Banca de DEFESA: MAURICIO MASSAKI ASANO

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : MAURICIO MASSAKI ASANO
DATE: 28/03/2025
TIME: 15:30
LOCAL: APRESENTAÇÃO HÍBRIDA (sala A1 12/6 - Anexo SG 12 e sala virtual do Microsoft Teams)
TITLE:

Urban Traffic Forecasting in the Federal District: Application of Traffic Forecasting Neural Networks Using Spatio-Temporal Data

KEY WORDS:

Traffic forecasting, deep learning, Graph-CNN-LSTM, urban mobility, spatiotemporal analysis

PAGES: 122
BIG AREA: Engenharias
AREA: Engenharia de Transportes
SUMMARY:

This study addresses short-term traffic forecasting using a Graph-CNN-LSTM-based model, which integrates spatial and temporal information to improve prediction accuracy. The data were collected by the Department of Highways of the Federal District (DER-DF) and include traffic volume records at 15-minute intervals from March to August 2024. During preprocessing, steps such as normalization, outlier treatment, and the construction of a dynamic adjacency matrix were performed, allowing for the modeling of interdependencies between road segments. The predictive model was evaluated using metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), demonstrating that prediction performance decreases as the forecast horizon increases. The average MAE values across all road segments were 13.70 (15 min), 14.30 (30 min), 19.39 (45 min), and 21.78 (60 min). The results indicate that high-traffic periods show greater variability in prediction errors, whereas lower-traffic hours allow for more accurate forecasts. The analysis of segment correlations revealed strong spatial interdependencies in traffic patterns, highlighting the importance of graph-based modeling. This study contributes to the enhancement of urban mobility management by providing a predictive model that can assist in strategic decision-making to optimize traffic flow and reduce congestion.

COMMITTEE MEMBERS:
Externo à Instituição - ALEXANDRE DE BARROS BARRETO - GMU
Interno - 2487190 - ALAN RICARDO DA SILVA
Presidente - 1220587 - LI WEIGANG
Interno - 1552603 - PASTOR WILLY GONZALES TACO