Greenhouse intelligent dynamic coupling regulation and control method based on multi-modal data deep fusion

Abstract

The invention relates to a greenhouse intelligent dynamic coupling regulation and control method based on multi-modal data deep fusion, and the method comprises the steps: obtaining greenhouse environment and crop physiological data, mapping the greenhouse environment and crop physiological data to a multi-dimensional environment parameter vector space, and respectively constructing an environment supply manifold and a physiological adaptation manifold; calculating the overlapping degree to obtain an ecological niche coupling index; when the index is lower than a threshold value, predicting an evolution track and a physiological tolerance boundary along an environmental inertia direction by using a digital twin model, and calculating a safe buffer path length; and finally, generating a regulation and control instruction according to the length and the state recovery vector. According to the invention, the topological coincidence degree of environment supply and crop demand can be quantified, and accurate dynamic compensation regulation and control of the greenhouse environment can be realized based on the safe buffer space.

Description

Technical Field

[0001] This invention relates to the field of intelligent greenhouse control, and in particular to an intelligent dynamic coupling control method for greenhouses based on deep fusion of multimodal data. Background Technology

[0002] With the rapid development of facility agriculture technology, greenhouses have become an important carrier of modern agricultural production. Existing greenhouse environmental control systems typically use sensors to collect environmental parameters such as temperature, humidity, and light intensity, compare them with preset fixed thresholds, and then use simple logic to start or stop actuators such as fans, wet curtains, or supplemental lighting to maintain a relatively stable environment inside the greenhouse.

[0003] However, this traditional environmental control method has significant technical shortcomings in practical applications. First, existing technologies generally lack in-depth perception and feedback of the real-time physiological state of crops, often setting fixed environmental parameter targets based solely on experience, ignoring the dynamic changes in the environmental needs of crops as living organisms at different growth stages and physiological states. This leads to a frequent mismatch between environmental supply and the actual physiological needs of crops. Second, traditional control methods typically treat parameters such as temperature and humidity as independent variables for adjustment, lacking quantitative analysis of the coupling relationships between multidimensional environmental parameters, and failing to accurately assess the degree of fit between the overall environmental supply state and the complex needs of crops.

[0004] Furthermore, existing feedback control strategies primarily rely on ex-post correction, meaning that regulation is only triggered after environmental parameters have deviated from the set range. This delayed response mechanism not only easily leads to control overshoot and oscillations but also fails to predict the inertial trend of environmental changes, making it difficult to provide effective buffer protection before environmental deterioration. Simultaneously, this rigid regulation method often acts against the natural evolution of the environment, ignoring the physical inertia of the environment itself, resulting in excessive energy consumption and exacerbated mechanical wear and tear on equipment. Therefore, how to construct a greenhouse intelligent regulation method that can deeply integrate multimodal data, accurately quantify the supply-demand coupling relationship, and possess forward-looking predictive capabilities is a pressing technical problem that needs to be solved in the field of facility agriculture. Summary of the Invention

[0005] To address the shortcomings of existing technologies, embodiments of the present invention provide a method for intelligent dynamic coupling control of greenhouses based on deep fusion of multimodal data, comprising the following steps: Acquire real-time environmental supply data sequences and real-time crop physiological data sequences within a preset time sliding window in the greenhouse; use a pre-constructed physiological adaptation environmental demand mapping model to transform the real-time crop physiological data sequences into ideal environmental state vector sequences, and map them to a multi-dimensional environmental parameter vector space to construct a physiological adaptation manifold that represents the dynamic needs of crops; Using manifold reconstruction technology, an environmental supply manifold representing the distribution of real-time environmental supply status is constructed in a multidimensional environmental parameter vector space from the real-time environmental supply data sequence. The overlap metric between the physiologically adapted manifold and the environmental supply manifold in the multidimensional environmental parameter vector space is calculated, and the niche coupling index of the greenhouse is calculated based on the overlap metric. When the niche coupling index is less than a preset niche imbalance threshold, the greenhouse supply and demand relationship is determined to be unbalanced, and the following environmental compensation and regulation strategies are implemented: Step A: Determine the target steady-state center based on the physiological adaptation manifold, determine the real-time environmental state points based on the environmental supply manifold, and construct a state restoration vector pointing from the real-time environmental state points to the target steady-state center; Step B: Determine the direction of environmental inertial evolution based on the temporal change characteristics of the real-time environmental supply data sequence, call the preset digital twin model, and generate the environmental evolution prediction trajectory along the direction of environmental inertial evolution in the digital twin model with the real-time environmental state point as the initial starting point, and determine the physiological tolerance boundary point on the environmental evolution prediction trajectory. Step C: Calculate the safe buffer path length between the real-time environmental state point and the physiological tolerance boundary point on the predicted trajectory of environmental evolution. Based on the safe buffer path length and the state restoration vector, generate control instructions and execute environmental compensation control operations.

