| 基于主成分分析的人工智能台风路径预报模型 |
| |
| 引用本文： | 黄小燕,金龙.基于主成分分析的人工智能台风路径预报模型[J].大气科学,2013,37(5):1154-1164. |
| |
| 作者姓名： | 黄小燕 金龙 |
| |
| 作者单位： | 1.南京信息工程大学大气科学学院, 南京210044;广西壮族自治区气象减灾研究所, 南宁530022 |
| |
| 基金项目： | 广西青年基金项目2011GXNSFB018001;国家自然科学基金资助项目41065002;广西自然科学基金北部湾重大专项项目
2011GXNSFE018006;科技部公益性行业(气象)科研专项GYHY201106036 |
| |
| 摘 要： | 利用主成分分析可以从具有随机噪声干扰的气象场提取主要信号特征,排除随机干扰的能力,论文以1980～2010年共31年6～9月西行进入南海海域的台风样本为基础,综合考虑台风移动路径的气候持续因子和数值预报产品动力预报因子,采用主成分分析的特征提取与逐步回归计算相结合的预报因子信息数据挖掘技术,以进化计算的遗传算法,生成期望输出相同的多个神经网络个体,建立了一种新的非线性人工智能集合预报模型,进行了分月台风路径预报模型的预报建模研究。在预报建模样本、独立预报样本相同的情况下,分别采用人工智能集合预报方法和气候持续法进行了预报试验,试验对比结果表明,前者较后者在6、7、8和9月份台风路径预报中,平均绝对误差分别下降了7.4%、4.8%、12.4%、17.0%。另外,论文进一步在初选预报因子和样本个例相同的情况下,通过比较新模型与直接采用主成分分析方法选因子并分别运用逐步回归和遗传—神经网络集合预报模型进行计算的预报精度差异表明,前者具有更高的预报精度,其原因是该方法挖掘利用了全部备选预报因子的有用预报信息,而且遗传—神经网络集合预报模型的是由多个神经网络个体预报结果合成,集合模型的各个神经网络个体的网络结构,是通过遗传算法的优化计算确定的,因此,该集合预报模型的泛化能力显著提高,在实际天气预报中具有较好的实用性和推广价值。
|
| 关 键 词： | 主成分分析 遗传算法 集合预报 气候持续法 台风路径 |
| 收稿时间： | 4/9/2012 12:00:00 AM |
| 修稿时间： | 9/5/2012 12:00:00 AM |
An Artificial Intelligence Prediction Model Based on Principal Component Analysis for Typhoon Tracks |
| |
| HUANG Xiaoyan and JIN Long.An Artificial Intelligence Prediction Model Based on Principal Component Analysis for Typhoon Tracks[J].Chinese Journal of Atmospheric Sciences,2013,37(5):1154-1164. |
| |
| Authors: | HUANG Xiaoyan and JIN Long |
| |
| Institution: | 1.College of Atmospheric Science, Nanjing University of Information Science & Technology, Nanjing 210044;Guangxi Research Institute of Meteorological Disasters Mitigation, Nanning 5300222.Guangxi Climate Center, Nanning 530022 |
| |
| Abstract: | We developed a novel nonlinear artificial intelligence ensemble prediction (NAIEP) model based on multiple neural networks with identical expected output created by using the genetic algorithm (GA) of evolutionary computation. We extracted the main signal feature from the meteorological fields with random noise and eliminated the random disturbance by principal component analysis (PCA). We set up the NAIEP model based on data of typhoons that occurred in the South China Sea from June to September in the period 1980-2010. The predictors were selected by the stepwise regression method and PCA both in the predictors of climatology persistence and Numerical Weather Prediction (NWP) products to predict the typhoon tracks for each month. Under the condition of identical model samples and independent prediction sample cases, we compared the genetic neural network ensemble prediction (GNNEP) model by selecting the predictors with both the method of Stepwise regression with PCA and the climatology and persistance(CLIPER) prediction model. The result showed that the former method was more accurate than the latter, and the average absolute error of the typhoon track from June to September decreased by 7.4%, 4.8%, 12.4%, and 17.0%, respectively. Under the condition of identical primary predictors and sample cases, we compared the prediction accuracies of the new model, the model of Stepwise regression, and the model of GNNEP (using only the method of PCA for the input predictors), and theoretically proved that the new model is more accurate than the other two. In the method which uses forecast information in all the alternative predictors and in the GNNEP model in which the resultant prediction from the ensemble integrates the predictions of the multiple ensemble members, the network structure is determined through the optimizing computation of GA;therefore, the generalization capacity of the ensemble prediction model is improved, leading to better availability and improved weather prediction |
| |
| Keywords: | Principal component analysis Genetic algorithm Ensemble prediction Climatology and persistance Typhoon tracks |
| 本文献已被 万方数据 等数据库收录！ |
| 点击此处可从《大气科学》浏览原始摘要信息 |
| 点击此处可从《大气科学》下载免费的PDF全文 |
|
