Digital Twin Based on Neural Network for a Grid Connected Modular Multilevel Converters for HVDC Transmission

Modular Multilevel Converters (MMCs) are the preferred voltage source converters for HVDC transmission in wind energy systems, and their effective lifecycle management is crucial for grid stability and reliability. Digital twin technology can be used to describe and model characteristics, behavior and real-world performance of complex systems by constructing real-time mapping of them which can be used for real-time and health monitoring. This paper presents a feed-forward neural network (FF-NN) based digital twin (DT) for health monitoring of a grid-connected, MMC based inverter. Given the complex, non-linear dynamics of MMCs, a data-driven approach using an FF-NN is well-suited for capturing the full system behavior. This work details the training data generation methodology, encompassing the diverse scenarios required for real-time modeling and health monitoring. By analyzing the difference between the DT's predicted output and the physical twin's actual output, this work demonstrates the model's accuracy in replicating system behavior under steady-state operation and during transient events i.e. the AC and DC faults. The trained model can be used for health monitoring and can indicate parameter degradation