A framework for dynamic decision making by multi-agent cooperative fault pair algorithm (MCFPA) in retail shop application

Abstract

The paper gives the novel framework for dynamic decision making in the retail shop application based on proposed improved Nash Q-learning by Fault Pair Algorithm. Accordingly, this approach presents three retailer shops in the retail market. Shops must support each other to gain maximum revenue from cooperative knowledge via learning their own policies. The suppliers are the intelligent agents to utilize the cooperative learning to train in the situation. Assuming significant theory on the shop’s storage plan, restock time, arrival process of the customers, the approach is formed as Markov decision process model that makes it feasible to develop the learning algorithms. The proposed algorithms obviously learn changing market situation. Moreover, the paper illustrates results of cooperative reinforcement learning algorithms using improved Nash Q-learning by Fault Pair Algorithm for three shop agents for the period of one-year sale duration. Results obtained by two approaches—Nash Q-learning and improved Nash Q-learning by Fault Pair—are compared. An agent keeps Q-functions containing joint actions and carries out modifications depending on Nash equilibrium performance for the present Q-values. Paper discovers that the agents are intended to attain a joint best possible path with Nash Q-learning. The performance of both agents enhanced after using Fault pair Nash Q-learning.