Near Real-Time Event Detection & Prediction Using Intelligent Software Agents

摘要

This report results from a contract tasking University of Surrey as follows: The Surrey Space Center primary developer of the Disaster Monitoring Constellation (DMC) a network of satellites that provides users global natural and man-made event monitoring, seeks to monitor space and/or terrestrial source data streams for identifying interest-event occurrences. For the purposes of this research an event is defined as a significant interest item that occurs at a particular time and location, such as an individual volcano eruption, a flood or a forest fire. During- and postevent detection can often be achieved through one of several change detection algorithms, however pre-event detection introduces an entirely different challenge. Successful pre-event detection involves comparing temporal data against unique impending event data patterns. More concisely, successful pre-event detection involves combining time series analysis with robust event pattern recognition. While domain-specific methodologies have garnered varying success levels a general approach for this complex task has yet to be found and therefore motivates this research effort. Significant progress across the range of research goals and objectives has been achieved. Preliminary analysis results using one and two channelled data suggest the method is capable of identifying complex event-related data patterns and perhaps even predicting significant events. These results strengthen our conviction the method warrants further research and investigation.