Non-intrusive Load Monitoring for Consistent Shape Loads Based on Convolutional Neural Network

Abstract

As the key method for demand-side management in power grid, non-intrusive load monitoring (NILM) keep up to the power consumption of various users in real time and provides data to support the formulation of relevant power policies. In order to achieve accurate resident load monitoring, this paper proposes a NILM architecture focus on consistent shape loads (CSL). Loads in CSL meet the following conditions: 1) current waveform images of different load individuals in the same type are highly similar. 2) different types of load waveform images are different in shapes which are distinguishable. Besides, a non-intrusive load monitoring method based on convolutional neural network (CNN) to identify CSL load is proposed and carried out on actual users. Power consumption data of CSL with different operating environments is taken as training samples. The outcome of our experiment shows the effectiveness of the method in accurately distinguishing CSL and high-precision identification which reaches 97.06%. The method ensures the real-time performance and accuracy of load monitoring.