protocol conversion and lightweight container deployment, enabling unified scheduling and differentiated optimization of multi-factory light-temperature control, and introducing regional climate and facility structure constraints at the global level to improve cluster energy efficiency.

[0016] Preferably, the multi-modal photoperiod recognition in step S1 uses a convolution-cyclic hybrid neural network (CNN-LSTM), and the input includes: natural light intensity sequence, artificial light switch timing, crown layer near-infrared and red light band reflectance ratio, and the output is the current actual photoperiod category and confidence, and a short-time sliding window rejudgment mechanism is triggered when the light intensity suddenly changes.

[0017] Preferably, the cross-species light-temperature coupling model in step S2 takes crop gene database encoding, growth stage label, and environmental light intensity-temperature three-dimensional tensor as input, uses transfer learning to transfer existing crop variety model knowledge to new crops, and simultaneously predicts target temperature curve and expected photosynthetic active radiation utilization rate (PUE) through multi-task regression.

[0018] Preferably, the reinforcement learning optimizer of the cloud-edge collaborative control in step S3 uses the deep deterministic policy gradient (DDPG) algorithm, simulates various disturbance scenarios (device failure, power grid fluctuation, other factor sudden stress) in the digital twin environment, trains a robust temperature regulation strategy, and issues millisecond-level instructions through the MQTT protocol.

[0019] Preferably, the online self-updating mechanism in step S4 introduces crop physiological sensor (chlorophyll fluorescence instrument, photosynthetic rate in-situ monitor) data, adjusts model hyperparameters through Bayesian optimization, and automatically triggers model retraining process when detecting that yield estimation deviation exceeds threshold.

[0020] Preferably, the cloud platform in step S5 adopts a containerized micro-service architecture, supports automatic parsing and conversion of multiple industrial protocols such as OPC-UA, Modbus and LoRaWAN, and introduces a regional climate correction coefficient and a facility insulation performance decay model in cluster scheduling, thereby realizing differentiated energy-saving optimization of multiple factory areas.

[0021] A system for implementing any of the above methods, comprising:

[0022] A multi-source data acquisition module configured with natural light intensity sensors, artificial light supplement controllers, crown layer spectrum sensors and crop physiological sensors;

[0023] An edge computing node for real-time filtering, event detection and preliminary photoperiod determination of the collected data;

[0024] A cloud intelligent platform including a cross-species light-temperature coupling model library, a digital twin simulation environment, a reinforcement learning optimizer and a cluster scheduling engine;

[0025] An execution control module for receiving temperature adjustment strategies issued by the cloud and driving heating / cooling equipment, ventilation equipment and light supplement equipment to operate in linkage;

[0026] A model self-updating module connected to a yield / quality monitoring system and an incremental learning unit to realize online correction of model parameters.

[0027] Compared with the prior art, the plant factory light-temperature coupling energy-saving model and cloud control method for dynamic photoperiod temperature adjustment provided by the present application have the following beneficial effects:

[0028] 1. The plant factory light-temperature coupling energy-saving model and cloud control method for dynamic photoperiod temperature adjustment effectively overcome misjudgments caused by natural light fluctuation and light supplement deviation by fusing multi-modal data of natural light, light supplement timing and crown layer spectral reflectance and using a CNN-LSTM deep learning network, realize high-precision real-time photoperiod determination, and based on a general light-temperature coupling model constructed by transfer learning and multi-task regression, quickly adapt to the needs of different genotypes of crops and different growth stages, avoid the limitation of traditional models that are only applicable to a single crop, and significantly improve the application range and stability of the system.

[0029] 2. The plant factory light-temperature coupling energy-saving model and cloud control method for dynamic photoperiod temperature adjustment solves the problem of high time delay of traditional cloud computing by using an edge node to be responsible for real-time collection and preliminary processing, using reinforcement learning and digital twin for global optimization by the cloud, issuing strategies by using a low-latency communication protocol, ensuring accurate control at the moment of photoperiod switching, introducing crop physiological state data and yield / quality feedback, combining incremental learning and Bayesian optimization to realize continuous adaptive updating of the model, and avoiding long-term performance decay caused by genetic characteristics drift of crops or aging of facilities.

