Ramadhan, Ihsan Kurnia (2026) Implementasi Fuzzy Inference System (FIS) untuk Deteksi Pencurian Listrik Menggunakan Data Advanced Metering Infrastructure (AMI) Pada Jaringan Smart Grid. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6022241121-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6022241121-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (13MB) | Request a copy |
Abstract
Perkembangan teknologi di sektor ketenagalistrikan mendorong transformasi menuju sistem yang lebih cerdas dan efisien, salah satunya melalui penerapan Advanced Metering Infrastructure (AMI). AMI tidak hanya berfungsi sebagai alat ukur digital, tetapi juga sebagai sarana deteksi dini terhadap anomali pemakaian energi listrik yang dapat mengindikasikan pencurian listrik. PT PLN (Persero) Unit Induk Distribusi (UID) Banten telah menerapkan AMI secara komersial sejak tahun 2023 dengan total pelanggan terpasang mencapai 141.406 per April 2025. Namun, pemanfaatan data AMI untuk mendeteksi pencurian listrik masih terbatas pada analisis selisih arus netral dan arus fasa, dengan tingkat keberhasilan sekitar 2,4%. Penelitian ini bertujuan untuk mengembangkan metode deteksi pencurian listrik dengan menggunakan pendekatan Fuzzy Inference System (FIS). Berdasarkan hasil pengecekan lapangan, pelanggan yang terindikasi melakukan penyalahgunaan pemakaian energi listrik golongan III (P III) diuji pada model FIS yang telah dirancang. Metode FIS dapat mendeteksi pola yang mengindikasikan penyalahgunaan pemakaian energi listrik ke dalam tiga kategori, yaitu rendah, sedang, dan tinggi dengan memanfaatkan data pengukuran AMI, yaitu tegangan, arus fasa, arus netral, konsumsi daya, dan faktor daya. Data pengukuran diolah menggunakan MATLAB untuk menentukan nilai indikatif pencurian. Hasil validasi model FIS terhadap 120 pelanggan menghasilkan sensitivitas 68%, spesifisitas 71,4%, presisi 63%, akurasi 70%, dan F1-score 65,4%. Validasi silang 5-fold berstrata dengan 20 kali pengulangan mengonfirmasi kestabilan hasil tersebut dengan selisih tidak melebihi 3,1 poin persentase. Analisis separabilitas menunjukkan bahwa nilai AUC (Area Under the Curve) gabungan kelima parameter sebesar 0,717, dan presisi model telah menyamai plafon teoretis tersebut. Temuan ini menunjukkan bahwa kinerja model dibatasi oleh daya pisah parameter pengukuran sesaat, bukan oleh perancangan sistem inferensi fuzzy. Metode ini dapat diterapkan sebagai alat prioritisasi target pemeriksaan lapangan karena mampu meningkatkan efisiensi inspeksi hingga dua kali lipat dibandingkan dengan pemilihan target secara acak.
========================================================================================================================================
Technological developments in the electricity sector are driving a transformation towards smarter and more efficient systems, one of which is through the implementation of Advanced Metering Infrastructure (AMI). AMI functions not only as a digital measuring tool but also as a means of early detection of electricity usage anomalies that could indicate electricity theft. PT PLN (Persero) Unit Induk Distribusi Banten (UID) has been commercially implementing AMI since 2023, with a total of 141,406 customers as of April 2025. However, the use of AMI data to detect electricity theft is still limited to analyzing the difference between neutral and phase currents, with a success rate of around 2.4%. This research aims to develop an electricity theft detection method using a Fuzzy Inference System (FIS) approach. Based on the results of field checks, customers who are indicated to have misused electrical energy usage in class III (P III) are tested on the designed FIS model. The FIS method in this study can detect patterns that indicate misuse of electrical energy usage into three categories, low, medium, and high, by utilizing AMI measurement data, including voltage, phase current, neutral current, power consumption, and power factor. The measurement data is processed using MATLAB to determine the indicative value of theft. Validation of the proposed FIS model against 120 customers yielded a sensitivity of 68%, specificity of 71.4%, precision of 63%, accuracy of 70%, and an F1-score of 65.4%. A stratified 5-fold cross-validation repeated 20 times confirmed the stability of these results, with deviations not exceeding 3.1 percentage points. Separability analysis showed a combined AUC (Area Under the Curve) of 0.711 for the five parameters, and the model precision has reached this theoretical ceiling. These findings indicate that model performance is constrained by the discriminative power of instantaneous measurement parameters rather than by the design of the fuzzy inference system. The method can be applied as a prioritisation tool for field inspection targets, as it doubles inspection efficiency compared with random target selection.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | pencurian listrik, kerugian non-teknis, Fuzzy Inference System, electricity theft, non-technical losses, Fuzzy Inference System |
| Subjects: | 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: | Ihsan Kurnia Ramadhan |
| Date Deposited: | 31 Jul 2026 01:01 |
| Last Modified: | 31 Jul 2026 01:01 |
| URI: | http://repository.its.ac.id/id/eprint/140074 |
Actions (login required)
 |
View Item |