In this work, a temperature predictor has been designed and implemented based on different series of meteorological data. The prediction is built by an artificial neural network multilayer perceptron, using 5 samples as window size of meteorological data. Besides, the floating point algorithm was evaluated, reaching a mean square error of 0.35, meaning a variation of 0.28 Celsius degrees versus the real temperature. Different approaches will be applied in order to show our best proposal.
Download Full PDF Version (Non-Commercial Use)