Energy storage sizing for wind power: impact of the autocorrelation of day-ahead forecast errors

Availability of day-ahead production forecast is an important step towards better dispatchability of wind power production. However, the stochastic nature of forecast errors prevents a wind farm operator from holding a firm production commitment. In order to mitigate the deviation from the commitment, an energy storage system connected to the wind farm is considered. One statistical characteristic of day-ahead forecast errors has a major impact on storage performance: errors are significantly correlated along several hours. We thus use a data-fitted autoregressive model that captures this correlation to quantify the impact of correlation on storage sizing. With a Monte Carlo approach, we study the behavior and the performance of an energy storage system (ESS) using the autoregressive model as an input. The ability of the storage system to meet a production commitment is statistically assessed for a range of capacities, using a mean absolute deviation criterion. By parametrically varying the correlation level, we show that disregarding correlation can lead to an underestimation of a storage capacity by an order of magnitude. Finally, we compare the results obtained from the model and from field data to validate the model.