Nonlinear Autoregression Neural Network for Meteorological Parameters Forecasting to Evaluate the Behavior of Radioactive Material Release from Nuclear Power Plants

In this paper a Nonlinear Autoregression Neural Network with exogenous input (NARX) is used to forecast temperature and wind speed. Six cases are studied to find the best condition with minimum mean absolute percentage error (MAPR) for prediction. In the first three cases the input data to NARX are four meteorological parameters (time, dew point, pressure and humidity) and the outputs are two parameters (temperature (Temp.) and wind speed (W.S.)). The first case is tested for Temp. and W.S. together as outputs. The second and the third cases are tested for Temp. and W.S. respectively. The results show that the first case gives MAPE for forecasting the two parameters together less than considering each one separately. As a result the first case is taken as a basic approach to the following three cases. In the fourth, fifth and sixth cases the input to NARX is taken as three parameters only without (dew point, pressure and humidity) respectively in each case. The NARX outputs are two parameters (Temp. and W.S.). In the last three cases the MAPE in the output parameters when compared with practical measurement of meteorological parameters do not exceed 20%. These results confirm that NARX model used in the current paper have the potential for successful application to meteorological parameters forecasting. NARX based weather forecast shows improvement over guidance forecasts from numerical models and official local weather.