Presentations

Abstract

In this paper, we introduce a framework for estimating wake losses using SCADA (Supervisory Control and Data Acquisition) data. The proposed approach leverages the freestream turbines as sensors to reconstruct ambient wind conditions and employs machine learning techniques to compute the wake-free turbine production. Wake losses are defined as the difference between the power a turbine could generate in a wake-free condition, known as "freestream power", and its actual power production, which is directly read from the SCADA system. Estimating freestream power, which is a purely theoretical concept, is a crucial step in this assessment. In this paper, we introduce a framework that harnesses machine learning to estimate freestream power, as a function of the windfarm ambient wind conditions, and evaluate wake losses in operational wind farms. The developed framework takes as input raw 10-minute SCADA timeseries data, which are first cleaned to remove any erroneous data. Estimating the ambient wind conditions of the wind farm requires to flag the freestream turbine at each timestamp, using the wind direction inferred from the nacelle position channel of each operational turbine. Timeseries of ambient wind speed and turbulence intensity, are then constructed from the freestream turbines SCADA data. Subsequently, a predictive model is trained for each turbine to estimate the power output based on the reconstructed ambient conditions but considering only freestream conditions for the training. These models are then utilized to predict the freestream power of each turbine across the entire SCADA time series, enabling the estimation of wake losses.