Sistem transportasi cerdas dapat diterapkan dengan salah satu bagiannya yaitu model prediksi lalu lintas untuk mengoptimalkan arus kendaraan sehingga dapat mengurangi kemacetan. Di sisi lain, telah banyak penelitian-penelitian sebelumnya yang berfokus pada pembangunan model prediksi lalu lintas menggunakan metode gabungan yang mempelajari fitur spasial dan temporal secara serentak melalui pendekatan model berbasis graf sedangkan penelitian di Indonesia masih menggunakan metode pembelajaran mesin sederhana. Oleh karena itu, dibangun model prediksi arus lalu lintas dengan dataset yang telah diimputasi oleh model prediktif Graph Imputation Neural Network (GINN) menggunakan framework Spatio-Temporal Graph Neural Network (STGNN). Model yang diusulkan terdiri dari dilated spatio-temporal synchronous graph convolutional module dan dilated temporal convolutional networks untuk menangkap dinamika spasial-temporal baik secara global dan lokal maupun jangka pendek dan jangka panjang, serta dynamic graph learning untuk mempelajari karakteristik dinamis spatial-temporal secara lokal. Graf dinamis yang dihasilkan dapat menyesuaikan pola data lalu lintas dengan baik serta meningkatkan kinerja hasil prediksi. Eksperimen yang dilakukan pada tiga dataset yang telah dibangun dan satu real-world dataset membuktikan bahwa model yang diusulkan dapat melampaui kinerja model state-of-the-art dan metode imputasi yang menggunakan model prediktif mampu meningkatkan kinerja model. Abstract Intelligent transportation systems can be implemented with one of their components, namely traffic prediction models, to optimize vehicle flow and reduce congestion. On the other hand, many previous studies have focused on developing traffic prediction models using a combined method that simultaneously studies spatial and temporal features through a graph-based model approach, while research in Indonesia still uses simple machine learning methods. Therefore, a traffic flow prediction model was developed using a dataset that had been imputed by Graph Imputation Neural Network (GINN) predictive model using Spatio-Temporal Graph Neural Network (STGNN) framework. The proposed model consists of a dilated spatio-temporal synchronous graph convolutional module and dilated temporal convolutional networks to capture spatio-temporal dynamics both globally and locally as well as in the short and long term, as well as dynamic graph learning to learn local spatio-temporal dynamic characteristics. The resulting dynamic graph can adapt traffic data patterns well and improve prediction performance. Experiments conducted on three constructed datasets and one real-world dataset prove that the proposed model can surpass the performance of state-of-the-art models and that imputation methods using predictive models can improve model performance.
English
Sistem transportasi cerdas dapat diterapkan dengan salah satu bagiannya yaitu model prediksi lalu lintas untuk mengoptimalkan arus kendaraan sehingga dapat mengurangi kemacetan. Di sisi lain, telah banyak penelitian-penelitian sebelumnya yang berfokus pada pembangunan model prediksi lalu lintas menggunakan metode gabungan yang mempelajari fitur spasial dan temporal secara serentak melalui pendekatan model berbasis graf sedangkan penelitian di Indonesia masih menggunakan metode pembelajaran mesin sederhana. Oleh karena itu, dibangun model prediksi arus lalu lintas dengan dataset yang telah diimputasi oleh model prediktif Graph Imputation Neural Network (GINN) menggunakan framework Spatio-Temporal Graph Neural Network (STGNN). Model yang diusulkan terdiri dari dilated spatio-temporal synchronous graph convolutional module dan dilated temporal convolutional networks untuk menangkap dinamika spasial-temporal baik secara global dan lokal maupun jangka pendek dan jangka panjang, serta dynamic graph learning untuk mempelajari karakteristik dinamis spatial-temporal secara lokal. Graf dinamis yang dihasilkan dapat menyesuaikan pola data lalu lintas dengan baik serta meningkatkan kinerja hasil prediksi. Eksperimen yang dilakukan pada tiga dataset yang telah dibangun dan satu real-world dataset membuktikan bahwa model yang diusulkan dapat melampaui kinerja model state-of-the-art dan metode imputasi yang menggunakan model prediktif mampu meningkatkan kinerja model. Abstract Intelligent transportation systems can be implemented with one of their components, namely traffic prediction models, to optimize vehicle flow and reduce congestion. On the other hand, many previous studies have focused on developing traffic prediction models using a combined method that simultaneously studies spatial and temporal features through a graph-based model approach, while research in Indonesia still uses simple machine learning methods. Therefore, a traffic flow prediction model was developed using a dataset that had been imputed by Graph Imputation Neural Network (GINN) predictive model using Spatio-Temporal Graph Neural Network (STGNN) framework. The proposed model consists of a dilated spatio-temporal synchronous graph convolutional module and dilated temporal convolutional networks to capture spatio-temporal dynamics both globally and locally as well as in the short and long term, as well as dynamic graph learning to learn local spatio-temporal dynamic characteristics. The resulting dynamic graph can adapt traffic data patterns well and improve prediction performance. Experiments conducted on three constructed datasets and one real-world dataset prove that the proposed model can surpass the performance of state-of-the-art models and that imputation methods using predictive models can improve model performance.
Affiliations come from OpenAlex; journal OAI-PMH metadata carries none. Authors without an OpenAlex ID are matched by name, which can merge or split people.