作　　者： ; ; ; ;

机构地区： 武汉大学计算机学院

出　　处： 《武汉理工大学学报》 2006年第7期97-100,共4页

摘　　要： 提出了一种新的预测气溶胶光学厚度AOT的优化融合模型，它把基于全局数据的径向基函数神经网络与基于局部区域数据的前向神经网络相结合，并通过优化权重组合，使得预测误差最小化。利用该模型对多角度成像光谱辐射仪MISR所采集的2002年7～9月份的数据进行AOT预测实验。结果显示：对美国陆地上空的大多数轨道，优化融合模型预测准确率明显高于全局模型和局部模型，其总体预测均方差比后二者分别减少了12．9％～24．5％和3．3％～10．5％．这表明优化融合模型是一种更为准确的摩间数据预测方法。 A novel fusion modeling approach to predict aerosol optical thickness （AOT） was proposed where both RBF ANN global model and BP ANN local models were constructed, and a prediction was obtained by weighting their outputs. The optimal weighting factors were determined through minimizing the mean squared prediction error over a specific region. Next, the fusion model was evaluated by using MISR geospatial remote sensing data collected from July to September in 2002 over the continental US. Results showed in most orbits, fusion model outperforms global and local models alone. The overall mean square prediction errors of latter two models were reduced 12.9 % -24.5 % and 3.3 % -10.5 % respectively by fusion model. The results confirm the proposed model was a more accurate spatial data predictor.

关 键 词： 空间数据 气溶胶光学厚度 全局预测模型 局部预测模型 融合预测模型

领　　域： [自动化与计算机技术] [自动化与计算机技术]