Prediksi Daya Output Panel Surya Menggunakan Jaringan Syaraf Tiruan (ANN) Berbasis Data Cuaca Lokal: Studi Kasus di Gedung Lab Terpadu Universitas Muhammadiyah Surabaya

Krisis energi fosil dan perubahan iklim mendorong transisi menuju energi terbarukan, salah satunya adalah energi surya. Namun, daya keluaran panel photovoltaic (PV) bersifat fluktuatif karena dipengaruhi oleh variabel cuaca seperti radiasi matahari, suhu modul, dan suhu lingkungan. Penelitian ini mengembangkan model prediksi daya keluaran panel surya menggunakan Artificial Neural Network (ANN) berbasis data cuaca lokal di Gedung Laboratorium Terpadu Universitas Muhammadiyah Surabaya. Dataset bersumber dari Global Solar Atlas 2025 sebanyak 288 data yang mencakup Global Horizontal Irradiance (GHI), suhu lingkungan, dan suhu modul. Empat variasi arsitektur ANN diuji menggunakan algoritma pelatihan Levenberg-Marquardt. Hasil penelitian menunjukkan bahwa arsitektur ANN-3L-32N (tiga hidden layer, 32 neuron) merupakan konfigurasi optimal dengan koefisien determinasi (R²) sebesar 1,0000 dan Root Mean Square Error (RMSE) sebesar 0,000863 W pada data pengujian. GHI terbukti berpengaruh positif dominan terhadap daya keluaran, dimana kenaikan GHI dari 0 W/m2 pada malam hari hingga 916 W/m2 pada kondisi puncak di bulan September menghasilka P_max daro 0-78,152 W/m2, sedangkan suhu modul berpengaruh negatif terhadap efisiensi konversi melalui koefisien suhu γ = −0,0045°C sehingga kenaikan Tc dari 27°C-57,625°C menyebabkan penurunan efisiensi sebesar 14,67% pada kondisi puncak. Validasi pengujian pada kondisi ekstrem menunjukkan bahwa kondisi irradiasi puncak (916 W/m2). Dan menengah (560 W/m2), menghasilkan deviasi absolut sebesar 0,0002 W dan selisih relatifnya 0,0002% dan 0,0005% Berdasarkan hasil tersebut, direkomendasikan penerapan sistem PLTS Rooftop On-Grid tanpa baterai berkapasitas 18 kWp menggunakan 180 unit panel dengan estimasi daya rata-rata harian sebesar 6,20 kW dan produksi energi 27,91 kWh/hari. Rekomendasi operasional meliputi penjadwalan beban pada jam puncak iradiasi, pemeliharaan panel berkala, serta integrasi model ANN ke sistem pemantauan untuk mendukung optimalisasi kinerja PLTS rooftop di lingkungan kampus.

================================================================================

The fossil energy crisis and climate change have accelerated the transition toward renewable energy, particularly solar energy. However, the output power of photovoltaic (PV) panels is highly variable due to weather-related factors such as solar irradiance, module temperature, and ambient temperature. This study develops a solar panel output power prediction model using an Artificial Neural Network (ANN) based on local weather data collected from the Integrated Laboratory Building of Universitas Muhammadiyah Surabaya. The dataset, obtained from the Global Solar Atlas 2025, consists of 288 data samples, including Global Horizontal Irradiance (GHI), ambient temperature, and module temperature. Four ANN architectures were evaluated using the Levenberg–Marquardt training algorithm. The results indicate that the ANN-3L-32N architecture (three hidden layers with 32 neurons) achieved the best performance, with a coefficient of determination (R²) of 1.0000 and a Root Mean Square Error (RMSE) of 0.000863W on the testing dataset. GHI was found to have the most significant positive influence on output power, where an increase in GHI from 0 W/m² at night to 916 W/m² under peak conditions in September produced a maximum output power ranging from 0 to 78.152 W/m². In contrast, module temperature negatively affected conversion efficiency through the temperature coefficient (γ = −0.0045 °C⁻¹); an increase in module temperature from 27 °C to 57.625 °C resulted in a 14.67% reduction in efficiency under peak operating conditions. Validation under extreme operating conditions showed that peak irradiance (916 W/m²) and moderate irradiance (560 W/m²) produced absolute deviations of 0.0002 W, with relative errors of 0.0002% and 0.0005%, respectively. Based on these findings, the implementation of an 18 kWp grid-connected rooftop photovoltaic (PV) system without battery storage is recommended, utilizing 180 solar panels with an estimated average daily output power of 6.20 kW and daily energy production of 27.91 kWh. Operational recommendations include scheduling electrical loads during peak irradiance periods, performing regular panel maintenance, and integrating the ANN model into the monitoring system to support the optimization of rooftop PV system performance within the university campus.
Keywords: Artificial Neural Network (ANN), Global Horizontal Irradiance (GHI), Solar Power Plant (PLTS).