Optimizing Artificial Neural Networks for Multi Area Network : A Metaheuristic Perspective

Abstract

Regression, data classification and function approximation are among the most common applications that make use of machine learning models. However, due to the vast range of applications for machine learning, comprehensive understanding of how to choose a model based on machine learning, as well as how to choose its structure, training method, and performance analysis criterion, is frequently lacking. This key issue is addressed using a well-known load frequency control (LFC) problem which is instrumental in maintaining balance between power generation and load demand. The study relates to developing and contrasting the performance of the Meta heuristic optimization based approach to the artificial neural network (ANN) based machine learning topology for the LFC control against the optimized ANN controller. The selected performance indices are peak time and settling time of the frequency deviation. It is observed that Linear Neural Network (LNN) outperforms Feedforward Neural Network (FNN) and Recurrent Neural Network (RNN). The performance is enhanced when its parameters are optimized by the metaheuristic algorithms.