Greenhouse environment management and control method and system based on multi-parameter multi-node prediction

Abstract

The invention belongs to the technical field of intelligent agriculture, and discloses a greenhouse environment management and control method and system based on multi-parameter multi-node prediction, and the method comprises the steps: carrying out the preprocessing of collected indoor environment data and outdoor meteorological data, obtaining a sample set, and dividing the sample set into a training set and a test set; constructing a GJO-GCN-BiGRU-SAM greenhouse environment prediction model based on a golden litsea rotundifolia optimization algorithm, and predicting the greenhouse environment after a preset duration; inputting the training set into a GJO-GCN-BiGRU-SAM greenhouse environment prediction model for training, and performing verification by using the test set to obtain a trained greenhouse environment prediction model; inputting the indoor environment data and the outdoor meteorological data collected in real time into the trained greenhouse environment prediction model to obtain a prediction result after a preset duration; and according to the prediction result, controlling an adjusting system of the greenhouse, and adjusting the greenhouse environment. According to the method, the spatial distribution trend of the multi-point environment of the greenhouse at the future moment can be predicted more accurately, and a decision basis is provided for overall environment regulation and control of the greenhouse.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent agriculture, and relates to a greenhouse environment management and control method and system based on multi-parameter multi-node prediction. BACKGROUND

[0002] In recent years, facility agriculture has developed rapidly, and greenhouses are important places for cultivating crops in facility agriculture. The greenhouse environment has an important influence on the growth and yield of crops. In the regulation of the greenhouse environment, reasonable environmental parameters can effectively guarantee the efficient and high-quality growth of crops. Therefore, effective prediction of greenhouse environmental parameters is an important basis for realizing precise regulation of the greenhouse environment, and it has important significance for optimizing greenhouse production, improving crop yield and quality, and realizing large-scale and intensive production.

[0003] Although significant progress has been made in greenhouse technology, the greenhouse environment of the greenhouse is affected by a variety of independent time series characteristics, including outdoor air temperature, outdoor air relative humidity, indoor temperature, indoor air relative humidity, indoor light intensity, and soil temperature. Therefore, in order to integrate these feature information related to the greenhouse, identify potential relationships and spatial patterns, accurately predict the spatial distribution of various environmental parameters in the greenhouse, and regulate and control, there are still many challenges and deficiencies.

[0004] Traditional greenhouse environmental parameter prediction methods are mostly based on simple mathematical models or empirical formulas, and cannot accurately reflect the complex relationships and dynamic changes between the parameters in the greenhouse. Therefore, the prediction results often have large errors, and cannot meet the needs of modern agriculture for precise regulation of the greenhouse environment, so it is urgent to explore a greenhouse environment management and control method based on multi-parameter multi-node prediction. SUMMARY

[0005] The purpose of the present application is to provide a greenhouse environment management and control method based on multi-parameter multi-node prediction, which can meet the needs of precise regulation of the greenhouse environment.

[0006] Another purpose of the present application is to provide a greenhouse environment management and control system based on multi-parameter multi-node prediction.

[0007] In order to achieve the above purpose, the technical solution adopted by the present application is as follows:

[0008] A greenhouse environment management and control method based on multi-parameter multi-node prediction, comprising the following steps:

[0009] S1, at least one indoor monitoring node is arranged in the greenhouse, indoor environmental data is collected for each indoor monitoring node, at least one outdoor monitoring node is arranged outside the greenhouse, and outdoor meteorological data is collected at each outdoor monitoring node;

[0010] S2, preprocessing the collected indoor environment data and outdoor meteorological data to obtain a sample set, and dividing the sample set into a training set and a test set;

[0011] S3, constructing a GJO-GCN-BiGRU-SAM greenhouse environment prediction model based on the GJO optimization algorithm to predict the greenhouse environment after a predetermined time; the GJO-GCN-BiGRU-SAM greenhouse environment prediction model comprises an input layer, a convolutional neural network layer, a bidirectional gated recurrent unit network layer, a self-attention mechanism network layer, a first full connection layer and an output layer connected in turn;

[0012] S4, inputting the training set into the GJO-GCN-BiGRU-SAM greenhouse environment prediction model for training, and verifying with the test set to obtain a trained greenhouse environment prediction model;

[0013] S5, inputting the real-time collected indoor environment data and outdoor meteorological data into the trained greenhouse environment prediction model to obtain the prediction results of indoor temperature, indoor air relative humidity, indoor CO2 concentration and indoor soil temperature of each indoor monitoring node after a predetermined time;

[0014] S6, according to the prediction results, controlling the adjusting system of the greenhouse to adjust the greenhouse environment.

[0015] As a limitation, in step S1, the horizontal distance between the indoor monitoring nodes is 15-25m;

[0016] The indoor environment data includes indoor temperature, indoor air relative humidity, indoor light intensity, indoor soil temperature and indoor CO2 concentration;

[0017] The outdoor meteorological data includes outdoor air temperature, outdoor air relative humidity and outdoor CO2 concentration.

[0018] As a second limitation, in step S2, the specific process of preprocessing the collected indoor environment data and outdoor meteorological data to obtain a sample set is as follows:

[0019] Based on k-means clustering analysis, the missing values of the collected indoor environment data and outdoor meteorological data are imputed by combining the nearest neighbor interpolation and mean filling strategy; then the threshold is set to remove the outliers; finally, the data is normalized by Min-Max standardization method to obtain the sample set.

