Update Artikelilmiah: 51151

DAFFA AFIF NASRUDIN

NIM

Judul Artikel

Abstrak (Bhs. Indonesia)

Meningkatnya kebutuhan masyarakat akan layanan internet yang stabil membuat operator telekomunikasi perlu mengoptimalkan kinerja Base Transceiver Station (BTS), sementara sebagian besar BTS belum dilengkapi Power Monitoring System (PMS) sehingga data konsumsi arus aktual (DC Load) tidak tersedia, padahal sangat penting untuk perencanaan kapasitas baterai. Permasalahan utama penelitian ini meliputi cara memprediksi DC Load tanpa PMS, membandingkan akurasi penggunaan fitur tunggal dan multifitur, serta menilai kinerja model regresi berdasarkan MAE, MSE, RMSE, dan R². Penelitian ini bertujuan menghasilkan model prediksi DC Load yang akurat dan dapat digunakan sebagai dasar estimasi kebutuhan baterai. Metode penelitian mencakup pengumpulan data BTS Power, trafik, jumlah pengguna terkoneksi, dan PRB dari PT Indosat Ooredoo Hutchison Jawa Tengah, preprocessing, pembangunan model Linear Regression, Decision Tree Regressor, Random Forest Regressor, dan XGBoost, serta evaluasi performa pada data uji. Hasil menunjukkan regresi linear satu fitur menghasilkan R² 0,986 (small hub site) dan 0,969 (end site), sedangkan model multifitur terbaik adalah Random Forest Regressor dengan R² 0,995, MAE 0,941, RMSE 1,753 pada small hub site, serta R² 0,944 pada end site. Secara keseluruhan, model Random Forest Regressor terbukti paling akurat pada beban rendah 0–1,5 kWh, sementara Regresi Linear single-fitur lebih optimal pada beban tinggi >1,5 kWh, sehingga kombinasi keduanya membentuk pendekatan adaptif yang mampu memberikan estimasi DC Load paling presisi di seluruh tipe BTS.

Abtrak (Bhs. Inggris)

The increasing demand for stable and high-speed internet services compels telecommunications operators to optimize the performance of Base Transceiver Stations (BTS). However, most BTS sites are not equipped with a Power Monitoring System (PMS), resulting in the unavailability of actual current consumption (DC Load) data, which is essential for planning battery backup capacity. The main problems addressed in this study include how to predict DC Load without PMS, comparing the accuracy of single-feature and multi-feature approaches, and evaluating the performance of regression models using MAE, MSE, RMSE, and R² metrics. This study aims to develop an accurate DC Load prediction model that can serve as a reference for estimating battery requirements. The methodology includes collecting BTS Power, traffic, connected user count, and PRB data from PT Indosat Ooredoo Hutchison Central Java, followed by preprocessing, developing Linear Regression, Decision Tree Regressor, Random Forest Regressor, and XGBoost models, and evaluating their performance on test data. Results show that the single-feature linear regression model achieved R² scores of 0.986 for small hub sites and 0.969 for end sites, while the best multi-feature model was the Random Forest Regressor, with an R² of 0.995, MAE of 0.941, and RMSE of 1.753 for small hub sites, and an R² of 0.944 for end sites. Overall, the Random Forest Regressor proved to be the most accurate for low-load ranges of 0–1.5 kWh, whereas single-feature Linear Regression performed better at higher load levels above 1.5 kWh, resulting in an adaptive approach that provides the most precise DC Load estimation across all BTS site types.

Kata kunci

Pembimbing 1

Pembimbing 2

Pembimbing 3

Tahun

Jumlah Halaman