Handling Uncertainty in Self-adaptive Software Using Selflearning Fuzzy Neural Network

Uncertainty has posed great challenges to the development and application of self-adaptive software ( SAS). To handle uncertainty underneath SAS, the technique of fuzzy control method has been employed to model and develop SASs. Practices prove that fuzzy logic is powerful to handle uncertainty, especially fuzzy uncertainty, within SAS. However, fuzzy control based SAS needs software developers to set fuzzy rules of the system, which is rather experience-dependent and heavily increases development burden of software engineers. To some extent, the effect of handling uncertainty depends on experiences of software engineers. Besides, fuzzy control based SAS realizes self-adaptation logic using fixed fuzzy rules, lacking the ability to adapt to large changes (e.g., scenario switches). In order to make up the above shortages of fuzzy control based SAS, we present the Fuzzy-Learning SAS, attempting to construct self-adaptation logic using self-learning fuzzy neural network. By incorporating the model of fuzzy neural network, Fuzzy-Learning models SAS with two feedback loops, i.e., the self-adaptation loop and the self-learning loop, enabling SASs with the ability of adapting to dynamic changes and the ability of automatically constructing self-adaptation logic. We have experimentally evaluated effectiveness and efficiency of Fuzzy-Learning SAS with a motivating example. The experiment results confirmed that Fuzzy-Learning SAS can improve the effect of handling uncertainty and alleviate the development burden of software engineers with ill knowledge of fuzzy control. Besides, Fuzzy-Learning SAS can adapt to large changes (e.g., scenario switches) with the self-learning ability.