Machine-Learning Microclimate Forecasting for Adaptive Equipment Control via Web Integration in Open-Ventilated Greenhouses

Abstract

Open-ventilated greenhouses have reasonable setup costs and low operational costs for growers, which is crucial and most appealing for this research. These attributes fit developing nations like Thailand and other tropical regions. It is challenging to control the equipment intended to obtain an ideal microclimate. This research was conducted in an actual greenhouse setting for data collection and experiments, with a proposed system for adaptive equipment control via web integration. Also, the proposed multivariate multistep LSTM was forecasted over 1 h and cooperated with sensor data. Additional sensors, like a leaf wetness sensor and a CO2 sensor, were installed for detecting plant-level precision for vaporization, rather than greenhouse-level. The proposed system can optimize the indoor temperature within 34.5 to 36 °C with a 39 to 40 °C outdoor temperature. Also, humidity was still at the ideal level of 68 to 70%; more precisely, the wetness value was below 300 throughout the experiment. The model accuracy achieved a sufficient RMSE (0.49) and R2 (0.9788). This proposed system architecture and MM-LSTM model has potential as one dimension of a fully smart greenhouse system development in open-ventilated greenhouse settings in tropical regions and Southeast Asian nations for a better yield rate and less human interaction.