Prediksi Distribusi Probabilistik Potensi Radiasi Surya Harian menggunakan Copula dan XGBoost Berdasarkan Data Cuaca di Pulau Jawa

Tsuraya, Shafa Fariha (2026) Prediksi Distribusi Probabilistik Potensi Radiasi Surya Harian menggunakan Copula dan XGBoost Berdasarkan Data Cuaca di Pulau Jawa. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2043221118-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Permintaan energi yang meningkat di Pulau Jawa dan ketergantungan Indonesia pada bahan bakar fosil menimbulkan tantangan bagi ketahanan energi dan keberlanjutan lingkungan. Energi surya merupakan solusi strategis, namun karakteristiknya yang sangat dipengaruhi kondisi cuaca membuat peramalan dayanya kompleks. Penelitian ini mengembangkan model prediksi distribusi probabilistik potensi radiasi surya berbasis Copula dan XGBoost, menggunakan data cuaca spasial-temporal di wilayah terpilih Pulau Jawa selama 2019–2023. Hasil menunjukkan bahwa hubungan antara radiasi surya dan variabel meteorologi harian bersifat kompleks dan bervariasi antar sub-klaster cuaca. Copula bivariat mampu menangkap ketergantungan antar variabel, dengan seluruh parameter signifikan secara statistik. Pemilihan Copula multivariat optimal menunjukkan perbedaan karakteristik struktur ketergantungan, dengan Copula Gaussian merepresentasikan hubungan yang relatif simetris, sementara Copula Student-t menunjukkan adanya tail dependence pada kejadian radiasi ekstrem. XGBoost Classifier mengklasifikasikan dengan akurasi 0,926, sedangkan XGBoost Regressor menghasilkan prediksi dengan RMSE pengujian 0,0632 dan R² sebesar 0,8990. Evaluasi probabilistik menunjukkan kinerja yang andal dengan CRPS 0,0346, Pinball Loss rendah pada P10 (0,0116), P50 (0,0245), dan P90 (0,0102), serta sharpness 90% sebesar 0,1914. Secara keseluruhan, pendekatan Copula dan XGBoost menghasilkan prediksi probabilistik radiasi surya harian yang valid dan mendukung pengambilan keputusan berbasis risiko untuk perencanaan dan pengelolaan PLTS di Pulau Jawa.
=========================================================================================================================================
Increasing energy demand in Java Island and Indonesia’s reliance on fossil fuels pose significant challenges to energy security and environmental sustainability. Solar energy represents a strategic solution; however, its strong dependence on weather conditions makes power forecasting complex. This study develops a probabilistic distribution prediction model for daily solar radiation potential using a Copula and XGBoost framework based on spatio-temporal weather data from selected regions in Java Island during 2019–2023. The results indicate that the relationship between solar radiation and daily meteorological variables is complex and varies across weather sub-clusters. Bivariate copulas effectively capture the dependence structure among variables, with all parameters being statistically significant. Optimal multivariate copula selection reveals differences in dependence characteristics, where the Gaussian copula represents relatively symmetric dependence, while the Student-t copula indicates the presence of tail dependence in extreme solar radiation events. The XGBoost Classifier achieves an accuracy of 0.926, while the XGBoost Regressor provides reliable solar radiation predictions with a testing RMSE of 0.0632 and an R² value of 0.8990. Probabilistic evaluation demonstrates robust performance, reflected by a CRPS of 0.0346, low Pinball Loss values at P10 (0.0116), P50 (0.0245), and P90 (0.0102) quantiles, and a 90% sharpness of 0.1914. Overall, the proposed Copula and XGBoost approach produces valid probabilistic forecasts of daily solar radiation and supports risk-based decision-making for solar power plant planning and management in Java Island.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Copula, Probabilistik, XGBoost
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA278 Cluster Analysis. Multivariate analysis. Correspondence analysis (Statistics)
Divisions:	Faculty of Vocational > 49501-Business Statistics
Depositing User:	Shafa Fariha Tsuraya
Date Deposited:	28 Jul 2026 02:59
Last Modified:	28 Jul 2026 02:59
URI:	http://repository.its.ac.id/id/eprint/138439

Actions (login required)

View Item