Intelligent islanding detection with grid topology adaptation and minimum non-detection zone

Islanding is the condition in which a distributed generator (DG) continues to power an area even though the electrical grid power is no longer present. This can be extremely dangerous to utility workers, and many techniques are dedicated to detect such a situation. This work presents a novel technique for islanding detection based on intelligent tools. Initially, the S-Transform is used to extract the frequency spectrum and calculate the energy from the signals of the three-phase voltages. Thus, the linear combinations of the energies for each phase are submitted to a feature selection algorithm in order to reduce the data dimensionality. Afterwards, the reduced subset of attributes is used as inputs for a predictive model based on Extreme Learning Machine. Very interesting results are presented and compared to conventional tools. The main contributions of the proposed approach are: (i) a fast islanding detection method incurring in low computational burden; and (ii) great generalization capability concerning the topology adaptation. These characteristics result in a reliable solution for islanding detection, which justifies its use in a real-time application.