In Economic Sciences, as well as in Engineering and Natural Sciences, phenomena occurs that depends on observing data at time intervals during a specific period. These observations of values are called Time Series. The techniques available to analyze these observations are called Time Series Analysis. This set of techniques aims to construct a model with adequate number of parameters estimated in order to fit the model to the time series. Obtaining a suitable model is extremely important because it can reveal some characteristics of the time series that help predict future values. This ability is of great practical relevance, especially in the generation of wind energy. Several techniques to perform the time series prediction are used, but all have a forecast error. The challenge is then to minimize forecast error by constructing a model that best fits the data. In this way, the objective of this work is to create prediction models of the speed and direction of the winds, considering that there are historical data. These models will first be created using Artificial Neural Network (RNA), and then the NARMA (Nonlinear Autoregressive Moving Average) model is used together with RNA for comparison of results. The forecast is made for one and six steps ahead. The use of RNA alone resulted in satisfactory results in all cases, with the exception of the prediction for six steps ahead of speed. However, when the hybrid model (NARMA + RNA) was used, all results improved considerably. The wind forecast series samples are from Wind Farms in Rio Grande do Norte, owned by the Brookfield Renewable Energy Company, and are collected at the base of 10 minutes.