Posters

Abstract

This presentation will explore use of Neural Networks to predict operational issues of turbine components using high level SCADA data. It will present the early results of Fichtner's investigation into the failure and operational issues of a number of wind farms. Machine learning techniques have numerous applications in various fields, ranging from facial recognition technologies to being used to predict share prices. Yet these methods are only just beginning to be used in the energy industry. Fichtner is capable of using a range of machine learning techniques and one that has proved particularly valuable is the use of Artificial Neural Networks. Artificial Neural Networks replicate a biological neural network in the brain and can be used effectively for modelling non-linear problems which are affected by many variables. Some patterns in datasets are obvious such as the correlation between the amount of sunlight and the power output of solar panels. However, other data patterns might be hidden and not easily predictable such as on wind farms, the exact effect wind speed combined with ambient temperature has on power output. A Neural Network will be able to quickly identify these hidden patterns and will be able to correlate the means in which many different variables can influence an output. Fichtner has successfully used Neural Networks to make accurate predictions of degradation of Solar PV modules and failure modes of major components and is currently creating models to use predictive maintenance to increase the run time of wind turbines. However, these models can be easily adapted for a range of applications by changing the inputs and minor alterations to the model. Background and early results of the investigations to date are presented.