Deep learning based electricity price forecasting in Finnish energy market

Abstract

Accurately forecasting electricity prices is crucial for market stability, risk management, and strategic decision-making in the Finnish energy sector. This thesis develops a deep learning- based forecasting model utilizing Bidirectional Long Short-Term Memory (BLSTM) networks to predict 24-hour electricity prices in Finnish energy market. Historical market data from Nordpool serves as the primary dataset for training and evaluation. The performance of the BLSTM is measured against traditional statistical models, including Autoregressive Integrated Moving Average (ARIMA) and Generalized Autoregressive Conditional Heteroskedasticity (GARCH), and a Deep Neural Network (DNN) model. Various evaluation metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) are used to measure the model accuracy. Results indicate that the BLSTM model significantly outperforms ARIMA and GARCH, particularly in capturing nonlinear market dynamics and adapting to volatile price fluctuations. The findings highlight the potential of deep learning techniques in improving electricity price forecasting, ultimately aiding energy providers, policymakers, and traders in optimizing market strategies and mitigating financial risks.