Combining RNN and Linear Structures in Day-Ahead Electricity Price Forecasting

Souhir Ben Amor, University of Duisburg-Essen

Co-authors: Florian Ziel, University of Duisburg-Essen

Abstract: his paper presents a combined forecasting model that integrates linear expert models with recurrent neural network to accurately predict day-ahead

electricity prices. In our methodology, we consider hourly data from the German electricity market between 2019 and 2024. Our approach takes advantage of the linear expert model’s capability to incorporate external market fundamentals efficiently and RNNs’ ability to capture non-linear patterns and temporal dynamics.

The combined model is developed using PyTorch and optimised using Optuna. Results show that the proposed model outperforms standalone forecasting models