Sistem Prediksi Tinggi Alun Menggunakan Metode Long Short Term Memory Untuk Mitigasi Risiko Keselamatan Kapal Sandar

Winata, Kinara Lingga (2026) Sistem Prediksi Tinggi Alun Menggunakan Metode Long Short Term Memory Untuk Mitigasi Risiko Keselamatan Kapal Sandar. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040211033-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Keselamatan operasional sandar kapal di PT Pertamina Energy Terminal sangat dipengaruhi oleh fluktuasi tinggi alun. Gelombang alun ekstrem yang tidak terprediksi sering memicu beban tarik berlebih yang berisiko memutus tali mooring, sedangkan sistem pemantauan yang ada saat ini masih bersifat reaktif. Untuk mengatasi permasalahan tersebut, penelitian ini merancang sistem prediksi tinggi alun berbasis algoritma deep learning Long Short-Term Memory LSTM yang diintegrasikan ke dalam Instrumen Peringatan Dini dengan proyeksi 3 jam ke depan. Data historis dari sensor LiDAR WeMon dipra-pemrosesan melalui resampling temporal 1 jam, pembersihan outlier statistik Interquartile Range (IQR), interpolasi PCHIP terbatas limit=12 jam, dan oversampling progresif. Hasil pengujian menunjukkan bahwa Model D LSTM Peak Hunter dengan Single Layer 128 units dan Dense Layer 32 nodes menjadi arsitektur paling superior untuk proyeksi t+3. Model ini berhasil mencatatkan nilai Mean Absolute Error (MAE) sebesar 20,7" cm" , Root Mean Squared Error (RMSE) 25" cm" , dan durasi pelatihan 104,2" detik" , mengungguli arsitektur GRU dengan hasil MAE 20,8" cm" , RMSE 25,5" cm" terutama dalam melacak puncak gelombang ekstrem >150" cm" . Integrasi luaran model pada antarmuka EWS berbasis zona keselamatan PIANC memberikan lead time operasional selama 3 jam, sehingga otoritas pelabuhan dapat mengeksekusi tindakan preventif untuk mencegah insiden terputusnya tali mooring secara proaktif. =========================================================================================================================================
Port operational safety during ship berthing at PT Pertamina Energy Terminal is highly influenced by swell height dynamics. Unpredictable extreme swells often trigger excessive tensile loads that risk snapping mooring lines, whereas current monitoring systems remain reactive. To address these issues, this study designs a swell height prediction system based on the Long Short-Term Memory (LSTM) deep learning algorithm integrated into an Early Warning System (EWS) with a 3-hour lead time forecasting horizon. Historical data from the WeMon LiDAR sensor were preprocessed using 1-hour temporal resampling, Interquartile Range (IQR) statistical outlier filtering, limited PCHIP interpolation limit=12 hours, and progressive oversampling. Experimental results indicate that Model D LSTM Peak Hunter, configured with a single layer of 128 units and a 32-node dense layer, was the most superior architecture for t+3 forecasting. This model achieved a Mean Absolute Error (MAE) of 20.7" cm" , a Root Mean Squared Error (RMSE) of 25" cm" , and a training time of 104.2" seconds" , outperforming the GRU model with MAE 20.8" cm" , RMSE 25.5" cm" , particularly in tracking extreme wave peaks >150" cm" . Integrating the model output into the EWS interface based on PIANC safety thresholds provides a 3-hour operational lead time, enabling port authorities to proactively execute preventive measures to prevent mooring line failure incidents.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Prediksi Tinggi Alun, Long Short-Term Memory, MAE, RMSE, Instrumen Peringatan Dini, Keselamatan Kapal Sandar. Swell Height Prediction, Long Short-Term Memory, MAE, RMSE, Early Warning System, Berthing Ship Safety.
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Kinara Lingga Winata
Date Deposited:	10 Aug 2026 01:03
Last Modified:	10 Aug 2026 01:03
URI:	http://repository.its.ac.id/id/eprint/144235

Actions (login required)

View Item