Forecasting of wind turbine synthetic signals based on convolutional neural networks

Abstract

The importance and future prospects of offshore wind power generation invite great efforts and investments to make it an efficient technology. A crucial aspect is the development of efficient control strategies, which in many cases require models to identify the state of the turbine at a given time accurately. These models must be simple enough not to increase the computational complexity of the control algorithm while being able to capture the nonlinearity and coupling of wind systems. In this work we study the possibility of using neural networks to identify a wind turbine model to predict its power output. Two models, with different number of inputs, have been proposed. LSTM (Long-Short Term Memory) and RNN (Recurrent Neural Network) have been compared, with satisfactory results in terms of model accuracy on an offshore 5MW WT.