Abstract:

The environment of selfadaptive systems is often uncertain, and the changes are difficult to predict. To develop such complex selfadaptive software systems has become a great challenge in the domain of software engineering. Reinforcement learning is an important branch of machine learning. A reinforcement learning system can learn the optimal mapping policy from the states of environment to the actions by means of trailanderror. Aiming to deal with the uncertainty of environments, this paper combines the agent technology and the reinforcement learning technology together, and proposes an adaptive mechanism based on reinforcement learning and the corresponding approach to construct complex selfadaptive systems that can adapt to the changes of uncertain environments. A case is illustrated to validate the effectiveness of the proposed mechanism and approach.

Key words: reinforcement learning;selfadaptive system;selfadaptive multiagent system