基于SVM-BP神经网络的气象能见度数据缺失值预估

自动气象站能见度检测仪多采用光学装置采样,雨雪、粉尘等天气因素会对部分仪器镜头造成污染,导致能见度要素数据缺测.针对能见度数据缺失问题,本文选用安徽部分气象站的历年数据,首先运用灰色关联分析方法筛选出与能见度密切相关的其他气象要素,通过支持向量机和BP神经网络单一预估方法预估不同地形的能见度缺失值,然后采用最优权重组合将两种方法预估的能见度值进行组合,并与单一预估方法进行对比.结果表明组合方法的预估结果误差均值小、整体准确度高,可以保证台站观测资料的完备性,为短时天气预报、实况分析和气象公共服务工作提供有效依据.

SVM-BP neural network based meteorological visibility data filling

Most automatic weather stations sample visibility with optical devices, which are vulnerable to interference from rain, snow and dust, resulting in the inaccuracy or missing of visibility data.To address this and provide complete data for meteorological prediction, this paper proposes a Support Vector Machine-Back Propagation (SVM-BP) neural network based data quality control method for visibility data correction and filling.First, the grey correlation analysis is used to select meteorological elements closely related with visibility.Second, the visibility data of different terrains are estimated by SVM and the BP neural network independently, which are then combined by optimal combination weights.Historical weather visibility data from Anhui Meteorological Bureau are used to verify the proposed method.The results show that compared with the independent SVM or the BP neural network, the combined estimation has smaller mean error and higher overall accuracy.The proposed SVM-BP neural network method provides an effective tool for visibility data filling, thus lays theoretical basis for weather forecasting, weather analysis, meteorological research and public service.