Modelling harmonics drawn by nonlinear loads

This paper is devoted to obtain a model of nonlinear loads (NLL) connected to LV electric networks, in order to predict the consequences of including parallel active or passive filters. The model is based on Neural Networks (NNs) and its purpose is the prediction of harmonic currents generated by a certain load when supplied by a network with significant series impedance, disturbed by other random and unknown neighbour loads. The NNs have been trained by using data obtained from several circuit simulations of a network supplying several unknown neighbour loads beside the load of interest, consisting of a set of rectifiers. The model is validated using the same network structure, with different neighbour loads and different load conditions.