[0020] As a further limitation, the division ratio of the training set and the test set is 8:2.

[0021] As a third limitation, in step S3, the convolutional neural network layer comprises a first convolutional layer, a first max-pooling layer, a second convolutional layer, a second max-pooling layer, a padding layer and a second full connection layer connected in sequence;

[0022] The first convolutional layer and the second convolutional layer are both one-dimensional convolutional layers, the convolution kernel size is 3x3, and the convolutional neural network layer adopts a Relu activation function.

[0023] The pooling size of the first max-pooling layer and the second max-pooling layer is both 2.

[0024] As a fourth limitation, in step S3, the golden cat optimization algorithm comprises:

[0025] (1) Prey search, the mathematical model of the relative position of each individual in the prey by the golden cat is as follows:

[0026] Y1(t) = Y M (t) - E x |Y M (t) - rl x Prey(t)|

[0027] Y2(t) = Y FM (t) - E x |Y FM (t) - rl x Prey(t)|

[0028] Wherein, Y1(t) is the corresponding male golden cat position after updating the position of tracking the prey; Y2(t) is the corresponding female golden cat position after updating the position of tracking the prey; Y M (t) is the initial position of the male golden cat; Y FM (t) is the initial position of the female golden cat; t is the current iteration number; Prey(t) is the prey position; Y M (t) - rl x Prey(t) is the relative distance between the male golden cat and the prey in the tracking stage, Y FM (t) - rl x Prey(t) is the relative distance between the female golden cat and the prey in the tracking stage.

[0029] E is the escape energy of the prey, and the calculation formula is:

[0030] E = E1E0

[0031] Wherein, E1 is the initial energy when the prey escapes; E0 is the energy decay state.

[0032] E0 = 2r-1

[0033] Wherein, r is a random number in [0, 1];

[0034]

[0035] Wherein, c1 is a constant with a value of 1.5, that is, E1 is linearly attenuated from 1.5 to 0 in the escape process of the prey; T is the maximum number of iterations;

[0036] rl is a random number based on the Levy distribution, and rl*Prey(t) represents the movement mode of the prey escaping;

[0037] rl=0.05*LF(y)

[0038]

[0039] Wherein, μ, v are random numbers in [0, 1]; β is a constant 1.5; LF(y) is a Levy flight function;

[0040] (2) as the escape energy of the prey continues to attenuate, the male and female golden jackals surround the prey and implement capture, and the updated male golden jackal position and female golden jackal position are obtained;

[0041] Y1(t)=Y M (t)-E*|rl*Y M (t)-Prey(t)|

[0042] Y2(t)=Y FM (t)-E*|rl*Y FM (t)-Prey(t)|

[0043] The position of the golden jackal is updated as:

[0044]

[0045] The application also provides a greenhouse environment management and control system based on multi-parameter multi-node prediction, which is used for implementing the greenhouse environment management and control method based on multi-parameter multi-node prediction.

[0046] The first multi-source sensor network is used for setting at least one indoor monitoring node in the greenhouse, and indoor environment data is collected by the first multi-source sensor for each indoor monitoring node;

[0047] The second multi-source sensor network is used for setting at least one outdoor monitoring node outside the greenhouse, and outdoor meteorological data is collected by the second multi-source sensor for each outdoor monitoring node;

[0048] The data transmission module is used for transmitting the indoor environment data and the outdoor meteorological data to the Internet of Things platform;

[0049] The Internet of Things platform comprises a data preprocessing unit, a model construction unit, a model prediction unit and a display unit.

[0050] The data preprocessing unit is configured to preprocess the collected indoor environment data and outdoor meteorological data to obtain a sample set, and divide the sample set into a training set and a test set.

[0051] The model construction unit is configured to construct a GJO-GCN-BiGRU-SAM greenhouse environment prediction model based on the GJO optimization algorithm, and predict the greenhouse environment after a preset time period. The GJO-GCN-BiGRU-SAM greenhouse environment prediction model comprises an input layer, a convolutional neural network layer, a bidirectional gated recurrent unit network layer, a self-attention mechanism network layer, a first full connection layer and an output layer connected in sequence. The training set is input into the GJO-GCN-BiGRU-SAM greenhouse environment prediction model for training, and the test set is used for verification to obtain a trained greenhouse environment prediction model.

[0052] The model prediction unit is configured to input the real-time collected indoor environment data and outdoor meteorological data into the trained greenhouse environment prediction model to obtain the prediction results of the indoor temperature, the indoor air relative humidity, the indoor CO2 concentration and the indoor soil temperature content of each indoor monitoring node after a preset time period.

[0053] The display unit is configured to display the prediction results.

[0054] The alarm unit is configured to alarm when the prediction results exceed a threshold.

[0055] The greenhouse environment adjustment module is configured to control the adjustment system of the greenhouse according to the prediction results to adjust the greenhouse environment.

[0056] As a limitation, the distance between the indoor monitoring nodes arranged in the greenhouse is 15-25 m.

[0057] The first multi-source sensor comprises at least one first air temperature and humidity sensor for collecting indoor temperature and indoor air relative humidity, at least one soil temperature sensor for collecting indoor soil temperature, a first CO2 sensor for collecting indoor CO2 concentration, and a light sensor for collecting indoor light intensity. The first air temperature and humidity sensor and the soil temperature sensor are arranged in layers in the vertical direction of the indoor monitoring node. The height range of the first air temperature and humidity sensor is 0.5-2.5 m, and the embedding depth of the soil temperature sensor is 10-30 cm.