[0030] 3. The photoperiod dynamic temperature adjustment plant factory light-temperature coupling energy-saving model and cloud control method, the cloud platform uses containerized microservices and multi-protocol analysis, supports plug-and-play of different manufacturers' equipment, and can perform unified scheduling and differentiated optimization of multi-factory cluster, fully utilizes regional climate and facility characteristics to further improve overall energy-saving benefits, through dynamic precise matching and global optimization scheduling of light-temperature coupling, can reduce energy consumption by more than 30% under the premise of guaranteeing high yield of crops, and promotes the development of plant factory to low carbonization and intelligentization. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be described below in a clear and complete manner. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0032] EMBODIMENT

[0033] Embodiment of the photoperiod dynamic temperature adjustment plant factory light-temperature coupling energy-saving model and cloud control method

[0034] The photoperiod dynamic temperature adjustment plant factory light-temperature coupling energy-saving model and cloud control method comprises the following steps:

[0035] S1: Multi-modal photoperiod recognition: based on multi-source data of fusion natural light intensity time series, artificial light supplement time sequence and crop canopy spectral reflectance, the actual photoperiod type and its dynamic turning point are determined in real time through a deep learning network, and the influence of natural light fluctuation and light supplement time sequence deviation on photoperiod recognition is eliminated;

[0036] S2: Cross-species light-temperature coupling modeling: a universal light-temperature coupling model is constructed with crop genotype identification, growth stage label and environmental light intensity-temperature history data as input, and a transfer learning and multi-task regression algorithm is used to adaptively map the light-temperature requirements of different crops and different growth stages, and output a dynamic target temperature curve;

[0037] S3: Cloud-edge collaborative control: deploy edge computing nodes in the plant factory for real-time collection of environmental parameters and execution of preliminary filtering and event detection, upload key state data and model inference requests to the cloud; the cloud uses a reinforcement learning optimizer combined with a digital twin simulation environment to generate high-precision temperature adjustment strategies and issue them to the edge nodes for immediate execution, achieving millisecond-level closed-loop response;

[0038] S4: Online self-updating mechanism: Continuously collect crop physiological state data and yield / quality feedback information, periodically correct parameters and fine-tune structure of the light-temperature coupling model through incremental learning and Bayesian optimization, and maintain the matching of the model and the actual needs of the crop;

[0039] S5: Cluster expansion and heterogeneous compatibility: The cloud platform supports heterogeneous device protocol conversion and lightweight container deployment, realizes unified scheduling and differentiated optimization of multi-factory area light-temperature control, and introduces regional climate and facility structure feature constraints at the global level to improve cluster energy saving benefits.

[0040] Specifically, the multi-modal photoperiod recognition in step S1 uses a convolution-cyclic hybrid neural network (CNN-LSTM), and the input includes: natural light intensity sequence, artificial light supplement switch timing, crown layer near-infrared and red band reflectance ratio, and the output is the current actual photoperiod category and confidence, and a short sliding window rejudgment mechanism is triggered when the light intensity suddenly changes.

[0041] Specifically, the cross-species light-temperature coupling model in step S2 takes crop gene database encoding, growth stage label, and environmental light intensity-temperature three-dimensional tensor as input, uses transfer learning to transfer existing crop variety model knowledge to new crops, and simultaneously predicts target temperature curve and expected photosynthetic active radiation utilization (PUE) through multi-task regression.

[0042] Specifically, the reinforcement learning optimizer of the cloud-edge collaborative control in step S3 uses the deep deterministic policy gradient (DDPG) algorithm, simulates various disturbance scenarios (device failure, power grid fluctuation, sudden stress) in the digital twin environment, trains a robust temperature regulation strategy, and realizes millisecond-level command issuance through the MQTT protocol.

[0043] Specifically, the online self-updating mechanism in step S4 introduces crop physiological sensor (chlorophyll fluorescence instrument, photosynthetic rate in-situ monitor) data, adjusts model hyperparameters through Bayesian optimization, and automatically triggers model retraining process when detecting that yield estimation deviation exceeds threshold.

[0044] Specifically, the cloud platform in step S5 uses a containerized microservice architecture, supports automatic parsing and conversion of multiple industrial protocols such as OPC-UA, Modbus, and LoRaWAN, and introduces regional climate correction coefficients and facility insulation performance decay models in cluster scheduling to realize differentiated energy saving optimization in multi-factory area.

[0045] A system for implementing any of the above methods, comprising:

[0046] A multi-source data acquisition module configured with natural light intensity sensors, artificial light supplement controllers, crown layer spectral sensors, and crop physiological sensors;

[0047] Edge computing nodes for real-time filtering of collected data, event detection, and preliminary photoperiod determination;

[0048] A cloud intelligent platform including a cross-species light-temperature coupling model library, a digital twin simulation environment, a reinforcement learning optimizer, and a cluster scheduling engine;

[0049] An execution control module that receives temperature control strategies issued by the cloud and drives the coordinated operation of heating / cooling equipment, ventilation equipment, and light supplementing equipment;

[0050] A model self-updating module that connects the yield / quality monitoring system with the incremental learning unit to realize online correction of model parameters.

