Stochastic Rule-Based Learning Approach to Dynamic Task Scheduling Problems

Abstract

A decision-making method has been proposed for distributed dynamic task scheduling controllers that involves automatic acquisition of scheduling rules through learning. The proposed method consists of stochastic rule-based decision making and its learning modification by stochastic matrix learning automata (SLA) theory : The action of determining a machine to be allocated is performed independently at each of the distributed agents, where the stochastic rule that consists of a condition and multiple actions is applied and one action is selected according to the probability vectors. SLA theory is applied to the learning modification of these probability vectors, which is proven to maximize a given objective function under a certain environment. To ensure the effectiveness of this learning for dynamic task scheduling, computer simulations are conducted, the results of which show that the method can obtain the same performance as that of a conventional dispatching rule, and that under a highly stochastic manufacturing environment, it can obtain hither performance than that using the dispatching rule.