Aplikasi IoT Smart Mobile Power Supply Berbasis PV 1,5 kWp dan Baterai 10 kWh untuk Kondisi Darurat Bencana dengan Integrasi Prediksi Cuaca
Bernardus Vico Shava Andika, Prof. Dr.Eng. Ir. F. Danang Wijaya, S.T., M.T.; Ir. Bambang Sugiyantoro, M.T.
2026 | Skripsi | TEKNIK ELEKTRO
Digitalisasi sistem pembangkit listrik tenaga surya (PV) memegang peranan krusial dalam mendukung keandalan transisi energi. Penelitian ini mengusulkan rancangan sistem akuisisi data Internet of Things (IoT) terintegrasi menggunakan mikrokontroler ESP32, protokol MQTT, Node-RED, dan basis data MongoDB untuk menggabungkan pembacaan sensor kelistrikan lokal dengan data meteorologi eksternal (API OpenMeteo) secara real-time. Setelah melalui tahap preprocessing untuk mengeliminasi anomali dan mengidentifikasi iradiasi shortwave sebagai prediktor cuaca dominan, data dieksekusi untuk membangun model forecasting energi PV. Tiga algoritma machine learning, yakni Support Vector Regression (SVR), Decision Tree (DT), dan Random Forest (RF), dioptimasi untuk memprediksi dua target: kalkulasi tegangan-arus per menit (target_vi) dan diferensial energi per menit (target_diff). Hasil evaluasi membuktikan bahwa model arsitektur pohon keputusan secara signifikan mengungguli SVR dalam memetakan fluktuasi non-linear sistem. Algoritma Random Forest tampil sebagai model paling optimal pada target_diff dengan pencapaian MSE terendah sebesar 3,0039 dan R^2 tertinggi sebesar 0,6352. Sistem ini disempurnakan melalui implementasi dashboard interaktif berbasis web yang memusatkan pemantauan kelistrikan real-time, visualisasi rekam jejak historis 96 jam, serta penyajian proyeksi energi untuk 48 jam ke depan.
The digitalization of photovoltaic (PV) power systems plays a crucial role in ensuring the reliability of the energy transition. This research proposes an integrated Internet of Things (IoT) data acquisition system utilizing an ESP32 microcontroller, MQTT protocol, Node-RED, and a MongoDB database to seamlessly merge local electrical sensor readings with external meteorological data from the OpenMeteo API in real-time. Following a preprocessing phase to eliminate data anomalies and identify shortwave irradiation as the dominant weather predictor, the dataset was utilized to develop PV energy forecasting models. Three machine learning algorithms, Support Vector Regression (SVR), Decision Tree (DT), and Random Forest (RF) were optimized to predict two distinct targets: a per-minute voltage-current calculation ($target\_{vi}$) and a per-minute energy differential ($target\_{diff}$). Evaluation results demonstrate that tree-based architecture models significantly outperform SVR in capturing the system's non-linear fluctuations. The Random Forest algorithm emerged as the most robust model for forecasting $target\_{diff}$, achieving the lowest Mean Square Error (MSE) of 3.0039 and the highest Coefficient of Determination ($R^2$) of 0.6352. Finally, the proposed system is complemented by an interactive web-based dashboard that centralizes real-time electrical monitoring, visualizes 96-hour historical data, and presents 48-hour future energy projections in a unified platform.
Kata Kunci : Fotovoltaik, Internet of Things, Machine Learning, Forecasting, Random Forest, Dashboard