An AI plant light spectrum adjustment method for optimizing plant photosynthesis

Abstract

The present invention relates to the technical field of spectral regulation, and discloses an AI plant light spectrum regulation method for optimizing plant photosynthesis. Among them, the method includes: collecting original plant physiological data through an intelligent spectral sensor network and performing preprocessing to obtain a plant spectral interaction feature data set; constructing a three-dimensional spectral response mathematical model using the plant spectral interaction feature data set; using the three-dimensional spectral response mathematical model as a teacher network, and training a student network through knowledge distillation to obtain a plant spectral matching network; processing real-time plant physiological data through the plant spectral matching network, and performing plant growth stage evaluation to generate a plant growth stage identifier, personalized spectral requirement parameters, and lighting timing requirement parameters; performing a differential co-evolution algorithm to generate an LED dimming scheme. This method can perform real-time correction according to the actual spectral deviation and plant photosynthetic efficiency deviation, ensuring the accuracy and stability of spectral regulation.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral regulation, and in particular to an AI plant light spectrum regulation method for optimizing plant photosynthesis. Background Art

[0002] As the core process of plant growth, the efficiency of plant photosynthesis directly affects crop yield and quality. Traditional plant spectrum regulation systems usually adopt fixed spectrum formulas and cannot be dynamically adjusted according to the plant growth stage and real-time physiological state, resulting in low light energy utilization efficiency and difficulty in meeting the specific spectrum requirements of different plant varieties at different growth stages.

[0003] Existing plant spectrum regulation technologies face multiple challenges. Due to the complex light scattering, reflection, and absorption characteristics within plant tissues, there are significant differences in the spectral responses of different plant varieties. Existing systems are difficult to accurately capture these differences and make corresponding adjustments. Secondly, plant photosynthesis is highly sensitive to environmental conditions, and the spectral requirements change with the growth stage, circadian rhythm, and external environment. Existing technologies lack the ability to adapt to this dynamic change. In addition, the spectrum regulation process involves multi-objective optimization problems. How to balance photosynthetic efficiency, energy consumption, and economic cost has become an urgent technical problem to be solved. Summary of the Invention

[0004] The present invention provides an AI plant light spectrum regulation method for optimizing plant photosynthesis. The present invention can perform real-time correction according to the actual spectral deviation and plant photosynthetic efficiency deviation to ensure the accuracy and stability of spectrum regulation.

[0005] In a first aspect, the present invention provides an AI plant light spectrum regulation method for optimizing plant photosynthesis. The AI plant light spectrum regulation method for optimizing plant photosynthesis includes:

[0006] Collecting original plant physiological data through an intelligent spectrum sensor network and performing preprocessing to obtain a plant spectrum interaction feature dataset;

[0007] Using the plant spectrum interaction feature dataset to construct a three-dimensional spectral response mathematical model;

[0008] Taking the three-dimensional spectral response mathematical model as a teacher network and training a student network through knowledge distillation to obtain a plant spectrum matching network;

[0009] Processing real-time plant physiological data through the plant spectrum matching network and performing plant growth stage assessment to generate a plant growth stage identifier, personalized spectrum requirement parameters, and lighting timing requirement parameters;

[0010] Execute the differential co-evolution algorithm according to the plant growth stage identifier, the personalized spectral requirement parameters, and the lighting time sequence requirement parameters to generate an LED dimming scheme.

[0011] In the technical solution provided by the present invention, the complex three-dimensional spectral response mathematical model is compressed into a lightweight plant spectral matching network through knowledge distillation technology, reducing the number of parameters of the student network, significantly reducing the computational resource requirements, while maintaining the model accuracy, enabling the system to operate efficiently in resource-constrained environments. A plant growth stage evaluator with a three-unit parallel structure comprehensively analyzes morphological, physiological, and metabolite indicators, processes time series features through feature fusion and gated recurrent unit networks, and realizes the accurate identification of typical plant growth stages. A complete optimization model is constructed through four objective functions (spectral matching degree, photosynthetic efficiency, energy consumption, economic cost), combined with the differential co-evolution algorithm and adaptive mutation operation, taking into account energy consumption and economic cost while ensuring photosynthetic efficiency, and realizing multi-dimensional balanced optimization. The hierarchical spectral adjustment controller combines real-time spectral monitoring and plant response monitoring to form a closed-loop feedback control system, which can perform real-time correction according to the actual spectral deviation and plant photosynthetic efficiency deviation, ensuring the accuracy and stability of spectral adjustment. When a light source failure, environmental anomaly, or plant stress condition is detected, the system automatically switches to a preset safe spectral mode to avoid irreversible damage to plants caused by inappropriate spectra, improving the reliability and safety of the system. Calculate the personalized adjustment coefficient according to the individual characteristics of the plant, customize the spectral requirements of different plant individuals, improve the pertinence and accuracy of spectral adjustment, and meet the specific spectral requirements of different plant varieties and individuals.

[0012] Other features and advantages of the present invention will be described in the following specification, and some will become obvious from the specification, or be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, claims, and drawings.

[0013] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given, and in conjunction with the accompanying drawings, the detailed description is as follows. Description of the Drawings

[0014] Figure 1 It is a schematic diagram of an embodiment of the AI plant light spectrum adjustment method for optimizing plant photosynthesis in an embodiment of the present invention. Detailed Embodiments

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] As used in the embodiments of the present invention, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes other steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products, or devices.

[0017] To facilitate the understanding of this embodiment, first, a detailed introduction will be given to an AI plant light spectrum adjustment method for optimizing plant photosynthesis disclosed in the embodiments of the present invention. As Figure 1 shown, this method includes the following steps:

[0018] 101. Collect original plant physiological data through an intelligent spectrum sensor network and perform preprocessing to obtain a plant spectrum interaction feature dataset;

[0019] It can be understood that the execution subject of the present invention can be an AI plant light spectrum adjustment device for optimizing plant photosynthesis, or a terminal or a server. Specifically, no limitation is made here. The embodiments of the present invention will be described by taking the server as the execution subject as an example.

