Grass seedling factory intelligent master control system and method thereof

Abstract

The invention provides an intelligent main control system and an intelligent main control method for a seedling grass factory, which are used for solving the problem that the automation level and the intelligent level of the existing seedling grass factory control system are insufficient, and the system comprises a remote updating module, a parameter setting module, a video and image data processing module, a sensor data processing module and an intelligent control strategy optimization calculation module, the optimal control strategy model based on a DDPG reinforcement learning algorithm is trained by analyzing seedling grass growth vigor, environment variables and equipment operation states and taking the lowest cost-efficiency ratio as a control strategy optimization target, the optimal control strategy is obtained through real-time calculation of the intelligent control strategy optimization calculation module, and accurate dynamic regulation and control of the seedling grass production process are achieved.

Description

Technical Field

[0001] The present invention relates to the field of automation and intelligent technology of facility agriculture plant factories, and in particular to an intelligent main control system and method of a seedling and grass factory. Background Art

[0002] Existing grass seedling factory control systems mostly use manual operations or simple process automation, such as timed application of water and fertilizer, timed application of supplementary lighting, and fixed target value control of temperature and humidity. They lack automated operation control of new vertical agricultural planting equipment, and lack real-time data collection, model calculation and precise regulation capabilities for the production process. This has seriously affected the production efficiency of the grass seedling factory and the variety of grass seedlings produced, and has been unable to adapt to the actual needs of producing higher yields and higher quality forage, restricting the industrial expansion of the grass seedling factory and supporting the demand for high-quality forage in the livestock industry. Summary of the invention

[0003] The present invention provides an intelligent main control system and method for a seedling grass factory, which are used to solve the problem of insufficient automation and intelligence level of the existing seedling grass factory control system and realize accurate dynamic regulation of the seedling grass production process.

[0004] On the one hand, in order to achieve the above-mentioned precise dynamic regulation of the seedling production process in the seedling factory, the present invention provides an intelligent main control system for the seedling factory. The main control system obtains the optimal control strategy through real-time calculation of the analysis data of the growth of seedlings in the seedling factory, the collection data of various environmental variables, and the collection data of equipment operation status variables through the intelligent control strategy optimization calculation module, and further drives or controls the corresponding execution unit through the PLC control module. The regulation targets include water and fertilizer control, supplementary lighting control, environmental control, and seedling planting equipment operation control.

[0005] The present invention provides an intelligent main control system for a seedling and grass factory, which includes: a remote update module, a parameter setting module, a video and image data processing module, a PLC control module, a sensor data processing module, and an intelligent control strategy optimization calculation module.

[0006] The remote update module of the main control system is used to update the main system software and parameter files. The binary file of each version of the main system software generates a specific version verifier through MD5. Similarly, the binary file of each model parameter generates a specific model verifier through MD5. When the main control system is started, the main control system will carry the remote update module to send a data packet carrying a specific version verifier and a specific model verifier to the remote server via Ethernet. After verification with the latest version verifier and model verifier stored in the remote server, if they are consistent, the update instruction will not be triggered. If they are inconsistent, the update instruction will be triggered. The update instruction is sent back to the remote update module by the remote server. After receiving the update instruction, the remote update module starts the update subroutine to download the new version of the main system software and the new version of the parameter file.

[0007] Furthermore, the remote update module can also perform validation and new version update on the running program of the PLC control module of the main control system.

[0008] The parameter setting module of the main control system is used to configure the startup parameters, runtime parameters and algorithm model parameters required by the main control system. After the main control system is powered on and the remote update is completed, the parameter setting module of the main control system reads the startup parameters and runtime parameters pre-stored in the memory, and sends the system parameter setting instructions based on the Modbus protocol to all PLC control modules through the serial port. After receiving the instructions, the PLC control module updates its operating parameters and returns the same instructions to the parameter setting module; after receiving the returned instructions, the parameter setting module verifies whether it is the same as the sent system parameter setting instructions. If they are the same, it means that the parameter setting is successful. If they are different, the system parameter setting instructions are resent.

[0009] Furthermore, the parameter setting module of the main control system reads the algorithm model parameters pre-stored in the memory, and provides them to the intelligent control strategy optimization calculation module of the main control system in the form of an internal interface for initialization and use.

[0010] The video and image data processing module of the main control system is used to collect image and video information of key positions of the seedling and grass planting equipment, and provide it to the intelligent control strategy optimization calculation module through an internal interface to calculate the evaluation value of the seedling and grass growth status.

