Abstract:

Large-scale reservoirs provide operational flexibility to water managers by storing water during times with higher surface water availability and releasing water when it is most needed. Most large-scale reservoirs serve for multipurpose demands, such as water supply for agricultural, urban and environmental users, hydropower, recreation, fisheries and transportation. Due to its low operating cost, hydropower generation is often maximized in energy systems with mixed hydro and thermal sources. Hydropower generation is also used to meet peak demand by advantage of operating in short notice. This study aims to simulate reservoir operations, including release schedule and hydropower operations of the Yamula Dam and hydropower plant using machine learning. Located on the Kızılırmak River, the Yamula Dam is a large-scale multipurpose reservoir with its 3476 million cubic meters of storage capacity. Turbine release decisions are learned with Random Forests algorithm using only reservoir inflow and upstream streamflow conditions. The developed model successfully predicts reservoir releases between 2006 and 2015, with a coefficient of determination value of 0.87. Model prediction results are provided, and then hydropower load, generation and revenue are calculated and results are presented. Based on simulation results, the Yamula Dam generates about 362.3 gigawatts hour of energy per year, with an annual average revenue of 14.1 million Dollars. With the developed model, reservoir operations under different upstream hydrological conditions can also be simulated.

Keywords: