Sukanta Roy1, Arif Sarwat1, Milad Behnamfar1, Anjan Debnath1, Mohd Tariq1, and Patrick McCluskey2
1Dept. of ECE, Florida International University, Miami, FL, USA
2Dept. of ME, University of Maryland, College Park, MD, USA
For more information about this article and related research, please contact Prof. Patrick McCluskey.
Abstract:
A significant concern in grid-scale photovoltaic power plants is the abnormal fluctuation of inverter electrical performance over its operational lifetime. Environmental conditions contribute to the degradation of various inverter components to different degrees, necessitating the construction of a digital twin for analysis and prediction. Initially, an H-bridge inverter with an output LC filter is constructed, and its experimental dataset is used to build a circuit-level ‘switching’ digital model of the inverter using particle swarm optimization. The tuned model is then degraded to generate a large dataset, including filter component parasitic resistance—a key degradation parameter for grid-tied inverters. Subsequently, supervised machine learning (ML) models are trained and tested to implement a digital twin of the inverter capable of accurately estimating degradation-induced reliability issues. Two ML algorithm-driven results are compared, with the random forest model emerging as the best-fit digital twin for the constructed inverter, achieving an R2 value of 0.99 and an RMSE as low as 1.155 × 10-6.
This article is available online here and to CALCE Consortium Members for personal review.
.