Commodity sales volume prediction method

Abstract

A commodity sales volume prediction method comprises the following steps: a data acquisition step: acquiring historical sales volume data of commodities; a data processing step of processing the historical sales volume data: performing data exploration processing, replacing abnormal values and missing values, exploring the correlation between each element of the historical sales volume data and the commodity sales volume, and taking the high-correlation element as a feature; carrying out feature engineering processing, and constructing derivative features from the features; data aggregation processing: aggregating the historical sales volume data into training samples according to time granularity; a model construction step: constructing a fusion model to carry out short-period prediction, and adopting a time sequence model to carry out medium-and-long-period prediction; and a sales volume prediction step of inputting the training samples, the features and the derivative features into a fusion model or a time sequence model to obtain a prediction result of the commodity sales volume. According to the invention, the method can meet the prediction requirements of the commodity sales volume of retail places such as stores, warehouses, manufacturers and the like at the same time, can achieve the targeted prediction of different periods, and improves the prediction accuracy.

Description

technical field [0001] The invention relates to a method for predicting commodity sales. Background technique [0002] Store sales forecasting has always been one of the core issues that the retail industry needs to solve. Retailers often reserve a large amount of inventory in order to avoid loss of orders and market share decline due to untimely product delivery, which leads to capital occupation, and reasonable forecasts The results and application are conducive to improving the matching degree of supply and demand and store sales, reducing temporary transfers between stores, and reducing costs. [0003] On the other hand, sales forecast can also coordinate the process of goods from warehouse to store, help drive the planned development of production, procurement, and capital management, organically combine all links in the industrial chain, and form a good cooperation to win Maximize efficiency and effectiveness. [0004] However, in the existing forecasting of commodit...

AI Technical Summary

Problems solved by technology

[0004] However, in the existing forecasting of commodity sales, the forecasting model is only a neural network and ARIMA model (differential integrated moving average autoregressive model), so the forecasting model is relatively single, and it is not possible to consider the factors that affect commodity sales from many aspects

Moreover, since such a model is only good at judging trends and seasonality, the time granularity of forecasting results using existing methods can only reach a few weeks, which cannot meet the needs of short-period forecasting

Images