[0006] According to a preferred embodiment, the real-time environmental supply data sequence includes: temperature, humidity, light intensity, and CO2 concentration within a continuous time period in the greenhouse; the real-time crop physiological data sequence includes: leaf temperature, stomatal conductance, photosynthetic rate, and stem flow rate of the crop within a continuous time period in the greenhouse.

[0007] According to a preferred embodiment, the physiological adaptation environment requirement mapping model is constructed based on deep neural networks or Gaussian process regression, and is used to establish the feature adaptation mapping relationship between crop physiological state and ideal environmental supply parameters. It characterizes the theoretically optimal combination of temperature, humidity, light intensity and CO2 concentration required to maintain or optimize the physiological state when the crop exhibits specific stem flow rate, photosynthetic rate, leaf temperature and stomatal conductance.

[0008] According to a preferred embodiment, mapping real-time crop physiological data sequences to a multi-dimensional environmental parameter vector space using a physiological adaptation environment demand mapping model to construct a physiological adaptation manifold includes: The physiological adaptation environment requirement mapping model is used to map each sampling point in the real-time crop physiological data sequence to an ideal environment state vector, forming an ideal environment state vector sequence, wherein the dimension of each ideal environment state vector is defined by the output parameters of the model. Construct a multidimensional environmental parameter vector space with the same dimension as the ideal environmental state vector, and map the sequence of the ideal environmental state vectors to a set of discrete coordinate points in this space; Based on the topological structure of the discrete coordinate point set, a discrete state vector sequence of the physiologically adapted manifold is defined, and the discrete state vector sequence is used as the numerical basis for constructing the topological structure of the continuous manifold.

[0009] According to a preferred embodiment, constructing an environmental supply manifold in a multidimensional environmental parameter vector space using manifold reconstruction technology for a real-time environmental supply data sequence includes: The acquired real-time environmental supply data sequence is normalized. Based on the dimension definition of the multidimensional environmental parameter vector space, the corresponding parameter values ​​are extracted synchronously from the normalized data to construct the measured environmental state vector sequence. The measured environmental state vector sequence is mapped to the multidimensional environmental parameter vector space to form an environmental evolution trajectory that evolves over time. By using the sliding window technique to extract a local segment of the environmental evolution trajectory, its discrete coordinate point set is defined as a discrete state vector sequence of the environmental supply manifold.

[0010] According to a preferred embodiment, calculating the niche coupling index of the greenhouse based on the overlap metric includes: Using the multidimensional kernel density estimation method, the discrete state vector sequences of the environmental supply manifold and the physiological adaptation manifold are smoothed with the Gaussian kernel function as the kernel basis, respectively. This process reconstructs a probability density function field with a continuous topological structure in the multidimensional environmental parameter vector space, which is defined as the environmental supply manifold field and the physiological adaptation manifold field, respectively. Perform full-space overlap integral operations on the two manifold fields to calculate the Batachaya coefficients of the overlapping region of the two continuous probability density function fields in the multidimensional environmental parameter vector space; The Batachaya coefficient is defined as an overlap metric, and the overlap metric is normalized and mapped as a niche coupling index. The niche coupling index is used to quantify the substantial overlap in topological structure between the probability distribution space of environmental supply and the probability distribution space of crop demand.

[0011] According to a preferred embodiment, determining the state restoration vector and the direction of environmental inertial evolution includes: Calculate the global probability density maximum point of the physiological adaptation manifold field in the multidimensional environmental parameter vector space, and define it as the target steady-state center; Extract the terminal vector with the current time stamp from the discrete state vector sequence of the environmental supply manifold and define it as the real-time environmental state point; A vector is constructed pointing from the real-time environmental state point to the target steady-state center, and defined as the state restoration vector; Based on the real-time environmental supply data sequence, the first and second derivatives of each environmental supply parameter as a function of time within a sliding time window are calculated. A weighted feature vector is constructed based on the first and second derivatives to obtain a time-series trend vector characterizing the evolution of environmental physical inertia. The direction of the time-series trend vector is determined as the direction of environmental inertia evolution.

[0012] According to a preferred embodiment, an environmental evolution prediction trajectory is generated in the digital twin model along the direction of environmental inertial evolution, and the physiological tolerance boundary points on the environmental evolution prediction trajectory are determined by: Construct a digital twin model that includes greenhouse thermodynamic equations and environmental dynamic parameters; Starting from the real-time environmental state point, the digital twin model is invoked to iteratively deduce along the inertial evolution direction of the environment. In each iteration, the deduction step size is dynamically adjusted according to the rate of change of the environmental supply parameters to generate a series of virtual environmental state points. Connect the virtual environment state points in time sequence to construct an environment evolution prediction trajectory in the multidimensional environment parameter vector space; The environmental evolution prediction trajectory is mapped onto the physiologically adapted manifold field, and the probability density value of each virtual environmental state point on the environmental evolution prediction trajectory in the continuous probability density function field is calculated. The gradient of probability density value along the predicted trajectory of environmental evolution is monitored, and the virtual environmental state point corresponding to the absolute value of the probability density change rate exceeding the preset gradient threshold or the probability density value decaying to the preset critical threshold is defined as the physiological tolerance boundary point.

