Towards Enhancing River Bed Stability Assessment: A Comparative Study of LSTM, GRU, and Transformer Predictive Models

As decision support methods (including as Artificial Intelligence supported decision making) progress, new ways and frameworks are emerging to enhance our understanding of sediment transport processes (via improved monitoring), but also better modeling those phenomena. This study offers a preliminary view of how deep learning models can link with real-time data from instrumented sediment particles, to predict the risk of bed surface destabilization in channels and rivers, which can lead to infrastructure scour. 
Specifically, three deep learning models are analyzed, herein: a) Long Short-Term Memory (LSTM), b) Gated Recurrent Units (GRU), and c) Transformers. These models were compared according to their accuracy, computational efficiency, and suitability for real-time applications.This study integrates data from specifically designed sediment stability monitoring sensors [1-3], with three deep learning models to predict the possibility that sediment is transported along the bed surface of the river [4], in real time. This is important for a series of applications, such as flood risk management, assessment of hazards to hydraulic infrastructure and water resource management, helping achieve resilient and sustainable development under a changing climate change. Future studies can explore further improving the efficiency of sensor enabled novel hydroinformatics approaches.

 

References
[1] Al-Obaidi, K., Xu, Y., & Valyrakis, M. (2020). The design and calibration of instrumented particles for assessing water infrastructure hazards. Journal of Sensor and Actuator Networks, 9(3), 36.
[2] AlObaidi, K., & Valyrakis, M. (2021). Linking the explicit probability of entrainment of instrumented particles to flow hydrodynamics. Earth Surface Processes and Landforms, 46(12), 2448-2465.
[3] Al-Obaidi, K., & Valyrakis, M. (2021). A sensory instrumented particle for environmental monitoring applications: Development and calibration. IEEE Sensors Journal, 21(8), 10153-10166.
[4] Valyrakis, M., Diplas, P., & Dancey, C. L. (2011). Prediction of coarse particle movement with adaptive neuro‐fuzzy inference systems. Hydrological Processes, 25(22), 3513-3524.