A non-intrusive load identification method based on convolution neural network

Load identification have shown significant performance gains in smart meters. Most existing load identification algorithms are based on electrical characteristics of steady or transient state, which are therefore limited by feature selection. In recent years, due to the strong self-learning capability of deep learning, it has been widely applied and has produced satisfactory results in many fields, but non-intrusive load monitoring (NILM) has seen few attempts. From the perspective of image processing, this paper uses the Convolution neural network (CNN) of deep learning to analyze the potential characteristics of the load current waveform. This work starts with building the current data set of different appliances' on and off states, and then the structure and parameter of CNN are designed based on current waveform features to identify the load. This method is simple to operate, and a series of experiment verify its feasibility and accuracy.