This dissertation presents a study of electricity load aggregated in time intervals into the same day. This type of aggregation has been used by the Brazilian´s electrical sector in the form of three load levels called low, middle and high. However, these load levels were always indirectly achieved from the hourly load aggregation, and it was not found any direct treatment of them as a series up to this publication. The work developed in this dissertation describes mathematically the calculation of the three levels of load and makes an analysis of the series formed by these variables checking their own characteristics, the relationship among themselves and the influence of exogenous variables. It also shows the modeling of some forecast methods for such series employing techniques of both statistics in computational intelligence, introduces the level profile concept and proposes a hybrid model of forecasting, formed by dynamic regression, pattern classification and fuzzy logic, to predict the load level pattern. In the proposed model, the dynamic regression is used in the forecasting of the daily global load that is used to match the resulting pattern, described in a normalized way to the actual load values. The profiles are obtained from a classifier based on the subtractive clustering technique. The combination of the profiles that compose the level pattern forecast to the desired horizon is carried out by a fuzzy logic system that uses the temperature as input variable intrinsically treating non-linear relationships between load level and temperature variables.