Calibration of a Wind Farm Wind Speed Model With Incomplete Wind Data

Publicly available information today on wind and wind generation is incomplete for deep understanding of their characteristics. The owner or operator of a wind farm typically only releases minutes-level turbine wind speed data. Although seconds-level wind power data may be obtained from the SCADA system, they only reflect aggregated performance at the wind farm level and thus cannot be used to obtain the covariance/correlation of wind speeds experienced by individual turbines at the seconds-level, which is a key to in-depth understanding of the characteristics of wind for wind farm control studies. To address this issue, this paper presents an empirical framework that uses incomplete wind data to calibrate a wind farm wind speed model. First, two such models are constructed from an ARIMA state-space model. Next, Monte Carlo filters and likelihood function maximization are incorporated to calibrate the models, reconstruct the unobserved states, and estimate the parameters. Finally, we test the model using synthetic wind data and demonstrate its effectiveness in statistically recovering missing information in the data.