Time Series Prediction of Humidity in Greenhouse Using GRU Neural Network Architectures

Abstract

Regulating the climate in greenhouses is one of the main obstacles faced in precision farming, mainly because the factors that affect the environment change in a very dynamic and nonlinear way. The current research introduces a forecasting structure that relies on a Gated Recurrent Unit (GRU) neural network, which is specifically designed for identifying temporal trends in time-series data. The aim is to forecast three vital microclimate parameters outdoor temperature, indoor temperature, and indoor relative humidity based on actual measurements from an experimental greenhouse. Data preprocessing, GRU model creation, model training and evaluation are the parts of the methodology. The performance of the GRU model in providing predictions that are both accurate and consistent has been demonstrated through the results, which further revealed its effectiveness in dealing with complex interactions that govern the climate in greenhouses. The prospected reliability of the suggested technique as a dependable tool for smart monitoring of greenhouses and the creation of automated climate control systems has been underscored by these results.