Toward Z-number Valued Reinforcement Learning Problem
摘要
Real-world decision-making problems are characterized by a fusion of fuzzy and probabilistic uncertainties. In view of this, Zadeh introduced the concept of Z-number to describe imprecision and partial reliability of decision relevant information. In this paper we proposed an approach to solving Q-learning problem where rewards and constraints over actions are described by using Z-numbers. A typical decision problem is used to illustrate the approach.