Flexible solar greenhouse temperature control method, system and device and storage medium

Abstract

The invention discloses a flexible sunlight greenhouse temperature control method, system and device and a storage medium, and relates to the technical field of greenhouse environment, the method comprises the following steps: calling a heat transfer model to carry out temperature field simulation calculation according to greenhouse internal thermal field distribution data to obtain a first predicted temperature value; calling a greenhouse temperature prediction model to perform time sequence analysis on the external environment temperature historical time sequence data to obtain a second predicted temperature value; performing data fusion according to the first predicted temperature value and the second predicted temperature value to obtain a comprehensive predicted temperature value; and according to a deviation value between the comprehensive prediction temperature value and a preset greenhouse air temperature target value, adjusting operation parameters of the greenhouse equipment. By simulating thermal field distribution in a greenhouse and processing time sequence data of an external environment, a deterministic rule and random fluctuation are covered, a deviation value with a target value is calculated according to an obtained comprehensive prediction value, and equipment is dynamically adjusted based on prediction deviation, so that ineffective energy consumption and equipment loss are reduced.

Description

Technical Field

[0001] This invention relates to the field of greenhouse environment technology, specifically to a method, system, equipment, and storage medium for temperature control in a flexible solar greenhouse. Background Technology

[0002] In the field of facility agriculture, greenhouse temperature is a core environmental parameter determining crop yield, quality, and market cycle, and is also a major driver of greenhouse energy consumption and carbon emissions. With the increasing demand for "low-carbon and high-efficiency agriculture," greenhouse structures are rapidly evolving from the traditional brick-frame-glass model to "multi-source heat storage flexible solar greenhouses." These greenhouses utilize complementary energy sources such as solar, geothermal, and biomass energy for heat storage and release, and employ flexible, lightweight materials like polymers and fibers to construct deformable frames and covering layers. They possess significant advantages such as high insulation, high heat storage, low cost, easy assembly and disassembly, and environmental friendliness, and have become a cutting-edge development direction in facility horticulture technology. However, due to the complex interplay of factors such as multi-source energy coupling, flexible structural deformation, geothermal frequency conversion control, and the porous and non-uniform nature of the covering layer, heat transfer between the inside and outside of the greenhouse exhibits highly nonlinear, time-varying, and multi-scale coupling characteristics. This makes accurate temperature prediction and real-time control a bottleneck restricting the improvement of greenhouse energy efficiency and production efficiency.

[0003] Existing greenhouse temperature prediction technologies are mainly divided into two categories: (1) Physical model-based methods: relying on the principles of thermodynamics and mass transfer, a set of energy balance differential equations is established, and the predicted temperature value is obtained through numerical solution. This type of method has a clear mechanism and strong interpretability, which is convenient for supporting the optimization of greenhouse structure and the formulation of control strategies; however, its model framework is difficult to fully characterize complex factors such as multi-source heat storage, dynamic deformation of flexible structures, frequency conversion regulation of ground heat and non-uniformity of materials, which causes the prediction accuracy and robustness to decrease significantly with the change of operating conditions. (2) Data model-based methods: using machine learning or deep learning technology, nonlinear mapping relationships are extracted from historical data to achieve temperature prediction. This type of method is more adaptable to dynamic random operating conditions, but due to the multimodal, high-dimensional, sparse and distribution offset characteristics of greenhouse historical data, the model is prone to overfitting and poor generalization; at the same time, due to the lack of physical constraints and prior knowledge, its interpretability and credibility are insufficient, making it difficult to provide a reliable basis for greenhouse optimization design and control strategies.

[0004] Therefore, there is an urgent need for a new technology for greenhouse temperature prediction and control that can integrate the advantages of physical mechanisms and data-driven approaches to achieve efficient, accurate, interpretable, and adaptive intelligent regulation of the greenhouse environment. Summary of the Invention

[0005] The main objective of this invention is to provide a flexible solar greenhouse temperature control method, system, device, and storage medium. By simulating the thermal field distribution inside the greenhouse and processing time-series data of the external environment, it covers deterministic patterns and random fluctuations. Then, based on the obtained comprehensive prediction value, it calculates the deviation from the target value and dynamically adjusts the device based on the prediction deviation to reduce ineffective energy consumption and equipment wear.

[0006] To achieve the above objectives, the embodiments of this application provide the following technical solutions: According to a first aspect of the embodiments of this application, a method for controlling the temperature of a flexible solar greenhouse is provided, applied to greenhouse equipment, the method comprising: Acquire environmental data for a flexible solar greenhouse; the environmental data for the flexible solar greenhouse includes data on the distribution of the internal thermal field of the greenhouse and historical time-series data on the external ambient temperature. The heat transfer model is invoked to perform temperature field simulation calculations based on the thermal field distribution data inside the greenhouse, and the first predicted temperature value is obtained. The greenhouse temperature prediction model is invoked to perform time-series analysis on the historical time-series data of the external environment temperature to obtain a second predicted temperature value. A comprehensive predicted temperature value is obtained by fusing the first predicted temperature value and the second predicted temperature value. The operating parameters of the greenhouse equipment are adjusted based on the deviation between the comprehensive predicted temperature value and the preset greenhouse air temperature target value.

[0007] According to a second aspect of the embodiments of this application, a flexible solar greenhouse temperature control system is provided, applied to greenhouse equipment, the system comprising: The data acquisition module is used to acquire environmental data of the flexible solar greenhouse; the environmental data of the flexible solar greenhouse includes the thermal field distribution data inside the greenhouse and the historical time series data of the external environment temperature; The heat transfer model calculation module is used to call the heat transfer model to perform temperature field simulation calculations based on the heat field distribution data inside the greenhouse, and obtain the first predicted temperature value. The temperature prediction module is used to call the greenhouse temperature prediction model to perform time-series analysis on the historical time-series data of the external environment temperature to obtain a second predicted temperature value. The fusion module is used to fuse the data based on the first predicted temperature value and the second predicted temperature value to obtain a comprehensive predicted temperature value. The greenhouse equipment control module is used to adjust the operating parameters of the greenhouse equipment based on the deviation between the comprehensive predicted temperature value and the preset greenhouse air temperature target value.

[0008] According to a third aspect of the present application, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in the first aspect above.

[0009] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided having computer-readable instructions stored thereon, which can be executed by a processor to implement the method described in the first aspect above.

[0010] In summary, this application provides a method, system, device, and storage medium for temperature control in a flexible solar greenhouse. The method involves acquiring environmental data of the flexible solar greenhouse, including internal thermal field distribution data and historical time-series data of external ambient temperature. A heat transfer model is invoked to simulate and calculate the temperature field based on the internal thermal field distribution data, yielding a first predicted temperature value. A greenhouse temperature prediction model is then invoked to perform time-series analysis on the historical time-series data of the external ambient temperature, yielding a second predicted temperature value. The first and second predicted temperature values ​​are then fused to obtain a comprehensive predicted temperature value. Based on the deviation between the comprehensive predicted temperature value and a preset target greenhouse temperature value, the operating parameters of the greenhouse equipment are adjusted. By simulating the internal thermal field distribution and processing the time-series data of the external environment, both deterministic patterns and random fluctuations are covered. The deviation from the target value is calculated based on the obtained comprehensive predicted value, and the equipment is dynamically adjusted based on the predicted deviation, reducing ineffective energy consumption and equipment wear. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0012] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0013] Figure 1 A schematic diagram of a temperature control method for a flexible solar greenhouse provided in this application embodiment; Figure 2A schematic diagram of a flexible solar greenhouse temperature control system provided in this application embodiment; Figure 3 This paper shows a structural diagram of an electronic device provided in an embodiment of this application; Figure 4 A diagram of a computer-readable storage medium provided in an embodiment of this application is shown.

[0014] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0017] Furthermore, in this invention, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0018] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0019] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0020] In recent years, some new technical methods have emerged, which help to overcome the shortcomings of traditional greenhouse temperature prediction methods. These mainly include the following three aspects: (1) Hybrid model linkage architecture, which utilizes the advantages of physical models and data models to realize greenhouse temperature prediction. By effectively integrating and coordinating physical models and data models, a unified greenhouse temperature prediction model is formed, which can reflect the inherent laws and mechanisms of greenhouses and capture the dynamics and randomness of greenhouses, thus achieving the goal of efficient, accurate, flexible and interpretable greenhouse temperature prediction. (2) Multi-scale coordination method, which utilizes different scales and perspectives to model and analyze the heat transfer process of greenhouses. By decomposing the heat transfer process of greenhouses into different scales and levels—microscale, mesoscale and macroscale or spatial scale, temporal scale and frequency scale, etc., and then establishing corresponding models and algorithms at each scale, the predicted value of greenhouse temperature is obtained by coordinating and integrating the results of different scales. Its advantage is that it can fully consider the influence of complex factors such as the porosity and non-uniformity of greenhouses, and improve the accuracy and stability of the model. (3) Adaptive weight generation method, which utilizes the characteristics and objectives of data to dynamically adjust the weights of the model. By using machine learning techniques such as Gaussian mixture models, convolutional neural networks, and gradient-weighted class activation mapping, data is clustered, features are extracted, and visualization optimizations are performed. Then, based on different clustering results and optimization objectives, different weight vectors are adaptively generated to dynamically adjust the contribution weights of various factors to greenhouse temperature prediction, thereby enhancing the model's generalization and transfer capabilities. The advantage is that it can adapt to different data distributions and prediction objectives, improving the model's flexibility and adaptability.

