Offline Power Quality Management and Control Using Neural Networks

In this study, the power quality (PQ) of a single-phase system is evaluated using a neural network (NN). The NN-based backpropagation (BP) technique is used to do offline assessment. The disturbances are produced using Simulink in MATLAB. The dyadic analysis filter bank receives these disturbances to acquire signal’s features. For training and testing NN, these extracted characteristics are used as input data. For single-phase evaluation without a filter bank, the conjugate gradient descent backpropagation algorithm (CGB) is initially utilized. However, training and accurate results require 100 neurons. For accurate findings, which are highly desired, the number of neurons is reduced by using a filter bank from 100 to 20. Hence, the feature extraction uses a filter bank. Using the same input data, seven backpropagation techniques train the network. The scaled conjugate gradient algorithm (SCGB)-based NN has been tested and shown to be superior to other algorithms. For training and testing the NN with the NN tool in MATLAB, the extracted features from the filter bank are utilized as the input data against the target data. The novelty to this work is that PQ disturbance signals are fed to a dyadic analysis filter bank and the number of neurons is reduced to 20. Also, here the SCGB-based NN is used for the first time for evaluating the accuracy of the algorithm. It is found from simulation results that SCGB-based NN gives 100% accuracy with minimum no. of neurons (20) and mean square error (MSE) during training (0.0034071).