Prediction of Air Temperature and Relative Humidity in Greenhouse based on Neural Network Model

doi:10.21203/rs.3.rs-4958200/v1

Greenhouse environmental factors such as temperature and humidity are critical environmental factors influencing plant development, quality, and production quantity, therefore, it is crucial to predict the temperature and humidity. In this study, a back propagation(BP)neural network was proposed to model the temperature and relative humidity of a greenhouse and have obtained a high accuracy pridiction modole. The model considered both internal and external environmental factors influencing crop growth, the best pridiction result accurred when input featurears are outdoor temperature, relative humidity, irradiation intensity and previors indoor temperature and relative humidity,with 9d of training data set, in which the R² of temperature is 0.999, the RMSE is 0.299℃,and error range is[ -0.240℃, 0.781℃], and the R² of relative humidity is 0.999, the RMSE is 0.657%, and the error range is [-0.766%, 1.995%]. This model can be used to forecast the temperature and humidity of a greenhouse and guide the control of the temperature and humidity of a greenhouse.

BP neural network

Levenberg-Marquardt

temperature prediction

relative humidity prediction

Greenhouses are a widely used system for artificially creating a suitable growth microclimate environment for plants. Greenhouse environmental factors such as temperature and humidity are critical environmental factors influencing plant development, quality, and production quantity[1]. More specifically, the exposure of plants to extreme temperatures can heat damage or frost damage and it is considered that the increase in relative humidity may exacerbate fungal diseases as well as affect the uptake and utilization of calcium and the normal water balance of plants[2]. Therefore, these factors need to be taken into account when growing and managing plants, and appropriate measures need to be taken to protect plants from the adverse effects of high humidity environments. Greenhouse climatic control necessitates the consideration of complex, non-linear and strong coupling systems, where variables are highly dependent on external climate conditions as well as structure type, design and orientation [3]. Obviously, building a precise model of inside greenhouse climate is an important way to achieve efficient microclimate management[4–6].

Modeling of solar greenhouses can either be based on physical laws or system identification method. As far as control is concerned, modeling using system identification methods, is more advantageous than physical modeling and is more responsive to the requirements of environmental factor regulation[7–9].With the rapid development of artificial intelligence technology, Machine learning is a widely used method to predict the temperature and humidity in greenhouses[10–12].ANN(Artificial neural network) models are powerful forecasting tools for analyzing nonlinear systems due to their ability to model systems without making assumptions implicit in most traditional statistical approaches. One of the most important advantages of ANN models over other classes of nonlinear models is that the former are universal approximators that can approximate a large class of functions with a high degree of accuracy. These have been applied to the prediction of greenhouse climatic data with superior results than those of physical models[13, 14]. Castañeda&Castaño et al[15]created a multilayer perceptron Neural Network with Levenberg-Marquardt as a training algorithm, which can predict the indoor temperature, with the calculated coefficients of determination being equal to 0.9549 and 0.9590, for the winter and summer season, respectively, to control the frost inside the greenhouse. Theodoros Petrakis et al[16]used a BP neural network to predict temperature and relative humidity in a greenhouse, and the inputs were 10 variables including historical indoor temperature and relative humidity. The number of nodes in the hidden layer of the BP neural network was derived through extensive experimental comparison. Zhang Yongfang et al[17] used the SSA (Sparrow Search Algorithm) to optimize the RBF network, to carry out a simulation study on the temperature and humidity of greenhouse, the coefficients of determination(R²) of the temperature and humidity were around 0.86, but the algorithm has a relatively large number of input variables. Hayoung Choi et al[18] predicted the indoor temperature and relative humidity of an eight-span greenhouse using multilayer perceptron (MLP), inputs for the MLP were greenhouse inside and outside environment data, and set-up and operating values of environment control devices. Discussed in detail the effects of different numbers of hidden layers and nodes on the prediction accuracy, and found the optimal network structure with four hidden layers and 128 nodes for air temperature (R² = 0.988) and with four hidden layers and 64 nodes for relative humidity (R² = 0.990), However, too many hidden layer nodes can lead to an overly large network structure, making the computation slower or less pervasive[19]. He F. et al[20]proposed A BP neural network based on principal component analysis (PCA) for modeling the internal greenhouse humidity in winter of North China,The input variables of the network were selected using principal component analysis and the prediction was good. Liu et al [21]established a BP neural network model to predict the temperature variation trend in the greenhouse for the future 1 to 7 days, based on the daily meteorological data of Tianjin from 2011 to 2013 in winter.Youjun Yue et al[22]proposed a model to predict the temperature and humidity of a greenhouse based on improved LM-RBF, The model used the inside and outside meteorological data of the greenhouse as input, and the temperature and humidity in a greenhouse as output, but they did not consider the effect of historical indoor temperature and humidity on the prediction accuracy of the system .Yu et al[23] established a new prediction model based on least squares support vector machine (LSSVM) using environmental factors of Shandong Province to predict the temperature of a greenhouse, they took into account the Previous inside temperature and predicted temperature well, but did not predict indoor humidity. Zou Weidong et al [24]took the previous inside humidity and inside temperature and the external meteorological data as the input vector of the prediction model, established the prediction model of temperature and humidity in greenhouse, but the range of error in the prediction was slightly large.

