Short-Term Wind Power Prediction via Spatial Temporal Analysis andDeep Residual Networks

Abstract

This study embarks on a rigorous replication of the innovative research conducted by Huajin Li on "Short-Term Wind Power Prediction via Spatial Temporal Analysis and Deep Residual Networks." The original study leveraged graph neural networks (GNNs) and deep residual networks (DRNs) to forecast short-term wind power accurately, focusing on enhancing grid stability and operational efficiency in the wind energy sector. Utilizing the identical dataset and experimental conditions as the original author, this reproduction affirms the robustness and applicability of the advanced machine learning techniques in handling the complex dynamics of wind power data. The models trained, including Neural Networks (NN), Support Vector Regression (SVR), Extreme Learning Machines (ELM), and Deep Residual Networks (DRN), demonstrated almost identical outcomes to the original study, thereby underscoring the predictability and consistency of these methods. This paper provides a detailed comparative analysis, highlighting slight improvements in accuracy and model fit, which illustrates the progression in computational capabilities and optimization techniques since the initial research. The successful replication not only validates the original findings but also enriches the scientific community's confidence in using sophisticated AI-driven approaches for renewable energy forecasting.