Short term daily electricity forecasting is helpful to construct power market and improve power supply service level. With the help of the accurate identification of different users’ characteristics, a short-term daily electricity demand prediction modeling of industrial users based on doubly decomposition is proposed. Firstly, statistical tools are used to identify the seasonality, temperature correlation and holiday correlation of users; Then, according to the recognition results, the seasonal-trend decomposition procedure based on regression is customized; After that, the electricity demand is decomposed by doubly decomposition combined with the regression cycle method and the ensemble empirical mode decomposition; Next, based on the characteristics of the obtained components, long/short term memory networks, weighted least squares support vector machine and convolutional neural network is used for prediction; Finally, the prediction results of each component are superimposed, and the holiday prediction results are adjusted by using the probability distribution to obtain the final prediction value. Through the analysis of practical examples, the proposed method has high prediction accuracy.