Intelligent temperature control method and system for planting winter jujubes based on greenhouse

Abstract

The invention provides an intelligent temperature control method and system for planting winter jujubes based on a greenhouse. The method comprises the following steps: collecting a dynamic environment data set of the winter jujubes; a greenhouse three-dimensional heat flow field digital twinborn model is constructed, and a winter jujube photosynthesis heat generation and soil heat conduction module is integrated; using an LSTM neural network model to generate temperature prediction data in future n hours, calculating an abnormal level in combination with a growth stress index, and generating a multi-level regulation and control instruction set; and formulating an optimal thermal environment regulation and control path according to the instruction set, and driving greenhouse temperature control equipment to regulate and control. According to the method, a data set is formed by fusing multi-modal data, a soil-root system composite heat transfer model is constructed in combination with a related model, and multi-physical domain interaction of a three-dimensional heat flow field digital twinborn model is accurately described; generating temperature prediction data through an LSTM model, and introducing a growth stress index to improve the precision; lSTM model parameters are calibrated by using an incremental learning algorithm, and triple optimization of temperature control precision, equipment energy consumption economy and crop growth safety is realized.

Description

Technical Field

[0001] This invention relates to the field of economic crop cultivation technology, and in particular to an intelligent temperature control method and system for growing winter jujubes in greenhouses. Background Technology

[0002] Temperature control technology in greenhouse jujube cultivation is gradually evolving from traditional manual, experience-based regulation to intelligent systems. Current technologies primarily employ single-dimensional data monitoring (such as temperature or humidity alone), combined with fixed-threshold trigger-based equipment control. Some advanced systems are beginning to integrate multi-sensor data acquisition and apply basic predictive models (such as the ARIMA time-series model) to predict temperature trends. Regarding model building, some studies attempt to establish simplified greenhouse heat flow models using CFD simulation software or to simulate crop photosynthetic heat effects using the Farquhar model, but these often focus on independent analyses of single physical fields or crop physiological processes. In actual production, temperature control equipment still relies mainly on "empirical parameters + manual intervention," lacking dynamic optimization strategies involving multi-device collaboration. Furthermore, feedback on environmental parameters after regulation is often used for simple threshold comparisons, failing to establish a closed-loop iterative mechanism for model parameters, resulting in low accuracy of predictive models. Summary of the Invention

[0003] This invention aims to at least solve the technical problems existing in the prior art, and in particular, it innovatively proposes an intelligent temperature control method and system for growing winter jujubes in greenhouses.

[0004] To achieve the above-mentioned objectives of this invention, this invention provides an intelligent temperature control method for growing winter jujubes in a greenhouse, the method comprising: S1. Collect multimodal data on the growth of winter jujubes to form a dynamic environment dataset; S2. Based on the dynamic environment dataset, construct a three-dimensional heat flow field digital twin model of the greenhouse, integrate the winter jujube photosynthetic heat generation module and the soil heat conduction module to form a dynamic simulation system for heat balance; S3. Based on the thermal balance dynamic simulation system, the LSTM neural network model is used to generate temperature prediction data for the next n hours. The abnormality level of the temperature prediction data is calculated by combining the growth stress index, and a multi-level control instruction set is generated. S4. Based on the multi-level control instruction set, formulate the optimal thermal environment control path; S5. Based on the optimal thermal environment control path, drive the temperature control equipment in the greenhouse to perform control actions; at the same time, collect the environmental parameters and physiological state data of winter jujubes after control in real time, and feed them back to the thermal balance dynamic simulation system to perform online iterative updates and parameter corrections on the LSTM neural network model.

[0005] On the other hand, the present invention also provides an intelligent temperature control system for growing winter jujubes in a greenhouse, the system comprising: processor; Memory used to store processor-executable instructions; The processor is configured to implement the intelligent temperature control method for growing winter jujubes in a greenhouse when executing the executable instructions.

[0006] The beneficial effects of this invention are as follows: This invention uses multimodal data fusion to form a dynamic environmental dataset, and combines the Farquhar model to simulate the photosynthetic heat effect of leaves, the heat diffusion equation and the Campbell thermal conductivity formula to construct a soil-root composite heat transfer model, realizing a three-dimensional heat flow field digital twin model to accurately characterize the interaction of multiple physical domains; it generates temperature prediction data for the next n hours through an LSTM neural network model, and innovatively introduces a growth stress index (integrating three-dimensional physiological indicators: photosynthetic rate decay rate, stomatal conductance change rate and fruit expansion rate) to calculate the temperature anomaly level, greatly improving the prediction accuracy; finally, through a closed-loop feedback mechanism of "prediction-regulation-feedback-update", it uses an incremental learning algorithm to realize online iterative calibration of LSTM model parameters, forming a triple optimization of temperature control accuracy, equipment energy consumption economy and crop growth safety, significantly improving the adaptive capability of the temperature control system for greenhouse winter jujube cultivation.

[0007] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0008] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of an intelligent temperature control method for growing winter jujubes in a greenhouse, according to the present invention. Detailed Implementation

[0009] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0010] Example 1 like Figure 1 As shown, an intelligent temperature control method for growing winter jujubes in a greenhouse is described, the method comprising: S1. Collect multimodal data on the growth of winter jujubes to form a dynamic environment dataset; In step S1, it is necessary to explain in detail that, in this embodiment, a multimodal sensor network is deployed in the greenhouse to collect data on the growth environment of winter jujubes, including leaf temperature data, soil temperature data, air temperature and humidity data, light intensity data, CO2 concentration data, and winter jujube fruit growth rate data, forming a dynamic environmental dataset. The multimodal sensor network specifically consists of temperature and humidity sensors, light sensors, carbon dioxide sensors, infrared thermometers, and digital calipers. Each sensor is interconnected through an IoT gateway, and edge computing nodes are used for preliminary data processing to reduce data transmission latency. Leaf temperature data is collected non-contactly using an infrared thermometer, soil temperature data is measured layer by layer using a buried temperature probe, air temperature and humidity data are collected synchronously with light intensity data, CO2 concentration data is monitored in real time through a gas analysis module, and fruit growth rate data is measured and recorded daily using digital calipers.

[0011] S2. Based on the dynamic environment dataset, construct a three-dimensional heat flow field digital twin model of the greenhouse, integrate the winter jujube photosynthetic heat generation module and the soil heat conduction module to form a dynamic simulation system for heat balance; S3. Based on the thermal balance dynamic simulation system, the LSTM neural network model is used to generate temperature prediction data for the next n hours. The abnormality level of the temperature prediction data is calculated by combining the growth stress index, and a multi-level control instruction set is generated. S4. Based on the multi-level control instruction set, formulate the optimal thermal environment control path; S5. Based on the optimal thermal environment control path, drive the temperature control equipment in the greenhouse to perform control actions; at the same time, collect the environmental parameters and physiological state data of winter jujubes after control in real time, and feed them back to the thermal balance dynamic simulation system to perform online iterative updates and parameter corrections on the LSTM neural network model.

