Electronic Theses and Dissertation

Peningkatan kebutuhan energi listrik mendorong pemanfaatan energi terbarukan, khususnya energi surya. Namun, sifat radiasi matahari yang fluktuatif menyebabkan ketidakpastian daya keluaran sistem fotovoltaik (PV) dan berpotensi mengganggu perencanaan operasi sistem tenaga. Oleh karena itu, diperlukan model prediksi yang akurat dan adaptif terhadap dinamika kondisi cuaca. Penelitian ini bertujuan mengembangkan model prediksi daya keluaran PV berbasis Bidirectional Long Short-Term Memory (Bi-LSTM) dengan memanfaatkan data meteorologis Kota Banda Aceh. Data yang digunakan meliputi Global Horizontal Irradiance (GHI), Direct Normal Irradiance (DNI), Diffuse Horizontal Irradiance (DHI), dan suhu udara periode 2019–2020 dengan resolusi per jam. Data diolah dalam bentuk deret waktu, kemudian dilakukan perhitungan Plane of Array Irradiance (G_poa) menggunakan pustaka pvlib dengan mempertimbangkan kemiringan dan orientasi panel. Nilai G_poa digunakan untuk mengestimasi daya keluaran PV (P_ac) sebagai variabel target. Dataset dibentuk menggunakan skema sliding window 24 jam dan dibagi menjadi data latih dan data uji. Model Bi-LSTM dikembangkan dengan dua lapisan berurutan, dropout, serta mekanisme early stopping, dan dilatih menggunakan optimasi ADAM dengan fungsi kehilangan MSE. Evaluasi kinerja dilakukan menggunakan metrik RMSE, MAE, dan MAPE. Hasil menunjukkan bahwa model Bi-LSTM menghasilkan performa prediksi yang lebih baik dibandingkan LSTM, dengan nilai RMSE 3,905 W, MAE 1,84 W, dan MAPE 4,14%. Model mampu merepresentasikan fluktuasi daya PV secara stabil mengikuti pola radiasi harian. Penelitian ini menyimpulkan bahwa Bi-LSTM efektif digunakan untuk prediksi daya PV di wilayah tropis dan berpotensi mendukung perencanaan sistem energi surya berbasis data.

The increasing demand for electrical energy encourages the utilization of renewable energy sources, particularly solar energy. However, the fluctuating nature of solar radiation introduces uncertainty in the power output of photovoltaic (PV) systems and may disrupt power system operation planning. Therefore, accurate and adaptive prediction models are required to address the dynamic behavior of weather conditions. This study aims to develop a PV power output prediction model based on Bidirectional Long Short-Term Memory (Bi-LSTM) using meteorological data from Banda Aceh. The dataset consists of Global Horizontal Irradiance (GHI), Direct Normal Irradiance (DNI), Diffuse Horizontal Irradiance (DHI), and air temperature for the period 2019–2020 with an hourly resolution. The data were processed into time-series format, and the Plane of Array Irradiance (Gₚₒₐ) was calculated using the pvlib library by considering panel tilt and orientation. The Gₚₒₐ values were then used to estimate the PV output power (P_ac) as the target variable. The dataset was constructed using a 24-hour sliding window approach and divided into training and testing sets. The Bi-LSTM model was developed using two stacked layers, dropout, and an early stopping mechanism, and trained using the ADAM optimizer with Mean Squared Error (MSE) as the loss function. Model performance was evaluated using RMSE, MAE, and MAPE metrics. The results indicate that the Bi-LSTM model outperforms the conventional LSTM model, achieving an RMSE of 3.905 W, MAE of 1.84 W, and MAPE of 4.14%. The model successfully captures the temporal patterns of PV power fluctuations and produces stable predictions following the daily solar radiation profile. This study concludes that Bi-LSTM is effective for PV power prediction in tropical regions and has strong potential to support data-driven solar energy system planning.