Probabilistic Wind-Power Forecasting Using Weather Ensemble Models

During the past one to two decades, the probabilistic forecasting of a wind-power generation has been regarded as a necessary input to decisions made for the purpose of reliable and economic power systems operations, especially since the penetration of the renewable energy has begun to grow rapidly. Probabilistic forecasting differs from traditional deterministic forecasting in which it takes uncertainty into account. This paper proposes a modified nonparametric method for constructing reliable prediction intervals (PIs). The lower upper bound estimation (LUBE) method is adapted to construct PIs for the wind-power generation, based on the ensemble wind-speed data from the numerical weather prediction system of the Central Weather Bureau of Taiwan. The charged system search (CSS) is used to adjust parameters in LUBE. The performance of the proposed method is examined using datasets from several wind farms in Taiwan. Simulation results demonstrate that the quality of PIs output by the proposed model significantly exceeded that of those constructed using the persistence model with a 1-h-ahead time horizon.