Greenhouse plane field environment simulation system based on machine learning

Abstract

The invention relates to the technical field of greenhouse environment intelligent control, in particular to a greenhouse plane field environment simulation system based on machine learning, which comprises a data acquisition unit used for acquiring greenhouse environment parameter data and control equipment operation data, and the control equipment operation data comprises working state combinations of a plurality of pieces of control equipment; and the data processing cloud platform is in communication connection with the data acquisition unit, and is used for processing the greenhouse environment parameter data and the control equipment operation data, generating a greenhouse environment prediction model, and predicting the greenhouse environment by using the greenhouse environment prediction model to obtain temperature distribution data of different positions in the greenhouse at the next moment. According to the system, high-precision prediction of temperature distribution at different positions in the greenhouse is realized, a scientific basis is provided for greenhouse environment optimization regulation and crop planting layout, and the intelligent level of greenhouse environment management is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of greenhouse environment intelligent control, and in particular to a greenhouse plane field environment simulation system based on machine learning. Background Art

[0002] Intelligent control of greenhouse environment is an important technical field in modern agricultural production. Environmental parameters such as temperature, humidity, light intensity and carbon dioxide concentration in the greenhouse have a direct impact on the growth and yield of crops. However, in the prior art, the control of the greenhouse environment usually relies on simple feedback control systems or experience-based adjustment methods. These methods are difficult to cope with the complex changes in the environment inside and outside the greenhouse, especially when there is a high degree of coupling and nonlinear relationship between environmental parameters. The modeling and prediction capabilities of traditional methods are insufficient. For example, the temperature in the greenhouse is not only affected by external climatic conditions (such as wind speed, radiation, rainfall, etc.), but also closely related to internal parameters such as humidity, light and carbon dioxide concentration. This complex coupling relationship makes it difficult for traditional control systems to achieve accurate prediction and efficient regulation.

[0003] In addition, the greenhouse environmental control systems in the prior art generally lack refined modeling of the spatial distribution of environmental parameters in the greenhouse, and are usually only able to predict single points or average values, which cannot accurately reflect the temperature distribution at different locations in the greenhouse. This limitation leads to uneven greenhouse environmental regulation, which may cause local overheating or overcooling, thus affecting the growth and yield of crops. At the same time, the prior art also has deficiencies in data processing. For example, outliers or missing values ​​often appear in the data collected by sensors. Failure to effectively handle these problems will significantly reduce the accuracy and reliability of the prediction model. Summary of the invention

[0004] In view of this, the present invention proposes a greenhouse plane field environment simulation system based on machine learning to solve the problems of low prediction accuracy, insufficient spatial distribution modeling and limited data processing capability of the greenhouse environment control system in the prior art.

[0005] The technical solution of the present invention is implemented as follows: The present invention provides a greenhouse plane field environment simulation system based on machine learning, comprising:

[0006] A data acquisition unit, used to collect greenhouse environmental parameter data and control device operation data, wherein the control device operation data includes a combination of working states of multiple control devices;

[0007] A data processing cloud platform, which is in communication with the data acquisition unit, is used to process the greenhouse environment parameter data and the control device operation data and generate a greenhouse environment prediction model, and use the greenhouse environment prediction model to predict the greenhouse environment to obtain temperature distribution data at different locations in the greenhouse at the next moment;

[0008] The data processing cloud platform includes:

[0009] A data processing unit, used for preprocessing the collected data;

[0010] A greenhouse environment prediction model training unit is used to train and generate multiple greenhouse environment prediction sub-models using pre-processed data, each prediction sub-model corresponds to a working state combination, and all greenhouse environment prediction sub-models constitute a greenhouse environment prediction model;

[0011] The prediction execution unit is used to select the corresponding prediction sub-model according to the working state combination of the current control device, and input the current environmental parameters into the selected prediction sub-model to obtain the temperature distribution prediction data of different positions in the greenhouse at the next moment;

[0012] The visualization display unit is used to receive the temperature distribution prediction data output by the prediction execution unit and generate a temperature distribution thermal map of the greenhouse plane field for display.

[0013] On the basis of the above technical solution, preferably, the data acquisition unit includes:

[0014] Weather station sensors to collect data on outdoor environmental conditions;

[0015] Environmental sensors in the greenhouse are used to collect environmental parameter data at different locations; the environmental sensors in the greenhouse include temperature sensors, humidity sensors, light radiation sensors and carbon dioxide sensors. The greenhouse is divided into several layers in the longitudinal direction, and each layer is provided with several temperature sensors and humidity sensors. The light radiation sensors and carbon dioxide sensors are provided on the top layer of the greenhouse;

[0016] Control device sensors are used to collect control device operation data.

