Intelligent power quality monitoring on distributed generation systems

Abstract

Increased consumer demand for electrical energy has forced the incorporation of distributed generation sources into traditional electrical power system schemes, resulting in changes in the way these distributed generation systems (DGSs) are studied. The constant monitoring of power quality (PQ) in these systems is an important aspect of the study of DGSs because the electronic devices inherent in these schemes produce various phenomena that aect them, causing serious problems for the consumers. On the other hand, the development of new data acquisition technologies has given rise to an increase in the amount of data extracted from the DGS that requires advanced techniques for its analysis, where machine learning (ML) has recently been playing a crucial role, making it a current research issue. This thesis focuses on PQ monitoring in distributed generation electrical power systems, with an emphasis on disturbance classication using an intelligent approach based on ML-based methods. Here, some typical problems with traditional machine learning techniques, such as the manual feature selection for disturbance generalization, are addressed to enhance disturbance classication. Moreover, a dataset generation protocol is proposed to standardize the assessment of ML-based classication methods. For this aim, three dierent dataset types, i.e., synthetic, simulated, and real-world measured, are used to train and assess the classication model. Finally, a comprehensive study about unbalanced data in uences on model training was conducted in this thesis to contribute to the classication model training data paradigm. Results obtained in this thesis open new opportunities for the development of future research work on PQ monitoring.