Multi-step wind power forecasting over 24 hours using a CNN-LSTM hybrid model

Long Duong Hoang Vu; Quan Phung Mac; Kien Tran Ngoc; Van Nguyen Dinh; Ninh Nguyen Tuan

doi:10.64032/mca.v29i4.345

Authors

Long Duong Hoang Vu School of Electrical and Electronic Engineering, Hanoi University of Science and Technology
Quan Phung Mac School of Electrical and Electronic Engineering, Hanoi University of Science and Technology
Kien Tran Ngoc Queensland University of Technology
Van Nguyen Dinh School of Electrical and Electronic Engineering, Hanoi University of Science and Technology
Ninh Nguyen Tuan School of Electrical and Electronic Engineering, Hanoi University of Science and Technology

DOI:

https://doi.org/10.64032/mca.v29i4.345

Keywords:

Convolutional Neural Network, Long Short-Term Memory, Wind Power Prediction, Multi-Step Forecasting

Abstract

This study investigated three deep learning architectures for wind power forecasting: LSTM, CNN, and hybrid CNN-LSTM. Traditional statistical models often fail to capture the nonlinear temporal dependencies inherent in wind power data. In contrast, neural-network-based approaches offer improved accuracy, with hybrid models showing potential for further enhancement. Experiments using real-world data from a wind power plant in Soc Trang, Vietnam, showed that the CNN-LSTM model consistently outperformed standalone CNN and LSTM networks. Although the results align with the findings of prior research, this study provides a more focused and systematic evaluation of deep learning techniques for short-term wind power prediction.