In this research, we employed BP neural network under different input features to predict the greenhouse temperature and relative humidity. First, we constructed an experimental platform to obtain greenhouse environmental data. Furthermore, we used the Pearson correlation coefficient(PCC) to identify the external meteorological factors influencing the temperature and relative humidity variation inside the greenhouse, and incorporated previous values of temperature and relative humidity in greenhouse into the input data. Then, we established a dataset and used BP neural network with various input features achieve prediction of greenhouse temperature and relative humidity. Finally, the accuracy and error of the prediction results were contrasted and found the best input features. In addition, the minimum amount of necessary data was determined. The results showed that introduce previous temperature and relative humidity could significantly improve the prediction accuracy of temperature and humidity. This study offered technical support for precise temperature and humidity regulation and early warning of damage in greenhouses.

2.1 Greenhouse and measurements

All experiments were based on a greenhouse (experimental area: 25m × 8m, 200 m²) located in Hohhot, Inner Mongolia, northwestern China, Hailiu Village Science and Technology Park of Inner Mongolia Agricultural University (lat. 40.68_N, long. 111.38_E). The greenhouse is equipped with inner and outer double-layer non-dripping film and inner and outer thermal insulation quilts, the back wall is brick-concrete structure and has an external thermal insulation layer of 8 cm thick. As shown in the Fig. 1, in addition, the greenhouse has added a solar heating system. Compared with other types of greenhouses, the experimental greenhouse’s minimum temperature of solar greenhouse is above zero, it is suitable for the cold, dry areas of northern China. When the sunshine is sufficient at about nine o 'clock in the morning in winter, the outer quilt, the inner quilt and the inner plastic film are opened so that the solar greenhouse can absorb more solar radiation, which can promote plant growth and make the solar greenhouse temperature rise rapidly, the top vent is opened at noon. At about 4 p.m., the solar radiation gradually weakened, and the inner and outer quilts and inner film were put down, which slowed down the heat loss of the solar greenhouse to outside and played a role in heat preservation. At the same time, the solar heating circulation system was opened at about 8 p.m., and the ground temperature at night was basically higher than 10°C, the air temperature was almost all above 9°C. The change trend of relative humidity was opposite to that of temperature.

Experiments were conducted in winter from December 21, 2020 to March 1, 2023, the crop planted is celery. The inside (Fig. 1b) climate (temperature, relative humidity, CO₂, illumination and soil temperature & moisture content) was measured by the corresponding sensors (Fig. 2) consisting of a CO₂ temperature and humidity three-in-one transmitter ( Shandong Renke : RS-CO₂WS-N01-2 type ), a total solar radiation sensor( Shandong Renke : RS-RA-N01-AL type ), soil temperature and moisture sensor(Kong Saien KE-N01-TR-1 type), the temperature range is − 10 ~ 50°C, the accuracy is ± 0.5°C, the relative humidity range is 0 ~ 95%, the accuracy is ± 3%, the carbon dioxide concentration range is 0 ~ 5000 ppm, and the accuracy is ± 40 ppm. The environment data was collected every half an hour and saved to a database. Exterior conditions were measured by portable small automatic weather station (HOBO U30 NRC) in the United States, the data is transmitted through the USB communication interface, and the measurement information is read by the HOBO ware Pro software on the computer side. The structure of the greenhouse environmental factor data acquisition system is shown in Fig. 2. The CO₂, air temperature and humidity three-in-one sensor, and the solar radiation sensor are 1.8 m high from the ground, which are located at the central point of the east-west and north-south directions of the greenhouse. Soil temperature and moisture sensor was arranged 5cm below the soil in the middle of the greenhouse to measure the surface temperature.

