Pengembangan Sistem Optimasi Pemuatan Logistik Berbasis Deep Reinforcement Learning di Perusahaan X

Pramudya, Yanuar Eka (2026) Pengembangan Sistem Optimasi Pemuatan Logistik Berbasis Deep Reinforcement Learning di Perusahaan X. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221049-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Proses pemuatan kontainer secara manual sering menyebabkan rendahnya utilisasi ruang, terbentuknya ruang kosong yang tidak dapat dimanfaatkan kembali, dan meningkatnya risiko ketidakstabilan susunan barang, terutama pada skenario kedatangan barang secara bertahap (semi-online). Untuk mengatasi permasalahan tersebut, tugas akhir ini mengembangkan sistem optimasi Online Three-Dimensional Container Loading Problem (3D-CLP) menggunakan pendekatan Hierarchical Deep Reinforcement Learning (H-DRL) yang mengintegrasikan holding buffer, validasi kestabilan berbasis Local Bottom Convex Polygon (LBCP), dan mekanisme repacking. Hasil pengujian menunjukkan bahwa model menghasilkan utilization sebesar 40,5%, success rate sebesar 52,5%, dan stability rate sebesar 88,7%, serta menurunkan jumlah deadlock menjadi 216.077 kejadian. Dibandingkan BLF-Only dan A3C-Only, model meningkatkan utilisasi ruang masing-masing sebesar 18,4% dan 9,5%. Pada pengujian zero-shot transfer menggunakan RS Dataset, model memperoleh utilisasi sebesar 68,8%, lebih tinggi dibandingkan A3C tanpa buffer sebesar 59,6%. Hasil tersebut menunjukkan bahwa kombinasi H-DRL, holding buffer, dan validasi kestabilan fisik mampu menghasilkan sistem optimasi pemuatan logistik yang lebih adaptif, stabil, dan memiliki kemampuan generalisasi yang baik pada skenario pemuatan kontainer semi-online.
===================================================================================================================================
Manual container loading often results in low space utilization, fragmented empty spaces, and unstable cargo arrangements, especially in semi-online environments where items arrive sequentially. To address these challenges, this study proposes a logistics loading optimization system for the Online Three-Dimensional Container Loading Problem (3D-CLP) using a Hierarchical Deep Reinforcement Learning (H-DRL) approach integrated with a holding buffer, physical stability validation, and repacking mechanisms. Experimental results show that the proposed model achieves a space utilization of 40.5%, a success rate of 52.5%, and a stability rate of 88.7%, while reducing the total number of deadlocks to 216,077 occurrences. The model improves space utilization by 18.4% compared with BLF-Only and by 9.5% compared with A3C-Only. In zero-shot transfer experiments on the RS Dataset, the proposed model achieves a utilization of 68.8%, outperforming the A3C model without a holding buffer, which achieves 59.6%. These findings demonstrate that integrating H-DRL, holding buffer mechanisms, and physical stability constraints can improve loading quality and provide an adaptive and stable optimization framework for semi-online container loading scenarios

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Reinforcement Learning, Heuristic Tree Search, 3D Container Loading Problem, Repacking Strategy, Semi-online
Subjects:	T Technology > T Technology (General) > T57.6 Operations research--Mathematics. Goal programming T Technology > T Technology (General) > T57.62 Simulation T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Yanuar Eka Pramudya
Date Deposited:	23 Jul 2026 14:49
Last Modified:	23 Jul 2026 14:49
URI:	http://repository.its.ac.id/id/eprint/137243

Actions (login required)

View Item