Hybrid self-learning controller for restoration of voltage power quality using optimized multilayer neural network

摘要

The main objective of this paper is to develop a hybrid predictor based on intelligence techniques for dynamic voltage restorer (DVR) to estimate the reference load voltage and self-tuned voltage regulation to enhance voltage power quality issues. The best fitted predictor model is obtained by using the potential merits of metaheuristic algorithms coupled with Antlion Optimization (ALO) and Genetic Algorithm (GA). This article proposes ALO optimized multilayer perceptron (MLP) neural network (NN) control algorithm for appropriate selection of weights and biases to generate the subsequent switching signal with lower error rates for load voltage estimation. The ANFIS-GA control algorithm is implemented for automatic tuning of fuzzy rules with fewer membership functions (MFs) to obtain the perfect ANFIS (Takagi-Sugeno) predictive models for effective DC and AC link voltage regulation. The proposed hybrid controller counteracts the drawbacks of traditional ANN such as early convergence, slow learning mechanism, less probability of local entrapment, and difficulties to achieve the estimated target under parametric uncertainties, namely, sag, swell, unbalance, and distortion. The accuracy of the prediction model is computed by considering the error indices for the training and testing data set. The result confirms that the hybrid system ALO-MLP NN establishes satisfactory performance in extracting the reference with the least error values.