Pulp Sales Forecasts Based On Neural Network For The K Company

The K company is an international trade company which is engaged in import sales of X brand pulp.With the continuous development of domestic pulp market and an increasing pressure of competition of the company.It is a worth research topic that how we can use the high quality sales forecast to make a decision for company.Firstly,this article analysis the current more classic forecasting technique:expert forecasting method、time series method、regression analysis model、grey system modeling and neural network model.By comparing the relative merits of all kinds of forecasting method,to determine use the neural network prediction model to solve the problems of K company.Then the article introduces the working principle of neural network,algorithm design and two kinds of neural network model which is applied widely:BPNN,RBFNN principle and RBFNN branch of GRNN.Last according to the actual of K company,choosing the GRNN and the BPNN to making forecasting model.Through the various factors analysis of pulp' s sales market.Screen out the most obvious impact factors on sale:the sale price of new order on next month,the sale price of pulp on this month,the sales volume of pulp on this month,different times,he rate between RMB and dollar,pulp' s the total number of import during this month,average unit price of import and quantity demand of local major dents.Inputting this eight factors will output the sale volume of pulp in next month in GRNN and BP network.After that need to check the other parameters,for example,smooth factor、layers of network、number of neurons in each layer、training algorithm etc.All of them build up the GRNN forecasting model and BPNN forecasting for pulp sale.In the end,this topic is adopted to MATLAB software simulation build the BPNN model and GRNN model.Using the different region sales data which from Jan.2010 to Jun.2015 as the sample to train the network.Meanwhile,using the sales data which from July 2015 to Dec.2015 to check the forecasting results.After that,input test data to observe the forecast results in network.Compared the two neural network forecasting models,the result shows that both of them are good forecast:GRNN network to predict the MAPE is 12.52%;BP network to predict the MAPE is 16.19%.After get the results,this article analyze all of the forecasting results according to the different region.Also discussion about why caused the error.Draw a conclusion,due to the complex of the sales process,two kinds of neural network is good in overall forecast effect although there is still a more error in part of the test points.So we still think this application experiment is successful.The sales forecasting of GRNN and BPNN can partly solve the issue of K company.They can provide a good assistance and support on supply chain management rely on their strong nonlinear mapping ability.