Prediksi Output Daya Jangka Pendek PLTB Sidrap Berbasis LSTM-Informer Dan Data SCADA

Putro, Unggul Wahyu Tri Purnomo (2026) Prediksi Output Daya Jangka Pendek PLTB Sidrap Berbasis LSTM-Informer Dan Data SCADA. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241036-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (15MB) | Request a copy

Abstract

Integrasi pembangkit listrik tenaga bayu ke dalam sistem tenaga listrik memerlukan prediksi daya yang andal karena output PLTB bersifat fluktuatif dan dipengaruhi oleh perubahan kecepatan serta arah angin. Pada sistem kelistrikan Sulawesi, PLTB Sidrap berperan sebagai salah satu sumber energi terbarukan, tetapi variabilitas output dayanya dapat menimbulkan ketidakpastian dalam perencanaan operasi, kesiapan cadangan daya, dan pengambilan keputusan dispatch. Penelitian ini bertujuan mengembangkan strategi prediksi output daya jangka pendek PLTB Sidrap berbasis data SCADA dengan pendekatan multi-horizon dan horizon-aware LSTM–Informer.Data yang digunakan merupakan data historis SCADA PLTB Sidrap beresolusi 15 menit dengan variabel utama output daya, kecepatan angin, dan arah angin. Prediksi dilakukan untuk horizon 1 jam, 3 jam, dan 6 jam ke depan. Model yang dievaluasi meliputi SVR, XGBoost, LSTM, dan Informer, dengan evaluasi menggunakan RMSE, MAE, nRMSE, R², distribusi error, uji Diebold-Mariano, dan analisis stabilitas. Selain itu, dilakukan analisis sensitivitas fitur untuk melihat pengaruh kompleksitas input terhadap performa model.Hasil penelitian menunjukkan bahwa model terbaik bergantung pada horizon prediksi. LSTM memberikan performa terbaik pada horizon 1 jam dengan RMSE 8,871 MW, sedangkan Informer memberikan performa terbaik pada horizon 3 jam dan 6 jam dengan RMSE masing-masing 13,604 MW dan 16,538 MW. Uji Diebold-Mariano menunjukkan bahwa perbedaan LSTM dan Informer pada horizon 1 jam tidak signifikan, sedangkan Informer signifikan lebih baik pada horizon 3 jam dan 6 jam. Berdasarkan hasil tersebut, strategi horizon-aware diformulasikan dengan LSTM untuk prediksi 1 jam dan Informer untuk prediksi 3 jam serta 6 jam. Secara operasional, hasil prediksi diarahkan untuk mendukung near-real-time monitoring, dispatch awareness, dan intraday planning. Penelitian ini juga menghasilkan prototype dashboard Grafana berbasis historical playback sebagai bukti pemanfaatan hasil prediksi untuk mendukung interpretasi operasi PLTB.
======================================================================================================================================
The integration of wind power plants into power systems requires reliable power forecasting because wind power output is highly variable and influenced by changes in wind speed and wind direction. In the Sulawesi power system, Sidrap Wind Power Plant contributes as one of the renewable energy sources; however, its fluctuating output may introduce uncertainty in operational planning, reserve preparation, and dispatch decision-making. This study aims to develop a short-term power output forecasting strategy for Sidrap Wind Power Plant based on SCADA data using a multi-horizon and horizon-aware LSTM–Informer approach.
The dataset consists of historical SCADA data with a 15-minute resolution, using wind power output, wind speed, and wind direction as the main variables. Forecasting is conducted for 1-hour, 3-hour, and 6-hour ahead horizons. The evaluated models include SVR, XGBoost, LSTM, and Informer. Model performance is assessed using RMSE, MAE, nRMSE, R², error distribution analysis, the Diebold-Mariano test, and training stability analysis. Feature sensitivity analysis is also conducted to examine the effect of input complexity on forecasting performance.The results show that the best-performing model depends on the forecasting horizon. LSTM achieves the best performance for the 1-hour ahead horizon with an RMSE of 8.871 MW, while Informer achieves the best performance for the 3-hour and 6-hour ahead horizons with RMSE values of 13.604 MW and 16.538 MW, respectively. The Diebold-Mariano test indicates that the difference between LSTM and Informer at the 1-hour horizon is not statistically significant, whereas Informer is significantly better at the 3-hour and 6-hour horizons. Based on these findings, the proposed horizon-aware strategy uses LSTM for 1-hour ahead forecasting and Informer for 3-hour and 6-hour ahead forecasting. Operationally, the forecasting results support near-real-time monitoring, dispatch awareness, and intraday planning. This study also develops a historical playback-based Grafana dashboard prototype as a proof of concept for operational forecasting support.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	horizon-aware forecasting, Informer , LSTM, prediksi daya angin, SCADA, Horizon-aware forecasting, Informer, LSTM, SCADA, Wind power forecasting
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > TJ Mechanical engineering and machinery > TJ820 Wind power T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1001 Production of electric energy or power
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Unggul Wahyu Tri Purnomo Putro
Date Deposited:	31 Jul 2026 02:03
Last Modified:	31 Jul 2026 02:03
URI:	http://repository.its.ac.id/id/eprint/140357

Actions (login required)

View Item