As natural gas accounts for an increasing proportion of energy consumption, how to accurately predict the future natural gas consumption is of great significance to the rational planning of natural gas. For this problem,a short?term natural gas load forecasting model based on wavelet transform and deep learning was proposed. First,the collected natural gas load was decomposed by using different wavelets , and then normalized it.Secondly, the data wes trained and predictd by using the deep learning algorithm Long Short?Term Memory (LSTM); then the predicted data was separately integrated by using wavelet reconstruction.Finally, the average absolute percentage error, average absolute error and root mean square error were used as evaluation indicators to evaluate the prediction results of different wavelets, and the optimal order and number of layers of the optimal wavelet were calculated.The examples show that the 22nd?order 6th layer of Fk wavelet transforms has higher prediction accuracy than other wavelets transforms and direct use of LSTM for prediction.