Abstract: Energy disaggregation plays an important role in the rational use and management of energy within a whole building. However, the current energy disaggregation methods encounter the problems of low accuracy to classify and long computational time to choose appropriate parameters. To solve this difficulty, this paper proposes a novel energy disaggregation method based multi-output extreme learning machine (MO-ELM) to analysis low-frequency monitoring data gathered by meters distributed in a building. The MO-ELM whose parameters of feature mapping need not be tuned and can be fixed once randomly generated requires fewer optimization constraints with the objective function and results in simpler implementation compared to SVM. The evaluation results using a dataset of power traces collected in real-world home setting shows that proposed method have a satisfied classification accuracy, training speed and generalization performance and less computational time.