[0020] Specifically, spectral analysis sensors, plant physiological parameter sensors, and growth parameter sensors are configured in the plant growth environment to form an intelligent spectral sensor network. The role of the spectral analysis sensor is to monitor the spectral conditions in the plant's surrounding environment in real time, record the light intensity at different wavelengths, and the impact of these lighting conditions on the plant. The plant physiological parameter sensor is used to detect the real-time physiological state of the plant, such as physiological indicators like photosynthesis rate and chlorophyll content. This data helps to understand how the plant responds to different lighting conditions. The growth parameter sensor, on the other hand, is used to monitor the growth of the plant, including parameters directly related to plant growth such as plant height, number of leaves, and leaf area. Through the collaborative work of these sensors, data on the growth and physiological responses of the plant under different lighting conditions are comprehensively collected, forming an intelligent spectral sensor network. Through the intelligent spectral sensor network, raw plant physiological data containing spectral conditions and corresponding plant physiological responses are collected. This data includes the outputs of multiple sensors, such as the light intensity at different wavelengths at a specific moment, the plant's physiological responses to these spectral conditions (such as photosynthesis rate, plant health status, etc.), and the growth status. Outlier detection and outlier correction are performed on the raw plant physiological data to identify and eliminate data points that do not conform to the normal growth pattern of the plant. Outlier detection is achieved through statistical methods such as standard deviation detection, Z-score detection, or algorithms based on machine learning. The detected outliers are caused by external environmental interference or sensor errors and are repaired through linear interpolation or other correction algorithms to ensure the smoothness and consistency of the data, resulting in cleaned plant physiological data. Multiscale time regularization processing is performed on the cleaned plant physiological data. By using regularization algorithms to handle the fluctuations at different time scales in the data, the data becomes more stable over time. The growth and physiological responses of plants are a long-term changing process but are affected by short-term environmental fluctuations (such as instantaneous changes in light intensity). Time regularization helps to smooth out these short-term fluctuations, making the data more conform to the long-term growth pattern of the plant. Through multiscale time regularization, the information at different time scales in the data is effectively balanced, resulting in standardized time series data. Spectral response curve feature points, photosynthetic efficiency indicators, and growth phenotype features are extracted from the standardized time series data. These feature points represent the plant's responses under specific lighting conditions. The spectral response curve feature points include the response intensities at different lighting wavelengths and can reflect the photosynthesis efficiency and absorption characteristics of the plant under different wavelength lights. The photosynthetic efficiency indicators are used to quantitatively measure the photosynthesis efficiency of the plant under different lighting conditions, while the growth phenotype features include morphological indicators of the plant, such as plant height, number of leaves, and leaf area. Through these extracted features, the physiological performance of the plant under different lighting conditions is comprehensively characterized. The plant physiological feature vectors are paired with the corresponding spectral condition data in time series to form a plant spectral interaction feature dataset.The plant physiological feature vector contains information about the photosynthesis efficiency and growth phenotype of the plant under different light conditions, while the spectral condition data contains information about the actual light conditions in the environment. By pairing these two in time series, the formed plant spectral interaction feature dataset comprehensively describes the response law of the plant to spectral conditions.

[0021] 102. Construct a three-dimensional spectral response mathematical model using the plant spectral interaction feature dataset;

[0022] Specifically, spectral parameters are discretized, converting complex lighting conditions into processable discrete data. The wavelength range of a spectrum is typically continuous, but it is discretized for computation and modeling purposes. Assuming the wavelength range is from 300 nm to 800 nm, this range is divided into A bands, each representing a range of light wavelengths. The light intensity range is discretized into B levels to describe varying light intensities. The duration of light exposure is also an important parameter, as plants' light exposure requirements vary over time. Time is divided into C time periods. The entire light condition space is described as a three-dimensional discrete parameter space, where each point is composed of light wavelength, light intensity, and light duration. Based on the plant spectral interaction feature dataset, the corresponding photosynthetic efficiency index is calculated for each three-dimensional coordinate point. Photosynthetic efficiency is the ability of a plant to photosynthesize under specific light conditions and is a key indicator of spectral response. Initial spectral response data is obtained by calculating the photosynthetic efficiency index for each three-dimensional coordinate point (i.e., a specific combination of wavelength, light intensity, and duration) in the spectral interaction feature dataset. These photosynthetic efficiency indices accurately reflect the photosynthetic performance of plants under different light conditions. Three-dimensional thin-plate spline interpolation is performed on the initial spectral response data to smooth and complete the response data. Three-dimensional thin-plate spline interpolation is a multidimensional data smoothing method that generates a smooth response surface based on discrete response data points. This interpolation process considers the influence of adjacent data points to produce a continuous, smooth response surface, allowing the system to smoothly transition between light conditions. This method produces a complete response surface that describes the changes in plant photosynthesis under different light conditions. To improve the accuracy of the model, a growth stage correction factor is introduced. Different plants have different light requirements at different growth stages. Therefore, the growth stage correction factor helps the model adjust the spectral response based on the plant's current growth stage. This allows the response surface to not only consider light conditions but also reflect the plant's growth and development, resulting in a four-dimensional extended model. The four-dimensional extended model incorporates the plant's developmental stage dimension, allowing it to dynamically adjust the plant's photosynthetic efficiency while also taking light conditions into account. This four-dimensional extended model more accurately describes the plant's response to light conditions at different growth stages. To reduce model complexity and improve computational efficiency, principal component analysis (PCA) was performed on the four-dimensional extended model. PCA is a dimensionality reduction method that converts raw high-dimensional data into low-dimensional data through linear transformation while preserving as many important features as possible. PCA reduces the dimensions of the four-dimensional extended model to a lower dimension while preserving the key features of the spectral response. After PCA processing, the reduced spectral response features were obtained. A deep autoencoder network was used to perform nonlinear mapping on the reduced spectral response features, resulting in a three-dimensional spectral response mathematical model.An autoencoder is an unsupervised learning algorithm that compresses input data into a low-dimensional space through an encoder and then restores it through a decoder. A deep autoencoder is composed of multiple layers of neural networks, enabling it to learn the non-linear relationships in the data. By training the deep autoencoder network, the system automatically extracts the complex response patterns of plants to light conditions from the spectral response features after dimensionality reduction and constructs a three-dimensional spectral response mathematical model.

[0023] 103. Use the three-dimensional spectral response mathematical model as the teacher network and train the student network through knowledge distillation to obtain the plant spectral matching network;

[0024] Specifically, based on the three-dimensional spectral response mathematical model, a teacher network and a student network are constructed. The teacher network consists of 12 fully connected layers, where each layer is a computational unit for feature learning and transformation of the input data. This deep network structure can effectively extract deep features from complex spectral response data and adapt to complex lighting adjustment tasks. The student network adopts a 5-layer fully connected structure, which is simpler in structure than the teacher network, enabling the student network to perform calculations and inferences more efficiently. Especially in practical applications, the low computational cost of the student network can improve the operating efficiency of the system. Parallel feature extraction is performed on the plant physiological characteristic data and the spectral response model parameters to obtain a plant physiological characteristic vector and a spectral response characteristic vector. The plant physiological characteristic data includes the physiological responses of plants under different lighting conditions, such as photosynthesis rate, chlorophyll content, plant height, etc., while the spectral response model parameters describe the response characteristics of plants under different lighting conditions. These two feature sets respectively represent the relationship between the physiological state of plants and the external lighting conditions. Through parallel feature extraction, different information from these two feature sets is extracted. The alternating attention mechanism is realized by calculating the similarity matrix between the plant physiological characteristic vector and the spectral response characteristic vector to obtain an aligned feature representation. The calculation of the similarity matrix can reflect the correlation between the plant physiological state and the spectral response, revealing which spectral conditions are more suitable for the current physiological needs of the plant. The alternating attention mechanism is an optimization algorithm that iteratively adjusts the mutual relationship between the two feature vectors to achieve feature alignment. This process can help the system find the optimal match between the plant physiological characteristics and the spectral response characteristics, enabling spectral adjustment to more precisely meet the growth needs of plants. The direct splicing method and the cross-modal self-attention mechanism are used to perform multi-modal fusion on the aligned feature representation. The cross-modal self-attention mechanism weights the features of different modalities through the self-attention mechanism to learn the correlation between different modalities. The self-attention mechanism enables the model to automatically focus on the features that are most important for spectral adjustment, making the effect of feature fusion more precise. Through the above fusion method, the obtained fusion features contain the joint representation of plant physiological information and spectral response information. Based on the fusion features, the teacher network is trained to generate high-precision spectral adjustment parameters, and the student network is trained to generate adjustment parameters based on the compressed features. The teacher network learns the best adjustment strategy for plants under different lighting conditions through the deep network structure and outputs high-precision spectral adjustment parameters. These parameters are used as the knowledge output of the teacher network to guide the training of the student network. The student network generates relatively simplified spectral adjustment parameters through learning the compressed features, which effectively reduce the computational burden and are suitable for practical application scenarios. During the training process, the soft target knowledge distillation method is used to construct the loss function, enabling the student network to learn to generate spectral adjustment parameters by mimicking the behavior of the teacher network.Soft target knowledge distillation guides the student network to learn the knowledge of the teacher network at a lower computational complexity by using the output of the teacher network as soft labels and passing them to the student network. The loss function optimizes the parameters of the student network by calculating the difference between the output of the student network and the output of the teacher network, making it as close as possible to the output of the teacher network. The parameters of the student network are optimized and adjusted to obtain the final plant spectral matching network. Through optimization and adjustment, the student network gradually approaches the performance of the teacher network while maintaining computational efficiency. The obtained plant spectral matching network can dynamically adjust the spectral output according to the physiological characteristics of plants and lighting conditions, achieving the best optimization of plant photosynthesis.