2.2 Data and Methodology

The dataset was selected for indoor and outdoor environmental parameters, including outdoor temperature($\:T$o), relative humidity $\:（Huo）$,wind speed༈$\:V$༉, indoor air temperature༈$\:Ti$༉,humidity༈$\:Hu$i༉, soil temperature༈$\:Ts$༉, soil water content༈$\:Hs$༉, irradiation intensity༈$\:I$i༉, the time-step of the data is equal to 10 min.The original dataset was separated into a training set and a test set. The training set was used for model training, and the test set was used to evaluate model performance.

For the case where the original data acquisition system has a missing value at a certain time, the interpolation method is used to fill in, the variables of continuous changes such as temperature and humidity is filled by the cubic spline interpolation method, missing value for wind speed and solar irradiance are replaced by the nearest neighbor principle.

It is clear that environmental factors have neither the same units nor the same orders of magnitude. Thus, variables with higher values will contribute more to the output error, with the result that the algorithm gives more weight to them and less to variables with a smaller range of values [25]. For this reason, data must be normalized before being presented to the network. Data normalization compresses the range of the data between 0 and 1. The normalization in this study was carried out by means of the following expression given by the Eq. (1) [26].

$$\:{X}_{n}=\frac{{X}_{r}-{X}_{min}}{{X}_{max}-{X}_{min}}$$

1

Where $\:{X}_{n}\:$is the normalized value of $\:{X}_{r}$,$\:\:{X}_{r}$ is a real value,$\:\:{X}_{max}$ and $\:{X}_{min}$ denote the maximum and minimum values in the data respectively.

2.3 Predictive model evaluation indicators

To verify the predictive performance of the BP neural network model for environmental factors in solar greenhouses, the determination coefficient (R²), Root Mean Square Error (RMSE) were selected as model evaluation indexes. The calculation equations of each evaluation index are as follows[27].

1.Root Mean Squared Error (RMSE)

The root mean square error, also known as the standard error, reflects the degree of deviation between the true value and the predicted value.

$$\:RMSE=\sqrt{\frac{{\sum\:}_{i=1}^{n}{({y}_{i}-{\widehat{y}}_{i})}^{2}}{n}}$$

2

2. Determination coefficient($\:{R}^{2}$)

$\:{R}^{2}$ indicates the closeness of the linear relationship between the predicted and measured value. The closer$\:\:{R}^{2}$is to 1, the better the model effect is. When the sum of squares of prediction residuals is greater than the sum of squares of variance of the true value, $\:{R}^{2}\:$becomes negative, indicating that the fitting effect is extremely poor, and the prediction effect is not as good as the direct mean value.

$$\:{R}^{2}=1-\sum\:_{i=1}^{n}{({y}_{i}-{\widehat{y}}_{i})}^{2}/\sum\:_{i=1}^{n}{({y}_{i}-\stackrel{-}{y})}^{2}$$

3

In the above equations, n is the number of samples, $\:{y}_{i}$ is the measured value of the i th sample, $\:{\widehat{y}}_{i}$is the predicted value of the i th sample, and $\:\stackrel{-}{y\:}$ is the average of the measured values of the n samples.

2.4 BP Neural Network Model

The BP (back propagation) neural network is a typical multilayer network of hierarchical type with input, hidden and output layers, and the layers are mostly fully connected to each other and network trained according to the error back propagation algorithm [28, 29], the network was trained and predicted with the Levenberg-Marquardt algorithm, Levenberg-Marquardt algorithm is one of improved algorithms and it combines advantages of the method of grads descent and Newton. It absorbs not only the target function’s first derivative information, but also the second, which make the learning time shorter and speed up the convergence process. It is an efficient method for nonlinear forecasting. The advantages of the BP neural network model lie in the few parameters required and the high simulation accuracy.

