| dc.description.abstract | The Russian invasion of Ukraine has significantly impacted energy resources across Europe, leading to a widespread increase in electricity prices. In an effort to mitigate this, European nations have significantly expanded their renewable energy projects, aiming to lessen their reliance on Russian energy and to confront the global climate crisis. As a result of these initiatives, electricity price volatility has dramatically increased across the continent. Due to the interconnected nature of the European electricity market, the surge in prices and increased volatility also affect Norway, which in recent years has become even more coupled with the European market due to new subsea HV-DC connectors.
The new market dynamics and elevated prices have made forecasting electricity prices increasingly challenging. This thesis evaluates the effectiveness of machine learning models, particularly focusing on Long Short-Term Memory (LSTM), for forecasting day-ahead prices in Norway's NO2 bidding zone under these difficult conditions. Alongside the LSTM, a Deep Neural Network (DNN) model is developed, and both are compared with an established benchmark. Following an extensive hyperparameter optimisation process, including cross-validation and multi-objective Sequential Model-Based Optimisation, both models achieved similar performance, with the LSTM model attaining a mean absolute error of 15.48 EUR/MWh and the DNN model reaching 15.62 EUR/MWh. The performance of our models aligns with similar studies, though direct comparison is limited due to our models' strict alignment with the day-ahead market timings, an aspect often overlooked.
% in other research.
Furthermore, the thesis employs Explainable Artificial Intelligence techniques, specifically SHapley Additive exPlanations (SHAP), to analyse electricity price predictions. It examines two separate Extreme Gradient Boosting models, each using data from different periods, one before and one after the onset of the energy crisis. This analysis highlights the influence of key market-specific variables, such as residual loads, hydro reservoir levels, and gas prices, on electricity price forecasts. By integrating SHAP values, this study quantifies the importance of these features, enhancing the transparency and understanding of the driving factors behind electricity prices in Norway. The dual-dataset approach further allows for an examination of market differences pre- and post-energy crisis, providing insights into the changing market dynamics.
Our results indicate oil and gas prices as the primary drivers of electricity prices in Norway, with the influence of gas prices becoming even more pronounced following the onset of the energy crisis. As anticipated, there is also an increased dependency on German residual load after the crisis, which coincides with the introduction of the NordLink subsea HV-DC cable between Norway and Germany. Additionally, before the crisis, the results reveal an odd non-linear dependency between electricity prices and the net position. Prices decrease when exports are below 2000 MW and sharply increase above this threshold. Post-crisis, this behaviour becomes more consistent with a pure economic view where imports lower prices, while exports raise them. | |
