Automated classification of power signals

Abstract

The Non-Intrusive Load Monitor (NILM) is a device that utilizes voltage and current measurements to monitor an entire system from a single reference point. The NILM and associated software convert the V/I signal to spectral power envelopes that can be searched to determine when a transient occurs. The identification of this signal can then be determined by an expert classifier and a series of these classifications can be used to diagnose system failures or improper operation. Current NILM research conducted at Massachusetts Institute of Technology's Laboratory for Electromagnetic and Electronic Systems (LEES) is exploring the application and expansion of NILM technology for the use of monitoring shipboard systems. This thesis presents the ginzu application that implements a detect-classify-verify loop that locates the indexes of transients, identifies them using a decision-tree based expert classifier, and then generates a summary event file containing relevant information. The ginzu application provides a command-line interface between streaming preprocessed power data (PREP) and an included graphical user interface. This software was developed using thousands of hours of archived data from the Coast Guard Cutters ESCANABA (WMEC-907) and SENECA (WMEC-906). A validation of software effectiveness was conducted as the software was installed onboard ESCANABA.