[0012] In this embodiment, the principle of an intelligent temperature control method for growing winter jujubes in a greenhouse is as follows: First, multi-dimensional data on the growth environment and physiological state of winter jujubes are collected through a multimodal sensor network to construct a three-dimensional digital twin model of the heat flow field, including the photosynthetic heat effect of leaves and the soil-root composite heat transfer, thereby achieving accurate simulation of the multi-physical domain interaction process of the greenhouse thermal environment; second, an LSTM neural network model is used to predict future temperature changes based on historical and real-time data, and the growth stress index is calculated by integrating the photosynthetic rate decay rate, stomatal conductance change rate, and fruit expansion rate, thereby classifying the temperature anomaly levels and generating a multi-level control instruction set; then... By integrating control priorities and equipment constraints through a spatiotemporal weighting algorithm, a multi-objective joint optimization function is constructed. A genetic algorithm is used to solve for the Pareto optimal solution. Combined with a digital twin path planning module, a control path map is generated and verified through virtual debugging, outputting the optimal control path. Finally, the temperature control equipment is driven to execute control actions. Simultaneously, real-time environmental and physiological data are used for online iterative updates of the LSTM model, forming a fully closed-loop intelligent temperature control mechanism of "data acquisition - model simulation - predictive control - feedback optimization." This ensures that the winter jujube growing environment is always in an optimal thermal state, balancing temperature control accuracy, energy efficiency, and crop growth safety. This mechanism achieves deep integration of virtual and reality through digital twin technology, not only avoiding the stress of abnormal temperatures on winter jujube growth in advance but also significantly reducing the energy consumption cost of temperature control equipment. It provides an efficient and feasible technical solution for the intelligent and precise management of winter jujube cultivation in greenhouses.

[0013] As an optional embodiment of the present invention, optionally, in step S2, a three-dimensional heat flow field digital twin model of the greenhouse is constructed based on the dynamic environment dataset, integrating the winter jujube photosynthetic heat generation module and the soil heat conduction module to form a dynamic simulation system for heat balance, including: S201. Preprocess the dynamic environment dataset to generate a standardized environment parameter matrix; In step S201, it is necessary to explain in detail that the preprocessing of the dynamic environment dataset includes three core steps: missing value imputation, outlier filtering, and data standardization. First, for missing values ​​caused by sensor offline or signal interference, a categorical interpolation strategy is adopted. For time-continuous data such as air temperature and humidity, and leaf surface temperature, linear interpolation is used to imput missing points based on valid data within 5 minutes before and after the data point. For discrete physiological data such as fruit growth rate, the corresponding data of three adjacent jujube trees at the same growth stage are selected and weighted averaged to imput the missing values. Second, outlier filtering is performed. The 3σ principle is used to identify outlier data that exceeds the mean ± 3 standard deviations (such as nighttime light intensity being falsely reported as 10000 lux, or soil temperature exceeding the reasonable range of -5℃ to 50℃), and these outliers are replaced with the historical median of the parameter. Finally, data standardization was performed. To address the dimensional differences of different parameters, the min-max normalization method was used to map all data to the [0,1] interval. After standardization, a standardized environmental parameter matrix with dimensions M×N was generated, where M represents the number of sampling time points and N represents the total dimensions of multimodal parameters (e.g., 2-dimensional air temperature and humidity + 1-dimensional leaf temperature + 3-dimensional soil stratification temperature + 1-dimensional CO2 concentration + 1-dimensional fruit growth rate, for a total of 8 dimensions). Each row corresponds to all standardized parameters at a given time point.

[0014] S202. Based on the standardized environmental parameter matrix, construct a basic model of the three-dimensional thermal flow field of the greenhouse using a simulation platform; In step S202, it is necessary to explain in detail that the basic model is constructed using the COMSOL Multiphysics simulation platform. First, a geometric model is completed based on the actual CAD drawings of the greenhouse, including a main structure with a span of 8m, a length of 50m, and a ridge height of 3.5m. The covering layer uses PO agricultural film (thermal conductivity 0.042W / (m·K), light transmittance 92%). The interior includes a crop canopy area (height 1.2-1.8m) and installation points for temperature control equipment (top circulating fan, side wall water curtain / hot air furnace). Next, the mesh is discretized using a structured hexahedral mesh to divide the overall space, including the crop canopy, the air outlets of the temperature control equipment, and... The top 0-30cm soil layer was meshed (mesh size refined from 0.5m to 0.1m), with a total of approximately 1.2 million meshes. Initial boundary conditions were then set: the convective heat transfer coefficient of the top membrane layer was calculated using the Nusselt number formula based on real-time wind speed data (taken from a dynamic environment dataset), the ground boundary adopted a thermal conductivity boundary (soil thermal conductivity was taken as 1.1W / (m·K)), and natural ventilation openings were set on the side walls (the opening degree was linked to the external temperature and humidity). Finally, the air temperature, relative humidity, and light intensity in the standardized environmental parameter matrix were used as the initial input variables of the model. The completed three-dimensional heat flow field basic model can realize steady-state and transient simulation of the temperature distribution inside the greenhouse.

[0015] S203. Based on the three-dimensional heat flow field basic model, integrate the winter jujube photosynthetic heat generation module, use the Farquhar model to calculate the leaf-scale photosynthetic heat effect, and generate a leaf heat generation rate distribution map. In step S203, it is necessary to explain in detail that, firstly, real-time photosynthetically active radiation (PAR), atmospheric CO2 concentration, leaf temperature (value collected by infrared thermometer), and relative humidity are extracted from the dynamic environment dataset as the core driving variables of the Farquhar model; at the same time, jujube variety-specific parameters (such as the maximum carboxylation rate of Rubisco enzyme Vcmax and the maximum electron transport rate Jmax, which were calibrated to the full bloom period through previous field trials) are introduced to construct the leaf-scale net photosynthetic rate calculation equation. Secondly, the net photosynthetic rate Pn is solved through the three-limitation analysis of the Farquhar model (Rubisco carboxylation limitation, RuBP regeneration limitation, and triose phosphate utilization limitation), and the leaf heat production rate is derived based on the energy conversion relationship: about 95% of the absorbed light energy is converted into heat energy during photosynthesis, so the leaf heat production rate Qp=Pn×(1-η) / η (η is the photosynthetic efficiency, which is taken as 0.05±0.01 on average during the growth period of jujube). Next, the spatial expansion of heat generation rate at the canopy scale was achieved: the two-dimensional distribution of leaf area index (LAI) was calculated using the digital image analysis module (based on canopy RGB images from the dynamic environment dataset). Soil stratification temperature data was combined to correct for differences in leaf activity at different canopy heights. The heat generation rate of a single leaf was multiplied by the leaf area density (LAD) at the corresponding location to generate a three-dimensional heat generation rate distribution map of the canopy region. Finally, this distribution map was embedded as a volumetric heat source term into the energy conservation control equation of the three-dimensional heat flow field model. A user-defined field function from COMSOL Multiphysics was used to couple the photosynthetic heat effect with the greenhouse convection and conduction heat exchange processes. The time update step of the heat generation rate was kept consistent with the transient simulation step of the model (updated every 10 minutes) to ensure the dynamic accuracy of the model's contribution to canopy heat.