[0017] On the basis of the above technical solution, preferably, the control device includes a fan, a skylight, a side window, a sunshade net and an air conditioner, and the working state combination of the control device is expressed by the following formula:

[0018] S={(s1,s2,s3,s4,s5)|s i ∈{0,1},i=1,2,3,4,5}

[0019] Among them, s1 represents the fan, s2 represents the skylight, s3 represents the side window, s4 represents the sunshade net, s5 represents the air conditioner, 0 represents off, and 1 represents on.

[0020] On the basis of the above technical solution, preferably, the method for constructing the greenhouse environment prediction sub-model includes:

[0021] S1. Acquire greenhouse environmental parameter data under the same working state combination of the control device, and divide it into training data and test data;

[0022] S2, preprocessing greenhouse environmental parameter data;

[0023] S3, constructing a greenhouse environment prediction sub-model using a BP neural network, and training the prediction sub-model based on training data;

[0024] S4. Use test data to verify the prediction accuracy of the prediction sub-model, optimize the model parameters by calculating the error between the predicted value and the measured value, and output the greenhouse environment prediction sub-model.

[0025] On the basis of the above technical solution, preferably, the greenhouse environment parameter data includes:

[0026] Outdoor environmental parameters: wind force, wind speed, total radiation, accumulated rainfall, instantaneous rainfall, daily rainfall, air temperature, air humidity and carbon dioxide concentration;

[0027] Indoor environmental parameters: air temperature, air humidity and carbon dioxide concentration at different locations.

[0028] Based on the above technical solution, preferably, in step S2, preprocessing the greenhouse environment parameter data includes:

[0029] S21, performing outlier detection on the acquired raw data, and eliminating data that does not meet the preset conditions;

[0030] S22. Filling missing values ​​based on data types, where filling methods include mode filling method, linear interpolation filling method and weighted average filling method;

[0031] S23, normalizing the filled data to obtain preprocessed data.

[0032] On the basis of the above technical solution, preferably, the wind force, wind speed and instantaneous rainfall adopt the mode filling method, and the formula of the mode filling method is:

[0033] x 填充 =Mod(x i |i indicates that data needs to be filled in on that day)

[0034] Among them, Mod represents the operation of finding the majority, x i Indicates the value on the day when the data needs to be filled;

[0035] The total radiation, accumulated rainfall, daily rainfall, air temperature, air humidity and carbon dioxide concentration are filled using a linear interpolation method, and the formula for the linear interpolation method is:

[0036]

[0037] Among them, N2 is the number of sampling points within the preset time range, x t-i and x t+i are the data of the i-th sampling point before and after the missing point respectively.

[0038] On the basis of the above technical solution, preferably, the greenhouse temperature and greenhouse humidity are filled by weighted average, including:

[0039] (1) Determine the eight nearest valid sampling points around the missing point, denoted as P1, P2, ... P8;

[0040] (2) Collect data at the same relative position in other plane fields at different time points, calculate the standardized regression coefficient of each sampling point and the center point, normalize the standardized regression coefficient, and obtain the weight coefficient. The calculation formula of the standardized regression coefficient is:

[0041]

[0042] Among them, x i is the historical data sequence of the i-th sampling point, is the independent variable x i The standard deviation of y is the historical data sequence of the center point, S y is the standard deviation of the independent variable y, and the standard deviation calculation formula is:

[0043]

[0044] x i represents the i-th sample point, represents the mean of the sample; n represents the number of samples;

[0045] (3) Calculate the weighted average filling of missing values ​​based on the weight coefficient.

[0046] Based on the above technical solution, preferably, in step S3, the structure of the greenhouse environment prediction sub-model includes:

[0047] The input layer is used to receive greenhouse environmental parameter data and control equipment operation data; the hidden layer adopts a double hidden layer structure, the number of nodes in the first hidden layer is 2n+1, and the number of nodes in the second hidden layer is n+1, where n is the number of nodes in the input layer; the number of nodes in the output layer corresponds to the number of temperature points to be predicted in the greenhouse.

