An Enhanced Signal Processing Toolbox for Electrical Energy Monitoring

Abstract

A nonintrusive load monitor (NILM) aims to perform power system analysis with a minimally invasive sensor profile. A wealth of literature exists for load identification and energy disaggregation under ideal, healthy conditions. However, a significant value proposition of nonintrusive load monitoring comes from fault detection and diagnostics. Early detection of electromechanical faults aids safety, reduces energy waste, and saves money. However, load identification and energy disaggregation are complicated by faulty or time-varying load operation profiles. This thesis extends previous thesis work by the author that addresses this issue. A new, “multistream” feature extraction approach to nonintrusive power monitoring is presented. This approach enables targeted electrical data analysis on non-stationary electrical systems.