Sensitivity Analysis of Stationarity Tests' Outcome to Time Series Facets and Test Parameters

Abstract

Time series stationarity is vital for the effective implementation of forecasting models. Time series of renewable generation rich power system input variables such as photovoltaic generations, wind power generations, load power, and ambient temperature have inherent time series facets such as trend, seasonality, and volatility. These inherent facets, combined with the length of the time series or time series clustering, have a propensity to bias the stationarity tests' outcome. This research conducts a rigorous comparative analysis to assess the tests' sensitivity to different time series facets. A seasonal, nonstationary load power time series and its derived time series with synthetically embedded trend and volatility effects are used to help the study capture tests' sensitivity to the above time series facets. This comprehensive analysis and discussion, via a set of well-delineated figures and tables, are expected to assist novice researchers in choosing a group of suitable tests for checking time series stationarity.