Plant factory intelligent management system based on digital twin

Abstract

The present invention discloses an intelligent management system for a plant factory based on digital twins, comprising: a perception and execution layer, comprising various sensors, energy equipment, cameras and control equipment, for real-time sensing, monitoring and regulation of various elements of the physical scene of the plant factory; a network transmission layer, comprising a convergence node, an intelligent gateway and a cloud server, and setting an artificial intelligence algorithm and a multi-path routing algorithm, for providing an intelligent network transmission environment, adopting a multi-path routing algorithm according to different functions and different subtasks for high-speed transmission and accurate distribution of data, and realizing two-way communication between a virtual model of a plant factory and a physical scene by setting different software interfaces, and real-time dynamic mapping of monitoring data; a twin function layer, for establishing a digital twin model of a plant factory with virtual-real mapping, and based on the digital twin model, adopting a deep learning algorithm for intelligent regulation and control decisions of energy equipment and control equipment, prediction of growth of different plants and identification of plant diseases.

Description

Technical Field

[0001] The present invention belongs to the technical field of plant factories, and in particular relates to an intelligent management system for plant factories based on digital twins. Background Art

[0002] Digital twin is a cutting-edge new technology that has emerged in recent years. Simply put, it uses physical models, IoT sensors, and simulation to construct a mapping of the physical world in the digital world in the virtual space through digital means to reflect the entire production process of the corresponding physical equipment. However, in the field of plant factory management, digital twin technology does not yet have a complete set of system processes to analyze the current most suitable growth environment for plants, control equipment regulation, and plant production prediction.

[0003] In the field of plant factory technology, digital twin technology can integrate multiple algorithms. Compared with traditional machine learning algorithms, digital twin technology can more accurately analyze the environmental status and growth analysis of plant factories, and achieve more refined simulation and intelligent management of real plants.

[0004] However, the existing technology only manages plant factories from a single perspective, conducting data mining based on monitoring data. It lacks the integration of multiple information such as the physical information of plants and environmental change information on the virtual end, and cannot intuitively manage plant factories intelligently and finely.

[0005] Based on the above technical problems, it is necessary to design a new plant factory intelligent management system based on digital twins. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a plant factory intelligent management system based on digital twins.

[0007] In order to solve the above technical problems, the technical solution of the present invention is:

[0008] The present invention provides a plant factory intelligent management system based on digital twins, which comprises: a perception execution layer, a network transmission layer and a twin function layer; the perception execution layer and the twin function layer are communicatively connected through the network transmission layer;

[0009] The perception execution layer includes various sensors, energy equipment, cameras and control equipment, which are used to sense and monitor various elements of the physical scene of the plant factory in real time and regulate the energy equipment and control equipment according to decision instructions;

[0010] The network transmission layer includes a convergence node, an intelligent gateway and a cloud server, and is provided with an artificial intelligence algorithm and a multi-path routing algorithm to provide an intelligent network transmission environment for data transmission and processing. A multi-path routing algorithm is used for high-speed data transmission and accurate distribution according to different functions and different subtasks. By setting different software interfaces, two-way communication between the plant factory virtual model and the physical scene is achieved, and real-time dynamic mapping of monitoring data is achieved;

[0011] The twin function layer is used to utilize the system's historical data and real-time operation data in a data-driven manner, update and correct the mathematical model, and establish a digital twin model of the plant factory with virtual-to-real mapping. Based on the digital twin model, a deep learning algorithm is used to perform intelligent regulation and decision-making on energy equipment and control equipment, predict the growth of different plants, and identify plant diseases.

[0012] Furthermore, the various sensors include at least air temperature and humidity sensors, CO2 sensors, light intensity sensors and soil temperature and humidity sensors, which are used to collect multi-source global real-time data of the plant factory; the energy equipment includes at least renewable energy power generation equipment, heat pumps, cogeneration units, energy storage devices and refrigeration units, which are used to supply electricity, heat and cold energy in the plant factory; the control equipment includes at least fill lighting equipment, ventilation equipment, dehumidification equipment, CO2 generation or release equipment, which are used to execute the regulation strategy of twin functional layer transmission.

[0013] Furthermore, in the network transmission layer, the data collected by various sensors are sent from the aggregation node to the intelligent gateway through the network, and then the intelligent gateway performs data preprocessing and transmits the data to the cloud server together with the video image information collected by the camera for storage and processing; the network adopts a combination of 3G, 4G, 5G, WIFI, NB-IOT wireless networks and Ethernet-based wired transmission; the data processing adopts multi-source data fusion technology, and the data collected by the sensors are input into mathematical models established by different algorithms for comprehensive processing;

[0014] The method realizes two-way communication between the virtual model of the plant factory and the physical scene by setting different software interfaces, and monitors the real-time dynamic mapping of data, including: setting different software interfaces to realize two-way communication between the virtual model of the plant factory and the physical scene, exchanging the actual scene process of the plant factory, and calculating and analyzing the acquired data, while monitoring the real-time dynamic mapping of data, completing data sending and receiving, instruction transmission and message synchronization, and issuing control strategies to the physical entity of the plant factory.