[0011] Furthermore, the video and image data processing module can also count the planting trays of the seedling and grass planting equipment based on image recognition, and the count value is used to correct the counting error of the infrared counting sensor and the Hall proximity switch sensor of the seedling and grass planting equipment.

[0012] Furthermore, the video and image data processing module can also display the real-time camera video on the relevant display window interface of the main control system through the Hikvision video recorder, and further send the video stream data to the remote server through Ethernet for remote monitoring and call.

[0013] The PLC control module of the main control system is used for data collection of various sensors in the Miaocao factory and operation control of various equipment. The equipment connected to the PLC control module includes variable frequency high-power motors, LED lighting equipment, ultraviolet sterilization equipment, spraying equipment, cleaning equipment, pallet flipping equipment, and environmental control equipment; the sensors connected to the PLC control module include temperature sensors, humidity sensors, carbon dioxide sensors, and light intensity sensors required for environmental measurement of the Miaocao factory, and inverter torque feedback sensors, infrared counting devices, Hall proximity switches, water pressure gauges, and barometers required for equipment status monitoring. The above equipment and sensors are controlled by the PLC control module according to the set operating parameters and operation logic, thereby realizing the automated production of the Miaocao factory.

[0014] Furthermore, the PLC control module of the main control system can also record the total operating time of each device according to the start and stop status of each device, and calculate and record its actual power according to the operating current and voltage values ​​of each device, and further provide it for use by the sensor data processing module of the main control system.

[0015] The sensor data processing module of the main control system is used to preprocess the Miaocao factory environmental measurement data and Miaocao factory equipment operating parameters collected by the PLC control module. The preprocessing process includes Kalman filtering, data normalization, and data compression to form the Miaocao factory environmental status and equipment operating status with timestamps.

[0016] Furthermore, the sensor data processing module may also include a data pre-processing process, such as correction of zero drift error and temperature compensation for light intensity sensor data.

[0017] The intelligent control strategy optimization calculation module of the main control system is used to implement the model deployment, parameter reading, real-time calculation and result output of the intelligent control strategy optimization method. It calls the model configuration file to complete the deployment when the main control system is turned on, and receives the model parameter interface provided by the parameter setting module; when running, it takes the processing results of the sensor data processing module and the video and image data processing module as input, and the input values ​​include the environmental status of the seedling factory and the equipment operation status and the apparent status of the seedling, and the output values ​​include water and fertilizer strategy parameters, fill light strategy parameters, environmental control strategy parameters, and seedling planting equipment control strategy parameters. The parameters in the above output values ​​are sent by the main control system through the serial port based on the Modbus protocol control instructions to update the operating parameters in the PLC control module operation logic.

[0018] On the one hand, in order to realize the precise dynamic regulation of the seedling production process in the seedling factory, the present invention also provides a method for an intelligent main control system of the seedling factory, which adopts the artificial intelligence deep reinforcement learning theory to establish an intelligent control strategy optimization model, including the following steps:

[0019] Step S001: Initialize parameters. According to the expert knowledge of grass seedling cultivation, the growth cycle of grass seedlings is divided into multiple stages, and the growth condition parameters of grass seedlings in each growth stage are initialized, specifically represented by the initial control parameters of water, fertilizer, temperature, humidity, and supplementary light.

[0020] Step S002: Obtain a training sample data set. For any growth stage, the input is the current state of the grass plant environment, the equipment operation state, and the apparent state of the grass. On the basis of fixing other control parameters, the optimization parameter is perturbed by x and used as the action input. x is randomly selected in the interval of [-10%, +10%]. The apparent state of the grass is observed in the next growth stage as the output, which is used as a set of training data.

[0021] Step S003: Training the intelligent control strategy optimization model. Based on the training sample data set described in step S002, the intelligent control strategy optimization model is established using artificial intelligence deep reinforcement learning theory to determine the reward and punishment values ​​and state transition information, action space and corresponding action value of the reinforcement learning agent; offline training and learning are used to obtain a trained intelligent control strategy optimization model.

[0022] Step S004: Calculate and output the control strategy. According to the current state of the seedling and grass factory environment, the equipment operation state, and the apparent state of the seedling and grass, the trained intelligent control strategy optimization model is used to output the water and fertilizer strategy, the lighting strategy, the environmental control strategy, and the equipment operation strategy, and the actuator is controlled to complete the corresponding action.

