Transformasi Proses Bisnis Pencatatan Meter Menggunakan Aplikasi Berbasis Deep Learning di PLN Cianjur Kota

Alfasyah, Rendy Alfasyah (2026) Transformasi Proses Bisnis Pencatatan Meter Menggunakan Aplikasi Berbasis Deep Learning di PLN Cianjur Kota. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241046-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Proses pencatatan meter listrik pascabayar di PLN Cianjur Kota saat ini hanya memverifikasi tren konsumsi numerik tanpa validasi visual, sehingga foto meter yang tidak valid (miring, buram, jarak jauh, atau bukan foto meter) dapat lolos dan menjadi dasar penagihan. Celah ini berisiko menimbulkan underbilling maupun overbilling. Penelitian ini bertujuan untuk mentransformasikan proses bisnis pembacaan meter dengan mengembangkan sistem klasifikasi citra sebagai lapisan gatekeeper visual yang memvalidasi kualitas foto meter ke dalam lima kelas (Standar, Miring, Jauh, Buram, Non-Meter) sebelum foto diteruskan ke tahap verifikasi numerik. Dengan transformasi ini, seluruh foto yang masuk dapat divalidasi secara visual (cakupan 100%), dibandingkan kondisi eksisting yang hanya memverifikasi tren numerik dan melewatkan validasi foto pada lebih dari 85% pelanggan. Model dilatih dan diuji pada 6.633 citra lapangan dari PLN ULP Cianjur Kota menggunakan skema transfer learning dua fase. Komparasi terhadap lima arsitektur CNN menunjukkan bahwa MobileNetV2 memberikan performa terbaik dengan akurasi 97,18%, F1-score makro 0,971, dan false negative rate 1,39% pada kelas Standar. Validasi lapangan pada 150 citra unseen menghasilkan akurasi 95,3% dengan 0% false negative, mengonfirmasi keandalan model pada data aktual. Analisis Grad-CAM menunjukkan model secara konsisten berfokus pada fitur struktural meter, bukan latar belakang citra. Hasil uji coba menunjukkan bahwa sistem gatekeeper yang diusulkan dapat menjadi solusi untuk menutup celah validasi visual pada proses pencatatan meter pascabayar, sehingga berpeluang mencegah kecurangan pelaporan sekaligus mengurangi beban verifikasi manual oleh petugas back-office. Model juga terbukti layak dioptimalkan untuk deployment pada perangkat dengan sumber daya terbatas, membuka peluang penerapan langsung di sisi edge (smartphone petugas lapangan). Penelitian ini memberikan kontribusi berupa rancangan transformasi proses bisnis pencatatan meter yang dapat diadopsi PLN untuk meningkatkan integritas penagihan listrik pascabayar.
================================================================================================================================
The postpaid electricity meter recording process at PLN Cianjur Kota currently only verifies numerical consumption trends without visual validation, allowing invalid meter photos (blurry, tilted, distant, or non-meter images) to pass through and become the basis for billing. This gap risks causing both underbilling and overbilling. This study aims to transform the meter reading business process by developing an image classification system as a visual gatekeeper layer that validates meter photo quality into five classes (Standard, Tilted, Distant, Blurry, Non-Meter) before the photos are forwarded to the numerical verification stage. With this transformation, all incoming photos can be visually validated (100% coverage), compared to the existing condition which only verifies numerical trends and skips photo validation for more than 85% of customers. The model was trained and tested on 6,633 field images from PLN ULP Cianjur Kota using a two-phase transfer learning scheme. Comparison of five CNN architectures showed that MobileNetV2 achieved the best performance with an accuracy of 97.18%, a macro F1-score of 0.971, and a false negative rate of 1.39% on the Standard class. Field validation on 150 unseen images yielded an accuracy of 95.3% with 0% false negatives, confirming the model’s reliability on actual data. Grad-CAM analysis shows that the model consistently focuses on structural meter features rather than image backgrounds. Experimental results indicate that the proposed gatekeeper system can serve as a solution to close the visual validation gap in the postpaid meter recording process, thereby offering the potential to prevent reporting fraud while reducing the manual verification burden on back-office officers. The model is also shown to be feasible for optimization toward deployment on resource-constrained devices, opening opportunities for direct edge-side implementation (field officers’ smartphones). This study contributes a business process transformation design for meter recording that can be adopted by PLN to improve the integrity of postpaid electricity billing.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	CNN, deteksi kecurangan, meter listrik pascabayar, MobileNetV2, transformasi proses bisnis, business process transformation, fraud detection, MobileNetV2, postpaid electricity meter.
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK351 Electric measurements.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Rendy Alfasyah
Date Deposited:	31 Jul 2026 04:20
Last Modified:	31 Jul 2026 04:20
URI:	http://repository.its.ac.id/id/eprint/140666

Actions (login required)

View Item