New Zealand Statistical Association 2024 Conference

Hannah Yun

University of Auckland

Advanced methods for time series data applied to prediction of operating modes for wind turbines

This is joint work with Ciprian Doru Giurcăneanu, Gill Dobbie

Wind turbines can be characterised by distinct operating modes that reflect efficiency of the turbine under various conditions. In this talk, we focus on the forecasting problem for univariate discrete-valued time series of operating modes of a wind turbine. We define three prediction strategies to overcome the difficulties associated with missing data. The first strategy is to ignore missing values and to focus solely on available data. The second strategy imputes the missing values by replacing them in the time series with an estimate. The last strategy treats missing values as an additional operating mode. These strategies are evaluated through experiments using five forecasting methods across two real-life datasets. Two of the forecasting methods have been introduced in the statistical literature as extensions of the well-known context algorithm: variable length Markov chains and Bayesian context tree. Additionally, we consider a Bayesian method based on conditional tensor factorisation and two different smoothing techniques from the classical tools for time series forecasting: Exponential smoothing and Whittaker smoother. Each combination of prediction strategy and forecasting method is evaluated in terms of prediction accuracy versus computational complexity. We provide guidance on the methods suitable for forecasting the time series of operating methods in wind turbines. The prediction results demonstrate that high accuracy can be achieved with reduced computational resources.