Plant growth full-period dynamic cooperative regulation and control method and system based on Internet of Things

Abstract

The invention discloses a plant growth full-cycle dynamic cooperative regulation and control method and system based on the Internet of Things, and relates to the technical field of intelligent agriculture. The technical scheme is characterized in that dynamic coupling modeling, multi-protocol redundant communication, incremental learning closed loop and constrained optimization are carried out; according to the method, the defects of environment regulation lag, poor model generalization, low communication reliability and lack of adaptive ability in a traditional method are overcome, the method has remarkable advantages in the aspects of yield improvement, resource saving and decision interpretability, and a complete technical closed loop from data perception to intelligent control is provided for intelligent agriculture.

Description

Technical Field

[0001] The present invention relates to the technical field of smart agriculture, and more specifically, to a method and system for dynamically and collaboratively regulating the entire growth cycle of plants based on the Internet of Things. Background Art

[0002] The purpose of regulating environmental parameters during the plant growth process is to achieve precise control of its growth state, physiological metabolism, and yield and quality by actively adjusting environmental factors closely related to plant growth, such as temperature, humidity, light, CO2 concentration, soil conditions, etc.

[0003] The existing plant growth regulation methods have the following defects: (1) Static data limitation: relying on environmental parameters (such as temperature and humidity) at a single time point for fixed threshold control, ignoring the dynamic response of plant physiological characteristics; (2) Fragmented analysis: insufficient modeling of the correlation between environmental parameters and plant growth characteristics, without considering the interaction of multiple variables and the differences in growth stages; (3) Regulation lag: the generation of control instructions depends on manual experience or simple feedback, lacking a forward-looking optimization mechanism.

[0004] Therefore, how to research and design a method and system for dynamically and collaboratively regulating the entire growth cycle of plants based on the Internet of Things that can overcome the above defects is an urgent problem for us to solve currently. Summary of the Invention

[0005] To solve the deficiencies in the prior art, the purpose of the present invention is to provide a method and system for dynamically and collaboratively regulating the entire growth cycle of plants based on the Internet of Things. By means of dynamic coupling modeling, multi-protocol redundant communication, incremental learning closed-loop, and constrained optimization, it solves the defects of environmental regulation lag, poor model generalization, low communication reliability, and lack of adaptability in traditional methods. It has significant advantages in terms of yield improvement, resource conservation, and decision interpretability, providing a complete technical closed-loop from data perception to intelligent control for smart agriculture.

[0006] The above technical purpose of the present invention is achieved through the following technical solutions:

[0007] In the first aspect, a method for dynamically and collaboratively regulating the entire growth cycle of plants based on the Internet of Things is provided, including the following steps:

[0008] Extract the underlying features of color, shape, texture, and growth rate from plant images at adjacent times to generate a non-zero sequence of feature variables;

[0009] Perform exponential decay weighted average processing on the environmental data within the time window corresponding to the adjacent times to generate a sequence of environmental variables;

[0010] Combine the characteristic variables and environmental variables into discrete point data, and fit the coupling relationship through a bivariate Bayesian ridge regression model containing cross terms to obtain a coupling model;

[0011] Load the positive limit value and environmental parameter boundary according to the current growth stage of the plant, incorporate them into the coupling model, and use the interior point method to solve the approaching point optimization problem to obtain environmental reference variables;

[0012] Convert the environmental reference variables into multi-protocol control instructions and adaptively send them to the execution device based on the link quality;

[0013] Collect the regulation feedback data and trigger the incremental learning mechanism to update the coupling model to form a closed-loop control.

[0014] Further, the specific calculation formula for the exponentially decaying weighted average processing is as follows:

[0015]

[0016] Among them, ΔE j represents the weighted average change of the j-th environmental parameter within the time window; m represents the number of sampling times within the time window; represents the instantaneous measurement value of the j-th environmental parameter at time point k; represents the instantaneous measurement value of the j-th environmental parameter at time point k - 1; λ represents the decay factor, which is negatively correlated with the growth rate of the plant species; t k represents the time interval from time point k to the current moment.

