Content area
Full Text
Abstract - Self-adaptive systems can change their sub-goals and behaviors to achieve an ultimate goal in a changing environment. Existing approaches can be adapted in environments using pre-defined utility functions and human strategies; however, human designers cannot perfectly assume and predict all possible system environments during design time. We propose a new method of dynamic decision-making for real-time adaptive strategies through the following steps. First, we design a dynamic decision network (DDN) with environmental factors and goal model;. Second, we evaluate and predict goal satisfaction using the DDN. We further propose a dynamic reflection method that changes the model using real-time data. We applied the proposed method in Robocode and verified its effectiveness by comparing it with static decision-making.
Keywords: self-adaptive system, dynamic decision-making, dynamic decision network, reflection model
(ProQuest: ... denotes formulae omitted.)
1 Introduction
Self-adaptive systems (SASs) can change their own sub-goals and behavior without a human operator to achieve an ultimate goal in a variety of environments [1]. To change sub-goals and behavior, researchers have proposed goal-model-based SASs because they provide information for the evaluation of goal satisfaction and the formulation of rational decisions [1-2].
Existing decision-making techniques for goal-model-based SASs have pre-defined utility functions and the strategies depend on assumptions at the design time [3-9]. However, during the design time, system designers cannot perfectly assume and predict all possible system environments that the system will be deployed in; as a result, neither goal achievement nor proper adaptability can be guaranteed. To cope with this limitation, SASs require dynamic-decision-making that can consider the variety of runtime environments that can be known after deployment.
Among Artificial Intelligence (AI) techniques, the dynamic decision network (DDN) [10], which is used in our research, is a proper model for the making of dynamic decisions. It can design alternatives and outcome-utility values for all the alternatives using environmental information. Consequently, it can make the best possible decisions within changing environments.
In this paper, we propose a method of dynamic decision-making for the deployment of SASs in runtime environments that are not known prior to deployment. To represent the system goals and environment, we designed a DDN using the goal model and environmental information so that decision-making can occur during runtime. After system deployment, our DDN...