Studi Komparatif Prophet dan Gradient Boosting untuk Prediksi Beban Jangka Pendek PLTU Boiler Pulverized Coal dengan Validasi Operasional
Rahwindhy Mazhar Chasnanda, Prof. Ir. Nazrul Effendy, S.T., M.T., Ph.D., IPM., ASEAN Eng.; Dr.-Ing. Ir. Awang Noor Indra Wardana, S.T., M.T., M.Sc., IPM.
2026 | Tesis | MAGISTER TEKNIK FISIKA
PLTU batubara masih mendominasi bauran pembangkitan listrik
di Indonesia, termasuk dalam sistem Jamali. Meskipun berperan sebagai
pembangkit beban dasar, PLTU harus menyesuaikan pembebanan berdasarkan dispatch.
Pengelolaan bahan bakar juga semakin kompleks karena coal switching dan co-firing
biomassa yang sejalan dengan target NZE. Selain itu, sistem penyiapan bahan
bakar unit PLTU dari coal bunker hingga furnace menyebabkan
perubahan komposisi bahan bakar tidak langsung tercermin pada keluaran daya.
Oleh karena itu, prediksi beban diperlukan untuk mendukung operasi PLTU.
Penelitian ini membandingkan Prophet dan Gradient
Boosting dalam memprediksi beban mingguan Unit 1 PLTU Pelabuhan Ratu
menggunakan data beban netto periode 1 April–2 November. Evaluasi dilakukan
melalui skema walk-forward dengan variasi data latih 1–6 bulan pada empat
periode pengujian Oktober 2025. Perbandingan kinerja dianalisis menggunakan scoring
multi-kriteria berdasarkan metrik akurasi, jumlah data latih, dan waktu
komputasi. Model terbaik divalidasi melalui konfirmasi dispatcher, lalu
dibandingkan terhadap dispatch aktual dan rencana eksisting.
Hasil menunjukkan bahwa Gradient Boosting unggul pada
dua periode, sedangkan Prophet unggul pada dua periode lainnya. Skema
terbaik diperoleh pada data latih 1–2 bulan. Rata-rata runtime Prophet
5 kali lebih cepat dibandingkan Gradient Boosting. Validasi operasional
menghasilkan deviasi rata-rata 4,14%, lebih rendah dibandingkan rencana
eksisting sebesar 10,40%.
Coal-fired steam power plants still dominate Indonesia’s
electricity generation mix, including in the Jamali system. Although operated
as baseload power plants, coal power plant units must adjust their load
according to dispatch. Fuel management has also become more complex due
to coal switching and biomass co-firing in line with the NZE target.
In addition, the fuel preparation system, from coal bunker to furnace,
causes changes in fuel composition to not be immediately reflected in power output.
Therefore, load forecasting is required to support operation.
This study compares Prophet and Gradient Boosting
for weekly load forecasting at PLTU Pelabuhan Ratu Unit 1 using load
data from 1 April to 2 November. The evaluation was conducted using a
walk-forward scheme with 1–6 months of training data across four testing
periods in October 2025. Model performance was analyzed using multi-criteria scoring
based on accuracy metrics, training data size, and computation time. The best
model was validated through dispatcher confirmation and then compared
with actual dispatch and the existing plan.
The results show that Gradient Boosting outperformed
in two periods, while Prophet performed better in the other two periods.
The best schemes were obtained using 1–2 months of training data. Prophet’s
average runtime was five times faster than Gradient Boosting.
Operational validation produced an average deviation of 4.14%, lower than the
existing plan at 10.40%.
Kata Kunci : Prediksi Beban Mingguan, PLTU, Prophet, Gradient Boosting, Perencanaan Beban.