Hybrid Data-Enabled Predictive Control for Optimal Operation of Urban Drainage Systems

<p dir="ltr">Data‐Enabled Predictive Control (DeePC) directly computes control policies from measured input–output data, unifying system identification and predictive control in a single convex optimization. In this study, we adapt DeePC to urban drainage real‐time control, where stormwater inflows are highly non‐stationary and non‐repetitive. We first expand the data library via a copula‐based rain synthesizer to cover diverse storm profiles. We then deploy a two‐regime DeePC scheme, distinguishing dry and wet periods to ensure appropriate control logic under rapidly changing conditions. Custom regularization further biases solutions toward robust, parsimonious responses. This approach is applied to a small drainage system to prevent flooding and maintain outfall discharge below a design limit. The results indicate that, when combined with targeted data augmentation and a regime‐aware framework, DeePC can achieve satisfactory performance in controlling complex, non‐stationary systems. Nevertheless, its application in such time‐varying environments requires further investigation.</p><p dir="ltr">This paper was presented at the 21st Computing and Control in the Water Industry Conference (CCWI 2025) at the University of Sheffield (1st - 3rd September 2025).</p>