[0051] Through the above technical solutions, in the present application, by fusing multi-modal data of natural light, light supplement timing, and canopy spectral reflectance, and using a CNN-LSTM deep learning network, the misjudgment caused by natural light fluctuation and light supplement deviation is effectively overcome, high-precision real-time photoperiod determination is realized, the general light-temperature coupling model constructed based on transfer learning and multi-task regression can quickly adapt to the needs of different genotypes of crops and different growth stages, avoiding the limitation of traditional models that are only applicable to a single crop, significantly improving the application range and stability of the system, through the edge node responsible for real-time collection and preliminary processing, the cloud uses reinforcement learning and digital twin for global optimization, and issues strategies using a low-latency communication protocol, solving the problem of high latency in traditional cloud computing, ensuring accurate regulation at the moment of photoperiod switching, introducing crop physiological state data and yield / quality feedback, combining incremental learning and Bayesian optimization to realize continuous adaptive updating of the model, avoiding long-term performance degradation caused by genetic characteristics drift of crops or aging of facilities, the cloud platform uses containerized microservices and multi-protocol parsing to support plug-and-play of equipment from different manufacturers and unified scheduling and differentiated optimization of multi-factory clusters, making full use of regional climate and facility characteristics to further improve overall energy saving benefits, through dynamic and accurate matching of light-temperature coupling and global optimization scheduling, the energy consumption can be reduced by more than 30% compared with traditional models while ensuring high-quality and high-yield crops, promoting the development of plant factories towards low-carbon and intelligent.

[0052] Example 1: Cross-species application of lettuce and medicinal plants

[0053] Natural light intensity sensors, LED light supplement controllers, canopy spectral reflectometers, and portable chlorophyll fluorometers are arranged in the plant factory.

[0054] The multi-modal photoperiod recognition module collects the natural light intensity curve, light supplement start / stop log, and canopy red / near-infrared ratio for 72 consecutive hours, and sends them to the CNN-LSTM network, which outputs the actual photoperiod of the day as "long day" with a confidence level of >95%.

[0055] The cross-species light-temperature coupling model uses the existing lettuce model knowledge to migrate to the medicinal plant according to the crop genetic library code (lettuce variety A and medicinal plant B), the current growth stage (lettuce is rosette stage and medicinal plant is seedling stage) and the environmental light intensity-temperature historical data, and outputs the dynamic target temperature curve: the lettuce section is 25±0.5℃ during the day and 18±0.3℃ at night; and the medicinal plant section is 22±0.5℃ during the day and 16±0.3℃ at night.

[0056] When the edge node detects that the photoperiod is switched from'short day' to 'long day', the state is immediately uploaded to the cloud, the cloud reinforcement learning optimizer simulates the cloudy to sunny and light supplement delay scene in the digital twin environment, generates the optimal temperature rising strategy and issues it to the execution module within 50ms through MQTT, and the heating device starts 0.8℃ in advance to avoid lag.

[0057] After harvesting, the yield and nutrition index are fed back to the model self-updating module, and the Bayesian optimization adjusts the temperature sensitivity coefficient of the model to the seedling stage of the medicinal plant, so that the prediction yield deviation of the next cycle is reduced from 8% to 2%.

[0058] Embodiment 2: Multi-factory cluster scheduling

[0059] Three plant factories distributed in different climate zones are connected to the same cloud platform, and the platform analyzes the LoRaWAN and Modbus device data of each factory area and automatically converts it into a unified internal protocol.

[0060] The cloud scheduling engine sets a gentler temperature rising curve for the northern alpine factory area to reduce heat loss according to the local weather forecast and facility insulation performance decay model, and strengthens the ventilation linkage control for the southern hot and humid factory area to prevent high temperature and high humidity stress.

[0061] The global optimization result shows that the comprehensive energy consumption of the three factory areas is reduced by about 36% compared with independent control, and the consistency of crop quality is improved by 15%.

[0062] As can be seen from the above embodiments, the present application is significantly superior to the prior art in terms of multi-source data fusion, cross-species modeling, real-time control, self-updating and cluster expansion, and can truly realize dynamic and precise energy-saving control of light-temperature coupling.