[0058] The second multi-source sensor comprises a second air temperature and humidity sensor for collecting outdoor air temperature and outdoor air relative humidity, and a second CO2 sensor for collecting outdoor CO2 concentration.

[0059] As a second limitation, the data transmission module simultaneously supports cellular mobile communication and LoRa remote communication, and transmits indoor environment data and outdoor meteorological data to the Internet of Things platform through a star topology.

[0060] As a third limitation, the regulation system includes a ventilation system, an irrigation system and a thermal insulation system.

[0061] Compared with the prior art, the technical progress achieved by the present application is that:

[0062] (1) The greenhouse environment management and control method based on multi-parameter multi-node prediction of the present application can more comprehensively cover the environmental conditions at different positions inside and outside the greenhouse, and improve the accuracy and representativeness of data acquisition, by respectively collecting indoor environment data from at least one indoor monitoring node and collecting outdoor meteorological data from at least one outdoor monitoring node.

[0063] (2) The greenhouse environment management and control method based on multi-parameter multi-node prediction of the present application introduces a convolutional neural network, a bidirectional gated recurrent unit and a self-attention mechanism, and constructs a GJO-GCN-BiGRU-SAM greenhouse environment prediction model based on a golden jackal optimization algorithm. This model structure combines the advantages of multiple deep learning technologies, can capture more complex environmental change patterns, and improve prediction accuracy and robustness. By predicting multiple environmental parameters, more precise and flexible control strategies can be implemented to meet the specific needs of different crops at different growth stages. In the traditional method, manual parameter adjustment is required, and the model has limited generalization ability. In the present application, the golden jackal optimization algorithm automatically optimizes hyperparameters (such as learning rate and hidden layer neuron number) by simulating the hunting behavior of a golden jackal, solving the shortcomings of traditional LSTM models in modeling complex nonlinear relationships.

[0064] (3) In the data preprocessing aspect of the greenhouse environment management and control method based on multi-parameter multi-node prediction of the present application, the collected indoor environment data and outdoor meteorological data are subjected to k-means clustering analysis, missing value interpolation, abnormal value removal and normalization processing, ensuring the high quality and consistency of the input data, reducing the impact of data missing and abnormality, and improving the stability and robustness of the model. Compared with the traditional simple interpolation method, the data preprocessing of the present application improves the model training convergence speed by 30%, and stabilizes the prediction error fluctuation (CV-RMSE) below 5%, laying a reliable data foundation for high-precision modeling.

[0065] (4) The data transmission module of the greenhouse environment management and control system based on multi-parameter multi-node prediction of the present application simultaneously supports cellular mobile communication and LoRa remote communication, and transmits indoor environment data and outdoor meteorological data to the Internet of Things platform through a star topology, optimizing the coverage range and transmission efficiency, and making the packet loss rate close to zero.

[0066] (5) In the greenhouse environment control system based on multi-parameter and multi-node prediction of the present invention, at least one indoor monitoring node is set in the greenhouse, and the spacing between indoor monitoring nodes is 15 to 25 meters. In each indoor monitoring node, the height range of the first air temperature and humidity sensor is 0.5 to 2.5 meters, and the burial depth of the soil temperature sensor is 10 to 30 cm. Through multi-node layered deployment, the temporal and spatial distribution differences of environmental parameters can be accurately captured, which solves the problem of insufficient modeling of spatial heterogeneity by traditional single-node models and provides a high-resolution data basis for prediction.

[0067] In summary, the present invention can more accurately predict the spatial distribution trend of multiple-point environments in a greenhouse at future times, providing a decision-making basis for the overall environmental control of the greenhouse. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 Shown is a flow chart of the method of Example 1 of the present invention;

[0069] Figure 2 FIG2 is a structural diagram of the GJO-GCN-BiGRU-SAM greenhouse environment prediction model in Example 1 of the present invention;

[0070] Figure 3 Shown is a comparison chart of comprehensive performance evaluation of indoor CO2 concentration prediction data in Example 1 of the present invention;

[0071] Figure 4 Shown is a comparison chart of comprehensive performance evaluation of indoor air relative humidity prediction data in Example 1 of the present invention;

[0072] Figure 5 Shown is a comparison chart of comprehensive performance evaluation of indoor temperature prediction data in Example 1 of the present invention;

[0073] Figure 6 Shown is a comparison chart of comprehensive performance evaluation of indoor soil temperature prediction data in Example 1 of the present invention;

[0074] Figure 7 Shown is a structural block diagram of embodiment 2 of the present invention. DETAILED DESCRIPTION

[0075] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0076] Example 1 A greenhouse environment control method based on multi-parameter multi-node prediction

[0077] like Figure 1 As shown, this embodiment is a greenhouse environment control method based on multi-parameter multi-node prediction, comprising the following steps:

[0078] S1, setting at least one indoor monitoring node in a greenhouse, collecting indoor environment data for each indoor monitoring node respectively, setting at least one outdoor monitoring node outside the greenhouse, and collecting outdoor meteorological data at each outdoor monitoring node respectively;