[0025] 104. Process real-time plant physiological data through the plant spectral matching network, and perform plant growth stage assessment to generate plant growth stage identifiers, personalized spectral requirement parameters, and lighting timing requirement parameters;

[0026] Specifically, a plant growth stage evaluator is constructed, which is a composite module containing three parallel units: a morphological evaluation unit, a physiological evaluation unit, and a metabolite evaluation unit. The morphological evaluation unit evaluates the growth stage of the plant by collecting morphological data of the plant, such as the height of the plant, leaf area, number of branches, etc. The physiological evaluation unit evaluates the health status and growth progress of the plant by monitoring physiological parameters of the plant, such as photosynthesis rate, chlorophyll content, etc. The metabolite evaluation unit evaluates the metabolic state and energy reserve of the plant by detecting metabolites in the plant, such as sugars, amino acids, etc. Through these three parallel evaluation units, information from multiple aspects is integrated to accurately evaluate the growth stage of the plant. The real-time plant physiological data collected is processed using a plant spectral matching network to obtain multi-stage feature vectors. These feature vectors represent the physiological state and light requirements of the plant at different growth stages and under different environmental conditions. Through feature fusion of these feature vectors, the feature information of each stage is combined into a comprehensive plant feature representation. Through the plant growth stage evaluator, the obtained comprehensive plant feature representation is mapped to the probability distribution of D typical plant growth stages. This mapping process uses the model of the plant growth stage evaluator, takes the comprehensive plant feature representation as the input, and calculates the probability distribution of the plant belonging to each growth stage. By selecting the growth stage corresponding to the maximum probability value, it is determined which growth stage the plant is currently in. This process is comprehensively evaluated based on the physiological characteristics, morphological indicators, and metabolite data of the plant, so it has high accuracy and can reflect the actual growth status of the plant in real time. After determining the growth stage identifier of the plant, the initial spectral requirement parameters are extracted from a preset stage-spectral requirement mapping table based on this identifier. The stage-spectral requirement mapping table is established through a large amount of experimental data and plant photosynthesis theory research. The table lists parameters such as the required proportion of different wavelength spectra, required light intensity, and light cycle for plants at different growth stages. By according to the growth stage identifier of the plant, the initial spectral requirement parameters corresponding to the current growth stage of the plant are extracted from this mapping table. Using the plant spectral matching network, the initial spectral requirement parameters are adjusted according to the current individual characteristics of the plant to generate personalized spectral requirement parameters and light timing requirement parameters. The light requirements of plants are not only affected by the growth stage but also closely related to their individual characteristics, such as plant species, health status, leaf area, etc. Therefore, through the plant spectral matching network, combined with the individual characteristic data of the plant, the initial spectral requirement parameters are adjusted to generate personalized spectral requirement parameters and light timing requirement parameters. The personalized spectral requirement parameters include the specific required proportion of different wavelength lights for the plant, while the light timing requirement parameters include the required light intensity and light cycle for the plant. These parameters help the system accurately adjust the spectral output of the light source to ensure that the plant obtains the best light conditions at different growth stages.

[0027] 105. Execute a differential co-evolution algorithm based on plant growth stage identifiers, personalized spectral requirement parameters, and lighting timing requirement parameters to generate an LED dimming scheme.

[0028] Specifically, define the set of optimization variables and obtain the basic variables for spectral optimization. The basic variables for spectral optimization include parameters such as light intensity, spectral ratio, and light cycle. These variables are the core elements for adjusting the spectral output of LED lights. The selection of optimization variables directly affects the optimization effect of the dimming scheme. Therefore, it is necessary to ensure that these variables cover all the key factors required for spectral adjustment. Based on the plant growth stage identifier, personalized spectral requirement parameters, and lighting timing requirement parameters, construct a multi-objective optimization model. The model includes multiple objective functions, such as the spectral matching degree objective function, photosynthetic efficiency objective function, energy consumption objective function, and economic cost objective function. These four objective functions respectively describe the optimization requirements in different aspects of the lighting adjustment process. The spectral matching degree objective function is used to measure the matching degree between the actual lighting conditions and the plant's required spectrum, ensuring that the plant can obtain the most suitable lighting conditions; the photosynthetic efficiency objective function evaluates the impact of lighting adjustment on the plant's photosynthesis, with the goal of maximizing the photosynthesis efficiency; the energy consumption objective function ensures the energy-saving nature of the dimming scheme by controlling the light intensity and cycle to reduce unnecessary energy waste; the economic cost objective function focuses on the economy of the dimming scheme, ensuring that while meeting the plant's needs, the overall operating cost is reduced. Through the design of these four objective functions, simultaneously optimize the plant's lighting requirements, energy consumption, and economic cost. Set constraints for the multi-objective optimization model to obtain a complete optimization problem. The constraints include the power limit of the LED light source, the maximum and minimum values of the light intensity, the limit of the light cycle, etc. These limitations ensure that the optimization process is feasible in practical applications and will not exceed physical and economic limitations while meeting the plant's needs. According to the basic variables for spectral optimization, execute the differential co-evolution algorithm for the complete optimization problem. The differential co-evolution algorithm is an optimization method based on population evolution, which searches for the optimal solution by simulating the evolution process of populations in nature. According to the basic variables for spectral optimization, execute the differential co-evolution algorithm for the complete optimization problem to generate an initial evolutionary population. The initial evolutionary population consists of a set of randomly generated individuals (i.e., possible spectral adjustment schemes), and these individuals will perform subsequent evolutionary operations. The fitness of each individual is measured by its performance in terms of spectral matching degree, photosynthetic efficiency, energy consumption, and economic cost. Perform evolutionary operations on the initial evolutionary population. The evolutionary operations include steps such as selection, crossover, and mutation. In this process, adopt an elitist retention strategy based on non-dominated sorting to ensure that the optimal individuals in each generation can be passed on to the next generation. This strategy selects those individuals that are relatively superior in each objective by comparing the performance of each individual in multiple objectives and retains them. Introduce an adaptive mutation operation to enhance the diversity of the search process and avoid the algorithm falling into a local optimal solution. Adaptive mutation can automatically adjust the mutation rate according to the current state of the population, making the exploration in the search space more uniform and able to avoid prematurely converging to the vicinity of a certain solution.As evolution progresses, the differential co-evolution algorithm gradually generates a target population, which are the spectral adjustment schemes that perform optimally after multiple generations of evolution. Select the non-dominated solution set from the target population. The non-dominated solution set refers to a set of solutions where, for all objective functions, no single solution is better than all other solutions in all objectives. By selecting these non-dominated solutions, the system finds the optimal dimming scheme from multiple valid solutions. The non-dominated solution set obtained by analysis using the fuzzy comprehensive evaluation method is used to generate the final LED dimming scheme. The fuzzy comprehensive evaluation method is a multi-criteria decision-making method that can handle the complex relationships between multiple objectives and synthesize the results of multiple objective functions into a final decision result through weighted averaging and other means. Fuzzy evaluation can tolerate uncertainty and ambiguity to a certain extent, enabling the final dimming scheme to balance multiple objectives such as spectral matching, photosynthetic efficiency, energy consumption, and economic cost, and to adapt to the changing requirements in practical applications.

