Plant growth cycle regulation method and system based on natural environment trajectory reconstruction

Abstract

The application discloses a plant growth cycle regulation method and system based on natural environment trajectory reconstruction, and relates to the technical field of plant physiological ecology regulation. The method comprises the following steps: collecting multi-dimensional environment data and phenotype image data in a natural habitat, generating environment time series data and phenotype image time series data; identifying and marking a plurality of growth stage transition nodes; establishing a coupling relationship model between environment parameters and growth stages; performing time axis reconstruction processing on the environment time series data according to a preset regulation target, and generating a target environment regulation trajectory; and dynamically adjusting an actuator in an indoor controllable environment by using a closed-loop control algorithm, so that the actual environment parameters in the indoor environment track and reproduce the target environment regulation trajectory. The system comprises a natural habitat learning unit, a growth stage identification unit and a coupling modeling unit. The application can realize early flowering, yield improvement and off-season production, and provides a precise regulation tool for a plant factory.

Description

Technical Field

[0001] This invention relates to the field of intelligent agricultural equipment technology, specifically to a method and system for regulating plant growth cycles based on natural environmental trajectory reconstruction. Background Technology

[0002] Existing greenhouse environmental control systems mostly use fixed parameters or empirical models for environmental regulation. Their core control logic is static maintenance: maintaining a constant temperature and humidity, and even if there are changes, they are based solely on a simple time-on / off mode. This regulation method cannot accurately reproduce the succession trajectory of plant growth stages in the natural environment, lacks system modeling of the coupling effects of multidimensional environmental factors, and cannot achieve precise control over the timing of growth stage transitions, such as actively advancing or delaying the flowering period.

[0003] Furthermore, existing technologies cannot specifically enhance or extend certain growth stages, such as accelerating growth during the vegetative growth period or extending the fruiting period to increase yield. Therefore, there is an urgent need for a method and system that can learn from the laws of the natural environment and achieve designable and predictable regulation of the entire plant growth cycle through timeline editing. Summary of the Invention

[0004] This invention aims to provide a method and system that can establish a coupling relationship between growth stages and environmental parameters based on long-term monitoring data of outdoor natural habitats, and achieve precise regulation of plant growth cycles through time axis reconstruction.

[0005] To achieve the above objectives, this invention provides a method for regulating plant growth cycles based on natural environmental trajectory reconstruction, comprising the following steps:

[0006] S1. Deploy sensor arrays and image acquisition devices in the natural habitat of plants to continuously collect multidimensional environmental data and phenotypic image data, and generate environmental time series data and phenotypic image time series data.

[0007] S2. Based on the environmental time series data and the phenotypic image time series data, identify and mark multiple growth stage transition nodes, including the germination period, vegetative growth period, flower bud differentiation period and fruiting period.

[0008] S3. Based on the identified reproductive stage transition nodes, establish a coupling relationship model between environmental parameters and reproductive stages;

[0009] S4. Based on the preset control target, the environmental time series data is reconstructed on a time axis to generate a target environmental control trajectory.

[0010] S5. In a controllable indoor environment, a closed-loop control algorithm is used to dynamically adjust the actuator so that the actual indoor environmental parameters track and reproduce the target environment control trajectory.

[0011] Furthermore, the identification and marking of reproductive stage transition nodes in step S2 specifically includes: using a Hidden Markov Model (HMM) for stage identification, wherein the HMM takes the environmental observation vector as input and the reproductive stage as the hidden state, and decodes it using the Viterbi algorithm to obtain the most probable stage sequence.

[0012] Furthermore, the observation probability distribution of the Hidden Markov Model (HMM) is fitted using a Gaussian Mixture Model (GaM), the parameters of which are obtained by training on historical data using an expectation-maximization algorithm. The state transition probability matrix of the HMM is dynamically adjusted according to an accumulated temperature model, the calculation formula of which is... Where GDD(t) is the effective accumulated temperature at time t; For integration time variable, for Real-time air temperature at any given moment This is the reference temperature.

[0013] Furthermore, the time-axis reconstruction processing of the environmental time-series data in step S4 specifically includes: performing a Fourier transform on the environmental time-series data within each reproductive stage to obtain the environmental spectrum of each reproductive stage.

