2 - Disturbance dataset development for machine-learning-based power quality monitoring in distributed generation systems: a practical guide

Abstract

The lack of an official benchmark for the generation of PQ disturbance datasets for the training and assessing of classification models inspired this chapter to develop a practical guide for the development of PQ disturbance datasets. In this regard, the assessment of classification methods commonly includes three types of PQ disturbance dataset for the model training and testing, i.e., the synthetic, the simulated, and the real-world datasets. In this chapter, a practical description of the development process for the abovementioned datasets is carried out. The synthetic dataset is generated from the computation of well-known disturbances numerical models, through their parameter variations according to the existing PQ standards. The development of simulated disturbance dataset employs the professional software PSCAD to generate more complex disturbances than in the synthetic dataset. Finally, some recommendations for the development of real-world datasets are given, such as parameters to be measured, monitoring devices and their location, and the common parameters range for the data acquisition. In addition, the public real-world disturbance datasets are reviewed, and the class imbalance issue in the real-world datasets is addressed.