Pattern Detection in Status Codes as an Optimization Tool in Offshore Wind Farms

Renewable, sustainable energy is an evolving field and will soon become a desirable necessity to sustain an ever-growing growing population. One viable source is wind power which is trending towards large farms with many involved turbines. The sheer size of the analytical data derived from these sizeable wind farms poses both a challenge and an opportunity for farm level optimization. Data driven analytics and machine learning are making the larger and more useful data sets available to be analyzed. One method based on these techniques, pattern detection, is already used very successfully in fraud detection and many other big data industries. One source of ascertaining the state of a turbine is through the appropriate understanding of its status codes. Such codes can indicate a myriad of outcomes from operational events to alarm conditions. It is expected that these codes follow consistent patterns and being able to extract these patterns from the data can help us understand how certain sequences relate to the turbine behavior and subsequently analysis of historically linked patterns can aid in predicting certain events. For example if codes A and B and C tend strongly to occur within the same time window then following an A-B pattern one could confidently predict a corresponding C event within the time window. Such an understanding enables the error codes to reveal more than simply a snippet of information, but a productivity- enhancing, cost-beneficial operational regime. These could then be used to track and anticipate failure events. For such high level computing to occur you must take the data into a parallelized environment making it scalable to an entire wind farm over the course of several years. In this study the effects of a varied time window on how the patterns manifested themselves was analyzed by frequency of occurrence and subsequently validated by physical insights into turbine behavior. This approach and the results extracted are based on real data of a full offshore wind farm and could be harnessed as a simple yet powerful tool for large scale wind farm optimization.