Order Forecast of Cigarette Distribution Center

The thesis forecasts order of Cigarette Distribution Center by period, brand, district, from several dimensions. The thesis uses different models and analyses forecast results. When forecasting year's order amount, the mean accuracy of Logarithm Regression Model is highest, gets 98.45%. When forecasting month's order amount of cigarette, the thesis uses Genetic Algorithm (GA) to optimize BP neural networks, and overcomes the shortcomings that Neural Networks apt to be trapped in local optimum when searching values of weights. The thesis also uses Regression Models, Grey Model and Self- adaptive Secondary Exponent Smooth Model, The mean accuracy of Linear Regression Model whose effect of forecasting is better is 96.9%. Considering the practical work period of Cigarette Distribution Center and the seasonal influences, when forecasting day's order amount, the thesis introduces PROPERTY_TAG to revise the forecast, improve the accuracy of forecast and easy to realize programming. GA Optimized BP Neural Network Model gets the expected accuracy which is 98%. The mean accuracy of Proportion Model Based-on Lunar Calendar, Regional Trend Model and ARMA Model are 90.87%, 95.61%, 96.61% respectively. When forecasting week's order amount of Cigarette, the accuracy of Accumulation of Mean Order Amount per Day Based-on PROPERTY TAG is 94.15% and cost time is 14s. The thesis develops Order Forecast Web System of Cigarette Distribution Center using Struts framework and makes it possible to share demand information for whole tobacco supply chain.