[0013] According to a preferred embodiment, calculating the safe buffer path length between the real-time environmental state point and the physiological tolerance boundary point on the environmental evolution prediction trajectory includes: Determine whether the probability density value of the real-time environmental state point in the continuous probability density function field is lower than a preset critical threshold. If the value is below the preset critical threshold, the length of the safety buffer path will be set to zero. If it exceeds the preset critical threshold, then in the multidimensional environmental parameter vector space, the path integral length between the real-time environmental state point and the physiological tolerance boundary point along the predicted environmental evolution trajectory is calculated. The path integral length is defined as the safe buffer path length; the safe buffer path length represents the remaining evolutionary buffer space required for the current greenhouse environment to evolve to the point where the crop leaves its suitable physiological niche.

[0014] According to a preferred embodiment, generating control commands based on the safety buffer path length and state restoration vector, and performing environmental compensation control operations includes: When the length of the safety buffer path is less than the preset emergency threshold, it is determined to be an emergency reset state, and the state restoration vector is directly selected as the main control command. When the length of the safety buffer path is greater than or equal to the preset emergency threshold, it is determined to be in a warning and correction state. Based on the length of the safety buffer path, the inertial weight coefficient is determined, and an inertial guidance vector with a unit time step along the inertial evolution direction of the environment is constructed. The state restoration vector and the inertial guidance vector are weighted and synthesized using the inertial weighting coefficient to generate a flexible correction vector; A multi-device coupled control matrix for greenhouses is constructed, and a multivariable decoupling control algorithm is used to convert the master control command or flexible correction vector into the execution power command of the greenhouse environmental control equipment.

[0015] The present invention has the following beneficial effects: 1. By constructing a manifold overlap model, the degree of matching between environmental supply and crop physiological needs is accurately quantified from the perspective of probability distribution. This method overcomes the limitation of traditional single threshold regulation that ignores the synergy of multiple parameters, ensuring that the greenhouse environment always maintains a high degree of overlap with the optimal growth niche of crops.

[0016] 2. By using digital twin models to deduce the direction of environmental inertia evolution and calculate the safe buffer path from the physiological tolerance boundary, this forward-looking prediction mechanism can detect and mitigate risks in advance, effectively eliminating the lag of traditional feedback control and preventing environmental abrupt changes from causing stress to crops.

[0017] 3. A hierarchical control strategy based on a safety buffer space is adopted to achieve dynamic switching between "emergency reset" and "flexible correction". This strategy utilizes vector synthesis technology to fine-tune in accordance with the natural inertia of environmental changes, significantly reducing energy consumption and minimizing equipment mechanical wear while ensuring control accuracy.

[0018] 4. A physiological adaptation and environmental demand mapping model was constructed based on deep learning, enabling inverse deduction from the crop's intrinsic physiological characteristics to external environmental parameters. This allows the control system to directly respond to the crop's real-time vital sign requirements, eliminating reliance on human experience and effectively improving crop photosynthetic efficiency. Attached Figure Description

[0019] Figure 1 A flowchart of a greenhouse intelligent dynamic coupling control method based on deep fusion of multimodal data is provided as an exemplary embodiment. Detailed Implementation

[0020] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0021] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0022] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0023] See Figure 1 The intelligent dynamic coupling control method for greenhouses based on deep fusion of multimodal data, as described in this invention, includes: S1. Obtain the real-time environmental supply data sequence and real-time crop physiological data sequence within a preset time sliding window in the greenhouse; using a pre-constructed physiological adaptation environmental demand mapping model, transform the real-time crop physiological data sequence into an ideal environmental state vector sequence, and map it to a multi-dimensional environmental parameter vector space to construct a physiological adaptation manifold representing the dynamic demand of crops.

[0024] Optionally, the real-time environmental supply data sequence refers to the physical environmental conditions currently provided by the greenhouse, including temperature, humidity, light intensity and CO2 concentration within the greenhouse over a continuous period of time. These environmental supply parameters constitute the external supply side for crop growth.

[0025] Optionally, real-time crop physiological data sequences refer to the real-time physiological state exhibited by crops as living organisms. Leaf temperature, stomatal conductance (characterizing gas exchange capacity), photosynthetic rate (characterizing energy accumulation efficiency), and stem flow rate (characterizing water transport status) are collected using non-invasive sensors.

