Machine learning for transient test sequences in closed-loop hydraulic turbine rigs: optimization of pump operation for stable head

Paper in proceeding, 2025

This study utilizes machine learning methods to alleviate head oscillation and shorten the response time during start-up sequences of a Kaplan turbine in a closed-loop test rig. A large amount of experimental data is collected from the test rig. Artificial neural networks (ANNs) are implemented to describe the non-linear relationship between the head, and other operational parameters, such as pump speeds, guide vane opening, etc., during the transient start-up sequences. Then a proportional-integral-derivate (PID) controller is designed to optimize the pump speed operation under a fixed runner blade angle and predetermined change of guide vane opening during the start-up sequences. With the help of the ANN prediction model and the PID controller, a proper pump speed operation is recommended to alleviate head fluctuations. The numerical results are validated and compared against the experimental data in terms of accuracy and usability. The pros and cons of the proposed method are also discussed.