Novel Computational-Index as a Representative Feature for Non-Intrusive Load Monitoring

This paper presents the development of novel extraction feature used for load disaggregation from aggregated demand profile. Non-intrusive load monitoring tackles the issue of appliances identification inside a residential building. New extraction features used for load identification are based on power, V-I mutual locus and wave-shape features for specific appliances which offers better or generally comparable results in comparison to popular methods. A novel dynamic computational index is developed to achieve a robust feature extraction using V-I mutual locus. Power patterns are used for identification of multi-state devices. Specific screening and classification of extraction features are so developed to get better load disaggregation and lesser computational time. Multi-label classifications approach has been used for the same. The methodology is compared based on precision, classification accuracy, and robustness using experimental data.