Using the Energy Error to improve the performance of Unsupervised Approach in Non-Intrusive Load Monitoring

Hader Aguiar Dias Azzini¹; Luiz Carlos Pereira da Silva¹

¹ UNICAMP

doi:10.20906/CPS/SBSE2016-0175

Resumo

This paper proposes the use of the percentage of Energy Error as feedback to improve the performance for an unsupervised approach for Non-Intrusive Load Monitoring. In this approach, different values of thresholds for the composition of clusters in ISODATA algorithm are tested, looking for the value which minimizes the percentage of Energy Error. This approach reduces the percentage of Total Energy Error from around 25% to around 15%, reaching the level of supervised machine learning approaches. The use of unsupervised approach could improve the commercialization of Non-Intrusive Load Monitoring technology because it reduces, or even eliminates, the training phase.

Palavras-chave: nonintrusive load disaggregation; nonintrusive load monitoring; smart meter; unsupervised machine learning