Cigarette Sales Forecast Based On Neural Network Research And Application Of The Model

At first this paper has basically analyzed the market of the tobacco industry, and analyzed several basic environmental factors influencing the cigarettes saleroom: income level, population, festivals and holidays, etc. , and endued that the factors influencing the cigarettes saleroom have many aspects which interacting on each other. Cigarette sale has the character of double trends of time series, namely whole trend change and season fluctuation. Data mining is to extract the hidden, implicit, valid, novel and interesting spatial or non-spatial patterns or rules from large-amount, incomplete, noisy, fuzzy, random, and practical data. This paper expatiates the course and basic implementation step of data mining, then, especially in the model of data mining, has introduces BP neural network, time series model, CBP network model, neural network model based on time series, and introduces their theoretical foundation and the principle.Under fully considering the time relativity characteristic of double trends time series of the predicted data source (cigarette saleroom of some Yunnan tobacco company), based on CBP neural network, we set up CCBP predict model. CCBP model is the product association of two sub models: LCBP and PCBP. LCBP model predicts the weight of trend on terms that have considered such environmental factors as the economic income level, population, age composition level and flow of floating population, etc.. According to the characteristic of time dependence, PCBP model, which based on time abatable methods, predicts the weight of fluctuation. Thus the CCBP predict model very properly depicts the data characteristic: trend increase and season fluctuate, which lies in the cigarette sales data. This model has overcome the shortcoming of single CBP model and other models limited in ability to express with double trend prediction; also separated the time dependence and environmental factor, and respectively treated the two weight of the double trend, this model have fully utilized the advantage of the CBP isotropism characteristic.Finally, in order to study and test the CCBP model, we take the past 6 years cigarette saleroom of Yunnan some region tobacco company as data sample, and set up APMSTS: the ANN predict model system of tobacco sales, which goes on real-time increment collect datum from this.company database, and take this regional yearbook data as the environment factor. After building and^training J3P,' CBP and CCBP model, the result indicates that the prediction result of CCBP model is superior to BP and CBP model. Prove that, in the prediction aspect of time series double trend, CCBP network model, which separated time dependence and environmental factor and respectively treat dual two weight of trend, can even more accord with the characteristic of cigarette sell which influenced by economic environmental factor and time dependence.