Posters

Abstract

A novel fault detection model based on a deep probabilistic multi-layer perceptron is presented. The model assumes that the wind turbine output in any 10-minute interval comes from a Gaussian distribution with an unknown mean and variance. The deep learning capability is utilized to capture complex patterns within many input variables obtained from SCADA data. To provide a comprehensive evaluation, comparative analyses are conducted with various existing probabilistic methods, such as Gaussian processes and Bayesian neural networks. The results showcase the superior efficacy of our proposed method over others, illustrating the precise posterior predictive coverage probabilities that guarantee an excellent fail-safe fault detection system.