Probabilistic Forecasting of Wind Power Ramps Using Autoregressive Logit Models

• Activity Number: 623
• Type: Invited
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): James W. Taylor*+
• Companies: University of Oxford
• Keywords: wind power ramps ; probability forecasting ; autoregressive logit models

Abstract

A challenge for the efficient operation of power systems and wind farms is the occurrence of wind power ramps, which are sudden large increases or decreases in the power output from a wind farm. This paper considers the probabilistic forecasting of a ramp, defined as exceedance beyond a specified threshold. In this paper, we directly model the exceedance probability using autoregressive logit models fitted to wind power time series. These models can be estimated by maximizing a Bernoulli likelihood. To try to capture the extent to which an observation does or does not exceed the threshold, we also consider a likelihood based on the asymmetric Laplace density, which has previously been employed for quantile estimation. We simultaneously estimate the probabilities of exceeding different thresholds using a categorical distribution. To model jointly the probability of ramps at more than one wind farm, we use a spatial autoregressive logit model, estimated using a multivariate Bernoulli density. We evaluate one step-ahead probability forecasts for hourly wind power data using the Brier score and a test for conditional coverage.