Pengembangan Sistem Prediksi Gangguan Aset Pembangkit Berbasis Machine Learning Menggunakan Data Historis Service Request

Nissa Shafadilla Saputri, Dr. Sahirul Alam, S.T., M.Eng.

2026 | Tugas Akhir | D4 TEKNOLOGI JARINGAN

Keandalan aset pembangkit penting untuk menjaga kontinuitas operasi, tetapi data historis Service Request masih belum dimanfaatkan secara optimal dan umumnya hanya menjadi arsip. Proyek akhir ini mengembangkan sistem prediksi risiko gangguan aset pembangkit berbasis machine learning menggunakan data Service Request sebagai sumber utama dan Asset Criticality Ranking (ACR) sebagai pendukung prioritas aset. Sistem memprediksi risiko gangguan dalam horizon tujuh hari melalui preprocessing, integrasi data, pembentukan fitur, dan pemodelan klasifikasi. Hasil evaluasi menunjukkan XGBoost terpilih sebagai model terbaik dengan ROC-AUC 0,8746, precision 0,77, recall 0,71, dan F1-score 0,74. Sistem kemudian diimplementasikan dalam dashboard berbasis web menggunakan Streamlit pada AWS EC2. Pengujian menunjukkan sistem mampu menjalankan prediksi dengan rata-rata warm run response time 11,02 detik dan rata-rata CPU utilization 0,64% pada kondisi load, sehingga dapat mendukung identifikasi aset prioritas secara lebih terarah.

The reliability of power plant assets is important for maintaining operational continuity, but historical Service Request data has not been utilized optimally and is generally used only as archival data. This final project develops a machine learning-based risk prediction system for power plant asset failures using Service Request data as the main source and Asset Criticality Ranking (ACR) as supporting data for asset prioritization. The system predicts failure risk within a seven-day horizon through preprocessing, data integration, feature engineering, and classification modeling. The evaluation results show that XGBoost was selected as the best model, achieving a ROC-AUC of 0.8746, precision of 0.77, recall of 0.71, and F1-score of 0.74. The system was then implemented in a web-based dashboard using Streamlit on AWS EC2. Testing results show that the system is able to perform predictions with an average warm-run response time of 11.02 seconds and an average CPU utilization of 0.64% under load, thereby supporting more focused identification of priority assets.

Kata Kunci : Sistem prediksi gangguan, aset pembangkit, machine learning, cloud computing, service request history