[0016] S204. Based on the leaf heat generation rate distribution map coupled with the soil heat conduction module, a soil-root composite heat transfer model is constructed using the heat diffusion equation and Campbell's thermal conductivity formula. The expression for the thermal diffusion equation is: , ; in, Indicates soil temperature, Indicates time, Indicates thermal diffusivity, Represents the Laplace operator. Indicates thermal conductivity. Indicates soil density, Indicates the specific heat capacity of the soil; Campbell's thermal conductivity formula: ; in, Indicates soil thermal conductivity. , , These are empirical coefficients and need to be calibrated using experimental data on soil types. For example, coefficients for different soil textures such as black loess, loess, and sandy loess need to be calibrated separately to ensure that the formula's prediction error is ≤10%. This indicates the volumetric water content of the soil, which is the percentage of water by volume in the soil pores. Indicates soil density; In step S204, it is necessary to explain in detail that, firstly, the soil area inside the greenhouse is stratified. Based on the stratified measurement data of the multimodal sensor network, the soil is divided into three functional layers: 0-10cm (top layer), 10-30cm (root active layer), and 30-60cm (deep layer). The soil texture parameters (such as density and specific heat capacity) of each layer are calibrated by physical property analysis of field samples (for example, the density of the top layer black soil is 1.35g / cm³, and the specific heat capacity is 0.85kJ / (kg·K); the density of the root active layer loess is 1.28g / cm³, and the specific heat capacity is 0.92kJ / (kg·K)). Secondly, the root respiration heat effect of jujube is introduced: based on the root biomass distribution data in the dynamic environment dataset (obtained in the previous micro-CT root scanning experiment), the root respiration heat generation rate is calculated by the formula Qr=Rroot×C, where Rroot is the root respiration rate (measured by the closed root chamber method, unit μmolCO2 / (g·h)) and C is the respiration heat conversion coefficient (value is 468kJ / molCO2); the root respiration heat is embedded as an additional heat source term into the corresponding layer of the heat diffusion equation to realize the composite heat transfer simulation of soil-root system.

[0017] Next, a coupling mechanism between the soil model and the three-dimensional heat flow field model is established: the temperature distribution results of each layer output by the soil-root composite heat transfer model are used as dynamic update terms of the ground boundary conditions of the three-dimensional heat flow field basic model, replacing the original static heat conduction boundary; at the same time, a feedback link between soil temperature and canopy thermal environment is constructed. When the temperature of the active root layer is higher than 28℃, the simulation weight of heat conduction from the soil to the canopy is enhanced, and vice versa, so as to realize the dynamic linkage of multi-domain thermal interaction.

[0018] Finally, online calibration of model parameters is performed: using real-time soil stratification temperature data collected by a multimodal sensor network, the deviation between measured values ​​and model predictions is fitted using the least squares method. The empirical coefficients a, b, and c in the Campbell thermal conductivity formula are updated every 24 hours (for example, after calibration, a=0.12, b=0.75, and c=0.03 for surface black soil), ensuring that the soil temperature prediction error is controlled within ±0.5℃. The simulation step size after model coupling is kept consistent with the overall thermal balance system (iteration every 10 minutes), ultimately forming a complete dynamic thermal balance simulation system that includes photosynthetic heat effect, root respiration heat, and soil stratification heat transfer, providing high-precision multi-physics domain interactive simulation support for subsequent temperature prediction and regulation.

[0019] S205. Perform multi-physics domain joint simulation on the soil-root composite heat transfer model to obtain the multi-physics domain joint simulation results. In step S205, it is necessary to explain in detail that, firstly, the real-time multi-dimensional data of the dynamic environment dataset (photosynthetically active radiation, atmospheric CO2 concentration, air temperature and humidity, soil volumetric water content, etc.) and the physiological state data of winter jujube (leaf temperature, leaf area index, root biomass distribution, etc.) are synchronously input into the coupled thermal balance dynamic simulation system to start the multi-physics domain joint simulation process; during the simulation, the system completes a full-domain thermal interaction calculation every 10 minutes, and sequentially executes the canopy photosynthetic heat generation rate update, soil-root composite heat transfer iteration, and greenhouse three-dimensional heat flow field transient simulation to realize the dynamic linkage of multiple physical processes such as photosynthetic heat effect, soil heat conduction, and air convection heat transfer. Secondly, real-time verification of simulation results is conducted: measured data such as canopy temperature, soil stratification temperature, and air temperature distribution collected by the multimodal sensor network are extracted and compared point-by-point with the corresponding parameters output by the simulation system, and the root mean square error (RMSE) and coefficient of determination (R²) are calculated. If the canopy temperature RMSE > 1.0℃ or the soil root layer temperature RMSE > 0.8℃, emergency calibration of model parameters is triggered (the least squares method in S204 is used to quickly update the Campbell coefficients) to ensure that the simulation accuracy meets the control requirements. Finally, multi-physics domain joint simulation results are output: including a three-dimensional spatial temperature field distribution map inside the greenhouse (resolution 0.1m×0.1m×0.1m), a two-dimensional distribution of canopy leaf heat generation rate, temperature variation curves of each soil stratum over time, and dynamic data of heat flow field and heat exchange flux in the soil domain. These results will serve as one of the input features of the LSTM neural network model in step S3, and at the same time provide a high-fidelity thermal environment simulation basis for the virtual debugging of the digital twin system.

[0020] S206. Based on the multi-physics domain joint simulation results, the root mean square error and coefficient of determination are used to quantitatively verify the prediction accuracy of the soil-root composite heat transfer model. If the verification results meet the preset accuracy threshold, the three-dimensional heat flow field basic model, the winter jujube photosynthetic heat generation module and the soil-root composite heat transfer model are deeply integrated to form a dynamic simulation system of heat balance that includes the interaction of multiple elements of environment, crop and soil. If the accuracy threshold is not met, the process is backtracked to steps S202 to S205, the corresponding coefficients are adjusted, and the model construction and simulation process is re-executed until the simulation results meet the preset accuracy requirements.

