Commodity sales prediction method and system

Abstract

The invention discloses a commodity sales prediction method and system and belongs to the technical field of deep learning. The method comprises the steps of: obtaining the historical sales data of acommodity; preprocessing the historical sales data to obtain preprocessed sales data; performing first feature extraction on the preprocessed sales data to obtain first feature sales data; performingtraining according to the first feature sales data and the preprocessed sales data to obtain a sales initial prediction model; performing second feature extraction on the first feature sales data andthe preprocessed sales data to obtain second feature sales data; and training the sales initial prediction model according to the first feature sales data, the preprocessed sales data and the second feature sales data to obtain a commodity sales prediction model so as to predict the sales of the commodity. According to the commodity sales prediction model, a wavenet network and an LSTM network arefused; and the average absolute percentage error of the commodity sales model can reach five percent under the condition of sufficient historical sales data.

Description

technical field [0001] The present invention relates to the technical field of deep learning, in particular to a method and system for predicting commodity sales. Background technique [0002] Commodity sales forecast is based on the full consideration of various influencing factors in the future, based on historical sales and changes in market demand for products, scientific predictions and speculations on changes in product sales in a certain period of time in the future. [0003] In the prior art, the forecast of commodity sales is mainly based on three types of models. One is the traditional statistical model (such as arima, garch model, etc.), the disadvantage of which is that this type of model can only predict a single commodity in a single store. The number of given multiple models to predict, very troublesome. Moreover, most of the sales forecasts of this type of model for a certain day are based on the sales data of the previous few days, that is, the model has n...

AI Technical Summary

Problems solved by technology

However, compared to LSTM, the disadvantage of wavenet is that it cannot use a two-way network like the LSTM model, so that the model has the ability to extract future information (such as: the weather in the next few days (this can be roughly known through weather forecasts) ), store sales plans for the next few days, and holiday arrangements for the next few days (holidays, holidays caused by special weather warnings), etc.

Images