Wind farms can be analyzed using state estimation methods, which can be used to obtain its running state, including several aspects that cannot be easily obtained using other methods (e.g., capacitor bank aging) Using these methods on these types of networks is strongly affected by decoupling between active and reactive power and by a radial configuration, which is typical. For example, this decoupling affects its observability and robustness as well as the technical feasibility of the results. To overcome these drawbacks, an extended state estimation method is proposed in which the models for the different wind turbine technologies have been incorporated. These models have been mainly generated from measurement data using neural networks and polynomial fitting; these models do not require parameter values, which are rarely available from manufacturers. Furthermore, the resulting equations for modeling wind turbines are easily integrated into the state estimator due to their simplicity and derivatives.Thus, a method that guarantees feasible results, at least for wind turbines, was generated with increased observability robustness. The method was tested using measurement data from the Sotavento Wind Park, which has wind turbines with different types of technologies. (C) 2014 Elsevier Ltd. All rights reserved.