Training requirements of a neural network used for fatigue load estimation of offshore wind turbines

To estimate fatigue loads of wind turbines via neural networks, costly load measurements are needed for training. Thus, our aim was to assess the minimum needed size of the training sample. We focused on the prediction of flapwise blade root bending moments with a neural network of eight inputs. Next to statistical testing of the training sample size, their representativeness compared to a one-year measurement as well as seasonal effects were investigated. Our results showed that training samples of about 2016 records of 10-minute statistics are representative and enable a reliable prediction independent from seasonal effects.