A Q-learning-based swarm optimization algorithm for economic dispatch problem 基于Q学习的经济调度群优化算法.pdf

Neural Comput & Applic DOI 10.1007/s00521-015-2070-1 PREDICTIVE ANALYTICS USING MACHINE LEARNING A Q-learning-based swarm optimization algorithm for economic dispatch problem Yi-Zeng Hsieh1,2 • Mu-Chun Su2 Received: 10 March 2015 / Accepted: 5 October 2015 The Natural Computing Applications Forum 2015 Abstract In this paper, we treat optimization problems as 1 Introduction a kind of reinforcement learning problems regarding an optimization procedure for searching an optimal solution as There are two different approaches to optimization algo- a reinforcement learning procedure for finding the best rithms. While one is a derivative-based approach, the other policy to maximize the expected rewards. This viewpoint one is a population-based approach. The derivative-based motivated us to propose a Q-learning-based swarm opti- approach (e.g., steepest descent method and Newton’s mization (QSO) algorithm. The proposed QSO algorithm is method) fully utilizes an objective function’s derivative a population-based optimization algorithm which inte- information to determine search directions. On the con- grates the essential properties of Q-learning and particle trary, the population-based approach does not depend on swarm optimization. The optimization procedure of the functional derivative information to search the parameter QSO algorithm proceeds as each individual imitates the space of the objective function. It adopts a set of operators behavior of the global best one in the swarm. The best or some kind of mechanisms to manipulate a population of individual is chosen based on its accumulated performance individuals to evolve good solutions generation by gener- instead of its momentary performance at each evaluation. ation. The genetic