[0014] in Let be the environmental parameter vector function for the i-th reproductive stage; determine the time axis reconstruction coefficients for each reproductive stage based on the preset control targets. And perform frequency domain scaling on the environmental spectrum. Then through inverse Fourier transform Generate reconstructed environmental trajectories for each reproductive stage; then, stitch together the reconstructed environmental trajectories for each reproductive stage in chronological order to obtain the target environmental regulation trajectory.

[0015] Furthermore, the time axis reconstruction coefficients The Bayesian optimization algorithm is used to determine that the Bayesian optimization takes yield or growth cycle as the objective function and the reconstruction coefficients of each stage as hyperparameters, and iteratively searches for the optimal combination of reconstruction coefficients based on a Gaussian process surrogate model.

[0016] Furthermore, the frequency domain scaling process also includes independent control of specific frequency bands: the environmental spectrum Apply frequency domain filter By preserving or enhancing specific frequency components, a filtered spectrum is obtained. .

[0017] Furthermore, after splicing the reconstructed environmental trajectories of each reproductive stage in chronological order, the method also includes smoothing the splicing points using a wavelet threshold-based denoising method. By performing wavelet decomposition on the discontinuous points at the splicing points and filtering out high-frequency noise, a continuous and smooth target environmental regulation trajectory is obtained.

[0018] Furthermore, the closed-loop control algorithm in step S5 adopts an iterative learning control algorithm. The iterative learning control algorithm uses the deviation between the actual environmental trajectory of the previous growth cycle and the target environmental regulation trajectory to correct the control input of the current growth cycle, so that the actual indoor environmental parameters successively approach and reproduce the target environmental regulation trajectory.

[0019] Furthermore, the iterative learning control algorithm employs a PD-type learning law. ,in This is the control input for the kth growth cycle. The deviation between the actual environmental trajectory and the target environmental control trajectory. and To learn the gain matrix.

[0020] Another objective of this invention is to provide a plant growth cycle regulation system based on natural environmental trajectory reconstruction, the system comprising:

[0021] Natural habitat learning units are deployed in the natural habitats of plants to continuously collect multidimensional environmental data and phenotypic image data, and generate environmental time series data and phenotypic image time series data.

[0022] A reproductive stage identification unit, connected to the natural habitat learning unit, is used to receive the environmental time series data and the phenotypic image time series data, and to identify and mark multiple reproductive stage transition nodes;

[0023] A coupling modeling unit, connected to the reproductive stage identification unit, is used to establish a coupling relationship model between environmental parameters and reproductive stages based on the identified reproductive stage transition nodes;

[0024] The time axis reconstruction unit, connected to the coupled modeling unit, is used to reconstruct the environmental time series data according to the preset control target and generate the target environmental control trajectory.

[0025] An indoor closed-loop control unit is deployed in the indoor controllable environment and connected to the time axis reconstruction unit. It is used to receive the target environment control trajectory and dynamically adjust the actuator using a closed-loop control algorithm so that the actual indoor environmental parameters track and reproduce the target environment control trajectory.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] 1. This invention establishes a coupling relationship model between environmental parameters and growth stages by collecting long-term multidimensional environmental data and phenotypic image data in natural habitats. This makes indoor regulation no longer dependent on fixed empirical parameters, but based on the real natural growth laws of plants, which significantly improves the biological rationality and success rate of regulation schemes.

[0028] 2. This invention transforms the plant development time from an irreversible physical variable into an editable engineering parameter by compressing or extending the time axis of the original environmental time series. It can compress the time axis to advance the flowering period and shorten the production cycle, or extend the time axis to prolong the fruiting period and increase biomass accumulation, thereby achieving active control over the growth cycle.

[0029] 3. This invention allows for the setting of independent time axis reconstruction coefficients for the germination period, vegetative growth period, flower bud differentiation period, and fruiting period. In the frequency domain reconstruction process, a frequency domain filter is introduced to adjust the gain of specific frequency components, so as to realize independent control of the development rate of each stage and fine shaping of environmental fluctuation characteristics, thereby meeting diverse production goals.

[0030] 4. This invention constructs a complete technical link through a natural habitat learning unit, a reproductive stage identification unit, a coupled modeling unit, a time axis reconstruction unit, and an indoor closed-loop control unit, enabling the system to have full automation capabilities from natural learning to precise indoor execution, providing a complete intelligent solution for facility agriculture. Attached Figure Description

[0031] Figure 1 This is a system structure block diagram according to an embodiment of the present invention;

[0032] Figure 2 This is a schematic diagram of data acquisition according to an embodiment of the present invention;

[0033] Figure 3 This is a flowchart illustrating the process of identifying reproductive nodes according to an embodiment of the present invention.

