SATISFy: Towards a Self-Learning Analyzer for Time Series Forecasting in Self-Improving Systems

摘要

Self-adaptive systems can adapt their managed resources to reflect changes in their environment or the resources themselves. However, sometimes these systems cannot handle situations due to uncertainty. Self-improvement enables the adaptation of the decision logic of such systems for coping with new situations. Proactive analysis predicts the need for self-improvement as well as reduces the delay for self-adaptation. However, implementing proactive analysis is a complex task which requires developers to analyze different algorithms and parameter combinations for finding the best fitting setting for the given data. This paper addresses this issue by presenting a model for a self-learning analyzer for proactive reasoning based on time series forecasting which can support self-improvement at runtime. We present a prototype implementation of such an analyzer and evaluate its performance for traffic prediction in an adaptive traffic management system.