[0017] Further, the expression of the bivariate Bayesian ridge regression model is as follows:

[0018] Δf i = β0 + β1ΔE j + β2ΔE j 2 + β3ΔE j ΔE q + ε;

[0019] Among them, Δf i represents the i-th characteristic variable; β0 is the intercept term, representing the baseline characteristic change when the environmental variable is zero; β1 is the first-order effect coefficient of the environmental parameter, reflecting the linear influence of a single environmental factor; ΔE j represents the weighted average change of the j-th environmental parameter within the time window; β2 is the quadratic term coefficient of the environmental parameter, capturing the non-linear effect; β3 represents the cross-term coefficient, quantifying the synergistic or antagonistic effect of multiple environmental parameters; ΔE q represents the weighted average change of the q-th environmental parameter within the time window; ε represents the error term, which follows a normal distribution.

[0020] Furthermore, the current growth stage of the plant is achieved in any of the following ways:

[0021] Based on the time series feature classification of the LSTM network, the input is the environmental data and image features of the time series, and the output is the stage probability;

[0022] By analyzing images or monitoring key morphological and / or physiological indicators with sensors, trigger stage switching when the preset threshold is reached;

[0023] Combined with the preset time window and feature verification, perform fault tolerance determination through a state machine.

[0024] Furthermore, the objective function of the approach point optimization problem is specifically:

[0025]

[0026] Among them, g(ΔE) represents the fitting function, describing the mapping relationship between the change ΔE of the environmental variable and the change Δf of the feature variable; Δf lim represents the positive limit value of the feature variable in the current growth stage; represents the square of the L2 norm; η represents the regularization coefficient; ||·|| represents the L1 norm.

[0027] Furthermore, the generation of the multi - protocol control instruction includes:

[0028] Map the environmental reference variable to the device absolute threshold or incremental instruction;

[0029] Encapsulate the device absolute threshold or incremental instruction into an MQTT message in JSON format and a Modbus RTU binary frame;

[0030] Add a timestamp and / or priority mark to the MQTT message and the Modbus RTU.

[0031] Furthermore, the multi - protocol control instruction is sent using dual - link redundant transmission.

[0032] Furthermore, the main link in the dual - link uses LoRaWAN, with a spreading factor of 10 and a bandwidth of 125 kHz;

[0033] And, the backup link in the dual - link uses 4G DTU and automatically switches when the signal strength of the main link is less than - 110 dBm.

[0034] Furthermore, the execution process of the incremental learning mechanism includes:

[0035] Regression coefficient correction based on online Bayesian update;

[0036] If the model prediction error continuously exceeds the set percentage, start the retraining of the entire model;

[0037] Moreover, the updated model parameters are stored on the blockchain to ensure immutability.

[0038] In a second aspect, a dynamic collaborative regulation system for the whole growth cycle of plants based on the Internet of Things is provided. This system is used to implement the dynamic collaborative regulation method for the whole growth cycle of plants based on the Internet of Things described in any one of the first aspects, including:

[0039] A feature analysis module, which is used to extract the underlying features of color, shape, texture, and growth rate from plant images at adjacent times, and generate a non-zero sequence of feature variables;

[0040] An environment analysis module, which is used to perform exponentially weighted moving average processing on the environmental data within the time window corresponding to the adjacent times, and generate a sequence of environmental variables;

[0041] A variable coupling module, which is used to combine the feature variables and the environmental variables into discrete point data, and fit the coupling relationship through a bivariate Bayesian ridge regression model including cross terms to obtain a coupling model;

[0042] An optimization solving module, which is used to load the positive limit value and the environmental parameter boundary according to the current growth stage of the plant, incorporate them into the coupling model, and use the interior point method to solve the approaching point optimization problem to obtain the environmental reference variable;

[0043] An instruction control module, which is used to convert the environmental reference variable into multi-protocol control instructions and adaptively send them to the execution device based on the link quality;

[0044] A feedback control module, which is used to collect the regulation feedback data and trigger an incremental learning mechanism to update the coupling model to form a closed-loop control.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The dynamic collaborative regulation method for the whole growth cycle of plants based on the Internet of Things provided by the present invention solves the defects of lagging environmental regulation, poor model generalization, low communication reliability, and lack of adaptability in traditional methods through dynamic coupling modeling, multi-protocol redundant communication, incremental learning closed-loop, and constrained optimization. It has significant advantages in yield improvement, resource conservation, and decision interpretability, providing a complete technical closed-loop from data perception to intelligent control for smart agriculture;