[0021] In summary, this invention proposes a greenhouse temperature prediction method based on a hybrid model linkage architecture. This method combines the advantages of heat transfer models and machine learning models, utilizing advanced techniques such as multi-scale coordination, spatiotemporal feature extraction, adaptive weight generation, and visualization optimization to improve the greenhouse's heat preservation and storage capacity, reduce energy consumption, and extend the production season. It provides an efficient, accurate, flexible, and interpretable solution for greenhouse temperature prediction. It comprehensively utilizes the advantages of physical and data models to achieve efficient, accurate, flexible, and interpretable prediction and control of greenhouse temperatures. It should fully consider the influence of various complex factors in greenhouses—such as multi-source heat storage, flexible structures, geothermal frequency conversion control, porosity, and non-uniformity—to improve the model's accuracy and stability. Simultaneously, it should fully utilize information from historical and future greenhouse data—such as spatiotemporally related features, dynamic and stochastic patterns, and adaptive and optimized weights—to improve the model's data fitting and predictive capabilities. Furthermore, it should provide model visualization and optimization, activating regional and gradient information, correlation strength and influence direction, optimization objectives, and control strategies, thereby improving the model's interpretability and reliability.

[0022] Figure 1 This application illustrates a temperature control method for a flexible solar greenhouse, applied to greenhouse equipment. The method includes: Step S101: Obtain environmental data of the flexible solar greenhouse; the environmental data of the flexible solar greenhouse includes the thermal field distribution data inside the greenhouse and the historical time series data of the external ambient temperature; Step S102: Call the heat transfer model to perform temperature field simulation calculation based on the heat field distribution data inside the greenhouse, and obtain the first predicted temperature value; Step S103: Call the greenhouse temperature prediction model to perform time series analysis on the historical time series data of the external environment temperature to obtain the second predicted temperature value; Step S104: Perform data fusion based on the first predicted temperature value and the second predicted temperature value to obtain a comprehensive predicted temperature value; Step S105: Adjust the operating parameters of the greenhouse equipment according to the deviation between the comprehensive predicted temperature value and the preset greenhouse temperature target value.

[0023] In one possible implementation, in step 103, the greenhouse temperature prediction model is invoked to perform time-series analysis on the historical time-series data of the external environment temperature to obtain a second predicted temperature value, according to the following formula: in, This is the second predicted temperature value. It is historical time-series data of external environmental temperature. These are the parameters of the machine learning model. It is an activation function. Characterizes the convolution operation. Representation pooling operation.

[0024] In one possible implementation, in step 102, the heat transfer model is invoked to perform temperature field simulation calculations based on the internal thermal field distribution data of the greenhouse, to obtain a first predicted temperature value, according to the following formula: Where M is the mass matrix, C is the damping matrix, K is the stiffness matrix, ρ is the material density, c is the specific heat capacity, and k is the thermal conductivity. It is the coefficient of thermal expansion, and T is the temperature. This is the reference temperature, and t is the time. It is gravitational acceleration. is a unit vector in the vertical direction, u is the nodal displacement vector, F is the nodal load vector, u̇ is the nodal velocity vector, q is the heat source term, and ∇ is the gradient operator. It is the resistance coefficient of the porous media in the greenhouse. It is the thermal diffusivity. It is a heat convection source term.

[0025] In one possible implementation, the construction of the greenhouse temperature prediction model includes the following steps: Historical time-series data of external ambient temperature is divided into spatiotemporal data blocks according to spatial and temporal region numbers. Spatial regions are divided according to sensor network topology, and temporal data is divided according to fixed time windows. Spatiotemporal feature vectors are extracted for each spatiotemporal data block, and all spatiotemporal feature vectors are combined into a spatiotemporal feature matrix. A multi-scale convolutional neural network is called to calculate the predicted value based on the spatiotemporal feature matrix. The multi-scale convolutional neural network is trained using the predicted value and the actual values ​​in the historical time-series data of external ambient temperature as training data to obtain the greenhouse temperature prediction model.

[0026] In one possible implementation, calling a multi-scale convolutional neural network to calculate a predicted value based on the spatiotemporal feature matrix includes: converting the spatiotemporal feature matrix into a multi-channel two-dimensional tensor as the input layer of the multi-scale convolutional neural network; extracting multi-scale features through several convolutional and pooling layers of the multi-scale convolutional neural network; converting the multi-scale features into a one-dimensional vector; and converting the one-dimensional vector into a predicted value.

[0027] In one possible implementation, the method further includes: using a Gaussian mixture model to probabilistically model the historical time-series data of the external environment temperature to obtain Gaussian model parameters and a posterior probability matrix; assigning Gaussian component labels to each data point according to the posterior probability matrix to obtain a label vector; assigning target values ​​to each Gaussian component according to the label vector and actual temperature data, grouping them by labels to obtain a target value vector; generating a weight vector for each data point according to the label vector and the target value vector; and optimizing the greenhouse temperature prediction model parameters according to the weight vector, the actual temperature data, and the greenhouse temperature prediction model parameters.

[0028] In one possible implementation, the step of using a Gaussian mixture model to probabilistically model the historical time-series data of the external environment temperature to obtain Gaussian model parameters and a posterior probability matrix includes the following steps: Step 1: Assume the input data follows a Gaussian mixture distribution, where each data point comprises multiple Gaussian components. Each Gaussian component is represented by a mean vector, a covariance matrix, and mixing coefficients. The Gaussian mixture distribution is expressed by the following formula: in, It is the input data. It is the number of Gaussian components. It is the first The mixing coefficients of the Gaussian components satisfy the following conditions: , It is the first The probability density function of Gaussian components, It is the first The mean vector of Gaussian components, It is the first The covariance matrix of Gaussian components; Step 2: Initialize the parameters of the Gaussian mixture model, setting the initial values ​​for the mean vector, covariance matrix, and mixing coefficients of the Gaussian components, denoted as . , ; Step 3: Calculate the posterior probability of each data point belonging to each Gaussian component. The posterior probability matrix is ​​expressed by the following formula: in, It is the first The data point belongs to the th data point The posterior probability of each Gaussian component It is the first The latent variables for each data point indicate which Gaussian component it belongs to. It is the first Observations for each data point It is the number of iterations; Step 4: Update the parameters of the Gaussian mixture model based on the current posterior probability, such that the mean vector, covariance matrix, and mixing coefficients of the Gaussian component that maximizes the expectation of the log-likelihood function of the input data with respect to the latent variables are obtained. The parameter update is expressed by the following formula: in, It is the first Updated values ​​of the mean vector, covariance matrix, and mixing coefficients of each Gaussian component. It is the number of input data; Step 5: Determine whether the increase in the log-likelihood function of the input data is less than a given threshold, or whether the number of iterations exceeds a given upper limit. If so, stop the iteration and output the parameters of the Gaussian mixture model; otherwise, return to continue the iteration. The log-likelihood function of the input data is expressed by the following formula: in, It is the input data. These are the parameters of the Gaussian mixture model. It is the number of Gaussian components.

[0029] In one possible implementation, optimizing the greenhouse temperature prediction model parameters based on the weight vector, the actual temperature data, and the greenhouse temperature prediction model parameters includes: in, It is the weight vector of the i-th data point. It is the first The target value of the Gaussian component to which each data point belongs. It is the first Labels for the Gaussian components of each data point; These are the parameters of the machine learning model. It is the number of input data. It is the first The weight vector of each data point It is the first The actual temperature of each data point It is the first Predicted temperature for each data point It is the first The squared error of each data point It is the weighted mean square error. It is an operation for solving minimization problems.

