Sistem Monitoring Real-Time untuk Fault Detection pada PLTS menggunakan Model Gradient Boosted Decision Tree

Ulhaq, Haritsa Dhiya (2026) Sistem Monitoring Real-Time untuk Fault Detection pada PLTS menggunakan Model Gradient Boosted Decision Tree. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009221014-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Pembangkit Listrik Tenaga Surya (PLTS) memerlukan pemantauan secara kontinu karena perubahan kondisi lingkungan dan gangguan kelistrikan dapat menurunkan daya keluaran. Penelitian ini bertujuan membangun model klasifikasi fault, mengevaluasi performanya, serta mengintegrasikan model terbaik ke dalam sistem monitoring PLTS secara real-time. Sistem akuisisi data dibangun menggunakan ESP32 yang terhubung dengan sensor TSL2591, DHT21, DS18B20, dan PZEM-017. Data dikirim melalui protokol MQTT, diproses menggunakan Node-RED, disimpan pada MongoDB, dan ditampilkan melalui dashboard web. Data hasil monitoring melalui tahap filtering, feature engineering, cleaning, dan labelling untuk membentuk empat kelas kondisi, yaitu normal, Thermal Derating, Open-Circuit, dan Soiling. Model XGBoost dan LightGBM dilatih menggunakan pembagian data 80% training dan 20% testing, kemudian dioptimasi menggunakan Randomized Search CV dan Bayesian Search CV. Hasil evaluasi menunjukkan bahwa seluruh varian model memperoleh ROC-AUC sebesar 0,98. LightGBM dengan Bayesian Search CV menghasilkan performa terbaik dengan akurasi 0,94 serta F1-score sebesar 0,94 pada kelas normal, 0,94 pada Thermal Derating, 1,00 pada Open-Circuit, dan 0,89 pada Soiling. Analisis SHAP menunjukkan bahwa keputusan model didominasi oleh temperatur modul pada Thermal Derating, arus pada Open-Circuit, dan iradiasi pada Soiling, sehingga sesuai dengan karakteristik fisik masing-masing kondisi. Model terbaik berhasil diintegrasikan ke dalam dashboard untuk menampilkan parameter operasional, status kondisi, jenis fault, dan confidence level dengan sampling time lima menit. Dengan demikian, sistem yang dikembangkan mampu melakukan monitoring dan klasifikasi fault pada PLTS secara real-time.
======================================================================================================================================
Photovoltaic (PV) systems require continuous monitoring because changes in environmental conditions and electrical disturbances can reduce their power output. This study aims to develop a fault classification model, evaluate its performance, and integrate the best-performing model into a real-time PV monitoring system. The data acquisition system was developed using an ESP32 connected to TSL2591, DHT21, DS18B20, and PZEM-017 sensors. The acquired data were transmitted via the MQTT protocol, processed using Node-RED, stored in MongoDB, and displayed on a web-based dashboard. The monitoring data underwent filtering, feature engineering, data cleaning, and labelling to form four operating-condition classes: normal, Thermal Derating, Open-Circuit, and Soiling. XGBoost and LightGBM models were trained using an 80% training and 20% testing data split, and their hyperparameters were optimized using Randomized Search CV and Bayesian Search CV. The evaluation results showed that all model variants achieved a ROC-AUC score of 0.98. LightGBM optimized using Bayesian Search CV achieved the best performance, with an accuracy of 0.94 and F1-scores of 0.94 for the normal class, 0.94 for Thermal Derating, 1.00 for Open-Circuit, and 0.89 for Soiling. SHAP analysis showed that the model’s decisions were primarily influenced by module temperature for Thermal Derating, current for Open-Circuit, and irradiance for Soiling, which is consistent with the physical characteristics of each condition. The best-performing model was successfully integrated into the dashboard to display operational parameters, operating status, fault type, and confidence level at a five-minute sampling interval. Therefore, the developed system is capable of performing real-time monitoring and fault classification in PV systems

Item Type:	Thesis (Other)
Uncontrolled Keywords:	PLTS, Fault Detection, Internet of Things, Machine Learning, Gradient Boosted Decision Tree
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > TD Environmental technology. Sanitary engineering > TD890 Global Environmental Monitoring System T Technology > TJ Mechanical engineering and machinery > TJ808 Renewable energy sources. Energy harvesting. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1087 Photovoltaic power generation
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Haritsa Dhiya Ulhaq
Date Deposited:	01 Aug 2026 04:35
Last Modified:	01 Aug 2026 04:35
URI:	http://repository.its.ac.id/id/eprint/141404

Actions (login required)

View Item