[0048] Based on the above technical solution, preferably, the execution process of the predictive execution unit is as follows:

[0049] Step 1: Receive the current working status information (s1, s2, s3, s4, s5) of each control device, convert it into a standard binary state vector, and match and call the corresponding prediction sub-model from the temperature prediction model according to the state vector;

[0050] Step 2: Obtain greenhouse environmental parameter data at the current moment and perform preprocessing;

[0051] Step 3: Input the preprocessed greenhouse environmental parameter data into the selected prediction sub-model, and the prediction sub-model generates a standardized prediction result of a preset detection point in the greenhouse at the next moment through forward propagation calculation based on the input data;

[0052] Step 4: De-normalize the standardized prediction results and convert them into actual temperature values. According to the spatial coordinate information of the preset detection points, complete temperature distribution prediction data is constructed.

[0053] The greenhouse plane field environment simulation system based on machine learning of the present invention has the following beneficial effects compared with the prior art:

[0054] (1) The present invention realizes high-precision prediction of temperature distribution at different locations in a greenhouse by combining a data acquisition unit, a data processing cloud platform and a visualization display unit. The system can dynamically select a prediction sub-model based on the combination of greenhouse environmental parameters and the operating status of the control equipment to generate greenhouse temperature distribution data at the next moment, thereby providing a scientific basis for optimizing and controlling the greenhouse environment and crop planting layout, and improving the intelligent level of greenhouse environmental management.

[0055] (2) The data acquisition unit of the present invention combines data collection of the outdoor environment, multi-point environment in the greenhouse and the operating status of the control equipment, comprehensively covering the core influencing factors of greenhouse environmental changes; by vertically arranging sensors in layers to collect accurate indoor temperature, humidity and radiation data, it provides high-quality data support for greenhouse environmental prediction; at the same time, through data preprocessing technology, including outlier detection, missing value filling and normalization processing, the integrity and consistency of the data are ensured, the accuracy and robustness of the environmental prediction model are further improved, and the reliability of the system is enhanced.

[0056] (3) The present invention significantly improves the adaptability and prediction ability of the model to complex and changing environments by designing a variety of greenhouse environment prediction sub-models based on the combination of control device states and using BP neural networks for training and optimization. The prediction execution unit can quickly match and call the corresponding sub-model after obtaining the operating status of the current control device in real time, and work together with the input environmental parameters to generate temperature distribution data at different locations in the greenhouse at the next moment. This modular construction and dynamic selection mechanism effectively improves the agility and pertinence of the simulation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 It is a structural schematic diagram of a greenhouse plane field environment simulation system based on machine learning of the present invention;

[0059] Figure 2 A system architecture diagram of a greenhouse environment of the present invention;

[0060] Figure 3 A system architecture diagram for measuring the weather station of the present invention;

[0061] Figure 4 It is a schematic flow chart of the method for constructing the greenhouse environment prediction sub-model of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] like Figure 1 and Figure 2 As shown, the present invention provides a greenhouse plane field environment simulation system based on machine learning, comprising:

[0064] A data acquisition unit, used to collect greenhouse environmental parameter data and control device operation data, wherein the control device operation data includes a combination of working states of multiple control devices;

[0065] The data processing cloud platform is communicatively connected with the data acquisition unit, and is used to process the greenhouse environmental parameter data and the control equipment operation data and generate a greenhouse environmental prediction model, and use the greenhouse environmental prediction model to predict the greenhouse environment to obtain the temperature distribution data of different positions in the greenhouse at the next moment.

[0066] The data acquisition unit is used to comprehensively and real-time collect greenhouse environmental parameter data and control equipment operation data, providing high-quality basic data support for the greenhouse environment simulation system. The unit includes weather station sensors, greenhouse environment sensors and control equipment sensors.

[0067] Specifically, Figure 3 As shown, the weather station sensor is used to collect environmental condition data outside the greenhouse, including but not limited to wind force, wind speed, total radiation, accumulated rainfall, instantaneous rainfall, daily rainfall, air temperature, air humidity and carbon dioxide concentration. By arranging the weather station sensor outside the greenhouse, the changes of external meteorological conditions can be monitored in real time.