[0079] In this step, the horizontal spacing between each indoor monitoring node is 20 m; the indoor environment data collected by each indoor monitoring node includes indoor temperature, indoor air relative humidity, indoor light intensity, indoor soil temperature and indoor CO2 concentration; the outdoor meteorological data collected by each outdoor monitoring node includes outdoor air temperature, outdoor air relative humidity and outdoor CO2 concentration;

[0080] Each indoor monitoring node is provided with a first multi-source sensor, and each outdoor monitoring node is provided with a second multi-source sensor, the first multi-source sensor includes at least one first air temperature and humidity sensor for collecting indoor temperature and indoor air relative humidity, at least one soil temperature sensor for collecting indoor soil temperature, a first CO2 sensor for collecting indoor CO2 concentration and a light sensor for collecting indoor light intensity; the first air temperature and humidity sensor and the soil temperature sensor are arranged in layers in the vertical direction of the indoor monitoring node, the height range of the first air temperature and humidity sensor is 0.5-2.5 m, and the embedding depth of the soil temperature sensor is 10-30 cm; the second multi-source sensor includes a second air temperature and humidity sensor for collecting outdoor air temperature and outdoor air relative humidity and a second CO2 sensor for collecting outdoor CO2 concentration.

[0081] S2, preprocessing the collected indoor environment data and outdoor meteorological data to obtain a sample set, and dividing the sample set into a training set and a test set, wherein the division ratio of the training set and the test set is 8:2;

[0082] In this step, the specific process of preprocessing the collected indoor environment data and outdoor meteorological data to obtain a sample set is as follows:

[0083] For the collected indoor environment data and outdoor meteorological data, based on k-means clustering analysis, combined with the interpolation and mean filling strategy, the missing values are supplemented; then the threshold is set to remove the abnormal values; finally, the Min-Max standardization method is used to normalize the data to obtain the sample set; for example, when removing the abnormal values, the threshold of temperature can be set to-20℃-50℃, etc.

[0084] S3, constructing a GJO-GCN-BiGRU-SAM greenhouse environment prediction model based on the golden jackal optimization algorithm to predict the greenhouse environment after a preset time; the GJO-GCN-BiGRU-SAM greenhouse environment prediction model comprises an input layer, a convolutional neural network layer, a bidirectional gated recurrent unit network layer, a self-attention mechanism network layer, a first full connection layer and an output layer connected in sequence, as shown in Figure 2 ; the convolutional neural network layer is denoted as GCN, the bidirectional gated recurrent unit network layer is denoted as BiGRU, the self-attention mechanism network layer is denoted as SAM, and the golden jackal optimization algorithm is denoted as GJO;

[0085] In this step, the golden jackal optimization algorithm is used to iteratively select the best hyperparameters of the model, such as the learning rate, the number of BiGRU hidden layer neurons, and the number of SAM key-value pairs, aiming to find the best parameter combination to improve the performance and effect of the model; in the golden jackal optimization algorithm, the population size is 15, the maximum number of iterations is 150, and the learning rate search range is [10 -4 , 10 -2 ];

[0086] The golden jackal optimization algorithm comprises:

[0087] (1) Prey search, the mathematical model of the relative position of each individual in the prey by the golden jackal is as follows:

[0088] Y1(t) = Y M (t) - E x |Y M (t) - rl x Prey(t)|

[0089] Y2(t) = Y FM (t) - E x |Y FM (t) - rl x Prey(t)|

[0090] wherein Y1(t) is the corresponding male golden jackal position after updating the position of the prey; Y2(t) is the corresponding female golden jackal position after updating the position of the prey; Y M (t) is the initial position of the male golden jackal; Y FM (t) is the initial position of the female golden jackal; t is the current iteration number; Prey(t) is the prey position; Y M (t) - rl x Prey(t) is the relative distance between the male golden jackal and the prey in the tracking stage, and Y FM (t) - rl x Prey(t) is the relative distance between the female golden jackal and the prey in the tracking stage;

[0091] E is the escape energy of the prey, and the calculation formula is:

[0092] E = E1E0

[0093] Wherein, E1 is the initial energy when the prey escapes; E0 is the energy attenuation state;

[0094] E0 = 2r-1

[0095] Wherein, r is a random number in [0, 1];

[0096]

[0097] Wherein, c1 is a constant with a value of 1.5, that is, E1 linearly decays from 1.5 to 0 during the escape of the prey; T is the maximum number of iterations;

[0098] rl is a random number based on the Levy distribution, and rl×Prey(t) represents the movement mode of the prey escaping;

[0099] rl = 0.05×LF(y)

[0100]

[0101] Wherein, μ, v are random numbers in [0, 1]; β is a constant 1.5; LF(y) is a Levy flight function;

[0102] (2) As the energy of the prey escaping continues to decay, the male and female golden jackals surround the prey and implement capture, and the updated male golden jackal position and female golden jackal position are obtained;

[0103] Y1(t) = Y M (t)-E×|rl×Y M (t)-Prey(t)|

[0104] Y2(t) = Y FM (t)-E×|rl×Y FM (t)-Prey(t)|

[0105] The position of the golden jackal is updated as:

[0106]

[0107] In this step, the convolutional neural network layer includes a first convolutional layer, a first max pooling layer, a second convolutional layer, a second max pooling layer, a padding layer and a second full connection layer connected in turn; the first convolutional layer and the second convolutional layer are both one-dimensional convolutional layers, the convolution kernel size is 3×3, and the convolutional neural network layer adopts a Relu activation function; the pooling size of the first max pooling layer and the second max pooling layer is both 2; the convolutional neural network layer is used to extract features from input data, eliminate potential unstable factors and enhance the expression ability of the model to input data; the convolutional neural network layer performs well in processing data with grid structure;

