Prediction of CO2 Concentration via Long Short-Term Memory Using Environmental Factors in Greenhouses
Taewon Moon, Ha Young Choi, Dae Ho Jung, Se Hong Chang, Jung Eek Son
Abstract
In greenhouses, photosynthesis efficiency is a crucial factor for increasing crop production. Since plants use CO2 for photosynthesis, predicting CO2 concentration is helpful for improving photosynthetic efficiency. The objective of this study was to predict greenhouse CO2 concentration using a long short-term memory (LSTM) algorithm. In a greenhouse where mango trees (Mangifera indica L. cv. Irwin) were grown, temperature, relative humidity, solar radiation, atmospheric pressure, soil temperature, soil humidity, and CO2 concentration were measured using complex sensor modules. Nine sensors were installed in the greenhouse. The averages of environmental factors from the nine sensors were used as inputs, and the average CO2 concentration was used as an output. In this experiment, LSTM, one of the recurrent neural networks, predicted changes in CO2 concentration from the present to 2 h later using historical data. The data were measured every 10 min from February. 1, 2017 to May 31, 2018, and missing data were interpolated with a linear method and multilayer perceptron. In this study, LSTM predicted the 2-h change in CO2 concentrations at an interval of 10 min with adequate test accuracy (R-2 = 0.78). Therefore, the trained LSTM can be used to predict the future CO2 concentration and applied to efficient CO2 enrichment for photosynthesis enhancement in greenhouses.