[0023] Preferably, the appearance state of the grass seedlings in step S002 and step S004 is obtained by analyzing the grass seedling visual appearance model, and the steps are as follows:

[0024] Step S011-1: Collect images of grass seedlings at various growth stages under initial control parameters.

[0025] Step S011-2: Establish standard characteristic parameter values ​​for each growth stage of grass. For the collected grass images at each growth stage, the characteristic parameter value G0(t) of the grass at that growth stage is obtained through the pre-trained grass visual appearance model. The characteristic parameter values ​​include grass height, density, chromaticity (the green channel data value of the image in the quantized color space), etc.

[0026] Step S011-3: Real-time calculation of characteristic parameter values ​​and growth stages of the grass to be detected. The characteristic parameter value g(t) of the grass to be detected is obtained by using the pre-trained grass visual appearance model for the grass image collected in real time; at the same time, the growth stage of the grass is obtained based on the count value of the planting tray where the grass is located.

[0027] Step S011-4: Calculate the characteristic vector of the current grass seedling apparent state. Compare and analyze the characteristic parameter values ​​of the grass seedling to be detected with the standard characteristic parameter values ​​corresponding to its growth stage to obtain the current grass seedling growth state evaluation value, including the extent of the growth state lagging behind or ahead, and describe it as the characteristic vector K(g(t), G0(t)).

[0028] Preferably, the environmental state of the Miaocao factory described in step S002 and step S004 is obtained by analyzing the Miaocao factory environmental model, whose input is the data collected by the temperature sensor, humidity sensor, light intensity sensor, and CO2 concentration sensor. After Kalman filtering, the discrete values ​​are combined into an environmental feature vector En.

[0029] Preferably, the equipment operation status in step S002 and step S004 is obtained by analyzing the equipment operation status model, and the equipment mainly includes water and fertilizer equipment, supplementary lighting equipment, environmental control equipment, and planting equipment, and the steps are as follows:

[0030] Step S012-1: The PLC control module collects the instantaneous power P of each device d and the total running time W d .

[0031] Step S012-2: Calculate the estimated state value of the device, the formula is C d =K*P d +T(W d ), where K is the unit energy cost and T is the equipment depreciation loss calculation formula with the total operating time as a parameter.

[0032] Step S012-3: C of all devices d The values ​​are normalized to form the state feature vector E d .

[0033] Preferably, the artificial intelligence deep reinforcement learning theory described in step S002 establishes an intelligent control strategy optimization model, including: action space, action value function, state transfer module, and reward function.

[0034] Preferably, the action space perturbs the water and fertilizer parameters, the fill light parameters, the environmental control parameters, and the equipment control parameters by x, respectively, where x is a random value in the interval of [-10%, +10%].

[0035] Preferably, the action value function is used to evaluate the current state of the seedling factory environment, the equipment operation state and the seedling appearance characteristics, output the value of the corresponding different control actions under the current state, and select the optimal control action according to the current state of the seedling factory. The value function can be fitted using a deep neural network model.

[0036] Preferably, the state transfer module implements state transfer on the environmental state, equipment operation state and seedling appearance in the seedling factory at the current moment according to the state information of the seedling factory at the previous moment and the action output by the action value function.

[0037] Preferably, the reward function is determined to minimize the cost-effectiveness ratio according to the final control strategy optimization goal of the seedling factory. Let the seedling yield be expressed as a function P(si, ai) of the current state si of the seedling factory and the current action ai, then the final optimization goal is to minimize the cost-effectiveness ratio: CER=Min(Ed / P(si, ai)), where Ed is determined by step S012-3.

[0038] The present invention provides a method for an intelligent main control system of a seedling factory. The method adopts a deep convolutional neural network as the basic model of a seedling visual appearance model. The collected seedling image data is input to obtain a feature vector description of the seedling appearance characteristics; then the seedling appearance feature vector, the seedling factory environmental state feature vector, and the equipment operation state feature vector are used as inputs, and the CER predicted value is used as output to train an optimal control strategy model based on a DDPG (deep deterministic policy gradient) reinforcement learning algorithm. The action strategy with the smallest CER value in 300 model iterations is selected as the optimal control strategy.

[0039] Beneficial effects of the present invention:

[0040] 1. The characteristic expression of the production status of the Miaocao factory is established. The present invention acquires the characteristic expression of the production status of the Miaocao factory by collecting video images of the Miaocao factory, collecting the environmental status of the Miaocao factory, and collecting the equipment status during the production process of the Miaocao factory, and establishes an analysis model of the three sets of data, which can be used to evaluate the production status of the Miaocao factory in a digital way and used by the intelligent control algorithm.

