Hybrid deep learning models for water demand forecasting in greenhouses: Exploring the energy Nexus in Urban agriculture

Abstract

Precise water demand forecasting (WDF) is crucial for sustainable irrigation and resource efficiency in urban greenhouse systems. This study introduces a cutting-edge hybrid deep learning approach designed for short-term WDF, while also considering the energy nexus between water, energy, and environmental factors. The model integrates the least squares generative adversarial network (LSGAN) for data pre-processing and noise reduction, convolutional neural networks (CNN) for feature selection, and bidirectional long short-term memory (BiLSTM) for time-series state modeling, and named as LSGANCBiLSTM. Using real-world data from the Wageningen Research Centre in Bleiswijk, Netherlands, the model significantly outperformed benchmark approaches, achieving an R-value of 99.57 % with minimal forecasting errors. The model demonstrated exceptional stability, minimal bias, and strong handling of environmental variability, improving short-term WDF accuracy, optimizing water management in urban agriculture, enhancing sustainable irrigation, and addressing the energy nexus for efficient resource use.