[0030] In one possible implementation, it also includes: The convolutional layer is selected as the target layer according to the greenhouse temperature prediction model, and the matrix of the output feature map is obtained. The gradient is calculated by backpropagation based on the predicted value and the feature map to obtain the gradient matrix. The global average value of the gradient of each feature map of the target layer is calculated along the spatial dimension based on the gradient matrix, and the weight vector is obtained. The weight vector of each feature map of the target layer and the feature map matrix are weighted and superimposed channel by channel to obtain the activation region matrix. The activation region matrix is ​​normalized to obtain the normalized heatmap.

[0031] In one possible implementation, the method further includes: performing backpropagation processing based on the predicted value and the original input data to obtain a gradient mapping value; performing element-wise multiplication of the gradient mapping value and the normalized heatmap to obtain an optimized heatmap; and performing weighted fusion of the optimized heatmap and the original image to obtain a greenhouse temperature correlation map, wherein the greenhouse temperature correlation map is used to characterize the degree of correlation between temperature and various factors.

[0032] In one possible implementation, the construction of the heat transfer model includes the following steps: constructing a greenhouse geometric model based on greenhouse parameters, wherein the greenhouse geometric model is divided into finite element meshes, and the nodes are connected by element stiffness matrices, mass matrices, and damping matrices; the greenhouse parameters include the shape parameters, size parameters, structural parameters, and material parameters of the greenhouse, as well as the crop parameters, soil parameters, and moisture parameters within the greenhouse; based on the geometric model, constructing a nonlinear heat conduction model, and outputting the heat conduction model parameters and temperature field; establishing a heat convection model using a multi-scale coordination method based on the parameters of the greenhouse geometric model and the temperature field, and outputting heat convection parameters, wherein the heat convection parameters include air velocity field, pressure field, and heat convection source term; and forming a heat transfer model based on the parameters of the greenhouse geometric model, the heat conduction model parameters, and the heat convection parameters.

[0033] In one possible implementation, a thermal convection model is established using a multi-scale coordination method based on the parameters of the greenhouse geometry and the temperature field, according to the following formula: in, , It's pressure. It is dynamic viscosity. It is the coefficient of thermal expansion. It is gravitational acceleration. This is a reference temperature. It is a unit vector in the vertical direction. It is the porosity coefficient of the greenhouse. It is the thermal diffusivity. It is a heat convection source term.

[0034] It can be seen that the present invention aims to at least solve one of the technical problems existing in the prior art or related technologies, such as low efficiency and accuracy of greenhouse temperature prediction, weak insulation and heat storage capacity of greenhouses, high energy consumption, and short production season. It optimizes and improves greenhouse temperature prediction methods based on physical models or data models by comprehensively using technologies such as hybrid model linkage architecture, multi-scale coordination method, spatiotemporal feature extraction method, adaptive weight generation method, and visualization optimization method. This overcomes the shortcomings of existing technical solutions, improves the efficiency and accuracy of greenhouse temperature prediction, and enhances the insulation and heat storage capacity of greenhouses, reduces energy consumption, and extends the production season.

[0035] The greenhouse temperature control method provided in the embodiments of this application will be further described in detail below.

[0036] This method combines the advantages of heat transfer models and machine learning models, utilizing advanced techniques such as multi-scale coordination, spatiotemporal feature extraction, adaptive weight generation, and visualization optimization to effectively improve the insulation and heat storage capacity of greenhouses, reduce energy consumption, and extend the production season. It provides an efficient, accurate, flexible, and interpretable solution for greenhouse temperature prediction. Specifically, this method includes the following five steps: S1: A heat transfer model is established using the nonlinear finite element method and the multi-scale coordination method. It considers factors such as multi-source heat storage, flexible structure, frequency conversion control of geothermal heat, porosity and non-uniformity of the greenhouse, describes the heat transfer process inside and outside the greenhouse, provides physical constraints and prior knowledge, and provides a basis for greenhouse temperature prediction.

[0037] S2: A machine learning model is established using multi-scale convolutional neural networks (MSCNN) and spatiotemporal feature extraction methods. Leveraging their powerful feature extraction and expression capabilities, spatiotemporal related features are extracted from historical greenhouse data to capture the complex and nonlinear relationship between greenhouse temperature and various factors.

[0038] S3: Optimize the machine learning model using Gaussian Mixture Model (GMM) and adaptive weight generation method. Based on different input data and output targets, cluster the data using Gaussian Mixture Model. Based on different clustering results, adaptively generate different weight vectors to dynamically adjust the contribution weights of each factor to greenhouse temperature prediction, thereby enhancing the model's generalization and transfer capabilities.

[0039] S4: The machine learning model is visualized using Grad-weighted class activation mapping (Grad-CAM) and backpropagation visualization optimization methods. By visualizing the activation regions and gradient information of the model, the Grad-CAM results are optimized to eliminate noise and artifacts, and to intuitively show the correlation strength and influence direction between greenhouse temperature and various factors, thereby improving the interpretability and credibility of the model.

[0040] S5: By utilizing the outputs of the heat transfer model and the machine learning model, combined with physical constraints and data fitting, a hybrid model is formed to achieve the prediction and control of greenhouse temperature.

[0041] The following is a detailed explanation of each of the above steps: The first aspect: Constructing a heat transfer model.

[0042] A heat transfer model is established using the nonlinear finite element method and a multi-scale coordination method, considering factors such as multi-source heat storage, flexible structure, frequency conversion control of underground heat, porosity, and non-uniformity of the greenhouse, to describe the heat transfer process inside and outside the greenhouse. This step provides physical constraints and prior knowledge to lay the foundation for greenhouse temperature prediction. The nonlinear finite element method solves complex nonlinear partial differential equations and is suitable for the nonlinear heat conduction problem in greenhouses. The multi-scale coordination method models the heat convection problem of the greenhouse at different scales, considering the greenhouse's porosity and non-uniformity, thus improving the model's accuracy and stability. The technical solution of this invention is: A greenhouse heat transfer model is established using the nonlinear finite element method and a multi-scale coordination method, including the following steps: Step 1.1: Establish the geometric model of the greenhouse, including parameters such as its shape, size, structure, and materials, as well as parameters such as crops, soil, and moisture within the greenhouse. The geometric model of the greenhouse is generated using a finite element mesh, with each mesh element represented by a node. The connection relationships between nodes are described by the element stiffness matrix and the element mass matrix. The geometric model of the greenhouse is expressed by the following formula: Where M is the total mass matrix, C is the total damping matrix, K is the total stiffness matrix, u is the nodal displacement vector, F is the nodal load vector, and u̇ and ü are the nodal velocity vector and nodal acceleration vector, respectively.

[0043] Step 1.2: Establish a heat conduction model for the greenhouse, described by nonlinear partial differential equations. Consider factors such as the greenhouse's nonlinear heat conduction coefficient, heat capacity, heat radiation, and heat convection, as well as boundary conditions including temperature, humidity, wind speed, and solar radiation inside and outside the greenhouse. The greenhouse heat conduction model is expressed by the following formula: Where ρ is density, c is specific heat capacity, T is temperature, t is time, k is thermal conductivity, q is the heat source term, ∇ is the gradient operator, and ⋅ is the dot product operation.

[0044] Step 1.3: Establish a thermal convection model for the greenhouse. This model is described using a multi-scale coordinated method, considering the greenhouse's porosity and non-uniformity, as well as factors such as airflow, humidity changes, and water vapor condensation within the greenhouse. The greenhouse thermal convection model is expressed by the following formula: in, , It's pressure. It is dynamic viscosity. It is the coefficient of thermal expansion. It is gravitational acceleration. This is a reference temperature. It is a unit vector in the vertical direction. It is the porosity coefficient of the greenhouse. It is the thermal diffusivity. It is a heat convection source term.

[0045] Step 1.4: Couple the greenhouse's geometric model, heat conduction model, and heat convection model to obtain the greenhouse heat transfer model. This model describes the heat transfer process inside and outside the greenhouse, providing physical constraints and prior knowledge for greenhouse temperature prediction. The greenhouse heat transfer model is expressed by the following formula: The meanings of each symbol are the same as those described above.

[0046] The second aspect: learning spatiotemporal features.

[0047] A machine learning model is built using a multi-scale convolutional neural network (MSCNN) and spatiotemporal feature extraction methods. Leveraging its powerful feature extraction and representation capabilities, this model extracts spatiotemporally relevant features from historical greenhouse data, capturing the complex and nonlinear relationships between greenhouse temperature and various factors. This step provides data fitting and prediction capabilities, supporting greenhouse temperature forecasting. The MSCNN deep learning model utilizes multi-scale convolutional kernels and pooling layers to extract multi-level features from the data, enhancing the model's expressive and predictive abilities. The spatiotemporal feature extraction method, a data processing technique, utilizes temporal and spatial information to extract spatiotemporally relevant features from historical greenhouse data, enhancing the model's input information and output quality.

