This paper presents a model for short-term forecasting of solar potential on a horizontal surface. This study is carried out in to the context of valuing of energy production from photovoltaic solar sources in the Sahelian zone. In this study, Autoregressive Moving Average (ARMA) process is applied to predict global solar potential upon 24 hours ahead. The ARMA (p, q) is based on finding optimum parameters p and q to better fit considered variable (sunshine). Data used for the model calibrating are measured at the station of Ecole Supérieure Polytechnique of Dakar. Records are hourly and range from October 2016 to September 2017. The choice of this model is justified by its robustness and its applicability on several scales through the world. Simulation is done using the RStudio software. The Akaike information criterion shows that ARMA (29, 0) gives the best representation of the data. We then applied a white noise test to validate the process. It confirms that the noise is of white type with zero mean, variance of 1.252 and P-value of about 26% for a significant level of 5%.Verification of the model is doneby analyzing some statistical performance criteria such the RMSE =0.629 (root mean squared error), the R² = 0.963 (Coefficient of determination), the MAE=0.528 (Mean Absolut Error) and the MBE=0.012 (Mean BiasError). Statistics criteria show that the ARMA (29,0) is reliable; then, can help to improve planning of photovoltaic solar power plants production in the Sahelian zone.