In order to ensure the provision of high-quality power quality and the smooth operation of the power system, aiming at the problem of low prediction accuracy of power quality indices, this paper proposes an online prediction method of power quality indices based on VMD-LSTM residual analysis. Firstly, the VMD-LSTM is used to simulate and predict the normal power quality signal; then the predicted signal data is different from the actual power quality signal data to obtain the residual signal, and the residual signal is input into the one-dimensional convolutional neural network for training, and finally the power quality indices prediction model is obtained. Finally, taking voltage flicker as an example, simulation experiments compared with other prediction models verify the better effectiveness and accuracy of the proposed method.