Abstract:The forecast of poor visibility has been paid more attention than good visibility. A risk neural network model was proposed based on following approach: poor visibility was assigned a higher risk value and good visibility was assigned a lower risk value. Observation of 6 meteorological factors, monitoring concentrations of SO2, NO2, PM10, and visible distances were chosen as the input data. The visibility after 24 hours was predicted as the output. A case study with the data from 2003 to 2007 in Tianjin region showed that the risk neural network model performed better than the traditional neural network models as well as linear regression model in terms of correlation and relative error.
王恺, 赵宏, 刘爱霞, 韩斌, 白志鹏. 基于风险神经网络的大气能见度预测[J]. 中国环境科学, 2009, 29(10): 1029-1033.
WANG Kai, ZHAO Hong, LIU Ai-Xia, HAN Bin, BAI Zhi-Peng. Development and validation of visibility forecast technique based on the risk neural network.. CHINA ENVIRONMENTAL SCIENCECE, 2009, 29(10): 1029-1033.