Appliance-Level Anomaly Detection in Nonintrusive Load Monitoring via Power Consumption-Based Feature Analysis

In the majority of current supervised nonintrusive load monitoring (NILM) techniques the characteristics of the training and test datasets are assumed to be constant. However, in practice, a new appliance may be added to the set of electrical appliances, an appliance may be replaced by a new one, the efficiency of an appliance may change during the time, or a fault may occur. All these issues of appliances cause abnormal consumption in total signal and not detecting them reduces the NILM accuracy. In this sense, submetering is widely used to address the above challenge. However, it leads to additional costs for consumers as it requires additional meters, disturbs consumers’ comfort and privacy, and is not scalable. In this paper, the non-intrusive load monitoring approach is utilized to detect the above anomalies based on two main factors: 1) the power distribution and 2) the participation index of appliances. Since the proposed method is nonintrusive it is cost-efficient and accommodates the consumer’s privacy and comfort. Furthermore, detecting all the above appliance anomalies not only increases the NILM accuracy but also reduces energy waste and avoids appliance breakdown. The experimental results prove that the proposed method yields outstanding performance in detecting the aforementioned anomalies.