[0026] Optionally, the physiological adaptation environment requirement mapping model is trained based on a deep neural network (DNN) or Gaussian process regression (GPR). Its input is a physiological indicator, and its output is a vector of ideal environmental states theoretically required to maintain or optimize that physiological state (i.e., the most suitable combination of temperature, humidity, etc. at that time). This is not just a simple linear correspondence, but an adaptation mapping that includes nonlinear features.

[0027] Preferably, the physiological adaptation environment demand mapping model is constructed based on deep neural networks or Gaussian process regression, and is used to establish the feature adaptation mapping relationship between crop physiological state and ideal environmental supply parameters. It represents the theoretically optimal combination of temperature, humidity, light intensity and CO2 concentration required to maintain or optimize the physiological state when the crop exhibits specific stem flow rate, photosynthetic rate, leaf temperature and stomatal conductance.

[0028] Optionally, the system transforms each physiological data point within the time window into a series of ideal environmental state vectors using a model. These vectors are then mapped to a multi-dimensional environmental parameter vector space (e.g., a 4-dimensional space with temperature, humidity, light, and air as axes). Since crop requirements are continuously changing, these points will form a topological structure in space.

[0029] Preferably, the process of mapping real-time crop physiological data sequences to a multi-dimensional environmental parameter vector space using a physiological adaptation environment requirement mapping model to construct a physiological adaptation manifold includes: The physiological adaptation environment requirement mapping model is used to map each sampling point in the real-time crop physiological data sequence to an ideal environment state vector, forming an ideal environment state vector sequence, wherein the dimension of each ideal environment state vector is defined by the output parameters of the model. Construct a multidimensional environmental parameter vector space with the same dimension as the ideal environmental state vector, and map the sequence of the ideal environmental state vectors to a set of discrete coordinate points in this space; Based on the topological structure of the discrete coordinate point set, a discrete state vector sequence of the physiologically adapted manifold is defined, and the discrete state vector sequence is used as the numerical basis for constructing the topological structure of the continuous manifold.

[0030] Optionally, the physiological adaptation manifold characterizes the dynamic environmental requirements of crops ("demand-side" profile). It is not derived from externally set fixed values ​​(such as "temperature 25℃"), but rather from the real-time physiological data of the crop itself (leaf temperature, stomatal conductance, photosynthetic rate, stem flow rate). It represents the theoretically optimal conditions that the external environment should provide to maintain or optimize a crop in a specific physiologically active state.

[0031] The physiologically adapted manifold represents the optimal coupling relationship of multidimensional parameters. It is not a linear index of a single parameter, but a geometric body with a topological structure located in the multidimensional environmental parameter vector space (a coordinate system composed of temperature, humidity, light, and CO2 concentration).

[0032] Physiologically adaptive manifolds characterize how these environmental supply parameters interact to meet the needs of crops. For example, under strong light conditions, crops may require higher humidity and specific temperatures to maximize photosynthesis. Physiologically adaptive manifolds are the continuous spatial distribution of these optimal combinations.

[0033] S2. Using manifold reconstruction technology, an environmental supply manifold representing the distribution of real-time environmental supply status is constructed in a multi-dimensional environmental parameter vector space from the real-time environmental supply data sequence.

[0034] Preferably, constructing an environmental supply manifold in a multidimensional environmental parameter vector space using manifold reconstruction technology for the real-time environmental supply data sequence includes: The acquired real-time environmental supply data sequence is normalized. Based on the dimension definition of the multidimensional environmental parameter vector space, the corresponding parameter values ​​are extracted synchronously from the normalized data to construct the measured environmental state vector sequence. The measured environmental state vector sequence is mapped to the multidimensional environmental parameter vector space to form an environmental evolution trajectory that evolves over time. By using the sliding window technique to extract a local segment of the environmental evolution trajectory, its discrete coordinate point set is defined as a discrete state vector sequence of the environmental supply manifold.

[0035] Optionally, the environmental supply manifold characterizes the supply-side reality of the environment. It is constructed from objective data (temperature, humidity, light intensity, CO2 concentration) collected in real time by sensors installed inside the greenhouse. It objectively records what kind of physical environment the greenhouse is currently providing for the crops. If the physiological adaptation manifold is what the crops want, then the environmental supply manifold is what the greenhouse provides.

[0036] Optionally, the environmental supply manifold characterizes the temporal evolution of environmental supply parameters. It is not merely a static data point, but rather a series of data sequences constructed based on a preset time sliding window (e.g., the past 30 minutes). It forms an "evolutionary trajectory" or distribution band in multidimensional space.

[0037] The system determines whether the greenhouse environment meets the crop's needs by calculating the overlap between the "environmental supply manifold" and the "physiological adaptation manifold" in multidimensional space. If the distribution area of ​​the environmental supply manifold falls entirely within the range of the physiological adaptation manifold, it indicates that the greenhouse environment supply perfectly matches the crop's physiological needs.

