E-commerce merchants usually want to bear low inventory costs while ensuring that the goods meet the demand, in order to sell a large number of goods, merchants will also carry out a variety of promotional activities, so e-commerce merchants need to accurately predict the promotional shipments of goods. In order to accurately predict the promotional shipments of goods, taking the double eleven data as an example, this paper adopts ARIMA time series analysis model and LSTM long and short-term memory model to predict the shipments of June promotions; the accuracy rate of 1-wmape of ARIMA time series model is 0.752, and the accuracy rate of 1-wmape of LSTM is 0.834, and the accuracy rate of LSTM is higher, the results are predicted by LSTM model. As a result, one of the merchants' promotional sales in June predicted by using LSTM model is 18, 19, 15, 12, 15, 7, 16, 22, 11, 32, 22, 25, 18, 16, 16, 19, 20, 22, 15, 13. However, the degree of recognition between the ARIMA time series model and the LSTM model prediction results is 80%, which can prove that the LSTM model reliability and rationality of the LSTM model.