Weather Forecasting Models Using Ensembles of Neural Networks

Abstract

This paper examines applicability of Hopfield Model (HFM) for weather forecasting in southern Saskatchewan, Canada. The model performance is contrasted with multi-layered perceptron network (MLPN), Elman recurrent neural network (ERNN) and radial basis function network (RBFN). The data of temperature, wind speed and relative humidity were used to train and test the four models. With each model, 24-hr ahead forecasts were made for winter, spring, summer and fall seasons. Moreover, ensembles of these models were generated by choosing the best values among the four predicted outputs that were closest to the actual values. Performance and reliabilities of the models were then evaluated by a number of statistical measures. The results indicate that the HFM was relatively less accurate for the weather forecasting problem. In comparison, the ensembles of neural networks and RBFN produced the most accurate forecasts.