Photovoltaic power systems connected to the electrical grid are becoming an important element in the adoption of solar energy as a meaningful power source. However, radiation nonlinear behavior considerably affects photovoltaic systems performance and certainty. Solar radiation forecasting analysis is presented as an alternative to improve photovoltaic systems’ certainty and robustness. The training of an autoregressive neural network applied to solar radiation forecasting supported by wavelet processing is presented in this paper. Solar radiation data from 2010 to 2012 corresponding to Cajica, Colombia is used as input data to the system in order to get predicted radiation data from the following year. Validation of the results is performed by analysis of the real 2013 radiation data through mean squared error and linear regression. The results showed an accuracy of 93%, indicating that the proposed system can be used as a tool for the design of photovoltaic power systems in Colombia.