A fast multiple appliance detection algorithm for non-intrusive load monitoring

Non-intrusive Load Monitoring (NILM) is a hallmark of monitoring technologies. It provides appliance-level energy usage feedback with minimal sensor deployments. How-ever, existing NILM systems are not designed with embedded systems (e.g. Smart Meters) in mind. In this paper, we consider a Bayesian solution which efficiently computes the Log-Likelihood Ratio (LLR) of an appliance´s on/off state, then combines it with historical estimates to yield a new estimate for improved accuracy. The detection of multiple appliances is achieved through an iterative method which also deals with unidentified appliances. With minimal multiplication and division, the algorithm is computationally lightweight and can easily be implemented in embedded systems using low-power processors. Despite the simplicity, results show promising disaggregation performance.