Reinforcement learning based PID controller design for LFC in a microgrid. Issue 4 (3rd July 2017)

• Record Type: Journal Article
• Authors: Esmaeili, Mehran; Shayeghi, Hossein; Mohammad Nejad, Hamid; Younesi, Abdollah
• Is Part Of: Compel. Volume 36:Issue 4(2017)
• Page Start: 1287
• Page End: 1297
• Publication Date: 2017-07-03
• Subjects: Design optimization -- Control theory -- Adaptive fuzzy logic control -- Electrical power systems -- Microgrid -- Reinforcement learning; Electrical engineering -- Data Processing -- Periodicals; Electrical engineering -- Mathematics -- Periodicals; Electrical engineering -- Periodicals; Electronics -- Data Processing -- Periodicals; Electronics -- Mathematics -- Periodicals; 621.3
• DOI: 10.1108/COMPEL-09-2016-0408
• Languages: English
• ISSNs: 0332-1649

Abstract

Abstract : Purpose: This paper aims to propose an improved reinforcement learning-based fuzzy-PID controller for load frequency control (LFC) of an island microgrid. Design/methodology/approach: To evaluate the performance of the proposed controller, three different types of controllers including optimal proportional-integral-derivative (PID) controller, optimal fuzzy PID controller and the proposed reinforcement learning-based fuzzy-PID controller are compared. Optimal PID controller and classic fuzzy-PID controller parameters are tuned using Non-dominated Sorting Genetic Algorithm-II algorithm to minimize overshoot, settling time and integral square error over a wide range of load variations. The simulations are carried out using MATLAB/SIMULINK package. Findings: Simulation results indicated the superiority of the proposed reinforcement learning-based controller over fuzzy-PID and optimal-PID controllers in the same operational conditions. Originality/value: In this paper, an improved reinforcement learning-based fuzzy-PID controller is proposed for LFC of an island microgrid. The main advantage of the reinforcement learning-based controllers is their hardiness behavior along with uncertainties and parameters variations. Also, they do not need any knowledge about the system under control; thus, they can control any large system with high nonlinearities.