Skip to Main Content
A novel technique, namely optimal feature selection in the wavelet domain and supervised neural network-fault classifier is developed. An output signal of the speed deviations of each generator of the multi-area multi-machines system is taken as the input for the wavelet analysis. The "oscillation signature" for each of the 4 machines in a 'no fault condition', 'fault' with the PSS and without the PSS is recorded at various fault locations for fault detection using multi resolution analysis (MRA) wavelet transforms. The MRA decomposes the signal into different resolutions allowing a detailed analysis of its energy content and characteristics. It is then used as a feature for classes and locations of the fault. Three classifiers are used, namely the generalised regression neural network (GRNN); the probabilistic neural network (PNN), and the adaptive network fuzzy inference system (ANFIS), to train and find the fault location and classification and the results obtained are compared. The two-area 4-machine system with a double circuit transmission lines between the two areas is modified to include a fictitious bus for the study. To control the oscillation at various fault locations, a lookup table is devised using Simulink® for various values of the gain and the time constant of the conventional power system stabiliser. The integral square error and multiple objective functions are used as a fitness function during the minimization operation. Results show that the proposed control of the PSS is more robust in damping the oscillations as compared to the fixed conventional PSS.