Ice prediction for wind turbine rotor blades with time series data and a deep learning approach

Abstract

Reducing ice formation on wind turbine rotor blades is mandatory to increase the economic viability of wind energy in cold regions. In our research, we propose a predictive anti-icing system as the solution to this problem. This system consists of a machine learning model which uses a combination of data from numerical weather prediction (NWP) and from supervisory control and data acquisition systems (SCADA). In the first part of this contribution, we briefly summarize our past research. In the second part of this contribution, we evaluate our best-performing model, a one-dimensional convolutional neural network with two inputs, using new data from winter 2020/2021, discuss the model's shortcomings, and implement changes to overcome these shortcomings. The results suggest that weather conditions across different winters differ significantly, and that therefore data from several winters would be necessary to train a generalized ice prediction model.