[0034] Figure 4 This is a schematic diagram of time axis reconstruction according to an embodiment of the present invention;

[0035] Figure 5 This is a diagram of the indoor control closed-loop structure according to an embodiment of the present invention. Detailed Implementation

[0036] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit the invention.

[0037] This embodiment uses strawberry as an example to illustrate the specific implementation process of the method of the present invention.

[0038] Step S1: Deploy sensor arrays and image acquisition devices in the natural habitat of plants to continuously collect multidimensional environmental data and phenotypic image data, and generate environmental time series data and phenotypic image time series data.

[0039] This step is the data acquisition and infrastructure construction phase, which provides raw data support for all subsequent analyses. A sensor array refers to a network of multiple physical sensors deployed in the plant's natural growth environment, specifically including air temperature sensors for measuring air temperature, soil moisture sensors for measuring soil moisture, and light intensity sensors for measuring light intensity. Multidimensional environmental data refers to the simultaneous collection of multiple environmental physical quantities, including at least air temperature, soil moisture, and light intensity; these are the core environmental factors driving plant growth and development.

[0040] Image acquisition devices refer to high-definition industrial cameras or multispectral cameras used to periodically capture visible light or multispectral images of the plant canopy, forming phenotypic image data, which is an image sequence reflecting the external morphological characteristics of the plant (such as plant height, leaf area, and leaf color). Environmental time-series data refers to environmental parameter sampling values ​​arranged in chronological order, such as a sequence of air temperature values ​​recorded every 10 minutes; phenotypic image time-series data refers to plant images arranged in chronological order, such as a canopy photograph taken at noon each day. The principle of this step is that the plant's growth process is the comprehensive result of the long-term effects of environmental factors. By synchronously acquiring environmental parameters and phenotypic images over a long period of time and at high frequency, the complete life cycle of the plant from germination to maturity can be fully recorded, laying the data foundation for subsequently establishing an environment-stage mapping relationship.

[0041] like Figure 2 As shown, the data acquisition process is carried out synchronously on the timeline: from day 1 to day n, environmental data (air temperature, soil moisture, light intensity) and plant images are collected every day, and finally environmental time series data and phenotypic image time series data are generated.

[0042] Specifically, in this embodiment, a weather station (to collect air temperature, humidity, and light) and a soil sensor (to collect soil moisture) were set up in a natural open-air strawberry field. A high-definition network camera was also installed 1.5 meters directly above the plants. Starting from autumn planting, data was collected continuously for 90 days, recording 24-hour environmental data and plant images at noon each day, forming a dataset containing a 90-day temperature sequence and 90 images of the strawberry canopy.

[0043] Step S2: Based on environmental time series data and phenotypic image time series data, identify and label multiple growth stage transition nodes, including germination period, vegetative growth period, flower bud differentiation period, and fruiting period.

[0044] This step is the automatic identification stage of plant growth stages, which transforms raw data into biologically meaningful stage labels. Growth stage transition nodes refer to the critical time points when a plant transitions from its current growth stage to the next, such as the moment of transition from vegetative growth to reproductive growth. Germination refers to the stage where seeds absorb water and swell, and the radicle breaks through the seed coat; vegetative growth refers to the stage where the plant's growth is primarily focused on vegetative organs such as roots, stems, and leaves; flower bud differentiation refers to the stage where the shoot apex meristem transforms from a vegetative growth cone to a reproductive growth cone and begins to form flower primordia; fruiting is the entire stage in the plant's growth and development cycle, from flowering and pollination to fruit ripening, falling off, or harvesting. The principle behind this step is that plant growth stage transitions are driven by environmental factors (such as accumulated temperature and photoperiod) and manifested through external morphological characteristics (such as leaf number, plant height, and budding). Therefore, by analyzing the cumulative effects and rates of change in environmental time-series data, combined with changes in morphological features in image data, multimodal fusion can accurately determine stage transition points.

[0045] like Figure 3 As shown, the process of identifying reproductive nodes includes an environmental data analysis path and an image analysis path. The features of the two paths are fused and input into a hidden Markov model. The Viterbi algorithm is used to decode the most likely stage sequence, and finally the reproductive stage transition node is marked.