[0038] S3. Calculate the overlap metric between the physiologically adapted manifold and the environmentally supplied manifold in the multidimensional environmental parameter vector space, and calculate the niche coupling index of the greenhouse based on the overlap metric.

[0039] Optionally, calculating the niche coupling index of the greenhouse based on the overlap metric includes: Using the multidimensional kernel density estimation method, the discrete state vector sequences of the environmental supply manifold and the physiological adaptation manifold are smoothed with the Gaussian kernel function as the kernel basis, respectively. This process reconstructs a probability density function field with a continuous topological structure in the multidimensional environmental parameter vector space, which is defined as the environmental supply manifold field and the physiological adaptation manifold field, respectively. Perform full-space overlap integral operations on the two manifold fields to calculate the Batachaya coefficients of the overlapping region of the two continuous probability density function fields in the multidimensional environmental parameter vector space; The Batachaya coefficient is defined as an overlap metric, and the overlap metric is normalized and mapped as a niche coupling index. The niche coupling index is used to quantify the substantial overlap in topological structure between the probability distribution space of environmental supply and the probability distribution space of crop demand.

[0040] Alternatively, a higher niche coupling index indicates that the distribution of the "environment provided by the greenhouse" and the "environment required by the crop" is highly overlapping, and the crop is in the optimal suitable area; a lower value indicates a mismatch between supply and demand, which requires regulation.

[0041] S4. When the niche coupling index is less than the preset niche imbalance threshold, the greenhouse supply and demand relationship is determined to be unbalanced, and the following environmental compensation and regulation strategy is executed: Step A: Determine the target steady-state center based on the physiological adaptation manifold, determine the real-time environmental state point based on the environmental supply manifold, and construct a state restoration vector pointing from the real-time environmental state point to the target steady-state center.

[0042] Preferably, the global probability density maximum point of the physiological adaptation manifold field in the multidimensional environmental parameter vector space is calculated and defined as the target steady-state center; Extract the terminal vector with the current time stamp from the discrete state vector sequence of the environmental supply manifold and define it as the real-time environmental state point; A vector is constructed pointing from the real-time environmental state point to the target steady-state center, and defined as the state restoration vector.

[0043] Optionally, the state restoration vector is a straight line vector pointing from the real-time environmental state point to the target steady-state center. This represents the theoretically fastest path to return to the optimal state.

[0044] S5, Step B: Determine the direction of environmental inertial evolution based on the temporal change characteristics of the real-time environmental supply data sequence, call the preset digital twin model, generate the environmental evolution prediction trajectory along the direction of environmental inertial evolution in the digital twin model with the real-time environmental state point as the initial starting point, and determine the physiological tolerance boundary point on the environmental evolution prediction trajectory.

[0045] Preferably, determining the state restoration vector and the direction of environmental inertial evolution includes: Based on the real-time environmental supply data sequence, the first and second derivatives of each environmental supply parameter as a function of time within a sliding time window are calculated. A weighted feature vector is constructed based on the first and second derivatives to obtain a time-series trend vector characterizing the evolution of environmental physical inertia. The direction of the time-series trend vector is determined as the direction of environmental inertia evolution.

[0046] Preferably, in the digital twin model, an environmental evolution prediction trajectory is generated along the direction of environmental inertial evolution, and the physiological tolerance boundary points on the environmental evolution prediction trajectory are determined, including: Construct a digital twin model that includes greenhouse thermodynamic equations and environmental dynamic parameters; Starting from the real-time environmental state point, the digital twin model is invoked to iteratively deduce along the inertial evolution direction of the environment. In each iteration, the deduction step size is dynamically adjusted according to the rate of change of the environmental supply parameters to generate a series of virtual environmental state points. Connect the virtual environment state points in time sequence to construct an environment evolution prediction trajectory in the multidimensional environment parameter vector space; The environmental evolution prediction trajectory is mapped onto the physiologically adapted manifold field, and the probability density value of each virtual environmental state point on the environmental evolution prediction trajectory in the continuous probability density function field is calculated. The gradient of probability density value along the predicted trajectory of environmental evolution is monitored, and the virtual environmental state point corresponding to the absolute value of the probability density change rate exceeding the preset gradient threshold or the probability density value decaying to the preset critical threshold is defined as the physiological tolerance boundary point.

[0047] Optionally, the physiological tolerance boundary point characterizes the turning point when the environmental state changes from "suitable for growth" to "causing physiological damage". Before this point, the crop is in the "comfort zone" or "adjustable zone"; once this point is crossed, the crop will be out of the suitable physiological niche and may begin to suffer from negative effects such as high temperature inhibition, stomatal closure or photosynthetic stagnation.

