Turbine-level clustering for improved short-term wind power forecasting

Abstract

At the present time, new types of data are collected at a turbine level, and can be used to enhance the skill of short-term wind power forecasts. In particular, high resolution measurements such as wind power and wind speed are gathered using SCADA systems. These data can be used to build turbine-tailored forecasting models, but at a higher computational cost to predict the production of the overall wind farm compared to a single farm-level model. Thus, we explore the potential of the DBSCAN clustering algorithm to group wind turbines and build forecasting models at a cluster-level to find a middle ground between forecasting accuracy and computational cost. The proposed approach is evaluated using SCADA data collected in two Irish wind farms.