[0046] As a preferred implementation, a Hidden Markov Model (HMM) can be used for stage identification. An HMM is a statistical model used to process time-series data. It assumes that the system has a hidden state sequence that cannot be directly observed (in this case, the plant's growth stage), and that this hidden state emits observable values ​​(in this case, environmental observation vectors) with a certain probability. The principle behind this step is that the plant's actual growth stage (such as whether it is in flower bud differentiation) cannot be directly measured, but we can observe the related environmental data. The HMM models this relationship through two core parameters: the state transition probability, i.e., the probability of transitioning from one growth stage to the next; and the observation probability, i.e., the probability of observing a specific environmental vector value at a certain growth stage. Based on the observed environmental time series, the Viterbi algorithm can be used to dynamically solve for the hidden state sequence most likely to produce the observed sequence, i.e., to obtain the most likely growth stage for each day.

[0047] In this embodiment, a Hidden Markov Model (HMM) is constructed, with the hidden state set consisting of the vegetative growth stage, flower bud differentiation stage, flowering stage, and fruiting stage. Based on agronomical knowledge, the state transition probabilities are set as follows: 0.01 probability daily for the vegetative growth stage to transition to the flower bud differentiation stage, 0.05 probability daily for the flower bud differentiation stage to transition to the flowering stage, and 0.05 probability daily for the flowering stage to transition to the fruiting stage. The observation probabilities are learned from historical data. Inputting the daily environmental observation vectors for 90 days, the Viterbi algorithm outputs the most probable state for each day. Analysis of the soil moisture data from the first 5 days reveals that on day 3, soil moisture decreased by 10% from saturation. Simultaneously, the emergence of the radicle was detected in images from days 3-5 (using image recognition algorithms), and combined with accumulated temperature calculations reaching 30℃·d, the system determines day 3 as the germination node. On day 45, the day length shortens to less than 12 hours, and a small protrusion at the stem tip is detected in the image; the system determines day 45 as the flower bud differentiation node.

[0048] Furthermore, the observation probability distribution of the Hidden Markov Model (HMM) can be fitted using a Gaussian mixture model (GaJM) to more accurately describe the various typical environmental patterns that may occur within the same stage. A Gaussian mixture model uses a weighted sum of multiple Gaussian distributions to fit a probability density function of arbitrary shape, as shown in the formula: In the formula: Hidden state Environmental vectors observed during the reproductive stage The probability of; The number of Gaussian components in the mixture; For the first The weights of each Gaussian component; It follows a multivariate normal distribution; For the first The mean vector of Gaussian components; For the first The covariance matrix of Gaussian components, with parameters (weights) mean Covariance The accumulated temperature model is automatically learned from labeled historical data using an expectation-maximization algorithm. Simultaneously, the state transition probability matrix of the Hidden Markov Model can be dynamically adjusted based on the accumulated temperature model, incorporating biological mechanisms. The calculation formula for the accumulated temperature model is as follows: ,in For air temperature, The baseline temperature is used. As the GDD value gradually approaches the accumulated temperature threshold required for the next stage, the system dynamically increases the probability of transitioning to the next stage. In this embodiment, for the hidden state of the vegetative growth period, three Gaussian distributions are mixed: one corresponding to low-temperature, rainy weather (mean 10℃), one corresponding to warm, sunny weather (mean 18℃), and one corresponding to an intermediate state. The parameters of these three distributions are learned from 90 days of data using the expectation-maximization algorithm. Simultaneously, the biological zero temperature for strawberries is set. The accumulated temperature is 5℃, when the accumulated temperature is... As the temperature gradually increases from 0 to 200℃·d, the model linearly increases the probability of transitioning from the vegetative growth stage to the flower bud differentiation stage each day from 0.01 to 0.1, in order to simulate the biological laws of stage transition driven by accumulated temperature.

[0049] Step S3: Based on the identified reproductive stage transition nodes, establish a coupling relationship model between environmental parameters and reproductive stages.

[0050] This step is the environment-stage coupling modeling stage, which uses a mathematical model to describe what environmental processes lead to what stage evolution. A coupling relationship model refers to a mathematical model that can characterize the complex nonlinear relationship between multidimensional environmental factors and growth stages. The principle of this step is that the evolution of plant growth stages is not determined by a single environmental factor, but rather by the combined effects of the coupling of multiple environmental factors such as temperature, humidity, and light. Therefore, it is necessary to construct a multi-input, single-output dynamic model, taking the environmental time series as input and the current growth stage as output. This model is essentially a digital encapsulation of the environment-life history experienced by plants in their natural environment.

