Klasfikasi Karakteristik Gangguan Pada Jaringan Transmisi Berdasarkan Disturbance Fault Recorder (DFR) Dengan Convolutional Neural Network - Long Short Term Memory

Januar, Hafizh Tri (2026) Klasfikasi Karakteristik Gangguan Pada Jaringan Transmisi Berdasarkan Disturbance Fault Recorder (DFR) Dengan Convolutional Neural Network - Long Short Term Memory. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241064-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (13MB) | Request a copy

Abstract

Analisis penyebab gangguan jaringan transmisi berdasarkan rekaman Disturbance Fault Recorder (DFR) masih banyak dilakukan secara manual sehingga memerlukan waktu, bergantung pada pengalaman operator, dan berpotensi menghasilkan interpretasi yang tidak konsisten. Di sisi lain, sebagian besar penelitian klasifikasi gangguan berbasis deep learning masih menggunakan data simulasi dan representasi data tunggal, sehingga kemampuan model dalam menghadapi kompleksitas sinyal gangguan aktual masih terbatas. Penelitian ini bertujuan mengembangkan dan mengevaluasi kerangka klasifikasi penyebab gangguan jaringan transmisi berbasis data DFR aktual menggunakan arsitektur hybrid Convolutional Neural Network - Long Short-Term Memory (CNN–LSTM). Model yang diusulkan memanfaatkan dua representasi dari rekaman gangguan yang sama. CNN digunakan untuk mengekstraksi pola spasial dari visualisasi gelombang, sedangkan LSTM digunakan untuk mempelajari dinamika temporal sinyal arus dan tegangan. Fitur dari kedua cabang kemudian digabungkan untuk mengklasifikasikan gangguan petir, binatang, dan vegetasi. Mekanisme uncertainty rejection juga diterapkan untuk menandai prediksi dengan tingkat keyakinan rendah agar model tidak memaksakan hasil klasifikasi. Hasil pengujian menunjukkan bahwa model Hybrid CNN–LSTM mencapai akurasi sebesar 96,3%, lebih tinggi dibandingkan CNN Only dan LSTM Only yang masing-masing memperoleh 92,6%. Peningkatan sebesar 3,7 poin persentase menunjukkan potensi penggabungan fitur spasial dan temporal dalam menangani pola gangguan yang ambigu. Kerangka yang dikembangkan dapat menjadi dasar sistem pendukung keputusan untuk membantu operator mempercepat identifikasi penyebab gangguan, menentukan prioritas inspeksi, dan meningkatkan ketepatan penanganan gangguan pada jaringan transmisi.
======================================================================================================================================
Analysis of the causes of transmission network disturbances based on Disturbance Fault Recorder (DFR) recordings is still largely performed manually. This process is time-consuming, highly dependent on operator experience, and may lead to inconsistent interpretations. Meanwhile, most deep-learning-based fault classification studies continue to rely on simulated data and a single data representation, limiting the models’ ability to address the complexity of actual disturbance signals. This study aims to develop and evaluate a transmission network fault-cause classification framework using actual DFR data and a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) architecture.The proposed model utilizes two representations derived from the same disturbance recording. The CNN is employed to extract spatial patterns from waveform visualizations, while the LSTM is used to learn the temporal dynamics of voltage and current signals. The features extracted from both branches are subsequently fused to classify lightning-, animal-, and vegetation-related faults. An uncertainty rejection mechanism is also implemented to flag low-confidence predictions, thereby preventing the model from forcing potentially unreliable classifications. The experimental results demonstrate that the Hybrid CNN–LSTM model achieved an accuracy of 96.3%, outperforming the CNN-Only and LSTMOnly models, which each obtained an accuracy of 92.6%. The improvement of 3.7 percentage points indicates the potential benefit of integrating spatial and temporal features in addressing ambiguous fault patterns. The developed framework can serve as the foundation for a decision-support system that assists transmission system operators in accelerating fault-cause identification, prioritizing field inspections, and improving the accuracy of fault-handling actions within transmission networks.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	gangguan jaringan transmisi, klasifikasi gangguan, Disturbance Fault Recorder (DFR), transmission network faults, fault classification, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM).
Subjects:	T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Electrical Technology > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Hafizh Tri Januar
Date Deposited:	31 Jul 2026 04:14
Last Modified:	31 Jul 2026 04:14
URI:	http://repository.its.ac.id/id/eprint/140416

Actions (login required)

View Item