Plant factory planting automation management and control system

Abstract

This application discloses an automated management and control system for plant factory cultivation. The system includes: a data acquisition module, a digital twin model construction module, a growth state anomaly prediction module, and a strategy generation module. The data acquisition module collects and preprocesses environmental and phenotypic parameters. The digital twin model construction module forms a dynamic digital twin that can be iterated in real time based on these parameters and outputs plant growth state prediction results. The growth state anomaly prediction module compares key features, growth state prediction results, and ideal growth states, and outputs anomaly prediction signals. The strategy generation module generates management and control strategies. This application achieves intelligent management and control of the entire plant factory process, improves cultivation efficiency and control accuracy, and provides technical support for large-scale, intelligent cultivation.

Description

Technical Field

[0001] This application relates to the fields of smart agriculture and plant factory environmental control technology, and in particular to an automated management and control system for plant factory planting. Background Technology

[0002] Plant factories, as a modern agricultural production method with fully controllable environment, achieve efficient and year-round crop production by regulating environmental factors such as light, temperature, humidity, CO2 concentration, and nutrient solution. Among these factors, the synergistic regulation of multiple environmental factors is a key link in the stable operation of plant factories.

[0003] Currently, plant factories suffer from the following shortcomings in environmental control: reliance on fixed-parameter PID control makes it difficult to adapt to the dynamic needs of different crop growth stages, resulting in limited control precision; a lack of real-time prediction and early warning capabilities for plant growth processes prevents true closed-loop optimization control; and in complex scenarios such as vertical planting and multi-layered cultivation, traditional systems struggle to achieve zoned control of different environmental areas. To address these issues, we propose an automated plant factory planting management system capable of full-process automation, self-learning, and self-adaptive capabilities, aiming to improve the production efficiency, resource utilization, and intelligence level of plant factories. Summary of the Invention

[0004] The main purpose of this application is to provide an automated management and control system for plant factory cultivation, which aims to solve the technical problems of traditional plant factories such as insufficient environmental control precision, poor adaptability, and lack of growth prediction and closed-loop optimization capabilities. By constructing an automated management and control system for plant factory cultivation with digital twin as the core, the system can realize full-process automation and production of plant factories.

[0005] To achieve the above objectives, this application provides an automated management and control system for plant factory cultivation, which includes a data acquisition module, a digital twin model construction module, a growth status anomaly prediction module, and a strategy generation module.

[0006] Data acquisition module: This module is used to collect and preprocess plant growth environment parameters and phenotypic parameters. It includes a phenotypic data acquisition unit, an environmental data acquisition unit, and a data preprocessing unit. The phenotypic data acquisition unit collects data including at least plant height and multi-angle RGB images of the canopy. The environmental data acquisition unit collects data including at least light intensity, temperature, humidity, and CO2 concentration. The data preprocessing unit cleans the collected environmental and phenotypic parameters to remove outliers and noise, interpolates and fills in missing data, and normalizes data of different dimensions to ensure that the parameters are within a similar numerical range.

[0007] The digital twin model construction module is used to construct a digital twin model that reflects the real-time operating conditions of the plant factory and outputs future prediction results based on preprocessed core environmental and phenotypic parameters of plant growth using Spatiotemporal Graphical Convolutional Network (ST-GCN) technology. The specific steps are as follows: The spatiotemporal data construction unit integrates data to construct a plant growth data structure containing spatiotemporal information; the ST-GCN model training unit trains the model using this data structure, enabling it to learn the spatiotemporal characteristics of growth; the dynamic digital twin generation unit fuses real-time data to construct dynamic twins for representative plants, accurately simulating the growth process. In addition, the module also includes a 3D reconstruction and image segmentation auxiliary unit, a feature fusion unit, a prediction result output unit, and a trend extrapolation and calibration unit. These units respectively achieve spatial information supplementation, feature optimization, outputting prediction results with specific scalar indicators, and trend extrapolation and online model calibration for the next 3-7 days, ensuring prediction accuracy.