[0051] In this embodiment, the daily environmental vectors (air temperature, soil moisture, and light intensity) for the first 90 days are used as input features, and the stage labels marked in step S2 (the first 45 days are the vegetative growth period, and the last 45 days are the flower bud differentiation period and thereafter) are used as output labels to train a long short-term memory neural network. This network learns the implicit rule that when the accumulated light integral reaches X and the day length is less than Y, the flower bud differentiation period begins.

[0052] Step S4: Based on the preset control target, reconstruct the time axis of the environmental time series data to generate the target environmental control trajectory.

[0053] This step is the core innovation—the timeline reconstruction stage—which digitally edits the natural flow of time according to production needs. The preset control target refers to the production objective the user hopes to achieve, such as flowering 15 days earlier or increasing fruit yield by 20%. Timeline reconstruction processing involves performing a non-linear scaling transformation on the original environmental time series along the time dimension. The principle behind this step is that plant development rate is highly coupled with the rhythm of environmental change; if the rhythm of environmental change is artificially accelerated, the plant will correspondingly accelerate its development. Based on this principle, this invention introduces a time transformation function to mathematically deform the original environmental trajectory.

[0054] like Figure 4 As shown, the complete process of time axis reconstruction includes: performing Fourier transform on the original environmental trajectories of each reproductive stage, scaling (with optional filtering) in the frequency domain, obtaining the reconstructed trajectory after inverse transform, and then splicing and smoothing to finally generate the target environmental regulation trajectory.

[0055] As a preferred implementation method, the time axis reconstruction can be achieved using the frequency domain reconstruction method. Specifically, this includes performing a Fourier transform on the environmental time series data within each reproductive stage to obtain the environmental spectrum for each reproductive stage.

[0056] In the formula: For the first Spectral functions of environmental parameters at each reproductive stage; , The first The beginning and end times of each reproductive stage; For the first Environmental parameter vector function for each reproductive stage; Angular frequency; The imaginary unit is used; the time axis reconstruction coefficient for each reproductive stage is determined based on the preset control targets. And perform frequency domain scaling on the environmental spectrum. Then through inverse Fourier transform (In the formula: For the reconstructed first The reconstructed environmental trajectories for each reproductive stage are generated by using a vector function of environmental parameters for each reproductive stage. The reconstructed environmental trajectories for each reproductive stage are then spliced ​​together in chronological order to obtain the target environmental regulation trajectory.

[0057] The principle behind this step is based on the scaling property of the Fourier transform: compressing a signal in the time domain (shortening the time axis) is equivalent to expanding the spectrum in the frequency domain, and vice versa. Compared to direct interpolation compression in the time domain, the advantage of the frequency domain method is that it can preserve the waveform characteristics and spectral structure of the original environmental signal, avoiding distortions that may be introduced by time domain interpolation. It is especially suitable for environmental signal processing that includes periodic fluctuations (such as diurnal temperature range).

[0058] In this embodiment, in the original 90-day environmental trajectory, the temperature during the first 45 days of the vegetative growth period gradually increased from 10°C to 18°C, and included significant diurnal fluctuations with a 24-hour cycle. The user desired earlier flowering, therefore, a vegetative growth period compression coefficient α1 = 1.5 was set, compressing the environmental changes of the first 45 days into 30 days (i.e., accelerating the temperature rise by 1.5 times). The original vegetative growth period temperature data... Perform a Fourier transform to obtain the spectrum. ,in There is a peak at that point. Scaling the spectrum yields... This causes the frequency peak corresponding to the original 24-hour cycle to shift to At this point, the new cycle is 16 hours. The inverse transformation yields 30 days of temperature data. Its diurnal fluctuations accelerated, but the waveform shape remained intact. This, combined with the trajectory of the flower bud differentiation period over the following 45 days, forms a 75-day target environmental regulation trajectory.

