CN112364936A

CN112364936A - Greenhouse control method, device and equipment based on artificial intelligence and storage medium

Info

Publication number: CN112364936A
Application number: CN202011376267.2A
Authority: CN
Inventors: 安志成; 罗迪君; 李蓝青
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-11-30
Filing date: 2020-11-30
Publication date: 2021-02-12

Abstract

The application provides a greenhouse control method and device based on artificial intelligence, electronic equipment and a computer readable storage medium; relates to artificial intelligence technology; the method comprises the following steps: acquiring the growth state of plants in the greenhouse in a first period, the internal climate state of the greenhouse in the first period and the external weather state of the greenhouse in the first period; invoking a machine learning model based on the growth status of the first period, the interior climate status of the first period, and the exterior weather status of the first period to obtain environmental control information for controlling the greenhouse in the second period; the environmental control information is applied to the greenhouse during a second period. Through this application, can realize the intelligent control of vegetation in the greenhouse.

Description

Greenhouse control method, device and equipment based on artificial intelligence and storage medium

Technical Field

The present application relates to artificial intelligence technology, and in particular, to a greenhouse control method and apparatus based on artificial intelligence, an electronic device, and a computer-readable storage medium.

Background

Artificial Intelligence (AI) is a comprehensive technique in computer science, and by studying the design principles and implementation methods of various intelligent machines, the machines have the functions of perception, reasoning and decision making. The artificial intelligence technology is a comprehensive subject and relates to a wide range of fields, for example, natural language processing technology and machine learning/deep learning, etc., and along with the development of the technology, the artificial intelligence technology can be applied in more fields and can play more and more important values.

In the related art, the growth environment of plants in a greenhouse is adjusted in an artificial regulation and control mode based on the experience of planting experts so as to control the growth state of the plants, and an effective scheme for regulating and controlling the growth environment of the plants in the greenhouse based on artificial intelligence is lacked.

Disclosure of Invention

The embodiment of the application provides a greenhouse control method and device based on artificial intelligence, electronic equipment and a computer readable storage medium, and intelligent control of plant growth in a greenhouse can be achieved.

The technical scheme of the embodiment of the application is realized as follows:

the embodiment of the application provides a greenhouse control method based on artificial intelligence, which comprises the following steps:

acquiring a growth state of plants in a greenhouse in a first period, an internal climate state of the greenhouse in the first period, and an external weather state of the greenhouse in the first period;

invoking a machine learning model based on the growth status of the first time period, the internal climate status of the first time period, and the external weather status of the first time period to derive environmental control information for controlling the greenhouse in a second time period, wherein the second time period is later than the first time period;

applying the environmental control information to the greenhouse during the second time period.

An embodiment of the present application provides an image target recognition apparatus, including:

the greenhouse management system comprises an acquisition module, a management module and a management module, wherein the acquisition module is used for acquiring the growth state of plants in a greenhouse in a first period, the internal climate state of the greenhouse in the first period and the external weather state of the greenhouse in the first period;

a processing module for invoking a machine learning model based on the growth status of the first period, the interior climate status of the first period, and the exterior weather status of the first period to derive environmental control information for controlling the greenhouse in a second period, wherein the second period is later than the first period;

an application module for applying the environmental control information to the greenhouse during the second period.

In the foregoing technical solution, the processing module is further configured to execute the following processing based on the machine learning model:

determining an expected growth state during the second period that meets a planting goal based on the growth state of the first period, and determining an interior climate state of the second period at which the expected growth state is achieved;

determining environmental control information for controlling the greenhouse in the second time period based on a characteristic that an external weather condition of the second time period and environmental control information for controlling the greenhouse in the second time period act together on an internal climate condition of the second time period.

In the above technical solution, the machine learning model includes a first mapping network and a fusion network; the processing module is further used for carrying out mapping processing on the growth state in the first period based on the first mapping network to obtain an expected growth state meeting a planting target in the second period;

mapping the internal climate status of the second period to environmental control information for controlling the greenhouse in the second period based on a mapping relationship between the environmental control information and the external climate status of the greenhouse together included in the converged network and the internal climate status of the greenhouse.

In the above technical solution, the machine learning model further includes a second mapping network; the processing module is further configured to map the expected growth state of the second period based on a mapping relationship between the internal climate state of the greenhouse and the growth state of the plant, which is included in the second mapping network, to obtain the internal climate state of the second period when the expected growth state of the second period is achieved; alternatively, the first and second electrodes may be,

and performing state conversion processing on the state difference between the growth state of the first period and the expected growth state of the second period based on the mapping relation between the internal climate state of the greenhouse and the growth state change of the plants, wherein the mapping relation is included in the second mapping network, so as to obtain the internal climate state of the second period when the expected growth state of the second period is realized.

In the above technical solution, the convergence network includes a first convolution layer, a second convolution layer, a full connection layer, and a third convolution layer; the processing module is further configured to perform convolution processing on the external weather state of the second period based on the first convolution layer to obtain first state information corresponding to the external weather state;

performing convolution processing on the internal climate state of the second period based on the second convolution layer to obtain second state information corresponding to the internal climate state;

determining a difference value of the second state information and the first state information based on the fully-connected layer;

and carrying out convolution processing on the difference value based on the third convolution layer to obtain environment control information for controlling the greenhouse in the second period.

In the above technical solution, the apparatus further includes:

a training module for constructing training samples of the machine learning model based on growth status of plants in the greenhouse over a plurality of historical periods, internal climate status of the greenhouse over the plurality of historical periods, external weather status of the greenhouse over the plurality of historical periods;

and training the machine learning model based on the training samples to obtain the machine learning model for predicting the environmental control information.

In the above technical solution, the training module is further configured to execute the following processing for any one of the plurality of history periods:

acquiring a growth state of a next historical period of the historical period, an internal climate state of the next historical period of the historical period, and an external weather state of the next historical period of the historical period;

combining the growth status of plants in the greenhouse during the historical period, the internal climate status of the greenhouse during the historical period, and the external weather status of the greenhouse during the historical period into first status information of the historical period;

combining a growing state of a next one of the historical periods, an interior climate state of the next one of the historical periods, and an exterior weather state of the next one of the historical periods into second state information of the next historical period;

constructing training samples for the historical period based on the first status information for the historical period, the environmental control information for controlling the greenhouse in the historical period, and the second status information for the next historical period;

and combining the training samples in the historical periods to obtain the training samples of the machine learning model.

In the above technical solution, the apparatus further includes:

the storage module is used for storing the training samples in the historical period to a cache space;

the training module is further configured to obtain a plurality of training samples in the historical period from the cache space when the number of training samples in the historical period in the cache space reaches a set threshold, and train the machine learning model based on the plurality of training samples in the historical period.

In the above technical solution, the training module is further configured to construct an objective function of the machine learning model based on a training sample in any historical period of the training samples and the labeled evaluation parameter in the historical period;

and updating the parameters of the machine learning model until the target function converges, and taking the updated parameters of the machine learning model when the target function converges as the parameters of the machine learning model for predicting the environmental control information.

In the above technical solution, the processing module is further configured to call the machine learning model to perform prediction processing based on a training sample in any historical period in the training samples, so as to obtain predicted environment control information for controlling the greenhouse in the historical period;

obtaining a prediction evaluation parameter of the historical period based on the prediction environment control information;

the training module is further used for constructing an objective function of the machine learning model based on the training samples of the historical period, the prediction evaluation parameters of the historical period and the marking evaluation parameters of the historical period.

In the above technical solution, the application module is further configured to obtain a growth state of a next historical period in the training samples of the historical period;

and calling a plant simulator model based on the growth state of the next historical period to determine growth expectation information brought by the plant growth in the historical period, and using the growth expectation information as a labeled evaluation parameter of the historical period.

In the above technical solution, the application module is further configured to obtain environmental control information for controlling the greenhouse in the historical period;

invoking the plant simulator model based on environmental control information used to control the greenhouse during the historical period to determine resource information required to be consumed by the environmental control information;

and taking the difference value of the growth expectation information and the resource information as the marked evaluation parameter of the historical period.