[0021] In step S206, it is necessary to explain in detail the specific values ​​of the preset accuracy thresholds. For key control parameters, stratified thresholds are set: the root mean square error (RMSE) between the simulated and measured values ​​of the canopy temperature must be ≤1.0℃ and the coefficient of determination (R²) ≥0.92; the RMSE of the soil root active layer (10-30cm) temperature must be ≤0.8℃ and R² ≥0.90; and the average RMSE of the greenhouse air temperature field must be ≤1.2℃ and R² ≥0.88. All parameters must simultaneously meet the threshold requirements for verification to be considered successful. Secondly, a quantitative verification process is executed: 72 consecutive hours of multimodal measured data and corresponding simulation results are selected. Data windows are divided hourly, and the RMSE and R² of each key parameter within each window are calculated. The average value of all windows is taken as the final verification index. If the average index of a certain parameter does not reach the threshold, it is marked as an error source. Next, the verification results are processed: if all key parameters meet the thresholds, the model is deeply integrated, and the core algorithms of the three-dimensional heat flow field basic model, the winter jujube photosynthetic heat generation module, and the soil-root composite heat transfer module are encapsulated into independent functional modules. Real-time data interaction between modules is realized through a data bus (such as the canopy heat generation rate output by the photosynthesis module being transmitted to the heat source term of the heat flow field model in real time, and the ground temperature output by the soil module updating the boundary conditions of the heat flow field in real time). At the same time, the simulation engine is optimized, and GPU acceleration is used to compress the time of a single full-domain simulation to 3-5 minutes to meet the real-time requirements of the intelligent temperature control system. If the verification fails, a targeted adjustment strategy is developed based on the source of the error: For example, if the soil root zone temperature error exceeds the standard, backtrack to step S204, increase the number of calibrated soil samples of the same texture (from the original 3 groups to 5 groups), and refit the empirical coefficient of the Campbell thermal conductivity formula; if the canopy temperature error exceeds the standard, backtrack to step S203, optimize the spatial interpolation algorithm for leaf area density (LAD), and introduce a light attenuation coefficient in the canopy height direction to correct for differences in leaf activity; if the air temperature field error exceeds the standard, backtrack to step S202, adjust the grid densification area (e.g., increase the grid density of the sidewall water curtain vents) or correct the Nusselt number calculation formula for the convective heat transfer coefficient. After the adjustments are completed, the entire process from S202 to S205 is re-executed until the verification indicators of all key parameters meet the preset thresholds.

[0022] As an optional embodiment of the present invention, optionally, the expression for calculating the photosynthetic heat effect at the leaf scale using the Farquhar model in step S203 is as follows: , ; in, This represents the photosynthetic heat effect at the leaf scale, that is, the heat flux density generated per unit area of ​​leaf through photosynthesis. Indicates the heat production coefficient, Indicates net photosynthetic rate, This represents the slope of the light response curve, i.e., the rate of change of the photosynthetic rate with light intensity. This represents the photon flux density, which is the amount of photons reaching the surface of the leaf. This represents the slope of the temperature response curve, i.e., the rate of change of photosynthetic rate with temperature. This indicates the optimal temperature for photosynthesis of winter jujubes, which is dynamically adjusted according to the growth period (24-26℃ during full bloom and 25-27℃ during fruit enlargement). Leaf temperature is indicated by real-time monitoring using an infrared thermal imager. This represents the curvature parameter of the temperature response curve, controlling the degree of nonlinearity of the temperature response curve.

[0023] As an optional embodiment of the present invention, optionally, in step S3, based on the thermal balance dynamic simulation system, temperature prediction data for the next n hours is generated using an LSTM neural network model, and the anomaly level of the temperature prediction data is calculated by combining the growth stress index, generating a multi-level control instruction set including: S301. Based on the aforementioned thermal balance dynamic simulation system, acquire historical data of greenhouse environment and monitoring data of winter jujube physiological state, and generate a structured training dataset using data standardization and feature extraction. In step S301, it is necessary to explain in detail that the thermal balance dynamic simulation system specifically collects environmental parameters and physiological state data of winter jujubes within the greenhouse through a multimodal sensor network, including air temperature, humidity, carbon dioxide concentration, light intensity, soil stratification temperature, canopy leaf temperature, and leaf area index. After preprocessing, this data is normalized using the Z-score standardization method to eliminate the influence of dimensional differences on model training. Subsequently, feature variables are extracted based on the time-series sliding window technique, such as the temperature change trend over the past 24 hours, the cumulative amount of photosynthetically active radiation, and the temperature fluctuation amplitude of the root active layer. The final generated structured training dataset contains input features across multiple dimensions, used for the subsequent training and validation of the LSTM neural network model.

[0024] S302. Based on the structured training dataset, a bidirectional LSTM neural network model is used to train a temperature prediction model. The model parameters are optimized through the backpropagation algorithm to generate a temperature prediction model with time series prediction capabilities. In step S302, it needs to be explained in detail that the construction of the bidirectional LSTM neural network model includes an input layer, hidden layers, and an output layer. The input layer receives multidimensional feature variables from the structured training dataset, such as temperature change trends and cumulative photosynthetically active radiation. The hidden layer uses bidirectional LSTM units to capture long-term dependencies in the time series data through forward and backward directions, thereby improving prediction accuracy. During model training, the backpropagation algorithm is used to iteratively optimize the weight parameters, the objective function is set as mean squared error (MSE), and the learning rate is dynamically adjusted using the Adam optimizer to accelerate convergence. In addition, to prevent overfitting, a dropout mechanism is introduced in the hidden layer to randomly discard some neuron connections, enhancing the model's generalization ability. The final temperature prediction model can output high-precision prediction results for the greenhouse air temperature, canopy leaf temperature, and soil stratification temperature for the next n hours, and supports real-time updates and dynamic adjustments.

[0025] Next, the anomaly level of the temperature prediction data was assessed using the growth stress index. The growth stress index is a comprehensive indicator calculated based on monitoring data of the physiological state of winter jujubes, used to reflect the potential impact of environmental conditions on crop growth. Specifically, the temperature prediction data was compared with the suitable temperature ranges for different growth stages of winter jujubes. If the predicted temperature exceeded the suitable range, anomalies were classified according to the degree of deviation. For example, when the canopy temperature exceeded 30℃ or fell below 15℃, it was marked as a Level 1 anomaly; when it exceeded 35℃ or fell below 10℃, it was marked as a Level 2 anomaly. Simultaneously, the anomaly level classification was further refined by combining the temperature fluctuation range of the active soil root layer and the distribution of air temperature and humidity. The final multi-level control instruction set included specific control strategies for different anomaly levels, such as adjusting the opening of greenhouse vents, activating the sprinkler cooling system, or increasing the power of heating equipment, to ensure that the winter jujube growing environment is always in optimal condition.

[0026] S303. Based on the temperature prediction model, generate a temperature prediction sequence data for the next n hours using real-time monitoring data of the greenhouse. The temperature prediction sequence data includes the temperature change trend, peak time node and fluctuation range. In step S303, the generation process of the temperature prediction sequence data specifically includes the following steps: First, real-time monitoring data from the greenhouse is input into a trained bidirectional LSTM neural network model. This data includes the current air temperature, canopy leaf temperature, soil stratification temperature, and environmental parameters such as humidity, light intensity, and carbon dioxide concentration. Based on time-series features learned from historical data, the model combines the current input data to output a temperature prediction sequence for the next n hours. Second, the prediction results are post-processed to extract key information, such as the overall trend of temperature change (rising, falling, or stable), peak time nodes (i.e., the specific times when the highest or lowest temperature may occur within the prediction period), and the temperature fluctuation range (the difference between the highest and lowest temperatures). Furthermore, to improve prediction accuracy, a Kalman filter algorithm is introduced when generating the temperature prediction sequence to dynamically correct the initial prediction results, reducing errors caused by sensor noise or short-term environmental disturbances. The final generated temperature prediction sequence data not only contains detailed numerical information but is also presented visually, such as plotting temperature change curves and marking peak time nodes and fluctuation ranges, facilitating quick understanding and response by operators.