[0029] The design of the basic layer realizes LED drive control through Pulse Width Modulation (PWM) technology. PWM technology adjusts the brightness of LEDs by controlling the duty cycle of the current switch, and the light intensity of each LED channel is determined by calculating its PWM duty cycle. The basic control parameters of light intensity generated in this process provide preliminary brightness adjustment for each LED channel. These basic control parameters are jointly determined by the light timing demand parameters (such as sunlight cycle, light intensity change, etc.) and the demand model of plants. The basic layer is mainly responsible for precisely controlling the drive of LEDs to ensure that the brightness and light timing of each LED channel meet the growth requirements of plants. In the adaptation layer, adaptive adjustments are performed on the basic control parameters of light intensity to cope with the dynamic changes between light intensity and plant requirements. During different plant growth stages and environmental conditions, the light requirements of plants are constantly changing, so adaptive adjustments are made to the basic control parameters. The adjustments in the adaptation layer are not only based on the basic control parameters of light intensity but also need to combine the light timing demand parameters for daily variation spectrum adjustment, which means that plants have different light requirements during day and night and at different time periods. The adaptation layer ensures that plants always obtain suitable light conditions by adjusting the light parameters in real time, and these adjustments reflect the changing needs of plant growth. In the optimization layer, long-term optimization is carried out on the adjustment parameters of the adaptation layer. According to the long-term growth data of plants, environmental changes, and the long-term trend of light requirements, the optimization adjustment scheme is carried out. The optimization layer mainly focuses on long-term adjustment according to multiple objectives such as the overall photosynthesis efficiency of plants, spectrum matching degree, and energy consumption. By continuously optimizing the parameters of the adaptation layer with the accumulated historical data, the final spectrum adjustment is made to better conform to the light requirements and growth laws of plants and ensure that plants can maintain efficient photosynthesis under different climate conditions. To verify whether these adjustment parameters are effective, according to the spectrum demand parameters adjusted by the optimization layer, the actual output spectrum is measured with the help of a spectrum monitoring unit. The spectrum monitoring unit captures the spectrum output of the LED lamp in real time and compares the deviation between the actual output spectrum and the target spectrum. This process can ensure that the adjusted spectrum meets the photosynthesis requirements of plants and provides data support for subsequent adjustments. At the same time, a plant response monitoring unit is used to collect real-time response data of chlorophyll fluorescence and photosynthetic rate, which reflect the physiological responses of plants under the current light conditions. Chlorophyll fluorescence is the fluorescence response of plants to light stimulation, and its intensity is closely related to the photosynthesis efficiency; while the photosynthetic rate represents the photosynthesis efficiency of plants under specific light conditions. By collecting these plant response data in real time, the deviation between the actual spectrum and the target spectrum, as well as the deviation between the actual photosynthetic efficiency of plants and the expected efficiency, are calculated. The deviation values provide feedback signals for the system, forming a closed-loop control deviation signal. According to these deviation signals, the system generates correction coefficients to correct the adjustment parameters of the adaptation layer. The correction coefficients can be adjusted according to real-time feedback to ensure the accuracy and stability of the system in practical applications.When the system detects a light source failure, abnormal environment, or the plant is under stress, it can automatically switch to a preset safe spectral mode. The safe spectral mode is the basic lighting condition provided by the system to protect the plant's health in abnormal situations, avoiding the continuous growth of the plant in an unsuitable lighting environment, thereby preventing the plant from being damaged. Through this series of control processes, the system adjusts the spectral output in real time to ensure that the plant can obtain suitable lighting at each growth stage. The hierarchical control structure of the basic layer, adaptation layer, and optimization layer can effectively combine the real-time needs of the plant and environmental changes, making the plant lighting adjustment more refined and personalized. Through closed-loop control and real-time feedback, continuous optimization of lighting adjustment is achieved to ensure that the plant can achieve the best photosynthesis efficiency in any environment, promoting the healthy growth and efficient production of the plant.

[0030] In the embodiment of the present invention, the complex three-dimensional spectral response mathematical model is compressed into a lightweight plant spectral matching network through knowledge distillation technology, reducing the number of parameters of the student network, significantly reducing the computational resource requirements, while maintaining the model accuracy, enabling the system to operate efficiently in resource-constrained environments. The plant growth stage evaluator with a three-unit parallel structure comprehensively analyzes morphological, physiological, and metabolite indicators, processes temporal features through feature fusion and gated recurrent unit networks, and realizes the accurate identification of typical plant growth stages. A complete optimization model is constructed through four objective functions (spectral matching degree, photosynthetic efficiency, energy consumption, economic cost), combined with differential co-evolution algorithm and adaptive mutation operation, taking into account energy consumption and economic cost while ensuring photosynthetic efficiency, and realizing multi-dimensional balanced optimization. The hierarchical spectral adjustment controller combines real-time spectral monitoring and plant response monitoring to form a closed-loop feedback control system, which can perform real-time correction according to the actual spectral deviation and plant photosynthetic efficiency deviation, ensuring the accuracy and stability of spectral adjustment.

[0031] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0032] Configure spectral analysis sensors, plant physiological parameter sensors, and growth parameter sensors in the plant growth environment to form an intelligent spectral sensor network;

[0033] Collect original plant physiological data containing spectral conditions and corresponding plant physiological responses through the intelligent spectral sensor network;

[0034] Perform outlier detection and outlier correction on the original plant physiological data to obtain cleaned plant physiological data, and perform multi-scale time regularization processing on the cleaned plant physiological data to obtain standardized time series data;

[0035] Extract spectral response curve feature points, photosynthetic efficiency indicators, and growth phenotype features from the standardized time series data to obtain a plant physiological feature vector;

[0036] Pair the plant physiological feature vectors with the corresponding spectral condition data in time series to form a plant spectral interaction feature dataset.

