Application Of GA-BP Neural Network In Cigarette Sales Forecast

The forecast for cigarette sales is a very complex non-linear forecasting system that predicts tobacco demand over a certain period of time and is an important way for companies to understand market demand.The neural network algorithm has a strong nonlinear expression ability,which makes it very suitable for dealing with complex,unstable,random features,and nonlinear time series prediction problems.Therefore,this thesis studies and optimizes the application of GA-BP neural network in cigarette sales forecasting.In this thesis,the establishment and forecasting process of the tobacco sales forecasting model and the evaluation criteria of the forecasting model are analyzed from the macro level.Through the research of multiple hidden layer BP neural network model,generalized regression neural network model and genetic algorithm,different cigarette sales forecast models are constructed and optimized.Among them,in view of the lack of original sales data,the weighted average filling method was proposed in the experiment to improve the sales data and ensure the integrity of the sales data.Aiming at the uncertainty of the training parameter setting of the neural network model and the local optimal problem that the network often appears in the training process,an adaptive learning rate setting method is proposed in the experiment to optimize the multiple hidden layer BP neural network,so that the network In the training process,it can tend to converge at a relatively stable speed.At the same time,the genetic algorithm was used to optimize the initialization parameters of BP neural network,and the GA-BP neural network prediction model was constructed to predict the cigarette sales.In the experimental simulation,based on the sales volume of Jiaozilan in the Sichuan region from October 2006 to November 2017,the prediction of cigarette sales was realized by using neural network related algorithms.The experiment proves that the cigarette sales model based on GA-BP neural network combines the adaptive learning rate setting method in the training process,which has higher training efficiency than BP neural network and GRNN neural network model,and also improves the model's prediction accuracy,whichmakes the model's predicted mean absolute error lower than 0.72.Experiments prove that GA-BP neural network can well establish the nonlinear relationship mapping between different sales values in time series,and the neural network model based on genetic algorithm optimization has a good generalization.