[0027] S304. Based on the temperature prediction sequence data and the growth stress index, calculate the temperature anomaly level. The growth stress index is calculated by comprehensively considering the photosynthetic rate decay rate of winter jujube, the stomatal conductance change rate of leaves, and the fruit expansion rate. Finally, generate a multi-level control instruction set that includes temperature control priority, equipment start-up threshold, and control amplitude. The multi-level control instruction set is divided into three levels of instructions according to the temperature anomaly level: emergency control, routine control, and preventive control.

[0028] , ; , ; in, Indicates the photosynthetic rate decay rate. The baseline value for net photosynthetic rate under normal growth conditions of winter jujube needs to be calibrated through growth period experiments (such as the value calibrated at 24-26℃ during the full bloom period). The net photosynthetic rate, as measured in real time, is obtained through a photosynthesis measurement system. This represents the rate of change of porosity. This represents the reference value of stomatal conductance under normal growth conditions, calibrated using a leaf stomatal gauge. This indicates the porosity under real-time monitoring. Indicates the rate of change of fruit enlargement rate. This represents the daily growth rate of fruit diameter under normal growth conditions, measured and calibrated continuously using digital calipers. This indicates the daily growth rate of fruit diameter as monitored in real time. Indicates the growth stress index. , and The weighting coefficients are determined using the analytic hierarchy process (AHP). =0.5, =0.3, =0.2).

[0029] In step S304, the calculation process of the growth stress index specifically includes the following steps: First, based on real-time monitoring data of the photosynthetic rate decay rate, leaf stomatal conductance change rate, and fruit enlargement rate of jujube, normalization processing is performed using their respective benchmark values ​​to eliminate the influence of different dimensions on the comprehensive calculation. Second, according to preset weighting coefficients, the three normalized indicators are weighted and summed to obtain the final growth stress index. This index can dynamically reflect the comprehensive impact of environmental temperature changes on the physiological state of jujube. Next, combined with temperature prediction sequence data, the temperature anomaly level is determined through comparative analysis. For example, when the growth stress index exceeds 0.8, it is determined to be at the emergency control level; when it is between 0.5 and 0.8, it is determined to be at the routine control level; and when it is below 0.5 but there is potential risk, it is classified as at the preventive control level. Each control level corresponds to different equipment activation thresholds and control amplitudes. For example, in emergency control, the spray cooling system or heating equipment needs to be activated immediately, and the control range should be set to the maximum value; in routine control, the opening of the ventilation openings or the coverage ratio of the shading net should be adjusted according to the actual situation; and in preventive control, the focus is on fine-tuning environmental parameters to prevent the abnormal situation from deteriorating further. The final multi-level control instruction set not only includes specific control strategies but also clarifies the priority order, ensuring that the intelligent temperature control system can operate efficiently in complex environments and guarantee the optimal growth conditions for winter jujubes.

[0030] As an optional embodiment of the present invention, optionally, in step S4, formulating the optimal thermal environment control path based on the multi-level control instruction set includes: S401. Based on the multi-level control instruction set, extract the control priority matrix and the equipment parameter constraint set, use the spatiotemporal weight allocation algorithm to generate a spatiotemporal weight coefficient matrix containing three levels of instructions: emergency control, regular control, and preventive control, and integrate the power range of heating / cooling equipment, ventilation window opening limit, and shading net unfolding ratio to form the equipment parameter constraint set. The expression for the spatiotemporal weight allocation algorithm is: , ; ; in, express The time dimension weighting coefficient, with a value range of [0,1]. and Indicates the weighting coefficient ( + =1), calibrated using historical control effect data. Indicates the duration of the daily cycle. express The maximum deviation between the predicted temperature at any given time and the target temperature is obtained using an LSTM prediction model. express The spatial dimension weighting coefficient for coordinate points, with a value range of [0,1]. and Represents the weighting coefficient ( + =1), determined by crop distribution density. express The planting area for winter jujubes was obtained through a greenhouse layout map. This indicates the total area of ​​the greenhouse. express The Euclidean distance to the nearest temperature control device. This represents the spatiotemporal weighting coefficient matrix, where rows represent the time dimension (24 hours) and columns represent the three-level control instructions. This represents the Kronecker product operation, achieving spatiotemporal weight coupling. This represents the adjustment matrix for the control level. Indicates the first Level regulation in the first Correction coefficients for spatiotemporal units.

[0031] In step S401, it is necessary to explain in detail that the generation process of the control priority matrix is ​​based on the temperature anomaly level classification in the multi-level control instruction set. First, the three levels of instructions—emergency control, routine control, and preventive control—are mapped to corresponding time periods and spatial regions. For example, emergency control is usually targeted at periods within the prediction period where extreme temperature fluctuations may occur, while preventive control is more often applied when the current environmental parameters are close to the boundary of the suitable range but have not yet reached the abnormal standard. The construction of the equipment parameter constraint set comprehensively considers actual operating conditions such as the power range of heating / cooling equipment, ventilation window opening limits, and the deployment ratio of shading nets. These constraints are transformed into quantifiable parameter ranges through mathematical modeling methods to ensure that the generated control path can be executed within the equipment's capabilities.

[0032] In the spatiotemporal weighting algorithm, the calculation of the time dimension weight coefficient not only relies on the deviation between the temperature prediction data and the target temperature, but also incorporates historical control effect data for dynamic adjustment. For example, within a daily cycle, if the temperature deviation is large at a certain moment and historical data shows that the control effect during that period is significant, a higher time weight is assigned. The calculation of the spatial dimension weight coefficient fully considers the influence of crop distribution density and temperature control equipment layout inside the greenhouse. Specifically, areas closer to the temperature control equipment have relatively lower spatial weights due to their faster response speed, while areas farther away from the equipment are assigned higher weights due to increased control difficulty. The spatiotemporal weight coupling achieved through Kronecker product operations can effectively integrate information from both time and space dimensions, thereby generating a more refined control strategy.

[0033] The introduction of the regulation level correction matrix further enhances the adaptability of the intelligent temperature control system. The setting of the correction coefficient comprehensively considers the sensitivity of jujube to environmental temperature at different growth stages and the real-time monitored growth stress index. For example, during the fruit enlargement stage, because jujube is more sensitive to temperature changes, the correction coefficient will be increased accordingly to strengthen the regulation. The final spatiotemporal weight coefficient matrix not only contains the specific execution parameters of each level of regulation command, but also clarifies the priority order within each spatiotemporal unit.

