Research On Commodity Sales Forecast Based On Residual Optimization Of ARIMA Model

Accurate commodity sales forecast increases the turnover rate for merchants's benign inventory stocking,formulates cost budget items,and formulates effective strategies according to the time and local conditions,thus providing theoretical basis for reasonable commodity sales strategies.However,the traditional methods of forecasting are not comprehensive and the actual effect is not ideal.The ARIMA model is used to fit the original product sales time series,and the main related information in the product sales time series is extracted by linear regression and other methods.The trend effect fitting polynomial and season fitting polynomial are subtracted or removed from the sales time series,and the residual error series is obtained.The nonlinear relationship information between the time series residual error and the original time series sales value in the stacked automatic encoder neural network is iteratively fitted,and the original observed value is changed by this information.Then new information items are added to the calculation formula in the autocorrelation function,and the prediction is changed by using accurate parameter coefficient calculations.Experimental comparison shows that the model optimized by residual error can reflect the sales rules more accurately and comprehensively,and improve the accuracy of commodity sales prediction.In this paper,there are two ways to predict commodity sales by ARIMA model,which are:based on the traditional ARIMA model,carry out conventional smoothing data,determine the order and parameters of the model and then predict;Research on improving prediction accuracy by optimizing autocorrelation function based on time series residual neural network.In the research of optimizing autocorrelation function and increasing prediction accuracy based on stack self-coding neural network,the time series residuals are preprocessed and adjusted by empirical mode decomposition.On this basis,the neural network fitting experiment is carried out by using the nonlinear relationship between different kinds of residuals and the original time series sales value,and the stack self-coding neural network is added by using the characteristics of sliding window to extract the residual value of the nonlinear relationship information.Added to the calculation process of the sequence values of the autocorrelation function calculation formula,different parameter coefficients are obtained,and the final prediction experiment is carried out.Under the same data set,the performance of the optimized model is compared and analyzed.The experimental results show that the ARIMA model based on time series residual optimization has higher prediction accuracy than the single ARIMA method,and eliminates the influence of the single ARIMA method on the experimental data which cannot be nonlinear fitted and unreliable.