[0015] Furthermore, the establishment of a plant factory digital twin model with virtual-real mapping includes:

[0016] By establishing a two-way connection and real-time interactive channel between the physical entity of the plant factory and its virtual twin;

[0017] The physical entity of the plant factory is mapped to reality and data-driven. Based on the information of various sensors, cameras, energy equipment and control equipment in the physical entity of the plant factory, a 2D drawing of the plant factory and a 3D model scene of the plant are established. The model parameters are configured and the real-time operation status is simulated according to the relevant data information. A high-fidelity virtual digital twin model of the physical entity of the plant factory is established.

[0018] Identify the digital twin model of the plant factory, connect the real-time operation data of multiple working conditions of the plant factory to the established digital twin model, use the reverse identification method to adaptively identify and correct the simulation results of the digital twin model, and obtain the identified and corrected digital twin model of the plant factory;

[0019] The plant factory digital twin model includes the plant factory physical entity, virtual entity, twin data service and connection elements between the components;

[0020] The physical entity of the plant factory is the basis of the digital twin model and the data source driven by the entire digital twin model; the virtual entity of the plant factory is mapped one-to-one with the physical entity and interacts in real time, and the actual process of the physical entity is simulated by characterizing the elements of the physical space from multiple dimensions and scales; the twin data service integrates physical space information and virtual space information to ensure the real-time transmission of data, and at the same time provides knowledge base data including intelligent algorithms, models, rules and standards, and expert experience, and forms a twin database by integrating physical information, multi-time and space related information, and knowledge base data; the connection between the components is to realize the interconnection and interoperability of the components, and the real-time collection and feedback of data between the physical entity and the twin data service are realized through sensors and protocol transmission specifications; data is transmitted between the physical entity and the virtual entity through the protocol, and the physical information is transmitted to the virtual space in real time to update the correction model, and the virtual entity controls the physical entity in real time through the actuator; the information transmission between the virtual entity and the twin data service is realized through the database interface.

[0021] Furthermore, the intelligent control decision-making of energy equipment using a deep learning algorithm based on the digital twin model includes:

[0022] Based on the plant factory digital twin model, historical electricity, heat, water, and cold-related data of different plants in different growth cycles, weather data, and historical energy equipment control data are obtained as energy sample sets;

[0023] After normalization preprocessing, the data in the energy sample set is divided into a training set and a test set;

[0024] The initialization hyperparameters of the CNN-LSTM-ATTN network model are set, and the training set is input into the CNN-LSTM-ATTN network model. The model continuously optimizes the parameters adaptively, extracts the features of the input data through the convolution and pooling operations of the CNN model, and performs learning and training through the LSTM model. The prediction error of the prediction model caused by the difference in energy consumption of plants at different growth cycle stages is reduced by adding the ATTN self-attention mechanism until the loss function tends to converge and the prediction model reaches the preset accuracy or the preset number of training times.

[0025] The test set is input into the trained CNN-LSTM-ATTN network model to obtain the prediction results of the test set, calculate the performance evaluation index of the model, and obtain the energy management model with the best prediction performance;

[0026] Based on the energy management model, the electricity, heat, water and cold energy load data of different plants in different growth cycles are obtained, and then the energy equipment is regulated according to the energy use strategy of the energy equipment itself;

[0027] The energy use strategy of the energy device itself includes at least:

[0028] For renewable energy power generation equipment combined with energy storage equipment, when the power load of the plant factory is small and the power generation of renewable energy is large, the excess power is stored through the energy storage device, and when the power load of the plant factory is large, the power is released through the energy storage device; or adopt the peak and valley electricity price time-sharing strategy, purchase electricity during the period of low electricity price, store the excess power, and release the stored power during the period of high electricity price;

[0029] For heat pumps, cogeneration units or other heating units in plant factories combined with energy storage equipment, when the heat load in the plant factory is small, heating is provided by a single heating device; when the heat load in the plant factory is large, heating is provided by multiple heating devices in combination, and when there is excess heat, it is stored through energy storage equipment;

[0030] In response to the water and cooling demands of different plants in different growth cycles in the plant factory, water pumps and refrigeration units are regulated in different time periods.

