A Comparative Study for Multilayer Perception versus Convolutional Neural Network Based On the Wind Turbine Benchmark Model

The condition monitoring and fault detection are vital for the wind turbine to reduce the maintenance costs and improve operation level. In this paper two data-driven models that include a convolutional neural network (CNN) and a multilayer perceptual network (MLP) are built to detect the faults from the observed variables. Both models are trained based on the data from the wind turbine benchmark model and made a comparison in the recognition accuracy. The study shows the training time of CNN is longer than that of MLP based on the back propagation approach and the recognition accuracy of CNN is about 10% better than that of MLP. The simulation also shows both models have a good robustness on the transition process after a fault occurs.