DescriptionA challenge in operating low-carbon electricity networks is the procurement of backup power for fast, corrective rebalancing of electricity generation and consumption immediately following the failure of large generators, transmission lines or other network components. Wind farms are well placed to secure revenue by providing this backup power, however they must be capable of contracting ahead of time to provide a fixed amount of power headroom at a high reliability and long lead times, so called 'firm power'. As stochastic generators, the capability to provide firm power is defined in the tail of a probabilistic forecast of minimum instantaneous power within time intervals specified by the power contracts. This is a deviation from the current practice of probabilistic energy (average power) forecasting at wind farms. A methodology for creating and scoring day-ahead probabilistic forecasts of minimum instantaneous power at wind farms is presented. The benefits of forecasting minimum power directly rather than repurposing energy forecasts are shown to be large. This suggests that specific forecasts of instantaneous power should be used to manage ancillary services provision. The competition winning approach of gradient boosted decision trees is utilized to handle multiple input features and create non-parametric quantile forecasts.
This methodology, and a benchmark based on current practice, are demonstrated in a case study comprising three wind farms in Great Britain (GB). In GB, system security requirements necessitate firm power provision with a response time of 1 second to be contracted day-ahead, meaning probabilistic forecasts of minimum instantaneous power are required on lead-times of 11 to 35 hours ahead. The forecasting methodology is extended to create probabilistic forecasts of the complete cumulative distribution function of instantaneous power in one-hour periods. This process allows the integration of time limited constraints to firm power allocation strategies from wind farms including the allocation of storage capacity day-ahead. The resulting probabilistic forecasts are reliable (calibrated) in contrast to re-purposing conventional energy (average power) forecasts, which are not. The forecasting of aggregated firm power from multiple wind farms is also investigated to demonstrate spatio-temporal effects on the variability of instantaneous power and the resulting forecast skill.
|Period||30 Jun 2021|
|Event title||International Symposium on Forecasting|
|Degree of Recognition||International|
Documents & Links
Project: Research Studentship - Internally Allocated