[0047] 2. The present invention quantifies the regulation intensity of the environment on the feature variables through cross terms and Bayesian inference. Compared with traditional multiple linear regression, the yield prediction error is lower and the interpretability is enhanced;

[0048] 3. The present invention responds to changes in the plant growth stage and environmental fluctuations in real time through an exponential decay weighting and incremental learning mechanism;

[0049] 4. The present invention adopts dual-link redundant communication and blockchain evidence storage to ensure stable instruction transmission and model security;

[0050] 5. The present invention uses a lightweight MT-CNN network to achieve synchronous extraction of multiple features and is adapted to edge computing devices. Description of the Drawings

[0051] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0052] Figure 1 is the flowchart in Embodiment 1 of the present invention;

[0053] Figure 2 is the system block diagram in Embodiment 2 of the present invention. Detailed Embodiments

[0054] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the embodiments and the drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and do not constitute a limitation to the present invention.

[0055] Embodiment 1: A method for dynamic collaborative regulation of the entire plant growth cycle based on the Internet of Things, as Figure 1 shown, includes the following steps:

[0056] S1: Extract the underlying features of color, shape, texture, and growth rate from plant images at adjacent times to generate a non-zero sequence of feature variables;

[0057] S2: Perform exponential decay weighted average processing on environmental data within the time window corresponding to adjacent times to generate an environmental variable sequence;

[0058] S3: Combine the feature variables and environmental variables into discrete point data, and fit the coupling relationship through a bivariate Bayesian ridge regression model containing cross terms to obtain a coupling model;

[0059] S4: Load the positive limit value and environmental parameter boundary according to the current plant growth stage, incorporate them into the coupling model, and use the interior point method to solve the approach point optimization problem to obtain an environmental reference variable;

[0060] S5: Convert the environmental reference variable into a multi-protocol control instruction and adaptively send it to the execution device based on the link quality;

[0061] S6: Collect the regulation feedback data, trigger the incremental learning mechanism to update the coupling model, and form a closed-loop control.

[0062] In step S1, features such as color, shape, texture, and growth rate can be extracted by different methods.

[0063] In some examples, the color feature can be extracted using a color histogram or HSV space analysis. In some examples, the shape feature can be extracted using edge detection or contour analysis. In some examples, the texture can be extracted using a gray-level co-occurrence matrix (GLCM) or local binary pattern (LBP). In some examples, the growth rate can be extracted using an optical flow method or temporal difference.

[0064] In addition, considering that the method provided by the present invention runs on an Internet of Things device or an edge computing device, to reduce the computational complexity and enhance the real-time performance, the present invention designs a lightweight multi-task convolutional neural network (MT-CNN) to synchronously extract multi-dimensional features of plant images.

[0065] MT-CNN extracts the basic image features through a shared backbone network, then extracts the color, shape, texture, and growth rate features through independent branches respectively, and finally fuses them to generate a non-zero feature variable sequence. This method completes all feature extractions in a single forward propagation, ensuring high efficiency and consistency.

[0066] MT-CNN consists of an input layer, a shared backbone network, multi-task branches, a feature fusion layer, and a non-zero filtering layer.

[0067] In some examples, the input layer mainly inputs RGB image pairs, which include the first image at the first moment and the second image at the second moment.

[0068] In some examples, the shared backbone network consists of an initial convolutional layer, an inverted residual block, and a global average pooling; the convolutional kernel size of the initial convolutional layer is 3×3, and the stride is 2; the inverted residual block mainly realizes channel expansion, depthwise separable convolution, and compression functions; the global average pooling mainly outputs the feature map.

[0069] In some examples, the multi-task branches include a color branch, a shape branch, a texture branch, and a growth rate branch.

[0070] The color branch performs HSV histogram and color moment processing; for example, convert the RGB image to the HSV space, segment the H (hue) and S (saturation) channels; use bilinear interpolation to calculate the 10-bin histogram, generate a 20-dimensional vector (10 dimensions for each of H+S); determine the color moments: the first moment (mean), the second moment (variance), the third moment (skewness), a total of 3 dimensions; finally output: a 23-dimensional color feature vector.

