Modelling and Predicting Photovoltaic Power Generation in the EEX Market

Abstract

The importance of solar energy has been growing in recent years. This raises the need for efficient modelling and forecasting methods. The existing methods are predominantly based on weather predictions or forecast solar radiation, which is not easy to convert into production forecast. Instead we propose to directly model the photovoltaic power production in the EEX market in Germany by time series methods. To this end we test an autoregressive moving average (ARMA) model combined with three types of generalised autoregressive conditional heteroscedastic (GARCH) models for the univariate case of solar production aggregated over the whole country, and an vector autoregressive (VAR) model for the multivariate case of individual regions divided among four transmission system operators (TSOs). We compare the output from the models with forecasts provided by the producers. The study reveals that our models work very well compared to rather complex models used by the TSOs. In addition, our stochastic models provide valuable insight into the market and can be used as a building block for risk management purposes in energy markets.