| 摘 要: 台风是沿海地区影响最严重的自然灾害之一,准确预测风速极值对防灾减灾至关重要。采用自编码器进行异常值检测,识别数据中的极端值,基于最大熵原理拟合较大异常值分布,并分别与经典的 GEV 分布和Gumbel分布进行对比,进一步提取分位点进行区间估计。结果表明,最大熵分布在相同置信水平下表现最佳,区间覆盖率达到94.3%,相比 GEV 分布提高了27.2个百分点,且置信区间长度缩小了88.2%,同时,最大熵分布优于Gumbel分布,展现出良好的鲁棒性与推广价值,为极端风速建模提供了新的思路。 |
| 关键词: 风速极值预测 异常值 自编码器 最大熵分布 |
|
中图分类号: TP391
文献标识码: A
|
|
| Wind Speed Extreme Value Prediction Based on AutoEncoder and Maximum Entropy |
|
CHEN Yuxin, HE Xiaoxia, LI Chunli
|
(School of Mathematics and Systems Science, Wuhan University of Science and Technology, Wuhan 430065, China)
CHENYXKKK@wust.edu.cn; hexiaoxia@wust.edu.cn; lichunli@wust.edu.cn
|
| Abstract: Typhoon is one of the most severe natural disasters affecting coastal areas. Accurate prediction of wind speed extremes is crucial for disaster prevention and mitigation. An AutoEncoder is used for outlier detection to identify extreme values in the data. Based on the principle of maximum entropy, the distribution of larger outliers is fitted and compared with the classical GEV distribution and Gumbel distribution respectively. Further, quantiles are extracted for interval estimation. The results show that the maximum entropy distribution performs best at the same confidence level, with an interval coverage rate of 94.3% , which is 27.2 percentage points higher than the GEV distribution and the confidence interval length is reduced by 88.2% . At the same time, the maximum entropy distribution is superior to the Gumbel distribution, demonstrating good robustness and generalization ability, providing a new idea for extreme wind speed modeling. |
| Keywords: wind speed extreme value prediction outlier AutoEncoder maximum entropy distribution |