[0048] A machine learning model is built using a multi-scale convolutional neural network (MSCNN) and spatiotemporal feature extraction methods. The model includes the following steps: Step 2.1: Extract spatiotemporal related features from historical greenhouse data using a spatiotemporal feature extraction method. This method includes the following steps: Step 2.1.1: Divide the historical greenhouse data according to the dimensions of time and space to obtain multiple spatiotemporal data blocks. Each spatiotemporal data block contains greenhouse data such as temperature, humidity, wind speed, and solar radiation within a certain spatial range over a specific time period. The spatiotemporal data block is represented by the following formula: in, It is a spatiotemporal data block of the i-th spatial region and the j-th temporal region. It is the data vector of the spatiotemporal data block at time t, where T is the time length of the spatiotemporal data block.

[0049] Step 2.1.2: Perform spatiotemporal feature extraction on each spatiotemporal data block. Utilizing temporal and spatial information, extract spatiotemporally relevant features from the spatiotemporal data block—such as the mean, variance, maximum value, minimum value, autocorrelation coefficient, cross-correlation coefficient, and spectral analysis. The spatiotemporal feature extraction is expressed by the following formula: in, F is the spatiotemporal feature vector of the spatiotemporal data block of the i-th spatial region and the j-th time region, and F is the spatiotemporal feature extraction function.

[0050] Step 2.1.3: Combine the extracted spatiotemporal feature vectors into a spatiotemporal feature matrix, which serves as the input to the machine learning model. This spatiotemporal feature matrix reflects the spatiotemporal correlation of historical greenhouse data, enhancing the model's input information and output quality. The spatiotemporal feature matrix is ​​expressed by the following formula: Where F is the spatiotemporal feature matrix, I is the spatiotemporal feature vector of the spatiotemporal data block of the i-th spatial region and the j-th time region, where I is the number of spatial regions and J is the number of time regions.

[0051] Step 2.2: Extract multi-level features from the spatiotemporal feature matrix using a multi-scale convolutional neural network (MSCNN). Leveraging its powerful feature extraction and representation capabilities, the network captures the complex and nonlinear relationships between greenhouse temperature and various factors. This network includes the following steps: Step 2.2.1: The spatiotemporal feature matrix is ​​used as the input layer of a multi-scale convolutional neural network (MSCNN). This input layer transforms the spatiotemporal feature matrix into a multi-channel two-dimensional tensor, where each channel corresponds to a spatiotemporal feature vector and each element corresponds to a spatiotemporal feature value. The input layer is represented by the following formula: Where I is the input layer and F is the spatiotemporal feature matrix.

[0052] Step 2.2.2: Set up multiple convolutional and pooling layers, each using convolutional and pooling kernels of different scales to extract features of different scales from the input layer or the output of the previous layer, enhancing the model's expressive and predictive capabilities. The convolutional and pooling layers are represented by the following formulas: in, It is the output of the l-th convolutional layer. It is the output of the l-th pooling layer, and σ is the activation function. It is the convolution kernel of the l-th convolutional layer. This represents the bias term of the l-th convolutional layer, * represents the convolution operation, and max represents the pooling operation. It is the input of the nth convolutional layer.

[0053] Step 2.2.3: Set up the fully connected layer and the output layer. The fully connected layer converts the output of the final pooling layer into a one-dimensional vector, and the output layer converts the output of the fully connected layer into a scalar, which is used as the predicted value of the greenhouse temperature. The fully connected layer and the output layer are represented by the following formula: in, It is the output of the fully connected layer. It is the output of the output layer. It is an activation function. It is the weight matrix of the fully connected layer. It is the bias vector of the fully connected layer. This is the output of the final pooling layer. It is the weight vector of the output layer. It is the bias scalar of the output layer.

[0054] Step 2.3: Using historical greenhouse data and actual temperatures as training data, train the Multi-Scale Convolutional Neural Network (MSCNN) to optimize network parameters, minimizing the error between the network's output and the actual temperature, thereby improving the network's fitting and predictive abilities. The training process is represented by the following formula: in, These are network parameters. It is the number of training data. It is the first The actual temperature of each training data point. It is the first Predicted temperature based on training data It is the first The squared error of each training data point It is the mean squared error of the training data. It is an operation for solving minimization problems.

[0055] Step 2.4: Using real-time greenhouse data as test data, test the multi-scale convolutional neural network (MSCNN) to obtain the predicted greenhouse temperature value, and compare it with the actual temperature to evaluate the network's predictive performance using the following metrics: 1. Root Mean Square Error (RMSE): Represents the square root of the error between the predicted and actual temperatures, reflecting the accuracy of the prediction; a smaller RMSE is better. RMSE is expressed by the following formula: 2. Correlation coefficient (R): This represents the degree of linear correlation between predicted and actual temperatures, reflecting the consistency of the predictions. A value closer to 1 is better. R is expressed by the following formula: 3. Deterministic coefficient (R) 2 R: represents the degree of variation between predicted and actual temperatures, reflecting the goodness of fit of the prediction; the closer to 1, the better. 2 Expressed using the following formula: The meanings of the symbols are the same as those described above. and These are the average of the actual temperature and the predicted temperature, respectively.

[0056] The third aspect: model optimization and weight adjustment.

[0057] This study optimizes a machine learning model using a Gaussian Mixture Model (GMM) and an adaptive weight generation method. Based on different input data and output targets, the GMM clusters the data, and adaptively generates different weight vectors according to the different clustering results. This dynamically adjusts the contribution weights of each factor to greenhouse temperature prediction, enhancing the model's generalization and transfer capabilities. This step improves the model's adaptability and flexibility, providing a guarantee for greenhouse temperature prediction. The Gaussian Mixture Model (GMM) is a probabilistic model that uses a Gaussian distribution to cluster data, suitable for multimodal and nonlinear data distributions in greenhouses. The adaptive weight generation method optimizes the model by adaptively generating different weight vectors based on the clustering results of the GMM, dynamically adjusting the contribution weights of each factor to greenhouse temperature prediction, and improving the model's flexibility and adaptability.

[0058] Optimizing a machine learning model using a Gaussian mixture model (GMM) and an adaptive weight generation method includes the following steps: Step 3.1: Cluster the input data using a Gaussian Mixture Model (GMM) and perform probabilistic modeling of the data using a Gaussian distribution. This method is suitable for multimodal and nonlinear data distributions in greenhouses and includes the following steps: Step 3.1.1: Assume the input data follows a Gaussian mixture distribution, meaning each data point consists of one or more Gaussian components, and each Gaussian component is represented by a mean vector, a covariance matrix, and mixing coefficients. The Gaussian mixture distribution is expressed by the following formula: in, It is the input data. It is the number of Gaussian components. It is the first The mixing coefficients of the Gaussian components satisfy the following conditions: , It is the first The probability density function of Gaussian components, It is the first The mean vector of Gaussian components, It is the first The covariance matrix of Gaussian components.

[0059] Step 3.1.2: Estimate the parameters of the Gaussian mixture model using the Expectation-Maximization (EM) algorithm, i.e., solve for the mean vector, covariance matrix, and mixing coefficients of the Gaussian components that maximize the log-likelihood function of the input data. The Expectation-Maximization (EM) algorithm includes the following steps: Step 3.1.2.1: Initialize the parameters of the Gaussian mixture model, i.e., randomly or according to some heuristic rules, give the initial values ​​of the mean vector, covariance matrix, and mixing coefficients of the Gaussian components, denoted as . , .

[0060] Step 3.1.2.2: Calculate the posterior probability of each data point belonging to each Gaussian component based on the current parameter values, i.e., the expected value of the latent variable. This posterior probability is expressed by the following formula: in, It is the first The data point belongs to the th data point The posterior probability of each Gaussian component It is the first The latent variables for each data point indicate which Gaussian component it belongs to. It is the first Observations for each data point It represents the number of iterations.

[0061] Step 3.1.2.3: Update the parameters of the Gaussian mixture model based on the current posterior probability, such that the mean vector, covariance matrix, and mixing coefficients of the Gaussian component that maximizes the expectation of the log-likelihood function of the input data with respect to the latent variables are obtained. This parameter update is expressed by the following formula: in, It is the first Updated values ​​of the mean vector, covariance matrix, and mixing coefficients of each Gaussian component. This is the number of input data.

[0062] Step 3.1.2.4: Determine if the convergence condition has been met, i.e., whether the increase in the log-likelihood function of the input data is less than a given threshold, or whether the number of iterations exceeds a given upper limit. If yes, stop the iteration and output the parameters of the Gaussian mixture model; otherwise, return to S1.2.2 and continue iterating. The log-likelihood function of the input data is expressed by the following formula: in, It is the input data. These are the parameters of the Gaussian mixture model. It is the number of Gaussian components.