[0031] Furthermore, the intelligent control decision-making of the control device using a deep learning algorithm based on the digital twin model includes:

[0032] Based on the plant factory digital twin model, historical air temperature and humidity sensor, CO2 sensor, light intensity sensor and soil temperature and humidity sensor data, weather data, plant categories, plant characteristics, plant production status, and historical control operations of control equipment for different growth periods of different plants are obtained as control sample sets;

[0033] The DBN model is used to extract features from the data in the control sample set to obtain the factors affecting the regulation of different control devices;

[0034] Establish the RLSSVM model, and optimize the adjustment penalty parameter C and kernel function width σ of the RLSSVM model through intelligent optimization algorithm;

[0035] The extracted characteristic data of the control impact of different control devices are input into the RLSSVM model after parameter optimization to obtain the control strategy of each control device;

[0036] The intelligent optimization algorithm is an improved GSA algorithm based on chaos algorithm and time-varying weights, including: initializing individual positions with chaos algorithm; calculating individual fitness values; updating the gravitational force G(t) at time t, the mass M of individual i at time t, and i (t), the best and worst values ​​of the fitness value best(t) and worst(t); calculate the force and acceleration in all directions of the individual; calculate the adaptive weight; update the individual speed and position; determine whether the termination condition is met, if so, stop, otherwise recalculate the individual fitness value.

[0037] Furthermore, the digital twin model is used to predict the growth of different plants using a deep learning algorithm, including:

[0038] Based on the plant factory digital twin model, historical growth data, historical sensor data, energy equipment intelligent control data, control equipment intelligent control data, plant types and plant characteristics of different plants in different growth cycles are obtained as a growth sample library; the growth data includes at least the number of plant leaves, leaf length, leaf width and fresh weight;

[0039] Perform model feature selection on data of different growth cycles in the growth sample library, and establish growth prediction models of various deep learning algorithms;

[0040] Information entropy is used to perform weighted integration of multiple single growth prediction models to obtain integrated plant growth prediction values, including:

[0041] Calculate the relative error value of each intelligent prediction model according to the predicted value and expected value of each growth prediction model;

[0042] The entropy value of each growth prediction model is calculated and expressed as: L is the number of samples; p ud is the prediction relative error ratio of the u-th model for the q-th sample;

[0043] Calculate the weights of each growth prediction model, expressed as:

[0044] Calculate the weighted ensemble output of each growth prediction model: is the predicted value of the qth sample of the uth model;

[0045] Among them, the multiple different deep learning algorithms include at least XGBoost model, support vector regression SVR model and random forest model.

[0046] Furthermore, the plant disease identification using a deep learning algorithm based on the digital twin model includes:

[0047] Preprocessing the image of the plant to be identified collected by the camera, wherein the image of the plant to be identified contains plant disease pictures and label data of the plant disease;

[0048] Through the Labelme annotation tool, the plant parts belonging to disease identification are extracted from the pre-processed images according to the shape and height of the plants;

[0049] The plant parts belonging to disease identification are identified by a support vector machine model; the support vector machine model uses a grid search algorithm to optimize the model parameters: the feasible interval of the support vector machine model parameter C to be optimized and the kernel function width σ is divided into multiple intervals according to the set step size, and the optimal model parameters for disease identification are calculated using a cross-validation method.

[0050] The beneficial effects of the present invention are:

[0051] The present invention establishes a perception execution layer, including various sensors, energy equipment, cameras and control equipment, which are used to sense and monitor various elements of the physical scene of the plant factory in real time, and regulate the energy equipment and control equipment according to decision instructions; the network transmission layer, including aggregation nodes, intelligent gateways and cloud servers, is equipped with artificial intelligence algorithms and multi-path routing algorithms to provide an intelligent network transmission environment for data transmission and processing, and adopts multi-path routing algorithms according to different functions and different subtasks to achieve high-speed transmission and accurate distribution of data. By setting different software interfaces, two-way communication between the virtual model of the plant factory and the physical scene is realized, and real-time dynamic mapping of monitoring data is achieved; the twin layer The functional layer is used to use the system's historical data and real-time operation data in a data-driven manner, update and correct the mathematical model, and establish a digital twin model of the plant factory with virtual-real mapping. Based on the digital twin model, a deep learning algorithm is used to make intelligent regulation decisions for energy equipment and control equipment, predict the growth of different plants, and identify plant diseases. Through the three-layer plant factory technical architecture, the physical information of the plant factory, environmental change information, plant characteristics and other information are integrated, and the plant factory is mapped to the virtual and real and dynamically simulated based on the digital twin model, so as to realize intelligent and refined management of the plant factory based on the digital twin model and deep learning algorithm.

[0052] Other features and advantages will be described in the following description, and partly become apparent from the description, or understood by practicing the invention. The purpose and other advantages of the invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 This is a structural schematic diagram of a plant factory intelligent management system based on digital twins of the present invention;

[0056] Figure 2 This is a flow chart of the energy equipment control using a deep learning algorithm based on a digital twin model in the present invention;

[0057] Figure 3 This is a flow chart of the device control using a deep learning algorithm based on a digital twin model in the present invention;

[0058] Figure 4 This is a flow chart of the present invention using a deep learning algorithm to predict the growth of different plants based on a digital twin model;

[0059] Figure 5 This is a flow chart of plant disease identification using a deep learning algorithm based on a digital twin model. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Example 1

[0062] Figure 1 It is a structural schematic diagram of a plant factory intelligent management system based on digital twins involved in the present invention.