[0059] Furthermore, time axis reconstruction coefficients This can be determined using Bayesian optimization. Bayesian optimization is a black-box optimization method suitable for situations where the objective function expression is unknown and the evaluation cost is high. Its principle is: first, a Gaussian process surrogate model is constructed to approximate the unknown objective function; this model can provide arbitrary... The predicted mean and uncertainty of the objective function value are combined. Then, a data collection function is defined to balance exploration and utilization, recommending the next most promising one. The surrogate model is evaluated using a combination of factors. As the number of evaluations increases, the surrogate model becomes increasingly accurate, eventually converging to the global optimum. In this embodiment, the user aims to maximize yield, with the decision variables being the compression coefficient α1 during the vegetative growth stage and the compression coefficient α2 during flower bud differentiation, both ranging from [0.5, 2.0]. Bayesian optimization first randomly samples five groups (α1, α2) for a real planting experiment, recording the yield. Based on these five data points, a Gaussian process model is constructed, a sampling function recommends the next set of points to try, and the yield is evaluated again, updating the model. After 20 iterations, the model outputs the optimal coefficient combination.

[0060] Furthermore, frequency domain scaling also includes independent manipulation of specific frequency bands: the ambient spectrum. Apply frequency domain filter By preserving or enhancing specific frequency components, a filtered spectrum is obtained. The principle behind this step is that plants respond differently to changes in different frequencies in the environment. This is achieved by applying a filter to the reconstructed spectrum along the time axis in the frequency domain. This allows for independent, decoupled control of the rhythm and amplitude / characteristics of temperature changes. In this embodiment, to promote strawberry flower bud differentiation, it is desirable to enhance the diurnal temperature range stimulus; therefore, a bandpass filter is designed. This resulted in a gain of 1.5 near the 16-hour cycle corresponding to the frequency, and a gain of 1 at other frequencies. After applying the filter, the diurnal temperature difference of the environmental trajectory obtained by the inverse transform was larger, thereby accelerating growth and enhancing the temperature difference-induced effect.

[0061] Furthermore, after splicing the reconstructed environmental trajectories for each growth stage in chronological order, a wavelet threshold-based denoising method is used to smooth the splicing points. The wavelet threshold-based denoising method is a signal smoothing technique that utilizes wavelet transform. By performing wavelet decomposition on the discontinuities at the splicing points and filtering out high-frequency noise, a continuous and smooth target environmental regulation trajectory is obtained. In this embodiment, the temperature at the end of day 30 of the reconstructed 30-day vegetative period trajectory is 18℃, while the temperature at the beginning of day 31 of the flower bud differentiation period trajectory is 12℃, resulting in a 6℃ jump at the splicing point. Wavelet decomposition is performed on the signals from days 25 to 35 (11 days total), with a threshold set to 0.5℃. High-frequency detail coefficients below 0.5℃ are set to zero, and then the signal is reconstructed to obtain a smooth transition temperature curve for days 30-31.

[0062] Step S5: In a controllable indoor environment, a closed-loop control algorithm is used to dynamically adjust the actuator so that the actual indoor environmental parameters can track and reproduce the target environment control trajectory.

[0063] This step is the precision execution stage, which aims to play the edited natural film in a high-fidelity environment within an artificial setting. The indoor controllable environment refers to a plant factory, artificial climate chamber, or intelligent greenhouse, equipped with adjustable actuators. A closed-loop control algorithm is an algorithm that can adjust the control quantity in real time based on the deviation between the actual output and the target value. Actuators include heating / cooling equipment, humidification / dehumidification equipment, LED supplemental lighting, irrigation systems, etc. The principle of this step is that the target environment control trajectory is a setpoint sequence that changes over time. Indoor environments exhibit large inertia and large lag; the closed-loop control algorithm overcomes these dynamic characteristics by predicting future states and acting in advance, achieving high-precision trajectory tracking.

[0064] The indoor control closed-loop structure includes a target setting layer, a control layer, an execution layer, and a monitoring layer. For example... Figure 5As shown, the control layer can employ model predictive control or iterative learning control. Iterative learning control utilizes PD-type learning laws and stored historical errors to optimize the control input cycle by cycle, enabling the actual environment to gradually approach the target trajectory.

[0065] As a preferred implementation, an iterative learning control algorithm can be employed. Iterative learning control is a control method specifically designed for systems performing repetitive tasks. Its core idea is to use the deviation between the actual environmental trajectory and the target environmental control trajectory from the previous growth cycle to correct the control input for the current growth cycle, allowing the actual indoor environmental parameters to successively approximate and reproduce the target environmental control trajectory. In this embodiment, the first planting batch attempts to reproduce the target temperature trajectory over 75 days, but due to the large inertia of the air conditioning system, the actual temperature always lags behind the target temperature. The tracking error for the entire cycle is recorded. At the start of the second batch, the iterative learning controller corrected the air conditioner's control strategy based on the error signal, for example, increasing the heating power one hour in advance each time a temperature increase was needed. The deviation between the actual temperature trajectory and the target trajectory in the second batch was significantly reduced. After 3-4 iterations, the actual trajectory almost perfectly overlapped with the target trajectory.