The embodiment of the application provides an electronic device for image target recognition, the electronic device includes:

a memory for storing executable instructions;

and the processor is used for realizing the image target identification method provided by the embodiment of the application when the executable instructions stored in the memory are executed.

The embodiment of the application provides a computer-readable storage medium, which stores executable instructions for causing a processor to execute the method for identifying the image target provided by the embodiment of the application.

The embodiment of the application has the following beneficial effects:

the machine learning model is called to obtain the environment control information for controlling the greenhouse in the second period based on the growth state of the first period, the internal climate state of the first period and the external weather state of the first period, so that the growth environment of the plants in the greenhouse is dynamically adjusted and controlled in combination with the external weather state, and intelligent and accurate control of the plant growth is realized.

Drawings

FIG. 1 is a schematic view of an application scenario of a greenhouse control system provided by an embodiment of the present application;

FIG. 2 is a schematic structural diagram of an electronic device for greenhouse control provided by an embodiment of the present application;

3-5 are schematic flow charts of artificial intelligence-based greenhouse control methods provided by embodiments of the present application;

FIG. 6 is a schematic flow chart of an artificial intelligence based greenhouse control method provided by an embodiment of the present application;

FIG. 7 is a schematic structural diagram of a machine learning model provided by an embodiment of the present application;

FIG. 8 is an architectural diagram of a platform provided by an embodiment of the present application;

FIG. 9 is a schematic diagram of a crop planting simulation provided by an embodiment of the present application;

fig. 10 is a schematic workflow diagram of a data collection module according to an embodiment of the present application.

Detailed Description

In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be described in further detail with reference to the attached drawings, the described embodiments should not be considered as limiting the present application, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

In the following description, references to the terms "first", "second", and the like are only used for distinguishing similar objects and do not denote a particular order or importance, but rather the terms "first", "second", and the like may be used interchangeably with the order of priority or the order in which they are expressed, where permissible, to enable embodiments of the present application described herein to be practiced otherwise than as specifically illustrated and described herein.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the present application only and is not intended to be limiting of the application.

Before further detailed description of the embodiments of the present application, terms and expressions referred to in the embodiments of the present application will be described, and the terms and expressions referred to in the embodiments of the present application will be used for the following explanation.

1) Convolutional Neural Networks (CNN), Convolutional Neural Networks: one class of feed Forward Neural Networks (FNNs) that includes convolution calculations and has a deep structure is one of the algorithms that represent deep learning (deep learning). The convolutional neural network has a representation learning (representation learning) capability, and can perform shift-invariant classification (shift-invariant classification) on an input image according to a hierarchical structure of the input image.

2) Environment control information: for influencing environmental factors for plant growth, e.g. temperature setting, CO₂Concentration setting, switching time of a fluorescent lamp, irrigation time and the like.

3) And (3) growth state: for characterizing the stage at which a plant is growing or for characterizing the growth vigor of a plant, such as the height of the plant, the leaf area index of the plant, the fruit weight of the plant, etc.

4) External weather conditions: for characterizing weather conditions outside the greenhouse, such as outdoor temperature, outdoor humidity, etc.

5) Interior climate conditions: for characterising the climatic conditions in greenhouses, e.g. interior temperature of the greenhouse, CO in the greenhouse₂Concentration, etc.

The intelligent greenhouse can help people to finely control the planting process of crops, and an important research direction is provided for obtaining a more excellent planting control strategy. The sources of planting strategies in the related art are mainly divided into two main categories: the first type is from planting experience of agricultural planting experts, and agricultural experts set rules in the planting process by combining past planting data and professional knowledge per se, which is also the most widely adopted mode in the current greenhouse; the second type is derived from data mining, and the best planting strategy is obtained by mining the relation between the control strategy and the planting benefit on the past planting data set in the modes of deep learning and the like.

However, the planting strategy in the related art has the following problems:

1) the planting strategies of the agricultural planting experts are expensive in cost, in addition, the planting strategies provided by the agricultural experts are generally fixed logic strategies, cannot be dynamically changed along with the planting process, and are poor in adaptability, the control granularity of the planting strategies of the planting experts is also relatively coarse, the planting strategies are generally logic combinations, for example, when the real-time temperature exceeds a certain temperature threshold value, the temperature is reduced, and the like, and fine-grained control cannot be performed by combining all state information, for example, the specific illumination intensity and the temperature setting of each hour are performed;

2) although the planting strategy based on data mining does not need the intervention of expert knowledge, and the relation is directly established through greenhouse state data and planting result data in the planting process, the method needs a large amount of agricultural data for training and is difficult to collect, the collected agricultural planting data are generally planted according to certain fixed strategies, the change is small, a strategy better than a planting expert is difficult to find on the basis, data about crops in the agricultural planting data are also small, and the conditions of the crops are difficult to be used as an important factor to determine the next planting measures.

In order to solve the above problems, embodiments of the present application provide a greenhouse control method and apparatus based on artificial intelligence, an electronic device, and a computer-readable storage medium, which can implement intelligent control of plant growth in a greenhouse.

The method based on artificial intelligence and capable of intelligently controlling the plant growth environment in the greenhouse can be independently realized by a terminal/server; the terminal and the server may cooperate with each other, for example, the terminal solely undertakes an artificial intelligence based method capable of intelligently controlling the growing environment of plants in the greenhouse and applies the environment control information to the greenhouse in the second period through the control device, or the terminal transmits a control request for the greenhouse (including the growing state of the plants in the greenhouse in the first period, the internal climate state of the greenhouse in the first period, and the external weather state of the greenhouse in the first period) to the server, and the server performs the artificial intelligence based greenhouse control method according to the received control request for the greenhouse and applies the environment control information to the greenhouse in the second period through the control device in response to the control request for the greenhouse, so as to automatically control the growing of the plants in the greenhouse.

The electronic device for greenhouse control provided by the embodiment of the application can be various types of terminal devices or servers, wherein the server can be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and a cloud server for providing cloud computing service; the terminal may be, but is not limited to, a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and the like. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the application is not limited herein.

Taking a server as an example, for example, the server cluster may be deployed in a cloud, and open an artificial intelligence cloud Service (AI as a Service, AIaaS) to users, the AIaaS platform may split several types of common AI services, and provide an independent or packaged Service in the cloud, this Service mode is similar to an AI theme mall, and all users may access one or more artificial intelligence services provided by the AIaaS platform by using an application programming interface.

For example, one of the artificial intelligence cloud services may be a greenhouse control service, that is, a server in the cloud end encapsulates the greenhouse control program provided in the embodiment of the present application. A user calls a greenhouse control service in a cloud service through a terminal (running with a client, such as a greenhouse monitoring client and the like) to enable a server deployed at the cloud to call a packaged greenhouse control program, determines environment control information for controlling a greenhouse in a second period based on a growth state of a plant in the greenhouse in the first period, an internal climate state of the greenhouse in the first period and an external weather state of the greenhouse in the first period, and applies the environment control information to the greenhouse in the second period to monitor the growth of the plant through the environment control information in the greenhouse, for example, for a greenhouse monitoring application, an internal climate state of a history period is obtained through a sensor (such as a temperature sensor, a humidity sensor and the like) in the greenhouse, an external weather state of a current period is obtained through a third application, an internal climate state of a current period is obtained based on the internal climate state of the history period, The growth state of the greenhouse in the historical period and the weather forecast of the current period, determining environment control information for controlling the greenhouse in the current period, and applying the environment control information to the greenhouse in a second period through a control device (such as a humidifying device, a warming device, a light supplement lamp and the like) so as to automatically control the growth of plants in the greenhouse.

Referring to fig. 1, fig. 1 is a schematic view of an application scenario of a greenhouse control system 10 provided in an embodiment of the present application, a terminal 200 is connected to a server 100 through a network 300, and the network 300 may be a wide area network or a local area network, or a combination of the two.