[0071] The shape branch mainly realizes edge detection and geometric parameter calculation. For example, for edge calculation: input the feature map of the backbone network, generate an edge activation map through 1×1 convolution; use the Sobel operator to enhance the contour response and output a binary edge mask; calculate geometric parameters, including area, compactness, and aspect ratio; output a 3D shape feature vector.

[0072] The texture branch mainly realizes LBP-GLCM fusion. For example: calculate the rotation-invariant local binary pattern and generate an 8D histogram; extract contrast, energy, and homogeneity, a total of 3D; splice the LBP histogram and GLCM parameters into an 11D vector; output an 11D texture feature vector.

[0073] The growth rate branch mainly calculates the dense optical flow field through the lightweight FlowNetS network; motion statistics, including the average displacement amplitude and direction consistency (generally calculate the variance of the optical flow direction and take the reciprocal as the stability index); output a 2D velocity feature vector.

[0074] The feature fusion layer splices the color (23D), shape (3D), texture (11D), and velocity (2D) features into a 39D vector.

[0075] The non-zero filtering layer performs threshold judgment on each feature value, uses L1 regularization to constrain the feature vector, and further compresses the values close to zero.

[0076] In step S2, the specific calculation formula for the exponentially decaying weighted average process is:

[0077]

[0078] where, ΔE j represents the weighted average change amount of the jth environmental parameter within the time window, with the unit being the unit of the corresponding environmental parameter, such as ℃ for temperature and %RH for humidity; m represents the number of samples within the time window. For example, if sampling is performed every 10 minutes, m takes the value of 144 for a 24-hour window; represents the instantaneous measurement value of the jth environmental parameter at time point k; represents the instantaneous measurement value of the jth environmental parameter at time point k - 1; λ represents the decay factor, which is negatively correlated with the growth rate of plant species. For example, for fast-growing plants, λ takes 0.1 min -1 , and for slow-growing plants, λ takes 0.05 min -1 ; t k represents the time interval from time point k to the current moment, which is used to calculate the exponential weight.

[0079] The present invention assigns higher weights to recent environmental changes through exponential weights to more accurately capture the dynamic impact on plant growth and avoid the interference of early data.

[0080] In step S3, each discrete point is composed of a combination of a feature variable and an environmental variable, representing the state of a certain plant under a specific environment. The environmental variable is any one or more of environmental parameters such as temperature, humidity, and light intensity. In addition, a single plant can generate multiple discrete points, allowing the model to capture the independent effects of different feature-environment combinations.

[0081] In some examples, the expression of the bivariate Bayesian ridge regression model is specifically:

[0082] Δf i = β0 + β1ΔE j + β2ΔE j 2 + β3ΔE j ΔE q + ε;

[0083] where, Δf i represents the i-th feature variable; β0 is the intercept term, representing the baseline feature change when the environmental variable is zero; β1 is the first-order effect coefficient of the environmental parameter, reflecting the linear influence of a single environmental factor; ΔE j represents the weighted average change of the j-th environmental parameter within the time window; β2 is the quadratic term coefficient of the environmental parameter, capturing the non-linear effect; β3 represents the cross-term coefficient, quantifying the synergistic or antagonistic effects of multiple environmental parameters; ΔE q represents the weighted average change of the q-th environmental parameter within the time window; ε represents the error term, following a normal distribution.

[0084] In step S4, the current growth stage of the plant is achieved in any of the following ways: temporal feature classification based on an LSTM network, with the input being time-series environmental data and image features, and the output being the stage probability; triggering stage switching when key morphological and / or physiological indicators reach a preset threshold through image analysis or sensor monitoring; combining a preset time window with feature verification and performing fault tolerance determination through a state machine.

[0085] In some examples, for time-series feature classification based on the LSTM network, environmental data and image features need to be obtained. Environmental data such as temperature, humidity, light intensity, CO2 concentration, etc. are collected at fixed time intervals (such as every hour); image features can be obtained by using an RGB or multispectral camera to capture plant images, and leaf area, color histogram (HSV), texture features (GLCM contrast, energy), etc. are extracted; then, the environmental data is synchronized with the image acquisition timestamp, with an error less than or equal to 1 minute. For example, when capturing images at 9:00 every day, the environmental data corresponding to 9:00 is the average value from 9:00 to 9:10; then, the environmental data is standardized, while the image features are normalized; finally, the sliding window length is defined (such as 7 days), and each window contains data for consecutive time steps.