[0041] 2. A mathematical description method for the production capacity and production expenditure of the Miaocao factory was established, which was constructed as a cost-effectiveness function to evaluate the production efficiency of the Miaocao factory and serve as the optimization objective function of the intelligent control strategy optimization model.

[0042] 3. An optimization model and training method for the intelligent control strategy of the seedling grass factory were established. The initial control strategy was provided by the knowledge of agricultural experts, and then a large amount of actual data of the seedling grass factory at different growth stages was used to train the strategy optimization model to continuously improve the effectiveness of the intelligent control strategy.

[0043] The optimization model of the intelligent control strategy of the seedling factory is deployed on the main control system to realize the intelligent control of the seedling factory with the goal of maximizing production efficiency, including automatic control of water and fertilizer equipment, lighting equipment, environmental control equipment, and planting equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the prior art description. The drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0045] Figure 1 A schematic diagram of the intelligent main control system architecture of the seedling and grass factory provided by the present invention;

[0046] Figure 2 A schematic diagram of a method for implementing the intelligent main control system of a seedling and grass factory provided by the present invention;

[0047] Numbers in the figure: 101-main system host; 102-PLC control subsystem; 201-state acquisition; 202-optimization calculation of intelligent control strategy; 203-strategy execution. Specific implementation methods

[0048] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0049] In one embodiment, referring to Figure 1, provides an architecture diagram of the Miaocao factory intelligent main control system, which mainly includes the main system host and the PLC control subsystem. Among them: the main system host is an industrial control platform, running 5 modules, including remote update module, parameter setting module, video and image data processing module, sensor data processing module, and intelligent control strategy optimization calculation module; the PLC control subsystem is a programmable logic control platform, running the PLC control module, which is used to control the automated operation of the Miaocao factory. The main external devices and sensors connected and controlled include variable frequency high-power motors, LED fill light equipment, ultraviolet sterilization equipment, spray equipment, cleaning equipment, pallet flipping equipment, temperature sensors, humidity sensors, infrared counting devices, Hall proximity switches, etc.

[0050] The industrial control platform is an Advantech EPC-B2205 industrial computer equipped with an I7 processor, 32G memory and a 512G solid-state hard disk. The programmable logic control platform uses a ZAN A6 series PLC programmable controller. The industrial control platform is connected to the programmable logic control platform through an RS485 interface and completes data interaction through the Modbus protocol; the industrial control platform is connected to the equipment monitoring camera through a local area network, and collects, displays and stores relevant video image data of the seedling production process.

[0051] The remote update module implementation method of the main control system can be described by the following steps:

[0052] Step S101: Maintain a remote server, store the latest version of the main system software and model parameters in the server, and obtain the latest version verifier and model verifier by MD5 encryption of the binary files of the two.

[0053] Step S102: when the main control system is started, a specific version verification code is generated for the binary file of the main system software through MD5, and a specific model verification code is generated for the binary file of each model parameter through MD5.

[0054] Step S103: the remote update module sends the data packet carrying the specific version verification symbol and the specific model verification symbol to the remote server via Ethernet.

[0055] Step S104: If the specific version verifier and the specific model verifier received by the remote server are consistent with the latest version verifier and model verifier stored in the remote server, the update instruction is not triggered; if they are inconsistent, the update instruction is triggered.

[0056] Step S105: the update instruction is sent back to the remote update module by the remote server. After receiving the update instruction, the remote update module starts the update subroutine to download the new version of the main system software and the new version of the parameter file.

[0057] Step S106: the update subroutine is completed, and the remote update ends.

[0058] The parameter setting module implementation method of the main control system can be described by the following steps:

[0059] Step S201: After the main control system is powered on and the remote update is completed, the parameter setting module of the main control system reads the startup parameters and runtime parameters pre-stored in the memory.

[0060] Step S202: Load the system parameter configuration file, and send the system parameter setting instructions based on the Modbus protocol to all PLC control modules through the serial port. The system parameters include: fill light time, fill light interval, spray time, spray interval, number of seedling and grass planting equipment sets, number of trays for a single set of equipment, and system debugging parameters: turnover time, error interval time, and the corresponding parameter values ​​are displayed on the parameter setting interface.

[0061] Step S203: After receiving the instruction, the PLC control module updates its operating parameters and returns the same instruction to the parameter setting module.

[0062] Step S204: the parameter setting module completes the parameter setting after verifying that the returned data is consistent with the sent system parameter setting instruction, otherwise the instruction is resent.

