In this paper, we introduce a new approach to aggregate traffic data based on speed/acceleration probability distributions. The method allows to estimate energy consumption for vehicles on a road network subdivided into road segments. The key point of the approach is to aggregate data thanks to clustering techniques, to obtain meaningful class of traffic behavior. We also compare the energy consumption obtained to a reference one, using data from a traffic simulator. Numerical simulations include a study of the influence of several parameters of the method.