Aplikasi Machine learning LSTM-RNN untuk Prediksi Efisiensi dan Kinerja Sistem PLTS Terintegrasi Virtual reality

Apramada, Nara (2026) Aplikasi Machine learning LSTM-RNN untuk Prediksi Efisiensi dan Kinerja Sistem PLTS Terintegrasi Virtual reality. KODEPRODI30201#TEKNIK_FISIKA, Institut Teknologi Sepuluh Nopember. (Unpublished)

Text 5009221091-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Pemanfaatan energi surya yang meningkat dihadapkan pada tantangan keluaran daya intermiten akibat ketergantungan pada kondisi cuaca, sehingga monitoring kinerja Pembangkit Listrik Tenaga Surya (PLTS) secara real-time menjadi krusial namun masih terbatas pada visualisasi data numerik yang kurang interaktif. Penelitian ini bertujuan mengembangkan sistem monitoring PLTS berbasis Internet of Things (IoT) yang diintegrasikan dengan model machine learning sekuensial untuk memprediksi daya PV, efisiensi, dan kinerja/Performance Ratio (PR), serta menampilkannya melalui antarmuka Virtual reality (VR) tiga dimensi. Data lingkungan dan elektrik diakuisisi menggunakan ESP32 dengan sensor irradiansi, suhu, dan daya yang dikalibrasi dengan error di bawah 5% sesuai IEC 61724, lalu disimpan pada Firebase Realtime Database. Tiga arsitektur Recurrent neural network (RNN), Long short-term memory (LSTM), dan Gated recurrent unit (GRU) dievaluasi pada 36 kombinasi hyperparameter, meliputi jumlah layer 1–4, sequence length 32/64/128, dan unit 64/128/256, lalu divalidasi menggunakan nRMSE, koefisien determinasi (R²) dan anomaly score. Hasil terbaik diperoleh dari model GRU konfigurasi satu layer, sequence length 32, dan 256 unit, menghasilkan nRMSE PR 5,9% (R² = 0,937) serta nRMSE Daya PV 7,6–7,8%. Dari enam target prediksi yang diuji, enam target memenuhi kriteria nRMSE di bawah 15%. Analisis SHAP mengonfirmasi konsistensi model terhadap hubungan fisis dan mampu menjelaskan alur berfikir model. Sistem VR berbasis Blender dan Unity yang terintegrasi dengan dashboard React dan Firebase lulus seluruh tujuh skenario uji fungsional, menampilkan data real-time dan prediksi lima jam ke depan secara interaktif, mendukung pemantauan dan pengambilan keputusan operasional PLTS yang lebih informatif
=====================================================================================================================================
The growing utilization of solar energy faces a key challenge in intermittent power output caused by dependence on weather conditions, making real-time performance monitoring of Photovoltaic Power Plants (PLTS) essential, yet conventional monitoring remains limited to numerical data with poor interactivity. This study develops an Internet of Things (IoT)-based PLTS monitoring system integrated with sequential machine learning models to predict PV power, efficiency, and Performance Ratio (PR), displayed through a three-dimensional Virtual reality (VR) interface. Environmental and electrical data were acquired using ESP32 with irradiance, temperature, and power sensors, calibrated to under 5% error per IEC 61724, then stored in a Firebase Realtime Database. Three architectures, Recurrent neural network (RNN), Long short-term memory (LSTM), and Gated recurrent unit (GRU), were evaluated across 36 hyperparameter combinations each, covering 1–4 layers, sequence lengths of 32/64/128, and 64/128/256 units, then validated using normalized Root Mean Squared Error (nRMSE), coefficient of determination (R²), anomaly score, and sequence prediction accuracy. The best result came from a single-layer GRU with sequence length 32 and 256 units, yielding nRMSE of 5.9% (R² = 0.937) for PR and 7.6–7.8% for PV power. Of six prediction targets tested, six met the nRMSE below 15% criterion. SHAP analysis confirmed the model's consistency with the physical relationship. The Blender- and Unity-based VR system, integrated with a React dashboard and Firebase, passed all seven functional test scenarios, displaying real-time data and five-hour-ahead predictions interactively, supporting more informative PLTS monitoring and operational decision-making.

Item Type:	Other
Uncontrolled Keywords:	Energi Terbarukan, Internet of Things, machine learning, monitoring PLTS,Virtual reality, Gated recurrent unit, Renewable Energy, Internet of Things, machine learning, solar power plant monitoring, Virtual Reality, Gated Recurrent Unit
Subjects:	L Education > L Education (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1056 Solar power plants. Ocean thermal power plants T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1322.6 Electric power-plants
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Nara Pangestu Apramada
Date Deposited:	01 Aug 2026 06:01
Last Modified:	01 Aug 2026 06:01
URI:	http://repository.its.ac.id/id/eprint/140793

Actions (login required)

View Item