Ship Berthing-Unberthing Path Planning Generated by Bi-Long Short Term Memory (LSTM) and Bi-Gated Recurrent Unit (GRU) with Risk Index

Wandani, Dhinakara Yumna (2026) Ship Berthing-Unberthing Path Planning Generated by Bi-Long Short Term Memory (LSTM) and Bi-Gated Recurrent Unit (GRU) with Risk Index. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5019221023-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Navigasi di perairan terbatas dan dangkal seperti Alur Pelayaran Barat Surabaya(APBS) menghadirkan risiko kandas yang signifikan akibat kondisi hidrografi dinamis dan batasan olah gerak kapal. Penelitian ini bertujuan untuk mengembangkan sistem bangkitan rute berbasis UKC (Under Keel Clearance) dan prediksi probabilitas kapal kandas untuk meningkatkan keselamatan navigasi. Penelitian ini menggunakan arsitektur deep learning, yaitu Bidirectional Long Short-Term Memory (Bi-LSTM) dan Bidirectional Gated Recurrent Unit(Bi-GRU). Model dilatih menggunakan data lintasan historis Automatic Identification System(AIS) yang diintegrasikan dengan batimetri, penyesuaian sarat dinamis (squat effect), dan fluktuasi pasang surut. Rute yang dihasilkan dievaluasi berdasarkan kepatuhan spasial terhadap alur dan efisiensi waypoint, sementara probabilitas kandas dianalisis pada variasi kecepatan operasi 1 hingga 5 knot. Untuk menjamin kelayakan operasional, model divalidasi oleh tiga pakar maritim menggunakan Item-Level Content Validity Index (I-CVI). Hasil penelitian menunjukkan bahwa kedua model secara efektif mengoptimalkan jalur navigasi, berhasil mereduksi kepadatan data lintasan AIS yang fluktuatif menjadi 16 waypoint esensial yang mulus dan disiplin pada koridor aman. Dalam memprediksi risiko kandas, model Bi-GRU menunjukkan performa komputasi yang sedikit lebih superior dan optimis dibandingkan BiLSTM. Risiko kandas menunjukkan korelasi terbalik non-linear dengan kecepatan kapal. Berlayar dengan kecepatan 5 knot meminimalkan risiko (10⁻⁷) berkat stabilitas hidrodinamik yang optimal, sedangkan penurunan kecepatan menjadi 1 knot secara eksponensial meningkatkan probabilitas kandas (hingga 9,5 × 10⁻⁶) akibat lamanya waktu tempuh yang lebih lama. Lebih lanjut, validasi pakar menghasilkan Scale-Level CVI (S-CVI) sebesar 0,82, yang mengonfirmasi validitas tinggi sistem ini sebagai decision support tool navigasi untuk fase transit. Oleh karena itu, penelitian ini merekomendasikan penggunaan sistem untuk perencanaan pelayaran di alur terbatas, dengan pengecualian pada manuver sandar (berthing) jarak dekat yang sangat bergantung pada data real-time dan bantuan kapal tunda.
=================================================================================================================================
Navigating confined and shallow waterways such as the Surabaya West Access Channel(SWAC) presents significant grounding risks due to dynamic hydrographic conditions and vessel maneuverability constraints. This research aims to develop a UKC (Under Keel Clearance) route generation and grounding probability prediction system to enhance navigational safety. The study utilized deep learning architectures, specifically Bidirectional Long Short-Term Memory (Bi-LSTM) and Bidirectional Gated Recurrent Unit (Bi-GRU). The models were trained on historical Automatic Identification System (AIS) trajectory data integrated with bathymetry chart, dynamic draft adjustments (squat effect), and tidal fluctuations. The generated routes were evaluated based on spatial adherence to fairways and waypoint efficiency, while grounding probability was analyzed across operating speeds ranging from 1 to 5 knots. To ensure operational feasibility, the model was validated by three maritime experts using the Item-Level Content Validity Index (I-CVI). The results demonstrate that both models effectively optimized navigational paths, successfully compressing dense, fluctuating AIS trajectories into 16 essential, smoothed waypoints that strictly adhered to the safe corridor. In predicting grounding risks, the Bi-GRU model exhibited slightly superior and more computationally efficient performance compared to the Bi-LSTM. Grounding risk exhibits a non-linear inverse correlation with vessel speed. Operating at 5 knots minimizes risk (10⁻⁷) through optimal hydrodynamic stability, whereas reducing speed to 1 knot exponentially spikes grounding probability (up to 9.5 x 10⁻⁶) due to longer transit duration. Furthermore, the expert validation yielded a Scale-Level CVI (S-CVI) of 0.82, confirming the system's high validity as a navigational decision support tool for transit phases. Consequently, the research recommends the system for voyage planning in restricted fairways, while explicitly excluding close-quarters berthing maneuvers due to real-time latency and external tugboat dependencies.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bi-LSTM, Bi-GRU, Data AIS, Olah Gerak, Probabilitas Kandas AIS Data, Bi-GRU, Bi-LSTM, Maneuver, Risk Index
Subjects:	T Technology > T Technology (General) > T58.62 Decision support systems V Naval Science > VK > VK555 Navigation. V Naval Science > VK > VK570 Optimum ship routing.
Divisions:	Faculty of Marine Technology (MARTECH) > Marine Engineering > 36202-(S1) Undergraduate Thesis
Depositing User:	Dhinakara Yumna Wandani
Date Deposited:	03 Aug 2026 02:45
Last Modified:	03 Aug 2026 02:45
URI:	http://repository.its.ac.id/id/eprint/141919

Actions (login required)

View Item