Nonintrusive Load Monitoring Using Support Vector Machine

This paper presents the development of non-intrusive load monitoring (NILM) to identify loads using the multi-output support vector machine (MOSVM). A supervised load monitoring method is applied to identify three types of loads that are typically used in commercial buildings such as fluorescent light, air conditioning and personal computers. The basic power parameter provided by the smart meter and other details of the extracted power parameters are considered in this paper. Effective power features are determined by selecting appropriate feature combinations and also, a new feature extraction technique, named ‘time-time’ transformation has been used in this study. A systematic selection of the power parameter is carried out, in this case, to find the best combination for comparison purposes. In the case of commercial smart meter usage for the end- user sector, which is the majority in the low sampling rate, an experiment and studies have been employed under the condition of real power measurement with a low sampling rate. The low sampling rate suitable for NILM is evaluated according to the specification of the commercial smart meter with three conditions of the sampling rate; 1 min, 10 min and 30 min. A set of validation data with random load activities was used to test the effectiveness of the developed NILM method. Further, the load classification technique of MOSVM was used to compare with other techniques such as naive Bayes and KNN to evaluate the performance of the proposed MOSVM for NILM. The results using the proposed MOSVM method showed the best result with an accuracy of 99.94 % in identifying the load. Therefore, based on the sampling rate studied, 1 min sampling showed the best results for the implementation of load monitoring compared to the other sampling rates for NILM.