Stochastic time series reconstruction of future wind farm output

To study the potential impact of the ever-increasing wind farm on future power grid, realistic long term time series of future wind generation at individual sites as well as for the system as a whole are required. This paper proposes a method to simulate time series of hourly wind speed. An ARMA(p, q) model with transformation procedure to a stationary process is chose to simulate the hourly wind speed for it can reflect the time sequential, statistical and stochastic characteristics. Ten years of hourly wind speed data are used to collect characteristic indices such as seasonal and diurnal patterns, autocorrelation and partial autocorrelation parameters. Wind farm output data in a whole year are reconstructed with diurnal and monthly pattern features. Comparison is made between the generated and real series of wind power in an aspect of probability distribution. The result demonstrates that the method is proper.