[0086] In some examples, the U-Net model is used to segment leaf, flower, and fruit regions; the number of flower buds (morphological operations) and fruit volume (point cloud reconstruction) are calculated; if the feature values exceed the threshold for 3 consecutive days (such as the number of flower buds ≥ 10), it is determined to enter the next stage. In some examples, the threshold can also be adaptively updated according to the historical data sliding window to adapt to variety differences.

[0087] In some examples, feature conditions take precedence over time conditions. For example, even if the time has not arrived, if the fruit diameter meets the standard, it is forced to enter the fruiting stage. In some examples, the Isolation Forest algorithm can also be used to identify plants that deviate from the normal growth curve and mark them as "growth lag" or "early maturity".

[0088] The state transition logic is: current state → condition check → next state. For example, germination stage → time ≥ 7 days, or cotyledon unfolding rate ≥ 90% → seedling stage; seedling stage → time ≥ 30 days, and number of leaves ≥ 8 → flowering stage; flowering stage → time ≥ 60 days, or fruit diameter ≥ 1 cm → fruiting stage; exception handling: if the time exceeds the limit but the features do not meet the standard, an alarm is triggered and manual review is required.

[0089] The positive limit value refers to the maximum or minimum value that the characteristic variable of a plant is allowed to reach during a specific growth stage, which can prevent the characteristic variable from exceeding the physiological tolerance range of the plant; the environmental parameter boundary refers to the safe operating range allowed for environmental variables (such as temperature, humidity, light intensity, CO2 concentration, etc.) during the plant growth process. These boundaries ensure the safety, stability, and efficiency of the plant growth environment by restricting the output range of environmental control equipment.

[0090] In some examples, environmental parameter boundaries can be divided into static boundaries and dynamic boundaries according to their sources and dynamics. Static boundaries are generally determined by the basic tolerance of the plant and are usually fixed values; dynamic boundaries can be adjusted according to the plant growth stage, variety characteristics, or real-time feedback.

[0091] In some examples, the objective function of the approach point optimization problem is specifically:

[0092]

[0093] where g(ΔE) represents the fitting function, describing the mapping relationship between the change ΔE of the environmental variable and the change Δf of the characteristic variable; Δf lim represents the positive limit value of the characteristic variable in the current growth stage; represents the square of the L2 norm; η represents the regularization coefficient; ||·|| represents the L1 norm.

[0094] In step S5, the generation of the multi-protocol control instruction includes: mapping the environmental reference variable to the device absolute threshold or the increment instruction; encapsulating the device absolute threshold or the increment instruction into an MQTT message in JSON format and a Modbus RTU binary frame; adding a timestamp and / or a priority tag to the MQTT message and the shown Modbus RTU.

[0095] In some examples, the environmental reference variable mapping includes: determining the execution device corresponding to each environmental parameter and its control method. For example, the temperature reference variable ΔE_temperature = +3.8°C may need to be converted into the target temperature setting value of the thermostat or used as an increment to adjust the current temperature.

[0096] In some examples, generating device instructions: absolute threshold, such as setting the temperature to 25°C + 3.8°C = 28.8°C; increment instruction, such as increasing the temperature by 3.8°C; specific instruction formats need to be defined for different devices.

[0097] In some examples, encapsulating into an MQTT message requires designing a JSON structure, including the device ID, instruction type, value, timestamp, and priority.

[0098] In some examples, Modbus RTU uses a binary frame, including the address, function code, data, and CRC check.

[0099] In some examples, adding a timestamp includes: adding a timestamp in ISO 8601 format to the JSON of MQTT; in Modbus RTU, passing the timestamp through a register or relying on device clock synchronization.

[0100] In some examples, the priority tag (QoS level) includes: specifying the QoS level when publishing a message in MQTT; Modbus RTU processes the priority through the master station polling frequency or the message queue.

