Convolutional Gated Power Prediction Combined With Multiscale Multilabel Classification for Nonintrusive Load Monitoring

Load monitoring is an effective approach to reducing energy consumption in buildings. However, state identification when multiple devices which have different load properties are operating simultaneously is still a relatively new research area by nonintrusive load monitoring (NILM) algorithms. Besides, large-scale monitoring data requires consideration of the storage capacity of the device, making the development of lightweight models a significant challenge for NILM algorithms. This study proposed a NILM algorithmic for learning the load characteristics of each appliance from the total power sequence by the proposed multilevel extraction parallel convolutional gating load prediction model (MEPCG) in tandem with a multiscale multilabel classification model (MMCM). Proposed NILM algorithmic realize simultaneous capture of the running statuses of multiple appliances with diverse load profiles and to predict the operational statuses of multiple appliances in the future period in advance. This study validates the algorithm’s performance on three low-frequency public datasets. The results on the AMPds2 dataset show that the proposed MEPCG reduces the root-mean-square error in power series prediction experiments by 14.12% compared others model. The F1 scores of MMCM are 10.18%, 14.85%, and 16.59% higher than the optimal multilabel classification model on the AMPds2 dataset, REFIT dataset, and REDD dataset, respectively. The proposed multiscale multilabel classification model has 1/54 times the amount of model parameters of the optimal model, and the training time is decreased by 59.03% compared to other classification models.