Research on Non-intrusive Load Monitoring Based on Seq2point Model

In order to improve the training efficiency of load disaggregation and identification, this paper will use a deep neural network architecture based on sequence-to-point ( Seq2point ) to study the non-intrusive load disaggregation and identification method. Convolutional neural network is used to extract load characteristics and efficiently mine the temporal relationship between input and output. The model is compared with the mainstream model on the public dataset UK-dale, and the recall rate, precision rate, Matthews correlation coefficient(MCC) and Mean absolute error(MAE) of power estimation are used as evaluation indexes to evaluate the model. The results show that the model greatly shortens the training time while maintaining a low disaggregation error.