Applying Neural Networks for Day-Ahead Electricity Price Forecasting in the Nord Pool Market

Abstract

In the current volatile energy market, electricity price forecasting presents significant challenges, mainly because of the unique and unpredictable nature of electricity prices after the European energy crisis in 2022. While there has been considerable research on electricity price forecasting, this specific period after 2022 has not been extensively studied. This thesis addresses this by implementing and evaluating various forecasting models to predict the day-ahead system price in the Nord Pool market, using artificial neural networks and traditional benchmark models. Two neural networks have been implemented: the Multilayer Perceptron (MLP) and the Long Short-Term Memory (LSTM). The benchmarks include the naïve forecast, and the statistical time series models ARIMA and SARIMA. Results show that the neural networks significantly outperform the benchmark models across all error metrics for the entire evaluation period. The MLP emerged as the best-performing model among the neural networks, surpassing the LSTM in every error metric evaluated. Additionally, the MLP shows superior capability in handling price spikes, making it effective for the current market conditions.