Qualitative reasoning assisted empirical system identification

Abstract

The design and operation of modern IT-based systems especially cyber-physical systems (CPS), need modelbased approaches due to their complexity. However, the limited faithfulness of pure analytic models with speculative layouts prohibits their use in complex systems. Complexity necessitates empirical system identification from observations. Exploratory data analysis (EDA) is a main approach to observation-based system identification. EDA combines visual methods and summary statistics for initial data analysis. In traditional EDA, there is a gap between the thinking of the expert and the logic of the analysis method. In general, everyday and engineering thinking use a qualitative approach. Qualitative abstraction represents the system with a discrete model of the granularity of individual operation domains. This way, qualitative models avoid the complexity problems by identifying and focusing on the most important features observed. Qualitative modeling is a gradual, iterative process extracting more and more abstract details of the observations. As the individual model fragments merge into the evolving system model, each step needs validation and verification (V&V). The use of a discrete formalism supports V&V by logic reasoning. The paper presents a prototype implementation covering data profiling, continuous-qualitative abstraction, and Answer Set Programming (ASP) for logic reasoning.