[0037] Specifically, considering multiple key factors in plant growth, these factors include light conditions, the physiological state of plants, and plant growth parameters. The spectral analysis sensor can monitor the light conditions in the plant growth environment in real time, especially the light intensity at different wavelengths. The plant physiological parameter sensor is used to measure the physiological responses of plants, such as photosynthesis rate, chlorophyll content, etc., which can reflect the growth state of plants under specific light conditions. The growth parameter sensor is used to monitor the morphological characteristics of plants, such as plant height, number of leaves, and leaf area, etc. These parameters can help judge the health status and growth stage of plants. Through the collaborative work of these sensors, a comprehensive intelligent spectral sensor network is formed. Using the raw data collected by this intelligent spectral sensor network, the relationship data between spectral conditions and plant physiological responses is collected. These raw data include multiple aspects, such as the wavelength distribution of light, light intensity, photosynthetic efficiency of plants under different light conditions, plant growth parameters, etc. Outlier detection and outlier correction are performed on the raw data. Outliers refer to extreme data points generated due to equipment failures or abnormal fluctuations in the external environment, and these data will have an adverse impact on subsequent analysis. Outlier detection uses statistical methods, such as the Z-score method, box plot method, etc., to identify those values that deviate from the normal range. These outliers are repaired through interpolation or other correction techniques (such as median filtering) to ensure the smoothness and accuracy of the data. For example, if the chlorophyll content data measured at a certain time point is abnormally low and differs significantly from the surrounding data points, the average value or interpolation result of the surrounding time points is used to replace this abnormal data, thereby eliminating the impact caused by errors. The cleaned plant physiological data is processed by multi-scale time regularization. The growth and physiological responses of plants are a long-term dynamic process, but in the actual collected data, there will be short-term fluctuations due to environmental changes (such as temperature and humidity fluctuations) or inconsistent system acquisition frequencies. Multi-scale time regularization smooths the data at different time scales, making the data stable over a long period of time and eliminating short-term noise. This process involves filtering the raw data, such as using low-pass filters or wavelet transforms. In this way, standardized time series data is obtained. Based on the standardized time series data, the spectral response curve feature points, photosynthetic efficiency indicators, and growth phenotype characteristics are extracted to obtain the plant physiological feature vector. The spectral response curve feature points refer to the response data of plants under different light conditions, and these feature points help the system understand the optimal response of plants under different spectral conditions. For example, light at specific wavelengths (such as blue light and red light) has a significant impact on the efficiency of photosynthesis. Therefore, the feature points of photosynthetic efficiency in these bands are extracted from the dataset to help adjust the light source. The photosynthetic efficiency indicator is a measure of the ability of plants to perform photosynthesis under specific light conditions and is obtained by measuring the carbon dioxide absorption rate or oxygen release rate of plants.Growth phenotypic characteristics include morphological indicators such as plant height, leaf area, and root length, which reflect the growth status and health of plants. By extracting these important physiological characteristics from standardized time-series data, the growth and photosynthesis efficiency of plants under different light conditions are comprehensively characterized. The plant physiological feature vectors are paired with the corresponding spectral condition data in time series to form a plant spectral interaction feature dataset. Each plant physiological feature vector contains the physiological response information of the plant at a specific moment, while the corresponding spectral condition data reflects the environmental light conditions at that moment. By pairing these two, an interaction dataset containing plant physiological characteristics and light conditions is obtained.

[0038] In a specific embodiment, the process of performing step 102 may specifically include the following steps:

[0039] Discretize the spectral parameters, discretize the wavelength range into A bands, the light intensity range into B levels, and the light time dimension into C time periods to obtain a three-dimensional discrete parameter space;

[0040] Based on the plant spectral interaction feature dataset, calculate the corresponding photosynthetic efficiency index for each three-dimensional coordinate point to obtain the initial spectral response data;

[0041] Perform three-dimensional thin plate spline interpolation on the initial spectral response data to obtain a complete response surface, and expand the complete response surface by introducing a growth stage correction factor to obtain a four-dimensional extended model including the plant development stage dimension;

[0042] Perform principal component analysis on the four-dimensional extended model to obtain the spectral response characteristics after dimensionality reduction;

[0043] Use a deep autoencoder network to perform nonlinear mapping on the spectral response characteristics after dimensionality reduction to obtain a three-dimensional spectral response mathematical model.

[0044] Specifically, the core of the spectral parameter discretization process is to convert the three dimensions of the lighting conditions: wavelength, light intensity, and lighting time, into discrete numerical values. These three dimensions are key influencing factors in plant photosynthesis, and their discretization simplifies data processing, facilitating calculation and analysis. The wavelength range is typically between 400 - 730 nm, where the light wavelengths required for plant growth are mainly concentrated in the blue and red regions. Therefore, the wavelength range is discretized into 33 bands, each with a width of 10 nm. Considering the plant's light intensity requirements, the light intensity range from 0 to 1000 μmol·m⁻²·s⁻¹ is discretized into 20 levels to accurately represent different lighting intensities. Considering the impact of the lighting cycle on plant growth, the lighting time from 0 to 24 hours is discretized into 8 time periods. Through these discretization processes, a three-dimensional discrete parameter space is obtained, containing information on the three dimensions of wavelength, light intensity, and lighting time. Based on the plant spectral interaction feature dataset, for each three-dimensional coordinate point ( calculate the corresponding photosynthetic efficiency index (PEl). The photosynthetic efficiency index is an indicator obtained by weighted fusion of the photosynthetic rate (Pn), electron transport rate (ETR), and photochemical quenching coefficient (qP). The calculation formula for PEI is:

[0045]

[0046] Each index is normalized to ensure that indices with different units can be reasonably combined. The obtained initial spectral response data can reflect the photosynthetic efficiency of plants under different light conditions. The three-dimensional thin plate spline interpolation method is used to interpolate the initial spectral response data to fill in the sparse regions in the data, resulting in a complete response surface. The thin plate spline interpolation method is an effective smoothing interpolation technique that can ensure the smoothness and continuity of the data. Through this interpolation method, a more complete spectral response model is obtained. However, plants have different light requirements at different growth stages. Therefore, a growth stage correction factor G(s) is introduced to expand the response surface. This correction factor reflects the differences in light requirements of plants at different growth stages (such as the germination stage, vegetative growth stage, and reproductive growth stage). By introducing the growth stage dimension, the model can describe the photosynthesis efficiency under different light conditions and can also adapt to the development process of plants, obtaining a four-dimensional extended model that includes the plant development stage. For the obtained four-dimensional extended model, principal component analysis is used for dimensionality reduction. By extracting the main components of the data, the dimension of the data is reduced while retaining the important information in the data. After dimensionality reduction, the main characteristics of the spectral response are obtained, and redundant information is eliminated, which is helpful for the subsequent training and application of the model. Through principal component analysis, the obtained spectral response characteristics can explain more than 90% of the data variation, ensuring the accuracy and effectiveness of the model. A deep autoencoder network is used to perform a non-linear mapping on the spectral response characteristics after dimensionality reduction. An autoencoder is a neural network model that can learn the internal representation of data, which can map the input data to a low-dimensional space and reconstruct the original data. By performing a mapping on the reduced-dimensional features through a deep autoencoder, more complex non-linear relationships are learned, generating a more accurate three-dimensional spectral response mathematical model. This three-dimensional model can characterize the photosynthesis response of plants under different light conditions.

[0047] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0048] Based on the three-dimensional spectral response mathematical model, a teacher network and a student network are constructed. The teacher network consists of 12 fully connected layers, and the student network adopts a 5-layer fully connected structure;

[0049] Parallel feature extraction is performed on the plant physiological characteristic data and the spectral response model parameters to obtain a plant physiological characteristic vector and a spectral response characteristic vector;

[0050] An alternating attention mechanism is implemented by calculating the similarity matrix between the plant physiological characteristic vector and the spectral response characteristic vector to obtain an aligned feature representation;

[0051] A multimodal fusion is performed on the aligned feature representation using the direct splicing method and the cross-modal self-attention mechanism to obtain a fused feature;

[0052] Based on the fused features, train the teacher network to generate high-precision spectral adjustment parameters, and train the student network to generate adjustment parameters based on the compressed features. Construct a loss function through the soft target knowledge distillation method to obtain the parameters of the student network;

[0053] Optimize and adjust the parameters of the student network to obtain the plant spectral matching network.

