Research On Clothing Sales Forecast Based On Prophet-LSTM Combination Model

Generating product-level sales forecasts is an important prerequisite for the healthy development of fast-selling clothing brand companies.Due to the special highly dynamic and unstable business environment,the life cycle of a fast-selling clothing survival market is short,so there are stricter requirements for production distribution plans.In the fierce market competition,blindly reducing production or replenishing without adapting to the actual sales of the product will cause the output to not match the actual sales situation,resulting in a shortage of goods or a backlog of inventory,which will increase costs and affect the company development of.Through sales forecasting,on the basis of effective results,the production plan can be adjusted according to the dynamic changes of data,and dynamic and flexible contracting tasks can be performed according to the predicted results,which plays an important and non-negligible role in guiding business decisions of enterprises.In order to reduce the probability of supply-demand mismatch between offline stores across the country,this article analyzes the sales demand at different stages in different regions.The main work done to build a high-precision prediction model is as follows:1.Analyze the characteristics of the original data,filter related impact variables,pre-process the original data,and multi-dimensionally merge the purer data records after processing according to business needs.2.Decomposition the weighted terms for the existing Prophet time series model,analyze and optimize each component,and use a change point selection algorithm based on the combination of TSTKS algorithm and sliding window to mark the position of the change point in the time series and fuse them.Go to the core component trend item to optimize the performance of the Prophet model.3.Construct an LSTM model with appropriate parameters,and choose an appropriate weighting algorithm through comparison to find the best weighting factor.Combined with the optimized Prophet model,it draws on the respective advantages of the time series model and the neural network model to build a Prophet-based model.-LSTM optimized combination model.4.Use scientific indicators to verify the model's performance through experiments.The experimental results are as follows: Under the same experimental conditions,the MAE value of the traditional time series ARIMA model is 13.284 and the RMSE value is 21.294;the MAPE value of the Prophet model optimized by change point selection is 2.325 and the RMSE value is 3.642;The optimized combination model has a MAE value of 1.603 and an RMSE value of 2.518.The verification of experimental data shows that the traditional time series model cannot meet the accuracy requirements in this business environment,and the combined model based on Prophet-LSTM optimization reduces the probability of the single model being susceptible to external influences and the risk of mutation.To some extent,the accuracy of prediction is also improved.