In this research, we propose a multivariate time series data prediction model to forecast agricultural commodity price trends. We introduce two prominent deep learning networks for multivariate time series data prediction.
One is the stacked Long Short-Term Memory (LSTM) network, known for its exceptional performance in time series data prediction, and the other is the graph-based Spectral Temporal Graph Neural Network (StemGNN).
The multi-time series data comprises wholesale prices of agricultural commodities and weather data, and our models are all designed to capture the correlations between agricultural commodity prices and weather variables.
Experimental results demonstrate that, in the context of multivariate time series data prediction using agricultural commodity and weather data, the graph-based network, StemGNN, outperformed the commonly used LSTMbased network. Additionally, we conducted experiments by adjusting the input and output lengths of the data and analyzed the results. Consequently, we propose optimal data lengths for each agricultural commodity, considering the unique characteristics of agricultural data, to enhance prediction performance. Finally, in the StemGNN network, we extract correlation graphs trained using self-attention techniques to analyze the weather
factors that significantly impact each agricultural commodity.

* 줌회의실: 개별 공지