[0054] Specifically, define the architectures of the teacher network and the student network. The teacher network consists of 12 fully connected layers, with each layer containing 256 neurons, and the activation function is GELU. This structure can effectively capture complex spectral-physiological feature relationships. The structure of the student network is relatively simplified, using 5 fully connected layers, with each layer containing 64 neurons, and the activation function is ReLU. This simplified structure enables the student network to reduce the consumption of computing resources and is suitable for applications in resource-constrained environments. During the process of constructing the network, parallel feature extraction is performed on the plant physiological feature data and the spectral response model parameters. These data include physiological response data such as the photosynthesis efficiency, chlorophyll fluorescence, and photosynthetic rate of plants, as well as spectral characteristics under light conditions. These data are processed by two independent feature extraction modules respectively to generate the plant physiological feature vector P and the spectral response feature vector S. The plant physiological feature vector P is extracted through a convolutional neural network, while the spectral response feature vector S is extracted through a one-dimensional convolution and frequency domain analysis layer. Through this parallel processing, features are extracted from both the physiological and spectral aspects simultaneously. Calculate the similarity matrix between the plant physiological feature vector and the spectral response feature vector to implement the alternating attention mechanism and obtain the aligned feature representation. The calculation formula for the similarity matrix M is:

[0055]

[0056] where, and are the plant physiological feature vector and the spectral response feature vector respectively, is the dimension of the feature. By calculating the similarity matrix , the similarity between the two feature vectors is quantified, and the features are updated bidirectionally based on this similarity. Through 3 iterations, the feature representations will be fully aligned, thus ensuring the correlation between the physiological features and the spectral features. Use direct concatenation and cross-modal self-attention mechanism to perform multi-modal fusion on the aligned features. The direct concatenation method directly concatenates the plant physiological feature vector and the spectral response feature vector into a large vector, and then performs fusion through a multi-layer perceptron; while the cross-modal self-attention mechanism calculates the interactions between different modalities, enabling the model to automatically identify which features are most critical for spectral adjustment. Through these two fusion methods, the fused feature , these features can comprehensively consider the physiological needs of plants and light conditions. Based on the fused features, the teacher network is trained to generate high-precision spectral adjustment parameters , and the student network is trained to generate adjustment parameters through the knowledge distillation method . The output of the teacher network is the complete high-precision spectral adjustment parameters, while the student network generates the corresponding adjustment parameters from the compressed features. To enable the student network to effectively learn the knowledge of the teacher network, the loss function is constructed through the soft target knowledge distillation method. The loss function of soft target knowledge distillation is expressed as:

[0057]

[0058] where is the cross-entropy loss, representing the difference between the outputs of the student network and the teacher network, is the Kullback-Leibler divergence, measuring the distribution difference between the two, is the temperature coefficient, controlling the softening degree. Through this loss function, the student network can gradually approach the accuracy of the teacher network while maintaining a low computational complexity. The parameters of the student network are optimized and adjusted, and finally a computationally efficient and high-precision plant spectral matching network is obtained. This optimization process uses a standard gradient descent method, such as the Adam optimizer, to adjust the parameters of the student network by evaluating the network performance on the validation dataset, reducing the computational complexity during inference, and making the student network suitable for real-time spectral adjustment tasks. The trained plant spectral matching network can generate matching spectral adjustment parameters according to real-time plant physiological data, thereby optimizing the photosynthesis efficiency of plants.

[0059] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0060] Construct a plant growth stage evaluator, which includes three parallel units: a morphology evaluation unit, a physiology evaluation unit, and a metabolite evaluation unit;

[0061] Use the plant spectral matching network to process the collected real-time plant physiological data to obtain a multi-stage feature vector, and perform feature fusion on the multi-stage feature vector to obtain a comprehensive plant feature representation;

[0062] Through the plant growth stage evaluator, map the comprehensive plant feature representation to the probability distribution of D typical plant growth stages, and determine the plant growth stage identifier through the maximum probability value;

[0063] Based on the plant growth stage identifier, extract the initial spectral requirement parameters from the preset stage-spectral requirement mapping table. The initial spectral requirement parameters include the requirements for the spectral ratio of different wavelengths, light intensity, and light cycle;

[0064] Adjust the initial spectral requirement parameters according to the current plant individual characteristics using the plant spectral matching network to generate personalized spectral requirement parameters and lighting time sequence requirement parameters.

[0065] Specifically, the growth stage of plants is affected by multiple factors, including morphological characteristics, physiological status, and metabolic activities. The evaluator extracts features from these three dimensions respectively and conducts comprehensive analysis on them. The morphological evaluation unit analyzes growth indicators such as the height growth rate, leaf area expansion rate, and internode length of plants. Process the time series features through a gated recurrent unit (GRU) to obtain the stage feature vector reflecting the morphological development of plants. The physiological evaluation unit analyzes key physiological indicators such as chlorophyll content, photosynthetic rate change trend, and water use efficiency. These indicators can reveal the photosynthesis status and physiological health of plants and generate the stage feature vector. The metabolite evaluation unit generates the stage feature vector related to metabolism by analyzing the content changes of metabolites such as sugars, proteins, and secondary metabolites. These feature vectors provide data support for the evaluation of plants at different stages. Through parallel processing, each evaluation unit uses a 128-dimensional GRU hidden layer to extract time series features, and then fuses these three vectors through a feature fusion device. The fusion process calculates the weighted sum where is the corresponding weight matrix, and the optimization objective adjusts these weight matrices through the backpropagation algorithm. The fused feature vector is mapped to the probability distribution of 7 typical plant growth stages (such as seed germination stage, seedling stage, early vegetative growth stage, late vegetative growth stage, flower bud differentiation stage, flowering and fruiting stage, and maturity stage), and the current growth stage identifier of the plant is determined through the maximum probability value , and extract the initial spectral requirement parameters from the preset stage-spectral requirement mapping table. This mapping table contains the spectral requirement parameters corresponding to each growth stage, such as the proportion requirements of blue light (400 - 500 nm), green light (500 - 600 nm), red light (600 - 700 nm), and far red light (700 - 730 nm), as well as the corresponding light intensity, light cycle, and light rhythm, etc. The parameterization of spectral requirements provides spectral intensity ratio and time series information for each stage. Using the plant spectral matching network, adjust these initial spectral requirement parameters according to the physiological characteristics of the current plant individual. The adjustment of personalized spectral requirement parameters is achieved by calculating the adjustment coefficient with the formula:

[0066]

[0067] where is the individual feature vector, and Adjust the parameters of the network, is the Sigmoid function to generate personalized spectral demand parameters and lighting timing demand parameters . These parameters describe the spectral intensity change requirements and the change rules of light intensity of plants at different growth stages. Through this process, real-time plant physiological data can be processed, the growth stage of plants can be accurately evaluated, and personalized spectral demand parameters and lighting timing demand parameters can be generated.

[0068] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0069] Define the set of optimization variables to obtain the basic variables for spectral optimization;

[0070] Based on the plant growth stage identifier, personalized spectral demand parameters, and lighting timing demand parameters, construct a spectral matching degree objective function, a photosynthetic efficiency objective function, an energy consumption objective function, and an economic cost objective function to obtain a multi-objective optimization model;

[0071] Set constraints for the multi-objective optimization model to obtain a complete optimization problem;

[0072] According to the basic variables for spectral optimization, execute the differential co-evolution algorithm on the complete optimization problem to obtain an initial evolutionary population;

[0073] Perform evolutionary operations on the initial evolutionary population, adopt the elitist retention strategy based on non-dominated sorting, and introduce an adaptive mutation operation to obtain the target population;

[0074] Select the non-dominated solution set from the target population and use the fuzzy comprehensive evaluation method to analyze and obtain the LED dimming scheme.

