Abstract:
The potential of advanced neural networks (NNs) in HVDC transmission remains largely unexplored. Implementing these techniques on real-time digital simulators (RTDS) is challenging due to rapid computation requirements and risks of overfitting from large data sets generated in small time steps. To address these issues, the study explores various NN techniques using supervised and reinforced imitation learning to mimic a proposed controller with labelled data for real-time applications. An error-tracking mechanism is also incorporated to enhance the NN’s learning process. The study evaluates the most effective NN methods through offline processing for online feasibility. Both offline and online training, as well as real-time adjustments, are demonstrated to deliver a robust and easily implementable control solution. Extensive RTDS simulations and experimental investigations validate the proposed methodology for four-terminal HVDC systems.
Rahul Rane, TU Delft