Keynote - Model-driven Analytics with [email protected]: The Case of Cyber-Physical-Systems

Bits and bytes are governing an increasing number of areas in our lives and businesses. The exploration and simulation of what might happen and which action can be triggered is a fundamental part of intelligent systems such as smart grids, smart buildings, smart homes and any cyber-physical system. This new intelligence is supported by machine learning algorithms that, based on past data and runtime data, model the behavior of the system to predict its evolution. Recommendation systems, autonomous decision-support, prescriptive simulations have to be both scalable and highly accurate at runtime. It is paramount to develop new decision support services that should (at least partly) relieve the users from the overwhelming load of information and the growing number of decisions to be taken in time. In that perspective, model-driven engineering offers a bridge between the knowledge of experts who best know which data are relevant , and the monitoring and control of software components and sensors. The presentation is about how MDE, and specifically [email protected], may become an enabler for designing and deploying easily domain-specific, scalable analytics for heterogeneous sources of timed data. Some problems still have to be solved and I will introduce some of them. Cyber-physical systems continuously analyze their surrounding environment and internal state, which together we refer to as the context of a system, in order to adapt itself to varying conditions. To yield accurate predictions, such systems not only rely on single numerical values, but also need structured data models aggregated from different sensors. Therefore, building appropriate context representations is of key importance. Over the past few years the [email protected] paradigm has shown the potential of models to be used not only at design-time but also at runtime to represent the context of cyber-physical systems, to monitor their runtime behavior and reason about it, and to react to state changes. However, reasoning about such contexts is a complex and time critical activity that needs to leverage near real-time analytics together with big data methods to quickly process the massive amount of data measured by these systems. Current modeling techniques do not allow to face all needed features for reasoning, such as distribution, large-scale and near real-time response time. In this talk I present two concepts that might push the limits of [email protected] for near-real time analytics a little further: 1) stream-based, distributed models and 2) historized models. I will present our results based on a …