[0034] S402. Based on the spatiotemporal weight coefficient matrix and the set of equipment parameter constraints, a set of optimization functions is constructed using a multi-objective optimization method. A joint optimization objective is formed by three sub-objective functions: minimizing temperature deviation, minimizing equipment energy consumption, and decaying growth stress exponential. In step S402, the construction process of the optimization function set includes the following steps: First, with the goal of minimizing temperature deviation, a first sub-objective function is defined. This function quantifies the degree of temperature deviation from the ideal state during the control process by calculating the sum of squares of the differences between the predicted temperature and the target temperature. Second, for minimizing equipment energy consumption, a second sub-objective function is designed. This function comprehensively considers the power consumption of heating and cooling equipment, as well as the operating energy consumption of mechanical devices such as ventilation windows and shading nets, and converts the energy consumption of various equipment into comparable numerical indicators through weighted summation. Finally, the decay of the growth stress index is introduced as a third sub-objective function, aiming to effectively reduce the level of physiological stress caused by abnormal environmental temperature through control measures. This function dynamically adjusts its weight in the joint optimization objective based on the real-time monitored rate of change of the growth stress index and the sensitivity of winter jujube to temperature at different growth stages. The three sub-objective functions are linearly combined after normalization to form the final joint optimization objective. To ensure the practicality of the optimization results, constraints were introduced when constructing the optimization function set. These constraints included hard constraints such as power range and opening limit from the equipment parameter constraint set, as well as soft constraints such as the suitable temperature range for jujube growth. These constraints were embedded into the optimization model using the penalty function method to avoid generating control paths that do not meet actual needs.

[0035] S403. Based on the optimization function set, a genetic algorithm is used to solve for the Pareto optimal solution, and an elite retention strategy and an adaptive crossover mutation operator are used to generate a non-dominated solution set. In step S403, it is necessary to explain in detail that the execution process of the genetic algorithm is specifically divided into five stages: initialization, selection, crossover, mutation, and termination condition judgment. First, in the initialization stage, a set of initial solutions satisfying the equipment parameter constraints is randomly generated as individuals in the population, each representing a possible thermal environment control path. Second, the performance of each individual is evaluated using a fitness function. The fitness value is calculated by the joint optimization objective and comprehensively reflects the performance of temperature deviation, equipment energy consumption, and growth stress index. In the selection stage, roulette wheel or tournament selection methods are used to select individuals with higher fitness to enter the next generation, while retaining some low-fitness individuals to maintain population diversity. In the crossover operation, an adaptive crossover operator is used to dynamically adjust the crossover probability, generating new individuals based on the current population distribution characteristics, ensuring a balance between search efficiency and global exploration capability. The mutation operation introduces random perturbations through an adaptive mutation operator to avoid the algorithm getting trapped in local optima. The elite retention strategy plays a crucial role throughout the process; after each iteration, the best individual in the current population is directly retained to the next generation, thus ensuring that the quality of the solution does not degrade. Finally, when the preset termination conditions are met, such as reaching the maximum number of iterations or the fitness value converges, the algorithm outputs a set of non-dominated solutions.

[0036] S404. Based on the non-dominated solution set, a thermal environment control path map is generated using a digital twin path planning module. The thermal environment control path map includes equipment control sequences in the time dimension, equipment coordination strategies in the spatial dimension, and energy consumption distribution curves in the energy dimension. In step S404, it is necessary to explain in detail that the operation mechanism of the digital twin path planning module is based on multiple feasible control schemes in the non-dominated solution set. A high-precision simulation model is used to dynamically simulate and evaluate the performance of each scheme in three dimensions: time, space, and energy. Specifically, in the time dimension, the module generates equipment control sequences based on temperature prediction sequence data, specifying the start and stop times and operating durations of heating, cooling, ventilation, and shading equipment. In the spatial dimension, combined with the layout of temperature control equipment inside the greenhouse and the crop distribution density, a coordinated equipment strategy is formulated to ensure that the control effect can evenly cover the entire planting area. In the energy dimension, through refined modeling of equipment energy consumption, an energy consumption distribution curve is plotted, intuitively displaying the energy consumption at each time period and its contribution to the overall control target. Furthermore, this module supports real-time interactive functions. Users can adjust key parameters, such as the target temperature range or equipment priority, through a visual interface to quickly obtain the updated control path map. The final generated thermal environment control path map is shown.

[0037] S405. Use a digital twin verification platform to virtually debug the thermal environment control path map. Verify the feasibility of the path by comparing the simulation results with the objective function threshold. If the verification is successful, output the optimal thermal environment control path. If the verification fails, backtrack to S402 to adjust the weight coefficients of the optimization function and solve it again.

[0038] In step S405, it is necessary to explain in detail that the virtual debugging process relies on the high-precision simulation capabilities of the digital twin verification platform. By inputting the thermal environment control path map into the platform, the temperature changes, equipment operating status, and energy consumption distribution within the actual greenhouse are simulated. First, the platform sets threshold ranges for various indicators of the objective function, such as the maximum allowable fluctuation range of temperature deviation, the upper limit of equipment energy consumption, and the decrease range of the growth stress index. Subsequently, by running the simulation model, dynamic performance data of the control path in the time, space, and energy dimensions are obtained and compared with preset thresholds. If the simulation results meet the requirements of all objective functions, the path is deemed feasible and output as the optimal thermal environment control path; if the verification fails, it is necessary to backtrack to step S402 and readjust the weight coefficients of each sub-objective in the optimization function to balance the relationship between temperature control, energy consumption management, and growth stress mitigation. In this process, the weight coefficient adjustment strategy combines historical control data and real-time monitoring information to ensure that the new solution direction is closer to actual needs. Furthermore, to improve verification efficiency, the platform supports parallel simulation of multiple schemes. By simultaneously testing the control paths corresponding to multiple non-dominated solution sets, the most suitable scheme can be quickly selected. This closed-loop verification mechanism enhances the reliability of the control path.

[0039] As an optional embodiment of the present invention, optionally, the expressions for the three sub-objective functions in step S402 are: , , ; in, This represents the target for temperature deviation optimization; a smaller value indicates more precise temperature control. Indicates the length of time. express Predicted temperature at any time express The target temperature trajectory is dynamically set according to the growth period (e.g., 24-26℃ during peak flowering). This represents the temperature baseline value, taken as the historical average temperature for the same period, used for normalization. This represents the energy consumption optimization target for the equipment; the smaller the value, the lower the energy consumption. Indicates the number of equipment categories. This represents the baseline energy consumption value, taken as the historical average energy consumption for the same period, used for normalization. Indicates the first Rated power of this type of equipment Indicates the first Operating time of such equipment This represents the target for attenuating the growth stress index; a higher value indicates a better stress relief effect. This indicates the initial growth stress index before regulation. This indicates the predicted growth stress index after regulation, which is predicted through digital twin simulation.

[0040] As an optional embodiment of the present invention, optionally, in step S5, based on the optimal thermal environment control path, the temperature control equipment in the greenhouse is driven to perform control actions; simultaneously, environmental parameters and physiological state data of winter jujubes after control are collected in real time and fed back to the thermal balance dynamic simulation system, and the online iterative update and parameter correction of the LSTM neural network model includes: S501. Based on the time-dimensional device control sequence in the optimal thermal environment control path, a precise control instruction set including device start-up time, operating power, and opening degree adjustment ratio is generated using the temperature control device drive module. In step S501, it is necessary to explain in detail that the temperature control device drive module generates specific control instruction sets based on the time-dimensional device control sequence in the optimal thermal environment control path. These instruction sets contain detailed parameters such as the start-up time, operating power, and opening adjustment ratio of each device to ensure that the control actions can be executed accurately. For example, for heating equipment, the instruction set will specify its operating power and start / stop times within a specific time period; for ventilation windows, it will be refined to the opening percentage and adjustment frequency. In addition, the drive module also optimizes the instructions based on the actual operating characteristics of the equipment to avoid equipment damage caused by frequent start-stops or power fluctuations. In this way, not only is the accuracy of control improved, but the service life of the equipment is also extended.

