Deterministic and probabilistic wind power forecasting using a hybrid method

摘要

This paper proposes a hybrid method for probabilistic wind power forecasting. The proposed approach consists of data classification, deterministic forecasting and probabilistic forecasting stages. In the data classification stage, a fuzzy k-means clustering algorithm is used to classify the historical time series of wind power into various wind classes. Several support vector regression (SVR) models that correspond to diverse wind speeds are then established to train the collected data in the deterministic forecasting stage. An enhanced harmony search (EHS) algorithm is presented to estimate the parameters for each SVR model. Using the wind speed forecasts given by Taiwan Central Weather Bureau (TCWB) for every three hours, a corresponding forecasting model is then used to produce wind power forecasts for future 3 hours in steps of 15 minutes. To assess the risk that is associated with forecasting errors, an EHS-based quantile regression (QR) method is used to provide the confidence intervals for forecasted values in the probabilistic forecasting stage. During testing on a practical wind power generation systems, the proposed method gives better forecasting accuracy and produces more reasonable confidence intervals than existing methods.