Abstract: Non-invasive load monitoring monitors the working status of each electric equipment by installing

an intelligent meter at the main power interface，aiming to collect and analyze users' power consumption

information in real time. According to different characteristics of household user load change in power system，a

non-invasive load monitoring method based on VMD-CNN was proposed. Firstly，the collected load signal was

preprocessed，and the original load power signal was decomposed into finite bandwidth subsequences with different frequencies by variational modal decomposition method. Secondly，the low-frequency signal features were extracted by Hilbert algorithm and input into convolutional neural network，after which different features of each modal component were learned through the automatic learning ability of the network. Finally，the full connection network was used to classify the electrical equipment，and the power consumption of each equipment was obtained.In order to verify this method，the UK-DALE data set was used for experimental verification，and to compared with different classification algorithms. The experimental results proved that the method is suitable for load monitoring，and the load monitoring accuracy reaches 0.98.

      Key words: non-invasive load monitoring（NILM）；power system；variational model decomposition（VMD）

algorithm；convolutional neural network（CNN）