In view of the large amount of gas consumption data and the low accuracy of the traditional machine learning model in processing large data, and the regular data in time, the unique memory ability of long short-term memory (LSTM) was used to predict the gas. In order to improve the accuracy of LSTM prediction model, empirical mode decomposition (EMD) algorithm was used to decompose the gas consumption data into several relatively stable intrinsic mode functions and a residual term r(t), a combined prediction model based on EMD-LSTM algorithm was proposed. The results show that compared with BP, EMD-BP and LSTM models, the gas consumption can be accurately predicted by using the method, which can provide reference for cost saving and gas allocation.