[0066] Furthermore, the iterative learning control algorithm employs a PD-type learning law. ,in This is the control input for the kth growth cycle. The deviation between the actual environmental trajectory and the target environmental control trajectory. and The learning gain matrix is ​​defined. The PD-type learning law utilizes both the current value and future trend of the error to correct the control input, resulting in faster convergence and better transient performance. In this embodiment, the control input is defined as follows: This represents the percentage of heating power of the air conditioner, with an error margin. This is the difference between the target temperature and the actual temperature. (Setting) , At some point on day 10 of the second batch, the tracking error... (Actual value is lower), and the rate of change of error (The error is increasing). According to the learning law, the heating power correction for the second batch at this moment is: This means increasing the heating power by 27% based on the heating power of the first batch, in order to suppress further expansion of the error in advance.

[0067] In addition, this embodiment provides a plant growth cycle regulation system based on natural environment trajectory reconstruction, which is used to implement the method of the above embodiment.

[0068] like Figure 1 As shown, the system includes:

[0069] Natural habitat learning unit 100 is deployed in the natural habitat of plants to continuously collect multidimensional environmental data and phenotypic image data, and generate environmental time series data and phenotypic image time series data.

[0070] The reproductive stage identification unit 200 is connected to the natural habitat learning unit 100 and is used to receive environmental time series data and phenotypic image time series data, and to identify and mark multiple reproductive stage transition nodes;

[0071] The coupling modeling unit 300 is connected to the reproductive stage identification unit 200 and is used to establish a coupling relationship model between environmental parameters and reproductive stages based on the identified reproductive stage transition nodes.

[0072] The time axis reconstruction unit 400 is connected to the coupled modeling unit 300 and is used to reconstruct the environmental time series data according to the preset control target to generate the target environmental control trajectory.

[0073] The indoor closed-loop control unit 500 is deployed in the indoor controllable environment and connected to the time axis reconstruction unit 400. It is used to receive the target environment control trajectory and dynamically adjust the actuator using a closed-loop control algorithm so that the actual indoor environmental parameters can track and reproduce the target environment control trajectory.

[0074] The connections between the units (including wired or wireless communication) form a complete pathway for data and instruction flow, enabling the entire process from natural learning to accurate indoor reproduction.

[0075] In a preferred embodiment, the time axis reconstruction unit 400 may include a frequency domain filter bank for applying independent gain coefficients to different frequency bands of the environmental spectrum to achieve personalized control of environmental fluctuation characteristics.

[0076] In a preferred embodiment, the system may further include a transfer learning module, which is connected to the physiological driving force analysis unit, the reproductive node identification unit, and the multivariate coupling modeling unit, respectively. The transfer learning module is used to store the natural habitat model parameters of multiple species, and when applied to a new species, it fine-tunes the multivariate coupling dynamic model of the existing species based on the short-term observation data of the new species to generate a target regulatory trajectory suitable for the new species.

[0077] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for regulating plant growth cycle based on natural environmental trajectory reconstruction, characterized in that, Includes the following steps: S1. Deploy sensor arrays and image acquisition devices in the natural habitat of plants to continuously collect multidimensional environmental data and phenotypic image data, and generate environmental time series data and phenotypic image time series data. S2. Based on the environmental time series data and the phenotypic image time series data, identify and mark multiple growth stage transition nodes, including the germination period, vegetative growth period, flower bud differentiation period and fruiting period. S3. Based on the identified reproductive stage transition nodes, establish a coupling relationship model between environmental parameters and reproductive stages; S4. Based on the preset control target, the environmental time series data is reconstructed on a time axis to generate a target environmental control trajectory. S5. In a controllable indoor environment, a closed-loop control algorithm is used to dynamically adjust the actuator so that the actual indoor environmental parameters track and reproduce the target environment control trajectory.

2. The plant growth cycle regulation method based on natural environment trajectory reconstruction according to claim 1, characterized in that, The identification and marking of reproductive stage transition nodes in step S2 specifically includes: using a Hidden Markov Model (HMM) for stage identification, wherein the HMM takes the environmental observation vector as input and the reproductive stage as the hidden state, and decodes it using the Viterbi algorithm to obtain the most likely stage sequence.

