Chapter Ten - Machine learning application to power system forecasting

Abstract

In recent days, analysis of renewable-rich power systems has shown greater interest as the integration of renewable generations is encouraged nationwide both at transmission and distribution levels. The increasing integration of such sources poses many technical challenges that must be considered when designing, planning, and operating modern power grids. Contrary to conventional generators where future generations are controllable, and forecasting is direct in potential schedules, nature plays a crucial role in deciding the intermittent and uncertain renewable power generations. The inability to store renewable energy in a technologically and economically efficient way demands favorable renewable generation forecasts. The present research in power system forecasting is mainly devoted to devising accurate forecasting models for renewable generations. Appropriate forecasting of renewable energy generations in an optimal manner can tremendously improve the decision-making process to control the potential risks caused due to unforeseen events. The extent of uncertainty can be modeled by using suitable forecasting techniques. The more realistic way to characterize these uncertainties is via probabilistic forecasting techniques to help make efficient renewable energy generation forecasts. Besides, the accuracy of the developed models strictly depends on the purity of the historical data at hand. This chapter mainly focuses on the importance of data preprocessing and probabilistic forecasting of uncertain power system inputs.