The terminal 200 (running with a client, such as a greenhouse monitoring client, etc.) may be used to acquire a control request for a greenhouse, for example, when a preset time point, i.e., a first period, is reached, the terminal 200 acquires an external weather state of the first period and an external weather state (weather forecast or real-time weather data) of a second period through a third application, the terminal 200 acquires an internal climate state of the first period (a historical period) and a growth state of the first period through a sensor in the greenhouse, and the control request for the greenhouse is automatically generated after the terminal 200 collects the internal climate state of the first period, the external weather state of the first period and the growth state of the first period.

In some embodiments, a client running in the terminal may be embedded with a greenhouse control plug-in for locally implementing an artificial intelligence-based greenhouse control method on the client. For example, after the terminal 200 acquires a control request for the greenhouse (including an internal climate state of the first period, an external weather state of the first period, and a growth state of the first period), the greenhouse control plug-in is invoked to implement an artificial intelligence based greenhouse control method, environmental control information for controlling the greenhouse in the second period is determined based on the growth state of the plants in the greenhouse in the first period, the internal climate state of the greenhouse in the first period, and the external weather state of the greenhouse in the first period, and the environmental control information is applied to the greenhouse in the second period to monitor the growth of the plants through the environmental control information in the greenhouse.

In some embodiments, after the terminal 200 acquires the control request for the greenhouse, the greenhouse control interface of the server 100 is invoked (which may be provided in the form of a cloud service, that is, a greenhouse control service), the server 100 invokes a machine learning model, determines environmental control information for controlling the greenhouse in the second period based on the growth state of plants in the greenhouse in the first period, the internal climate state of the greenhouse in the first period, and the external weather state of the greenhouse in the first period, and applies the environmental control information to the greenhouse in the second period to monitor the growth of the plants through the environmental control information in the greenhouse, for example, for a greenhouse monitoring application, the terminal 200 acquires the internal climate state of the historical period and the growth state of the historical period through sensors in the greenhouse, acquires the weather forecast of the historical period and the weather forecast of the current period through a third application, the method includes the steps that a control request for the greenhouse is automatically generated based on an internal climate state of a historical period, a growth state of the historical period and a weather forecast of the historical period, the control request for the greenhouse is sent to a server 100, the server 100 calls a machine learning model, environment control information used for controlling the greenhouse in the current period is determined based on the growth state of plants in the greenhouse in the historical period, the internal climate state of the greenhouse in the historical period and an external weather state of the greenhouse in the historical period, the environment control information in the current period is sent to a terminal 200, and the terminal 200 applies the environment control information to the greenhouse in the current period through a control device so as to automatically control the growth of the plants in the greenhouse.

The following describes a structure of an electronic device for greenhouse control provided in an embodiment of the present application, referring to fig. 2, fig. 2 is a schematic structural diagram of an electronic device 500 for greenhouse control provided in an embodiment of the present application, and taking the electronic device 500 as an example of a server, the electronic device 500 for greenhouse control shown in fig. 2 includes: at least one processor 510, memory 550, at least one network interface 520, and a user interface 530. The various components in the electronic device 500 are coupled together by a bus system 540. It is understood that the bus system 540 is used to enable communications among the components. The bus system 540 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 540 in fig. 2.

The Processor 510 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor, or the like.

The memory 550 may comprise volatile memory or nonvolatile memory, and may also comprise both volatile and nonvolatile memory. The non-volatile Memory may be a Read Only Memory (ROM), and the volatile Memory may be a Random Access Memory (RAM). The memory 550 described in embodiments herein is intended to comprise any suitable type of memory. Memory 550 optionally includes one or more storage devices physically located remote from processor 510.

In some embodiments, memory 550 can store data to support various operations, examples of which include programs, modules, and data structures, or subsets or supersets thereof, as exemplified below.

An operating system 551 including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

a network communication module 552 for communicating to other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 including: bluetooth, wireless compatibility authentication (WiFi), and Universal Serial Bus (USB), etc.;

in some embodiments, the greenhouse control device provided in the embodiments of the present application may be implemented in a software manner, for example, the greenhouse control device may be a greenhouse control plug-in the terminal described above, and may be a greenhouse control service in the server described above. Of course, without limitation, the greenhouse control device provided by the embodiments of the present application may be provided as various software embodiments, including various forms of application programs, software modules, scripts or code.

Fig. 2 shows a greenhouse control means 555 stored in a memory 550, which may be software in the form of programs and plug-ins, such as a greenhouse control plug-in, and comprises a series of modules, including an acquisition module 5551, a processing module 5552, an application module 5553, a training module 5554, and a storage module 5555; the obtaining module 5551, the processing module 5552, the application module 5553, and the storage module 5555 are configured to implement the greenhouse control function provided in the embodiment of the present application, and the training module 5554 is configured to train a machine learning model.

As described above, the artificial intelligence based greenhouse control method provided by the embodiments of the present application can be implemented by various types of electronic devices. Referring to fig. 3, fig. 3 is a schematic flowchart of an artificial intelligence based greenhouse control method provided in an embodiment of the present application, which is described with reference to the steps shown in fig. 3.

In the following steps, the external weather condition of the first period may be a real-time weather forecast acquired by the third application, that is, the external weather condition of the first period is a real weather condition of the first period, or may be past contemporary weather data, that is, the external weather condition of the first period is a simulated weather condition of the first period. Wherein the first period is a historical period relative to the second period, and the units of the first period and the second period may be months, weeks, days, or even hours.

The internal climate state is obtained by sensing the growth environment in the greenhouse through the sensor, and the growth state is obtained by sensing the growth vigor of plants in the greenhouse through the sensor.

In step 101, the growth status of plants in a greenhouse during a first period, the internal climate status of the greenhouse during the first period, and the external weather status of the greenhouse during the first period are obtained.

As shown in fig. 6, the terminal acquires the interior climate state and the growth state of the historical period (first period) with respect to the current period (second period) through sensors in the greenhouse, acquiring, by a third application, an external weather state for a first period and an external weather state for a second period, automatically generating a control request for the greenhouse based on the growth state for the first period, the internal weather state for the first period, and the external weather state for the first period, and sends a control request for the greenhouse to the server, after the server receives the control request for the greenhouse, analyzing a control request for the greenhouse to obtain a growth state in a first period, an internal climate state in the first period, and an external weather state in the first period, so that the machine learning model is invoked for dynamic optimization of the environmental control information based on the growth state of the first period, the interior climate state of the first period, and the exterior weather state of the first period.

In step 102, a machine learning model is invoked based on the growth status of the first period, the interior climate status of the first period, and the exterior weather status of the first period to derive environmental control information for controlling the greenhouse in a second period, wherein the second period is later than the first period.

For example, the growth state of the first period, the interior climate state of the first period, and the exterior weather state of the first period are input to the machine learning model such that the machine learning model learns the following strategies: the method comprises the steps of enabling environment control information and external climate states of a greenhouse to be used for the internal climate states of the greenhouse in a shared mode, enabling the internal climate states of the greenhouse to be used for the growth states of plants, conducting prediction processing through a machine learning model, obtaining environment control information used for controlling the greenhouse in a second period, namely environment control information used for controlling the greenhouse in the next period, combining the external climate states to dynamically adjust and control the growth environments of the plants in the greenhouse, and achieving intelligent and accurate control of the growth environments of the plants.

Referring to fig. 4, fig. 4 is a schematic flow chart of an alternative artificial intelligence-based greenhouse control method provided by an embodiment of the present application, and fig. 4 shows that step 102 in fig. 3 can be implemented by steps 1021 to 1023: performing the following processing based on the machine learning model: determining an expected growth state meeting the planting target in the second period based on the growth state of the first period in step 1021; in step 1022, the interior climate state for the second period when the expected growth state is achieved is determined; in step 1023, environmental control information for controlling the greenhouse in the second period is determined based on the characteristic that the external weather state of the second period and the environmental control information for controlling the greenhouse in the second period act together on the internal weather state of the second period.

