Review of preprocessing methods for univariate volatile time-series in power system applications

Outlier detection and correction of time-series referred to as preprocessing, play a vital role in forecasting in power systems. Rigorous research on this topic has been made in the past few decades and is still ongoing. In this paper, a detailed survey of different preprocessing methods is made, and the existing preprocessing methods are categorized. Also, the preprocessing capability of each method is highlighted. The well-established methods of each category applicable to univariate data are critically analyzed and compared based on their preprocessing ability. The result analysis includes applying the well-established methods to volatile time-series frequently used in power system applications. PV generation, load power, and ambient temperature time-series (clean and raw) of different time-step collected from various places/weather zones are considered for index-based and graphical-based comparison among the well-established methods. The impact of change in the crucial parameter(s) values and time-resolution of the data on the methods' performance is also elucidated in this paper. The pros and cons of methods are discussed along with the scope for improvisation.

Automated summary

This paper examines and categorizes preprocessing methods for time-series data, evaluating and comparing the capabilities of each method. The application of these methods to commonly used time-series data in power systems is discussed, along with the impact on method performance and potential for improvement.