[0063] Step S205: When setting the device system debugging parameters, first select the device serial number, and the parameter setting module sends the system debugging parameter setting instruction based on the Modbus protocol to the device PLC control module of the corresponding serial number through the 485 serial port. After receiving the instruction, the device PLC control module returns the same instruction. The parameter setting module completes the setting after verifying that the returned data is consistent with the sent system debugging parameter setting instruction, otherwise resends the instruction.

[0064] Step S206: Repeat step 6 until the main control system is powered off.

[0065] The video and image data processing module implementation method of the main control system can be described by the following steps:

[0066] Step S301: Initialize the video surveillance display module, initialize the Hikvision SDK, register the Hikvision NVR device and obtain the playback handle.

[0067] Step S302: Obtain the corresponding Hikvision surveillance camera number according to the seedling planting equipment number, start the video preview function and display it on the corresponding control of the video surveillance display interface.

[0068] Step S303: Setting a dedicated video image analysis thread to perform image recognition of the planting tray, and correcting the tray count value according to the result of the image recognition.

[0069] Step S303 - 1 : The seedling tray image is collected every 10 seconds. For each frame of the collected image, it is first identified whether there is a starting tray serial number mark “1”.

[0070] Step S303-2: If the starting pallet mark is not identified, the relative position relationship between the pallet and the pre-set mark line in the image is identified. If the distance between the pallet and the mark line is shorter than the distance in the previous frame image, it means that the pallet has not passed the mark line and the count value remains unchanged.

[0071] Step S303 - 3 : If the distance between the pallet and the marking line is longer than the distance in the previous frame image, it means that the pallet is a new pallet and the previous pallet has passed the marking line, and the count value is increased by one.

[0072] Step S303 - 4 : If the start tray identifier is identified, the tray count value is reset to 1.

[0073] Step S304: When switching the seedling and grass planting equipment number, first stop the preview function of the corresponding Hikvision surveillance camera, then obtain the Hikvision surveillance camera number corresponding to the switched seedling and grass planting equipment, start the video preview function and display it on the corresponding control of the video surveillance display interface.

[0074] Step S305: The video surveillance display module function ends. First, the current camera preview is stopped, then the Hikvision NVR device is deregistered, and the SDK resources are released.

[0075] The implementation method of the PLC control module of the main control system can be described by the following steps:

[0076] Step S401: After power-on, the PLC control module receives instructions from the parameter equipment module to initialize the operating parameters, and sets seven switch states, including high motor speed, fill light state, spray state, cleaning state, turning over state, grass collecting, and seeding, to off, sets the switch state of low motor speed to on, and sets the tray count value to 0.

[0077] Step S402: Set a dedicated status query thread, send a Modbus-based device switch status query instruction to the device PLC control module through the 485 serial port every 10 seconds, and send a Modbus-based pallet count value query instruction to the device PLC control module at the same time.

[0078] Step S403: After receiving the query instruction, the PLC control module returns the device switch status value based on the Modbus protocol to the status query dedicated thread, and the main control system status display interface updates the device switch status according to the received data.

[0079] Step S404: After receiving the pallet count value query instruction, the equipment PLC control system returns the pallet count value based on the Modbus protocol to the industrial control platform, and the system status display module updates the pallet count value on the corresponding display interface according to the received data.

[0080] The sensor data processing module implementation method of the main control system can be described by the following steps:

[0081] Step S501: The status query dedicated thread receives the return value of the PLC control module.

[0082] Step S502: pre-processing process, the collected data of the environmental status is corrected for zero drift error and temperature compensation.

[0083] Step S503: Perform Kalman filtering on the pre-processed environmental state data and the original device state data, and then perform normalization processing on the filtered data.

[0084] Step S504: read the current master control system timestamp, compress the data, and form a state vector with a timestamp.

[0085] The intelligent control strategy optimization calculation module of the main control system is used to implement the model deployment, parameter reading, real-time calculation and result output of the intelligent control strategy optimization method. It calls the model configuration file to complete the deployment when the main control system is turned on, and receives the model parameter interface provided by the parameter setting module; when running, it takes the processing results of the sensor data processing module and the video and image data processing module as input, and the input values ​​include the environmental status of the seedling factory, the equipment operation status and the apparent status of the seedling, and the output values ​​include water and fertilizer strategy parameters, fill light strategy parameters, environmental control strategy parameters, and seedling planting equipment control strategy parameters. The parameters in the above output values ​​are sent by the main control system through the serial port based on the Modbus protocol control instructions to update the operating parameters in the PLC control module operation logic.

