Chapter 9 - Intelligent Data Analytics for Time-Series Load Forecasting Using Fuzzy Reinforcement Learning (FRL)

Abstract

In recent years, load forecasting is becoming more-and-more important due to its numerous applications in the modern power system as well as in the virtual power plant (VPP) by virtue of the real-time analysis of ESS (Energy Storage System), DERs (Distributed Energy Resources), DSM (demand side management), and EVs (Electric vehicles). To overcome the real-time created/generated challenges and ensure accurate, reliable, and stable power generation for continuous time horizons in different time steps (i.e., yearly, monthly, weekly, daily, and hourly, etc.), an advanced intelligent model is proposed by using FQL (fuzzy-Q-learning) based FRL (fuzzy reinforcement learning) approach. In this chapter, a short-term load predictor, which is able to forecast the load for next 24 h, is presented. The proposed approach is tested by using real-time recorded historical data collected from GEFCom2012 and GEFCom2014 and simulated results show accurate and highly satisfactory performance. In this chapter four case studies have been performed for 1 month-ahead, week-ahead, day-ahead, and hour-ahead load forecasting. The proposed approach can predict the load for 1 month, 1 week, 1 day, as well as 1 h in advance, which shows high prediction accuracy with acceptable range of MAPE (mean absolute percentage error) for all four case studies.