A Novel Framework for Short-Term Wind Power Prediction with RL-based Hyper-parameter Optimization

A Novel Framework for Short-Term Wind Power Prediction with RL-based Hyper-parameter Optimization

Enhancing the accuracy and effectiveness of wind power forecasting is pivotal for optimizing renewable energy integration and grid stability. This paper proposes a pioneering methodology that uses deep convolutional neural network (DCNN), energized with hyper-parameters optimization via a deep Q-network (DQN) for on-shore wind power prediction. DCNN has the inherent capability to automatically learn complex temporal patterns and spatial correlations within data providing unprecedented advantages for time series forecasting. The integration of optimizing framework facilitates a balance between exploration and exploitation of hyper-parameters resulting in optimal solutions to enhance model performance. To assess the effectiveness of the proposed model, data sets of on-shore wind power generation (WPG) from two European nations are used where the proposed model outperforms the other benchmark models. This analysis reveals a remarkable improvement of the proposed model i.e. 42.10% and 75.22% in root mean squared error (RMSE) and mean absolute error (MAE) respectively compared to the persistence model. Additionally, our approach demands a higher computational time however, the remarkable leap in accuracy signifies a new horizon in on-shoreWPG forecasting.

The accessible file is the accepted version of the paper. Please refer to the URL for the published version.