[0086] In one embodiment, referring to Figure 2 , provides a method for implementing the intelligent main control system of the seedling factory, which can be described by the following steps:

[0087] Step S601: Initialize parameters. According to the expert knowledge of grass seedling cultivation, the growth cycle of grass seedlings is divided into 7 stages, and the growth condition parameters of grass seedlings in each growth stage are initialized, specifically represented by the initial control parameters of water, fertilizer, temperature, humidity, and supplementary light.

[0088] Step S602: Obtain a training sample data set. For any growth stage, the input is the current state of the grass plant environment, the equipment operation state, and the apparent state of the grass. On the basis of fixing other control parameters, the optimization parameter is perturbed by x and used as the action input. x is randomly selected in the interval of [-10%, +10%]. The apparent state of the grass is observed in the next growth stage as the output, which is used as a set of training data.

[0089] Step S603: Train the intelligent control strategy optimization model. Based on the training sample data set described in step S602, the intelligent control strategy optimization model is established using artificial intelligence deep reinforcement learning theory to determine the reward and punishment values ​​and state transition information, action space and corresponding action value of the reinforcement learning agent; and an offline training and learning method is used to obtain a trained intelligent control strategy optimization model.

[0090] Step S604: Calculate and output the control strategy. According to the current state of the seedling and grass factory environment, the equipment operation state, and the apparent state of the seedling and grass, the trained intelligent control strategy optimization model is used to output the water and fertilizer strategy, the lighting strategy, the environmental control strategy, and the equipment operation strategy, and the actuator is controlled to complete the corresponding action.

[0091] In one embodiment, the appearance state of the seedling grass in step S602 and step S604 is obtained by analyzing the seedling grass visual appearance model, and the steps are as follows:

[0092] Step S611 - 1 : Collect images of grass seedlings at various growth stages under initial control parameters.

[0093] Step S611-2: Establish standard characteristic parameter values ​​for each growth stage of grass. For the collected grass images at each growth stage, the characteristic parameter value G0(t) of the grass at that growth stage is obtained through the pre-trained grass visual appearance model. The characteristic parameter values ​​include grass height, density, chromaticity (the green channel data value of the image in the quantized color space), etc.

[0094] Step S611-3: Real-time calculation of characteristic parameter values ​​and growth stages of the grass to be detected. The characteristic parameter value g(t) of the grass to be detected is obtained by using the pre-trained grass visual appearance model for the grass image collected in real time; at the same time, the growth stage of the grass is obtained based on the count value of the planting tray where the grass is located.

[0095] Step S611-4: Calculate the characteristic vector of the current grass seedling apparent state. Compare and analyze the characteristic parameter values ​​of the grass seedling to be detected with the standard characteristic parameter values ​​corresponding to its growth stage to obtain the current grass seedling growth state evaluation value, including the extent of the growth state lagging behind or ahead, and describe it as the characteristic vector K(g(t), G0(t)).

[0096] The environmental status of the Miaocao factory described in step S602 and step S604 is obtained by analyzing the Miaocao factory environmental model, whose input is the data collected by the temperature sensor, humidity sensor, light intensity sensor, and CO2 concentration sensor. After Kalman filtering, the discrete values ​​are combined into an environmental feature vector En.

[0097] The equipment operation status in step S602 and step S604 is obtained by analyzing the equipment operation status model. The equipment mainly includes water and fertilizer equipment, supplementary lighting equipment, environmental control equipment, and planting equipment. The steps are as follows:

[0098] Step S612-1: The PLC control module collects the instantaneous power P of each device d and the total running time W d .

[0099] Step S612-2: Calculate the estimated value of the device state, the formula is C d =K*P d +T(W d ), where K is the unit energy cost and T is the equipment depreciation loss calculation formula with the total operating time as a parameter.

[0100] Step S612-3: C of all devices d The values ​​are normalized to form the state feature vector E d .

[0101] In one embodiment, the artificial intelligence deep reinforcement learning theory described in step S602 establishes an intelligent control strategy optimization model, including: action space, action value function, state transfer module, and reward function.

[0102] The action space perturbs the water and fertilizer parameters, the fill light parameters, the environmental control parameters, and the equipment control parameters by x, respectively, where x is a random value in {-10%, -6%, -2%, +2%, +6%, +10%}.