For example, after the server receives a control request for a greenhouse sent by the terminal, the control request for the greenhouse is analyzed, the growth state in the first period, the internal climate state in the first period and the external weather state in the first period are obtained, a machine learning model is called for prediction processing, based on the strategy learned by the machine learning model, the prediction processing of the growth state is firstly carried out based on the growth state in the first period, and an expected growth state meeting a planting target in the second period is obtained, wherein the planting target can be set in advance according to an actual scene, for example, the yield of the plant is maximized, the profit caused by the plant is maximized, and the like. After obtaining the expected growth state of the second period, determining the internal climate state of the second period when the expected growth state is realized, and finally, determining the environmental control information for controlling the greenhouse in the second period based on the characteristic that the external weather state of the second period learned by the machine learning model and the environmental control information for controlling the greenhouse in the second period act on the internal climate state of the second period together so as to intelligently control the growth environment of the plants in the greenhouse.

As an example, the growth state of the first period is a fruit maturity of the plant of 10%, the prediction process of the growth state is performed based on the growth state of the first period based on a planting goal that profit brought by the plant is maximized, the expected growth state of the second period is a fruit maturity of 15%, and in order to achieve the expected growth state of the second period, the internal climate state of the second period is a greenhouse temperature of 28 ℃ and a carbon dioxide concentration of 300 g/m, and the external weather state of the second period is an outdoor temperature of 20 ℃ obtained by a weather forecast applied by a third party, and the environmental control information of the second period is determined as a characteristic that the temperature is set to be on to reach 28 ℃ based on the internal climate state of the second period in which the external weather state of the second period and the environmental control information for controlling the greenhouse in the second period are commonly used in the second period, CO 2₂The concentration is set to on to reach 300 g/m and the on/off time of the fluorescent lamp is set to 3 hours.

In some embodiments, the machine learning model includes a first mapping network, a second mapping network, and a converged network; mapping the growth state of the first period based on the first mapping network to obtain an expected growth state meeting the planting target in the second period; correspondingly, mapping the expected growth state of the second period based on the mapping relation between the internal climate state of the greenhouse and the growth state of the plants, wherein the mapping relation is included in the second mapping network, so that the internal climate state of the second period when the expected growth state of the second period is realized is obtained; in contrast, the interior climate status of the second period is mapped to the environment control information for controlling the greenhouse in the second period based on the mapping relationship between the environment control information and the exterior climate status of the greenhouse included in the converged network together with the interior climate status of the greenhouse.

Taking the above example into account, as shown in fig. 7, the growth state of the first period is input to the first mapping network in the machine learning model, the growth state of the first period is mapped by the first mapping network to obtain the expected growth state satisfying the planting target in the second period, the expected growth state is input to the second mapping network, the mapping relationship between the interior climate state of the greenhouse and the growth state of the plant learned by the second mapping network is mapped by the second mapping network to obtain the interior climate state of the second period when the expected growth state of the second period is achieved, the interior climate state of the second period is input to the fusion network, the mapping relationship between the environment control information of the greenhouse and the exterior climate state learned by the fusion network and the interior climate state of the greenhouse is common, and the interior climate state of the greenhouse is mapped to the environment control information for controlling the greenhouse in the second period . Therefore, the environmental control information in the second period can be predicted through the multilayer network in the machine learning model, intelligent and accurate control of the plant growth environment is achieved, and the situation that the growth environment of plants in the greenhouse is adjusted in a manual regulation mode is avoided.

In some embodiments, the machine learning model includes a first mapping network, a second mapping network, and a converged network; mapping the growth state of the first period based on the first mapping network to obtain an expected growth state meeting the planting target in the second period; on the contrary, based on the mapping relation between the internal climate state of the greenhouse and the growth state change of the plants, which is included in the second mapping network, the state conversion processing is carried out on the state difference between the growth state in the first period and the expected growth state in the second period, and the internal climate state in the second period when the expected growth state in the second period is realized is obtained; in contrast, the interior climate status of the second period is mapped to the environment control information for controlling the greenhouse in the second period based on the mapping relationship between the environment control information and the exterior climate status of the greenhouse included in the converged network together with the interior climate status of the greenhouse.

As shown in fig. 7, the growth state of the first period is input into the first mapping network in the machine learning model, the growth state of the first period is mapped through the first mapping network to obtain the expected growth state satisfying the planting target in the second period, the expected growth state is input into the second mapping network, the mapping relationship between the internal climate state of the greenhouse learned through the second mapping network and the growth state change of the plants, the state conversion processing is carried out on the state difference between the growth state of the first period and the expected growth state of the second period to obtain the internal climate state of the second period when the expected growth state of the second period is realized, the internal climate state of the second period is input into the fusion network through the second mapping network, the mapping relationship between the environmental control information and the external climate state of the greenhouse learned through the fusion network and the internal climate state of the greenhouse is obtained, the interior climate status of the second time period is mapped to environmental control information for controlling the greenhouse in the second time period. Therefore, the environmental control information in the second period can be predicted through the multilayer network in the machine learning model, intelligent and accurate control of the plant growth environment is achieved, and the situation that the growth environment of plants in the greenhouse is adjusted in a manual regulation mode is avoided.

Taking the above example, if the growth state of the first period is 10% of the fruit maturity of the plant and the expected growth state of the second period is 15% of the fruit maturity, the change in the growth state of the plant is 5% of the fruit maturity, and in order to realize the change in the growth state of the plant, the internal climate state of the second period is such that the greenhouse temperature reaches 28 ℃ and the carbon dioxide concentration reaches 300 g/m, so as to learn the relationship between the internal climate state of the greenhouse and the change in the growth state of the plant in the actual growth process of the plant through the second mapping network, thereby accurately determining the internal climate state of the second period when the expected growth state of the second period is realized, so as to accurately locate the environmental control information for controlling the greenhouse in the second period in the following.

In some embodiments, the converged network includes a first convolutional layer, a second convolutional layer, a fully-connected layer, and a third convolutional layer; mapping the interior climate status of the second time period to environmental control information for controlling the greenhouse in the second time period, comprising: performing convolution processing on the external weather state of the second period based on the first convolution layer to obtain first state information corresponding to the external weather state; performing convolution processing on the internal climate state in the second period based on the second convolution layer to obtain second state information corresponding to the internal climate state; determining a difference value of the second state information and the first state information based on the full connection layer; and performing convolution processing on the difference value based on the third convolution layer to obtain environment control information for controlling the greenhouse in the second period.

As shown in fig. 7, after the second mapping network inputs the internal climate state of the second period into the fusion network, based on the learned mapping relationship between the environmental control information and the external climate state of the greenhouse and the internal climate state of the greenhouse, the external weather state of the second period is convolved by the first convolution layer in the fusion network to obtain the first state information (i.e. the first convolution vector) corresponding to the external weather state, the internal climate state of the second period is convolved by the second convolution layer in the fusion network to obtain the second state information (i.e. the second convolution vector) corresponding to the internal climate state, then the second state information is differentiated from the first state information by the full-connection layer to obtain the difference between the second state information and the first state information, and finally the difference is convolved by the third convolution layer, environmental control information is obtained for controlling the greenhouse during the second period.

In step 103, environmental control information is applied to the greenhouse for a second period of time.

For example, the terminal sends a control request for a greenhouse to the server, the server calls the machine learning model to perform prediction processing after receiving the control request for the greenhouse, environment control information in the second period is obtained and fed back to the terminal, the terminal applies the environment control information to the greenhouse in the second period through the control device, so that the internal climate state in the first period is converted into the internal climate state in the second period, and the growth state in the first period is converted into the growth state in the second period under the combined action of the internal climate state in the second period and the external climate state in the second period, and therefore growth of plants in the greenhouse is automatically controlled.

