Research On Forecasting For Quantity Of Commodity Barcodes Registration Based On ARIMA And Neural Network

The quantity of commodity barcodes registration is an important measure of the manufacturing and sale of new products. Change of the quantity of commodity barcodes registration can reflect not only the performance of the manufacturing and sale of products directly, but also the activity of business. In this case, if we carry on the forecasting of the quantity of commodity barcodes registration, we can judge broadly the condition of economic development. Further, we can formulate management and control policies according to the forecasting data. Therefore, the forecasting of the quantity of commodity barcodes registration has important instruction significance.After studying and analyzing the methods to time series forecasting, ARIMA model and neural network model are chosen to forecast the quantity of commodity barcodes registration. In addition, these two models are combined in order to improve the accuracy of forecasting results. This paper has mainly studied the following contents:Firstly, ARIMA model is established according to the autocorrelation and partial autocorrelation of the quantity of commodity barcodes registration. However, the forecasting results based on this model is unsatisfactory.Secondly, because of the nonlinear characteristic of the quantity of commodity barcodes registration, BP neural network and RBF neural network are applied to establish forecasting models. In the modeling process, neural network learning algorithm is found to be slow in convergence and easy to fall into local minimum value, and adaptive learning rate and increasing momentum item are applied to improve these defects. Using these two neural network models to forecasting the quantity of commodity barcodes registration, the results indicate that BP neural network model and RBF neural network model both have good forecasting ability, and their forecasting accuracies are higher than that of ARIMA model.Finally, in order to improve the forecasting accuracy, ARIMA model and neural network model are combined to establish ARIMA-BP model and ARIMA-RBF model. On the basis of these two integrated models, ARIMA-BP-RBF model is established. This model use BP-RBF model to forecast the residual of ARIMA model, and then add the results with the forecasting results of ARIMA model. The forecasting results based on the above three integrated models indicated that the integrated models perform better than single models in forecasting. Moreover, the forecasting accuracy of ARIMA-BP-RBF is higher than that of ARIMA-BP model and ARIMA-RBF model.