[0103] The action value function is used to evaluate the current state of the seedling factory environment, the operating state of the equipment, and the apparent characteristics of the seedlings, output the value of different control actions corresponding to the current state, and select the optimal control action according to the current state of the seedling factory. The value function can be fitted using a deep neural network model.

[0104] The state transfer module implements state transfer on the environmental state, equipment operation state and seedling appearance in the seedling factory at the current moment according to the state information of the seedling factory at the previous moment and the action output by the action value function.

[0105] The reward function is determined to minimize the cost-effectiveness ratio according to the final control strategy optimization goal of the seedling factory. Let the seedling yield be expressed as the function P(si, ai) of the current state si of the seedling factory and the current action ai. Then the final optimization goal is to minimize the cost-effectiveness ratio: CER=Min(Ed / P(si, ai)), where Ed is determined by step S612-3.

[0106] In one embodiment, the present invention provides a method for an intelligent main control system of a seedling factory, wherein the method uses a deep convolutional neural network as the basic model of the seedling visual appearance model, and the collected seedling image data is input to obtain a feature vector description of the seedling appearance characteristics; then the seedling appearance feature vector, the seedling factory environmental state feature vector, and the equipment operation state feature vector are used as inputs, and the CER prediction value is used as output to train an optimal control strategy model based on the DDPG (deep deterministic policy gradient) reinforcement learning algorithm. The action strategy with the smallest CER value in 300 model iterations is selected as the optimal control strategy.

Claims

1. An intelligent main control system for a seedling and grass factory, characterized in that: The system includes: a remote update module, a parameter setting module, a video and image data processing module, a PLC control module, a sensor data processing module, and an intelligent control strategy optimization calculation module; through the analysis data of the growth of seedlings in the seedling factory, the collection data of various environmental variables, and the collection data of equipment operation status variables, the intelligent control strategy optimization calculation module calculates in real time to obtain the optimal control strategy, and further drives or controls the corresponding execution unit through the PLC control module. The control targets include water and fertilizer control, supplementary lighting control, environmental control, and seedling planting equipment operation control.

2. The intelligent main control system for a seedling and grass factory as claimed in claim 1, characterized in that: The video and image data processing module of the main control system is used to collect image and video information of key positions of the seedling and grass planting equipment, and provide it to the intelligent control strategy optimization calculation module through an internal interface to calculate the evaluation value of the seedling and grass growth status; The video and image data processing module can also count the planting trays of the seedling and grass planting equipment based on image recognition, and the count value is used to correct the counting error of the infrared counting sensor and the Hall proximity switch sensor of the seedling and grass planting equipment; The video and image data processing module can also display the real-time camera video on the relevant display window interface of the main control system through the video recorder, and further send the video stream data to the remote server through Ethernet for remote monitoring and call.

3. The intelligent main control system for a seedling and grass factory as claimed in claim 1, characterized in that: The intelligent control strategy optimization calculation module of the main control system is used to implement the model deployment, parameter reading, real-time calculation and result output of the intelligent control strategy optimization method. It calls the model configuration file to complete the deployment when the main control system is turned on, and receives the model parameter interface provided by the parameter setting module; when running, it takes the processing results of the sensor data processing module and the video and image data processing module as input, and the input values ​​include the environmental status of the seedling factory and the equipment operation status and the apparent status of the seedling, and the output values ​​include water and fertilizer strategy parameters, fill light strategy parameters, environmental control strategy parameters, and seedling planting equipment control strategy parameters. The parameters in the above output values ​​are sent by the main control system through the serial port based on the Modbus protocol control instructions to update the operating parameters in the PLC control module operation logic.

4. The method of the intelligent main control system of the seedling and grass factory as claimed in claim 1, characterized in that: The intelligent control strategy optimization model is established using artificial intelligence deep reinforcement learning theory, including the following steps: Step S001: Initialize parameters. According to the expert knowledge of grass seedling cultivation, the growth cycle of grass seedlings is divided into multiple stages, and the growth condition parameters of grass seedlings in each growth stage are initialized, specifically represented by the initial control parameters of water, fertilizer, temperature, humidity, and supplementary light; Step S002: Obtain a training sample data set. For any growth stage, the input is the current state of the grass plant environment, the equipment operation state, and the apparent state of the grass. On the basis of fixing other control parameters, the disturbance x of the optimization parameter is implemented as the action input, and x is randomly selected in the interval of [-10%, +10%]. The apparent state of the grass is observed in the next growth stage as the output, which is used as a set of training data. Step S003: training the intelligent control strategy optimization model. Based on the training sample data set described in step S002, the intelligent control strategy optimization model is established by using the artificial intelligence deep reinforcement learning theory to determine the reward and punishment value and state transition information, action space and corresponding action value of the reinforcement learning agent; offline training and learning are used to obtain the trained intelligent control strategy optimization model; Step S004: Calculate and output the control strategy. According to the current state of the seedling and grass factory environment, the equipment operation state, and the apparent state of the seedling and grass, the trained intelligent control strategy optimization model is used to output the water and fertilizer strategy, the lighting strategy, the environmental control strategy, and the equipment operation strategy, and the actuator is controlled to complete the corresponding action.

