A Study on Highway Traffic Volume Forecasting Based on the SARIMA-GCN-LSTM Model

To effectively address the rapid increase in highway traffic load, this paper designs and proposes a SARIMA-GCN-LSTM hybrid model for short-term forecasting of highway traffic volume time series data. The model aims to reduce traffic congestion, prevent accidents, enhance driving experience, and promote high-quality urban development. The Seasonal Auto regressive Integrated Moving Average (SARIMA) model is used to model and fit the time series data, capturing the forecasted trend and seasonal components. The Graph Convolutional Network (GCN) model is employed to model the adjacency relationships within the time series data, extracting feature vectors from graph data, and then feeding the concatenated predictions from both models into the Long Short-Term Memory (LSTM) model. This process allows the LSTM to learn the dynamic characteristics and long-term dependencies of the time series data, resulting in the final forecast. Using real highway gantry data from the author's home province A as the experimental subject, the study demonstrates that the SARIMA-GCN-LSTM hybrid forecasting model outperforms individual models in predictive performance, indicating that the hybrid model is effective for short-term highway traffic volume forecasting.