Nowadays the consumption and generation profile estimation is of the greatest importance. New loads characterized by coincident peak of consumption (e.g., home charging of electric vehicles) or by high absorption peaks (heat pumps) are increasingly frequent. The presence of such loads must be carefully considered for network investments and for the optimization of asset management. Moreover, the massive diffusion of non-programmable renewable sources gives a leading role to the flexibility of demand, which is crucial for the success of the energy transition. The variety and difference of the electrical behaviour of LV customers, even nominally homogeneous, need stochastic methods for estimating the load profile on the LV/MV interfaces for the planning and the operation of distribution network, and for estimating the flexibility potential of demand. In this paper different techniques for modelling the demand composition are compared to evaluate the quality of the DSO models on real customers. In particular, the power peak of a given network section is calculated as key indicator for estimating the risk of overloading of lines and secondary substation transformers. Different methods of calculation have been applied on a dataset gathered with a recent measurement campaign in Italy by considering real LV distribution networks.