Back-propagation-based multivariate state estimation technique: A lightweight adaptive condition monitoring approach for wind turbine

Abstract

Aimed at the challenge of wind turbine condition monitoring with insufficient available data caused by the harsh conditions, this paper presents a lightweight adaptive condition monitoring approach based on the back-propagation algorithm and the multivariate state estimation technique. In order to maximize the utility of available data, the multivariate state estimation technique is used to extract the features from multiple historical data. Moreover, buffer space is adopted to update dynamically while avoiding the effect of outliers. Then, the operator is modified so that the model can be trained and updated by the back-propagation algorithm, thereby achieving lightweight. The model performs well in condition monitoring with small samples, even discontinuous samples, that are verified through experiments based on field wind turbine datasets. The experimental results and comparative analyses demonstrate the effectiveness, generality, and advantage over others.