Real-Time Event-Driven Learning in Highly Volatile Systems: A Case for Embedded Machine Learning for SCADA Systems

Extracting key system parameters and their impact on state transition is a necessity for knowledge data engineering. In Decision Support Systems, the quest yet more efficient faster methods of sensitivity analysis (SA) feature extraction in complex volatile systems persists. A new improved event tracking methodology, fastTracker, real-time SA large scale proposed this paper. The main fastTracker its high-frequency analytics using meager computational cost. It suitable processing prioritization embedded systems, Internet Things (IoT), distributed computing (e.g. lEdge computing) applications. presented algorithm’s underpinning rationale driven; objective to correctly succinctly quantify observable changes (output) with respect input variables. To demonstrate performance was deployed Supervisory control acquisition (SCADA) from real cement industry. has been verified by experts industrial application. Its compared other event-based techniques. comparison revealed savings 98.8% time per index 20% memory usage when EventTracker, closest rival. methodology accurate 80.9% than an entropy-based method. application recommended reinforced learning and/or formulating indicators raw data.