[0101] The present invention realizes precise regulation in the Internet of Things environment by mapping environmental reference variables into device instructions, encapsulating them into MQTT / Modbus protocol messages, and adding timestamps and priority tags.

[0102] In some examples, the multi - protocol control instructions are sent using dual - link redundant transmission. Specifically, the main link in the dual - link uses LoRaWAN with a spreading factor of 10 and a bandwidth of 125 kHz; and the backup link in the dual - link uses 4G DTU, which automatically switches when the signal strength of the main link is less than - 110 dBm.

[0103] In step S6, the execution process of the incremental learning mechanism includes: correcting the regression coefficient based on online Bayesian update; if the model prediction error continuously exceeds a set percentage (such as 5%), then initiate full - model retraining; and the updated model parameters are stored on the blockchain to ensure immutability.

[0104] The present invention realizes the closed - loop dynamic regulation of the plant growth environment by collecting multi - source feedback data, triggering incremental learning to update model parameters, and generating optimized control instructions. The present invention combines data - driven machine learning with the real - time control of physical systems, providing an extensible and highly reliable solution for smart agriculture.

[0105] The method for dynamically collaborative regulation of the entire plant growth cycle based on the Internet of Things recorded in the present invention can be applied to plants such as vegetables and fruit trees.

[0106] Example 2: A system for dynamically collaborative regulation of the entire plant growth cycle based on the Internet of Things, which is used to implement the method for dynamically collaborative regulation of the entire plant growth cycle based on the Internet of Things recorded in Example 1, as Figure 2 shown, includes a feature analysis module, an environment analysis module, a variable coupling module, an optimization solving module, an instruction control module, and a feedback control module.

[0107] Among them, the feature analysis module is used to extract the underlying features of color, shape, texture and growth rate from plant images at adjacent times, and generate a non-zero sequence of feature variables; the environment analysis module is used to perform exponentially weighted average processing on the environmental data within the time window corresponding to adjacent times to generate an environmental variable sequence; the variable coupling module is used to combine the feature variables and the environmental variables into discrete point data, and fit the coupling relationship through a bivariate Bayesian ridge regression model including cross terms to obtain a coupling model; the optimization and solution module is used to load the positive limit value and environmental parameter boundaries according to the current growth stage of the plant, incorporate them into the coupling model, and use the interior point method to solve the approach point optimization problem to obtain the environmental reference variable; the instruction control module is used to convert the environmental reference variable into a multi-protocol control instruction and adaptively send it to the execution device based on the link quality; the feedback control module is used to collect the regulation feedback data and trigger an incremental learning mechanism to update the coupling model to form a closed-loop control.

[0108] Working principle: The present invention solves the defects of environmental regulation lag, poor model generalization, low communication reliability and lack of adaptability in traditional methods through dynamic coupling modeling, multi-protocol redundancy communication, incremental learning closed-loop and constrained optimization. It has significant advantages in yield improvement, resource conservation and decision interpretability, providing a complete technical closed-loop from data perception to intelligent control for smart agriculture;

[0109] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0111] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in the function.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in the function.

[0113] The specific embodiments described above further elaborate on the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things, characterized in that: The following steps are involved: Extract the underlying features of color, shape, texture and growth rate from plant images at adjacent moments to generate a non-zero feature variable sequence; Performing exponential decay weighted averaging processing on the environmental data in the time window corresponding to the adjacent moments to generate an environmental variable sequence; The characteristic variables and environmental variables are combined into discrete point data, and the coupling relationship is fitted through a bivariate Bayesian ridge regression model including cross terms to obtain a coupling model; According to the current growth stage of the plant, the positive limit value and the environmental parameter boundary are loaded and integrated into the coupling model, and the approach point optimization problem is solved by the interior point method to obtain the environmental reference variable; Convert the environmental reference variables into multi-protocol control instructions, and adaptively send them to the execution device based on link quality; Collect control feedback data and trigger the incremental learning mechanism to update the coupling model to form a closed-loop control.

