Probabilistic Quantile Multi-step Forecasting of Energy Market Prices: A UK Case Study.

The transition from traditional dispatchable generation units to intermittent supply from renewable energy sources, as well as the continuous rise in energy demand, partially due to the growing popularity of electric vehicles (EVs), has sparked an upsurge in research interest for energy related forecasting in recent decades. The heavy reliance on weather conditions adds unpredictability in energy generation, resulting in fluctuations in the electricity system and, as a result, in electricity prices. Therefore, in order to support more efficient energy management, high-quality forecasts are required not just for energy demand and generation, but also for energy market prices. While most approaches aim to achieve point forecasts for the energy market prices, a probabilistic forecast approach could further assist the decision making process. This paper proposes a lightweight forecasting model for accurate multi-step forecasts of day-ahead and intra-day prices of the UK electricity market, while providing different quantiles of the forecast in order to estimate the potential uncertainty of price forecasts. The methodology focuses heavily on the feature engineering step by utilizing features extracted from numerical weather values, load and generation forecasts of the respective region, temporal features and historical values of day-ahead and intra-day prices. Furthermore, new metrics for evaluating the forecasted quantile intervals are introduced and defined in the analysis, in addition to the commonly used evaluation metrics implemented in time series forecasting.