Perspectives on explainable reinforcement learning (XRL) for hydraulic systems

Abstract

The complexity of simulating hydraulic systems, combined with the human and economic implications of forecasting flooding, controlling water quality, and anticipating power requirements, underscores the need to introduce novel techniques in the hydraulics field. In this article, the perspectives of introducing explainable deep learning in the hydraulics field are explored. Several problems of interest are presented, and specific implementations of explainable deep learning are discussed to improve the control of complex hydraulic systems under uncertain conditions. The explainable deep learning approach is introduced as a promising tool for reducing the impacts of river flooding, adjusting hydropower generation to grid demand by anticipating fluctuations in renewable energy production caused by meteorological variability, monitoring and controlling water quality using surface unmanned vessels, reducing friction losses in pipelines, and optimizing water distribution systems.