[0068] The indoor environment data sensor is used to collect environmental parameter data inside the greenhouse, including air temperature, air humidity, light radiation intensity and carbon dioxide concentration. In order to achieve three-dimensional perception of the internal environment of the greenhouse, the sensors are arranged in layers according to the longitudinal height of the greenhouse, and several temperature sensors and humidity sensors are evenly distributed on each layer. The top layer is used as the crop growth layer, and light radiation sensors and carbon dioxide sensors are set on the top layer of the greenhouse. Through the layered arrangement, the environmental information at different heights in the greenhouse can be fully covered to ensure the spatial resolution and accuracy of the collected data. More preferably, the greenhouse environment is divided into 5 layers, and 12 temperature sensors and humidity sensors are set on each layer.

[0069] Indoor control device sensors are used to collect operating status data of control devices in the greenhouse, including fans, skylights, side windows, sunshade nets and air conditioners. The operating status of each control device includes two states: on and off. The working status of each device is monitored in real time by sensors and converted into a standardized binary state vector representation. For example, the on / off state of a fan can be represented as "1" or "0", and the states of all control devices are combined to form a multidimensional vector, which is used to describe the current control conditions of the greenhouse.

[0070] Specifically, the working state combination of the control device is expressed by the following formula:

[0071] S={(s1,s2,s3,s4,s5)|s i ∈{0,1},i=1,2,3,4,5}

[0072] Among them, s1 represents the fan, s2 represents the skylight, s3 represents the side window, s4 represents the sunshade net, s5 represents the air conditioner, 0 represents off, and 1 represents on.

[0073] The data acquisition unit is linked to various sensors through the real-time acquisition module to collect outdoor meteorological data, indoor environmental data and control equipment status data at preset time intervals (for example, every 10 minutes). The collected data is transmitted to the data processing cloud platform through a wireless transmission module or a wired network and stored in a database. Through the collaborative work of multiple types of sensors, the data acquisition unit can provide high-precision and high-timeliness environmental parameter data and control equipment status data, fully considering the complexity and dynamics of the greenhouse environment, and can achieve all-round perception of the environment inside and outside the greenhouse. In addition, the data acquisition unit supports real-time display and export of data, which is convenient for users to monitor and analyze the greenhouse environment.

[0074] The data processing cloud platform includes a data processing unit, a greenhouse environment prediction model training unit, a prediction execution unit and a visualization display unit. The data processing unit is used to preprocess the collected data. The greenhouse environment prediction model training unit is used to train and generate multiple greenhouse environment prediction sub-models using the preprocessed data. Each prediction sub-model corresponds to a working state combination, and all greenhouse environment prediction sub-models constitute a greenhouse environment prediction model; the prediction execution unit is used to select the corresponding prediction sub-model according to the working state combination of the current control device, and input the current environmental parameters into the selected prediction sub-model to obtain the temperature distribution prediction data of different positions in the greenhouse at the next moment.

[0075] The visualization display unit is used to receive the temperature distribution prediction data output by the prediction execution unit, and generate a temperature distribution thermogram of the greenhouse plane field for display. The visualization display unit receives the temperature distribution prediction data at different locations in the greenhouse in real time through the communication interface with the prediction execution unit. These data are calculated based on the greenhouse environment prediction model, and contain the temperature values ​​of multiple preset detection points in the greenhouse and their corresponding spatial coordinate information. The visualization display unit processes these data to generate a thermogram reflecting the temperature distribution in the greenhouse, and displays it intuitively through a display device (such as a monitoring screen, a mobile terminal or a remote monitoring platform).

[0076] Specifically, Figure 4 As shown, the construction method of the greenhouse environment prediction sub-model includes:

[0077] S1. Acquire greenhouse environmental parameter data under the same working state combination of control devices, which are divided into training data and test data.

[0078] Specifically, a working state combination of the fixed control device is used to collect the greenhouse outdoor environment parameter data and indoor environment parameter data under this condition, and then the collected data is divided into training data and test data, wherein the training data is used for model training and the test data is used for model verification. The division of training data and test data usually follows a certain ratio (e.g., 8:2 or 7:3) to ensure that the model can fully learn the data features during the training process.

[0079] S2. Preprocess the greenhouse environmental parameter data, including data cleaning, data filling and data normalization.

[0080] The preprocessing specifically includes the following steps:

[0081] S21. Perform outlier detection on the acquired raw data and remove data that does not meet the preset conditions. Through outlier detection, abnormal data points in the raw data caused by sensor failure, environmental mutation or other uncontrollable factors are identified and removed. Outliers are usually manifested as extreme values ​​or unreasonable numerical changes that deviate greatly from the normal data distribution, such as the temperature in the greenhouse suddenly jumping to an abnormally high or low value, and the humidity sensor showing obvious discontinuous fluctuations. The detected outliers will be removed or marked as missing values ​​and enter the next step of filling processing.