Referring to fig. 5, fig. 5 is an alternative flowchart of an artificial intelligence based greenhouse control method provided by an embodiment of the present application, fig. 5 illustrates a training method for a machine learning model, and fig. 5 illustrates that fig. 3 further includes steps 104-105: in step 104, constructing training samples of a machine learning model based on the growth state of plants in the greenhouse in a plurality of historical periods, the internal climate state of the greenhouse in the plurality of historical periods, and the external weather state of the greenhouse in the plurality of historical periods; in step 105, the machine learning model is trained based on the training samples, and the machine learning model for environment control information prediction is obtained.

For example, the state of the current period (including the growth state, the interior climate state, and the exterior weather state) is taken as the state of the historical period of the next period. The states (including the growth state, the internal climate state, and the external weather state) in one growth cycle of the plant growth in the greenhouse can be used as training samples to train the machine learning model, that is, a plurality of historical periods belong to one growth cycle of the plant growth, for example, the plant growth cycle is one month, and the environmental control information is updated in units of days, so that the number of the plurality of historical periods is 30, that is, day 1, day 2, and day …, day 30.

The plurality of history periods may be configured by a plurality of cycles, for example, the plurality of history periods are a 1 st history period, a2 nd history period and a 3 rd history period, and the 1 st history period is any period in a 1 st growth cycle of the plant, the 2 nd history period is any period in a2 nd growth cycle of the plant, and the 3 rd history period is any period in a 3 rd growth cycle of the plant.

In some embodiments, constructing training samples of a machine learning model based on growth status of plants in the greenhouse over a plurality of historical periods, interior climate status of the greenhouse over the plurality of historical periods, exterior weather status of the greenhouse over the plurality of historical periods comprises: performing the following processing for any one of a plurality of history periods: acquiring a growth state of a next historical period of the historical period, an internal climate state of the next historical period of the historical period, and an external weather state of the next historical period of the historical period; combining a growth state of a plant in a greenhouse in a historical period, an internal climate state of the greenhouse in the historical period, and an external weather state of the greenhouse in the historical period into first state information of the historical period; combining the growing state of the next historical period of the historical period, the internal climate state of the next historical period of the historical period and the external weather state of the next historical period of the historical period into second state information of the next historical period; constructing training samples of a historical period based on first state information of the historical period, environmental control information for controlling the greenhouse in the historical period and second state information of the next historical period; and combining the training samples in the plurality of historical periods to obtain the training samples of the machine learning model.

For example, when the growth state of a certain historical period is x₁Interior climate State y₁External weather condition is z₁If so, the state information of the history period is s₁[x₁，y₁，z₁]. When the growth state of the next history period of the history period is x₂Interior climate State y₂External weather condition is z₂If so, the state information of the next historical period of the historical period is s₂[x₂，y₂，z₂]. When the environmental control information for controlling the greenhouse in the history period is a₁Then the growth state s of the history period is determined₁And the environmental control information for controlling the greenhouse in the historical period is a₁And status information s of the next one of the historical periods₂Combined as training samples [ s ] for the historical period₁，a₁，s₂]. When there are N historical periods, then the training sample is [ s ]₁，a₁，s₂]、…、[s_N-1，a_N-1，s_N]、[s_N，a_NAt the end state]Wherein N is a natural number greater than 2, and the termination state represents a training termination condition.

In the process of dynamically adjusting and controlling the growing environment of plants in the greenhouse, training samples in historical periods can be stored in the cache space to collect the training samples, so that when the number of the training samples in the historical periods in the cache space reaches a set threshold value, a plurality of training samples in the historical periods are obtained from the cache space, and the machine learning model is trained on the basis of the training samples in the historical periods.

In order to save the buffer space, the buffer space may be periodically cleaned, for example, when the validity period of the training sample in the buffer space reaches, the training sample is deleted; obtaining the importance degree of the training sample in the buffer space (for example, the change degree of the growth state in adjacent periods in the training sample indicates that the training sample is more important when the change degree is larger, which indicates that the plant growth is better), and deleting the training sample from the buffer space when the importance degree of the training sample is lower than an importance degree threshold.

In some embodiments, training the machine learning model based on training samples of plant growth in the greenhouse to obtain the machine learning model for environmental control information prediction comprises: constructing an objective function of the machine learning model based on the training samples in any historical period in the training samples and the labeled evaluation parameters in the historical period; and updating parameters of the machine learning model until the target function converges, and taking the updated parameters of the machine learning model when the target function converges as the parameters of the machine learning model for predicting the environmental control information.

For example, with the reinforcement learning algorithm, after determining the value of the objective function of the machine learning model based on the training samples of any historical period and the labeled evaluation parameters of the historical period (for example, the actual profit of the plant in the historical period, or the actual yield of the plant in the historical period), it may be determined whether the value of the objective function of the machine learning model exceeds a preset threshold, when the value of the objective function of the machine learning model exceeds the preset threshold, an error signal of the machine learning model is determined based on the objective function of the machine learning model, the error information is propagated in the machine learning model in a reverse direction, and the model parameters of each layer are updated in the process of propagation.

Describing backward propagation, inputting training sample data into an input layer of a neural network model, passing through a hidden layer, finally reaching an output layer and outputting a result, which is a forward propagation process of the neural network model, wherein because the output result of the neural network model has an error with an actual result, an error between the output result and the actual value is calculated and is propagated backward from the output layer to the hidden layer until the error is propagated to the input layer, and in the process of backward propagation, the value of a model parameter is adjusted according to the error; and continuously iterating the process until convergence. Wherein, the machine learning model belongs to a neural network model.

In some embodiments, before constructing the objective function of the machine learning model, the method further includes: calling a machine learning model to perform prediction processing based on a training sample in any historical period in the training samples to obtain predicted environment control information for controlling the greenhouse in the historical period; obtaining a prediction evaluation parameter of a historical period based on the prediction environment control information; and constructing an objective function of the machine learning model based on the training samples of the historical period, the prediction evaluation parameters of the historical period and the marking evaluation parameters of the historical period.

For example, in a training phase, calling a machine learning model to perform prediction processing to obtain predicted environment control information for controlling a greenhouse in a historical period, and obtaining a predicted evaluation parameter of the historical period based on the predicted environment control information, for example, predicting a growth state that a plant can reach in the greenhouse based on the predicted environment control information, and taking growth expectation information brought by the growth state that the plant can reach as the predicted evaluation parameter; and predicting the growth state which can be reached by the plants in the greenhouse based on the predicted environment control information, taking growth expectation information brought by the growth state which can be reached by the plants as a predicted evaluation parameter, predicting resource information which needs to be consumed by the environment control information, and taking the difference value between the growth expectation information and the resource information as a labeled evaluation parameter. And finally, constructing an objective function of the machine learning model based on the training samples of the historical period, the prediction evaluation parameters of the historical period and the marking evaluation parameters of the historical period.

As an example, the growth state of the history period is mapped based on the first mapping network in the machine learning model, an expected growth state satisfying the planting target in the next history period is obtained, and the internal climate state of the next history period is mapped to the predicted environmental control information for controlling the greenhouse in the next history period based on the mapping relationship between the environmental control information and the external climate state of the greenhouse, which are included in the fusion network in the machine learning model, and the internal climate state of the greenhouse.

For example, after determining the value of an objective function (e.g., a cross entropy loss function) of the machine learning model based on a training sample of any one of the historical periods, a predicted evaluation parameter of the historical period, and a labeled evaluation parameter of the historical period (e.g., a profit actually brought by a plant in the historical period, or a yield actually brought by the plant in the historical period), it may be determined whether the value of the objective function of the machine learning model exceeds a preset threshold, when the value of the objective function of the machine learning model exceeds the preset threshold, an error signal of the machine learning model is determined based on the objective function of the machine learning model, the error information is propagated in the machine learning model in reverse, and the model parameters of each layer are updated in the process of propagation.

