The objective of this study is to analyze 3 approaches to objectively set the configuration parameters of a support vector regression algorithm to forecast insolation. The approaches were based on techniques such as multifold cross-validation, grid search, and frequency analysis. The ${ u}$ support vector regression with a Gauss function as kernel function was used to forecast insolation. The configuration parameters set were the cost parameter, the gamma parameter in the kernel function, and the ${ u}$ parameter. The analysis was done using weather related data of Tsukuba city in Japan, from January to December of 2009. The results show that variations of the forecast errors caused by using the different approaches were small, 4.5% in the worst case. Moreover, any of the proposed approaches yielded forecasts of insolation with annual root mean square errors lower than $0.12kWh/m^2$ and mean absolute errors lower than $0.07kWh/m^2$, what shows their applicability.