Spillage Forecast in Hydroelectric Power Plants via Machine Learning

Abstract

Brazilian hydroelectric power plants often use telemetry stations to extract information about the environment. These equipment are usually installed in several strategic spots of rivers that "feed" the reservoir, and are capable of providing important information such as precipitation, river level, and water flow. This paper presents an analysis of Machine Learning applied to the forecasting of spillage occurrences over a set amount of time in a Brazilian power plant. To achieve this goal, telemetry stations' data were utilized together with the plant's operations historical, which provides information about previous spillages, turbines' flows, among others. The Machine Learning approach has shown to be promising in this problem, and the developed model presented the potential to effectively support decisions by helping the operators prepare for significant incoming water flows.