[0082] S22. Fill missing values ​​based on data types, and filling methods include majority filling method, linear interpolation filling method and weighted average filling method.

[0083] Specifically, wind force, wind speed and instantaneous rainfall change significantly in a short period of time and are discrete variables. The mode filling method is used. The formula of the mode filling method is:

[0084] x 填充 =Mod(x i |i indicates that data needs to be filled in on that day)

[0085] Among them, Mod represents the operation of finding the majority, x i Indicates the value on the day that the data needs to be filled.

[0086] Total radiation, accumulated rainfall, daily rainfall, air temperature, air humidity and carbon dioxide concentration are continuous variables. The linear interpolation filling method is used to calculate the average value of the sampling data within 10 minutes before and after. The formula of the linear interpolation filling method is:

[0087]

[0088] Among them, N2 is the number of sampling points within the preset time range, x t-i and x t+i are the data of the i-th sampling point before and after the missing point, The weight coefficient for average calculation ensures that the filling value is the average value of the previous and next sampling points.

[0089] The data of greenhouse air temperature and humidity are greatly affected by the surrounding environment. Such data are not only related to time, but also closely related to spatial location. Therefore, it is necessary to consider the influence of spatial sampling points around the missing points. Therefore, the data of greenhouse air temperature and humidity are estimated based on the surrounding sensors and their historical time series, by calculating the correlation and weighted averaging. The weighted average filling method specifically includes:

[0090] (1) In the greenhouse, the air temperature and humidity at the missing point can be estimated using data from multiple sensors around it. First, determine the eight valid sampling points closest to the missing point (denoted as P1, P2, ... P8). These points are data provided by environmental sensors arranged in the greenhouse. At the same time, outliers or invalid data should be eliminated to ensure that the sampling points involved in the estimation are reliable.

[0091] (2) In order to comprehensively consider the correlation and influence of the surrounding sensors and the missing point in spatial distribution, it is necessary to calculate the correlation coefficient between each surrounding sampling point and the missing point. Collect data at the same relative position in other plane fields at different time points, calculate the standardized regression coefficient of each sampling point and the center point, normalize the standardized regression coefficient, and obtain the weight coefficient. The calculation formula of the standardized regression coefficient is:

[0092]

[0093] Among them, x i is the historical data sequence of the i-th sampling point, is the independent variable x i The standard deviation of y is the historical data sequence of the center point, S y is the standard deviation of the independent variable y, and the standard deviation calculation formula is:

[0094]

[0095] x i represents the i-th sample point, represents the sample mean; n represents the sample size.

[0096] The calculated correlation coefficient indicates the similarity between the data of a certain sampling point and the missing point. The higher the correlation coefficient, the greater the influence of the sampling point on the missing point. Then the standardized regression coefficient is normalized to obtain the weight coefficient so that the sum of the weight coefficients is equal to 1. The calculation formula of the weight coefficient is:

[0097]

[0098] wi is the weight coefficient of the i-th sampling point, indicating the contribution of the sampling point to the missing point filling value, r i is the standardized regression coefficient of the i-th sampling point; is the sum of the standardized regression coefficients of all 8 sampling points, which is used to normalize the weights.

[0099] (3) According to the calculated weight coefficient, the temperature or humidity data of the surrounding sensors are weighted averaged to infer the value of the missing point. The formula for weighted average filling is:

[0100]

[0101] Among them, w i is the weight coefficient of the i-th sampling point, P i is the actual data value of the i-th sampling point.

[0102] For filling the temperature and humidity data in the greenhouse, we fully utilize the advantages of the dense sensor network in the greenhouse and effectively fill the missing values ​​with the compensatory data of the surrounding sensors. Compared with simple linear interpolation or fixed mean filling, the weighted average calculation fully considers the spatial correlation and historical behavior patterns of the greenhouse environment, which not only improves the authenticity of the predicted value, but also retains the spatial distribution characteristics of the temperature and humidity in the greenhouse, thereby providing higher quality input data for subsequent greenhouse environment predictions.

[0103] S23, normalizing the filled data to obtain preprocessed data.

[0104] The normalization formula is:

[0105]

[0106] Among them, is the normalized data value, ranging between [0,1], x is the original data value, x min is the minimum value in the data set, x max is the maximum value in the data set.

