DRLA: Dimensionality ranking in learning automata and its application on designing analog active filters

Abstract

In this paper, a learning automata (LA) method is introduced based on dimensionality ranking (called DRLA), and its application for designing analog active filters is investigated. The basic idea in the DRLA is that the number of iterations and ordering selection of dimensions for the next phase is according to effect of each dimension in the fitness improvement of the current phase or previous phases. This strategy improves the exploration of search space and escaping the local minimums. Also, to achieve the proper balance in exploration and exploitation, the control parameter τ is changed by a linear control manner. To validate the applicability, accuracy, and performance of the proposed approach, it is first evaluated on 23 benchmark functions and then, as a new application of LA, applied to the optimal design of two different analog active filter topologies as real-world engineering problems. The filter component values are estimated within E96 series for resistors and capacitors, and a single-objective function is also considered as average quality factors deviation and cut-off frequency deviation. Simulation results compared to some rival single-objective metaheuristic optimization methods reveal that the improved method has competitive performance in terms of fitness, convergence, statistical analyses, box plot, and bode curve as well as runtime.