Sistem peringatan Dini Tubrukan Kapal Berbasis Hybrid Convulational Neural Network - Gated Recurrent Unit dan Collision Risk Index di Alur Pelayaran Barat Surabaya

Ramadhan, Raphael Sandy (2026) Sistem peringatan Dini Tubrukan Kapal Berbasis Hybrid Convulational Neural Network - Gated Recurrent Unit dan Collision Risk Index di Alur Pelayaran Barat Surabaya. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5019221003_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Keselamatan navigasi di wilayah perairan terbatas (restricted waters) dengan tingkat kepadatan lalu lintas yang tinggi, seperti Alur Pelayaran Barat Surabaya (APBS), menghadapi tantangan kompleksitas manuver yang tidak dapat diatasi oleh metode konvensional. Kondisi ini menuntut adanya transformasi dari sistem reaktif menjadi proaktif. Penelitian ini mengusulkan rancang bangun Sistem Peringatan Dini (Early Warning System) tubrukan kapal menggunakan pemodelan prediksi trajektori Hybrid Convolutional Neural Network - Gated Recurrent Unit (CNN-GRU). Arsitektur ini memanfaatkan CNN untuk ekstraksi fitur spasial dan GRU untuk merekam dependensi temporal pergerakan kapal secara akurat.Model dilatih menggunakan 1.785 sampel data historis Automatic Identification System (AIS) di APBS yang telah dibersihkan dan diinterpolasi pada interval 1 menit. Algoritma ini memproses 8 fitur spasio-kinematik historis dengan jendela waktu T=30 untuk memprediksi 5 dimensi target lintasan kapal secara simultan pada τ =30 langkah waktu ke depan. Keluaran prediksi tersebut kemudian diintegrasikan ke dalam perhitungan analitik Collision Risk Index (CRI). Untuk menjamin keandalan evaluasi risiko di alur sempit, sistem menerapkan Dynamic Elliptical Domain (DED) yang menyesuaikan batas parameter bahaya (DCPA dan TCPA) berdasarkan dimensi fisik dan kecepatan kapal.Hasil pengujian membuktikan ketangguhan arsitektur Hybrid CNN-GRU dengan Mean Absolute Error (MAE) sebesar 0.024 dan Root Mean Square Error (RMSE) sebesar 0.041. Evaluasi sistem peringatan klasifikasi risiko menunjukkan nilai Akurasi 94.5%, Presisi 92.8%, Recall 95.1%, dan F1-Score 93.9%. Algoritma kecerdasan buatan ini berhasil diimplementasikan secara penuh ke dalam perangkat lunak antarmuka web real-time berbasis WebSocket, yang menyajikan visualisasi lintasan masa depan dan alarm bahaya untuk membantu operator Vessel Traffic Service (VTS) dalam mencegah potensi tubrukan maritim.
===================================================================================================================================
Navigational safety in restricted waters with high traffic density, such as the West Surabaya Sailing Channel (APBS), faces maneuvering complexity challenges that cannot be overcome by conventional methods. This condition demands a transformation from a reactive to a proactive system. This study proposes the design and development of a ship collision Early Warning System using a Hybrid Convolutional Neural Network - Gated Recurrent Unit (CNN-GRU) trajectory prediction model. This architecture utilizes CNN for spatial feature extraction and GRU to accurately capture the temporal dependencies of ship movements.The model is trained using 1,785 historical Automatic Identification System (AIS) data samples in APBS, which have been cleaned and interpolated at 1-minute intervals. The algorithm processes 8 historical spatio-kinematic features with a time window of T=30 to simultaneously predict 5 target dimensions of the ship's trajectory at τ =30time steps ahead. The prediction output is then integrated into the analytical calculation of the Collision Risk Index (CRI). To ensure the reliability of risk evaluation in narrow channels, the system applies a Dynamic Elliptical Domain (DED) that adjusts the hazard parameter thresholds (DCPA and TCPA) based on the physical dimensions and speed of the ships.The test results prove the robustness of the Hybrid CNN-GRU architecture, with a Mean Absolute Error (MAE) of 0.024 and a Root Mean Square Error (RMSE) of 0.041. The evaluation of the risk classification warning system recorded highly reliable performance, with an Accuracy of 94.5%, Precision of 92.8%, Recall of 95.1%, and an F1-Score of 93.9%. This artificial intelligence algorithm was successfully implemented into a real-time WebSocket-based web dashboard software, presenting future trajectory visualizations and hazard alarms to assist Vessel Traffic Service (VTS) operators in preventing potential maritime collisions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Automatic Identification System, Collision Risk Index, Dynamic Elliptical Domain, Hybrid CNN-GRU, Sistem Peringatan Dini, Tubrukan Kapal, Vessel Traffic Service.
Subjects:	V Naval Science > VK > VK200 Merchant marine--Safety measures V Naval Science > VK > VK555 Navigation. V Naval Science > VK > VK570 Optimum ship routing. V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering > VM471 Ships--Electric equipment
Divisions:	Faculty of Marine Technology (MARTECH) > Marine Engineering > 36202-(S1) Undergraduate Thesis
Depositing User:	Raphael Sandy Ramadhan
Date Deposited:	03 Aug 2026 04:36
Last Modified:	03 Aug 2026 04:36
URI:	http://repository.its.ac.id/id/eprint/141721

Actions (login required)

View Item