[0108] The first convolutional layer and the second convolutional layer are responsible for extracting spatial features from the input data, and can effectively capture local features by sharing parameters at different positions; the first max-pooling layer and the second max-pooling layer retain important features while reducing the spatial dimension, which helps to improve the robustness of the model and reduce overfitting; the second fully connected layer maps high-level features to target categories; due to the large number of parameters, some neurons are randomly set to zero during training to reduce the dependence between neurons and enhance the generalization ability of the model; however, the convolutional neural network layer cannot extract time series features, so a bidirectional gated recurrent unit network layer is introduced to learn the time characteristics;

[0109] In this step, the number of hidden layer neurons in the bidirectional gated recurrent unit network layer is 15, and the bidirectional gated recurrent unit network layer is used to analyze the input data and extract feature vectors, which helps to improve the understanding of the model for sequence data;

[0110] The bidirectional gated recurrent unit network layer is composed of a forward GRU unit and a backward GRU unit. In the greenhouse environment, the environmental conditions at consecutive time points are closely related, and deep feature extraction is required. By combining the forward GRU unit and the backward GRU unit, the bidirectional gated recurrent unit network layer can fully utilize the time dependence and explore the feature relationship, thereby improving the performance of the model.

[0111] In this step, the self-attention mechanism network layer is used to assign different weights to the feature vectors, highlighting the effective features and helping the model to more effectively utilize the feature information.

[0112] Attention mechanism is a resource allocation mechanism that simulates the attention mechanism of the human brain. Its core function is to enhance the attention and weight allocation of different positions in the input sequence. Although traditional attention mechanisms can improve the model's attention to key information to some extent, their attention allocation is heavily dependent on external information guidance. The self-attention mechanism network layer can capture long-distance dependencies in the input sequence and automatically focus on information from different positions. When dealing with large-scale data and building complex large-scale models, the self-attention mechanism network layer shows excellent scalability and adaptability. Unlike traditional attention mechanisms, the self-attention mechanism network layer can directly model the relationship between elements in the input sequence, thereby surpassing the limitations of traditional attention mechanisms that only focus on the interaction between input and output. By introducing the self-attention mechanism network layer into the bidirectional gated recurrent unit network layer, the mutual relationship between elements in the input sequence can be dynamically calculated, and the weighted input at each time step can be realized. The output of the bidirectional gated recurrent unit network layer is updated and iteratively calculated through the self-attention mechanism network layer to generate a better weight parameter matrix, highlighting the influence of different features in the data.

[0113] S4, input the training set into the GJO-GCN-BiGRU-SAM greenhouse environment prediction model for training, and use the test set for verification to obtain the trained greenhouse environment prediction model; wherein, during training, the batch size is 128, the time step is 5, and the regularization parameter is 10 -4 ;

[0114] S5, input the real-time collected indoor environment data and outdoor meteorological data into the trained greenhouse environment prediction model to obtain the prediction results of the indoor temperature, indoor air relative humidity, indoor CO2 concentration and indoor soil temperature of each indoor monitoring node after a preset time length; the preset time length in this step is one hour;

[0115] S6, according to the prediction results, control the adjustment system of the greenhouse to adjust the greenhouse environment;

[0116] In this step, the adjustment system includes a ventilation system, an irrigation system and a heat preservation system. For example, according to the prediction results, when the predicted indoor CO2 concentration of the indoor monitoring node is >1000ppm, start the fan in the ventilation system to ventilate; when the predicted indoor soil temperature is <12℃, start the water pump in the irrigation system; when the predicted indoor temperature at night is <10℃, close the ventilation system and start the heat preservation curtain.

[0117] In order to objectively evaluate the effectiveness of the GJO-GCN-BiGRU-SAM greenhouse environment prediction model (denoted as GCBS model) of the embodiment, the embodiment selects LSTM, BiGRU, CNN-BiGRU and CNN-BiGRU-SAM models for comparison, and compares and analyzes indoor temperature, indoor air relative humidity, indoor CO2 concentration and indoor soil temperature as prediction results.

[0118] Wherein, during comparison, five indoor monitoring nodes are arranged in the greenhouse, each indoor monitoring node is provided with a first multi-source sensor, and the first multi-source sensor includes five first air temperature and humidity sensors, three soil temperature sensors, a first CO2 sensor and a light sensor; the first air temperature and humidity sensors are arranged in layers on each indoor monitoring node, and the installation heights of the first air temperature and humidity sensors are 0.5m, 1.0m, 1.5m, 2.0m and 2.5m respectively; the soil temperature sensors are buried in layers on each indoor monitoring node, and the burial depths of the soil temperature sensors are 10cm, 20cm and 30cm respectively.

[0119] By comparing the key indicators such as MAE, MAPE, R 2 , CV-RMSE, etc., the results as shown in Tables 1 to 4, and Figures 3 to 6The results are shown. Among them, MAE represents the mean absolute error, which is used to measure the average deviation of the predicted value from the true value; MAPE represents the mean absolute percentage error, which reflects the relative error of the prediction; R 2 represents the determination coefficient, which is used to evaluate the goodness of fit of the model; CV-RMSE represents the coefficient of variation root mean square error, which is used to measure the relative volatility of the error.

