A Transformer Based Sales Prediction of Smart Container in New Retail Era

With the advent of the new retail era, the value of unmanned smart container is increasingly prominent. Fast and flexible self-service is favored by consumers. How to use accumulated historical sales data to predict sales in the future is an important part of smart container operation management. Reasonable sales prediction can not only reduce the inventory cost, but also reduce the shortage loss of the container. Based on the smart container sales data of Dalian Xiaode New Retail Co., Ltd., through detailed exploratory analysis in many aspects, this paper carries out the feature selection of sales prediction, and uses random forest, XGBoost, Transformer and other algorithms to predict sales. The experimental results show that the prediction accuracy of Transformer is better than traditional algorithms, whose MAPE is 14.67% lower than that of the worst one. Transformer can be well applied in the field of sales prediction of smart container. And in this experiment, compared with Transformer using sine and cosine functions for positional encoding, Transformer encoded by position index has better prediction performance and stronger stability.