Revealing the Diversity of Hydropeaking Patterns by Time-Series Data Mining

Abstract:

Operating hydro-electricity generation according to the hourly-adjusted electricity market has been widely applied due to its economic efficiency while also known for its significant impacts to the downstream for producing the sub-daily flow fluctuations called “hydropeaking”. To ascertain the downstream impacts of hydropeaking, features of hydropeaking have been analyzed with biologically-related hydrologic variables. However, studies on hydropeaking are constrained by the manual feature extraction and thus are limited to small temporal and spatial scales. Besides, hydropeaking has been treated as a broadly similar pattern regardless of the complex essence of the electricity market and natural settings. Therefore, this study sought to determine whether there exist significantly different hydropeaking patterns on a regional scale, as revealed by the temporal variation in hydropeaking with a long temporal scale (>five years). To fulfill this goal, a new algorithm, Hydropeaking Event Detection Algorithm (HEDA), was developed in R to automate the parameterization of hydropeaking from public flow records. Using HEDA outputs, classification was conducted to explore the differentiation and similarity of hydropeaking. Four distinct hydropeaking patterns were identified among 33 hydropeaking sites in California. Frequency and duration are two major variables that distinguish the four hydropeaking patterns. Timing of hydropeaking is not exclusively day or night but also includes the uniform and mixed. As for seasonality, hydropeaking mainly occurs in the dry season but tends to occur more frequently in other seasons if hydropeaking is intensive in the dry season.