Pendeteksian Anomaly dalam Load Profile AMR untuk Mengklasifikasi Pencurian Energy dan Kerusakan Meter Menggunakan Kecerdasan Buatan

Fachrurroji, Mochammad (2025) Pendeteksian Anomaly dalam Load Profile AMR untuk Mengklasifikasi Pencurian Energy dan Kerusakan Meter Menggunakan Kecerdasan Buatan. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022231082_Mochammad Fachrurroji_Tesis (1).pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Susut energi masih menjadi kendala yang signifikan dalam pengoperasian sistem distribusi listrik, yang mencakup komponen teknis dan non-teknis. Di antara semua ini, kehilangan non-teknis—yang sering kali disebabkan oleh pencurian listrik dan ketidakakuratan dalam pengukuran—menimbulkan tantangan serius terhadap efisiensi utilitas. Teknik deteksi anomali yang ada dalam sistem Automatic Meter Reading (AMR) sering kali gagal dalam memberikan identifikasi penyimpangan yang tepat waktu dan akurat. Penelitian ini mengeksplorasi dan membandingkan dua pendekatan kecerdasan buatan dengan supervised learning—Extreme Gradient Boosting (XGBoost) dan Multilayer Perceptron (MLP)—untuk mendeteksi anomali dalam data profil beban AMR, dengan fokus pada pengukuran tegangan dan arus dari pelanggan. Untuk mengatasi ketidakseimbangan jumlah data antar kelas, dilakukan proses penyeimbangan data menggunakan SMOTE (Synthetic Minority Over-sampling Technique). Dataset dibagi menjadi 80% data pelatihan dan 20% data pengujian. Evaluasi dilakukan menggunakan metrik akurasi, precision, recall, dan F1-score. Hasil penelitian menunjukkan bahwa model XGBoost secara konsisten memiliki performa terbaik, dengan akurasi hingga 99.98%, serta F1-score dan recall di atas 99.85% untuk seluruh pelanggan. Sementara itu, model MLP juga menunjukkan kinerja tinggi, namun cenderung mengalami penurunan performa pada pelanggan dengan proporsi anomali yang tinggi, dengan akurasi terendah tercatat sebesar 97.69%. Dengan hasil yang diperoleh, pendekatan AI berbasis XGBoost dinilai sangat efektif dan potensial untuk diterapkan dalam sistem monitoring smart meter guna mendukung deteksi otomatis terhadap kerusakan peralatan dan potensi pencurian listrik.
==================================================================================================================================
Energy losses remain a significant constraint in the operation of electricity distribution systems, which include both technical and non-technical components. Among these, non-technical losses—often caused by electricity theft and inaccuracies in metering—pose serious challenges to utility efficiency. Existing anomaly detection techniques in Automatic Meter Reading (AMR) systems often fail to provide timely and accurate identification of anomalies. This study explores and compares two supervised learning artificial intelligence approaches—Extreme Gradient Boosting (XGBoost) and Multilayer Perceptron (MLP)—to detect anomalies in AMR load profile data, focusing on voltage and current measurements from customers. To address the imbalance in the amount of data between classes, a data balancing process using SMOTE (Synthetic Minority Over-sampling Technique) was performed. The dataset was divided into 80% training data and 20% testing data. The evaluation was carried out using accuracy, precision, recall, and F1-score metrics The results showed that the XGBoost model consistently had the best performance, with an accuracy of up to 99.98%, and F1-score and recall above 99.85% for all customers. Meanwhile, the MLP model also showed high performance, but tended to experience a decrease in performance for customers with a high proportion of anomalies, with the lowest accuracy recorded at 97.69%. With the results obtained, the XGBoost-based AI approach is considered very effective and has the potential to be applied in smart meter monitoring systems to support automatic detection of equipment damage and potential electricity theft

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Susut Non Teknis, AMR, Deteksi Anomali, SMOTE, XGBoost, Multilayer Perceptron, Confussion Matrix
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 > 20101-(S2) Master Thesis
Depositing User:	Mochammad Fachrurroji
Date Deposited:	25 Jul 2025 03:14
Last Modified:	25 Jul 2025 03:14
URI:	http://repository.its.ac.id/id/eprint/121140

Actions (login required)

View Item