With an increasing interest in constructing new run-of-the-river (ROR) hydroelectric generation over the more traditional reservoir-based hydroelectric systems, there is an increasing
operational challenge due to the volatility of streamflow. The Snohomish County Public
Utilities District (SnoPUD) has recently invested in the construction and operation of 3 new
run-of-the-river projects in Northwestern Washington along Calligan Creek, Hancock Creek,
and Youngs Creek. In order to effectively plan generation dispatch, SnoPUD has expressed
interest in the development of an accurate forecasting tool to predict the generation capacity
for these ROR systems.
The following research project aims to use statistical learning models, namely Hidden
Markov Models (HMMs), to predict day-ahead generation capacities for the aforementioned
ROR systems. These models are constructed using 12 years of historical streamflow data
collected at the intake sites and precipitation data recorded at the National Oceanic and Atmospheric Administration (NOAA) Alpine Meadows station. Four methods of constructing the models are studied for their forecast accuracies, and are compared with the persistence model. Despite using only one set of observable variables, the HMMs are shown to have
slight improvements in accuracy over the persistence model approach, which shows great
optimism for future work.
