An Empirical Study on Missing Data Imputation Techniques for Transient Stability Assessment Using Four Sample Frequency Features Data set

Transient stability assessment (TSA) is a common routine performed during the planning and operation stages of power system. Artificial intelligence (AI) approaches such as machine learning (ML) and deep learning (DL) have been playing an important role in assessment because of the availability of large labelled data sets. However, one main concern in an AI-based approach is the selection of a set of appropriate input features for the AI model. This paper considers a simple data set for TSA. The data set is formed by sampling the generator bus frequency trajectory at four specific time stamps, assuming the presence of phasor measurement units (PMUs) at respective buses. An artificial neural network (ANN) has been trained and tested on this data set to illustrate the appropriateness of these input features for TSA. Further, the effect of any partial or complete missing of these data samples on trained AI model performance has been discussed in detail. An empirical study on the improvement of classifier accuracy by imputing these missing features using various simple imputation techniques has been carried out. IEEE 10-generator 39 bus test system is used for constructing the transient stability data base. The suitability of the selected features for TSA under (i) no missing samples and (ii) missing samples imputed by various simple approaches have been presented using numerous case studies throughout the sections of the paper.