Neural Networks in RSCAD: Enhancing MMC-Based HVDC Simulation With Advanced Machine Learning

Abstract

The potential of advanced neural networks (NNs) has yet to be explored in the field of HVDC transmission. Implementing such intelligent computational techniques on a real-time digital simulator (RTDS) is challenging due to the need for rapid computation and the risk of overfitting with extensive data generated at tiny time steps. To overcome these limitations, different NN techniques are studied using a supervised and reinforced imitation learning method to mimic the suggested controller with labeled data for real-time applications. Furthermore, the NN component does not necessarily just take a label, and therefore, the authors propose a more advanced approach by incorporating reinforced learning through an error-tracking mechanism into the NN, apart from its loss function. The initial offline processing identifies the best-suited NN technique for online computational feasibility. Both online and offline training methods as well as online adjustments are showcased to provide a robust control solution that is easy to implement. This work deals with developing an intuitive and versatile Toolbox installed on a real-time simulator platform that can integrate complex NN-based control strategies. Extensive simulations on the RTDS platform and experimental investigations of the four terminal HVDC systems validate the interest and viability of the proposed design methodology.