[0063] Figure 2 This is a flow chart of energy equipment control based on a digital twin model using a deep learning algorithm as involved in the present invention.

[0064] Figure 3 This is a flow chart of the device control using a deep learning algorithm based on a digital twin model involved in the present invention.

[0065] Figure 4 This is a flow chart of the present invention for predicting the growth of different plants using a deep learning algorithm based on a digital twin model.

[0066] Figure 5 This is a flow chart of plant disease identification based on a digital twin model using a deep learning algorithm as involved in the present invention.

[0067] like Figure 1-5 As shown, this embodiment 1 provides a plant factory intelligent management system based on digital twins, which includes: a perception execution layer, a network transmission layer and a twin function layer; the perception execution layer and the twin function layer are communicatively connected through the network transmission layer;

[0068] The perception execution layer includes various sensors, energy equipment, cameras and control equipment, which are used to sense and monitor various elements of the physical scene of the plant factory in real time and regulate the energy equipment and control equipment according to decision instructions;

[0069] The network transmission layer includes a convergence node, an intelligent gateway and a cloud server, and is provided with an artificial intelligence algorithm and a multi-path routing algorithm to provide an intelligent network transmission environment for data transmission and processing. A multi-path routing algorithm is used for high-speed data transmission and accurate distribution according to different functions and different subtasks. By setting different software interfaces, two-way communication between the plant factory virtual model and the physical scene is achieved, and real-time dynamic mapping of monitoring data is achieved;

[0070] The twin function layer is used to utilize the system's historical data and real-time operation data in a data-driven manner, update and correct the mathematical model, and establish a digital twin model of the plant factory with virtual-to-real mapping. Based on the digital twin model, a deep learning algorithm is used to perform intelligent regulation and decision-making on energy equipment and control equipment, predict the growth of different plants, and identify plant diseases.

[0071] It should be noted that a plant factory is a model that achieves crop planting by regulating environmental parameters in a closed indoor environment. Therefore, it is necessary to have planting conditions, build an indoor planting environment, and build an environmental monitoring module to achieve real-time feedback of indoor environmental parameters. Through corresponding control strategies, it is regulated and controlled to make the entire plant factory system operate normally, ensuring high-yield and high-quality industrial production of crops. For example, simulating crop growth refers to the use of computer technology to simulate the plant development process through certain algorithms and programs. The model established based on this is called a crop growth model. Modeling the growth conditions and different parts of crops at different growth stages is to better understand the growth and development process of crops. The results of the modeling can be used to develop a decision support system, thereby realizing the functions of environmental control of production equipment in facility agriculture, yield prediction, and production and planting plan formulation.

[0072] In the actual plant factory management process, environmental monitoring and control are important work contents. With the rapid development of Internet of Things technology, plant factory management also needs to introduce its core sensing technology, wireless communication technology, etc. to establish a comprehensive plant factory management system to ensure that producers or technical researchers can monitor and control the plant factory site in real time and realize green and intelligent production of plants.

[0073] In this embodiment, the various sensors include at least air temperature and humidity sensors, CO2 sensors, light intensity sensors and soil temperature and humidity sensors, which are used to collect multi-source global real-time data of the plant factory; the energy equipment includes at least renewable energy power generation equipment, heat pumps, cogeneration units, energy storage devices and refrigeration units, which are used to supply electricity, heat and cold energy in the plant factory; the control equipment includes at least fill lighting equipment, ventilation equipment, dehumidification equipment, CO2 generation or release equipment, which are used to execute the regulation strategy of twin functional layer transmission.

[0074] In this embodiment, the data collected by various sensors in the network transmission layer are sent from the aggregation node to the intelligent gateway through the network, and then the intelligent gateway performs data preprocessing and transmits the data to the cloud server together with the video image information collected by the camera for storage and processing; the network adopts a combination of 3G, 4G, 5G, WIFI, NB-IOT wireless networks and Ethernet-based wired transmission; the data processing adopts multi-source data fusion technology, and the data collected by the sensors are input into the mathematical model established by different algorithms for comprehensive processing;

[0075] The method realizes two-way communication between the virtual model of the plant factory and the physical scene by setting different software interfaces, and monitors the real-time dynamic mapping of data, including: setting different software interfaces to realize two-way communication between the virtual model of the plant factory and the physical scene, exchanging the actual scene process of the plant factory, and calculating and analyzing the acquired data, while monitoring the real-time dynamic mapping of data, completing data sending and receiving, instruction transmission and message synchronization, and issuing control strategies to the physical entity of the plant factory.

[0076] In this embodiment, the establishment of a plant factory digital twin model with virtual-real mapping includes:

[0077] By establishing a two-way connection and real-time interactive channel between the physical entity of the plant factory and its virtual twin;

[0078] The physical entity of the plant factory is mapped to reality and data-driven. Based on the information of various sensors, cameras, energy equipment and control equipment in the physical entity of the plant factory, a 2D drawing of the plant factory and a 3D model scene of the plant are established. The model parameters are configured and the real-time operation status is simulated according to the relevant data information. A high-fidelity virtual digital twin model of the physical entity of the plant factory is established.