[0120] Table 1 Prediction of indoor CO2 concentration

[0121]

[0122]

[0123] In the greenhouse, the change of indoor CO2 concentration is quite significant, resulting in obvious differences in prediction accuracy between different models. Table 1 compares the prediction accuracy of indoor CO2 concentration in the test set. As shown in Table 1, the GCBS model in this embodiment performs best in predicting indoor CO2 concentration, with an R 2 value of 0.987 and an MAE of 21.93 (mg / kg) in the average of the five indoor monitoring nodes in the test set. Compared with the LSTM, BiGRU, CNN-BiGRU and CNN-BiGRU-SAM models, the R 2 value of the GCBS model in this embodiment is increased by 10.9%, 4.9%, 3.2% and 1.4% respectively, and the MAE is reduced by 76.65%, 60.90%, 51.81% and 35.42% respectively.

[0124] Table 2 Prediction of indoor air relative humidity

[0125]

[0126] Due to the characteristics of the greenhouse environment, the indoor air relative humidity is usually maintained at a relatively high and stable level with little fluctuation. Table 2 shows a comparative analysis of the prediction accuracy of indoor air relative humidity by various models in the test set. As shown in Table 2, the GCBS model in this embodiment performs best in predicting indoor air relative humidity, with an R 2 value of 0.957 and an MAE of 0.492% in the average of the five indoor monitoring nodes in the test set. Compared with the LSTM, BiGRU, CNN-BiGRU and CNN-BiGRU-SAM models, the R 2 value of the GCBS model in this embodiment is increased by 17.4%, 8.86%, 3.6% and 1.6% respectively, and the MAE is reduced by 69.70%, 51.38%, 50.50% and 35.77% respectively.

[0127] Table 3 Prediction of indoor temperature

[0128]

[0129] The temperature inside the greenhouse generally shows regular and periodic fluctuations. Table 3 shows the comparative analysis of the prediction accuracy of indoor temperature by different models in the test set. The results show that the GCBS model of the present embodiment performs relatively well in predicting air temperature, with an R 2 value of 0.971 and an MAE of 0.556°C in the average of the five indoor monitoring nodes in the prediction set. Compared with the LSTM, BiGRU, CNN-BiGRU and CNN-BiGRU-SAM models, the R 2 of the GCBS model of the present embodiment is increased by 9.6%, 5.0%, 2.4% and 1.9% respectively, and the MAE is reduced by 60.11%, 50.09%, 33.17% and 23.63% respectively.

[0130] Table 4 Indoor soil temperature prediction

[0131]

[0132] The indoor soil temperature and the indoor temperature show consistent trends, showing a clear diurnal variation pattern. Table 4 shows the comparative analysis of the prediction accuracy of soil temperature by different models in the test set. The results show that the GCBS model of the present embodiment has lower prediction error for indoor soil temperature, with an R 2 value of 0.973 and an MAE of 0.044°C in the prediction set. Compared with the LSTM, BiGRU, CNN-BiGRU and CNN-BiGRU-SAM models, the R 2 of the GCBS model of the present embodiment is increased by 9.7%, 5.2%, 3.6% and 1.4% respectively, and the MAE is reduced by 55.56%, 38.03%, 31.25% and 21.43% respectively.

[0133] As Figures 3 to 6 shown, the GCBS model of the present embodiment is compared with the LSTM, BiGRU, CNN-BiGRU and CNN-BiGRU-SAM models in terms of the performance of four key factors in predicting the greenhouse environment, i.e. indoor temperature, indoor air relative humidity, indoor CO2 concentration and indoor soil temperature, according to the average of the five nodes. By constructing comprehensive performance evaluation performances including MAE, MAPE, MSE, CV-RMSE and R 2 , the prediction ability of each model is intuitively evaluated. As Figures 3 to 6 can be seen, the radar chart of the four comprehensive performance evaluations shows that the GCBS model of the present embodiment is superior to the other four models in the five indicators, and the relative comprehensive performance is more comprehensive, and the prediction result of the GCBS model of the present embodiment is better.

[0134] Embodiment 2 A greenhouse environment management and control system based on multi-parameter multi-node prediction

[0135] As shown in the figure, the embodiment is a greenhouse environment management and control system based on multi-parameter multi-node prediction, which is used to implement the greenhouse environment management and control method based on multi-parameter multi-node prediction in the embodiment, and comprises: Figure 7

[0136] A first multi-source sensor network is arranged to set at least one indoor monitoring node in the greenhouse, and indoor environment data is collected by the first multi-source sensor for each indoor monitoring node;

[0137] A second multi-source sensor network is arranged to set at least one outdoor monitoring node outside the greenhouse, and outdoor meteorological data is collected by the second multi-source sensor at each outdoor monitoring node;

[0138] A data transmission module is arranged to transmit the indoor environment data and the outdoor meteorological data to an Internet of Things platform;

[0139] The Internet of Things platform comprises a data preprocessing unit, a model construction unit, a model prediction unit and a display unit;

[0140] The data preprocessing unit is arranged to preprocess the collected indoor environment data and outdoor meteorological data to obtain a sample set, and the sample set is divided into a training set and a test set;