[0107] In the preprocessing stage, the accuracy and completeness of the data were significantly improved by using a classification processing filling method for greenhouse environmental parameter data according to their different characteristics, and the prediction accuracy and stability of the greenhouse environmental simulation system were enhanced. For data with small fluctuations and discrete characteristics such as wind force, wind speed and instantaneous rainfall, the majority filling method can quickly capture the dominant characteristic values ​​within the time period and compensate for the impact of data breakpoints; for data with continuous change characteristics such as total radiation, accumulated rainfall, air temperature, air humidity and carbon dioxide concentration, the linear interpolation filling method can better retain the trend and progressiveness of the data and avoid the accumulation of model errors caused by discontinuity; and for data such as greenhouse temperature and humidity that are greatly affected by spatial distribution, the weighted average filling method is used to dynamically weight the correlation of the sampling points around the missing points, so as to more accurately restore the missing values ​​and retain the spatial distribution characteristics and physical consistency of the data.

[0108] S3. Use BP neural network to construct a greenhouse environment prediction sub-model, and train the prediction sub-model based on training data.

[0109] Specifically, the structure of the greenhouse environment prediction sub-model based on the BP neural network includes an input layer, an output layer, and a hidden layer. The input layer is used to receive the environmental condition data outside the greenhouse (such as wind speed, air temperature, etc.), the environmental condition data inside the greenhouse (such as the current indoor temperature, humidity, etc.), and the working status data of each control device (such as fans, skylights, side windows, etc.). The number of input nodes is n, which represents the total dimension of environmental parameters and control status. The hidden layer adopts a double hidden layer structure, the number of nodes in the first hidden layer is 2n+1, and the number of nodes in the second hidden layer is n+1, where n is the number of nodes in the input layer. The number of nodes in the output layer corresponds to the number of temperature points to be predicted in the greenhouse, and each node outputs the predicted temperature value of a greenhouse detection point.

[0110] The entire network structure forms a complete mapping relationship through the hierarchical connection of input layer, hidden layer and output layer: the input data is connected to the hidden layer neurons through the fully connected layer, and each node performs weighted summation and nonlinear transformation; the hidden layer uses activation functions (such as Sigmoid or ReLU) to perform nonlinear activation on the data, thereby enhancing the expressive power of the model.

[0111] During the network training process, the input layer receives processed training data, including a variety of environmental parameters (outdoor meteorological data and greenhouse sensor data) and control equipment working state combinations. In each iteration of training, the network is calculated through forward propagation. After the forward propagation is completed, the network output value is compared with the actual value in the training data to calculate the prediction error; through the error back propagation algorithm, the weights and biases of each layer in the network are adjusted by gradient descent according to the error. The training process is carried out through multiple iterations until the network error function E converges or reaches the preset number of training times. After the training is completed, the optimized network model parameters are saved to form a greenhouse environment prediction sub-model under the current control equipment state combination.

[0112] S4. Use test data to verify the prediction accuracy of the prediction sub-model, optimize the model parameters by calculating the error between the predicted value and the measured value, and output the greenhouse environment prediction sub-model.

[0113] Specifically, the execution process of the predictive execution unit is as follows:

[0114] Step 1: Receive the current working status information of each control device (s1, s2, s3, s4, s5) and convert it into a standard binary state vector. The binary vector reflects the on / off status of each control device through encoding. The system matches the corresponding prediction sub-model in the temperature prediction model library based on this state vector and calls the model for subsequent prediction calculations;

[0115] Step 2: Obtain the greenhouse environmental parameter data at the current moment and perform preprocessing, including outlier detection, missing value filling and normalization, to eliminate invalid data, fill in missing data and standardize the scale of input variables;

[0116] Step 3: Input the preprocessed greenhouse environmental parameter data into the selected prediction sub-model, and the prediction sub-model generates a standardized prediction result of a preset detection point in the greenhouse at the next moment through forward propagation calculation based on the input data;

[0117] Step 4: De-standardize the standardized prediction results, restore the standardized values ​​to the actual temperature values, and generate complete greenhouse internal temperature distribution prediction data using interpolation technology or spatial mapping algorithms based on the spatial coordinate information of the preset detection points (such as the geometric position and height level of the detection points).

