Multistep Electricity Price Forecasting for Deregulated Energy Markets: GAN-Based Reinforcement Learning

Abstract

Electricity Price Forecasting (EPF) plays a vital role in smart grid applications for deregulated electricity markets. Most of the studies tend to investigate the electricity market influencers using forecasting techniques, often losing sight of significance on the sensibility of EPF models to the unstable real-time environment. This project will address a novel EPF based on deep reinforcement learning. The proposed approach uses generative adversarial networks (GAN) to collect synthetic data and increase training set effectively and increase the adaptation of the forecasting system to the environment. The data collected will be fed to a Deep Q learning to generate the final predictions. The proposed GAN-DQL will also be assessed on real data to prove the proposed model advantages compared to several machine learning solutions.