[0075] Specifically, define the set of optimization variables , where each variable represents the dimming level of the th type of LED light source, and the value range is [0, 1], indicating from completely off to maximum brightness. For example, the types of LEDs in the system , including deep blue light (440nm), blue light (460nm), green light (520nm), yellow light (590nm), red light (660nm), and far red light (730nm). These variables determine the output spectral intensity of each LED and are the basic variables for constructing spectral optimization. Based on the plant growth stage identifier, personalized spectral demand parameters, and lighting timing demand parameters, construct multiple objective functions. The spectral matching degree objective function is used to measure the similarity between the actual output spectrum and the target spectrum, and its form is:

[0076]

[0077] Among them, represents the actually output spectrum, is the target spectrum generated according to the requirements of the plant growth stage. The function measures the deviation between the actual spectrum and the target spectrum. The smaller the deviation, the higher the spectrum matching degree. Photosynthetic efficiency objective function Calculates the photosynthetic efficiency index (PEl) based on a three-dimensional spectral response model, and its form is:

[0078]

[0079] Among them, is the photosynthetic efficiency index, is the plant growth stage identifier, and this function aims to maximize the photosynthesis efficiency. Energy consumption objective function Calculates the total energy consumption of the system:

[0080]

[0081] Among them, is the power consumption of the th LED, is the dimming level of the LED. The objective of this function is to minimize the energy consumption. Economic cost objective function Is used to calculate the economic cost of the system:

[0082]

[0083] Among them, is the unit time usage cost of the th LED, is the dimming level of the LED. The objective is to reduce the operating cost of the system. Through the above objective functions, a multi-objective optimization model is constructed, aiming to simultaneously optimize the spectrum matching degree, photosynthetic efficiency, energy consumption and economic cost. In the multi-objective optimization model, constraint conditions are set, such as the total power constraint:

[0084]

[0085] And the light intensity constraint:

[0086]

[0087] These constraints ensure that the system meets the requirements in terms of power and safety while providing the required spectrum and light intensity. Use the differential co-evolution algorithm to solve this optimization problem. The differential co-evolution algorithm is an effective multi-objective optimization algorithm that searches for Pareto front solutions through population evolution. Define the initial population , the population size is , and each individual represents a set of LED dimming schemes. Based on the principle of differential co-evolution algorithm, evolutionary operations are performed. Each individual is updated according to the gradient of the objective function, and the evolution of the population is evaluated by calculating the fitness function to assess the quality of the solution. In each generation of evolution, an elitist retention strategy based on non-dominated sorting is adopted to ensure that the optimal solution of each generation is retained to the next generation. To improve the convergence speed of the algorithm, an adaptive mutation operation is introduced, and the mutation probability linearly decreases from the initial value of 0.1 to 0.01 as the number of iterations increases. After multiple generations of evolution, the target population is obtained, and the non-dominated solution set is selected from it through an improved Pareto front identification method. These non-dominated solution sets represent the trade-off equilibrium points between different objectives. Through the fuzzy comprehensive evaluation method, the optimal solution is selected from these non-dominated solutions , which achieves the best balance among the four objective functions of spectral matching degree, photosynthetic efficiency, energy consumption, and economic cost.

[0088] In a specific embodiment, the AI plant light spectrum adjustment method for optimizing plant photosynthesis further includes the following steps:

[0089] Design a hierarchical spectrum adjustment controller including a basic layer, an adaptation layer, and an optimization layer. The basic layer realizes LED drive control through pulse width modulation technology, calculates the PWM duty cycle for each LED channel, and obtains the basic light intensity control parameters;

[0090] Perform adaptive adjustment on the basic light intensity control parameters, and combine the light timing demand parameters to achieve daily variation spectrum adjustment, and obtain the adaptation layer adjustment parameters;

[0091] Perform long-term optimization on the adaptation layer adjustment parameters to obtain the optimized spectrum demand parameters after the optimization layer adjustment;

[0092] Measure the actual output spectrum according to the optimized spectrum demand parameters after the optimization layer adjustment through the spectrum monitoring unit, and collect the real-time response data of chlorophyll fluorescence and photosynthetic rate through the plant response monitoring unit;

[0093] Calculate the deviation between the actual spectrum and the target spectrum and the deviation between the actual photosynthetic efficiency of the plant and the expected efficiency according to the real-time response data to obtain the closed-loop control deviation signal;

[0094] Generate a correction coefficient based on the closed-loop control deviation signal, and perform a correction operation on the adaptation layer adjustment parameters. When a light source failure, environmental anomaly, or plant stress condition is detected, the system automatically switches to the preset safe spectrum mode to obtain the final real-time adjustment parameters.

[0095] Specifically, the hierarchical spectrum adjustment controller consists of three key levels: the basic layer, the adaptation layer, and the optimization layer. The basic layer realizes LED drive control through Pulse-Width Modulation (PWM) technology. The working principle of PWM is to control the output light intensity by adjusting the duty cycle of the LED. The frequency of PWM is set to 1200 Hz, and the dimming resolution is set to 12 bits, that is, the dimming level of each LED channel is determined by the value of the duty cycle, with a value range of 0 to 4095. For each LED channel , its PWM duty cycle The calculation formula is:

[0096]

[0097] where is the dimming level of the th type of LED. This formula controls the output light intensity of the LED through the PWM signal according to the optimized spectral demand parameters . The role of the adaptation layer is to perform dynamic adjustment of the spectrum according to real-time plant physiological data and lighting timing requirements. Every 30 minutes, the adaptation layer will perform adaptive adjustment on the basic light intensity control parameters. Based on the plant physiological data, the adaptive adjustment coefficient is calculated, and the adjustment formula is:

[0098]

[0099] where has a value range of [-0.15, 0.15], reflecting the changing trend of the plant photosynthetic efficiency. The adaptation layer combines the lighting timing demand parameters to achieve daily-changing spectrum adjustment, which simulates the characteristics of natural light changing over time, thereby optimizing the adjustment for different time periods of plant growth. The optimization layer is responsible for long-term spectrum adjustment optimization, with an execution period of 24 hours. The optimization layer optimizes the spectral parameters by comprehensively analyzing the plant physiological data and photosynthetic efficiency data of the past 24 hours and using the gradient descent method. The update formula for the spectral demand parameters is:

[0100]

[0101] where is the learning rate, set to 0.05, is the gradient of the photosynthetic efficiency with respect to the spectral demand parameters. Calculating this gradient obtains the influence of each spectral intensity on the change of photosynthetic efficiency, thereby optimizing the spectral demand parameters to obtain the optimized spectral configuration. According to the spectral demand parameters adjusted by the optimization layer, the actual output spectrum is measured by the spectral monitoring unit 。The spectral monitoring unit is equipped with a micro spectrometer, with a sampling frequency of once every 5 minutes, to monitor the changes in the output spectrum in real time. The plant response monitoring unit will collect the real-time physiological data of the plants, including chlorophyll fluorescence, photosynthetic rate, etc. By calculating the difference between the actual spectrum and the target spectrum and the difference between the actual photosynthetic efficiency of the plants and the expected efficiency the closed-loop control deviation signal is obtained. These deviation signals reflect the deviation between spectral regulation and plant response, providing feedback information for subsequent correction and optimization. Based on the closed-loop control deviation signal, a correction coefficient is generated, which is used to correct the adaptation layer adjustment parameters. The correction formula is:

[0102]

[0103] The corrected spectral parameters are sent into the system to achieve more accurate spectral regulation. When the system detects conditions such as light source failure, environmental abnormality, or plant stress, it automatically switches to the preset safe spectral mode. This mechanism can ensure that plants still receive sufficient light in harsh environments, while avoiding plant damage caused by excessive regulation. Through the above hierarchical control structure and feedback mechanism, the system can continuously adjust and optimize the spectral regulation parameters according to the real-time spectral monitoring data and plant response feedback, so as to achieve a dynamically optimal photosynthetic environment, ensure that plants can obtain the best light conditions at each growth stage, and improve their photosynthesis efficiency.