[0118] It should be noted that the greenhouse environment prediction simulation constructed by the present invention can develop from temperature modeling of a single point in the greenhouse crop layer, simulating and analyzing the general impact of greenhouse environmental conditions and control conditions on single-point sensors, to plane field temperature modeling at different heights in the greenhouse, simulating and analyzing the specific impact of different greenhouse environmental conditions and control conditions on temperatures at different heights in the greenhouse, and then rationally planning the spatial layout in the greenhouse to increase crop yields.

[0119] In both single-point greenhouse environment simulation and plane field greenhouse environment simulation, the input data are derived from greenhouse environment parameter data. However, due to the different goals of the two simulations, there may be differences in the data preprocessing stage. In single-point greenhouse environment simulation, the input data is mainly used to predict the temperature of a single sensor, and the temperature parameters related to the target sensor are directly used; while in plane field greenhouse environment simulation, the input data needs to be used to predict the temperature distribution of the entire plane field, and the spatial position information or other features of the sensor may need to be added to reflect the overall characteristics of the plane field.

[0120] In one embodiment, two days of data with stable control conditions in the greenhouse are selected as training data and test data of the model respectively, and the average value of the sampling data every 10 minutes in one day is selected as the characteristic value of the time period. The input data is a 144*n dimensional vector matrix generated by n items in the greenhouse environmental parameter data (including wind force, wind speed, total radiation, accumulated rainfall, instantaneous rainfall, daily rainfall, air temperature, air humidity, carbon dioxide, etc.), and the output data is a 144*1 dimensional vector matrix generated by the air temperature of a single sensor in the greenhouse.

[0121] In another embodiment, 2 days of data with stable control conditions in the greenhouse are selected as training data and test data of the model respectively. The data acquisition module selects the average value of the sampled data every 10 minutes in one day as the characteristic value of the time period. The input data is a 144*n dimensional vector matrix generated by n items in the greenhouse environmental parameter data (including wind force, wind speed, total radiation, accumulated rainfall, instantaneous rainfall, daily rainfall, air temperature, air humidity, carbon dioxide, etc.), and the output data is the air temperature of m sensors in a single layer in the greenhouse, that is, a 144*m dimensional vector matrix.

[0122] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A greenhouse plane field environment simulation system based on machine learning, characterized in that: include: A data acquisition unit, used to collect greenhouse environmental parameter data and control device operation data, wherein the control device operation data includes a combination of working states of multiple control devices; A data processing cloud platform, which is in communication with the data acquisition unit, is used to process the greenhouse environment parameter data and the control device operation data and generate a greenhouse environment prediction model, and use the greenhouse environment prediction model to predict the greenhouse environment to obtain temperature distribution data at different locations in the greenhouse at the next moment; Among them, the data processing cloud platform includes: A data processing unit, used for preprocessing the collected data; A greenhouse environment prediction model training unit is used to train and generate multiple greenhouse environment prediction sub-models using pre-processed data, each prediction sub-model corresponds to a working state combination, and all greenhouse environment prediction sub-models constitute a greenhouse environment prediction model; The prediction execution unit is used to select the corresponding prediction sub-model according to the working state combination of the current control device, and input the current environmental parameters into the selected prediction sub-model to obtain the temperature distribution prediction data of different positions in the greenhouse at the next moment; The visualization display unit is used to receive the temperature distribution prediction data output by the prediction execution unit and generate a temperature distribution thermal map of the greenhouse plane field for display.

2. A greenhouse plane field environment simulation system based on machine learning as claimed in claim 1, characterized in that: The data acquisition unit includes: Weather station sensors to collect data on outdoor environmental conditions; Environmental sensors in the greenhouse are used to collect environmental parameter data at different locations; the environmental sensors in the greenhouse include temperature sensors, humidity sensors, light radiation sensors and carbon dioxide sensors. The greenhouse is divided into several layers in the longitudinal direction, and each layer is provided with several temperature sensors and humidity sensors. The light radiation sensors and carbon dioxide sensors are provided on the top layer of the greenhouse; Control device sensors are used to collect control device operation data.

3. A greenhouse plane field environment simulation system based on machine learning as claimed in claim 1, characterized in that: The control equipment includes fans, skylights, side windows, sunshades and air conditioners. The working status combination of the control equipment is expressed by the following formula: s={(s1,s2,s3,s4,s5)|s i ∈{0,1},i=1,2,3,4,5} Among them, s1 represents the fan, s2 represents the skylight, s3 represents the side window, s4 represents the sunshade net, s5 represents the air conditioner, 0 represents off, and 1 represents on.

