Forecasting of Electricity Prices Using Deep Learning Networks

Estimating future electricity prices is a critical economic success factor for participants in the energy market. In this paper, a modular day-ahead electricity price forecasting approach is proposed. Feeding on historical price data, grid weather data and load forecasts, the model delivers precise estimations of hourly day-ahead electricity prices, for a chosen case study of the German bidding zone. A time series model, based on a combination of convolutional and recurrent neural networks, provides a predictive foundation. This basic estimation is enriched with predictions of wind and solar power generation in Germany, derived from a convolutional neural network feeding on multi-dimensional grid weather data. Together with a load estimation, provided by the German Transmission System Operators, all three input streams are aggregated in a gradient boosting regressor to produce a final estimation for the electricity prices of the upcoming day.