Earticle

초록

영어

Solar power generation provides significant environmental and economical advantages, in comparison to nuclear and fossil fuel. Although, due to the unpredictable and intermittent patterns in the data, it is difficult to forecast power generation effectively. Therefore, in this study, we proposed stacked Gated Recurrent Units (SGRU) and deep Convolutional Neural Networks (DCNN) for power generation forecasting. Initially, data preprocessing strategies are applied such as imputing missing values and data normalization, to convert the raw input data into refined formate. The proposed dual SGRUDCNN is then used to learn temporal pattern via SGRU and spatial pattern via DCNN, followed by a feature fusion layer, where the outputs vectors of both networks are integrated into a single representative feature vector and fed to fully connected layers for final forecasting. Furthermore, the effectiveness of the SGRU-DCNN is evaluated via two benchmarks where the SGRU-DCNN achieved optimal performance among state-of-the-art (SOTA) architectures.

목차

Abstract
1. Introduction
2. THE PROPOSED METHOD
3. Experiment result
3.1. Comparative analysis
3.2. Comparison with SOTA methods
4. Conclusions
Acknowledgment
Reference

저자

• Taimoor Khan [ Dept. of Computer Engineering Gachon University ]
• JunHo Yoon [ Dept. of Computer Engineering Gachon University ]
• Chang Choi [ Dept. of Computer Engineering Gachon University ] Corresponding Author

참고문헌