[0048] The physiological tolerance boundary point is a point on the environmental evolution prediction trajectory that meets any of the following conditions: The probability density value is too low: the value decays to a preset critical threshold (indicating that the environmental combination has extremely low fitness). The gradient of change is too large: the absolute value of the probability density change rate exceeds a preset gradient threshold (indicating that the environmental fitness is experiencing a precipitous drop).

[0049] Physiological tolerance boundary points are not points that have already occurred, but rather points that may be reached in the future. Specifically, physiological tolerance boundary points are critical warning points on the predicted trajectory where crops are about to slide from a "safe state" to a "dangerous state".

[0050] S6, Step C: Calculate the safe buffer path length between the real-time environmental state point and the physiological tolerance boundary point on the environmental evolution prediction trajectory, generate control instructions based on the safe buffer path length and the state restoration vector, and execute environmental compensation control operations.

[0051] Preferably, calculating the safe buffer path length between the real-time environmental state point and the physiological tolerance boundary point on the environmental evolution prediction trajectory includes: Determine whether the probability density value of the real-time environmental state point in the continuous probability density function field is lower than a preset critical threshold. If the value is below the preset critical threshold, the length of the safety buffer path will be set to zero. If it exceeds the preset critical threshold, then in the multidimensional environmental parameter vector space, the path integral length between the real-time environmental state point and the physiological tolerance boundary point along the predicted environmental evolution trajectory is calculated. The path integral length is defined as the safe buffer path length; the safe buffer path length represents the remaining evolutionary buffer space required for the current greenhouse environment to evolve to the point where the crop leaves its suitable physiological niche.

[0052] Preferably, the environmental compensation control operation, which generates control instructions based on the safety buffer path length and state restoration vector, includes: When the length of the safety buffer path is less than the preset emergency threshold, it is determined to be an emergency reset state, and the state restoration vector is directly selected as the main control command. When the length of the safety buffer path is greater than or equal to the preset emergency threshold, it is determined to be in a warning and correction state. Based on the length of the safety buffer path, the inertial weight coefficient is determined, and an inertial guidance vector with a unit time step along the inertial evolution direction of the environment is constructed. The state restoration vector and the inertial guidance vector are weighted and synthesized using the inertial weighting coefficient to generate a flexible correction vector; A multi-device coupled control matrix for greenhouses is constructed, and a multivariable decoupling control algorithm is used to convert the master control command or flexible correction vector into the execution power command of the greenhouse environmental control equipment.

[0053] The master control command or flexible correction vector is an abstract "environmental parameter adjustment quantity" (e.g., a 2°C temperature drop and a 5% humidification increase). However, in the physical world, greenhouse equipment is often coupled (e.g., opening the evaporative cooling pad both cools and humidifies, and opening the roof window both cools and humidifies). Therefore, decoupling transformation is necessary. A coupling control matrix is ​​constructed: the system pre-configured a multi-equipment coupling control matrix for the greenhouse. This matrix describes the specific impact of each unit action of each device (fan, evaporative cooling pad, shading net, supplemental lighting) on ​​various environmental parameters (temperature, humidity, light, and air).

[0054] Multivariable decoupling operation: The system utilizes a multivariable decoupling control algorithm (usually involving matrix inversion or pseudo-inversion operations) to deduce the power command to be executed by the equipment based on the "master control command" or "flexible correction vector". Final output: After decoupling, the abstract mathematical vector is transformed into specific physical actions, such as: fan: turn on 60% power, external shading: expand 30%, micro-mist humidification: turn off, supplementary lighting: maintain the status quo.

[0055] This application constructs a manifold overlap model to accurately quantify the degree of matching between environmental supply and crop physiological needs from a probability distribution perspective. This method overcomes the limitation of traditional single-threshold regulation that ignores the synergy of multiple parameters, ensuring that the greenhouse environment always maintains a high degree of overlap with the optimal growth niche of crops.

[0056] By using digital twin models to predict the direction of environmental inertia evolution and calculate safe buffer paths from physiological tolerance boundaries, this forward-looking predictive mechanism can detect and mitigate risks in advance, effectively eliminating the lag of traditional feedback control and preventing environmental abrupt changes from causing stress to crops.

[0057] A hierarchical control strategy based on a safety buffer space is adopted to achieve dynamic switching between "emergency reset" and "flexible correction". This strategy utilizes vector synthesis technology to fine-tune in accordance with the natural inertia of environmental changes, significantly reducing energy consumption and minimizing equipment mechanical wear while ensuring control accuracy.

[0058] A physiological adaptation model for environmental requirements was constructed based on deep learning, enabling inverse deduction from the crop's intrinsic physiological characteristics to external environmental parameters. This allows the control system to directly respond to the crop's real-time vital signs, eliminating reliance on human experience and effectively improving crop photosynthetic efficiency.

