Multi-Label Deep Blind Compressed Sensing for Low-Frequency Non-Intrusive Load Monitoring

Current non-intrusive load monitoring (NILM) algorithms are reasonably accurate when the sampling rate is high. However, in practical scenarios, when the sampling frequency is low, the performance of most algorithms deteriorates. This work proposes a solution for low-frequency NILM. We propose to modify the smart-meter such that it can transmit at low frequency using principles of compressed sensing (CS). From such CS samples, we propose to detect the state of the appliance by using a multi-label consistent version of deep blind compressed sensing. Comparison with existing techniques shows that our proposed approach yields considerably better results.