Search

Precise and robust wind power prediction can effectively alleviate the problem caused by the randomness and volatility of wind power. Ensemble learning can successfully improve forecasting precision and robustness, and quantify the uncertainty of the prediction. This paper presents a new ensemble probabilistic forecasting framework, based on modified randomized maximum a posteriori (MAP) sampling technique, echo state network (ESN) and generalized mixture (GM) function to bring superior forecasting results. The proposed model first trains a set of independent ESN models for probabilistic forecasting using the modified randomized MAP sampling technique, and then dynamically weighs and ensembles the base model forecasting through the GM function. The proposed model and other benchmark models have been implemented on four wind power datasets from different places to illustrate the advantage of the proposed method. The compared result indicates that the suggested model outperforms some state-of-the-art models and can successfully achieve dynamic ensemble probabilistic prediction.

Precise and robust wind power prediction can effectively alleviate the problem caused by the randomness and volatility of wind power. Ensemble learning can successfully improve forecasting precision and robustness, and quantify the uncertainty of the prediction. This paper presents a new ensemble probabilistic forecasting framework, based on modified randomized maximum a posteriori (MAP) sampling technique, echo state network (ESN) and generalized mixture (GM) function to bring superior forecasting results. The proposed model first trains a set of independent ESN models for probabilistic forecasting using the modified randomized MAP sampling technique, and then dynamically weighs and ensembles the base model forecasting through the GM function. The proposed model and other benchmark models have been implemented on four wind power datasets from different places to illustrate the advantage of the proposed method. The compared result indicates that the suggested model outperforms some state-of-the-art models and can successfully achieve dynamic ensemble probabilistic prediction., Competing Interests: The authors declare no conflict of interest., (© 2022 The Author(s).)