[0008] The abnormal growth state prediction module compares and comprehensively analyzes the predicted future growth results of plants with the ideal growth curves of the corresponding growth cycles, and outputs anomaly prediction signals. The module comprises four main units: an ideal growth curve storage unit with pre-set curves categorized by crop variety, plotted with time on the horizontal axis and four key growth indicators on the vertical axis, based on extensive standard environmental experimental data; a prediction result comparison unit that compares each predicted scalar indicator with the ideal curve, considering both numerical magnitude and trend differences analyzed using spatiotemporal characteristics captured by the ST-GCN model; an anomaly severity assessment unit that quantifies the degree to which the growth state deviates from the ideal state based on the comparison results, providing accurate judgment criteria; and an anomaly warning signal output unit that outputs tiered warning signals when values ​​exceed preset ranges and the deviation reaches the target level for multiple consecutive prediction cycles, providing support for the strategy generation module to formulate targeted control strategies.

[0009] Strategy generation module: Used to receive anomaly prediction signals and generate control strategies. When generating control strategies, it comprehensively considers three dimensions: plant growth rate, biomass accumulation, and resource utilization efficiency, and comprehensively evaluates the current growth status in order to generate more scientific and reasonable control strategies. Attached Figure Description

[0010] Figure 1 A block diagram of an automated control system for plant factory cultivation provided in one embodiment of this application; Figure 2 A flowchart of an automated control system for plant factory cultivation provided in one embodiment of this application; The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the automated control system for plant factory cultivation proposed in this application will be described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art will understand that the technical details and modifications in the following embodiments can achieve the technical solutions claimed in this application without departing from the technical concept of this application.

[0012] The automated management and control system for plant factory cultivation provided in this application collects and preprocesses plant growth environment parameters and phenotypic parameters through a data acquisition module. It then uses a digital twin model construction module to construct a dynamic digital twin based on spatiotemporal graph convolutional network technology to accurately predict the future growth of plants. The system further compares and analyzes the predicted results with the ideal growth curve through a growth anomaly prediction module, outputting anomaly prediction signals. Finally, the strategy generation module combines plant growth adaptation rules to generate scientific and reasonable management and control strategies, realizing intelligent management and control of the entire plant factory process, improving planting efficiency and control accuracy, and providing technical support for large-scale and intelligent planting.

[0013] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various implementation methods of this application will be described in detail below in conjunction with specific implementation scenarios. However, those skilled in the art will understand that many technical details have been presented in the various implementation methods of this application to enable readers to better understand this application. However, even without these technical details and various changes and modifications based on the following implementation methods, the technical solutions claimed in this application can be implemented.

[0014] One embodiment of this application relates to an automated management and control system for plant factory cultivation, referring to... Figure 1 Specifically, it includes: a data acquisition module, a digital twin model construction module, a growth state anomaly prediction module, and a strategy generation module. The data acquisition module communicates with the digital twin model construction module, the digital twin model construction module communicates with the growth state anomaly prediction module, and the growth state anomaly prediction module communicates with the strategy generation module. These modules work collaboratively in sequence to form a complete control process.

[0015] The data acquisition module is used to collect plant growth environment parameters and phenotypic parameters and perform preprocessing, then sends the preprocessed parameter data to the digital twin model construction module. This module includes a phenotypic data acquisition unit, an environmental data acquisition unit, and a data preprocessing unit. The phenotypic data acquisition unit is used to collect at least the plant height and canopy multi-angle RGB image phenotypic parameters. The environmental data acquisition unit is used to collect at least the light intensity, temperature, humidity, and CO2 concentration environmental parameters. The data preprocessing unit is used to clean, imput missing values, and normalize the collected parameters.

[0016] In one example, the environmental data acquisition unit uses a distributed sensor array deployed in the plant factory planting area. The sensor array includes at least a light intensity sensor, a temperature sensor, a humidity sensor, and a CO2 concentration sensor. The sensor acquisition frequency is set to 1Hz, and the acquired data is transmitted to the data preprocessing unit in real time via an RS485 bus.

[0017] In one example, the phenotypic data acquisition unit uses a combination of 3D LiDAR and high-definition camera. The 3D LiDAR scanning frequency is set to 1 scan per 5 minutes to collect plant phenotypic parameters; the high-definition camera is used to capture multi-angle RGB images of the canopy, and is triggered synchronously with the LiDAR acquisition to ensure spatiotemporal consistency of the data.

