Research On Sales Forecast Model Of Vending Machine

The application of vending machines in big cities is becoming more and more common,and vending machines have different characteristics from general retail.The sales of beverage vending machines are instant consumption,and the sales volume fluctuates greatly,which will be affected by the external environment;and due to the equipment capacity of the vending machine,machine failure,and untimely replenishment,there are some abnormalities in the data.Based on the sales characteristics of the above-mentioned vending machines,based on the research of relevant domestic and foreign literature,from the perspective of sales forecasts,this article proposes a sales forecast model and algorithm based on beverage vending machines One-year real sample data is used for time series forecasting,and an applicable model for sales forecasting based on different granularities is proposed.For the problem of data abnormalities,this paper proposes a method of wavelet transform combined with dispersion coefficients to identify and process the outliers of the sales series.This method can effectively identify hidden abnormal points that do not conform to the sales trend and periodicity.For detected abnormalities Point uses corresponding processing methods to replace outliers;in the analysis of external environmental factors affecting sales,this paper uses scatter plots and Pearson coefficients to analyze the correlation between temperature and sales.The results show that there is a relatively high relationship between temperature and sales.Strong correlation,and statistical distribution confirms the periodicity of sales.In the choice of sales forecasting method,this article combines the periodicity of sales and the characteristics of holiday sales will be different from workdays.Taking the temperature factor into consideration,the hybrid model of Prophet?LSTM is compared with the single Prophet model in order to effectively improve The accuracy of the forecast.Aiming at the volatility of sales volume,this paper adopts a time series forecasting method based on clustering analysis,using DTW(Dynamic Time Warping)distance similarity measurement formula combined with hierarchical clustering method to cluster the sales volume sequence of vending machine equipment,and get For devices of the same category with similar sales trends,use the idea of principal component analysis to extract the dominant sales sequence,use it together with the sales sequence of a single device as the input of the model,and use the Prophet and Prophet?LSTM hybrid models to predict separately.It can solve the problem of instability in forecast accuracy caused by the large volatility of the sales volume of the vending machine.Through empirical analysis,this article uses different conditions and methods to predict the results of different granularities.The results show that for the single-machine prediction of vending machines with high volatility,the prediction accuracy of the Prophet?LSTM model based on cluster analysis is compared.It is higher than other methods,and for the overall forecast,the volatility of the sales series is weak,so the direct use of the Prophet?LSTM hybrid model with added temperature factors has higher prediction accuracy.