A1170
Title: Neural network water inflow modelling: Predicting Colombian hydropower generation capacities
Authors: Johannes Schwenzer - Europa-Universität Viadrina Frankfurt(O) (Germany) [presenting]
Abstract: The Colombian energy system is heavily dependent on hydroelectric power. Hydropower plants generate up to 70\% of the electricity. The strong dependency on a single, weather-dependent source of energy generation introduces a certain vulnerability to the country's energy security. The increasing rate of climate change may increase those vulnerabilities drastically. Droughts brought by the warm phase of the ENSO phenomenon have led to significant strains on the Colombian energy system with outages and big price spikes in electricity costs. It highlights two important fields of research: pathways to a more diversified energy system and analyses of future climate change impact on the hydropower generation capacity. The aim is to contribute to both fields by applying state-of-the-art machine learning techniques to model the non-linear impact of temperature and precipitation variables on the water in-flows of selected hydro reservoirs. Additionally, explanatory algorithms are applied to quantify the impact of each input feature. It permits long-term forecasts for each future weather prediction under the respective representative concentration pathways (RCP) adopted by the IPCC. These results are vital to identify the optimal composition and magnitude for the expansion of the Colombian energy system to increase the security of supply, reduce dependency on weather phenomena and limit electricity-related CO2 emissions.
