紀要論文

同時進行する時系列データからのTransformerによる予測モデルの構築

Building a Forecasting Model with Transformer from Concurrent Time Series Data

jpn

24321192

https://doi.org/10.15002/00025265

法政大学大学院情報科学研究科

法政大学大学院紀要. 情報科学研究科編

17

1

6

2022-03-24

Version of Record

In this paper, we propose a method to forecast the future of time series data using Transformer. The forecast can be widely utilized in various areas, such as product demand prediction and pandemic analysis of infectious diseases. Many methods have been proposed to analyze the time series data such as ARIMA and neural network. Especially, the Transformers has been attracting researcher’s attention to analyze the time series data in recent years. In order to precisely forecast the future, a large amount of data is required, but in some cases, such as retail stores, enough data is not always available. Therefore, in this research, we propose a model that combines two types of Transformers ,such as time and space Transformers, to supplement prediction from the relationship between the data. The Time Transformer makes predictions based on the sales performance of the target product, and The Space Transformer infers the relationship between the target product and the others. In experiments, we showed that the proposed model improved the prediction performance over the ARIMA and the conventional Transformers in the task of predicting the volume of the sales one week ahead.

Article

AA12746425