Presentation Description: Analyzing SCADA data at operating wind projects is an important exercise for understanding their historical performance and informing future projections. Analysts are able to identify and quantify the various effects of shutdowns, sub-optimal performance, curtailments, or other physical occurrences. Categorizing the data manually can be extremely time consuming, and using simple logical filters can be prone to errors as they may miss legitimate differences between turbines and nuances in the data.
Standard supervised learning techniques have been employed to address this challenge, however the variation between turbine model, control strategies, geographic location, and more hinder the transferability of the trained models between projects. We propose two methods to deal with the transferability: ﬁrst, data normalization in the form of power curve alignment, and second, a robust method based on convolutional neural networks and feature-space extension.
We illustrate several examples of categorization, why it matters, and the success of machine learning methods using a set of real-world data sets.
Methodology: We’re flexible! An Interactive start and Q&A at the end may be best.