An Improved Forecasting Model Using Long Short-Term Memory And Support Vector Regression And Its Applications In The Tobacco Industry

Accurate forecasting of the tobacco industry's supply can guide the production and marketing planning of cigarettes,bringing more economic benefits to society.However,the prediction accuracy of existing tobacco supply-side currently forecasting methods needs to be improved.In order to obtain a high-precision forecast method applicable to such problems,this thesis built a hybrid forecasting model based on Fruit Fly Optimization Algorithm(FOA),Support Vector Regression(SVR)model,Singular Spectrum Analysis(SSA)and Long Short-Term Memory(LSTM)networks.To investigate the prediction accuracy of the proposed SSA-LSTM-SVR model,the experiments of forecasting short-term cigarette sales and the price of white cardboard were conducted.In the proposed SSA-LSTM-SVR model,the SSA is used to decompose an original series into a trend component and other detail components in the first stage.In the second stage,the LSTM is used to forecast the trend component,while the FOASVR is used to forecast the several detail series.In the third stage,prediction reconstruction of each deconstructed sequence is performed to complete the output of the final prediction.In addition,the excellent parameters of the SVR are optimized by the FOA algorithm with an adaptive search step size introduced.In order to evaluate the prediction performance of SSA-LSTM-SVR model,it is compared with the LSTM model,SVR model,ARIMA model,SSA-SVR model,SSA-LSTM model and other prediction models in two experiments.Finally,it is confirmed that the SSA-LSTM-SVR model can obtain more stable and accurate results than the other eight comparison models,which is suitable for the tobacco industry's supply forecasting problem.