[0063] Step 3.2: Optimize the machine learning model using an adaptive weight generation method. Based on different input data and output targets, utilize the clustering results of the Gaussian mixture model to adaptively generate different weight vectors, dynamically adjusting the contribution weights of each factor to greenhouse temperature prediction. This includes the following steps: Step 3.2.1: Based on the clustering results of the Gaussian mixture model, assign a label to each data point based on its Gaussian component. This label indicates which Gaussian component the data point is most likely to belong to. The label is expressed by the following formula: in, It is the first Labels of the Gaussian components of each data point. It is the first The data point belongs to the th data point The posterior probability of each Gaussian component.

[0064] Step 3.2.2: Assign a target value to each Gaussian component according to different output targets, representing the expected value of the greenhouse temperature corresponding to that Gaussian component. This target value is expressed by the following formula: in, It is the first The target value of each Gaussian component It is the objective function, which is set according to different output objectives. For example, the objective function for maximizing greenhouse temperature is set as follows: in, It is the first The actual temperature of each data point It is the first Labels for the Gaussian components of each data point.

[0065] Step 3.2.3: Based on the label and target value of the Gaussian component of each data point, generate a weight vector for each data point, which represents the contribution weight of that data point to the greenhouse temperature prediction, expressed by the following formula: in, It is the first The weight vector of each data point It is the first The target value of the Gaussian component to which each data point belongs. It is the first Labels for the Gaussian components of each data point.

[0066] Step 3.2.4: Optimize the machine learning model using the generated weight vector. Based on different input data and output targets, dynamically adjust the contribution weights of each factor to greenhouse temperature prediction to improve the model's generalization and transfer capabilities. The optimization process is expressed by the following formula: in, These are the parameters of the machine learning model. It is the number of input data. It is the first The weight vector of each data point It is the first The actual temperature of each data point It is the first Predicted temperature for each data point It is the first The squared error of each data point It is the weighted mean square error. It is an operation for solving minimization problems.

[0067] Fourth aspect: Visual model explanation.

[0068] This paper utilizes Gradient Weighted Class Activation Mapping (Grad-CAM) and backpropagation visualization optimization methods to visualize machine learning models. By visualizing the activation regions and gradient information of the model, the Grad-CAM results are optimized to eliminate noise and artifacts, intuitively demonstrating the strength and direction of the correlation between greenhouse temperature and various factors, thus improving the interpretability and reliability of the model. This step aims to enhance the transparency and reliability of the model, providing a basis for greenhouse temperature prediction. Gradient Weighted Class Activation Mapping (Grad-CAM) visualization technology uses the model's gradient information to generate the model's activation regions, expressing the model's focus and judgment criteria. The backpropagation visualization optimization method uses the gradient information from backpropagation to optimize the Grad-CAM results, eliminating noise and artifacts, and improving the model's visualization quality and interpretability.

[0069] Visualizing machine learning models using Gradient Weighted Class Activation Mapping (Grad-CAM) and backpropagation visualization optimization methods includes the following steps: Step 4.1: Visualize the machine learning model using Gradient Weighted Class Activation Mapping (Grad-CAM). Utilize the model's gradient information to generate activation regions, expressing the model's focus and judgment criteria. This includes the following steps: Step 4.1.1: Select the target layer, convolutional layer, or pooling layer as the object of visualization, and output the high-level features that reflect the model—such as spatial location, shape, and texture. The target layer is represented by the following formula: in, It is the output of the target layer. It is the first of the target layers. Each feature map It represents the number of feature maps in the target layer.

[0070] Step 4.1.2: Calculate the gradient of the target layer, which is the partial derivative of the target layer's output with respect to the predicted value of the output layer. This gradient reflects the contribution of each feature map in the target layer to the predicted value. The gradient is expressed by the following formula: in, It is the gradient of the target layer. It is the predicted value of the output layer. It is the first of the target layers. Each feature map It represents the number of feature maps in the target layer.

[0071] Step 4.1.3: Calculate the weights of the target layer, which are the global average gradients of each feature map in the target layer. These weights reflect the importance of each feature map in the target layer to the predicted value, and are expressed by the following formula: in, It is the first of the target layers. The weights of each feature map, It is the number of elements in the feature map. and These are the spatial coordinates of the feature map. It is the first feature map One element, It is the first feature map The gradient of each element.

[0072] Step 4.1.4: Calculate the activation region of the target layer, which is the element-wise summation of the product of the weights of each feature map in the target layer and the feature map itself. This reflects the model's focus and judgment criteria, and is expressed by the following formula: in, It is the activation region of the target layer. It is the first of the target layers. The weights of each feature map, It is the first of the target layers. Each feature map It represents the number of feature maps in the target layer.

[0073] Step 4.1.5: Post-process the activation regions of the target layer by setting the negative values ​​of the activation regions to zero and then normalizing them to obtain the gradient-weighted class activation map (Grad-CAM) of the target layer. This visually represents the activation regions of the model, expressed by the following formula: in, It is the gradient-weighted class activation mapping (Grad-CAM) of the target layer. It is the activation region of the target layer. This is an operation that sets the negative values ​​in the activated region to zero. It is the maximum value of the active region.

[0074] Step 4.2: Optimize the Grad-CAM results using a backpropagation visualization optimization method. This method utilizes gradient information from backpropagation to optimize the Grad-CAM results, eliminating noise and artifacts, and improving the model's visualization quality and interpretability. The steps include: Step 4.2.1: Calculate the gradient of the output layer. The partial derivative of the output layer's predicted value with respect to the input layer's data reflects the influence of each data point in the input layer on the predicted value, expressed by the following formula: in, It is the gradient of the output layer. It is the predicted value of the output layer. It is the first of the input layer Data points, It represents the number of data points in the input layer.

[0075] Step 4.2.2: Post-process the gradient of the output layer by setting the negative values ​​of the gradient to zero and then normalizing it to obtain the backpropagation gradient map (BP-Grad) of the output layer. This map reflects the direction of influence of each data point in the input layer on the predicted value. This map is expressed by the following formula: in, It is the gradient mapping (BP-Grad) of the backpropagation of the output layer. It is the gradient of the output layer. It is an operation that sets the negative value of the gradient to zero. It is the maximum value of the gradient.

[0076] Step 4.2.3: Fuse the results of Grad-CAM and BP-Grad to obtain the optimized gradient-weighted class activation map (Grad-CAM++). This map comprehensively considers the activation region and gradient information of the model, eliminates noise and artifacts, and improves the visualization quality and interpretability of the model. This map is expressed by the following formula: in, It is the optimized gradient-weighted class activation mapping (Grad-CAM++). It is the gradient-weighted class activation mapping (Grad-CAM) of the target layer. It is the gradient mapping (BP-Grad) of the backpropagation of the output layer. It is an element-wise multiplication operation.

[0077] Step 4.2.4: Overlay the optimized gradient-weighted class activation map (Grad-CAM++) with the input data to obtain a visualized greenhouse temperature prediction map. This map intuitively shows the correlation strength and direction of influence between greenhouse temperature and various factors, improving the interpretability and credibility of the model. The map is represented by the following formula: in, It is a visualized greenhouse temperature forecast map. It is the optimized gradient-weighted class activation mapping (Grad-CAM++). It is the input data. It's an operation that involves layering.

[0078] Fifth aspect: Hybrid model prediction.

[0079] By combining the outputs of heat transfer and machine learning models with physical constraints and data fitting, a hybrid model is formed to predict and control greenhouse temperatures. This step aims to leverage the advantages of both models to improve the efficiency and accuracy of greenhouse temperature prediction. The hybrid model integrates the outputs of heat transfer and machine learning models, along with physical constraints and data fitting, to form a unified greenhouse temperature prediction model, achieving efficient and accurate greenhouse temperature prediction and control.

[0080] By utilizing the outputs of the heat transfer model and the machine learning model, and combining physical constraints and data fitting, a hybrid model is formed, including the following steps: Step 5.1: Use a heat transfer model to make a physical prediction of greenhouse temperature. This model uses the greenhouse's geometric model, heat conduction model, and heat convection model to describe the heat transfer process inside and outside the greenhouse, providing physical constraints and prior knowledge for greenhouse temperature prediction. The model is expressed by the following formula: The meanings of each symbol are the same as those in S1 above.

