期刊
APPLIED SOFT COMPUTING
卷 148, 期 -, 页码 -出版社
ELSEVIER
DOI: 10.1016/j.asoc.2023.110873
关键词
Day-ahead wind power prediction; Ensemble model; Multiple indicators; Error correction
In this study, a novel ensemble model considering multiple indicators and error correction was proposed to improve the accuracy of day-ahead wind power prediction. The effectiveness of the model was verified using data from a real wind farm. The results showed that the proposed ensemble model outperformed other models in terms of prediction errors, and further improvements were achieved through error correction.
Accurate day-ahead wind power prediction is the basis for scheduling next-day generation in power systems with a high proportion of wind power. In this study, to overcome the adverse effects of changeable weather conditions and inaccurate numerical weather prediction data on day-ahead wind power prediction, a novel ensemble model considering multiple indicators combined with error correction was proposed. Four representative models that are frequently used for day-ahead wind power forecasting served as benchmark models. Then, considering the mean and variance of prediction errors, the nondominated sorting genetic algorithm was utilized to assemble the forecast outcomes from four benchmark models. Lastly, a statistics-based algorithm for error correction that considered temporal correlation was proposed to enhance prediction accuracy. The effectiveness of the proposed model was verified using data from a real wind farm. In a month of performing the successive day-ahead prediction task, the average errors of the proposed ensemble model were 0.1301 for the root mean square error and 0.1076 for the mean absolute error, which were 1.89%-14.52% and 2.62%-15.54% lower, respectively, than those of other models, including four other individual models and three ensemble models. Through error correction, the average forecast errors decreased to 0.1182 for the root mean square error and 0.0915 for the mean absolute error, indicating 9.14% and 14.96% reductions, respectively.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据