The different environmental factors had a nonlinear coupling relationship with Ti and Hui. In order to reduce redundant information about the input feature and the computational complexity of the model, PCC was employed to analyze the relationship among the Ti, Hui, and others. The computation results of PCC are demonstrated in Fig. 3. Ii, Hui, and Ti are very strongly correlated, To, Huo, V, and Ti are strongly correlated, Ts and Ti are weakly correlated, whereas Hs and Ti are very weakly correlated. Thus, the primary factors affecting greenhouse temperature are Ii, Hui, To, Huo, V and Ts. Similarly, Ti, Ii, and Hui are highly correlated, To, V, and Hui are strongly correlated, and Huo and Hui are weakly correlated, while Ts, Hs, and Hui are very weakly correlated. Therefore, the main environmental factors for establishing a humidity prediction model are Ti, Ii,To, V, and Huo.

On the basis of the above analysis and thinking, the outside temperature, air humidity, solar radiation, wind speed, soil temperature, and previous inside temperatures and humidity form the input vector of the multiple-input/multiple-output( MIMO ) forecasting model.In order to further optimize the number of input features, this study divides the features derived through PCC into several groups for experimentation. The details are shown in Table 1. Among them, G1 ~ G3 are the most relevant features in PCC analysis. According to the correlation coefficient, 3,4 and 5 input features are selected respectively. G4 and G5 increase the historical data of indoor temperature and relative humidity in the previous 10 min and 20 min respectively on the basis of G1.

Table 1

Grouping of input features
Input Group	G1	G2	G3	G4	G5
Input Features	To, Huo, Ii	To, Huo, Ii,V	To, Huo, Ii,V,Ts	To, Huo, Ii, Ti(t-t0),Hui(t-t0)	To, Huo, Ii,Ti(t-t0),Hui(t-t0), Ti(t-2t0),Hui(t-2t0)

Ti(t-t0),Hui(t-t0) is temperature and humidity data from 10 minutes ago,while Ti(t-2t0),Hui(t-2t0)is temperature and humidity data from 20 minutes ago.

3.1 Prediction and evaluation of indoor air temperature

In the following analysis, the indoor air temperature prediction models were compared, when the input data was G1, G2, G3, G4, G5, respectively. In addition, based on the developed environmental prediction accuracy, we compared the model development results by applying various training set conditions. In order to identify the minimum amount of training data required for model development, periods of 3, 5, 7, 9d was used as training data sets based on the dataset and for model development, the test set is the indoor temperature in the next 24 hours.Comparison result of indoor air temperature is shown in Table 2.

Table 2 shows that when the input features are the same, with the increase of the number of training samples, the overall trend of indoor temperature prediction results is that the coefficient of determination(R²) is higher, and the root mean square error(RMSE) and error range are smaller.When the number of training samples is the same, with the change of input features, G4 and G5 with the input features of previous temperature and relative humidity can obtain higher prediction accuracy. For example, when input is G5, all results showed relatively high accuracy for temperature prediction during varying number of training days. Specifically, while training data sets increased from 3 days to 9 days, R² increased from 0.994 to 0.999, RMSE decreased from 0.629℃ to 0.299℃, and the error range decreased from [ -1.064℃, 1.523℃] to [ -0.240℃, 0.781℃].

Table 2

prediction result comparison of indoor air temperature
Training data sets(day)	Input group	R²	RMSE(℃)	error range(℃)
3	G1 G2 G3 G4 G5	0.817 0.874 0.870 0.989 0.994	3.521 2.919 2.964 0.843 0.629	[-1.119, 9.271] [-1.588, 9.603] [-1.364, 8.392] [-1.002, 3.131] [-1.064, 1.523]
5	G1 G2 G3 G4 G5	0.918 0.934 0.927 0.994 0.997	2.300 2.075 2.179 0.614 0.436	[-3.835, 5.352] [-5.988, 6.827] [-2.265, 5.846] [-2.241, 1.387] [-1.617, 0.715]
7	G1 G2 G3 G4 G5	0.970 0.925 0.871 0.984 0.997	1.425 2.215 2.909 1.040 0.433	[-3.381, 4.526] [-4.500, 3.232] [-7.799, 3.555] [-1.646, 2.440] [-0.934, 0.808]
9	G1 G2 G3 G4 G5	0.964 0.907 0.913 0.999 0.999	1.528 2.131 2.067 0.302 0.299	[-2.548, 4.580] [-8.062, 2.933] [-7.936, 1.887] [-0.670, 0.778] [-0.240, 0.781]
*Bolded text shows the best results

In Fig. 4 and Fig. 5, the time series of the predicted and actual temperature values for the 24h used for testing the model are presented as well as the absolute errors between the respective values. As shown in Fig. 4, The 144 samples are considered as equivalent to 24h, the model has predicted temperature to a large extent, and the maximum errors occur during the daytime hours when the temperature presents high values but also abrupt changes due to ventilation(Fig. 5). However, such error has little effect on the design of temperature control system. The model can be further applied to the temperature control system of greenhouse.

