A comparative analysis of LSTM, SVM, and GSTANN models for enhancing solar power prediction
Solar power prediction is crucial for integrating renewable energy into the grid, but current methods often struggle with accuracy due to the limitations of machine learning algorithms. This study aims to enhance prediction accuracy by comparing the performance of Long Short-Term Memory (LSTM)...
保存先:
| 主要な著者: | , , , , , |
|---|---|
| フォーマット: | Proceeding Paper |
| 言語: | English |
| 出版事項: |
IEEE
2024
|
| 主題: | |
| オンライン･アクセス: | http://irep.iium.edu.my/115179/13/115179_%20A%20comparative%20analysis%20of%20LSTM.pdf http://irep.iium.edu.my/115179/ https://ieeexplore.ieee.org/document/10675536 https://doi.org/10.1109/ICSIMA62563.2024.10675536 |
| タグ: |
タグ追加
タグなし, このレコードへの初めてのタグを付けませんか!
|
| 要約: | Solar power prediction is crucial for integrating
renewable energy into the grid, but current methods often
struggle with accuracy due to the limitations of machine
learning algorithms. This study aims to enhance prediction
accuracy by comparing the performance of Long Short-Term
Memory (LSTM) and Support Vector Machine (SVM) models
using datasets from Hebei, China. The main objective is
identifying the most effective algorithm for precise solar power forecasting. The methodology involves training both models on historical solar power data and evaluating their performance against the Graph Spatial-Temporal Attention Neural Network (GSTANN) benchmark. The SVM model was selected for its superior metrics, achieving an MAE (Mean Absolute Error) of 0.5587, RMSE of 0.9741, and a training time of 0.0157 seconds.
Results show that SVM outperforms GSTANN in 45 and 60-
minute intervals, with MAE, MAPE, and RMSE improvements
of up to 68.62%, 42.65%, and 69.44%, respectively. These
findings suggest that SVM offers a more reliable solution for
solar power prediction, providing valuable insights for further
model enhancements. |
|---|