The net power output of a MicroGrid (MG) is often scheduled using optimization-based strategies. Recently, the new figure of the Aggregator (AG) has been introduced with the role of intermediate broker in the energy market, efficiently managing the interaction between a cluster of MGs and the system operators. To do that, the AG needs models of the related MGs to estimate both their power absorption/production, as a function of the energy prices, and the corresponding uncertainty ranges accounting for non-dispatchable generators and loads. To protect the MGs internal information and to reduce the complexity of the AG decision-making process, the problem of deriving these models from data is considered here. In order to cope with the problem nonlinearity and to quantify the uncertainty range, a nonlinear Set Membership approach is applied, and a new tuning method is described. The potentials of the proposed approach are tested with data obtained from a realistic MG model.