3.2 Prediction and evaluation of indoor relative humidity in greenhouses

The analysis and comparison of relative humidity prediction accuracy is the same as that of indoor temperature prediction. Table 3 presents the prediction performance of indoor relative humidity when the training data sets and input group change. It shows that when the input features are the same, with the increase of the number of training samples, the overall trend of indoor relative humidity prediction results is that the coefficient of determination(R²) is higher, and the root mean square error(RMSE) and error range are smaller,When the number of training samples is the same, with the change of input features, G4 and G5 with the input features of previous temperature and relative humidity can obtain higher prediction accuracy. For example, when input is G5, all results showed very high accuracy for relative humidity prediction during varying number of training days. Specifically, while training data sets is 3 d, the prediction was excellent with R² of 0.993, RMSE is 6.707%, and the error range is [-23.881%, 2.279%].Using 9 d, the R² reached 0.999, RMSE reduced to 0.657%, and the error range reduced to [-0.766%, 1.995%], a highly satisfactory performance, far better than the other prediction results.

Table 3

Comparison result of indoor relative humidity
Training data sets(day)	Input group	R²	RMSE(%)	error range(%)
3	G1 G2 G3 G4 G5	0.876 0.888 0.929 0.990 0.993	6.707 6.375 5.084 1.867 1.643	[-23.881, 2.279] [-25.507, 7.718] [-18.447, 4.723] [-4.482, 4.484] [-3.874, 3.704]
5	G1 G2 G3 G4 G5	0.812 0.890 0.912 0.996 0.998	7.848 5.991 5.375 1.212 0.713	[-14.309, 25.476] [-16.105, 24.586] [-13.949, 16.006] [-2.674, 5.187] [-1.600, 3.446]
7	G1 G2 G3 G4 G5	0.933 0.905 0.807 0.983 0.998	4.571 5.352 7.642 2.306 0.730	[-8.050, 17.707] [-10.671, 15.559] [-15.671, 28.491] [-1.435, 4.997] [-1.582, 2.738]
9	G1 G2 G3 G4 G5	0.967 0.910 0.912 0.998 0.999	3.460 5.355 5.292 0.825 0.657	[-4.952, 12.759] [-21.116, 9.997] [-8.485, 17.228] [-2.594, 1.923] [-0.766, 1.995]
*Bolded text shows the best results

Figure 6 illustrates the comparison findings between the actual and predicted relative humidity when input group is G5, and training set is 9d, great predictability is presented again, with the two curves being almost identical. According to Fig. 7, the maximum error occurs at the time when the humidity changes sharply at noon, and the error range is very small.

3.3 Discussion

The comparison results show that when the input features were only the most relevant G1 group (To, Huo and Ii), the prediction accuracy is generally low with maximum R² of 0.970(temperature), 0.967(relative humidity), with the increase of the training data from 3d to 9d, the coefficient of determination(R²) increases a bit. When the inputs were the G1 ~ G3 groups, the R² of the temperatures and relative humidity prediction change very little when the training set was same, there are even a decrease in the R² with the increase of the number of input features, which indicates that the two added input features V and Ts did not contribute significantly to the improvement of prediction accuracy.The corresponding R² of both G4 and G5 input groups are very high, in which the R² of temperature is above 0.984, the RMSE is below 1.040, and the error range is within [-1.646℃, 2.440℃], and the R² of relative humidity is above 0.983, the RMSE is below 2.306, and the error range is within [-1.435%, 4.997%] .

The above analysis of the prediction results for indoor temperature and relative humidity revealed that the best predictions were made in the G5 input group, which contains the previous 10-minute and 20-minute temperatures and humidity in addition to the three data that have the strongest correlation with indoor temperature and relative humidity, i.e., outdoor temperature(To), humidity(Huo), and intensity of irradiation(Ii). The change of internal temperature and relative humidity of greenhouse is closely related to the historical temperature and relative humidity. This result proves that for the data set with obvious time series characteristics, adding the data set of historical time will be of great help to the improvement of prediction accuracy.

