Replacing Classical Algorithms to Determine the Reliability of Power Electronic Converters: An AI Method Based on the Nonlinear Autoregressive with Exogeneous Inputs Artificial Neural Network

Power electronic converters (PECs) are one of the most crucial components in today's modern power systems. Therefore, the reliability of these converters is pivotal for maintaining the continuous and uninterrupted operation of electrical systems. Classical algorithms for determining the reliability of PECs, such as the Monte-Carlo algorithm, the Coffin-Manson model, and conventional probabilistic approaches, are parametric in nature and require intensive manual input. These algorithms do not account for PECs' dynamic mission profiles. In order to address these drawbacks, this paper proposes an innovative nonlinear autoregressive with exogeneous inputs artificial neural network (NARX-ANN) algorithm solely for enhancing the analysis of PEC reliability for power system applications. The proposed algorithm predicts the dynamic nature of PECs and provides a more flexible, cost-effective, and adaptive solution capable of real-time predictions. In order to account for uncertainty in dynamic mission profiles, the NARX-ANN is designed to output a range of estimated number of cycles to failure ranging from conservative to optimistic. The neural network training is supervised and apportioned into a training and testing set. The neural network results verify NARX-ANN's viability as an alternative to classical methods in assessing PEC reliability.