[0081] Step 5.2: Utilize a machine learning model to predict greenhouse temperature data. This model employs a multi-scale convolutional neural network (MSCNN) and spatiotemporal feature extraction methods to extract spatiotemporally relevant features from historical greenhouse data. It captures the complex and nonlinear relationships between greenhouse temperature and various factors, providing data fitting and predictive capabilities for greenhouse temperature forecasting. The model is expressed by the following formula: in, It is a predicted value for greenhouse temperature. It is a spatiotemporal characteristic matrix. These are the parameters of the machine learning model. It is an activation function. It's a convolution operation. It is a pooling operation.

[0082] Step 5.3: Fuse the outputs of the heat transfer model and the machine learning model to obtain the output of the hybrid model, which is the final predicted value of the greenhouse temperature. This predicted value comprehensively considers physical constraints and data fitting, improving the efficiency and accuracy of greenhouse temperature prediction. This predicted value is expressed by the following formula: in, This is the final predicted value for greenhouse temperature. It is the output of the machine learning model. This is the output of the heat transfer model. These are adjustable weighting coefficients used to balance the influence of the output results of the two models.

[0083] Step 5.4: Use the output of the hybrid model to control the greenhouse. Based on the final predicted greenhouse temperature, adjust the control parameters such as multi-source heat storage and geothermal frequency conversion control to ensure the greenhouse temperature reaches the expected target value. This achieves the prediction and control of the greenhouse temperature. The control process is expressed by the following formula: in, These are the control parameters for the greenhouse. It is a control function. This is the final predicted value for greenhouse temperature. It is the target value for greenhouse temperature.

[0084] The hybrid model architecture provided in this application effectively integrates and coordinates the heat transfer model and the machine learning model to form a unified greenhouse temperature prediction model. This model reflects both the inherent laws and mechanisms of greenhouses and captures their dynamics and randomness, achieving the goals of efficient, accurate, flexible, and interpretable greenhouse temperature prediction. This method overcomes the limitations of traditional physical and data models, comprehensively utilizing the advantages of both to improve the quality and efficiency of greenhouse temperature prediction.

[0085] The adaptive weight generation method provided in this application utilizes a Gaussian mixture model to cluster data. Based on different clustering results and optimization objectives, it adaptively generates different weight vectors, dynamically adjusting the contribution weights of various factors to greenhouse temperature prediction, thereby enhancing the model's generalization and transfer capabilities. This method overcomes the shortcomings of traditional fixed weights or manual weight adjustment, achieving intelligent and automated weight generation, and improving the model's flexibility and adaptability.

[0086] The method provided in this application improves the insulation and heat storage capacity of greenhouses, reduces energy consumption, and extends the production season. This method utilizes multiple energy sources (solar energy, geothermal energy, biomass energy, etc.) for heat storage and uses flexible materials (polymers, fibers, etc.) to construct the greenhouse's frame and covering layer, enabling adaptive adjustment and intelligent control of the greenhouse. It has advantages such as good insulation performance, strong heat storage capacity, lightweight structure, high plasticity, low cost, and good environmental performance, effectively reducing greenhouse energy consumption and environmental pollution, extending the greenhouse's production season, and improving greenhouse production efficiency and economic benefits.

[0087] The method provided in this application improves the efficiency and accuracy of greenhouse temperature prediction, enhancing the basis and reliability of greenhouse temperature regulation. This method combines the advantages of heat transfer models and machine learning models, utilizing advanced techniques such as multi-scale coordination, spatiotemporal feature extraction, adaptive weight generation, and visualization optimization to effectively improve the efficiency and accuracy of greenhouse temperature prediction. This method fully considers the influence of various complex factors in greenhouses—such as multi-source heat storage, flexible structures, geothermal frequency conversion control, porosity, and non-uniformity—to improve the model's accuracy and stability. Simultaneously, it fully utilizes information from historical and future greenhouse data—spatiotemporally relevant features, dynamic and stochastic patterns, adaptive and optimized weights, etc.—to improve the model's data fitting and predictive capabilities. Furthermore, this method provides model visualization and optimization, activating regional and gradient information, correlation strength and influence direction, optimization objectives and control strategies, etc., improving the model's interpretability and credibility, and providing a more scientific and effective basis and guarantee for greenhouse temperature regulation.

[0088] In specific implementation, the embodiments of this application include, but are not limited to, three main steps: Phase 1: Data Acquisition

[0089] To implement this invention, the following types of data need to be obtained: Environmental data inside and outside the greenhouse, including indicators such as temperature, humidity, light intensity, wind speed, wind direction, precipitation, and carbon dioxide, reflect the heat transfer and crop growth conditions within the greenhouse. This data is collected through sensor devices installed inside and outside the greenhouse, such as temperature and humidity sensors, light intensity sensors, wind speed and direction sensors, rainfall sensors, and carbon dioxide sensors.

[0090] Greenhouse control data, including parameters for heat storage, ventilation, and irrigation, reflects the greenhouse's energy consumption and interventions in crop growth. This data is collected through controller devices installed within the greenhouse, such as heat storage controllers, ventilation controllers, and irrigation controllers.

[0091] Greenhouse crop data, including indicators such as crop variety, growth stage, and growth status, reflects the greenhouse's production efficiency and crop quality. This data is collected through manual recording or by using technologies such as image recognition, for example, to identify crop variety, crop growth stage, and crop diseases and pests.

[0092] In addition, some data is obtained through public access or self-collection: Public access via the Internet: Relevant data inside and outside the greenhouse, such as meteorological data, agricultural data, and market data, are obtained using the Internet or other public platforms as a reference or supplement for greenhouse temperature forecasting.

[0093] Phase Two: Data Processing.

[0094] To enable the acquired data to be used for model building and prediction, data processing is required, including structured data entry, data cleaning, data standardization, and data analysis. The following Python libraries and functions are needed: Structured data import utilizes the pandas library (https: / / pandas.pydata.org / ) and the sqlite3 library (https: / / docs.python.org / 3 / library / sqlite3.html) to convert data into DataFrame format and store it in an SQLite database for convenient subsequent data processing and querying. The following functions are used: `pandas.read_csv()` reads a CSV file and returns a DataFrame object; `pandas.to_sql()` writes the DataFrame object to the SQLite database, creating a table; `sqlite3.connect()` connects to the SQLite database and returns a database connection object; `sqlite3.Cursor()` creates a database cursor object for executing SQL statements.

[0095] Data cleaning utilizes the NumPy library (https: / / numpy.org / ) and the SciPy library (https: / / www.sciPy.org / ) to handle missing values, outliers, and duplicates, improving data quality and integrity. The following functions are used: `numpy.isnan()` checks if data contains missing values ​​(NaN) and returns a boolean array; `numpy.nanmean()` calculates the mean of the data, ignoring missing values; `numpy.nanfill()` fills missing values ​​with specified values; `sciPy.stats.zscore()` calculates the Z-score of the data, used to detect outliers; `sciPy.stats.iqr()` calculates the interquartile range (IQR) of the data, also used to detect outliers; and `pandas.drop_duplicates()` removes duplicate values ​​from the data and returns a new DataFrame object.

[0096] Data standardization utilizes the sklearn library (https: / / scikit-learn.org / ) to normalize or standardize data, eliminating the influence of dimensions and scales, and improving data comparability and stability. The following functions are used: `sklearn.preprocessing.MinMaxScaler()` creates a min-max normalization transformer object, converting data to values ​​between 0 and 1; `sklearn.preprocessing.StandardScaler()` creates a standardization transformer object, converting data to values ​​with a mean of 0 and a standard deviation of 1; `sklearn.preprocessing.fit_transform()` fits and transforms the data, returning a new array object.

[0097] Data analysis utilizes the matplotlib library (https: / / matplotlib.org / ) and the seaborn library (https: / / seaborn.pydata.org / ) to visualize and statistically analyze data, exploring its distribution, correlation, trends, and other characteristics, providing a basis for model construction and prediction. Use the following functions: `matplotlib.pyplot.figure()` to create a figure object for drawing charts; `matplotlib.pyplot.plot()` to draw a line chart to show data trends; `matplotlib.pyplot.scatter()` to draw a scatter plot to show data distribution and correlation; `matplotlib.pyplot.bar()` to draw a bar chart to show data frequency and proportion; `matplotlib.pyplot.hist()` to draw a histogram to show data distribution and density; `matplotlib.pyplot.boxplot()` to draw a box plot to show data quantiles and outliers; `seaborn.heatmap()` to draw a heatmap to show the correlation coefficient matrix of data; `seaborn.pairplot()` to draw a pairwise plot to show pairwise relationships in data; and `seaborn.regplot()` to draw a regression plot to show linear relationships and confidence intervals in data.

[0098] Phase 3: Model building.