After extensive literature research, it was found that the prediction model for indoor temperature and relative humidity proposed in this study has smaller RMSE values and the coefficient of determination R² is very close to 1, combined with the prediction curves and error curves for temperature and relative humidity in Fig. 4-Fig. 7, proved that the model has excellent performance. In addition, the maximum error is not a widely used data comparison value, but for this study, this data is very important and its value is quite small.

In this study, we developed a BP neural network-based prediction method for greenhouse temperature and relative humidity, and obtained a high prediction accuracy. To ensure the accuracy of the prediction, The data pretreatment was made first,then analyzed the correlation between indoor temperature, relative humidity and other environmental factors by PCC, sorted and classified the input features into groups of G1 ~ G5 on the basis of PCC, and tested them separately. The effects of variations in the input features and the number of training samples on the prediction model were investigated. The best results of indoor temperature and relative humidity prediction were obtained when input is G5 group with 9d of training set data, in which R² of temperature and relative humidity are both 0.999, are very close to 1, as well as RMSE and error range are quite small. The above results prove that the BP neural network prediction model proposed in this study can can effectively predict the temperature and humidity inside greenhouse, and also provide theoretical support for the controller design of greenhouse environment.

Author Contributions: Caixia Yan: Conceptualization, Methodology, Software, Validation, Writing – original draft. Ta Na: Resources, Supervision, Project administration, Funding acquisition. Qi Zhen: Writing – review & editing, Project administration. Yunfeng Sun: Conceptualization, Project administration. Kunyu Liu: Formal analysis, Investigation. Xiaokai Li: Formal analysis, Data curation. All authors have read and agreed to the published version of the manuscript.

Funding: This research was funded by the University Basic Scientific Research Project of Inner Mongolia Autonomous Region Directly-Interdisciplinary Research Project (No. BR22-14-03), the Inner Mongolia Natural Science Foundation of China (No. 2022MS03040）.

Institutional Review Board Statement: Not applicable.

Informed Consent Statement: Not applicable.

Data Availability Statement: The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Conflicts of Interest: The authors declare no conflicts of interest.