[0104] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0105] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. And the foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0106] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An AI plant light spectrum adjustment method for optimizing plant photosynthesis, characterized in that, The method includes: Collecting original plant physiological data through an intelligent spectral sensor network and performing preprocessing to obtain a plant spectral interaction feature dataset; Constructing a three-dimensional spectral response mathematical model using the plant spectral interaction feature dataset; specifically including: discretizing spectral parameters, discretizing the wavelength range into A bands, the light intensity range into B levels, and the light duration dimension into C time periods to obtain a three-dimensional discrete parameter space; based on the plant spectral interaction feature dataset, calculating the corresponding photosynthetic efficiency index for each three-dimensional coordinate point to obtain initial spectral response data; performing three-dimensional thin plate spline interpolation on the initial spectral response data to obtain a complete response surface, and expanding the complete response surface by introducing a growth stage correction factor to obtain a four-dimensional extended model including the plant development stage dimension; performing principal component analysis on the four-dimensional extended model to obtain the spectral response features after dimensionality reduction; using a deep autoencoder network to perform non-linear mapping on the spectral response features after dimensionality reduction to obtain a three-dimensional spectral response mathematical model; Taking the three-dimensional spectral response mathematical model as a teacher network and training a student network through knowledge distillation to obtain a plant spectral matching network; Processing real-time plant physiological data through the plant spectral matching network and performing plant growth stage assessment to generate a plant growth stage identifier, personalized spectral requirement parameters, and lighting timing requirement parameters; Executing a differential co-evolution algorithm according to the plant growth stage identifier, the personalized spectral requirement parameters, and the lighting timing requirement parameters to generate an LED dimming scheme.

2. The AI plant light spectrum adjustment method for optimizing plant photosynthesis according to claim 1, wherein, The collecting original plant physiological data through an intelligent spectral sensor network and performing preprocessing to obtain a plant spectral interaction feature dataset includes: Configuring spectral analysis sensors, plant physiological parameter sensors, and growth parameter sensors in the plant growth environment to form an intelligent spectral sensor network; Collecting original plant physiological data including spectral conditions and corresponding plant physiological responses through the intelligent spectral sensor network; Performing outlier detection and outlier correction on the original plant physiological data to obtain cleaned plant physiological data, and performing multi-scale time regularization processing on the cleaned plant physiological data to obtain standardized time series data; Extracting spectral response curve feature points, photosynthetic efficiency indicators, and growth phenotype features from the standardized time series data to obtain a plant physiological feature vector; Pairing the plant physiological feature vector with the corresponding spectral condition data in time series to form a plant spectral interaction feature dataset.

3. The AI plant light spectrum adjustment method for optimizing plant photosynthesis according to claim 1, characterized in that, The taking the three-dimensional spectral response mathematical model as a teacher network and training a student network through knowledge distillation to obtain a plant spectral matching network includes: Constructing a teacher network and a student network based on the three-dimensional spectral response mathematical model, the teacher network consists of 12 fully connected layers, and the student network adopts a 5-layer fully connected structure; Performing parallel feature extraction on plant physiological feature data and spectral response model parameters to obtain a plant physiological feature vector and a spectral response feature vector; The alternating attention mechanism is implemented by calculating the similarity matrix between the plant physiological feature vector and the spectral response feature vector to obtain an aligned feature representation; The direct splicing method and the cross-modal self-attention mechanism are used to perform multi-modal fusion on the aligned feature representation to obtain a fused feature; Based on the fused feature, the teacher network is trained to generate high-precision spectral adjustment parameters, and the student network is trained to generate adjustment parameters based on the compressed feature. The loss function is constructed by the soft target knowledge distillation method to obtain the student network parameters; The student network parameters are optimized and adjusted to obtain a plant spectral matching network.

4. The AI plant light spectrum adjustment method for optimizing plant photosynthesis according to claim 1, characterized in that, Processing the real-time plant physiological data through the plant spectral matching network and performing plant growth stage assessment to generate a plant growth stage identifier, personalized spectral demand parameters, and lighting time sequence demand parameters, including: Constructing a plant growth stage evaluator, which includes three parallel units: a morphology evaluation unit, a physiology evaluation unit, and a metabolite evaluation unit; Using the plant spectral matching network to process the collected real-time plant physiological data to obtain a multi-stage feature vector, and performing feature fusion on the multi-stage feature vector to obtain a comprehensive plant feature representation; Through the plant growth stage evaluator, mapping the comprehensive plant feature representation to the probability distribution of D typical plant growth stages, and determining the plant growth stage identifier through the maximum probability value; Based on the plant growth stage identifier, extracting the initial spectral demand parameters from the preset stage-spectral demand mapping table, where the initial spectral demand parameters include the requirements for the spectral ratio of different wavelengths, light intensity, and light cycle; Using the plant spectral matching network to adjust the initial spectral demand parameters according to the current plant individual characteristics to generate personalized spectral demand parameters and lighting time sequence demand parameters.

5. The AI plant light spectrum adjustment method for optimizing plant photosynthesis according to claim 1, wherein, Performing a differential co-evolution algorithm according to the plant growth stage identifier, the personalized spectral demand parameters, and the lighting time sequence demand parameters to generate an LED dimming scheme, including: Defining a set of optimization variables to obtain the basic variables for spectral optimization; Constructing a spectral matching degree objective function, a photosynthetic efficiency objective function, an energy consumption objective function, and an economic cost objective function based on the plant growth stage identifier, the personalized spectral demand parameters, and the lighting time sequence demand parameters to obtain a multi-objective optimization model; Setting constraint conditions for the multi-objective optimization model to obtain a complete optimization problem; According to the basic variables for spectral optimization, performing a differential co-evolution algorithm on the complete optimization problem to obtain an initial evolutionary population; Performing evolutionary operations on the initial evolutionary population, adopting an elitist retention strategy based on non-dominated sorting, and introducing an adaptive mutation operation to obtain a target population; Selecting a non-dominated solution set from the target population and using the fuzzy comprehensive evaluation method to analyze and obtain an LED dimming scheme.

6. The AI plant light spectrum adjustment method for optimizing plant photosynthesis according to claim 1, characterized in that, The AI plant light spectrum adjustment method for plant photosynthesis optimization further includes: Design a hierarchical spectrum adjustment controller including a basic layer, an adaptation layer, and an optimization layer. The basic layer realizes LED drive control through pulse width modulation technology, calculates the PWM duty cycle for each LED channel, and obtains the basic light intensity control parameters. Perform adaptive adjustment on the basic light intensity control parameters, and combine with the light intensity timing requirement parameters to achieve daily variation spectrum adjustment, and obtain the adaptation layer adjustment parameters. Perform long-term optimization on the adaptation layer adjustment parameters to obtain the spectrum requirement parameters after adjustment by the optimization layer. Measure the actual output spectrum through the spectrum monitoring unit according to the spectrum requirement parameters after adjustment by the optimization layer, and collect the real-time response data of chlorophyll fluorescence and photosynthetic rate through the plant response monitoring unit. Calculate the deviation between the actual spectrum and the target spectrum and the deviation between the actual photosynthetic efficiency of the plant and the expected efficiency according to the real-time response data, and obtain the closed-loop control deviation signal. Generate a correction coefficient based on the closed-loop control deviation signal, and perform a correction operation on the adaptation layer adjustment parameters. When a light source failure, environmental anomaly, or plant stress condition is detected, the system automatically switches to a preset safe spectrum mode to obtain the final real-time adjustment parameters.

Images