عنوان مقاله [English]
This study, with the help of minimum temperature data, has addressed the prediction of frost during 21 years period by means of neural network in Kermanshah province. In order to forecast frost, data were converted to the values between 0 and 1 by means of a subjective and one to one (injective) function. We have used feed-forward neural network by one hidden interior layer with number of changeable neurons for each station to forecast and ultimately to determine frost spans. The algorithm in this investigation has used back propagation with batch training method and training functions such as Satlins, logsig and Satlin. Determination of frost and non- frost spans in each synoptic station and forecasting Precocious and serotinous frost are results of this study, and the designed network has had a convergence between 72.22 to 80.55 percent for each station. Results of this study reveal that, in spite of data limitations, MLP neural network has adequate ability in forecasting and estimating frost.