DESAIN MODEL MACHINE LEARNING UNTUK DETEKSI STABILITAS FREKUENSI PADA SISTEM TENAGA LISTRIK DENGAN METODE XGBOOST,RANDOM FOREST,DAN SUPPORT VECTOR MACHINE

Raja Kamal Ramadhan, Ir. Lesnanto Multa Putranto, S.T., M.Eng., Ph.D., IPM., SMIEEE ; Ir. Roni Irnawan, S.T., M.Sc., Ph.D., SMIEEE.

2025 | Tesis | S2 Teknik Elektro

Fequency stability detection merupakan bidang penelitian krusial dalam sistem tenaga listrik karena dampaknya yang signifikan terhadap keandalan dan stabilitas jaringan listrik yang saling terhubung. Metode konvensional berbasis model sistem tenaga semakin tergantikan berbasis data-driven model guna mengatasi beban komputasi yang tinggi. Penelitian ini mengusulkan kerangka kerja Fequency stability detection menggunakan model machine learning XGBoost untuk mengisi kesenjangan dalam penerapan teknik machine learning pada sistem tenaga berskala besar dan kompleks, termasuk jaringan listrik terhubung yang melampaui skala mikrogrid. Dataset sintetis dihasilkan dari IEEE 39-bus Test System, yang mensimulasikan kondisi operasi normal serta skenario gang guan N-1, mencakup variasi pada permintaan beban, daya pembangkitan, serta para meter frekuensi nadir dan frekuensi rebound. Model dievaluasi berdasarkan akurasi, presisi, recall, F1-score, serta efisiensi komputasi, dengan hasil menunjukkan bahwa Random Forest mencapai akurasi tertinggi 94.34?n recall 95% untuk kondisi tidak stabil, XGBoost memperoleh akurasi 93,08?ngan recall 93%, sementara SVM-RBF memiliki akurasi 89,62?n recall 90%. Hasil penelitian menunjukkan bahwa XGBo ost memberikan klasifikasi paling efektif, diikuti oleh Random Forest, dengan kedua nya menunjukkan kinerja yang lebih unggul dibandingkan SVM-RBF dalam mendeteksi ketidakstabilan frekuensi. Temuan ini memiliki implikasi penting dalam meningkatkan pemantauan dan manajemen stabilitas sistem tenaga secara real-time, serta menegaskan potensi XGBoost dan Random Forest sebagai solusi yang dapat diskalakan untuk peman tauan stabilitas frekuensi dalam sistem tenaga listrik yang kompleks.

Detection of frequency instability has become a critical area of research in power systems because of its significant impact on the reliability and stability of interconnected grids. Traditional methods based on complete power system models are increasingly repla ced by data driven approaches to address computational burdens. This paper proposes a frequency stability detection framework using the XGBoost machine learning model. It addresses the gap in applying machine learning techniques to larger and more complex power systems, including interconnected grids beyond microgrids. A synthetic dataset was generated from the IEEE 39-bus test system, simulating both normal operating condi tions and N-1 contingency scenarios. The dataset incorporates variations in load demand, generator dispatch power, and frequency nadir and rebound. The models were asses sed based on accuracy, precision, recall, F1-score, and computational efficiency. The Random Forest model achieved the highest accuracy at 94.34%, with a recall of 95% for unstable states, while the XGBoost model reached an accuracy of 93,08% and a recall of 93%. The SVM-RBF model achieved an accuracy of 89,62% and a recall of 90%. The results demonstrate that XGBoost offers the most effective classification, followed closely by Random Forest, with both models outperforming SVM-RBF in detecting frequency instability. These findings have important implications for improving real-time stability monitoring and management in power systems.. These findings underscore the potential of XGBoost and Random Forest for scalable frequency stability monitoring in complex power systems.

Kata Kunci : data-driven model, fequency stability detection, machine learning.