Tambun, Rejeki (2025) Analisis Data Harian Pelanggan AMR Dengan Algoritma K-Means Clustering Dan Isolation Forest Untuk Deteksi Anomali Dan Pengurangan Susut Listrik ( Studi Kasus: PT PLN (Persero) UP3 Mamuju ). Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6022231084-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6022231084-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy |
Abstract
Deteksi anomali konsumsi energi pelanggan menjadi aspek penting dalam upaya pengurangan susut non-teknis (Non-Technical Losses/NTL) di sektor ketenagalistrikan. Penelitian ini bertujuan untuk membandingkan performa dua algoritma unsupervised learning, yaitu K-Means Clustering dan Isolation Forest, dalam mendeteksi anomali pada data harian pelanggan Automatic Meter Reading (AMR) PLN UP3 Mamuju. Data yang digunakan mencakup pelanggan tegangan rendah (TR), tegangan rendah dengan CT (TR-CT), dan tegangan menengah (TM) pada periode 2021–2023, dengan variabel utama berupa tegangan, arus, dan arus tidak seimbang.
Evaluasi kinerja dilakukan menggunakan confusion matrix dengan metrik recall, precision, F1-score, accuracy, dan Cohen’s Kappa, serta perbandingan waktu eksekusi sebagai indikator efisiensi. Hasil penelitian menunjukkan bahwa K-Means unggul pada variabel tegangan dan arus tidak seimbang, dengan F1-score mencapai 100% pada pelanggan TR-CT tahun 2023. Sementara itu, Isolation Forest menunjukkan performa lebih baik pada variabel arus, khususnya untuk pelanggan TM, dengan F1-score hingga 88,89%. Dari sisi efisiensi, K-Means lebih cepat pada dataset kecil, sedangkan Isolation Forest lebih stabil untuk dataset besar. Hasil deteksi anomali dapat dimanfaatkan sebagai rekomendasi untuk melakukan pemeliharaan rutin, preventif, maupun korektif terhadap pelanggan terindikasi. Penelitian ini merekomendasikan pendekatan hybrid yang menggabungkan kekuatan kedua algoritma guna meningkatkan efektivitas pengawasan konsumsi energi dan pengendalian susut listrik di PLN.
========================================================================================================================
Anomaly detection in customer energy consumption is a critical component in reducing non-technical losses (NTL) in the power distribution sector. This study aims to compare the performance of two unsupervised learning algorithms, K-Means Clustering and Isolation Forest, in detecting anomalies within daily Automatic Meter Reading (AMR) data from PLN UP3 Mamuju. The dataset includes low voltage (TR), low voltage with CT (TR-CT), and medium voltage (TM) customers from 2021 to 2023, with key variables including voltage, current, and unbalanced current.
Performance evaluation was conducted using a confusion matrix with metrics such as recall, precision, F1-score, accuracy, and Cohen’s Kappa, along with execution time as a measure of computational efficiency. The results indicate that K-Means performs better for voltage and unbalanced current variables, achieving an F1-score of 100% for TR-CT in 2023. Conversely, Isolation Forest outperforms in detecting anomalies in current data, especially for TM customers, with F1-scores reaching up to 88.89%. In terms of efficiency, K-Means is faster for smaller datasets, while Isolation Forest is more stable for larger data volumes.
The identified anomalies serve as a practical reference for prioritizing routine, preventive, or corrective maintenance on suspect customers. This study recommends a hybrid approach that combines the strengths of both algorithms to enhance anomaly monitoring and support energy loss reduction strategies at PLN.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Anomali, AMR, K-Means, Isolation Forest, PLN, Non-Technical Losses, Anomaly, AMR, K-Means, Isolation Forest, PLN, Non-Technical Losses. |
| Subjects: | Q Science T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3030 Electric power distribution systems |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis |
| Depositing User: | Rejeki Tambun |
| Date Deposited: | 24 Jul 2025 07:25 |
| Last Modified: | 24 Jul 2025 07:25 |
| URI: | http://repository.its.ac.id/id/eprint/121139 |
Actions (login required)
 |
View Item |