Forecasting method of e-commerce cargo sales based on ARIMA-BP model

In order to prevent merchants from occupying a large amount of cash flow due to unreasonable replenishment and the waste of resources and loss of profits caused by goods in short supply, a reasonable forecast of sales volume is particularly important. This article used the ARIMA-BP nonlinear combination model to predict the sale of merchants' goods. Two separate prediction models, the BP neural network and the ARIMA, were used to predict the sales volume in the next 5 days, and then a mean square error model was established to weight the fitting and predictive results of the two single predictions. The weight obtained is brought into the BP neural network model for training, and one is obtained The dynamic weight model is used to predict the dynamic weight of two single prediction models in the next five days. Finally, the test data was substituted into the model, and the normalized error is 0.05363. The research shows that the model can adapt to the problem of forecasting the sales volume of e-commerce goods, more stable, and the error of the combined prediction is much smaller than a single prediction.