Electronic Theses and Dissertation

Perencanaan pasokan listrik yang akurat sangat penting untuk menjaga stabilitas, efisiensi dan penyiapan infrastruktur sistem kelistrikan khususnya di wilayah PT PLN (Persero) UP3 Banda Aceh. Penelitian ini bertujuan untuk memprediksi beban listrik jangka pendek menggunakan algoritma Extreme Gradient Boosting (XGBoost), yang mampu memodelkan hubungan nonlinier dan kompleksitas data historis. Data yang digunakan mencakup beban listrik per jam, harian, dan mingguan, serta fitur berbasis waktu dan variabel cuaca. Model dibangun melalui tahapan pengumpulan data, pra-pemrosesan, pelatihan, dan evaluasi menggunakan metrik MAE, RMSE, dan MAPE. Hasil evaluasi kuantitatif menunjukkan bahwa konfigurasi terbaik diperoleh pada learning rate 0.1 dengan nilai RMSE sebesar 1,817 MWh, MAE sebesar 1,198 MWh dan MAPE sebesar 4,52%, yang mengindikasikan tingkat kesalahan prediksi relatif rendah dan model layak digunakan untuk prakiraan beban jangka pendek. Sementara itu, pada learning rate 0.5 diperoleh RMSE 2,047 MWh, MAE 1,368 MWh, dan MAPE 5,17% yang juga menunjukkan performa model kurang optimal. Pada konfigurasi learning rate 0,1 menunjukkan bahwa model mampu merepresentasikan pola pembebanan dengan baik. Kondisi ini mencerminkan bahwa karakteristik beban lebih stabil dan konsisten, sehingga hubungan antara variabel input dan beban dapat dipelajari secara efektif oleh model XGBoost. Dengan tingkat kesalahan di bawah 5%, model ini dinilai cukup andal untuk mendukung pengambilan keputusan operasional dan perencanaan infrastruktur, seperti perencanaan kapasitas trafo distribusi, penguatan penyulang, serta prioritas pengembangan jaringan distribusi. Implementasi model prediksi berbasis XGBoost ini diharapkan dapat menjadi alat bantu analitis yang efektif bagi PT PLN (Persero) UP3 Banda Aceh dalam mewujudkan perencanaan ketenagalistrikan yang lebih efisien dan berbasis data.

Kata kunci: Prediksi beban listrik, XGBoost, Time series forecasting, Infrastruktur ketenagalistrikan, PLN UP3 Banda Aceh.

Accurate electricity supply planning is essential to maintain system stability, efficiency, and the preparedness of power infrastructure, particularly in the service area of PT PLN (Persero) UP3 Banda Aceh. This study aims to predict short-term electricity load using the Extreme Gradient Boosting (XGBoost) algorithm, which is capable of modeling nonlinear relationships and the complexity of historical data. The data used includes hourly, daily, and weekly load records, as well as time-based features and weather variables. The model was developed through stages of data collection, preprocessing, training, and evaluation using the MAE, RMSE, and MAPE. Quantitative evaluation results show that the best configuration was obtained at a learning rate of 0.1 with an RMSE value of 1,817 MWh, an MAE of 1,198 MW, and a MAPE of 4.52%, indicating a relatively low prediction error rate and a suitable model for short-term load forecasting. Meanwhile, at a learning rate of 0.5, an RMSE of 2,047 MWh, an MAE of 1,368 MWh, and an MAPE of 5.17% was obtained, also indicating suboptimal model performance. The learning rate configuration of 0.1 shows that the model is able to represent the loading pattern well. This condition reflects that the load characteristics are more stable and consistent, so that the relationship between input variables and the load can be learned effectively by the XGBoost model. With an error rate below 5%, the model is considered reliable to support operational decision-making and infrastructure planning, such as distribution transformer capacity planning, feeder reinforcement, and prioritization of distribution network development. The implementation of this XGBoost-based prediction model is expected to serve as an effective analytical tool for PT PLN (Persero) UP3 Banda Aceh in achieving more efficient and data-driven power system planning. Kata kunci: Electricity load prediction, XGBoost, time series forecasting, power infrastructure, PLN UP3 Banda Aceh.