黄小燕,何立,赵华生,黄颖,吴玉霜. 2021. Shapley—模糊神经网络方法在华南台风卫星云图的长时效滚动预测中的应用[J]. 气象学报, (0):-, doi:10.11676/qxxb2021.017
Shapley—模糊神经网络方法在华南台风卫星云图的长时效滚动预测中的应用
Application of Shapley-fuzzy neural network method in rolling forecasting long-time of typhoon satellite Image in South China
投稿时间:2019-09-25  修订日期:2020-07-26
DOI:10.11676/qxxb2021.017
中文关键词:  卫星云图,华南区域台风,Shapley—模糊神经网络,k-近邻互信息估计,多元回归
英文关键词:satellite  image, typhoon  in south  China, Shapley-fuzzy  neural network, k-nearest  neighbor mutual  information estimation, multiple  regression
基金项目:(41765002),广西自然科学基金重点项目(2017GXNSFDA198030),广西自然科学基金面上项目(2018GXNSFAA294128,2018GXNSFAA281229)
作者单位E-mail
黄小燕 广西区气象科学研究所 gx_huangxy@163.com 
何立 广西区气象科学研究所 gx_heli@sina.cn 
赵华生 广西区气象科学研究所 2006zhaohuasheng@163.com 
黄颖 广西区气象科学研究所 yinger2001@126.com 
吴玉霜 广西区气象科学研究所 gx_wuyushuang@163.com 
摘要点击次数: 176
全文下载次数: 116
中文摘要:
      [目的] 为了更好地利用大量的卫星云图观测资料来提高台风暴雨的预报能力,解决并提高对台风强降水云系变化的预报精度,加大对未来云系变化的预报时效,构建基于合作对策Shapley—模糊神经网络的华南区域台风卫星云图非线性智能计算滚动集合预测模型,对增强卫星云图资料在台风暴雨天气预报中的实用性和及时性具有重要意义。[资料和方法]依据2013-2016年期间华南区域台风影响过程的卫星云图,采用类似于数值预报模式的集合预报方法,通过对每间隔6小时的卫星云图云顶亮温样本序列作经验正交函数分解,将提取出的时间系数作为云图预报建模的预报分量。考虑台风云系的发展变化,主要是受到云团环境物理量场的影响,利用数值预报模式的物理量预报产品作为各预报分量的预报因子,并采用k-近邻互信息估计的分步式变量选择算法,通过两步过程分别实现相关变量的选择与弱相关变量的剔除,分别建立相应时间系数的Shapley—模糊神经网络集合预报模型,进一步将预报得到的各时间系数与空间向量合成,重构得到未来时刻的卫星云图预报图,[结果和结论]实现了云图6 h-72 h的长时效客观滚动预测。实际华南区域台风卫星云图预测试验结果表明,新方案所预测得到的云图与实况云图相关较高,重构云图的基本轮廓、纹理特征分布、清晰度以及云系强弱方面都比较接近原始云图。另外,本文还进一步基于相同的云图预报因子,针对同样的建模和预报样本采用多元线性回归方案进行和新方案一致的云图预测。对比结果表明,这种非线性预报模型比线性方案能更好地预报未来较长时效台风云团的发展、移动的主要特征和变化趋势,其描述的预测云图与实际云图的主要特征更相符。云图预报时效达到了72小时的长时效,具有业务实用预报意义。
英文摘要:
      In order to make better use of the observation data of a large number of satellite cloud maps to improve the forecasting ability of typhoon storms, to solve and improve the prediction accuracy of cloud-based changes of typhoon strong precipitation, and to increase the forecasting time for future cloud system changes, based on the Shapley-Fuzzy Neural Network for cooperative countermeasures is a non-linear intelligent computing rolling set prediction model for typhoon satellites in South China. It is of great significance to enhance the practicality and timeliness of satellite cloud image data in the typhoon storm forecast. According to the satellite image of the typhoon impact process in South China during 2013-2016, based on the ensemble prediction method similar to the numerical forecasting model, a nonlinear intelligent computing rolling set prediction model based on cooperative strategy Shapley-fuzzy neural network is constructed. The empirical time orthogonal function decomposition is performed on the infrared cloud top temperatures(TBB) sample sequence of the satellite image every 6 hours, and the extracted time coefficient is used as the forecast component of the satellite image prediction modeling. Considering the development and changes of the typhoon image system, it is mainly influenced by the physical quantity field of the cloud group environment. The physical quantity forecasting product using the numerical forecasting model is used as the forecasting factor of each forecast component, and the stepwise variable selection algorithm based on k-nearest neighbor mutual information estimation is adopted. Through the two-step process, the selection of related variables and the elimination of weakly correlated variables are respectively implemented, and the Shapley-fuzzy neural network ensemble prediction model with corresponding time coefficients is established respectively. The time coefficients of the forecast are combined with the space vector to reconstruct the future. At the moment, the satellite image forecast map realizes the long-time objective rolling prediction of the image at 6 h-72 h. The results of the prediction experiments of the actual satellite image of the typhoon in South China indicate that the cloud image predicted by the new scheme is highly correlated with the live cloud image, and the basic contour, texture feature distribution, definition, and cloud strength of the reconstructed cloud image are all the live cloud map is more consistent. In addition, based on the same cloud map forecasting factor, the multivariate linear regression scheme is used to predict the cloud image consistent with the new scheme for the same modeling and forecasting samples. The comparison results show that this nonlinear forecasting model can better predict the development of the long-term typhoon cloud cluster, the main characteristics of the movement and the trend of change than the linear scheme. The predicted cloud map described by it is more consistent with the main features of the actual cloud map. The timeliness of the forecast of the cloud map has reached a long-term effect of 72 hours, with the significance of a practical forecast for forecasting business.
查看全文   查看/发表评论  下载PDF阅读器
分享按钮