Peramalan Traffic Load Pada Jembatan Timbang Dengan Model Hybrid Convolutional Neural Network Dan Bidirectional Gated Recurrent Unit

Alifutri, Salwa Iqlima (2026) Peramalan Traffic Load Pada Jembatan Timbang Dengan Model Hybrid Convolutional Neural Network Dan Bidirectional Gated Recurrent Unit. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221098-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Jembatan timbang milik PT XYZ berperan penting sebagai penyedia jasa penimbangan komoditas dan barang yang keluar dari kapal maupun aktivitas logistik di sekitar area pelabuhan, sehingga ketepatan waktu pelaksanaan maintenance menjadi krusial agar tidak terjadi downtime yang dapat mengganggu kelancaran layanan. Fluktuasi volume lalu lintas truk yang tidak menentu menjadi tantangan tersendiri bagi manajemen operasional, karena dapat menyebabkan ketidakefisienan dalam penjadwalan maintenance. Tugas akhir ini mengusulkan model peramalan deret waktu berbasis pendekatan hybrid Convolutional Neural Network (CNN) dan Bidirectional Gated Recurrent Unit (BiGRU) untuk penjadwalan peramalan maintenance. Pada arsitektur ini, CNN mengekstraksi fitur dan pola lokal dari data historis, sementara BiGRU menangkap ketergantungan temporal dua arah untuk meningkatkan akurasi prediksi. Model dilatih menggunakan data historis volume lalu lintas truk pada jembatan timbang PT XYZ yang telah melalui proses diferensiasi untuk mengurangi fluktuasi nilai, kemudian dievaluasi melalui grid search sembilan hyperparameter pada tiga proporsi data latih-uji (60:40, 70:30, 80:20) dengan metrik Mean Absolute Percentage Error (MAPE). Hasil menunjukkan proporsi 80:20 memberikan performa terbaik pada ketiga arsitektur, dengan model hybrid CNN-BiGRU mencapai MAPE terendah sebesar 34,57%, mengungguli CNN tunggal (38,80%) dan BiGRU tunggal (38,39%). Akurasi model terbukti paling dipengaruhi oleh proporsi split data dan window size, sementara CNN unggul dari sisi efisiensi waktu komputasi karena strukturnya yang lebih sederhana.
========================================================================================================================================
The weighbridge facility owned by PT XYZ plays a critical role in providing weighing services for commodities and goods unloaded from ships as well as logistics activities in the port area, making the timeliness of maintenance execution crucial to prevent downtime that could disrupt service continuity. The unpredictable fluctuation in truck traffic volume poses a significant challenge for operational management, as it can lead to inefficiencies in maintenance scheduling. This final project proposes a time series forecasting model based on a hybrid approach combining Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BiGRU) to support maintenance scheduling forecasting. In this architecture, CNN extracts local features and patterns from historical data, while BiGRU captures bidirectional temporal dependencies to improve prediction accuracy. The model was trained using historical truck traffic volume data from PT XYZ's weighbridge, which had undergone a differencing process to reduce value fluctuations, and was then evaluated through a grid search of nine hyperparameters across three train-test split proportions (60:40, 70:30, 80:20) using the Mean Absolute Percentage Error (MAPE) metric. The results show that the 80:20 split proportion yielded the best performance across all three architectures, with the hybrid CNN-BiGRU model achieving the lowest MAPE of 34.57%, outperforming the standalone CNN (38.80%) and standalone BiGRU (38.39%). Model accuracy was found to be most influenced by the data split proportion and window size, while CNN demonstrated superior computational time efficiency owing to its simpler structure.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	BiGRU, CNN, Deep Learning, Forecasting, Jembatan Timbang, Weighbridge
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Salwa Iqlima Alifutri
Date Deposited:	29 Jul 2026 04:22
Last Modified:	29 Jul 2026 04:22
URI:	http://repository.its.ac.id/id/eprint/139490

Actions (login required)

View Item