Easy Transfer Learning-Based Model-Data-Hybrid-Driven Fault Detection for Battery Inverters
Yu Zeng, Ezequiel Rodríguez, Qingxiang Liu, Gaowen Liang, Huamin Jie, Josep Pou, Hebin Ruan, Janardhana Kotturu
Abstract
In this letter, a hybrid method of fault detection using data and models, based on easy knowledge transfer learning, is proposed. The proposed method is applied for multiple battery converters, where new systems that are integrated into a microgrid are trained using the knowledge acquired by the existing systems during the offline phase. The new Target classifier can detect both open-circuit faults and current sensor faults with a 60% dataset reduction. The effectiveness of the method has been experimentally corroborated in a microgrid with two three-phase two-level converters, one Source, and one Target. Different values in terms of voltage, capacity, and power rating of the batteries, are tested using hardware in the loop. The detection accuracy is 99.1%.