D. H. Jung, T.S.L.K., A deep learning model to predict evapotranspiration and relative humidity for moisture control in tomato greenhouses. Agronomy, 2022. 9(12): p. 2169.
Rabbi, B., Z.H. Chen and S. Sethuvenkatraman, Protected Cropping in Warm Climates: A Review of Humidity Control and Cooling Methods. Energies, 2019. 12(14): p. 2737.
El Ghoumari, M.Y.T.H., Non-linear constrained MPC: realtime implementation of greenhouse air temperature control. Comput. Electron. Agric. Computers & Electronics in Agriculture, 2005. 49(3): p. 345–356.
Kacira, E.T.A.M., Analysis of climate uniformity in a naturally ventilated greenhouse equipped with high pressure fogging system using computational fluid dynamics. ActaHorticulturae, 2013. 51(1008): p. 177–183.
Jiao, W., et al., Computational Fluid Dynamics-Based Simulation of Crop Canopy Temperature and Humidity in Double-Film Solar Greenhouse. Journal of Sensors, 2020(2020): p. 0-8874468.
Ramin, S.R., et al., Review of optimum temperature, humidity, and vapour pressure deficit for microclimate evaluation and control in greenhouse cultivation of tomato: a review. International Agrophysics, 2018. 32(2): p. 287-302.
Joao Israel Bernardo, M.S.S.M., Analytical modeling in greenhouse for simulation and optimized multivariable control. Computers in Agriculture and Natural Resources, 4th World Congress Conference, Orlando, Florida USA, , ASABE Publication, 2006. 701(2006): p. 612-617.
Lopez-Cruz, I., et al., Development and analysis of dynamical mathematical models of greenhouse climate: A review. European Journal of Horticultural Science, 2018. 83(5): p. 269-279.
Xu Lihong, S.Y.L.Y., Requirement and current situation of control-oriented microclimate environmental model in greenhouse system. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2013. 29(19): p. 1-15.(in Chinese with English abstract)
Ou Shichao, L.C.J.G., Application Of Artificial Neural Network Model In Greenhouse Plantation Environments. ASAE Annual International Meeting Presentation ,Las Vegas,Nevada,USA., 2003: p. 1-6.
Qi, L. and Z. Feng, Improved BP Neural Network of Heat Load Forecasting Based on Temperature and Date Type. Journal of System Simulation, 2018. 30(4): p. 1464-1472.
Zhang, S. and S. Liao, Simulation and Analysis of Spatialization of Mean Annual Air Temperature Based on BP Neural Network. Journal of Geo-Information Science, 2011. 13(4): p. 534-538.
Devi, C.J., ANN approach for weather prediction using back propagation. international journal of engineering trends & technology, 2012. 1(3): p. 19-23.
Jung, D.H., et al., Time-serial analysis of deep neural network models for prediction of climatic conditions inside a greenhouse. Computers and Electronics in Agriculture, 2020. 173(2020): p. 105402.
Alejandro Castañeda-Miranda A, V.M.C.B., Smart frost control in greenhouses by neural networks models. Computers and Electronics in Agriculture, 2017. 137(2017): p. 102-114.
Petrakis, T., et al., Neural Network Model for Greenhouse Microclimate Predictions. Agriculture, 2022. 12(6): p. 780.
ZHANG Yongfang, W.F., Study on temperature and humidity prediction model of solar greenhouse based on SSA-RBF network.Journal of Agricultural University of Hebei. 2021. 44(3): p. 115-121.(in Chinese with English abstract)
Hayoung Choi, T.M.D.H., Prediction of Air Temperature and Relative Humidity in Greenhouse via a Multilayer Perceptron Using Environmental Factors. Protected Horticulture and Plant Factory, 2019. 2(28): p. 95-103.
Xu, C., Optimization Analysis of Dynamic Sample Number and Hidden Layer Node Number Based on BP Neural Network. Advances in Intelligent Systems and Computing, 2013. 212(2013): p. 687-695.
He, F. and C. Ma, Modeling greenhouse air humidity by means of artificial neural network and principal component analysis. Computers & Electronics in Agriculture, 2010. 71(2010): p. 19-23.
Shu-Mei, L., et al., An air temperature predict model based on BP neural networks for solar greenhouse in North China. Journal of China Agricultural University, 2015. 20(1): p. 176-184.
Yue, Y., et al. The Prediction of Greenhouse Temperature and Humidity Based on LM-RBF Network. in 2018 IEEE International Conference on Mechatronics and Automation. 2018.
Yu, H., et al., Prediction of the temperature in a Chinese solar greenhouse based on LSSVM optimized by improved PSO. Comput. Electron. Agric., 2016. 122: p. 94–102.
Zou, W., et al., Verification and predicting temperature and humidity in a solar greenhouse based on convex bidirectional extreme learning machine algorithm. Neurocomputing, 2017: p. 72-85.
Bhanja, S. and A. Das, Impact of Data Normalization on Deep Neural Network for Time Series Forecasting. 2018.
Yongxia Yang A, B.P.G.A. and B.C.J.H. Yingying Liu A. Multistep ahead prediction of temperature and humidity in solar greenhouse based on FAM-LSTM model. Computers and Electronics in Agriculture, 2023,213 :p.10826.
Kamilaris, A.P.F.X., Deep learning in agriculture: a survey. Comput. Electron. Agric., 2018. 147: p. 70–90.
Liu, X.Y., Agricultural Irrigation Forecasting Based on Grey Relational Analysis and BP Neural Network. J. Math. Pract. Theory, 2020. 2020(50): p. 287–291.
Hsieh, K.L.L.Y., Model construction and parameter effect for TFT-LCD process based on yield analysis by using ANNs and stepwise regression. Expert Systems with Applications, 2008. 34(1): p. 717–724.

No competing interests reported.

Prediction of Air Temperature and Relative Humidity in Greenhouse based on Neural Network Model

Status:

Version 1

Abstract

Figures

1. Introduction

2. Materials and Methods

2.1 Greenhouse and measurements

2.2 Data and Methodology

2.3 Predictive model evaluation indicators

1.Root Mean Squared Error (RMSE)

2.4 BP Neural Network Model

3. Results and Discussion

3.1 Prediction and evaluation of indoor air temperature

3.2 Prediction and evaluation of indoor relative humidity in greenhouses

3.3 Discussion

4. Conclusions

Declarations

References

Additional Declarations

Status:

Version 1