3. The plant growth cycle regulation method based on natural environment trajectory reconstruction according to claim 2, characterized in that, The observation probability distribution of the Hidden Markov Model (HMM) is fitted using a Gaussian Mixture Model (GaM), the parameters of which are obtained by training on historical data using the Expectation-Maximization (EM) algorithm. The state transition probability matrix of the HMM is dynamically adjusted according to an accumulated temperature model, the calculation formula of which is as follows: ,in, for The effective accumulated temperature value at any given time; For integration time variable, for Real-time air temperature at any given moment This is the reference temperature.

4. The plant growth cycle regulation method based on natural environment trajectory reconstruction according to claim 1, characterized in that, Step S4, which involves reconstructing the environmental time series data along a time axis, specifically includes performing a Fourier transform on the environmental time series data for each reproductive stage to obtain the environmental spectrum for each reproductive stage. in Let be the environmental parameter vector function for the i-th reproductive stage; determine the time axis reconstruction coefficients for each reproductive stage based on the preset control targets. And perform frequency domain scaling on the environmental spectrum. Then through inverse Fourier transform Generate reconstructed environmental trajectories for each reproductive stage; then, stitch together the reconstructed environmental trajectories for each reproductive stage in chronological order to obtain the target environmental regulation trajectory.

5. The plant growth cycle regulation method based on natural environment trajectory reconstruction according to claim 4, characterized in that, The time axis reconstruction coefficient The Bayesian optimization algorithm is used to determine that the Bayesian optimization takes yield or growth cycle as the objective function and the reconstruction coefficients of each stage as hyperparameters, and iteratively searches for the optimal combination of reconstruction coefficients based on a Gaussian process surrogate model.

6. The plant growth cycle regulation method based on natural environment trajectory reconstruction according to claim 4, characterized in that, The frequency domain scaling process also includes independent control of specific frequency bands: the environmental spectrum Apply frequency domain filter By preserving or enhancing specific frequency components, a filtered spectrum is obtained. .

7. The plant growth cycle regulation method based on natural environment trajectory reconstruction according to claim 4, characterized in that, After splicing the reconstructed environmental trajectories of each reproductive stage in chronological order, the method further includes using a wavelet threshold-based denoising method to smooth the splicing points. By performing wavelet decomposition on the discontinuous points at the splicing points and filtering out high-frequency noise, a continuous and smooth target environmental regulation trajectory is obtained.

8. The plant growth cycle regulation method based on natural environment trajectory reconstruction according to claim 1, characterized in that, The closed-loop control algorithm in step S5 adopts an iterative learning control algorithm. The iterative learning control algorithm uses the deviation between the actual environmental trajectory of the previous growth cycle and the target environmental regulation trajectory to correct the control input of the current growth cycle, so that the actual indoor environmental parameters successively approach and reproduce the target environmental regulation trajectory.

9. The plant growth cycle regulation method based on natural environment trajectory reconstruction according to claim 8, characterized in that, The iterative learning control algorithm employs a PD-type learning law. ,in This is the control input for the kth growth cycle. The deviation between the actual environmental trajectory and the target environmental control trajectory. and To learn the gain matrix.

10. A plant growth cycle regulation system based on natural environmental trajectory reconstruction, characterized in that, The plant growth cycle regulation method based on natural environment trajectory reconstruction, applied to any one of claims 1-9, comprises: Natural habitat learning unit (100) is set up in the natural habitat of plants to continuously collect multidimensional environmental data and phenotypic image data, and generate environmental time series data and phenotypic image time series data. A reproductive stage identification unit (200) is connected to the natural habitat learning unit and is used to receive the environmental time series data and the phenotypic image time series data, and to identify and mark multiple reproductive stage transition nodes; A coupling modeling unit (300) is connected to the reproductive stage identification unit and is used to establish a coupling relationship model between environmental parameters and reproductive stages based on the identified reproductive stage transition nodes; A time axis reconstruction unit (400) is connected to the coupled modeling unit and is used to perform time axis reconstruction processing on the environmental time series data according to the preset control target to generate the target environmental control trajectory. An indoor closed-loop control unit (500) is deployed in the indoor controllable environment and connected to the time axis reconstruction unit. It is used to receive the target environment control trajectory and dynamically adjust the actuator using a closed-loop control algorithm so that the actual indoor environmental parameters track and reproduce the target environment control trajectory.

Images