[0059] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0060] The present invention discloses a non-transitory computer-readable storage medium storing computer instructions, which, when executed by a processor, cause the processor to perform the above-described method.

[0061] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware (e.g., processor, FPGA, ASIC, etc.), and the program can be stored in a readable storage medium, such as a read-only memory, a disk, or an optical disk. All or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module in the above embodiments can be implemented in hardware, such as by using integrated circuits to implement its corresponding function, or it can be implemented as a software functional module, such as by a processor executing a program / instruction stored in memory to implement its corresponding function. The embodiments of the present invention are not limited to any particular combination of hardware and software.

[0062] Furthermore, the functional units in the various embodiments of this document can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0063] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this article, in essence, or the part that contributes to the prior art, or all or part of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this article. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0064] The above description is only a preferred embodiment of the present invention and is not intended to limit 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 intelligent dynamic coupling control of greenhouses based on deep fusion of multimodal data, characterized in that, Includes the following steps: Acquire real-time environmental supply data sequences and real-time crop physiological data sequences within a preset time sliding window in the greenhouse; use a pre-constructed physiological adaptation environmental demand mapping model to transform the real-time crop physiological data sequences into ideal environmental state vector sequences, and map them to a multi-dimensional environmental parameter vector space to construct a physiological adaptation manifold that represents the dynamic needs of crops; Using manifold reconstruction technology, an environmental supply manifold representing the distribution of real-time environmental supply status is constructed in a multidimensional environmental parameter vector space from the real-time environmental supply data sequence. The overlap metric between the physiologically adapted manifold and the environmental supply manifold in the multidimensional environmental parameter vector space is calculated, and the niche coupling index of the greenhouse is calculated based on the overlap metric. When the niche coupling index is less than a preset niche imbalance threshold, the greenhouse supply and demand relationship is determined to be unbalanced, and the following environmental compensation and regulation strategies are implemented: Step A: Determine the target steady-state center based on the physiological adaptation manifold, determine the real-time environmental state points based on the environmental supply manifold, and construct a state restoration vector pointing from the real-time environmental state points to the target steady-state center; Step B: Determine the direction of environmental inertial evolution based on the temporal change characteristics of the real-time environmental supply data sequence, call the preset digital twin model, and generate the environmental evolution prediction trajectory along the direction of environmental inertial evolution in the digital twin model with the real-time environmental state point as the initial starting point, and determine the physiological tolerance boundary point on the environmental evolution prediction trajectory. Step C: Calculate the safe buffer path length between the real-time environmental state point and the physiological tolerance boundary point on the predicted trajectory of environmental evolution. Based on the safe buffer path length and the state restoration vector, generate control instructions and execute environmental compensation control operations.

2. The method according to claim 1, characterized in that, The real-time environmental supply data sequence includes: temperature, humidity, light intensity, and CO2 concentration within the greenhouse over a continuous period of time; the real-time crop physiological data sequence includes: leaf temperature, stomatal conductance, photosynthetic rate, and stem flow rate of crops within the greenhouse over a continuous period of time.

3. The method according to claim 2, characterized in that, The physiological adaptation environment requirement mapping model is constructed based on deep neural networks or Gaussian process regression. It is used to establish the feature adaptation mapping relationship between crop physiological state and ideal environmental supply parameters. It represents the theoretically optimal combination of temperature, humidity, light intensity and CO2 concentration required to maintain or optimize the physiological state when the crop exhibits specific stem flow rate, photosynthetic rate, leaf temperature and stomatal conductance.

4. The method according to claim 3, characterized in that, The physiological adaptation manifold is constructed by mapping real-time crop physiological data sequences to a multi-dimensional environmental parameter vector space using a physiological adaptation environmental demand mapping model. The physiological adaptation environment requirement mapping model is used to map each sampling point in the real-time crop physiological data sequence to an ideal environment state vector, forming an ideal environment state vector sequence, wherein the dimension of each ideal environment state vector is defined by the output parameters of the model. Construct a multidimensional environmental parameter vector space with the same dimension as the ideal environmental state vector, and map the sequence of the ideal environmental state vectors to a set of discrete coordinate points in this space; Based on the topological structure of the discrete coordinate point set, a discrete state vector sequence of the physiologically adapted manifold is defined, and the discrete state vector sequence is used as the numerical basis for constructing the topological structure of the continuous manifold.

5. The method according to claim 4, characterized in that, Using manifold reconstruction techniques to construct an environmental supply manifold in a multidimensional environmental parameter vector space from real-time environmental supply data sequences includes: The acquired real-time environmental supply data sequence is normalized. Based on the dimension definition of the multidimensional environmental parameter vector space, the corresponding parameter values ​​are extracted synchronously from the normalized data to construct the measured environmental state vector sequence. The measured environmental state vector sequence is mapped to the multidimensional environmental parameter vector space to form an environmental evolution trajectory that evolves over time. By using the sliding window technique to extract a local segment of the environmental evolution trajectory, its discrete coordinate point set is defined as a discrete state vector sequence of the environmental supply manifold.

