TA-LSTM and IoT Based Energy Management for PV-Powered Greenhouses

Abstract

Traditional photovoltaic powered greenhouses face significant challenges, including high operational costs, complex maintenance due to extensive sensor deployment, and requirement substantial energy storage capacity. These constraints hinder broader adoption, particularly in remote or resource limited environments. This study addresses these challenges by proposing an Internet of Things (IoT) based energy management framework for PV powered greenhouses utilizing a Temporal Attention Long Short Term Memory (TA-LSTM) model. The approach replaces traditional physical sensors with a data driven forecasting mechanism, wherein the TA-LSTM model is employed to capture the temporal dependencies and enhance the interpretability of environmental predictions. Using one year of greenhouse data, the model demonstrated high prediction accuracy with a MAPE of 1.33% for temperature and 1.24% for relative humidity. A priority based demand side management (DSM) algorithm is integrated to facilitate real time load shifting, that reduces energy storage and generation requirements. The complete system is implemented on a Raspberry Pi, to validate the effective adaptation to dynamic environmental conditions. This study provides a pathway for cost effective, AI driven energy management in smart greenhouse operations.