Neural network modelling with autoregressive inputs for wind turbine condition monitoring

Artificial neural networks enjoy popularity among different areas of modelling, including financial decision making, medical diagnosis, visualisation, and process control. This paper presents potential problems with the inclusion of autoregressive terms in a neural network model with specific reference to an application to wind turbine condition monitoring. The model´s ability to detect anomalies is explored by employing 10-minute supervisory control and data acquisition (SCADA) data from a commercial wind turbine gearbox. The issues associated with the inclusion of autoregressive inputs are assessed through an investigation of the weighting parameters for each neuron in the hidden and output layers and the outputs from these neurons.