File
Title Transcription
イデン アルゴリズム ニヨル セイヤク ツキ マルコフ ケッテイ カテイ ノ カイホウ
Title Alternative
A Solving Method of a MDP with Constraint by Genetic Algorithm
Authors
Abstract
We consider discrete time Markov decision process (MDP) with finite state space, finite action space and two kinds of immediate reward The problem is to maximize time average reward generated by on reward stream, subject to that the other reward is not smaller than a prescribed value. The probelm is analyzed in the range of pure stationary policies MDP with one optimality criterion and no constraint can be solved by usual policy improvement method. MDP with one reward constraint can be solved by linear programming, in the range of mixed policies. On the other hand, however, when we restrict the policies to pure polices the problem is some conbinatrial problem,
for which any solving method has not been discovered. In this paper, we propose an approach applying Genetic Algorithm in order to carry on a search process effectively
and to obtain a near optimal pure stationary policy. A numerical example is given to examine the effeciency of the approach proposed here.
Publisher
鳥取大学工学部
Content Type
Departmental Bulletin Paper
ISSN・ISBN
0385-8596
NCID
AN00174610
Journal Title
鳥取大学工学部研究報告
Current Journal Title
鳥取大学工学部研究報告
Volume
26
Issue
1
Start Page
295
End Page
302
Published Date
1995-11
Text Version
Publisher
Rights
注があるものを除き、この著作物は日本国著作権法により保護されています。 / This work is protected under Japanese Copyright Law unless otherwise noted.
Citation
鳥取大学工学部研究報告. 1995, 26(1), 295-302
Department
Faculty of Engineering/Graduate School of Engineering
Language
Japanese