Deep learning for sensing in wastewater treatment plants

This dissertation primarily focuses on employing deep learning with sparse data in wastewater treatment processes. Firstly, it proposes a novel deep soft-sensing model and develops a robust data sampling algorithm to train with sparse data. Secondly, it introduces two data synthesis techniques that enhance transfer learning performance for downstream applications. Finally, it develops a framework to learn a feature-centric latent representation that clusters high-dimensional data for state estimation of the processes. These contributions address the challenges of applying deep learning to the complex but sparse data, allowing resource optimization to advance decision support in complex, non-linear wastewater treatment systems.