Evaluating Deep Learning Methods for Virtual Sensors in Water Infrastructure Digital Twins

Water scarcity is an increasing global issue, with the demand for water potentially exceeding supply, and organizations failing to balance current and future needs. Continuous monitoring and accurate prediction of water availability, proactive leakage detection and optimising water consumption are crucial, necessitating advanced, cost-effective, and highly accurate methods for effective water management. The integration of the Internet of Things (IoT) with Artificial Intelligence (AI) in Industry 4.0 has significantly enhanced monitoring, automated industrial processes and enabled forecasting. High-quality data is essential to deliver precise insights and optimal performance. Virtual sensors, which digitally replicate failed IoT sensors, offer a smart solution for data generation, that impacts water infrastructure monitoring. This study employs deep learning models Generative Adversarial Networks (GAN), Long Short-Term Memory (LSTM), and Variational Autoencoders (VAE), to generate synthetic data for virtual sensors. The findings indicate that LSTM provide better outputs, paving the way for further development and implementation of virtual sensors in Water Infrastructure Digital Twins.