[0079] Identify the digital twin model of the plant factory, connect the real-time operation data of multiple working conditions of the plant factory to the established digital twin model, use the reverse identification method to adaptively identify and correct the simulation results of the digital twin model, and obtain the identified and corrected digital twin model of the plant factory;

[0080] The plant factory digital twin model includes the plant factory physical entity, virtual entity, twin data service and connection elements between the components;

[0081] The physical entity of the plant factory is the basis of the digital twin model and the data source driven by the entire digital twin model; the virtual entity of the plant factory is mapped one-to-one with the physical entity and interacts in real time, and the actual process of the physical entity is simulated by characterizing the elements of the physical space from multiple dimensions and scales; the twin data service integrates physical space information and virtual space information to ensure the real-time transmission of data, and at the same time provides knowledge base data including intelligent algorithms, models, rules and standards, and expert experience, and forms a twin database by integrating physical information, multi-time and space related information, and knowledge base data; the connection between the components is to realize the interconnection and interoperability of the components, and the real-time collection and feedback of data between the physical entity and the twin data service are realized through sensors and protocol transmission specifications; data is transmitted between the physical entity and the virtual entity through the protocol, and the physical information is transmitted to the virtual space in real time to update the correction model, and the virtual entity controls the physical entity in real time through the actuator; the information transmission between the virtual entity and the twin data service is realized through the database interface.

[0082] In this embodiment, the intelligent control decision of energy equipment using a deep learning algorithm based on the digital twin model includes:

[0083] Based on the plant factory digital twin model, historical electricity, heat, water, and cold-related data of different plants in different growth cycles, weather data, and historical energy equipment control data are obtained as energy sample sets;

[0084] After normalization preprocessing, the data in the energy sample set is divided into a training set and a test set;

[0085] The initialization hyperparameters of the CNN-LSTM-ATTN network model are set, and the training set is input into the CNN-LSTM-ATTN network model. The model continuously optimizes the parameters adaptively, extracts the features of the input data through the convolution and pooling operations of the CNN model, and performs learning and training through the LSTM model. The prediction error of the prediction model caused by the difference in energy consumption of plants at different growth cycle stages is reduced by adding the ATTN self-attention mechanism until the loss function tends to converge and the prediction model reaches the preset accuracy or the preset number of training times.

[0086] The test set is input into the trained CNN-LSTM-ATTN network model to obtain the prediction results of the test set, calculate the performance evaluation index of the model, and obtain the energy management model with the best prediction performance;

[0087] Based on the energy management model, the electricity, heat, water and cold energy load data of different plants in different growth cycles are obtained, and then the energy equipment is regulated according to the energy use strategy of the energy equipment itself;

[0088] The energy use strategy of the energy device itself includes at least:

[0089] For renewable energy power generation equipment combined with energy storage equipment, when the power load of the plant factory is small and the power generation of renewable energy is large, the excess power is stored through the energy storage device, and when the power load of the plant factory is large, the power is released through the energy storage device; or adopt the peak and valley electricity price time-sharing strategy, purchase electricity during the period of low electricity price, store the excess power, and release the stored power during the period of high electricity price;

[0090] For heat pumps, cogeneration units or other heating units in plant factories combined with energy storage equipment, when the heat load in the plant factory is small, heating is provided by a single heating device; when the heat load in the plant factory is large, heating is provided by multiple heating devices in combination, and when there is excess heat, it is stored through energy storage equipment;

[0091] In response to the water and cooling demands of different plants in different growth cycles in the plant factory, water pumps and refrigeration units are regulated in different time periods.

[0092] It should be noted that CNN is used to mine the internal connection between input data, the convolution layer is used to extract and filter the input data features, and then the pooling layer is used to filter the data features to reduce the data dimension; the activation function of the convolution layer uses RELU, which can suppress overfitting and has a fast operation speed; the key to the attention mechanism is to allow the neural network to pay attention to the places where it needs special attention at different time points, thereby improving the accuracy of model prediction, and the self-attention mechanism is to achieve attention interaction through the data information within the data features, and obtain the importance of time series data at different stages of plant growth cycles through the self-attention mechanism, thereby improving the accuracy of the energy management prediction model. The performance evaluation indicators of the model use root mean square error RMSE, mean absolute error MAE and mean absolute percentage error MAPE to measure the stability and performance of the model. The smaller the RMSE value, the better the prediction ability of the model and the closer it is to the actual value; the smaller the MAE and MAPE values, the better the stability of the model.

