A fully unsupervised non-intrusive load monitoring framework

Monitoring an individual electrical load's energy usage is of great significance in energy-efficient buildings as it underlies the sophisticated load control and energy optimization strategies. Non-intrusive load monitoring (NILM) provides an economical tool to access per-load power consumption without deploying fine-grained, large-scale smart meters. However, existing NILM approaches require training data to be collected by sub-metering individual appliances as well as the prior knowledge about the number of appliances attached to the meter, which are expensive or unlikely to obtain in practice. In this paper, we propose a fully unsupervised NILM framework based on Non-parametric Factorial Hidden Markov Models, in which per-load power consumptions are disaggregated from the composite signal with minimum prerequisite. We develop an efficient inference algorithm to detect the number of appliances from data and disaggregate the power signal simultaneously. We also propose a criterion, Generalized State Prediction Accuracy, to properly evaluate the overall performance for methods targeting at both appliance number detection and load disaggregation. We evaluate our framework by comparing against other multi-tasking schemes, and the results show that our framework compares favorably to prior work in both disaggregation accuracy and computational overhead.