Development of an AI-Based Model with Low Computational Complexity for Accurate Wind Energy Forecasting

Abstract

Most countries primarily relay on fossil fuel for electricity generation, leading to fossil fuel depletion and environmental pollution. The countries are developed technologies for renewable energy generation. The wind energy being promoted as a superior renewable energy. However, wind energy has its challengers, particularly uncertainty that can affect overall system stability. The accurate short-term forecasting of wind energy was crucial for ensuring grid stability. Both physical and AI-based models can effectively be utilized for wind energy prediction. AI-based methodologies have shown superior effectiveness, efficiency, and accuracy when compared to traditional physical models. The lightweight AI-based forecasting model was particularly significant for processing devices, enabling faster computations and substantially more cost-effective forecasting. The research utilized simulation software to develop an Artificial Neural Network (ANN) model, initially incorporating eight meteorological parameters. Four of these parameters showed weak correlations and were subsequently removed from the model. Further optimization was achieved through pruning and quantization techniques, significantly reducing computational complexity. The optimized model demonstrates a notable reduction in both training time by 92.69% and inference time by 63.83%, while maintaining accuracy with only a marginal decrease of 3.99% compared to the initial model. These improvements were achieved with minimal loss in predictive accuracy, significantly reducing computational complexity. The study concludes that the optimized ANN model is wellsuited for real-time wind power forecasting, offering a balance between accuracy and computational efficiency. This approach not only facilitates better grid management but also extends the applicability of AI-based forecasting to devices with limited processing capabilities. Future work could explore additional complexity reduction techniques and broader deployment scenarios.