An Integrated Stacked Sparse Autoencoder and CNN-BLSTM Model for Ultra-Short-Term Wind Power Forecasting with Advanced Feature Learning

With the increasing integration of renewable energy sources into the power grid, accurate and reliable ultra-short-term forecasting of wind power is critical for optimizing grid stability and energy efficiency, especially for a highly dynamic and variable environment. This paper combines Stacked Sparse Autoencoders (SSAE) with a Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BLSTM) architecture to address this challenge, which forms a novel deep learning framework, namely hybrid Stacked Sparse Autoencoder and Convolutional neural network-Bidirectional CNN-BLSTM with advanced Feature selection (SSACBF). The process starts with rigorous data preprocessing and key variable selection through a three-step approach based on expert and statistical methods. The framework employs a stacked sparse multi-layer CNN autoencoder to distil inputs into a robust feature set capturing complex temporal dependencies. These features are then processed by a CNN-BLSTM model, which leverages CNN layers for spatial-temporal nuances and BLSTM layers to simultaneously learn from past and future data. The approach significantly outperforms existing models in accuracy and efficiency, demonstrating potential for real-time applications in wind farm operational planning and energy management systems.