The distribution substations power curves are often affected by outliers: very discrepant measurements of the curve standard behavior. If the presence of outliers is very large, the power utilities internal studies and analysis developed from the history of the data collected may be compromised. In order to detect and correct atypical data, a complementary module for the supervisory system was developed. In the previous work of the authors, two techniques were developed based on artificial intelligence: fuzzy logic and artificial neural networks. This work presents a hybrid technique that uses the best performances of the previous ones to reduce the relative error, increasing the correction technique performance. Furthermore, it develops an outlier detection method based on the standard behavior of the curve and the distribution of the measurements. The performances of the techniques are compared using real data from a substation over 6 years.