Research On The Combination Forecasting Model Of Sales Of Dishes Based On Time Series And Neural Network

Sales forecast of dishes is a important link in the refined operation of catering enterprises.Accurate sales forecast of dishes is of great significance.It can help the catering enterprises reduce the waste of raw materials,control the procurement cost,increase the profit and enhance the market competitiveness of enterprises.At present,the vast majority of catering enterprises still rely on personal experience to make qualitative forecast on the sales of dishes,which has strong subjectivity and uncertainty.Based on the real transaction data in the backstage database of an online ordering system of a catering enterprise,this paper researches the forecast of the daily sales of dishes,and proposes an ARIMA-BP combined forecast model based on ARIMA time series model and BP neural network model.The main contents of this paper are as follows:(1)Based on the forecast goal,this paper collects and analyzes the data of dish sales.Through the research and analysis of the original data,we can find that the law presented by the sales data of dishes is relatively complex.These data contain linear components,showing a certain trend and cycle volatility,and it has obvious nonlinear characteristics due to the influence of weather,holidays,preferential activities and other external factors.(2)The ARIMA time series model and BP neural network model are respectively constructed to make short-term forecast for the sales of dishes sing the collected sales data.The results show that the two models have good prediction effect.But ARIMA time series model can only fit the linear characteristics of the data,and is not sensitive to the new trend of the data;BP neural network model can capture the non-linear characteristics and the change trend of sales data,but there are still some deviations at some time points.(3)In order to further improve the forecast accuracy,this paper combines the advantages of ARIMA time series model and BP neural network model in linear and non-linear feature fitting,aiming at the minimization of forecast error,using the method of error reciprocal weighted average,giving different weights to ARIMA time series model and BP neural network model respectively,and constructs ARIMA-BP combined prediction model.Through the comparative analysis of the forecast results,it is proved that the combined forecast model proposed in this paper has better forecast effect on the short-term forecast of the daily sales of dishes,and has higher forecast accuracy than the single forecast model.It has certain practical and guiding significance for the refined operation of catering enterprises.