[0141] The model construction unit is arranged to construct a GJO-GCN-BiGRU-SAM greenhouse environment prediction model based on a golden jackal optimization algorithm, and the greenhouse environment after a preset time is predicted; the GJO-GCN-BiGRU-SAM greenhouse environment prediction model comprises an input layer, a convolutional neural network layer, a bidirectional gated recurrent unit network layer, a self-attention mechanism network layer, a first full connection layer and an output layer connected in sequence; the training set is input into the GJO-GCN-BiGRU-SAM greenhouse environment prediction model for training, and the test set is used for verification to obtain the trained greenhouse environment prediction model;

[0142] The model prediction unit is arranged to input the real-time collected indoor environment data and outdoor meteorological data into the trained greenhouse environment prediction model to obtain the prediction results of the indoor temperature, the indoor air relative humidity, the indoor CO2 concentration and the indoor soil temperature content of each indoor monitoring node after a preset time;

[0143] The display unit is arranged to display the prediction results;

[0144] The alarm unit is arranged to alarm when the prediction results exceed the threshold value;

[0145] ​The greenhouse environment adjusting module is used for controlling an adjusting system of the greenhouse according to the prediction result, adjusting the greenhouse environment, and the adjusting system comprises a ventilation system, an irrigation system and a heat preservation system.

[0146] The distance between the indoor monitoring nodes arranged in the greenhouse is 15-25 meters, and in the embodiment, the distance between the indoor monitoring nodes arranged in each greenhouse is 20 meters.

[0147] The first multi-source sensor comprises at least one first air temperature and humidity sensor for collecting indoor temperature and indoor air relative humidity, at least one soil temperature sensor for collecting indoor soil temperature, a first CO2 sensor for collecting indoor CO2 concentration and an illumination sensor for collecting indoor illumination intensity; the first air temperature and humidity sensor and the soil temperature sensor are arranged in layers in the vertical direction of the indoor monitoring node, the height range of the first air temperature and humidity sensor is 0.5-2.5 m, and the embedding depth of the soil temperature sensor is 10-30 cm.

[0148] The second multi-source sensor comprises a second air temperature and humidity sensor for collecting outdoor air temperature and outdoor air relative humidity and a second CO2 sensor for collecting outdoor CO2 concentration.

[0149] The data transmission module simultaneously supports cellular mobile communication and LoRa remote communication, and transmits the indoor environment data and the outdoor meteorological data to the Internet of Things platform through a star topology.

[0150] It should be noted that the above only describes the preferred embodiments of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art can still modify the technical solutions recorded in the above embodiments or equivalently replace some technical features. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A greenhouse environment control method based on multi-parameter and multi-node prediction, characterized in that: The following steps are involved: S1. Set up at least one indoor monitoring node in the greenhouse, collect indoor environmental data for each indoor monitoring node, set up at least one outdoor monitoring node outside the greenhouse, and collect outdoor meteorological data for each outdoor monitoring node; S2. Preprocess the collected indoor environment data and outdoor meteorological data to obtain a sample set, and divide the sample set into a training set and a test set; S3. Constructing a GJO-GCN-BiGRU-SAM greenhouse environment prediction model based on the golden jackal optimization algorithm to predict the greenhouse environment after a preset time period; the GJO-GCN-BiGRU-SAM greenhouse environment prediction model includes an input layer, a convolutional neural network layer, a bidirectional gated recurrent unit network layer, a self-attention mechanism network layer, a first fully connected layer, and an output layer connected in sequence; S4. Input the training set into the GJO-GCN-BiGRU-SAM greenhouse environment prediction model for training, and use the test set for verification to obtain a trained greenhouse environment prediction model; S5. Input the real-time collected indoor environmental data and outdoor meteorological data into the trained greenhouse environment prediction model to obtain the prediction results of indoor temperature, indoor air relative humidity, indoor CO2 concentration and indoor soil temperature of each indoor monitoring node after a preset time period; S6. Based on the prediction results, control the greenhouse regulation system and adjust the greenhouse environment.

2. The greenhouse environment control method based on multi-parameter and multi-node prediction according to claim 1 is characterized in that: In step S1, the horizontal spacing between indoor monitoring nodes is 15 to 25 m; Indoor environmental data include: indoor temperature, indoor air relative humidity, indoor light intensity, indoor soil temperature and indoor CO2 concentration; Outdoor meteorological data include: outdoor air temperature, outdoor air relative humidity and outdoor CO2 concentration.

3. The greenhouse environment control method based on multi-parameter and multi-node prediction according to claim 1, characterized in that: In step S2, the collected indoor environment data and outdoor meteorological data are preprocessed to obtain a sample set. The specific process is as follows: The collected indoor environmental data and outdoor meteorological data are interpolated based on k-means cluster analysis, combined with neighbor interpolation and mean filling strategies to fill missing values; then outliers are removed by setting thresholds; finally, the Min-Max normalization method is used to normalize the data to obtain a sample set.

4. The greenhouse environment control method based on multi-parameter and multi-node prediction according to claim 3 is characterized in that: The ratio of the training set to the test set is 8:

2.

5. The greenhouse environment control method based on multi-parameter and multi-node prediction according to claim 1, characterized in that: In step S3, the convolutional neural network layer includes a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer, a padding layer, and a second fully connected layer connected in sequence; The first convolutional layer and the second convolutional layer are both one-dimensional convolutional layers with a convolution kernel size of 3×3. The convolutional neural network layer uses the ReLU activation function; The pooling size of the first and second maximum pooling layers is 2.

