Adaptive AI-Driven Control System for Real-Time Regulation of Kaplan Turbines in Small and Medium Hydropower Plants

Abstract

This paper proposes an adaptive, AI-driven control architecture for real-time regulation of Kaplan turbines in small and medium hydropower plants, targeting simultaneous power tracking, reservoir level stabilization, and cavitation-aware operation. A cascade control structure is adopted, where an outer level loop enforces admissible hydraulic constraints by shaping the discharge limit, while an inner power loop regulates guide vane opening to track the power reference under head-dependent flow bounds. The runner blade angle is scheduled by a baseline combinator (cam curve) and refined online via a lightweight learning layer based on recursive least squares, which provides corrective blade commands to improve efficiency while respecting cavitation margins. Cavitation risk is modelled using a Thoma-based criterion, augmented with a speed-dependent constraint surface to reflect the influence of rotor speed on required cavitation allowance. A Python-based nonlinear simulation framework is developed, incorporating tailwater rating effects, head losses, spill behaviour, and grid-connected speed droop dynamics. Comparative results against the conventional cam-curve strategy demonstrate that the proposed adaptive approach preserves power-tracking performance and level safety while increasing average efficiency and reducing cavitation constraint violations over representative inflow and dispatch scenarios. The study is presented as a simulation-based proof of concept. The obtained results are used to evaluate the potential of the proposed controller under representative hydraulic and dispatch scenarios, while validation with real plant measurements is identified as a necessary next step.