4. A greenhouse plane field environment simulation system based on machine learning as claimed in claim 3, characterized in that: The construction method of the greenhouse environment prediction sub-model includes: S1. Acquire greenhouse environmental parameter data under the same working state combination of the control device, and divide it into training data and test data; S2, preprocessing greenhouse environmental parameter data; S3, constructing a greenhouse environment prediction sub-model using a BP neural network, and training the prediction sub-model based on training data; S4. Use test data to verify the prediction accuracy of the prediction sub-model, optimize the model parameters by calculating the error between the predicted value and the measured value, and output the greenhouse environment prediction sub-model.

5. A greenhouse plane field environment simulation system based on machine learning as claimed in claim 4, characterized in that: The greenhouse environment parameter data include: Outdoor environmental parameters: wind force, wind speed, total radiation, accumulated rainfall, instantaneous rainfall, daily rainfall, air temperature, air humidity and carbon dioxide concentration; Indoor environmental parameters: air temperature, air humidity and carbon dioxide concentration at different locations.

6. A greenhouse plane field environment simulation system based on machine learning as claimed in claim 5, characterized in that: In step S2, the greenhouse environment parameter data is preprocessed, including: S21, performing outlier detection on the acquired raw data, and eliminating data that does not meet the preset conditions; S22. Filling missing values ​​based on data types, where filling methods include mode filling method, linear interpolation filling method and weighted average filling method; S23, normalizing the filled data to obtain preprocessed data.

7. A greenhouse plane field environment simulation system based on machine learning as claimed in claim 6, characterized in that: Wind force, wind speed and instantaneous rainfall adopt the mode filling method, and the formula of the mode filling method is: x 填充 =Mod(x i |i indicates that data needs to be filled in on the current day) Among them, Mod represents the operation of finding the majority, x i Indicates the value on the day when data needs to be filled; The total radiation, accumulated rainfall, daily rainfall, air temperature, air humidity and carbon dioxide concentration are filled using a linear interpolation method, and the formula for the linear interpolation method is: Among them, N2 is the number of sampling points within the preset time range, x t-i and x t+i are the data of the i-th sampling point before and after the missing point respectively.

8. The greenhouse plane field environment simulation system based on machine learning as claimed in claim 6, characterized in that: Greenhouse temperature and greenhouse humidity are populated using a weighted average, including: (1) Determine the eight nearest valid sampling points around the missing point, denoted as P1, P2, ..., P8; (2) Collect data at the same relative position in other plane fields at different time points, calculate the standardized regression coefficient of each sampling point and the center point, normalize the standardized regression coefficient, and obtain the weight coefficient. The calculation formula of the standardized regression coefficient is: Among them, x i is the historical data sequence of the i-th sampling point, is the independent variable x i The standard deviation of y is the historical data sequence of the center point, S y is the standard deviation of the independent variable y, and the standard deviation calculation formula is: x i represents the i-th sample point, represents the mean of the sample; n represents the number of samples; (3) Calculate the weighted average filling of missing values ​​based on the weight coefficient.

9. The greenhouse plane field environment simulation method based on machine learning according to claim 4, characterized in that: In step S3, the structure of the greenhouse environment prediction sub-model includes: The input layer is used to receive greenhouse environmental parameter data and control equipment operation data; the hidden layer adopts a double hidden layer structure, the number of nodes in the first hidden layer is 2n+1, and the number of nodes in the second hidden layer is n+1, where n is the number of nodes in the input layer; the number of nodes in the output layer corresponds to the number of temperature points to be predicted in the greenhouse.

10. The greenhouse plane field environment simulation system based on machine learning according to claim 5, characterized in that: The execution process of the speculative execution unit is as follows: Step 1: Receive the current working status information (s1, s2, s3, s4, s5) of each control device, convert it into a standard binary state vector, and match and call the corresponding prediction sub-model from the temperature prediction model according to the state vector; Step 2: Obtain greenhouse environmental parameter data at the current moment and perform preprocessing; Step 3: Input the preprocessed greenhouse environmental parameter data into the selected prediction sub-model, and the prediction sub-model generates a standardized prediction result of a preset detection point in the greenhouse at the next moment through forward propagation calculation based on the input data; Step 4: De-normalize the standardized prediction results and convert them into actual temperature values. According to the spatial coordinate information of the preset detection points, complete temperature distribution prediction data is constructed.

Images