Sistem Pemulihan Layanan Listrik Secara Real-time Menggunakan Rekonfigurasi Jaring Distribusi Listrik Pulau Bawean Berbasis Deep Reinforcement Learning

Fabian, Omar (2026) Sistem Pemulihan Layanan Listrik Secara Real-time Menggunakan Rekonfigurasi Jaring Distribusi Listrik Pulau Bawean Berbasis Deep Reinforcement Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221077-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Jaringan distribusi radial pada sistem kelistrikan Pulau Bawean rentan mengalami pemadaman meluas ketika terjadi gangguan saluran, karena area beban sehat di sisi hilir dapat ikut terputus setelah proses isolasi gangguan. Pemulihan layanan secara manual atau berbasis aturan tetap memiliki keterbatasan dalam menghadapi banyak kombinasi lokasi gangguan, variasi pembebanan harian, dan kebutuhan pengambilan keputusan yang cepat. Oleh karena itu, tugas akhir ini mengusulkan sistem pemulihan layanan listrik berbasis Deep Reinforcement Learning (DRL) melalui rekonfigurasi jaringan distribusi. Permasalahan restorasi diformulasikan sebagai Markov Decision Process (MDP) dengan state berupa status switch, jam operasi, status gangguan, dan rasio beban tidak tersuplai. Aksi restorasi dilakukan melalui operasi remote-controlled switch, sedangkan reward disusun berdasarkan tiga objektif, yaitu memaksimalkan beban tersuplai, meminimalkan rugi daya aktif, dan mengurangi penalti tegangan. Model DQN, Dueling DQN, dan FQF-DQN diintegrasikan dengan DIgSILENT PowerFactory dan Python, kemudian diuji pada modified IEEE 69-bus serta sistem distribusi Pulau Bawean selama 24 jam pada beberapa skenario gangguan. Hasil pengujian menunjukkan bahwa metode DRL mampu meningkatkan beban tersuplai dibandingkan kondisi pascaisolasi tanpa restorasi. Pada sistem IEEE 69-bus, FQF-DQN menunjukkan performa restorasi yang konsisten dengan beban tersuplai mendekati penuh pada beberapa skenario. Pada sistem Bawean, Dueling DQN dan FQF-DQN mampu mempertahankan beban tersuplai hingga lebih dari 98% pada skenario kritis. Waktu pengujian yang berada pada orde detik menunjukkan bahwa framework yang dikembangkan berpotensi mendukung pengambilan keputusan restorasi secara waktu nyata pada lingkungan simulasi.
======================================================================================================================================
Radial distribution networks in the Bawean Island power system are vulnerable to widespread outages when line faults occur, since healthy downstream load areas may become disconnected after fault isolation. Manual or rule-based service restoration has limitations in handling numerous fault-location combinations, daily load variations, and fast decision-making requirements. Therefore, this final project proposes a Deep Reinforcement Learning (DRL) based power service restoration system through distribution network reconfiguration. The restoration problem is formulated as a Markov Decision Process, where the state consists of switch status, operating hour, fault condition, and unserved load ratio. Restoration actions are performed through remote-controlled switching operations, while the reward function is designed using three objectives: maximizing served load, minimizing active power loss, and reducing voltage penalty. DQN, Dueling DQN, and FQF-DQN models are integrated with DIgSILENT PowerFactory and Python, then evaluated on the modified IEEE 69-bus system and the Bawean Island distribution system under 24-hour operating conditions and several fault scenarios. The testing results show that DRL-based methods improve the served load compared with the post-isolation condition without restoration. In the IEEE 69-bus system, FQF-DQN demonstrates consistent restoration performance with nearly full load recovery in several scenarios. In the Bawean system, Dueling DQN and FQF-DQN maintain more than 98% served load under critical fault scenarios. The testing time, which remains in the order of seconds, indicates that the proposed framework has the potential to support real-time restoration decision-making in a simulation-based environment.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Pemulihan Layanan, Rekonfigurasi Jaringan, Deep Reinforcement Learning (DRL), Markov Decision Process (MDP), Jaringan Distribusi, Service Restoration, Network Reconfiguration, Deep Reinforcement Learning (DRL), Markov Decision Process (MDP), Distribution Network.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3030 Electric power distribution systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Omar Fabian
Date Deposited:	22 Jul 2026 08:30
Last Modified:	22 Jul 2026 08:30
URI:	http://repository.its.ac.id/id/eprint/136312

Actions (login required)

View Item