In connection with the above example, before constructing the objective function of the machine learning model, the method further includes: and acquiring the growth state of the next historical period in the training samples of the historical periods, calling a plant simulator model based on the growth state of the next historical period to determine the growth expectation information brought by the plant growth in the historical periods, and taking the growth expectation information as the labeled evaluation parameter of the historical periods. For example, the actual yield of the plant during the historical period is used as the marker evaluation parameter for the historical period.

In connection with the above example, before constructing the objective function of the machine learning model, the method further includes: acquiring the growth state of the next historical period in the training samples of the historical period, and calling a plant simulator model based on the growth state of the next historical period to determine the growth expectation information brought by the plant growth in the historical period; acquiring environmental control information for controlling a greenhouse in a historical period; calling a plant simulator model based on environmental control information for controlling the greenhouse in the historical period to determine resource information required to be consumed by the environmental control information, and taking the growth expectation information as a labeled evaluation parameter of the historical period, wherein the labeled evaluation parameter comprises the following steps: and taking the difference value of the growth expectation information and the resource information as the marked evaluation parameter of the historical period. For example, the actual profit that the plant actually brings during the historical period serves as a parameter for the annotated assessment of the historical period.

In the following, an exemplary application of the embodiments of the present application in a practical greenhouse crop planting application scenario will be described.

As the population grows and urbanization progresses, fine agriculture has become more and more important. Through less human input, more intelligent greenhouse drive mode promotes agricultural productivity, reduces the consumption of resource, satisfies more population demands, makes people live more healthy life, is the new direction of agriculture that all countries are developing vigorously. A plurality of intelligent greenhouses are developed at present, the greenhouses sense the environment inside and outside the greenhouse through the use of a plurality of sensors, and intelligent management and control in the crop planting process are realized through a series of modes such as heaters, illuminating lamps and drip irrigation water pipes.

The embodiment of the application provides an accurate and perfect greenhouse planting model, can accurately model the external weather and the environment of an internal greenhouse, and model the complete growth cycle of crops, and even can obtain greenhouse state data and crop state data of minute level. After a complete growth cycle of the crop, the training of the planting strategy can be performed using reinforcement learning algorithms that, unlike supervised learning, emphasize environmental-based actions to maximize the desired benefit, and that seek a balance between environmental exploration and the already explored experience, well suited for the planning of the control strategy (i.e., the planting strategy). In the process of exploration of reinforcement learning, the limit of human beings can be broken through, and planting modes which are not discovered by the human beings are found, so that greater benefits are obtained.

The embodiment of the application provides a method and a platform (system) for greenhouse crop planting strategies, and the method and the platform can be applied to the fields of intelligent agriculture, greenhouse fine control and the like. A user establishes a specific simulated greenhouse planting environment by using the platform, trains and tests different planting strategies in the environment, compares the planting strategies, adjusts parameters and the like, and deploys the planting strategies into a real agricultural greenhouse after a mature and excellent-performance planting strategy is obtained. The planting strategy can make judgment based on the external weather condition of the greenhouse, the internal greenhouse state (internal climate state), the self growth state of crops and the like to heat or cool, and CO is used₂And the fine control of multiple dimensions such as release, irrigation, illumination and the like is realized, so that better planting benefits are obtained.

As shown in fig. 8, the platform provided in the embodiment of the present application is composed of 4 modules, which are an environment module, a policy module, a data collection module, and an experiment module. The environment module is used for creating a reinforcement learning environment simulating a real crop planting process; the strategy module provides a plurality of components for realizing and constructing the reinforcement learning strategy by reinforcement learning methods, and a user can create a planting strategy based on the module; the data collection module is used for controlling the interaction between the strategy module and the environment module and collecting data in the interaction process; the experiment module is the top-level package of the whole platform, is divided into two parts of a training experiment and a testing experiment, and regulates the whole process of the training and the testing, such as self-updating of the planting strategy after how many steps the strategy module interacts with the environment module in the training process, termination conditions of the training of the planting strategy, the number of interaction steps in the testing process and the like.

The following specifically describes a specific interaction flow of the environment module, the policy module, the data collection module, and the experiment module:

firstly, an environment module is used for generating a simulated agricultural planting environment, then an enhanced learning agricultural planting strategy is built based on a strategy module, then a data collection module controls the environment module to interact with the strategy module, the current environment state (including an external weather state, a greenhouse internal climate state, a crop growth state and the like) is provided for the planting strategy, the planting strategy calculates an action value (namely environment control information including control measures such as heating, ventilation, irrigation, lighting and the like) based on the state, the simulated environment receives the action value and executes the action value, then the current environment state is converted to a next state, an award value (namely an evaluation parameter, such as the yield of crops in the next state, the profit brought by the crops in the next state and the like) is returned to represent the single step effect of the current action, and when the converted next state is an environment termination state, the current environment will be reset to the initial state; and when the next state after the conversion is not the environment termination state, the next state after the conversion of the current environment is used as the input again and is transmitted into the strategy module, the process is continued, and the steps are repeated in a circulating way.

The data collection module continuously collects the records (the state value, the action value, the reward value, the environment termination identifier and the next state value) in the interaction process of the strategy module and the environment module and stores the records into a data cache pool (cache space), all the records in each complete planting period form a planting track, when the records in the data cache pool reach a certain set value, the interaction between the strategy module and the environment module is suspended, and the data in the data cache pool is used for training the planting strategy.

Continuing with the training process of reinforcement learning, to better illustrate the training process of planting strategy, the related definitions are symbolized as shown in table 1:

TABLE 1

Wherein, the strategy training process adopts a standard reinforcement learning strategy gradient algorithm, and the reinforcement learning aims at obtaining a planting strategyThis planting strategy can select an optimal action to maximize long-term value based on the current state. The strategy gradient algorithm models and optimizes this planting strategy directly, which can be expressed as a function pi parameterized by theta_θ(as) indicates that in the s state, the planting strategy will give a corresponding action value a. And the objective function to be maximized is the cumulative reward obtained by implementing the planting strategy pi, and the objective function is shown in formula (1):

J(θ)＝∑_s∈Sd^π(s)∑_a∈Aπ_θ(a|s)Q^π(s,a) (1)

wherein d is^π(s) denotes a planting strategy of π_θThe smooth distribution of the drawn markov chain (i.e., the probability of each state occurring in the state space).

To solve the maximization problem, a gradient ascending algorithm can be adopted to make the parameter theta gradient

Changing the given direction to find out the optimal theta and make the corresponding planting strategy pi_θThe accumulated prize of (2) is the largest. Through calculation and simplification, a general form of the gradient of the objective function can be obtained, as shown in equation (2):

the gradient strategy algorithm comprises the following steps:

step 1, randomly initializing planting strategy parameters theta

And 2, when the strategy is not converged, executing the following steps:

step a): one complete planting trajectory S generated using the current mid-planting strategy₁,A₁,R₁,S₂,A₂,R₂,…,S_T,A_T,R_T

Step b): for each time step T, where T is greater than or equal to 1 and less than or equal to T, the following steps are performed:

the method comprises the following steps: estimating cumulative returns

Step two: updating parameters

The embodiment of the present application is not limited to the above training process method, and may also be other training methods based on records (state value, action value, reward value, environment termination identifier, next state value) in the data cache pool.

The following specifically describes the implementation of each specific module in the platform:

(1) environment module

The environment module provides a simulated agricultural environment, and the core of the environment module is a greenhouse crop planting simulator. The simulator is built based on a real agricultural greenhouse, which is different from a real crop greenhouse in that the growth state of the crop is calculated by a planting strategy. The simulator operable part comprises a temperature setting, CO₂Concentration setting, daylight lamp on-off time, irrigation time and other 44 dimensions (action space) can be controlled comprehensively and finely, and a user can also set the outside weather of the greenhouse. The observation space of the simulator comprises the temperature and the humidity outside the greenhouse, the temperature inside the greenhouse and CO in the greenhouse₂Concentration, crop leaf area index, dry weight, wet weight of fruits, planting cost, crop profit and other 38 dimensions (state space), and can accurately describe the weather outside the greenhouse, the climate inside the greenhouse, the state of the crops, economic benefits and the like.