[0018] In a specific example, the data preprocessing unit performs multi-step preprocessing operations on the collected environmental and phenotypic parameters: First, data cleaning is performed, using the 3σ principle to remove outliers and noise. For environmental parameters, the mean μ and standard deviation σ of each parameter are calculated, and data exceeding the range [μ-3σ, μ+3σ] are identified as outliers and removed. Next, missing data imputation is performed. For a small number of missing data points with no more than 3 consecutive missing data points, linear interpolation is used for imputation. For missing environmental parameters with temporal correlation, the missing value is calculated by linear fitting of valid data at adjacent time points. For missing phenotypic parameters, interpolation is performed based on the consistency characteristics of plant growth in the same batch. Finally, normalization is performed, using the Min-Max normalization method to standardize data of different dimensions, mapping all parameters to the [0,1] interval, so that the parameters are in a similar numerical range, eliminating the influence of dimensional differences on subsequent model training. The normalization formula is: , in, These are the normalized eigenvalues. These are the original eigenvalues. and These are the historical minimum and maximum values ​​of this feature, respectively.

[0019] Reference Figure 2 The digital twin model construction module is used to construct a digital twin model based on preprocessed core environmental parameters and phenotypic parameters of plant growth. This model reflects the real-time operating conditions of the plant factory and outputs predictions of future plant growth. This module employs ST-GCN technology and specifically includes a spatiotemporal data construction unit, an ST-GCN model training unit, a dynamic digital twin generation unit, a 3D reconstruction and image segmentation auxiliary unit, a feature fusion unit, a prediction result output unit, and a trend extrapolation and calibration unit.

[0020] In one example, the spatiotemporal data construction unit builds a spatiotemporal data structure for plant growth that incorporates both temporal and spatial information, based on collected phenotypic and environmental data. Spatially, a graph structure is constructed using individual plants as nodes. Node features are feature vectors derived from the fusion of preprocessed environmental and phenotypic parameters. Edge weights are set based on the spatial distance between plants; the closer the plants are, the greater the edge weight, thus capturing the mutual influence of adjacent plant growth. Temporally, with a 5-minute time step, data from 12 consecutive time steps are combined into a time-series sequence, forming a spatiotemporally fused data structure for model training and prediction.

[0021] In one example, the ST-GCN model training unit uses a constructed spatiotemporal data structure to train the ST-GCN model, enabling it to learn the spatiotemporal characteristics of plant growth. The spatiotemporal data is divided into training, validation, and test sets in a 7:2:1 ratio. The Adam optimizer is used with an initial learning rate of 0.001, decaying by 50% every 200 epochs, and mean squared error (MSE) as the loss function. Training is iteratively stopped when the validation set loss shows no decrease for 30 consecutive epochs or when the maximum number of training epochs (1000 epochs) is reached. During training, a Dropout layer is used to suppress overfitting (dropout rate = 0.2), and early stopping ensures the model's generalization ability.

[0022] In a specific example, the 3D reconstruction and image segmentation auxiliary unit, based on multi-angle RGB images of the canopy, uses a binocular vision 3D reconstruction algorithm and a Mask R-CNN image segmentation algorithm to assist in obtaining the leaf area index and leaf tilt angle spatial structure information of the plant. First, feature point matching and stereo correction are performed on the multi-angle RGB images, and disparity maps are calculated. The 3D structure of the plant is then restored using the triangulation principle. Next, the Mask R-CNN algorithm is used to segment the 3D point cloud data, separating leaves and stem organs, accurately extracting the leaf area index and leaf tilt angle spatial structure parameters. This information is integrated into the spatiotemporal data structure to supplement spatial dimension features and improve the accuracy of the digital twin.

[0023] In a specific example, the feature fusion unit fuses the leaf area index and leaf tilt angle obtained from 3D reconstruction and image segmentation with spatiotemporal features learned through the ST-GCN model. Employing a fusion method combining feature concatenation and weighted summation, the spatial structure information and spatiotemporal features are first dimensionally concatenated, then adaptive weights are assigned based on feature importance, ultimately forming an optimized feature vector, further enhancing the representational capability of the dynamic digital twin.

[0024] In one example, the prediction output unit outputs the prediction results as scalar indicators: leaf area index, leaf tilt angle, photosynthetically active radiation utilization rate, and biomass accumulation rate. These scalar indicators are calculated based on a dynamic digital twin and the ST-GCN model to predict the future growth status of the plant: the leaf area index is calculated by converting the percentage of segmented leaf pixels to the actual area; the leaf tilt angle is calculated by the normal vector of the 3D point cloud data; the photosynthetically active radiation utilization rate is calculated based on the correlation model between light intensity and plant photosynthetic rate; and the biomass accumulation rate is derived from the growth rate of the phenotypic parameter and the biomass conversion coefficient.

