Intermittent renewable energy sources (RES) require forecasting of the power they will generate, which is particularly relevant in the operation of isolated microgrids. The objective of this work is to identify Autoregressive Integrated Moving Average (ARIMA) models for forecasting solar photovoltaic and wind generation in microgrid operations. The community of Cayo Carenas, an islet located in Cienfuegos Bay, Cuba, was used as a case study. The dataset was integrated by solar radiation, temperature, and wind speed measurements, recorded by a meteorological station near the islet. The technical parameters of the microgrid's solar and wind farms were used to estimate the power generated by each RES. Different ARIMA model configurations were evaluated using fit and complexity metrics. Forecast accuracy was evaluated for prediction horizons ranging from 5 to 20 min, with a 10-min horizon selected, yielding a mean absolute percentage error of 0.44% for the solar model and 1.51% for the wind model. The research highlights the importance of parameter selection and adapting the ARIMA model structure to the specific characteristics of each RES.