[0099] To implement this invention, the following three model components need to be constructed: A heat transfer model is used to describe the heat transfer process inside and outside the greenhouse. It considers factors such as the greenhouse's multi-source heat storage, flexible structure, frequency conversion control of geothermal energy, porosity, and non-uniformity, providing physical constraints and prior knowledge to lay the foundation for greenhouse temperature prediction. The construction steps of this model are as follows: Step A: Using the nonlinear finite element method and the multi-scale coordination method, establish the heat conduction equation and heat convection equation for the greenhouse to describe the solid-state heat conduction and fluid heat convection of the greenhouse, respectively. Step B: Using boundary conditions and initial conditions, determine the thermal boundary and initial thermal state of the greenhouse, including the external ambient temperature, internal heat storage temperature, and geothermal temperature of the greenhouse; Step C: Using numerical methods, solve the heat conduction equation and heat convection equation of the greenhouse to obtain the heat field distribution and heat flow velocity distribution of the greenhouse, as well as the average temperature and maximum temperature of the greenhouse.

[0100] The heat transfer model is constructed and solved using a database, with the following specific configuration: Spatial discretization: triangular element mesh (global size 0.01) is used, with local refinement at the greenhouse interior-exterior interface (γ); Function space: second-order Lagrangian finite element (P2) method, supporting high-order gradient calculations of the temperature field; Boundary and initial conditions: Boundary: external ambient temperature 20℃, internal heat storage temperature 35℃, geothermal temperature 30℃; Initial conditions: greenhouse interior 25℃, exterior 20℃; Physical parameters (dynamically loaded from the database): Thermal conductivity coefficient: interior k1=0.5, exterior k1=0.5. k2=0.2; Thermal convection coefficient: internal c1=0.8, external c2=0.4; Heat capacity: internal p1=1.0, external p2=0.8; Density: internal ρ1=1.2, external ρ2=1.0; Wind speed: internal u1=0.5m / s, external u2=0.2m / s; The governing equations are bilinear in form, describing the transient heat conduction-convection equations (including conduction, convection, and transient terms); The weighting function and the unknown function belong to the same P2 space; Solver: Smooth Aggregate Algebraic Multigrid (SA-AMG) method, configuration parameters: solution accuracy 1e⁻ 6 The projection space is P2; the smoothing parameter ω = 0.67, the maximum number of iterations is 100; the interface γ forces the flux continuity condition (k1∇T1·n = k2∇T2·n).

[0101] A machine learning model is used to extract spatiotemporally relevant features from historical greenhouse data, capturing the complex and nonlinear relationships between greenhouse temperature and various factors, providing data fitting and prediction capabilities, and supporting greenhouse temperature forecasting. The model is constructed as follows: Step a: Utilize Multi-Scale Convolutional Neural Networks (MSCNN) and spatiotemporal feature extraction methods to establish the network structure of the machine learning model, including an input layer, multi-scale convolutional layers, pooling layers, fully connected layers, and an output layer. Step b: Use the mean squared error (MSE) as the loss function of the machine learning model to measure the error in greenhouse temperature prediction. Step c: Use stochastic gradient descent (SGD) as the optimization algorithm for the machine learning model to update the network parameters for greenhouse temperature prediction.

[0102] A machine learning model was constructed and trained using historical greenhouse data (including multi-dimensional data such as thermal field distribution, heat flow rate distribution, average temperature, maximum temperature, external ambient temperature, internal heat storage temperature, and geothermal temperature) stored in a database. The model input data dimension was (None, 24, 64, 64, 8), and was processed sequentially through the following network layers: First convolutional layer: using 32 (3,3) convolutional kernels, ReLU activation function, and Same padding to extract spatial features; Second convolutional layer: using 16 one-dimensional convolutional kernels of size 3 (temporal processing), ReLU activation function, and Same padding; Two-stage pooling layer: first, spatial downsampling with a (2,2) window size and stride of 2, then compression of the temporal dimension with a window size of 2 and stride of 2; Fully connected layer: 128 neurons working with the ReLU activation function to achieve feature fusion; Finally, the model outputs the future average temperature and maximum temperature of the greenhouse, with an output dimension of (None, 2). The model training configuration is as follows: Optimizer: Stochastic Gradient Descent (SGD) with a learning rate of 0.01 and momentum of 0.9; Loss function: Mean Squared Error (MSE); Training settings: Batch size 32, number of training epochs 100, validation set ratio 0.2; After training, the model can directly receive current greenhouse data (without additional prediction parameters) to predict future temperatures.

[0103] A hybrid model is used to comprehensively utilize the outputs of heat transfer and machine learning models, combining physical constraints and data fitting to form a unified greenhouse temperature prediction model, thereby improving the efficiency and accuracy of greenhouse temperature prediction. The construction steps of this model are as follows: Step 1: Optimize the machine learning model using a Gaussian Mixture Model (GMM) and an adaptive weight generation method. Based on different input data and output targets, cluster the data using the Gaussian Mixture Model. Adaptively generate different weight vectors based on different clustering results, dynamically adjusting the contribution weights of each factor to greenhouse temperature prediction, thereby enhancing the model's generalization and transfer capabilities. Step 2: Visualize the machine learning model using Grad-Weighted Class Activation Mapping (Grad-CAM) and backpropagation visualization optimization methods. By visualizing the activation regions and gradient information of the model, optimize the Grad-CAM results to eliminate noise and artifacts. This visually demonstrates the strength and direction of the correlation between greenhouse temperature and various factors, improving the model's interpretability and credibility. Step 3: Utilize the outputs of the heat transfer model and the machine learning model, combined with physical constraints and data fitting, to form a hybrid model for predicting and controlling greenhouse temperature.

[0104] A Gaussian Mixture Model (GMM) is constructed using historical greenhouse data stored in a database. The core configuration is as follows: Model structure: 3 Gaussian components, covariance type is full (complete covariance matrix) to capture complex data distributions; Training and prediction: Training data: historical greenhouse data (no additional training parameters required); Prediction data: current greenhouse data (no additional prediction parameters required); Gradient analysis: Calculate the gradient (Jacobi matrix) of the machine learning model output with respect to the input, with the gradient variable being the model input data; Output gradient tensor, and visualize the gradient values ​​through a heatmap, using a jet color mapping scheme; Mixture model output: Array 1: Physical simulation results of the heat transfer model; Array 2: Prediction results of the machine learning model; Dynamic prediction visualization: Horizontal axis (x-axis): Time series; Vertical axis (y-axis): Temperature values; Line 1: Weighted fusion result of the mixture model (formula: Ŷ = λ·Y_machine learning + (1-λ)·T_physical model); Line 2: Real observation values, used to compare prediction accuracy.

[0105] In summary, this application provides a method for temperature control in a flexible solar greenhouse. The method involves acquiring environmental data of the flexible solar greenhouse, including internal thermal field distribution data and historical time-series data of external ambient temperature. A heat transfer model is invoked to simulate and calculate the temperature field based on the internal thermal field distribution data, yielding a first predicted temperature value. A greenhouse temperature prediction model is then invoked to perform time-series analysis on the historical time-series data of the external ambient temperature, yielding a second predicted temperature value. The first and second predicted temperature values ​​are then fused to obtain a comprehensive predicted temperature value. Based on the deviation between the comprehensive predicted temperature value and a preset target greenhouse temperature value, the operating parameters of the greenhouse equipment are adjusted. By simulating the internal thermal field distribution and processing the time-series data of the external environment, both deterministic patterns and random fluctuations are covered. The deviation from the target value is calculated based on the obtained comprehensive predicted value, and the equipment is dynamically adjusted based on the predicted deviation, reducing ineffective energy consumption and equipment wear.

[0106] Based on the same technical concept, this application also provides a flexible solar greenhouse temperature control system, applied to greenhouse equipment, such as... Figure 2 As shown, the system includes: Data acquisition module 201 is used to acquire environmental data of flexible solar greenhouse; the environmental data of flexible solar greenhouse includes thermal field distribution data inside the greenhouse and historical time series data of external ambient temperature; The heat transfer model calculation module 202 is used to call the heat transfer model to perform temperature field simulation calculation based on the heat field distribution data inside the greenhouse, and obtain the first predicted temperature value. Temperature prediction module 203 is used to call the greenhouse temperature prediction model to perform time series analysis on the historical time series data of the external environment temperature to obtain a second predicted temperature value; The fusion module 204 is used to fuse data based on the first predicted temperature value and the second predicted temperature value to obtain a comprehensive predicted temperature value. The greenhouse equipment control module 205 is used to adjust the operating parameters of the greenhouse equipment based on the deviation between the comprehensive predicted temperature value and the preset greenhouse air temperature target value.