2. The method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things according to claim 1, characterized in that: The calculation formula of the exponential decay weighted average processing is specifically: Where, ΔE j represents the weighted average change of the jth environmental parameter in the time window; m represents the number of sampling times in the time window; represents the instantaneous measurement value of the jth environmental parameter at time point k; represents the instantaneous measurement value of the jth environmental parameter at time point k-1; λ represents the attenuation factor, which is negatively correlated with the growth rate of the plant species; t k Represents the time interval from time point k to the current moment.

3. The method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things according to claim 1, characterized in that: The expression of the bivariate Bayesian Ridge regression model is specifically: Δf i =β0+β1ΔE j +β2ΔE j 2 +β3ΔE j D.E. q +e; Where Δf i represents the i-th characteristic variable; β0 is the intercept term, which represents the baseline characteristic change when the environmental variable is zero; β1 is the first-order effect coefficient of the environmental parameter, which reflects the linear influence of a single environmental factor; ΔE j represents the weighted average change of the jth environmental parameter in the time window; β2 is the quadratic term coefficient of the environmental parameter, capturing the nonlinear effect; β3 represents the cross-term coefficient, quantifying the synergistic or antagonistic effect of multiple environmental parameters; ΔE q It represents the weighted average change of the qth environmental parameter in the time window; ε represents the error term, which obeys the normal distribution.

4. The method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things according to claim 1, characterized in that: The current growth stage of the plant is achieved by any of the following methods: Time series feature classification based on LSTM network, the input is time series environmental data and image features, and the output is stage probability; Monitor key morphological and / or physiological indicators through image analysis or sensors, triggering stage switching when preset thresholds are reached; Combined with the preset time window and feature verification, fault tolerance determination is performed through the state machine.

5. The method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things according to claim 1, characterized in that: The objective function of the approach point optimization problem is specifically: Where g(ΔE) represents the fitting function, which describes the mapping relationship between the change of environmental variable ΔE and the change of characteristic variable Δf; Δf lim Indicates the positive limit value of the characteristic variable at the current growth stage; represents the L2 norm squared; η represents the regularization coefficient; ||·|| represents the L1 norm.

6. The method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things according to claim 1, characterized in that: The generation of the multi-protocol control instruction includes: Mapping the environmental reference variable to a device absolute threshold or increment instruction; Encapsulating the device absolute threshold or increment instruction into an MQTT message in JSON format and a Modbus RTU binary frame; Add timestamps and / or priority tags in the MQTT messages and Modbus RTU.

7. The method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things according to claim 1, characterized in that: The sending of the multi-protocol control instructions adopts dual-link redundant transmission.

8. The method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things according to claim 7, characterized in that: The main link in the dual link adopts LoRaWAN, with a spreading factor of 10 and a bandwidth of 125kHz; Furthermore, the backup link in the dual link adopts 4G DTU and automatically switches when the signal strength of the main link is less than -110dBm.

9. The method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things according to claim 1, characterized in that: The execution process of the incremental learning mechanism includes: Regression coefficient correction based on online Bayesian updating; If the model prediction error continues to be greater than the set percentage, the full model retraining will be initiated; In addition, the updated model parameters are stored on the blockchain to ensure that they cannot be tampered with.

10. A plant growth full-cycle dynamic coordinated control system based on the Internet of Things, characterized in that: The system is used to implement the method for dynamic coordinated regulation of plant growth throughout the entire cycle based on the Internet of Things as described in any one of claims 1 to 9, comprising: The feature analysis module is used to extract the underlying features of color, shape, texture and growth rate from the plant images at adjacent moments and generate a non-zero feature variable sequence; An environmental analysis module, used to perform exponential decay weighted average processing on the environmental data in the time window corresponding to the adjacent moments to generate an environmental variable sequence; The variable coupling module is used to combine the characteristic variables and environmental variables into discrete point data, and fit the coupling relationship through a bivariate Bayesian ridge regression model containing cross terms to obtain a coupling model; An optimization solution module is used to load the positive limit value and the environmental parameter boundary according to the current growth stage of the plant, integrate them into the coupling model, and use the interior point method to solve the approach point optimization problem to obtain the environmental reference variable; An instruction control module, used for converting the environmental reference variables into multi-protocol control instructions, and adaptively sending them to the execution device based on the link quality; The feedback control module is used to collect control feedback data and trigger the incremental learning mechanism to update the coupling model to form a closed-loop control.

Images