Posters

Abstract

Plenty of research has been devoted to wind power forecasting ([1], [2], [3]), which can be split into forecasting of weather, most importantly wind speed, and forecasting of turbine availability. Forecasting of availability is the more underrepresented of the two, and specifically forecasting of availability of turbines that are currently experiencing a technical malfunction. As the markets often operate on a day-ahead commitment principle, it is equally important to determine if turbines can be declared available to generate electricity 24 hours in advance, as it is to forecast wind resource. Forecasting the availability of turbines impacted by technical malfunctions often relies on the expert opinion of an experienced technician, which, albeit reliable, is not scalable or possible to do in real time. The goal of this presentation is to summarize research on data-driven approaches to predict when a turbine will become available again [4]. A data-driven approach is meant to provide an initial estimate which can be adjusted by experienced technicians if time allows. A method for nonparametric clustering of survival curves was developed and validated. The Weibull Accelerated Failure Time model with the clustered error codes and logarithm of energy produced in the month prior to failure is found to perform significantly better than alternatives. Validation shows that the approach predicts availability with quite well. The presentation will focus on: *  Providing context as to why availability forecasting is important: potential for forecast use in everyday operations * Explaining the required data preprocessing and feature engineering steps (in particular nonparametric clustering of turbine error codes) * Sharing model testing and validation results, with result accuracy * Demonstrating how time to recovery estimates can be made at scale via survival analysis, and used as an input to anyone’s availability forecasting, or provide the audience with a better understanding of what they need to request from third party forecasting services. The innovative content of this proposed presentation comes from three key aspects: * Innovative research into availability forecasting * Innovative use of survival analysis * Innovative clustering of error codes References: [1] Wang, Xiaochen et. al. (2011), A Review of Wind Power Forecasting Models [2] Foley, Aoife M. et. al. (2012), Current methods and advances in forecasting of wind power generation [3] Sideratos, George et. al. (2007), An Advanced Statistical Method for Wind Power Forecasting [4] Palets, Anton (2023), Wind Turbine Recovery Forecasting using Survival Analysis, Master’s Theses in Mathematical Sciences, Lund University