[0107] This application also provides an electronic device corresponding to the method provided in the foregoing embodiments. Please refer to... Figure 3 The diagram illustrates an electronic device provided by some embodiments of this application. The electronic device 70 may include: a processor 700, a memory 701, a bus 702, and a communication interface 703, wherein the processor 700, the communication interface 703, and the memory 701 are connected via the bus 702; the memory 701 stores a computer program that can run on the processor 700, and when the processor 700 runs the computer program, it executes the method provided by any of the foregoing embodiments of this application.

[0108] The memory 701 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one physical port (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0109] Bus 702 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The memory 701 is used to store programs. After receiving an execution instruction, the processor 700 executes the program. The method disclosed in any of the foregoing embodiments of this application can be applied to the processor 700, or implemented by the processor 700.

[0110] The processor 700 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 700 or by instructions in software form. The processor 700 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 701. Processor 700 reads the information in memory 701 and, in conjunction with its hardware, completes the steps of the above method.

[0111] The electronic devices and methods provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods they employ, operate, or implement.

[0112] This application also provides a computer-readable storage medium corresponding to the method provided in the foregoing embodiments. Please refer to... Figure 4 The computer-readable storage medium shown is an optical disc 80, on which a computer program (i.e., a program product) is stored, which, when run by a processor, performs the methods provided in any of the foregoing embodiments.

[0113] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0114] The computer-readable storage medium provided in the above embodiments of this application and the method provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

Claims

1. A temperature control method for a flexible solar greenhouse, applied to greenhouse equipment, characterized in that, The method includes: Acquire environmental data for a flexible solar greenhouse; the environmental data for the flexible solar greenhouse includes data on the distribution of the internal thermal field of the greenhouse and historical time-series data on the external ambient temperature. The heat transfer model is invoked to perform temperature field simulation calculations based on the thermal field distribution data inside the greenhouse, and the first predicted temperature value is obtained. The greenhouse temperature prediction model is invoked to perform time-series analysis on the historical time-series data of the external environment temperature to obtain a second predicted temperature value. A comprehensive predicted temperature value is obtained by fusing the first predicted temperature value and the second predicted temperature value. The operating parameters of the greenhouse equipment are adjusted based on the deviation between the comprehensive predicted temperature value and the preset greenhouse air temperature target value.

2. The method as described in claim 1, characterized in that, The greenhouse temperature prediction model is invoked to perform time-series analysis on the historical time-series data of the external environment temperature to obtain a second predicted temperature value, according to the following formula: in, This is the second predicted temperature value. It is historical time-series data of external environmental temperature. These are the parameters of the machine learning model. It is an activation function. Characterizes the convolution operation. Representation pooling operation.

3. The method as described in claim 1, characterized in that, The heat transfer model is invoked to perform temperature field simulation calculations based on the internal thermal field distribution data of the greenhouse, and the first predicted temperature value is obtained according to the following formula: Where M is the mass matrix, C is the damping matrix, K is the stiffness matrix, ρ is the material density, c is the specific heat capacity, and k is the thermal conductivity. It is the coefficient of thermal expansion, and T is the temperature. This is the reference temperature, and t is the time. It is gravitational acceleration. is a unit vector in the vertical direction, u is the nodal displacement vector, F is the nodal load vector, u̇ is the nodal velocity vector, q is the heat source term, and ∇ is the gradient operator. It is the resistance coefficient of the porous media in the greenhouse. It is the thermal diffusivity. It is a heat convection source term.

4. The method as described in claim 1, characterized in that, The construction of the greenhouse temperature prediction model includes the following steps: The historical time-series data of external ambient temperature is divided into spatiotemporal data blocks according to spatial region numbering and time region numbering; among them, spatial regions are divided according to sensor network topology, and time-series data are divided according to fixed time windows. For each spatiotemporal data block, extract spatiotemporal feature vectors and combine all spatiotemporal feature vectors into a spatiotemporal feature matrix; The predicted value is calculated based on the spatiotemporal feature matrix by calling a multi-scale convolutional neural network. The multi-scale convolutional neural network is trained using the predicted values ​​and the actual values ​​in the historical time series data of external environmental temperature as training data to obtain the greenhouse temperature prediction model.

5. The method as described in claim 4, characterized in that, Also includes: Gaussian mixture model is used to perform probabilistic modeling on the historical time series data of the external environment temperature to obtain Gaussian model parameters and posterior probability matrix; Assign Gaussian component labels to each data point based on the posterior probability matrix to obtain a label vector; Based on the label vector and the actual temperature data, target values ​​are assigned to each Gaussian component according to the label grouping, resulting in a target value vector. Generate a weight vector for each data point based on the label vector and the target value vector; The parameters of the greenhouse temperature prediction model are optimized based on the weight vector, the actual temperature data, and the parameters of the greenhouse temperature prediction model.

6. The method as described in claim 5, characterized in that, The step of using a Gaussian mixture model to probabilistically model the historical time-series data of the external environment temperature to obtain the Gaussian model parameters and the posterior probability matrix includes the following steps: Step 1: Assume the input data follows a Gaussian mixture distribution, where each data point comprises multiple Gaussian components. Each Gaussian component is represented by a mean vector, a covariance matrix, and mixing coefficients. The Gaussian mixture distribution is expressed by the following formula: ； in, It is the input data. It is the number of Gaussian components. It is the first The mixing coefficients of the Gaussian components satisfy the following conditions: , It is the first The probability density function of Gaussian components, It is the first The mean vector of Gaussian components, It is the first The covariance matrix of Gaussian components; Step 2: Initialize the parameters of the Gaussian mixture model, setting the initial values ​​for the mean vector, covariance matrix, and mixing coefficients of the Gaussian components, denoted as . , ; Step 3: Calculate the posterior probability of each data point belonging to each Gaussian component. The posterior probability matrix is ​​expressed by the following formula: ； in, It is the first The data point belongs to the th data point The posterior probability of each Gaussian component It is the first The latent variables for each data point indicate which Gaussian component it belongs to. It is the first Observations for each data point It is the number of iterations; Step 4: Update the parameters of the Gaussian mixture model based on the current posterior probability, such that the mean vector, covariance matrix, and mixing coefficients of the Gaussian component that maximizes the expectation of the log-likelihood function of the input data with respect to the latent variables are obtained. The parameter update is expressed by the following formula: ； ； ； in, It is the first Updated values ​​of the mean vector, covariance matrix, and mixing coefficients of each Gaussian component. It is the number of input data; Step 5: Determine whether the increase in the log-likelihood function of the input data is less than a given threshold, or whether the number of iterations exceeds a given upper limit. If so, stop the iteration and output the parameters of the Gaussian mixture model; otherwise, return to continue the iteration. The log-likelihood function of the input data is expressed by the following formula: in, It is the input data. These are the parameters of the Gaussian mixture model. It is the number of Gaussian components.

7. The method as described in claim 1, characterized in that, The construction of the heat transfer model includes the following steps: A greenhouse geometric model is constructed based on greenhouse parameters. The greenhouse geometric model is divided into finite element meshes, and the nodes are connected by element stiffness matrix, mass matrix, and damping matrix. The greenhouse parameters include the shape parameters, size parameters, structural parameters, material parameters of the greenhouse, as well as the crop parameters, soil parameters, and moisture parameters inside the greenhouse. Based on the geometric model, a nonlinear heat conduction model is constructed, and the heat conduction model parameters and temperature field are output. Based on the parameters of the greenhouse geometric model and the temperature field, a thermal convection model is established using a multi-scale coordination method, and thermal convection parameters are output, including air velocity field, pressure field, and thermal convection source term. A heat transfer model is formed based on the parameters of the greenhouse geometric model, the parameters of the heat conduction model, and the parameters of the heat convection model.

8. A flexible solar greenhouse temperature control system, applied to greenhouse equipment, characterized in that, The system includes: The data acquisition module is used to acquire environmental data of the flexible solar greenhouse; the environmental data of the flexible solar greenhouse includes the thermal field distribution data inside the greenhouse and the historical time series data of the external environment temperature; The heat transfer model calculation module is used to call the heat transfer model to perform temperature field simulation calculations based on the heat field distribution data inside the greenhouse, and obtain the first predicted temperature value. The temperature prediction module is used to call the greenhouse temperature prediction model to perform time-series analysis on the historical time-series data of the external environment temperature to obtain a second predicted temperature value. The fusion module is used to fuse the data based on the first predicted temperature value and the second predicted temperature value to obtain a comprehensive predicted temperature value. The greenhouse equipment control module is used to adjust the operating parameters of the greenhouse equipment based on the deviation between the comprehensive predicted temperature value and the preset greenhouse air temperature target value.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when running the computer program, performs an action to implement the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores computer-readable instructions that can be executed by a processor to implement the method as described in any one of claims 1-7.

Images