[0041] S502. Based on the precise control instruction set, use an IoT sensor array to collect environmental parameters after control in real time and generate a dynamic environmental feedback dataset. In step S502, it is necessary to explain in detail that the IoT sensor array is deployed at various key locations within the greenhouse to monitor environmental parameters in real time after regulation. These sensors include temperature sensors, humidity sensors, light intensity sensors, and carbon dioxide concentration sensors, which can comprehensively capture environmental changes inside the greenhouse. The collected data is transmitted to the data processing center via wireless communication technology to form a dynamic environmental feedback dataset. This dataset not only records the trends of environmental parameter changes over different time periods but also incorporates spatial distribution information to reflect the differences in regulation effects in different areas. To improve data reliability, the sensor array also has a self-calibration function, automatically correcting measurement errors by comparing with standard reference values. In addition, the dynamic environmental feedback dataset also integrates external meteorological data, such as outside temperature, wind speed, and rainfall, providing more comprehensive input conditions for subsequent regulation optimization.

[0042] S503. Based on the dynamic environment feedback dataset, use digital image analysis methods to synchronously collect physiological state data of winter jujube and generate a physiological state feedback dataset. In step S503, it is necessary to explain in detail that the digital image analysis method uses high-definition cameras and infrared imaging devices deployed inside the greenhouse to monitor the physiological state of jujube plants non-contactly. These devices can capture key indicators such as leaf color changes, fruit surface temperature distribution, and overall plant morphological characteristics, thereby generating a physiological state feedback dataset. For example, the degree of leaf yellowing can be quantified using image segmentation algorithms to assess the nutritional status of the jujubes; abnormal fluctuations in fruit surface temperature are detected using infrared thermal imaging technology, reflecting their adaptability to changes in environmental temperature. Furthermore, a deep learning-based image recognition model can further extract subtle features of plant growth stress, such as leaf curling or delayed fruit development, and convert them into quantifiable physiological state parameters.

[0043] S504. Based on the dynamic environment feedback dataset and physiological state feedback dataset, the LSTM neural network model is updated online iteratively. The model weights are optimized through incremental learning algorithm to generate an updated temperature prediction model.

[0044] In step S504, it is necessary to explain in detail that the online iterative update process of the LSTM neural network model relies on an incremental learning algorithm. Through joint analysis of the dynamic environmental feedback dataset and the physiological state feedback dataset, the model weights are continuously optimized. Specifically, the incremental learning algorithm first preprocesses the new data, removing outliers and noise interference to ensure the quality of the input data. Then, the algorithm inputs this data into the LSTM model in batches, adjusting the model's internal parameters through backpropagation to better fit the latest trends in environmental and physiological state changes. During this process, to avoid overfitting due to frequent updates, regularization constraints, such as L2 norm penalty terms, are introduced to limit the range of magnitude changes in the weight parameters. Furthermore, to improve the model's generalization ability, a sliding window technique is used, mixing historical and new data proportionally during training, thus quickly adapting to short-term fluctuations while retaining long-term memory. The final temperature prediction model not only has higher prediction accuracy but also reflects the complex coupling relationship between the environment and physiological state during the growth of winter jujubes in real time.

[0045] Example 2 A smart temperature control system for growing winter jujubes in greenhouses includes: processor; Memory used to store processor-executable instructions; The processor is configured to implement an intelligent temperature control method for growing winter jujubes in greenhouses when executing executable instructions.

[0046] It should be noted that the computer device includes a processor, a memory, and may also include one or more of a multimedia component, an input / output (I / O) interface, and a communication component.

[0047] The processor controls the overall operation of the computer device to complete all or part of the steps in the above-mentioned intelligent temperature control method for growing winter jujubes in greenhouses.

[0048] Memory is used to store various types of data to support the operation of the computer device. This data may include, for example, instructions for any application or method used to operate on the computer device, as well as application-related data. Memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0049] The multimedia component may include a screen and an audio component, wherein the screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals; for example, the audio component may include a microphone for receiving external audio signals, the received audio signals may be further stored in memory or transmitted via a communication component; the audio component may also include at least one speaker for outputting audio signals.

[0050] I / O interfaces provide interfaces between the processor and other interface modules, such as keyboards, mice, buttons, etc.; these buttons can be virtual buttons or physical buttons.

[0051] The communication component is used for wired or wireless communication between the computer device and other devices; wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G or 5G, or one or more combinations thereof, and the corresponding communication component may include: Wi-Fi module, Bluetooth module, NFC module, mobile communication module.

[0052] As a preferred embodiment, the computer device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the above-described intelligent temperature control method for growing winter jujubes in greenhouses.

[0053] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A smart temperature control method for growing winter jujubes in a greenhouse, characterized in that, The method includes: Collect multimodal data on the growth of winter jujubes to form a dynamic environment dataset; Based on the dynamic environment dataset, a three-dimensional digital twin model of the greenhouse heat flow field is constructed, integrating the winter jujube photosynthetic heat generation module and the soil heat conduction module to form a dynamic simulation system for heat balance. Based on the aforementioned thermal balance dynamic simulation system, the LSTM neural network model is used to generate temperature prediction data for the next n hours. The abnormality level of the temperature prediction data is calculated by combining the growth stress index, and a multi-level control instruction set is generated. Based on the multi-level control instruction set, the optimal thermal environment control path is formulated; Based on the optimal thermal environment control path, the temperature control equipment in the greenhouse is driven to perform control actions; at the same time, the environmental parameters and physiological state data of winter jujubes after control are collected in real time and fed back to the thermal balance dynamic simulation system to perform online iterative updates and parameter corrections on the LSTM neural network model.

2. The intelligent temperature control method for growing winter jujubes in a greenhouse as described in claim 1, characterized in that, The formation of a dynamic thermal equilibrium simulation system includes: The dynamic environment dataset is preprocessed to generate a standardized environment parameter matrix; Based on the standardized environmental parameter matrix, a basic model of the three-dimensional heat flow field of the greenhouse is constructed using a simulation platform. Based on the three-dimensional heat flow field basic model, the winter jujube photosynthetic heat generation module is integrated, and the Farquhar model is used to calculate the leaf-scale photosynthetic heat effect and generate a leaf heat generation rate distribution map. Based on the leaf heat generation rate distribution map coupled with the soil heat conduction module, a soil-root composite heat transfer model is constructed using the heat diffusion equation and Campbell's thermal conductivity formula. Multi-physics domain co-simulation was performed on the soil-root composite heat transfer model to obtain the multi-physics domain co-simulation results; Based on the results of multi-physics domain joint simulation, the root mean square error and coefficient of determination are used to quantitatively verify the prediction accuracy of the soil-root composite heat transfer model. If the verification results meet the preset accuracy threshold, the three-dimensional heat flow field basic model, the winter jujube photosynthetic heat generation module and the soil-root composite heat transfer model are deeply integrated to form a dynamic simulation system of heat balance that includes the interaction of multiple elements of environment, crop and soil.

