Stationarity Analysis and Supervised Machine Learning Techniques for Energy Management Forecasting

Abstract

One of the most realistic solutions to the problem of power outages in remote areas is hybrid energy sources (HES). Forecasting is necessary because HES uses renewable energy sources, which are often either intermittent or insufficient. By directly influencing planning and management techniques, forecasting plays a crucial part in energy systems. Due to electronic devices shutting down as a result of generating unwanted harmonics that degrade the system's quality, inaccurate forecasting can lead businesses to lose money. Choosing a trustworthy and efficient forecasting model is crucial. In this thesis, regression and classification are the two supervised machine learning techniques used for predicting. There are two sections to the thesis. The regression using the ARIMA model is covered in the first section. Three studies on ARIMA are presented in order to examine the stationarity of a time series. The third study suggests a novel technique that involves moving non-stationary data to a domain that deals with it as a stationary time series. The first two studies provide a new technique to test stationarity. The thesis' second section, on classification, focuses on a fresh approach to energy management. To create a new dataset that may be used to anticipate energy management (EM), the new technique advises gathering the outcomes of energy management from many decision-making procedures throughout a range of time periods.