Week-ahead Predictions of Wind Power Based on Weather Classification

At present, the researches on short-term wind power prediction mostly focus on three days, but the maintenance schedule of wind farm and the generation schedule of hydro-thermal power system both hope to get the wind power forecasting results week-ahead. Exploring the law of wind power fluctuation becomes one of the most crucial issues to extend the prediction time scale. This paper proposes a combined probabilistic prediction approach based on weather classification to forecast the wind power week-ahead. This approach considers that wind power has a similar fluctuation law under the same weather type, and the combination model is used to improve forecasting accuracy. In the approach, the GK fuzzy clustering algorithm based on subtraction clustering is used to divide the weather into several typical types. And the combination probabilistic forecasting model formed by the sparse Bayesian learning, kernel density estimation, and beta distribution estimation is trained for each weather type. Meanwhile, the Gaussian Case-based Reasoning is used to classify the forecasting sample into corresponding weather type and the corresponding combined model is used to make the forecast. Case studies employ the proposed approach as well as several benchmarks to achieve week-ahead predictions of a wind farm located in China and the superior performance demonstrates the effectiveness of the proposed approach.