전략적 경매를 위한 Temporal Fusion Transformer 모델 기반 경매 가격 다층 수평 예측

초록

This study focuses on predicting auction prices for agricultural products, which in South Korea are primarily traded through auctions. However, beginners face challenges such as price volatility, lack of information, and impulsive bidding risks. Previous studies have developed prediction models but were limited to single-step forecasts, which only identify the optimal price for the current day, thus hindering strategic bidding over multiple days. These studies also failed to fully account for price variations based on product characteristics. To address these issues, this study developed a prediction model based on the Temporal Fusion Transformer using data from agricultural and fishery trading platforms. The data was divided into observed variables, known variables, and static covariates and analyzed in a multi-layer horizon forecasting prediction model. The model developed in this study predicts auction prices for the next month, thereby providing overall trends and enabling strategic bidding.

키워드