5. The method of the intelligent main control system of the seedling and grass factory as claimed in claim 4, characterized in that: The appearance state of Miaocao is analyzed by Miaocao visual appearance model, and the steps are as follows: Step S011-1: collecting images of grass seedlings at various growth stages under initial control parameters; Step S011-2: Establish standard characteristic parameter values ​​for each growth stage of grass seedlings. For the collected grass seedling images at each growth stage, the characteristic parameter value G0(t) of the grass seedling at that growth stage is obtained through the pre-trained grass seedling visual appearance model. Its characteristic parameter values ​​include grass height, density, chromaticity (the green channel data value of the image in the quantized color space), etc. Step S011-3: Real-time calculation of characteristic parameter values ​​and growth stages of the grass to be detected. The characteristic parameter value g(t) of the grass to be detected is obtained by using the pre-trained grass visual appearance model for the grass image collected in real time; at the same time, the growth stage of the grass is obtained based on the count value of the planting tray where the grass is located; Step S011-4: Calculate the characteristic vector of the current grass seedling apparent state. Compare and analyze the characteristic parameter values ​​of the grass seedling to be detected with the standard characteristic parameter values ​​corresponding to its growth stage to obtain the current grass seedling growth state evaluation value, including the extent of the growth state lagging behind or ahead, and describe it as the characteristic vector K(g(t), G0(t)).

6. The method of the intelligent main control system of the seedling and grass factory as claimed in claim 4, characterized in that: The equipment operation status is obtained by analyzing the equipment operation status model. The equipment mainly includes water and fertilizer equipment, supplementary lighting equipment, environmental control equipment, and planting equipment. The steps are as follows: Step S012-1: The PLC control module collects the instantaneous power P of each device d and the total running time W d ; Step S012-2: Calculate the estimated state value of the device, the formula is C d =K*P d +T(W d ), where K is the unit energy cost, and T is the equipment depreciation loss calculation formula with the total operating time as a parameter; Step S012-3: C of all devices d The values ​​are normalized to form the state feature vector E d .

7. A method for an intelligent main control system for a seedling and grass factory as claimed in claim 4, wherein the artificial intelligence deep reinforcement learning theory establishes an intelligent control strategy optimization model, including: Action space, action value function, state transition module, reward function; The action space perturbs the water and fertilizer parameters, lighting parameters, environmental control parameters, and equipment control parameters by x, where x is a random value in the range of [-10%, +10%]; The action value function is used to evaluate the current state of the seedling factory environment, the operating state of the equipment, and the apparent characteristics of the seedlings, output the value of different control actions corresponding to the current state, and select the optimal control action according to the current state of the seedling factory. The value function can be fitted using a deep neural network model; The state transfer module implements state transfer on the environment state, equipment operation state and seedling appearance in the seedling factory at the current moment according to the state information of the seedling factory at the previous moment and the action output by the action value function; The reward function is determined to minimize the cost-effectiveness ratio according to the final control strategy optimization goal of the seedling factory. Let the seedling yield be represented by the function P(si, ai) of the current state si of the seedling factory and the current action ai, and Ed represents the equipment state feature vector. Then the final optimization goal is to minimize the cost-effectiveness ratio: CER=Min(Ed / P(si, ai)).

8. A method for an intelligent main control system of a seedling factory as described in claim 7, wherein the method uses a deep convolutional neural network as the basic model of the seedling visual appearance model, and the collected seedling image data is input to obtain a feature vector description of the seedling appearance characteristics; then the seedling appearance feature vector, the seedling factory environmental state feature vector, and the equipment operation state feature vector are used as inputs, and the CER prediction value is used as output to train the optimal control strategy model based on the DDPG (deep deterministic policy gradient) reinforcement learning algorithm. The action strategy with the smallest CER value in 300 model iterations is selected as the optimal control strategy.

Images