Application of Neural Networks to Weather Forecasting with Local Data

摘要

Weather forecasting has been made along History in very different ways, many of them try to establish a relation between meteorological data at time t and before, with data after time t. In this paper propose to use a multilayer Perceptron, with error backpropagation training procedure, as a classification tool working with local meteorological data to produce weather prediction. Changing the set of magnitudes at input and/or neural network's structure, we have another classifier, and consequently we need a method that allows to select the best among them. We present a parameter that we call efficiency of a classifier, and according to it, we can vary the neural network's structure and/or meteorological magnitudes involved to improve the results. The results obtained by this method vary from 87.4% in rain forecasting for next 6 hours, and 49.5% in variation of minimum daily temperature for next 96 hours.