6. The greenhouse environment control method based on multi-parameter and multi-node prediction according to claim 1, characterized in that: In step S3, the golden jackal optimization algorithm includes: (1) Prey search: The mathematical model of the golden jackal's relative position to each individual prey item is expressed as follows: Y1(t)=Y M (t)-E×|Y M (t)-rl×Prey(t)| Y2(t)=Y FM (t)-E×|Y FM (t)-rl×Prey(t)| Among them, Y1(t) is the position of the male golden jackal after the position of the prey is updated; Y2(t) is the position of the female golden jackal after the position of the prey is updated; Y M (t) is the initial position of the male golden jackal; Y FM (t) is the initial position of the female golden jackal; t is the current iteration number; Prey(t) is the prey position; Y M (t)-rl×Prey(t) is the relative distance between the male golden jackal and the prey during the tracking phase, Y FM (t)-rl×Prey(t) is the relative distance between the female golden jackal and the prey during the tracking phase; E is the escape energy of the prey, and the calculation formula is: E=E1E0 Among them, E1 is the initial energy when the prey escapes; E0 is the decay state of energy; E0＝2r-1 Where r is a random number in [0, 1]; Where c1 is a constant with a value of 1.5, that is, E1 decays linearly from 1.5 to 0 during the prey's escape process; T is the maximum number of iterations; rl is a random number based on the Levy distribution, and rl×Prey(t) reflects the movement mode of the prey’s escape; rl=0.05×LF(y) Where, μ and v are random numbers in [0, 1]; β is a constant 1.5; LF(y) is the Levy flight function; (2) As the prey's escape energy continues to decay, the male and female golden jackals surround the prey and capture it, obtaining updated positions of the male and female golden jackals; Y1(t)=Y M (t)-E×|rl×Y M (t)-Prey(t)| Y2(t)=Y FM (t)-E×|rl×Y FM (t)-Prey(t)| The Golden Jackal's location has been updated to:

7. A greenhouse environment control system based on multi-parameter and multi-node prediction, used to implement the greenhouse environment control method based on multi-parameter and multi-node prediction according to any one of claims 1 to 6, characterized in that: include: A first multi-source sensor network is configured to set up at least one indoor monitoring node in the greenhouse, and collect indoor environmental data from each indoor monitoring node using a first multi-source sensor; A second multi-source sensor network is configured to set up at least one outdoor monitoring node outside the greenhouse, and collect outdoor meteorological data at each outdoor monitoring node through a second multi-source sensor; Data transmission module, used to transmit indoor environmental data and outdoor meteorological data to the Internet of Things platform; IoT platform, including data pre-processing unit, model building unit, model prediction unit and display unit; The data preprocessing unit is used to preprocess the collected indoor environmental data and outdoor meteorological data to obtain a sample set, and divide the sample set into a training set and a test set; The model construction unit is used to construct a GJO-GCN-BiGRU-SAM greenhouse environment prediction model based on the golden jackal optimization algorithm to predict the greenhouse environment after a preset time period; the GJO-GCN-BiGRU-SAM greenhouse environment prediction model includes an input layer, a convolutional neural network layer, a bidirectional gated recurrent unit network layer, a self-attention mechanism network layer, a first fully connected layer and an output layer connected in sequence; the training set is input into the GJO-GCN-BiGRU-SAM greenhouse environment prediction model for training, and the test set is used for verification to obtain a trained greenhouse environment prediction model; The model prediction unit is used to input the real-time collected indoor environmental data and outdoor meteorological data into the trained greenhouse environment prediction model to obtain the prediction results of indoor temperature, indoor air relative humidity, indoor CO2 concentration and indoor soil temperature content of each indoor monitoring node after a preset time period; A display unit, used to display the prediction results; An alarm unit, used to issue an alarm when the prediction result exceeds a threshold; The greenhouse environment adjustment module is used to control the greenhouse regulation system and adjust the greenhouse environment according to the prediction results.

8. The greenhouse environment control system based on multi-parameter and multi-node prediction according to claim 7 is characterized in that: The distance between indoor monitoring nodes set up in the greenhouse is 15 to 25 meters; The first multi-source sensor includes at least one first air temperature and humidity sensor for collecting indoor temperature and indoor air relative humidity, at least one soil temperature sensor for collecting indoor soil temperature, a first CO2 sensor for collecting indoor CO2 concentration, and a light sensor for collecting indoor light intensity. The first air temperature and humidity sensor and the soil temperature sensor are arranged in layers in the vertical direction of the indoor monitoring node. The height range of the first air temperature and humidity sensor is 0.5 to 2.5 meters, and the burial depth of the soil temperature sensor is 10 to 30 cm. The second multi-source sensor includes a second air temperature and humidity sensor for collecting outdoor air temperature and outdoor air relative humidity and a second CO2 sensor for collecting outdoor CO2 concentration.

9. The greenhouse environment control system based on multi-parameter and multi-node prediction according to claim 7, characterized in that: The data transmission module supports both cellular mobile communications and LoRa remote communications, and transmits indoor environmental data and outdoor meteorological data to the Internet of Things platform through a star topology.

10. The greenhouse environment control system based on multi-parameter and multi-node prediction according to claim 7, characterized in that: The conditioning system includes a ventilation system, an irrigation system and a heat preservation system.

Images