Prediksi Output Daya PLTS Hybrid Pulau Selayar Berbasis Integrasi Data Onsite Dan Reanalysis ERA5 Menggunakan Algoritma LSTM

Ashshiddiq, Hasbi Nur (2026) Prediksi Output Daya PLTS Hybrid Pulau Selayar Berbasis Integrasi Data Onsite Dan Reanalysis ERA5 Menggunakan Algoritma LSTM. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241056-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (16MB) | Request a copy

Abstract

Peningkatan penetrasi Pembangkit Listrik Tenaga Surya (PLTS) pada sistem kelistrikan terisolasi menimbulkan tantangan operasional akibat sifat intermiten dan variabilitas energi surya. Pada sistem kelistrikan Pulau Selayar, integrasi PLTS Hybrid dengan pembangkit konvensional dan sistem penyimpanan energi membutuhkan perencanaan operasi yang akurat agar keseimbangan antara suplai dan beban tetap terjaga. Ketidakpastian output daya PLTS yang dipengaruhi oleh perubahan irradiance dan kondisi meteorologis menjadi salah satu faktor penting yang perlu diprediksi secara andal. Penelitian ini bertujuan untuk menganalisis dan memprediksi output daya PLTS Hybrid Pulau Selayar dengan memanfaatkan integrasi data multi-sumber, yaitu data pengukuran onsite dan data reanalysis ERA5. Pendekatan berbasis deep learning menggunakan algoritma Long Short-Term Memory (LSTM) diterapkan sebagai model utama karena kemampuannya dalam menangkap hubungan nonlinier dan dependensi temporal pada data deret waktu. Selain itu, model Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), dan Feedforward Neural Network (FFN) digunakan sebagai model pembanding. Metodologi penelitian meliputi pengumpulan data output daya PLTS, irradiance onsite, serta variabel meteorologis ERA5 yang terdiri atas irradiance, cloud cover, temperature, wind speed, dan precipitation. Data diproses melalui tahapan pembersihan data, penanganan missing value, normalisasi min-max, analisis korelasi fitur, dan pembentukan data deret waktu menggunakan metode sliding window dengan panjang input 48 jam. Tiga skenario input digunakan, yaitu skenario data onsite, data ERA5, dan data hybrid. Model dikembangkan untuk prediksi jangka pendek multi-horizon dengan keluaran t+1, t+2, dan t+3 jam. Evaluasi kinerja dilakukan menggunakan Root Mean Square Error (RMSE), Mean Absolute Error (MAE), dan Normalized Root Mean Square Error (NRMSE). Hasil penelitian menunjukkan bahwa irradiance merupakan variabel paling dominan terhadap output daya PLTS. Pada model LSTM, skenario data hybrid memberikan kinerja prediksi terbaik dibandingkan skenario data onsite dan data ERA5. Pada horizon t+1, data Hybrid-LSTM menghasilkan RMSE sebesar 74,00 kW, MAE sebesar 41,28 kW, dan NRMSE sebesar 7,17%. Pada horizon t+2 dan t+3, NRMSE masing-masing sebesar 9,11% dan 9,89%. Integrasi data onsite dan data ERA5 memberikan nilai error yang sedikit lebih rendah dibandingkan penggunaan data secara terpisah, meskipun peningkatannya relatif kecil.
=================================================================================================================================
The increasing penetration of Solar Photovoltaic Power Plants (PV) in isolated power systems creates operational challenges due to the intermittent and variable nature of solar energy. In the Selayar Island power system, the integration of the Hybrid PV system with conventional power plants and energy storage systems requires accurate operational planning to maintain the balance between supply and demand. The uncertainty of PV power output, influenced by changes in irradiance and meteorological conditions, is therefore an important factor that needs to be reliably predicted. This study aims to analyze and predict the power output of the Selayar Island Hybrid PV system by integrating multi-source data, namely onsite measurements and ERA5 reanalysis data. A deep learning approach using the Long Short-Term Memory (LSTM) algorithm is applied as the main model because of its ability to capture nonlinear relationships and temporal dependencies in time-series data. Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), and Feedforward Neural Network (FFN) are used as comparison models. The research data consist of PV power output, onsite irradiance, and ERA5 meteorological variables, including irradiance, cloud cover, temperature, wind speed, and precipitation. The data are processed through data cleaning, missing value handling, min-max normalization, feature correlation analysis, and time-series construction using a sliding window with a 48-hour input length. Three input scenarios are evaluated: onsite, ERA5, and hybrid. The models are developed for short-term multi-horizon forecasting at t+1, t+2, and t+3 hours. Performance is evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Normalized Root Mean Square Error (NRMSE). The results show that irradiance is the most dominant variable affecting PV power output. For the LSTM model, the hybrid scenario provides the best forecasting performance compared with the onsite and ERA5 scenarios. At the t+1 horizon, Hybrid-LSTM produces an RMSE of 74.00 kW, an MAE of 41.28 kW, and an NRMSE of 7.17%. At the t+2 and t+3 horizons, the NRMSE values are 9.11% and 9.89%, respectively. The integration of onsite and ERA5 data produces slightly lower error values than the use of either data source separately, although the improvement is relatively small.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	PLTS Hybrid, Prediksi Daya, LSTM, ERA5, Data Onsite. Hybrid PV System, Power Forecasting, LSTM, ERA5, Onsite Data.
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1001 Production of electric energy or power T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1087 Photovoltaic power generation
Divisions:	Faculty of Electrical Technology > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Hasbi Nur Ashshiddiq
Date Deposited:	30 Jul 2026 01:41
Last Modified:	30 Jul 2026 01:41
URI:	http://repository.its.ac.id/id/eprint/139245

Actions (login required)

View Item