Pengembangan Sistem Deteksi Otomatis Objek KWh Meter pada Data Audit Pelanggan PT PLN (Persero) Menggunakan Pendekatan Fine-Tuned MobileNetV2

Sudrajab, Achmad Fajri (2025) Pengembangan Sistem Deteksi Otomatis Objek KWh Meter pada Data Audit Pelanggan PT PLN (Persero) Menggunakan Pendekatan Fine-Tuned MobileNetV2. Project Report. [s.n.], [s.l.]. (Unpublished)

Text 5025221104-Project_Report.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

PT PLN (Persero) menghadapi tantangan efisiensi dalam proses validasi ribuan foto KWh meter pelanggan yang dikirimkan oleh petugas lapangan setiap harinya. Proses verifikasi manual yang berjalan saat ini memakan waktu lama dan rentan terhadap kesalahan manusia (human error), sehingga berpotensi meloloskan data anomali yang tidak valid. Penelitian ini bertujuan merancang bangun sistem deteksi otomatis objek KWh meter untuk memisahkan gambar valid dan tidak valid secara akurat dan cepat. Metode yang digunakan adalah Deep Learning dengan pendekatan Transfer Learning menggunakan arsitektur MobileNetV2. Model dikembangkan melalui strategi pelatihan dua tahap (Two-Stage Training) yang mencakup pembekuan bobot dasar dan fine-tuning pada 30 lapisan terakhir untuk meningkatkan sensitivitas terhadap karakteristik fisik meteran listrik. Berdasarkan pengujian pada data uji, model berhasil mencapai akurasi global sebesar 99,65% dengan nilai Precision dan Recall di atas 0,99. Selanjutnya, pengujian skala besar (stress testing) pada 20.000 gambar lapangan menunjukkan bahwa sistem mampu memproses data dengan kecepatan rata-rata 36,53 gambar per detik pada perangkat komputasi standar. Hasil ini membuktikan bahwa sistem yang dibangun mampu meningkatkan efisiensi waktu validasi secara signifikan dan siap diimplementasikan sebagai alat pendukung keputusan dalam kegiatan audit operasional PLN.
===================================================================================================================================
PT PLN (Persero) faces efficiency challenges in the validation process of thousands of customer KWh meter photos submitted daily by field officers. The current manual verification process is time-consuming and prone to human error, thereby potentially allowing invalid anomaly data to pass through. This study aims to design and build an automated KWh meter object detection system to accurately and rapidly distinguish between valid and invalid images. The method employed is Deep Learning with a Transfer Learning approach utilizing the MobileNetV2 architecture. The model was developed through a Two-Stage Training strategy, which includes freezing base weights and fine-tuning the last 30 layers to enhance sensitivity towards the physical characteristics of electric meters. Based on evaluation using test data, the model successfully achieved a global accuracy of 99.65% with Precision and Recall values above 0.99. Furthermore, large-scale stress testing on 20,000 field images demonstrated that the system is capable of processing data at an average speed of 36.53 images per second on standard computing devices. These results demonstrate that the developed system significantly improves validation time efficiency and is ready for implementation as a decision support tool in PLN's operational audit activities.

Item Type:	Monograph (Project Report)
Uncontrolled Keywords:	Deep Learning, Transfer Learning, MobileNetV2, KWh Meter Detection, Image Classification, Deep Learning, Klasifikasi Gambar, MobileNetV2, Transfer Learning, Deteksi KWh Meter.
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	ACHMAD FAJRI SUDRAJAB
Date Deposited:	12 Jan 2026 04:16
Last Modified:	12 Jan 2026 04:16
URI:	http://repository.its.ac.id/id/eprint/129473

Actions (login required)

View Item