[0063] Although embodiments of the present application 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 therein without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A light-temperature coupled energy-saving model and cloud control method for plant factories with dynamic photoperiodic temperature regulation, characterized by: Includes the following steps: S1: Multimodal photoperiod recognition: Based on multi-source data that integrates natural light intensity time series, artificial supplementary lighting time series, and crop canopy spectral reflectance, a deep learning network is used to determine the actual photoperiod type and its dynamic turning point in real time, eliminating the impact of natural light fluctuations and supplementary lighting time series deviations on photoperiod recognition. S2: Cross-species light-temperature coupling modeling: Construct a generalized light-temperature coupling model with crop genotype identifiers, growth stage labels and historical ambient light intensity-temperature data as inputs. Use transfer learning and multi-task regression algorithms to adaptively map the light-temperature requirements of different crops and different growth stages, and output dynamic target light-temperature curves. S3: Cloud-Edge Collaborative Control: Deploy edge computing nodes in the plant factory to collect environmental parameters in real time and perform preliminary filtering and event detection, and upload key status data and model inference requests to the cloud; The cloud-based reinforcement learning optimizer, combined with a digital twin simulation environment, generates a high-precision light-temperature adjustment strategy and distributes it to edge nodes for immediate execution, achieving millisecond-level closed-loop response. S4: Online self-updating mechanism: Continuously collect crop physiological status data and yield / quality feedback information, and perform periodic parameter correction and structural fine-tuning of the light-temperature coupling model through incremental learning and Bayesian optimization to maintain the model's match with the actual needs of the crop; S5: Cluster Expansion and Heterogeneous Compatibility: The cloud platform supports protocol conversion for heterogeneous devices and lightweight container deployment, enabling unified scheduling and differentiated optimization of light-temperature control across multiple plant areas. It also introduces regional climate and facility structure characteristics constraints at the global level to improve the overall energy-saving efficiency of the cluster.

2. The photoperiodic dynamic temperature regulation plant factory light-temperature coupling energy-saving model and cloud control method according to claim 1, characterized in that: The multimodal photoperiod recognition in step S1 uses a convolutional-recurrent hybrid neural network (CNN-LSTM). The inputs include: natural light intensity sequence, artificial illumination switching timing, and the ratio of near-infrared to red light band reflectivity of the canopy. The output is the current actual photoperiod category and confidence level, and a short-time sliding window re-judgment mechanism is triggered when the light intensity changes abruptly.

3. The photoperiodic dynamic temperature regulation plant factory light-temperature coupling energy-saving model and cloud control method according to claim 1, characterized in that: The cross-species light-temperature coupling model described in step S2 takes crop gene database encoding, growth stage tags, and ambient light intensity-temperature three-dimensional tensor as input. It uses transfer learning to transfer knowledge of existing crop variety models to new crops and combines multi-task regression to simultaneously predict the target temperature curve and the expected photosynthetically active radiation utilization rate (PUE).

4. The photoperiodic dynamic temperature regulation plant factory light-temperature coupling energy-saving model and cloud control method according to claim 1, characterized in that: In step S3, the reinforcement learning optimizer for cloud-edge collaborative control adopts the Deep Deterministic Policy Gradient (DDPG) algorithm to simulate various disturbance scenarios (equipment failure, power grid fluctuations, and sudden stress from other factors) in a digital twin environment, trains a robust temperature regulation strategy, and implements millisecond-level command issuance through the MQTT protocol.

5. The photoperiodic dynamic temperature regulation plant factory light-temperature coupling energy-saving model and cloud control method according to claim 1, characterized in that: In step S4, the online self-updating mechanism introduces data from crop physiological sensors (chlorophyll fluorometer, photosynthetic rate in situ monitor), combines Bayesian optimization to adjust the model hyperparameters, and automatically triggers the model retraining process when the yield prediction deviation is detected to exceed the threshold.

6. The photoperiodic dynamic temperature regulation plant factory light-temperature coupling energy-saving model and cloud control method according to claim 1, characterized in that: In step S5, the cloud platform adopts a containerized microservice architecture, supports automatic parsing and conversion of various industrial protocols such as OPC-UA, Modbus, and LoRaWAN, and introduces regional climate correction coefficients and facility insulation performance attenuation models in cluster scheduling to achieve differentiated energy-saving optimization for multiple plant areas.

7. A system for implementing the method according to any one of claims 1-6, characterized in that: include: The multi-source data acquisition module is equipped with a natural light intensity sensor, an artificial lighting controller, a canopy spectral sensor, and a crop physiological sensor. Edge computing nodes are used for real-time filtering, event detection, and preliminary optical period determination of the collected data; The cloud-based intelligent platform includes a cross-species light-temperature coupling model library, a digital twin simulation environment, a reinforcement learning optimizer, and a cluster scheduling engine. The execution control module receives temperature control strategies from the cloud and drives the heating / cooling equipment, ventilation equipment, and supplementary lighting equipment to operate in conjunction. The model self-updating module connects the yield / quality monitoring system with the incremental learning unit to achieve online calibration of model parameters.