In summary, the simulator can perform a complete cycle and complete description on crop planting, and the simulation process is shown in fig. 9, and the basic logic is that greenhouse control and external weather affect the greenhouse state, and the greenhouse state affects the crop growth state.

The platform in the embodiment of the application is further packaged on the basis of the simulator, when the platform is actually used by a user, a state space and an action space of an environment can be selected in a configuration file mode, a new action value function can be customized outside a default action value function, and simulated planting environments created by the environment module all meet the interface standard of OpenAI Gym.

(2) Policy module

The strategy module is used for helping a user build a planting strategy of reinforcement learning, and the reinforcement learning tends to approximate an action value function or a state value function by using a neural network. Therefore, the strategy module can provide a plurality of neural network components, and a user can utilize the components to rapidly realize the reinforcement learning strategy so as to more fully utilize data.

In addition, the strategy module also implements reinforced learning algorithms, such as depth Deterministic strategy Gradient (DDPG), dominant action review algorithm (A2C), dual delay Deep Deterministic strategy Gradient (TD 3), etc., which can be directly used for training and testing as a baseline for comparison experiments.

(3) Data collection module

As shown in FIG. 10, the data collection module provides data collection and storage functions, including data collectors and data buffers (including buffer pools). The data collector defines the interaction process of the strategy module and the environment module, and the data buffer is used for storing records of state values, action values, reward values, environment termination marks and the like in the interaction process. The data collector and the data buffer have data additional processing, for example, the data collection module also provides a parallel collection function, so that the collection speed can be increased.

(4) Experiment module

The experiment module is the top-level package of the whole platform and controls the interaction logic among the modules in the platform. The experiment module comprises a training part and a testing part, wherein the training part: the experiment module specifies interactive logic between the data collector and the strategy module, for example, updating of the planting strategy after how many steps are collected, and specifies various parameters in the training process, such as the maximum capacity of a cache pool, a convergence threshold value and the like; test part: interaction is carried out between the test strategy module and the test environment module through control, and selected observation values in the interaction are recorded.

In summary, the method and the platform for greenhouse crop planting strategy provided by the embodiment of the application have the following beneficial effects:

1) the platform provides a refined greenhouse agricultural planting simulation environment and a relevant reinforcement learning tool, can help a user to rapidly develop and test a reinforcement learning agricultural planting strategy, and the screened planting strategy can be deployed in a real greenhouse to improve the planting benefit of the real greenhouse;

2) the method has a wider strategy search space, almost does not need actual planting data or experience of agricultural experts, can carry out self-learning purely through an algorithm, makes up the disadvantages of agricultural planting strategy development in the related technology, reduces the cost and improves the development efficiency;

3) the core of the agricultural planting environment provided by the platform is a greenhouse crop planting simulator, and the simulator can simulate the real greenhouse environment and the crop growth process more comprehensively and accurately.

The artificial intelligence based greenhouse control method provided by the embodiment of the application has been described in connection with the exemplary application and implementation of the server provided by the embodiment of the application. In practical applications, each functional module in the greenhouse control apparatus may be cooperatively implemented by hardware resources of an electronic device (such as a terminal device, a server, or a server cluster), such as computing resources like a processor, communication resources (such as for supporting communications in various manners like optical cables and cellular), and a memory. Fig. 2 shows a greenhouse control means 555 stored in a memory 550, which may be software in the form of programs and plug-ins, e.g. software C/C + +, software modules designed in a programming language such as Java, application software designed in a programming language such as C/C + +, Java, or dedicated software modules in a large software system, application program interfaces, plug-ins, cloud services, etc., as exemplified below for different implementations.

Example one greenhouse control device is a Mobile-side application and Module

The greenhouse control device 555 in the embodiment of the present application may be provided as a software module designed using a programming language such as software C/C + +, Java, or the like, and may be embedded in various mobile applications based on systems such as Android or iOS (stored in a storage medium of the mobile terminal as an executable instruction and executed by a processor of the mobile terminal), so as to directly use computing resources of the mobile terminal itself to complete related information recommendation tasks, and periodically or aperiodically transmit processing results to a remote server through various network communication methods, or locally store the processing results in the mobile terminal.

Example two, the greenhouse control device is a Server application and platform

The greenhouse control device 555 in the embodiment of the present application may be provided as application software designed using a programming language such as C/C + +, Java, or the like, or a dedicated software module in a large-scale software system, and run on the server side (stored in a storage medium of the server side in the form of executable instructions and run by a processor of the server side), and the server uses its own computing resources to complete the relevant information recommendation task.

The embodiment of the application can also provide a method for forming an information recommendation platform (used for a recommendation list) and the like for individuals, groups or units to use by carrying a customized and easily interactive network (Web) Interface or other User Interfaces (UI) on a distributed and parallel computing platform formed by a plurality of servers.

Third, the greenhouse control device is a server side Application Program Interface (API) and a plug-in

The greenhouse control device 555 in the embodiment of the present application can be provided as an API or a plug-in on a server side for a user to call, so as to execute the greenhouse control method based on artificial intelligence in the embodiment of the present application, and be embedded into various application programs.

Example four greenhouse control device is Mobile device client API and plug-in

The greenhouse control apparatus 555 in the embodiment of the present application may be provided as an API or a plug-in on the mobile device side for a user to call to execute the artificial intelligence based greenhouse control method in the embodiment of the present application.

Example five greenhouse control device is cloud open service

The greenhouse control device 555 in the embodiment of the application can provide information recommendation cloud service developed for a user, so that individuals, groups or units can obtain recommendation lists.

The greenhouse control device 555 includes a series of modules, including an obtaining module 5551, a processing module 5552, an application module 5553, a training module 5554, and a storage module 5555. The following continues to describe how each module in the greenhouse control device 555 provided in the embodiment of the present application cooperates to implement a greenhouse control scheme.

An obtaining module 5551, configured to obtain a growth status of plants in a greenhouse in a first period, an internal climate status of the greenhouse in the first period, and an external weather status of the greenhouse in the first period; a processing module 5552 for invoking a machine learning model based on the growth status of the first period, the interior climate status of the first period, and the exterior weather status of the first period to derive environmental control information for controlling the greenhouse in a second period, wherein the second period is later than the first period; an application module 5553 for applying the environmental control information to the greenhouse during the second period.

In some embodiments, the processing module 5552 is further configured to perform the following based on the machine learning model: determining an expected growth state during the second period that meets a planting goal based on the growth state of the first period, and determining an interior climate state of the second period at which the expected growth state is achieved; determining environmental control information for controlling the greenhouse in the second time period based on a characteristic that an external weather condition of the second time period and environmental control information for controlling the greenhouse in the second time period act together on an internal climate condition of the second time period.

In some embodiments, the machine learning model includes a first mapping network and a converged network; the processing module 5552 is further configured to map the growth status of the first period based on the first mapping network, so as to obtain an expected growth status meeting a planting target in the second period; mapping the internal climate status of the second period to environmental control information for controlling the greenhouse in the second period based on a mapping relationship between the environmental control information and the external climate status of the greenhouse together included in the converged network and the internal climate status of the greenhouse.

In some embodiments, the machine learning model further comprises a second mapping network; the processing module 5552 is further configured to map the expected growth state of the second period based on a mapping relationship between the internal climate state of the greenhouse and the growth state of the plant, which is included in the second mapping network, to obtain the internal climate state of the second period when the expected growth state of the second period is achieved; or, based on the mapping relationship between the internal climate state of the greenhouse and the change of the growth state of the plant, which is included in the second mapping network, performing state conversion processing on the state difference between the growth state of the first period and the expected growth state of the second period to obtain the internal climate state of the second period when the expected growth state of the second period is realized.