[0025] In a specific example, the trend extrapolation and calibration unit extrapolates the trend of each scalar indicator based on historical time-series data to predict its value for the next 3 to 7 days. A linear trend extrapolation method is used, by fitting the indicator change curve over the most recent 14 days: Y(t) = at + b, in a For growth rate, b The intercept is used to extrapolate the predicted value for the next 7 days.

[0026] Simultaneously, a synchronization error threshold is set, with the default average relative error being less than or equal to 5%. The average relative error between the predicted value and the latest measured value of each scalar indicator is calculated every 24 hours, using the following formula: , in, n The number of predictions within 24 hours. Y pred,i For the first i Secondary predicted value Y meas,i For the first i The measured values ​​are used as follows. When the average relative error of any scalar indicator exceeds the configurable threshold, an online calibration procedure is triggered. Utilizing the updatable nature of the ST-GCN model, an incremental learning approach is used to locally correct the prediction model's parameters, with the correction magnitude not exceeding 10%, ensuring the stability of the prediction accuracy.

[0027] The dynamic digital twin generation unit, based on a trained ST-GCN model, integrates real-time acquired and preprocessed data to construct a dynamic digital twin for each representative plant. One representative plant is selected from every ten plants in the planting area to account for regional differences and ensure that the digital twin comprehensively reflects the plant growth status of the entire planting area. The dynamic digital twin maps the plant's growth status in real time, including current plant height, leaf area index, leaf tilt angle phenotypic parameters, and growth performance parameters such as photosynthetically active radiation utilization rate and biomass accumulation rate. It is updated every 5 minutes in sync with the real-time acquired data to more accurately simulate the plant growth process.

[0028] The abnormal growth status prediction module is connected to the digital twin model construction module. It is used to compare and comprehensively analyze the plant's future prediction results with the ideal growth curve of the corresponding growth cycle, and output an anomaly prediction signal. This module includes an ideal growth curve storage unit, a prediction result comparison unit, an anomaly degree assessment unit, and an anomaly early warning signal output unit.

[0029] In one example, the ideal growth curve storage unit pre-sets an ideal growth state curve corresponding to a crop variety. This curve is plotted with time on the horizontal axis and leaf area index, leaf tilt angle, photosynthetically active radiation utilization rate, and biomass accumulation rate on the vertical axis. The ideal curve is plotted through numerous experiments, combined with growth data of different plant varieties under standard conditions, with a light intensity of 1000 μmol / m². The optimal conditions were: temperature 22℃, humidity 60% RH, and CO2 concentration 1000ppm. Ideal growth curve libraries were constructed for different crop varieties (such as lettuce, tomato, and cucumber) and different growth stages (germination, seedling, mature plant, and harvesting) to ensure the curves were targeted and scientifically sound.

[0030] In one example, the prediction comparison unit compares each scalar indicator in the prediction results with the ideal curve item by item. During the comparison process, not only the magnitude of the values ​​is considered, but also the absolute deviation between the predicted value and the corresponding value on the ideal curve at each time point is calculated. Furthermore, by combining the ST-GCN model to grasp the spatiotemporal characteristics of plant growth, the differences between the trend of indicator changes and the ideal curve were analyzed. The slope of the predicted value sequence was calculated. The slope of the segment corresponding to the ideal curve ,when In this way, we can determine if the trend deviates and avoid misjudgments caused by comparing a single value.

[0031] In a specific example, the anomaly assessment unit uses the analytic hierarchy process (AHP) to construct an anomaly assessment system based on comparison results, evaluating the degree to which plant growth deviates from the ideal state. The assessment system includes three dimensions: numerical deviation magnitude, trend deviation degree, and duration. Numerical deviation magnitude is graded according to the proportion of deviation to the ideal value: less than or equal to 5% = 1 point, 5%-10% = 2 points, 10%-20% = 3 points, and >20% = 4 points. Trend deviation degree is graded according to slope difference: no deviation = 1 point, slight deviation = 2 points, moderate deviation = 3 points, and severe deviation = 4 points. Duration is graded according to the number of predicted periods of continuous deviation: 1 period = 1 point, 2-3 periods = 2 points, 4-6 periods = 3 points, and more than or equal to 7 periods = 4 points. The total anomaly degree score is a weighted sum of the three scores, providing a more accurate basis for the subsequent output of anomaly warning signals.

