Predicting Day-Ahead Electricity Prices in Western Denmark (Dk1) Using Bayesian Deep Learning: Robust Intervals and The Role of Renewable Energy

In the competitive electricity market, accurate prediction of electricity price is crucial for market participants to develop optimal bidding and offering strategies. However, due to the influence of various factors, electricity price prediction remains a challenging task. In addition, the increasing integration of renewable energy sources over the past two decades has significantly amplified price variability. Hence, to address this challenge, this study proposes a Bayesian-optimized hybrid forecasting model combining Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN). Indeed, the LSTM model is employed to capture long-term dependencies in electricity price patterns, while the CNN model is used to extract significant features and detect sudden price fluctuations. To enhance model accuracy, Bayesian Optimization (BO) is utilized for hyperparameter tuning, improving the model’s efficiency and performance. Additionally, the Bayesian technique is applied to quantify the uncertainty of predictions and determine robust prediction intervals with confidence levels of 90%, 80%, and 65%. A key contribution of this study is the customized forecasting model designed for electricity price prediction in the western region of Denmark (DK1). In this model, amounts of total load, solar power, onshore and offshore wind power are used as independent variables. Moreover, the CNN layer in the multivariate model is designed with a dilated structure, which enhances the model’s ability to capture dependencies between variables and their long-term effects in the time dimension. The model is evaluated using various error metrics and the Diebold-Mariano (DM) test. Furthermore, Bayesian uncertainty analysis is used to determine the uncertainty in the results obtained with the forecasting models in different intervals. The results indicate that the Bayesian-optimized hybrid model outperforms traditional forecasting models and provides more reliable and accurate electricity price predictions. Additionally, the study highlights the significant impact of renewable energy generation on forecasting accuracy and reliability, demonstrating that incorporating renewable energy data into the model improves price prediction performance.