Increasing the Accuracy and Speed of Universal Non-Intrusive Load Monitoring (UNILM) Using a Novel Real- Time Steady-State Block Filter

Non-intrusive load monitoring (NILM) is a research field focused on developing algorithms that can accurately track constituent electrical loads in a system using only the aggregate signal alone (i.e., smart meter). It is widely understood that having a clean signal free of noise and transient behaviour, whether for event-based or state-based methods, can lead to more accurate solutions that will eventually solve the NILM problem. We propose a fast and highly reliable method for producing a block-like representation of signals. Using the same data and disaggregation technique, we compare our algorithm with a recent similar effort and show significant improvements in accuracy (98 % vs. 94 % tracked energy over three appliances) and run-time (143ms vs. 891s). Application of our method to raw mains power data shows it can generalize to more complex cases.