HUTOMO, KHANSA 'AAFIYA (2026) PEMANFAATAN ALGORITMA BiLSTM UNTUK FORECASTING DAYA KELUARAN PLTS BERBASIS DATA SATELIT NASA POWER. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
COVER.pdf Download (646kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
BAB I.pdf Restricted to Registered users only Download (63kB) |
|
|
Text (BAB II)
BAB II.pdf Restricted to Registered users only Download (225kB) |
|
|
Text (BAB III)
BAB III.pdf Restricted to Registered users only Download (227kB) |
|
|
Text (BAB IV)
BAB IV.pdf Restricted to Registered users only Download (373kB) |
|
|
Text (BAB V)
BAB V.pdf Restricted to Registered users only Download (25kB) |
|
|
Text (DAFTAR PUSTAKA)
DAFTAR PUSTAKA.pdf Restricted to Registered users only Download (27kB) |
|
|
Text (LAMPIRAN)
DAFTAR LAMPIRAN.pdf Restricted to Registered users only Download (292kB) |
Abstract
This study aims to apply the Bidirectional Long Short-Term Memory (BiLSTM) algorithm to forecast the power output of a Photovoltaic (PV) Power Plant using meteorological data from the National Aeronautics and Space Administration Prediction of Worldwide Energy Resources (NASA POWER). The dataset consists of 9,048 hourly observations covering the period from January 2025 to January 2026. The research data are obtained by combining meteorological data with PV power plant operational data, with Total Solar Power (W) used as the target variable. The meteorological variables used as input features include Clear Sky Surface Shortwave Downward Radiation, Solar Zenith Angle, Air Temperature, Specific Humidity, Precipitation, and Surface Pressure. The research stages consist of data collection and integration, Exploratory Data Analysis (EDA), data preprocessing, chronological data splitting, normalization using MinMaxScaler, sequence formation using a 24-hour window, BiLSTM model training, validation, testing, and model evaluation. A 24-hour window is used so that the model utilizes the previous 24 hourly observations to predict the power output at the following time step. The model architecture consists of two Bidirectional LSTM layers with 128 and 64 units, respectively, Dropout layers with a rate of 0.3, a Dense layer with 32 units using the ReLU activation function, and a single output neuron for predicting PV power output. The testing results show that the BiLSTM model achieved a Mean Absolute Error (MAE) of 305.77 W, a Root Mean Square Error (RMSE) of 660.95 W, and a coefficient of determination (R²) of 0.7365. These results indicate that the model can explain approximately 73.65% of the variation in PV power output in the testing data. The prediction visualization also shows that the predicted values generally follow the pattern of the actual power output, although differences remain during several periods. Based on these results, the BiLSTM algorithm can be applied to forecast PV power output using meteorological data from NASA POWER. Keywords: PV Power Output, BiLSTM, NASA POWER, Time Series Forecasting, Meteorological Data. Penelitian ini bertujuan untuk menerapkan algoritma Bidirectional Long Short-Term Memory (BiLSTM) dalam melakukan forecasting daya keluaran Pembangkit Listrik Tenaga Surya (PLTS) menggunakan data meteorologi dari NASA Prediction of Worldwide Energy Resources (NASA POWER). Data yang digunakan terdiri atas 9.048 pengamatan dengan interval waktu per jam yang mencakup periode Januari 2025 hingga Januari 2026. Data penelitian merupakan gabungan antara data meteorologi dan data operasional PLTS, dengan variabel target berupa Total Solar Power (W). Variabel meteorologi yang digunakan sebagai masukan meliputi Clear Sky Surface Shortwave Downward Radiation, Solar Zenith Angle, temperatur udara, specific humidity, precipitation, dan surface pressure. Tahapan penelitian meliputi pengumpulan dan penggabungan data, Exploratory Data Analysis (EDA), preprocessing, pembagian data berdasarkan urutan waktu, normalisasi menggunakan MinMaxScaler, pembentukan sequence menggunakan windowing 24 jam, pelatihan model BiLSTM, validasi, pengujian, dan evaluasi model. Pembentukan window 24 jam digunakan agar model memanfaatkan 24 pengamatan sebelumnya untuk memprediksi daya keluaran pada waktu berikutnya. Arsitektur model terdiri atas dua lapisan Bidirectional LSTM dengan masing-masing 128 dan 64 unit, Dropout sebesar 0,3, Dense layer sebanyak 32 unit dengan fungsi aktivasi ReLU, serta satu neuron output untuk menghasilkan nilai prediksi daya. Hasil pengujian menunjukkan bahwa model BiLSTM memperoleh nilai Mean Absolute Error (MAE) sebesar 305,77 W, Root Mean Square Error (RMSE) sebesar 660,95 W, dan koefisien determinasi (R²) sebesar 0,7365. Nilai tersebut menunjukkan bahwa model mampu menjelaskan sekitar 73,65% variasi daya keluaran PLTS pada data pengujian. Hasil visualisasi prediksi juga menunjukkan bahwa pola prediksi secara umum mengikuti pola daya aktual, meskipun masih terdapat selisih pada beberapa periode. Berdasarkan hasil tersebut, algoritma BiLSTM dapat digunakan untuk melakukan forecasting daya keluaran PLTS berbasis data meteorologi NASA POWER. Kata kunci: Daya Keluaran PLTS, BiLSTM, NASA POWER, Time Series Forecasting, Data Meteorologi.
Actions (login required)
 |
View Item |