In some embodiments, the converged network includes a first convolutional layer, a second convolutional layer, a fully-connected layer, and a third convolutional layer; the processing module 5552 is further configured to perform convolution processing on the external weather state in the second period based on the first convolution layer, so as to obtain first state information corresponding to the external weather state; performing convolution processing on the internal climate state of the second period based on the second convolution layer to obtain second state information corresponding to the internal climate state; determining a difference value of the second state information and the first state information based on the fully-connected layer; and carrying out convolution processing on the difference value based on the third convolution layer to obtain environment control information for controlling the greenhouse in the second period.

In some embodiments, the greenhouse control device 555 further comprises: a training module 5554 for constructing training samples of the machine learning model based on the growth status of plants in the greenhouse over a plurality of historical periods, the internal climate status of the greenhouse over the plurality of historical periods, and the external weather status of the greenhouse over the plurality of historical periods; and training the machine learning model based on the training samples to obtain the machine learning model for predicting the environmental control information.

In some embodiments, the training module 5554 is further configured to perform the following for any one of the plurality of historical time periods: acquiring a growth state of a next historical period of the historical period, an internal climate state of the next historical period of the historical period, and an external weather state of the next historical period of the historical period; combining the growth status of plants in the greenhouse during the historical period, the internal climate status of the greenhouse during the historical period, and the external weather status of the greenhouse during the historical period into first status information of the historical period; combining a growing state of a next one of the historical periods, an interior climate state of the next one of the historical periods, and an exterior weather state of the next one of the historical periods into second state information of the next historical period; constructing training samples for the historical period based on the first status information for the historical period, the environmental control information for controlling the greenhouse in the historical period, and the second status information for the next historical period; and combining the training samples in the historical periods to obtain the training samples of the machine learning model.

In some embodiments, the greenhouse control device 555 further comprises: a storage module 5555, configured to store the training samples of the historical period in a buffer space; the training module 5554 is further configured to, when the number of training samples in the historical period in the cache space reaches a set threshold, obtain a plurality of training samples in the historical period from the cache space, and train the machine learning model based on the plurality of training samples in the historical period.

In some embodiments, the training module 5554 is further configured to construct an objective function of the machine learning model based on training samples of any historical period in the training samples and the labeled evaluation parameters of the historical period; and updating the parameters of the machine learning model until the target function converges, and taking the updated parameters of the machine learning model when the target function converges as the parameters of the machine learning model for predicting the environmental control information.

In some embodiments, the processing module 5552 is further configured to invoke the machine learning model for prediction processing based on a training sample of any historical period in the training samples, resulting in predicted environmental control information for controlling the greenhouse in the historical period; obtaining a prediction evaluation parameter of the historical period based on the prediction environment control information; the training module 5554 is further configured to construct an objective function of the machine learning model based on the training samples of the historical time period, the predicted evaluation parameters of the historical time period, and the labeled evaluation parameters of the historical time period.

In some embodiments, the application module 5553 is further configured to obtain a growth status of a next historical period in the training samples of the historical period; and calling a plant simulator model based on the growth state of the next historical period to determine growth expectation information brought by the plant growth in the historical period, and using the growth expectation information as a labeled evaluation parameter of the historical period.

In some embodiments, the application module 5553 is further configured to obtain environmental control information for controlling the greenhouse during the historical period; invoking the plant simulator model based on environmental control information used to control the greenhouse during the historical period to determine resource information required to be consumed by the environmental control information; and taking the difference value of the growth expectation information and the resource information as the marked evaluation parameter of the historical period.

Embodiments of the present application provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the artificial intelligence based greenhouse control method according to the embodiment of the application.

Embodiments of the present application provide a computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to perform an artificial intelligence based greenhouse control method provided by embodiments of the present application, for example, the artificial intelligence based greenhouse control method as shown in fig. 3-5.

In some embodiments, the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.

In some embodiments, executable instructions may be written in any form of programming language (including compiled or interpreted languages), in the form of programs, software modules, scripts or code, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

By way of example, executable instructions may correspond, but do not necessarily have to correspond, to files in a file system, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).

By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.

The above description is only an example of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. An artificial intelligence based greenhouse control method, characterized in that the method comprises:

2. The method of claim 1, wherein invoking a machine learning model based on the growth status of the first time period, the interior climate status of the first time period, and the exterior weather status of the first time period to derive environmental control information for controlling the greenhouse in a second time period comprises:

performing the following processing based on the machine learning model:

3. The method of claim 2,

the machine learning model comprises a first mapping network and a fusion network;

said determining an expected growth state meeting a planting goal in said second period based on the growth state of said first period comprises:

mapping the growth state of the first period based on the first mapping network to obtain an expected growth state meeting a planting target in the second period;

said determining environmental control information for controlling said greenhouse in said second period of time comprises:

4. The method of claim 3,

the machine learning model further comprises a second mapping network;

said determining the interior climate state for the second period of time at which the expected growth state is achieved comprises:

mapping the expected growth state of the second period based on the mapping relation between the internal climate state of the greenhouse and the growth state of the plant, wherein the mapping relation is included in the second mapping network, so that the internal climate state of the second period when the expected growth state of the second period is realized is obtained; alternatively, the first and second electrodes may be,

5. The method of claim 3,

the fusion network comprises a first convolution layer, a second convolution layer, a full-connection layer and a third convolution layer;

said mapping the interior climate status of the second time period to environmental control information for controlling the greenhouse in the second time period comprises:

performing convolution processing on the external weather state of the second period based on the first convolution layer to obtain first state information corresponding to the external weather state;

6. The method of claim 1, further comprising:

constructing training samples of the machine learning model based on growth states of plants in the greenhouse over a plurality of historical periods, internal climate states of the greenhouse over the plurality of historical periods, and external weather states of the greenhouse over the plurality of historical periods;

7. The method of claim 6, wherein constructing training samples of the machine learning model based on the growth status of plants in the greenhouse over a plurality of historical periods, the internal climate status of the greenhouse over the plurality of historical periods, and the external weather status of the greenhouse over the plurality of historical periods comprises:

performing the following processing for any one of the plurality of history periods:

8. The method of claim 7,

the method further comprises the following steps:

storing the training samples of the historical period in a cache space;

the training the machine learning model based on the training samples includes:

when the number of the training samples in the historical period in the cache space reaches a set threshold value, obtaining a plurality of training samples in the historical period from the cache space, and training the machine learning model based on the plurality of training samples in the historical period.

9. The method of claim 7, wherein training a machine learning model based on training samples of plant growth in the greenhouse, resulting in the machine learning model for environmental control information prediction, comprises:

constructing an objective function of the machine learning model based on training samples of any historical period in the training samples and the labeled evaluation parameters of the historical period;

10. The method of claim 9,

before the constructing the objective function of the machine learning model, the method further includes:

calling the machine learning model to perform prediction processing based on a training sample in any historical period in the training samples to obtain predicted environment control information for controlling the greenhouse in the historical period;

the constructing of the objective function of the machine learning model comprises:

and constructing an objective function of the machine learning model based on the training samples of the historical period, the prediction evaluation parameters of the historical period and the marking evaluation parameters of the historical period.

11. The method of claim 9 or 10, wherein before constructing the objective function of the machine learning model, further comprising:

acquiring the growth state of the next historical period in the training samples of the historical period;

12. The method according to claim 11, wherein before the using the growth expectation information as the labeled evaluation parameter of the historical period, further comprising:

obtaining environmental control information for controlling the greenhouse during the historical period;

the using the growth expectation information as the labeled evaluation parameter of the historical period comprises:

13. A greenhouse control apparatus, characterized in that the apparatus comprises:

14. An electronic device, characterized in that the electronic device comprises:

a memory for storing executable instructions;

a processor for implementing the artificial intelligence based greenhouse control method of any one of claims 1 to 12 when executing executable instructions stored in the memory.

15. A computer-readable storage medium storing executable instructions for implementing the artificial intelligence based greenhouse control method of any one of claims 1 to 12 when executed by a processor.