基于风险神经网络的大气能见度预测

Development and validation of visibility forecast technique based on the risk neural network

针对空气污染导致大气能见度降低的预测研究,构建了一个风险神经网络模型,模型以6个气象因子、3种主要污染物（SO2,NO2,PM10）浓度和能见度作为输入因子,输出为24h后能见度的预测值.该模型对低能见度情况的数据给予相对较高的风险值,而对高能见度情况的数据则给予相对较低的风险值.以天津市2003-2007年的气象数据对模型进行检验,结果表明该风险神经网络模型优于传统神经网络模型和线性回归模型.

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.