The Risk Neural Network Based Visibility Forecast
Kai Wang, Hong Chao Zhao, Aixia Liu, Zhipeng Bai · 2009
Currently, the measurement of the visibility mainly depends on the human eyes, so that the objectivity is relatively poor. In general, the higher the visibility is, the greater the measurement error is. On the other hand, in practice, as opposed to high visibility situations, low visibility situation is more notable. Therefore, for the same forecast error, low visibility should have a higher risk value. Based on this principle, a risk neural network model is proposed, in which the relatively high risk is given for the low visibility case, while the relatively low risk is given for the high-visibility case. Experimental results show that the risk neural network model is superior to the standard one and linear regression model, which provide a support for our work.