The integration of photovoltaic systems (PVS) into the electrical grid involves the use of various types of converters and inverters based on power electronics, along with the inherent disadvantages of solar energy, resulting in the generation of harmonics in the electrical signal of the grid, producing negative effects such as overheating of conductors or reduction of the lifespan of electrical equipment, due to the stochastic nature of THD levels, making prediction difficult with climatic or electrical variables due to their low correlation. The objective of this research was to predict current THD indices during irradiance hours. For this purpose, clustering algorithms were used that allowed increasing the level of correlation and prediction by finding clusters with days that have similar statistical profiles of THD and thus allowed developing better forecasting models based on deep learning or machine learning, in which values of coefficient of determination \({\text{R}}^{2}\) greater than 0.90 were obtained with test days, thus validating the proposed methodology and overcoming the difficulties of low correlation.