[0093] In this embodiment, the intelligent control decision of the control device using a deep learning algorithm based on the digital twin model includes:

[0094] Based on the plant factory digital twin model, historical air temperature and humidity sensor, CO2 sensor, light intensity sensor and soil temperature and humidity sensor data, weather data, plant categories, plant characteristics, plant production status, and historical control operations of control equipment for different growth periods of different plants are obtained as control sample sets;

[0095] The DBN model is used to extract features from the data in the control sample set to obtain the factors affecting the regulation of different control devices;

[0096] Establish the RLSSVM model, and optimize the adjustment penalty parameter C and kernel function width σ of the RLSSVM model through intelligent optimization algorithm;

[0097] The extracted characteristic data of the control impact of different control devices are input into the RLSSVM model after parameter optimization to obtain the control strategy of each control device;

[0098] The intelligent optimization algorithm is an improved GSA algorithm based on chaos algorithm and time-varying weights, including: initializing individual positions with chaos algorithm; calculating individual fitness values; updating the gravitational force G(t) at time t, the mass M of individual i at time t, and i (t), the best and worst values ​​of the fitness value best(t) and worst(t); calculate the force and acceleration in all directions of the individual; calculate the adaptive weight; update the individual speed and position; determine whether the termination condition is met, if so, stop, otherwise recalculate the individual fitness value.

[0099] It should be noted that the feature extraction of DBN model includes two stages: pre-training and fine-tuning. In the pre-training stage, the network parameters are learned by layer-by-layer unsupervised method; in the fine-tuning stage, the BP network is set in the last layer of DBN, the output feature vector of DBN is received as its input feature vector, and the gradient descent algorithm is used to fine-tune the weight of the entire network. RLSSVM is an improvement on the standard SVM, which improves the control speed of the prediction model and reduces the calculation difficulty, thereby improving the control efficiency of the control equipment. The parameters in the RLSSVM model include the penalty parameter C and the parameters in the kernel function. The RBF kernel function width is used to affect the performance of RLSSVM. If the kernel function width is small, the RLSSVM algorithm will have over-learning phenomenon, thereby reducing its generalization ability. Otherwise, under-learning may occur. The performance of RLSSVM can be controlled by adjusting the penalty parameter and the kernel function width. These parameters affect the number of support vector machines and the distance between two samples, and affect the generalization ability of the model. The proposed improved GSA algorithm based on chaos algorithm and time-varying weight, G(t) is the universal gravitation at time t;

[0100]

[0101]

[0102] Among them, M i (t), fit i (t) is the quality and fitness value of individual i at time t.

[0103] In this embodiment, the growth prediction of different plants using a deep learning algorithm based on the digital twin model includes:

[0104] Based on the plant factory digital twin model, historical growth data, historical sensor data, energy equipment intelligent control data, control equipment intelligent control data, plant types and plant characteristics of different plants in different growth cycles are obtained as a growth sample library; the growth data includes at least the number of plant leaves, leaf length, leaf width and fresh weight;

[0105] Perform model feature selection on data of different growth cycles in the growth sample library, and establish growth prediction models of various deep learning algorithms;

[0106] Information entropy is used to perform weighted integration of multiple single growth prediction models to obtain integrated plant growth prediction values, including:

[0107] Calculate the relative error value of each intelligent prediction model according to the predicted value and expected value of each growth prediction model;

[0108] The entropy value of each growth prediction model is calculated and expressed as: L is the number of samples; p ud is the prediction relative error ratio of the u-th model for the q-th sample;

[0109] Calculate the weights of each growth prediction model, expressed as:

[0110] Calculate the weighted ensemble output of each growth prediction model: is the predicted value of the qth sample of the uth model;

[0111] Among them, the multiple different deep learning algorithms include at least XGBoost model, support vector regression SVR model and random forest model.

[0112] In this embodiment, the plant disease identification using a deep learning algorithm based on the digital twin model includes:

[0113] Preprocessing the image of the plant to be identified collected by the camera, wherein the image of the plant to be identified contains plant disease pictures and label data of the plant disease;

[0114] Through the Labelme annotation tool, the plant parts belonging to disease identification are extracted from the pre-processed images according to the shape and height of the plants;

[0115] The plant parts belonging to disease identification are identified by a support vector machine model; the support vector machine model uses a grid search algorithm to optimize the model parameters: the feasible interval of the support vector machine model parameter C to be optimized and the kernel function width σ is divided into multiple intervals according to the set step size, and the optimal model parameters for disease identification are calculated using a cross-validation method.

[0116] It should be noted that cross-validation is to divide the training set into k groups, and use the k-1 subsets as the training set, and then use the trained support vector machine model to test the remaining subsets, and take the average of the recognition accuracy of k tests as the accuracy corresponding to this group of parameters C and the kernel function width σ, which can effectively avoid overfitting or underfitting.

