Time series forecasting models are designed to meet the demands of various sectors of society, such as industry, service and agriculture. Such models are used, for example, to evaluate stock, define product logistics and in weather forecast. In this context, this thesis proposes the GPFISForecast+ model, which allows the automatic design of a linguistically interpretable prediction system. The GPFIS-Forecast + has as its starting point the previously developed GPFIS-Forecast, a result of a hybridization of genetic programming techniques and type-1 fuzzy systems. The rule base of the type-1 fuzzy inference system (SIFT1) is optimized through Multigenic Genetic Programming. In the construction of the GPFIS-Forecast + model, type-2 fuzzy sets are used, which provide a model with greater flexibility and, according to the literature, a greater capacity of dealing with noisy data. It also adds the possibility of using exogenous variables to increase genetic variability in the stochastic optimization process. Results of the comparison between the GPFIS-Forecast and GPFIS-Forecast+ models show that the second presents the smallest errors in general. When compared to other models in the literature, the GPFIS-Forecast+ presents higher accuracy for series without tendency and without heteroscedesticity.