Improved CGAN for Short-Term Wind Power Forecasting based on Heat Maps and LSTM-CNN

Wind power generation plays a crucial role in maintaining grid stability and achieving sustainable energy goals, making accurate forecasting of electricity generation capacity essential for efficient power system operation. Traditional Conditional Generative Adversarial Networks (CGAN) face limitations in condition selection and spatiotemporal feature extraction, affecting their forecasting accuracy. This paper presents a modified model that addresses these challenges through an innovative integration of deep learning architectures. The proposed model employs a data-driven approach to select weather conditions with the highest correlation as conditional information, utilizing heat map analysis to identify the most significant predictive factors. Temporal characteristics are captured through Long Short-Term Memory (LSTM) networks, while spatial features are extracted using Convolutional Neural Networks (CNN), creating a comprehensive framework for feature analysis. The model's effectiveness is validated through extensive experiments across different seasons using real-world data from operational wind farms. Results demonstrate superior forecasting precision compared to conventional methods, with significant improvements in Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The enhanced generalization capability and robust performance make this model particularly suitable for power grid scheduling applications, contributing to more reliable and efficient renewable energy integration.