3. The intelligent temperature control method for growing winter jujubes in a greenhouse as described in claim 2, characterized in that, The expression for calculating the photosynthetic heat effect at the leaf scale using the Farquhar model is as follows: ， ； in, This represents the photosynthetic heat effect at the leaf scale. Indicates the heat production coefficient, Indicates net photosynthetic rate, This represents the slope of the light response curve. Represents the quantum flux density of light. This represents the slope of the temperature response curve. This indicates the optimal temperature for photosynthesis in winter jujubes. Indicates leaf temperature, This represents the curvature parameter of the temperature response curve.

4. The intelligent temperature control method for growing winter jujubes in a greenhouse as described in claim 1, characterized in that, The generation of multi-level control instruction sets includes: Based on the aforementioned thermal balance dynamic simulation system, historical data of the greenhouse environment and monitoring data of the physiological state of winter jujubes are obtained, and a structured training dataset is generated by using data standardization and feature extraction. Based on the structured training dataset, a bidirectional LSTM neural network model is used to train a temperature prediction model. The model parameters are optimized through the backpropagation algorithm to generate a temperature prediction model with time series prediction capabilities. Based on the temperature prediction model, real-time monitoring data of the greenhouse is used to generate a temperature prediction sequence data for the next n hours. The temperature prediction sequence data includes the temperature change trend, peak time node and fluctuation range. The temperature anomaly level is calculated based on the temperature prediction sequence data and the growth stress index. The growth stress index is calculated by comprehensively considering the photosynthetic rate decay rate of jujube, the stomatal conductance change rate of leaves, and the fruit enlargement rate. Finally, a multi-level control instruction set is generated, which includes temperature control priority, equipment start-up threshold, and control amplitude. The multi-level control instruction set is divided into three levels of instructions according to the temperature anomaly level: emergency control, routine control, and preventive control.

5. The intelligent temperature control method for growing winter jujubes in a greenhouse as described in claim 4, characterized in that, The expression for calculating the growth stress index is: ； ； ； ； in, Indicates the photosynthetic rate decay rate. This represents the baseline value of net photosynthetic rate under normal growth conditions for winter jujubes. This represents the net photosynthetic rate as monitored in real time. Indicates the rate of change of porosity. This represents the baseline value of stomatal conductance under normal growth conditions. This indicates the porosity under real-time monitoring. Indicates the rate of change of fruit enlargement rate. This represents the daily growth rate of fruit diameter under normal growth conditions. This indicates the daily growth rate of fruit diameter as monitored in real time. Indicates the growth stress index. , and This represents the weighting coefficient.

6. The intelligent temperature control method for growing winter jujubes in a greenhouse as described in claim 1, characterized in that, Determining the optimal thermal environment control path includes: Based on the multi-level control instruction set, the control priority matrix and equipment parameter constraint set are extracted. The spatiotemporal weight allocation algorithm is used to generate a spatiotemporal weight coefficient matrix containing three levels of instructions: emergency control, regular control, and preventive control. The power range of heating / cooling equipment, ventilation window opening limit, and shading net deployment ratio are integrated to form the equipment parameter constraint set. Based on the spatiotemporal weight coefficient matrix and the set of equipment parameter constraints, a set of optimization functions is constructed using a multi-objective optimization method. The joint optimization objective is formed by three sub-objective functions: minimizing temperature deviation, minimizing equipment energy consumption, and decaying growth stress exponential. Based on the aforementioned set of optimization functions, a genetic algorithm is used to find Pareto optimal solutions, and an elite retention strategy and an adaptive crossover mutation operator are used to generate a non-dominated solution set. Based on the non-dominated solution set, a thermal environment control path map is generated using a digital twin path planning module. The thermal environment control path map includes equipment control sequences in the time dimension, equipment coordination strategies in the spatial dimension, and energy consumption distribution curves in the energy dimension. The thermal environment control path map is virtually debugged using a digital twin verification platform. The feasibility of the path is verified by comparing the simulation results with the objective function threshold. If the verification is successful, the optimal thermal environment control path is output.

7. The intelligent temperature control method for growing winter jujubes in a greenhouse as described in claim 6, characterized in that, The expression for the spatiotemporal weight allocation algorithm is: ； ； ； in, express Time dimension weighting coefficient and Indicates the weighting coefficient. Indicates the duration of the daily cycle. express The maximum deviation between the predicted temperature data and the target temperature at any given time. express Spatial dimension weighting coefficient of coordinate points and Represents the weighting coefficients. express The area planted with winter jujubes This indicates the total area of ​​the greenhouse. express The Euclidean distance to the nearest temperature control device. Represents the spatiotemporal weighting coefficient matrix. This represents the Kronecker product operation. This represents the adjustment matrix for the control level. Indicates the first Level regulation in the first Correction coefficients for spatiotemporal units.

8. The intelligent temperature control method for growing winter jujubes in a greenhouse as described in claim 6, characterized in that, The expressions for the three sub-objective functions are: ； ； ； in, This represents the target for temperature deviation optimization; a smaller value indicates more precise temperature control. Indicates the length of time. express Predicted temperature at any time express The target temperature trajectory at any time. Indicates the temperature reference value. Indicates the equipment energy consumption optimization target. Indicates the number of equipment categories. This represents the baseline value for energy consumption. Indicates the first Rated power of this type of equipment Indicates the first Operating time of such equipment This indicates the optimization target for the decline of the growth stress index. This represents the initial growth stress index before regulation. This indicates the predicted growth stress index after regulation.

9. The intelligent temperature control method for growing winter jujubes in a greenhouse as described in claim 1, characterized in that, Online iterative updates and parameter correction of LSTM neural network models include: Based on the time-dimensional device control sequence in the optimal thermal environment control path, a precise control instruction set including device start-up time, operating power, and opening degree adjustment ratio is generated using the temperature control device drive module. Based on the precise control instruction set, an IoT sensor array is used to collect environmental parameters in real time after control and generate a dynamic environmental feedback dataset. Based on the dynamic environmental feedback dataset, digital image analysis methods are used to synchronously collect physiological state data of winter jujubes to generate a physiological state feedback dataset. Based on the dynamic environmental feedback dataset and the physiological state feedback dataset, the LSTM neural network model is updated online iteratively. The model weights are optimized through incremental learning algorithm to generate an updated temperature prediction model.

10. An intelligent temperature control system for growing winter jujubes in a greenhouse, characterized in that, The system includes: processor; Memory used to store processor-executable instructions; The processor is configured to implement, when executing the executable instructions, the intelligent temperature control method for growing winter jujubes in a greenhouse, as described in any one of claims 1 to 9.

Images