Power-LSTM for smart greenhouse: a novel deep learning approach to temperature prediction in a Mexican case study

Abstract

This paper addresses the challenge of predicting internal temperature in greenhouse environments, a critical aspect of optimizing crop growth and ensuring resource efficiency. While machine learning (ML) techniques have been widely applied to predict greenhouse climates, deep learning (DL) methods offer the potential to capture more complex relationships within the data. In this study, we present a comprehensive evaluation of ML and DL models, along with our proposed power-long short-term memory (PLSTM) model, to predict the internal temperature of a greenhouse using a database from Mexico. We compared traditional ML models such as linear regression (LR) and extreme gradient boosting (XGBoost) with DL architectures like gated recurrent unit (GRU), artificial neural networks (ANN), hybrid LSTM-ANN and LSTM-RNN architectures. Our proposed PLSTM model outperformed both ML and DL models, achieving the R² score of 0.9710, and root mean square error (RMSE) equal to 0.1710, highlighting its superior ability to predict complex time-series data.