Pengimplementasian Energy Production Forecast Model Dan Soiling Estimation Pada Sistem Plts Pt. Atap Surya Nusantara Untuk Mendukung Predictive Maintenance

Rahadiansyah, Nur Azka (2026) Pengimplementasian Energy Production Forecast Model Dan Soiling Estimation Pada Sistem Plts Pt. Atap Surya Nusantara Untuk Mendukung Predictive Maintenance. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027221064-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Penumpukan kotoran (soiling) pada panel surya menurunkan kinerja Pembangkit Listrik Tenaga Surya (PLTS), khususnya di iklim tropis Indonesia yang memiliki variabilitas cuaca dan curah hujan tinggi. Penelitian ini membangun sistem pendukung operasional pada PLTS Grand Kamala Lagoon yang terdiri atas dua model prediktif yang ditopang perhitungan Soiling Ratio berbasis fisika, dihitung deterministik melalui pemodelan akumulasi massa debu Coello Boyle dan fungsi error yang diperkuat kajian Fernández Solas, tanpa pelatihan machine learning. Model Peramalan Energi meramalkan daya keluaran secara soiling-aware melalui dekomposisi Ensemble Empirical Mode Decomposition, prediksi komponen frekuensi tinggi dengan Long Short-Term Memory dan frekuensi rendah dengan XGBoost, digabung berbobot melalui Snake Optimizer, dengan hyperparameter dituning menggunakan Optuna. Model Pemeliharaan Prediktif mengidentifikasi sumber rugi energi melalui deteksi anomali berbasis Median Absolute Deviation dan persistensi Performance Index, membedakan Fault Peralatan berbasis pohon prioritas dari Fault Energy yang mengfaktorkan rugi soiling dapat pulih terhadap gangguan kelistrikan berdasarkan Soiling Ratio. Kedua model diuji pada data operasional nyata selama tujuh belas bulan (Januari 2025-Mei 2026). Model Peramalan Energi mencapai Root Mean Square Error ternormalisasi sekitar 9,46 persen terhadap kapasitas, dengan studi ablasi menunjukkan pemisahan model per komponen belum terbukti konsisten unggul. Model Pemeliharaan Prediktif dievaluasi jujur melalui distribusi kategori gangguan, mengingat tidak tersedianya catatan gangguan resmi sebagai pembanding independen. Studi ini menemukan tingkat soiling sangat kecil pada iklim tropis basah lokasi penelitian akibat pembersihan alami hujan yang sering, dengan Soiling Ratio bertahan di atas 0,99, dilaporkan sebagai karakteristik lapangan. Penelitian ini memvalidasi sistem secara end-to-end sebagai proof-of-concept pada lokasi penelitian.
===============================================================================================================================
Soiling accumulation on solar panels reduces the performance of Photovoltaic Power Plants, particularly in Indonesia's tropical climate with high weather variability and rainfall. This research develops an operational support system for the Grand Kamala Lagoon PV plant, consisting of two predictive models supported by a physics-based Soiling Ratio calculation, computed deterministically through the Coello-Boyle dust mass accumulation model and an error function whose validity is reinforced by Fernández Solas et al., without requiring machine learning training. The Energy Forecasting Model predicts power output in a soiling-aware manner through Ensemble Empirical Mode Decomposition, with high-frequency components predicted using Long Short-Term Memory and low-frequency components using XGBoost, combined through weighted fusion optimized by Snake Optimizer, with hyperparameters tuned using Optuna. The Predictive Maintenance Model identifies energy loss sources through anomaly detection based on Median Absolute Deviation and Performance Index persistence rules, distinguishing Equipment Fault classified through a rule-based priority tree from Energy Fault, which attributes recoverable soiling losses from electrical faults based on Soiling Ratio. Both models were developed and tested using seventeen months of real operational data (January 2025-May 2026). The Energy Forecasting Model achieved a normalized Root Mean Square Error of approximately 9.46 percent of capacity, with an ablation study showing that separating models by frequency component has not been consistently proven superior. The Predictive Maintenance Model was honestly evaluated through the distribution of fault categories, given the absence of official fault logs as an independent benchmark during the study period. This study found that soiling levels were very low at the research site's humid tropical climate due to frequent natural rain cleaning, with Soiling Ratio remaining above 0.99, reported as-is as a field characteristic. This research validates the system end-to-end on tropical operational data as a proof-of-concept at the research site.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Soiling Ratio, Peramalan Energi, Pemeliharaan Prediktif, Ensemble Empirical Mode Decomposition, Median Absolute Deviation, Pembangkit Listrik Tenaga Surya. Soiling Ratio, Energy Forecasting, Predictive Maintenance, Ensemble Empirical Mode Decomposition, Median Absolute Deviation, Photovoltaic Power Plant
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Nur Azka Rahadiansyah
Date Deposited:	02 Aug 2026 19:12
Last Modified:	02 Aug 2026 19:12
URI:	http://repository.its.ac.id/id/eprint/141242

Actions (login required)

View Item