Research on non-intrusive load identification based on VMD-LSTM

Non-intrusive load identification technology is important for optimising the energy structure, improving power efficiency and saving resources by installing smart meters at the main power supply connector to collect and analyse the user's electricity information in real time. For the different characteristics of household load changes in the power system, a non-intrusive load identification approach based on VMD-LSTM is proposed. The acquired data were first preprocessed by decomposing the signal into K IMF components using the variational modal decomposition (VMD). Secondly, the relevance coefficient of each component with the normalized signal are calculated, and the three components with the largest correlation coefficients are selected to form the load feature. Then, it is input into the trained LSTM neural network for identification. Finally, the experimental results indicate that the recognition rate of the proposed means in the public data set is more than 96 %, and the accuracy of load recognition is significantly improved.