Using rule-based systems and autoencoders for anomaly detection in supervisory control of wind turbines.

The arising demand for renewable sources of energies has led to a major focus on wind turbine based power production systems. Due to the high maintenance costs it is required the development of monitoring strategies of the status of the wind turbines to detect failures and anomalies. The wind farm taken into consideration in this work of thesis is composed of wind turbines that are all equipped with a SCADA systems for the collection of operative data. A rule-based model was developed to spot anomalies in the SCADA data. The goal of the thesis is the development of a machine learning model for anomaly detection that could overcome the limitation of the rule-based model. The adopted strategy is based on autoencoders, which are a special type of neural network, to substitute the rule-based model or improve its results. Moreover, to enhance the usability of the proposed model, some interpretability strategies are tested