[0032] In a specific example, the anomaly warning signal output unit is set with a preset percentage range. When the predicted values ​​for multiple consecutive prediction periods deviate from the preset percentage range corresponding to the ideal curve value, and the anomaly severity assessment unit's evaluation result reaches a certain threshold, an anomaly warning signal is output. Based on the degree of anomaly, the warning signal is divided into three levels: mild anomaly (total score 1-2 points), moderate anomaly (total score 2-3 points), and severe anomaly (total score greater than or equal to 3 points), so that the strategy generation module can generate control strategies more effectively. The anomaly warning signal includes the name of the out-of-tolerance feature, Numerical values, comparisons of predicted and ideal values, anomaly scores, and early warning levels are sent to the strategy generation module via the MQTT protocol.

[0033] The strategy generation module receives anomaly prediction signals and generates control strategies. When generating control strategies, it comprehensively considers multiple dimensions such as plant growth rate, biomass accumulation, and resource utilization efficiency to fully assess the current growth status and generate more scientific and reasonable control strategies.

[0034] In a specific example, after receiving the anomaly prediction signal output by the growth state anomaly prediction module, the strategy generation module first analyzes the name of the out-of-range feature, the anomaly level, and the ΔY value in the signal. Combining this with the plant growth adaptation rules for the corresponding growth stage, it determines the control target. Differentiated control strategies are adopted for different levels of anomaly signals: for mild anomalies, only the environmental parameters corresponding to the out-of-range feature are fine-tuned, with the adjustment range controlled within 5%; for moderate anomalies, in addition to adjusting the environmental parameters corresponding to the out-of-range feature, related parameters are adjusted in conjunction, with an adjustment range of 5%-10%; for severe anomalies, the plant growth status is comprehensively assessed, and environmental parameters such as light intensity, temperature, humidity, and CO2 concentration are systematically adjusted, with the adjustment range dynamically determined according to the degree of anomaly, not exceeding 15%.

[0035] In one example, the strategy generation process fully considers the correlation between plant growth and resource utilization efficiency. For instance, when the predicted leaf area index is lower than the ideal curve, in addition to increasing light intensity, CO2 concentration and nutrient solution supply are adjusted simultaneously to promote photosynthesis and biomass accumulation. At the same time, the resource consumption cost of the control scheme is calculated, and control combinations with high resource utilization are selected first to ensure that energy consumption and resource waste are reduced while improving plant growth status.

[0036] The automated management and control system for plant factory cultivation provided in this application includes a data acquisition module for accurate acquisition and standardized preprocessing of multi-source parameters; a digital twin model construction module based on spatiotemporal graph convolutional network technology, combined with 3D reconstruction and image segmentation-assisted optimization, to construct a high-precision dynamic digital twin, enabling accurate prediction and trend extrapolation of future plant growth; a growth anomaly prediction module that compares prediction results with ideal growth curves to accurately identify and provide graded early warnings of abnormal states; and a strategy generation module that comprehensively considers plant growth needs and resource utilization efficiency to generate scientific and reasonable management and control strategies. This system effectively solves the technical pain points of traditional plant factory environmental control, such as insufficient precision, poor adaptability, and lack of growth prediction and anomaly early warning capabilities. It achieves intelligent management and control of the entire plant factory process, improves planting efficiency and control precision, and provides reliable technical support for large-scale, intelligent cultivation.

[0037] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk—any medium capable of storing program code.

[0038] The above embodiments are provided for those skilled in the art to implement and use this application. Those skilled in the art can make various modifications or changes to the above embodiments without departing from the inventive concept of this application. Therefore, the protection scope of this application is not limited to the above embodiments, but should conform to the maximum scope of the innovative features mentioned in the claims.

Claims

1. An automated control system for plant factory cultivation, comprising a data acquisition module, a digital twin model construction module, a growth status anomaly prediction module, and a strategy generation module, characterized in that: The data acquisition module is used to collect plant growth environment parameters and phenotypic parameters and perform preprocessing. The digital twin model construction module is used to construct a digital twin model based on preprocessed environmental parameters and phenotypic parameters. The digital twin model is used to reflect the working conditions of the plant factory in real time and output the prediction results of the future plants. The abnormal growth state prediction module is used to compare the plant's future prediction results with the ideal growth curve of the corresponding growth cycle, and output an abnormal prediction signal based on the analysis results. The strategy generation module is used to receive anomaly prediction signals and generate control strategies.

