The following experiments were performed in a 30–60 day period (which give us a time series of 720–1,440 measurements). The training and validation sets were created using 70% and 30% of these data, respectively. After the training process, we obtained 24 trained nets. The GA parameters used to perform the training process are shown in Table 3. The inputs of the ANN contain between 10 and 70 past observations. This means that GA will search within this range; the same case applies for the number of neurons. In the case of the mutation probability, it means that 55% of the offspring will mutate, but only less than 1% of its genes are mutated. The approximate time to train a single net is about 90 minutes.