| dc.contributor.advisor | Annisa | |
| dc.contributor.advisor | Hardhienata, Medria Kusuma Dewi | |
| dc.contributor.author | Suganda, Noer Hanifah | |
| dc.date.accessioned | 2026-08-12T04:49:37Z | |
| dc.date.available | 2026-08-12T04:49:37Z | |
| dc.date.issued | 2026 | |
| dc.identifier.uri | http://repository.ipb.ac.id/handle/123456789/178417 | |
| dc.description.abstract | Radiasi penyinaran merupakan variabel faktor dinamis pada INA Agro-GARLIC sebagai sistem rekomendasi kesesuaian lahan bawang putih pada kawasan prioritas pengembangan lahan bawang putih di Indonesia. Namun, saat ini proses modeling prediksi untuk radiasi matahari belum dilakukan pada sistem INA Agro-GARLIC sehingga perubahan radiasi matahari dari waktu ke waktu belum sepenuhnya memengaruhi hasil evaluasi lahan bawang putih. Oleh karena itu, diperlukan model prediksi radiasi matahari yang mampu merepresentasikan karakteristik data yang bersifat nonlinear serta memiliki ketergantungan temporal dan spasial. Model Long Short-Term Memory (LSTM) telah banyak digunakan untuk memodelkan ketergantungan temporal pada data deret waktu, namun belum mengakomodasi hubungan spasial wilayah pengamatan. Penelitian ini bertujuan membangun serta membandingkan kinerja model Long Short-Term Memory (LSTM) dan Convolutional Long Short-Term Memory (ConvLSTM) dalam memprediksi radiasi matahari. Data yang digunakan merupakan data radiasi matahari hasil statistical downscaling berbasis Random Forest pada domain spasial berukuran 3×3 grid. Model dibangun menggunakan data historis sepanjang 252 langkah waktu untuk memprediksi 12 langkah waktu berikutnya. Optimasi hiperparameter dilakukan menggunakan Optuna, sedangkan evaluasi model dilakukan menggunakan Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), koefisien determinasi (R²), serta paired Moving Block Bootstrap untuk mengevaluasi perbedaan kinerja kedua model. Hasil penelitian menunjukkan bahwa model LSTM memperoleh nilai MAE sebesar 40,0817 W/m², RMSE sebesar 59,1422 W/m², MAPE sebesar 23,12%, dan R² sebesar 0,8061. Sementara itu, model ConvLSTM memperoleh nilai MAE sebesar 39,7404 W/m², RMSE sebesar 58,7890 W/m², MAPE sebesar 22,55%, dan R² sebesar 0,8084. Hasil paired Moving Block Bootstrap menunjukkan bahwa interval kepercayaan 95% untuk selisih MAE, RMSE, dan R² masih mencakup nol, sedangkan interval kepercayaan selisih MAPE tidak mencakup nol. Hasil tersebut menunjukkan bahwa pemodelan struktur spasio-temporal menggunakan ConvLSTM belum menghasilkan peningkatan kinerja yang signifikan pada seluruh metrik dibandingkan LSTM untuk domain spasial 3×3 yang digunakan. | |
| dc.description.abstract | Solar radiation is a dynamic variable factor in INA Agro-GARLIC as a garlic land suitability recommendation system in priority areas for garlic land development in Indonesia. Currently, however, the predictive modeling process for solar radiation has not yet been implemented in the INA Agro-GARLIC system, meaning temporal changes in solar radiation do not yet fully influence the garlic land suitability evaluation results. Therefore, a solar radiation prediction model is needed that is capable of representing nonlinear data characteristics as well as having temporal and spatial dependencies. The Long Short-Term Memory (LSTM) model has been widely used to model temporal dependencies in time series data, but it has not accommodated the spatial relationships of the observation region. This study aims to build and compare the performance of Long Short-Term Memory (LSTM) and Convolutional Long Short-Term Memory (ConvLSTM) models in predicting solar radiation. The data used are solar radiation data resulting from Random Forest-based statistical downscaling on a 3×3 grid spatial domain. The models are built using historical data over 252 time steps to predict the next 12 time steps. Hyperparameter optimization is performed using Optuna, while model evaluation is conducted using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), coefficient of determination (R²), and paired Moving Block Bootstrap to evaluate the performance differences between the two models. The results show that the LSTM model achieves an MAE of 40.0817 W/m², an RMSE of 59.1422 W/m², a MAPE of 23.12%, and an R² of 0.8061. Meanwhile, the ConvLSTM model achieves an MAE of 39.7404 W/m², an RMSE of 58.7890 W/m², a MAPE of 22.55%, and an R² of 0.8084. The paired Moving Block Bootstrap results show that the 95% confidence intervals for the differences in MAE, RMSE, and R² still include zero, whereas the confidence interval for the difference in MAPE does not include zero. These results indicate that spatio-temporal structure modeling using ConvLSTM has not yielded a significant performance improvement across all metrics compared to LSTM for the 3×3 spatial domain used. | |
| dc.description.sponsorship | | |
| dc.language.iso | id | |
| dc.publisher | IPB University | id |
| dc.title | Model Prediksi Radiasi Penyinaran untuk Evaluasi Kesesuaian Lahan Bawang Putih (INA Agro-GARLIC) | id |
| dc.title.alternative | Radiation Prediction Model for Garlic Land Suitability Evaluation (INA Agro-GARLIC) | |
| dc.type | Skripsi | |
| dc.subject.keyword | long short term memory | id |
| dc.subject.keyword | convolutional lstm | id |
| dc.subject.keyword | deep learning | id |
| dc.subject.keyword | solar radiation | id |
| dc.subtype | Undergraduate Theses | |