2. The plant factory growing automation management system of claim 1, wherein, The data acquisition module includes: The phenotypic data acquisition unit collects data including at least plant height and multi-angle RGB images of the canopy. The environmental data acquisition unit collects data including at least light intensity, temperature, humidity, and CO2 concentration. The data preprocessing unit cleans the collected environmental and phenotypic parameters to remove outliers and noise, interpolates and fills in missing data, and normalizes data of different dimensions to make each parameter fall within a similar value range.

3. The plant factory growing automation management system of claim 1, wherein, The digital twin model construction module employs spatiotemporal graph convolutional network technology, specifically including: The spatiotemporal data construction unit constructs a spatiotemporal data structure for plant growth that includes both temporal and spatial information, based on the collected phenotypic and environmental data. The ST-GCN model training unit uses the constructed spatiotemporal data structure to train the ST-GCN model, enabling it to learn the spatiotemporal characteristics of plant growth. The dynamic digital twin generation unit, based on a trained ST-GCN model, integrates real-time collected data to construct a dynamic digital twin of each representative plant, thereby more accurately simulating the plant growth process.

4. The automated control system for plant factory cultivation according to claim 3, characterized in that, The digital twin model construction module also includes: The 3D reconstruction and image segmentation auxiliary unit, based on the multi-angle RGB image of the canopy, uses 3D reconstruction and image segmentation algorithms to help obtain the leaf area index and leaf tilt angle spatial structure information of the plant, and integrates this information into the spatiotemporal data structure to improve the accuracy of the digital twin. The feature fusion unit integrates the spatial structure information obtained from 3D reconstruction and image segmentation with the spatiotemporal features learned through the ST-GCN model to further optimize the dynamic digital twin.

5. The automated control system for plant factory cultivation according to claim 3, characterized in that, The digital twin model construction module also includes: The prediction result output unit outputs the prediction results in the form of leaf area index, leaf tilt angle, photosynthetically active radiation utilization rate, and biomass accumulation rate. These scalar indicators are calculated based on dynamic digital twins and combined with the ST-GCN model to predict the future growth status of plants.

6. The automated control system for plant factory cultivation according to claim 5, characterized in that, The digital twin model construction module also includes: The trend extrapolation and calibration unit extrapolates the trend of each scalar indicator based on historical time series data to predict the value in the next 3 to 7 days; and sets a synchronization error threshold. When the average relative error between the predicted value and the latest measured value of any scalar indicator exceeds the configurable threshold, the online calibration procedure is triggered, and the local parameters of the prediction model are corrected by utilizing the updatable characteristics of the ST-GCN model.

7. The automated control system for plant factory cultivation according to claim 1, characterized in that, An abnormal growth state prediction module, connected to the digital twin model construction module, includes: The ideal growth curve storage unit is pre-set with the ideal growth status curve corresponding to the crop variety. The curve is plotted with time as the horizontal axis and leaf area index, leaf tilt angle, photosynthetically active radiation utilization rate, and biomass accumulation rate as the vertical axis. The ideal curve is plotted by experiments and combined with the growth data of different plant varieties under standard conditions.

8. The automated control system for plant factory cultivation according to claim 7, characterized in that, The abnormal growth state prediction module also includes: The prediction result comparison unit compares each scalar indicator in the prediction result with the ideal curve item by item. In the comparison process, it not only considers the magnitude of the value, but also combines the ST-GCN model to grasp the spatiotemporal characteristics of plant growth and analyzes the difference between the indicator change trend and the ideal curve. The anomaly assessment unit evaluates the degree to which the plant's growth status deviates from the ideal state based on the comparison results, providing a more accurate basis for the subsequent output of anomaly warning signals.

9. The automated control system for plant factory cultivation according to claim 8, characterized in that, The abnormal growth state prediction module also includes: The abnormal warning signal output unit outputs an abnormal warning signal when the predicted values ​​for multiple consecutive prediction periods deviate from the preset percentage range of the corresponding values ​​of the ideal curve, and the evaluation result of the abnormality degree assessment unit reaches a certain threshold. At the same time, the warning signal is divided into different levels according to the different degrees of abnormality, so that the strategy generation module can generate control strategies in a more targeted manner.

10. The automated control system for plant factory cultivation according to claim 1, characterized in that, When generating management and control strategies, the strategy generation module comprehensively considers three dimensions: plant growth rate, biomass accumulation, and resource utilization efficiency, and fully assesses the current growth status to generate more scientific and reasonable management and control strategies.

Images