6. The method according to claim 5, characterized in that, The calculation of the niche coupling index of the greenhouse based on the aforementioned overlap metric includes: Using the multidimensional kernel density estimation method, the discrete state vector sequences of the environmental supply manifold and the physiological adaptation manifold are smoothed with the Gaussian kernel function as the kernel basis, respectively. This process reconstructs a probability density function field with a continuous topological structure in the multidimensional environmental parameter vector space, which is defined as the environmental supply manifold field and the physiological adaptation manifold field, respectively. Perform full-space overlap integral operations on the two manifold fields to calculate the Batachaya coefficients of the overlapping region of the two continuous probability density function fields in the multidimensional environmental parameter vector space; The Batachaya coefficient is defined as an overlap metric, and the overlap metric is normalized and mapped as a niche coupling index. The niche coupling index is used to quantify the substantial overlap in topological structure between the probability distribution space of environmental supply and the probability distribution space of crop demand.

7. The method according to claim 6, characterized in that, Determining the state restoration vector and the direction of environmental inertial evolution includes: Calculate the global probability density maximum point of the physiological adaptation manifold field in the multidimensional environmental parameter vector space, and define it as the target steady-state center; Extract the terminal vector with the current time stamp from the discrete state vector sequence of the environmental supply manifold and define it as the real-time environmental state point; A vector is constructed pointing from the real-time environmental state point to the target steady-state center, and defined as the state restoration vector; Based on the real-time environmental supply data sequence, the first and second derivatives of each environmental supply parameter as a function of time within a sliding time window are calculated. A weighted feature vector is constructed based on the first and second derivatives to obtain a time-series trend vector characterizing the evolution of environmental physical inertia. The direction of the time-series trend vector is determined as the direction of environmental inertia evolution.

8. The method according to claim 7, characterized in that, In the digital twin model, an environmental evolution prediction trajectory is generated along the direction of environmental inertial evolution. The physiological tolerance boundary points on the environmental evolution prediction trajectory are determined, including: Construct a digital twin model that includes greenhouse thermodynamic equations and environmental dynamic parameters; Starting from the real-time environmental state point, the digital twin model is invoked to iteratively deduce along the inertial evolution direction of the environment. In each iteration, the deduction step size is dynamically adjusted according to the rate of change of the environmental supply parameters to generate a series of virtual environmental state points. Connect the virtual environment state points in time sequence to construct an environment evolution prediction trajectory in the multidimensional environment parameter vector space; The environmental evolution prediction trajectory is mapped onto the physiologically adapted manifold field, and the probability density value of each virtual environmental state point on the environmental evolution prediction trajectory in the continuous probability density function field is calculated. The gradient of probability density value along the predicted trajectory of environmental evolution is monitored, and the virtual environmental state point corresponding to the absolute value of the probability density change rate exceeding the preset gradient threshold or the probability density value decaying to the preset critical threshold is defined as the physiological tolerance boundary point.

9. The method according to claim 8, characterized in that, The calculation of the safe buffer path length between the real-time environmental state point and the physiological tolerance boundary point on the environmental evolution prediction trajectory includes: Determine whether the probability density value of the real-time environmental state point in the continuous probability density function field is lower than a preset critical threshold. If the value is below the preset critical threshold, the length of the safety buffer path will be set to zero. If it exceeds the preset critical threshold, then in the multidimensional environmental parameter vector space, the path integral length between the real-time environmental state point and the physiological tolerance boundary point along the predicted environmental evolution trajectory is calculated. The path integral length is defined as the safe buffer path length; The safe buffer path length represents the remaining evolutionary buffer space required for the current greenhouse environment to evolve to the point where crops are no longer in their suitable physiological niche.

10. The method according to claim 9, characterized in that, Based on the safety buffer path length and state restoration vector, control commands are generated, and environmental compensation control operations are performed, including: When the length of the safety buffer path is less than the preset emergency threshold, it is determined to be an emergency reset state, and the state restoration vector is directly selected as the main control command. When the length of the safety buffer path is greater than or equal to the preset emergency threshold, it is determined to be in a warning and correction state. Based on the length of the safety buffer path, the inertial weight coefficient is determined, and an inertial guidance vector with a unit time step along the inertial evolution direction of the environment is constructed. The state restoration vector and the inertial guidance vector are weighted and synthesized using the inertial weighting coefficient to generate a flexible correction vector; A multi-device coupled control matrix for greenhouses is constructed, and a multivariable decoupling control algorithm is used to convert the master control command or flexible correction vector into the execution power command of the greenhouse environmental control equipment.

Images