To improve the accuracy of electricity price forecasting and the stability of forecasting models, a short-term electricity price forecasting method based on improved VMD-PSO-CNN-LSTM is proposed. Firstly, after studying the correlation between variational mode decomposition(VMD) and the influencing factors of electricity prices, and introducing the maximum information coefficient, a parameter optimization model for VMD is constructed. Secondly, convolutional neural networks(CNN) and long short-term memory(LSTM) neural networks are used to predict the modal components obtained by VMD decomposition. As for the convolution in CNN, a extraction structure with multi-scale convolution feature is constructed, on the basis of the depth-wise separable convolution combined with the time law of electricity prices. Particle swarm optimization algorithm is then used to optimize parameters including the number of CNN convolutional layers, the number of CNN convolutional neurons, the number of LSTM hidden layers, LSTM memory time, and the number of fully connected layers, so as to improve the prediction accuracy and stability of the model. Finally, the analysis and prediction of the day-ahead electricity prices in the Australian electricity market are carried out and compared with the algorithm. The results show that the proposed algorithm has higher accuracy and better stability.