[0117] In several embodiments provided in the present application, it should be understood that the disclosed system and method can also be implemented in other ways. The system embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0118] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0119] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0120] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A plant factory intelligent management system based on digital twins, characterized in that: It includes: A perception execution layer, a network transmission layer and a twin function layer; the perception execution layer and the twin function layer are communicatively connected via the network transmission layer; The perception execution layer includes various sensors, energy equipment, cameras and control equipment, which are used to sense and monitor various elements of the physical scene of the plant factory in real time and regulate the energy equipment and control equipment according to decision instructions; The network transmission layer includes a convergence node, an intelligent gateway and a cloud server, and is provided with an artificial intelligence algorithm and a multi-path routing algorithm to provide an intelligent network transmission environment for data transmission and processing. A multi-path routing algorithm is used for high-speed data transmission and accurate distribution according to different functions and different subtasks. By setting different software interfaces, two-way communication between the plant factory virtual model and the physical scene is achieved, and real-time dynamic mapping of monitoring data is achieved; The twin function layer is used to use the historical data and real-time operation data of the system in a data-driven manner to update and correct the mathematical model, establish a virtual-real mapping digital twin model of the plant factory, and use a deep learning algorithm based on the digital twin model to make intelligent regulation decisions for energy equipment and control equipment, predict the growth of different plants, and identify plant diseases; The plant factory digital twin model with virtual-real mapping is established, including: By establishing a two-way connection and real-time interactive channel between the physical entity of the plant factory and its virtual twin; The physical entity of the plant factory is mapped to reality and data-driven. Based on the information of various sensors, cameras, energy equipment and control equipment in the physical entity of the plant factory, a 2D drawing of the plant factory and a 3D model scene of the plant are established. The model parameters are configured and the real-time operation status is simulated according to the relevant data information. A high-fidelity virtual digital twin model of the physical entity of the plant factory is established. Identify the digital twin model of the plant factory, connect the real-time operation data of multiple working conditions of the plant factory to the established digital twin model, use the reverse identification method to adaptively identify and correct the simulation results of the digital twin model, and obtain the identified and corrected digital twin model of the plant factory; The plant factory digital twin model includes the plant factory physical entity, virtual entity, twin data service and connection elements between the components; The physical entity of the plant factory is the basis of the digital twin model and the data source driven by the entire digital twin model; the virtual entity of the plant factory is mapped one-to-one with the physical entity and interacts in real time, and the actual process of the physical entity is simulated by characterizing the elements of the physical space from multiple dimensions and scales; the twin data service integrates physical space information and virtual space information to ensure the real-time transmission of data, and at the same time provides knowledge base data including intelligent algorithms, models, rules and standards, and expert experience, and forms a twin database by integrating physical information, multi-time and space related information, and knowledge base data; the connection between the components is to realize the interconnection and interoperability of the components, and the real-time collection and feedback of data between the physical entity and the twin data service are realized through sensors and protocol transmission specifications; data is transmitted between the physical entity and the virtual entity through the protocol, and the physical information is transmitted to the virtual space in real time to update the correction model, and the virtual entity controls the physical entity in real time through the actuator; the information transmission between the virtual entity and the twin data service is realized through the database interface.

2. The plant factory intelligent management system according to claim 1, characterized in that: The various sensors include at least air temperature and humidity sensors, CO2 sensors, light intensity sensors and soil temperature and humidity sensors, which are used to collect multi-source global real-time data of the plant factory; the energy equipment includes at least renewable energy power generation equipment, heat pumps, cogeneration units, energy storage devices and refrigeration units, which are used to supply electricity, heat and cold energy in the plant factory; the control equipment includes at least fill lighting equipment, ventilation equipment, dehumidification equipment, CO2 generation or release equipment, which are used to execute the regulation strategy of twin functional layer transmission.

3. The plant factory intelligent management system according to claim 1, characterized in that: In the network transmission layer, the data collected by various sensors are sent from the aggregation node to the intelligent gateway through the network, and then the intelligent gateway performs data preprocessing and transmits the data to the cloud server together with the video image information collected by the camera for storage and processing; the network adopts a combination of 3G, 4G, 5G, WIFI, NB-IOT wireless networks and Ethernet-based wired transmission; the data preprocessing adopts multi-source data fusion technology, and the data collected by the sensor is input into the mathematical model established by different algorithms for comprehensive processing; The method realizes two-way communication between the virtual model of the plant factory and the physical scene by setting different software interfaces, and monitors the real-time dynamic mapping of data, including: setting different software interfaces to realize two-way communication between the virtual model of the plant factory and the physical scene, exchanging the actual scene process of the plant factory, and calculating and analyzing the acquired data, while monitoring the real-time dynamic mapping of data, completing data sending and receiving, instruction transmission and message synchronization, and issuing control strategies to the physical entity of the plant factory.

