Nord pool Day-ahead Electiricty Price Forecasting using machine learning.

Abstract

Electricity stands apart from other commodities due to its unique characteristics, forecasting requires a deep understanding of precise equilibrium between supply and demand. Since the enactment of the Energy Act in 1991, Norway's electricity market has undergone profound transformations, which eventually results in a regulated and interconnected marketplace. One of the notable impacts is the evolution of market bid and offers based on expected time-ahead supply and demand. In this scenario, suppliers strategically craft bidding plans, leveraging electricity price predictions in order to optimize profits. Note that, with the advent of renewable energy sources (such as solar, wind, hydroelectric), suppliers are witnessing significant uncertainties and heightened volatility in this space. Therefore, the primary objective of this thesis is to develop state-of-the-art electricity price forecasting models that boast robustness and accuracy, delivering estimates closely aligned with real-time values. This will result in improved decision making, and eventually improve economic gains for stakeholders involved in the electricity sector. While numerous statistical studies have been proposed in the past, their estimation capabilities have generally been limited. This is largely due to the highly non-linear and uncertain nature of exogenous variables within electricity markets. Recently, Deep Learning-based approaches have gained significant attention for their ability to efficiently estimate functions, even in scenarios involving highly non-linear exogenous variables and complex data spaces. Therefore, inspired by the proven success of Deep Learning in navigating highly non-linear data spaces, this thesis attempts to investigate its applicability in energy price forecasting. It rigorously follows the established practices of Machine Learning, including data preprocessing, model training, and analysis, using the NordPool energy market dataset as a primary source of examination. A range of Deep Learning methodologies were explored, including Feedforward Neural Networks, Artificial Neural Networks (ANN), Deep Neural Networks (DNN), Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and Long Short-Term Memory (LSTM) networks. As an innovative aspect of this research, various dropout techniques and parameters were also investigated. The results indicate that the ANN model exhibited the highest performance, followed by LSTM and RNN. These findings suggest that these models can serve as robust and reliable tools for forecasting electricity price.