4. The plant factory intelligent management system according to claim 1, characterized in that: The intelligent control decision-making of energy equipment using a deep learning algorithm based on the digital twin model includes: Based on the plant factory digital twin model, historical electricity, heat, water, and cold-related data of different plants in different growth cycles, weather data, and historical energy equipment control data are obtained as energy sample sets; After normalization preprocessing, the data in the energy sample set is divided into a training set and a test set; The initialization hyperparameters of the CNN-LSTM-ATTN network model are set, and the training set is input into the CNN-LSTM-ATTN network model. The model continuously optimizes the parameters adaptively, extracts the features of the input data through the convolution and pooling operations of the CNN model, and performs learning and training through the LSTM model. The prediction error of the prediction model caused by the difference in energy consumption of plants at different growth cycle stages is reduced by adding the ATTN self-attention mechanism until the loss function tends to converge and the prediction model reaches the preset accuracy or the preset number of training times. The test set is input into the trained CNN-LSTM-ATTN network model to obtain the prediction results of the test set, calculate the performance evaluation index of the model, and obtain the energy management model with the best prediction performance; Based on the energy management model, the electricity, heat, water and cold energy load data of different plants in different growth cycles are obtained, and then the energy equipment is regulated according to the energy use strategy of the energy equipment itself; The energy use strategy of the energy device itself includes at least: For renewable energy power generation equipment combined with energy storage equipment, when the power load of the plant factory is small and the power generation of renewable energy is large, the excess power is stored through the energy storage device, and when the power load of the plant factory is large, the power is released through the energy storage device; or adopt the peak and valley electricity price time-sharing strategy, purchase electricity during the period of low electricity price, store the excess power, and release the stored power during the period of high electricity price; For heat pumps, cogeneration units or other heating units in plant factories combined with energy storage equipment, when the heat load in the plant factory is small, heating is provided by a single heating device; when the heat load in the plant factory is large, heating is provided by multiple heating devices in combination, and when there is excess heat, it is stored through energy storage equipment; In response to the water and cooling demands of different plants in different growth cycles in the plant factory, water pumps and refrigeration units are regulated in different time periods.

5. The plant factory intelligent management system according to claim 1, characterized in that: The intelligent control decision-making of the control device using a deep learning algorithm based on the digital twin model includes: Based on the plant factory digital twin model, historical air temperature and humidity sensor, CO2 sensor, light intensity sensor and soil temperature and humidity sensor data, weather data, plant categories, plant characteristics, plant production status, and historical control operations of control equipment for different growth periods of different plants are obtained as control sample sets; The DBN model is used to extract features from the data in the control sample set to obtain the factors affecting the regulation of different control devices; Establish a RLSSVM model, and optimize the adjustment penalty parameter C and kernel function width σ of the RLSSVM model through an intelligent optimization algorithm; The extracted characteristic data of the control impact of different control devices are input into the RLSSVM model after parameter optimization to obtain the control strategy of each control device; The intelligent optimization algorithm is an improved GSA algorithm based on chaos algorithm and time-varying weights, including: initializing individual positions with chaos algorithm; calculating individual fitness values; updating the algorithm universal gravitation G(t) at time t, the mass M of individual i at time t i (t), the best and worst values ​​of the fitness value best(t) and worst(t); calculate the force and acceleration in all directions of the individual; calculate the adaptive weight; update the individual speed and position; determine whether the termination condition is met, if so, stop, otherwise recalculate the individual fitness value.

6. The plant factory intelligent management system according to claim 1, characterized in that: The digital twin model is based on which a deep learning algorithm is used to predict the growth of different plants, including: Based on the plant factory digital twin model, historical growth data, historical sensor data, energy equipment intelligent control data, control equipment intelligent control data, plant types and plant characteristics of different plants in different growth cycles are obtained as a growth sample library; the growth data includes at least the number of plant leaves, leaf length, leaf width and fresh weight; Perform model feature selection on data of different growth cycles in the growth sample library, and establish growth prediction models of various deep learning algorithms; Information entropy is used to perform weighted integration of multiple single growth prediction models to obtain integrated plant growth prediction values, including: Calculate the relative error value of each intelligent prediction model according to the predicted value and expected value of each growth prediction model; The entropy value of each growth prediction model is calculated and expressed as: L is the number of samples; p ud is the prediction relative error ratio of the u-th model for the q-th sample; Calculate the weights of each growth prediction model, expressed as: Calculate the weighted ensemble output of each growth prediction model: is the predicted value of the qth sample of the uth model; Among them, the multiple different deep learning algorithms include at least XGBoost model, support vector regression SVR model and random forest model.

7. The plant factory intelligent management system according to claim 1, characterized in that: The plant disease identification using a deep learning algorithm based on the digital twin model includes: Preprocessing the image of the plant to be identified collected by the camera, wherein the image of the plant to be identified contains plant disease pictures and label data of the plant disease; Through the Labelme annotation tool, the plant parts belonging to disease identification are extracted from the pre-processed images according to the shape and height of the plants; The plant parts belonging to disease identification are identified by a support vector machine model; the support vector machine model uses a grid search algorithm to optimize the model parameters: the feasible interval of the support vector machine model parameter C to be optimized and the kernel function width σ is divided into multiple intervals according to the set step size, and the optimal model parameters for disease identification are calculated using a cross-validation method.

Images