Wind power generation prediction model based on secondary reconstruction decomposition and CNN-Transformer-AcfPSO-BiLSTM in extreme weather conditions

Extreme weather can cause significant fluctuations in renewable energy output, potentially jeopardizing power system stability and reliability. Existing methods for predicting short-term renewable energy output struggle to capture the nonlinear dynamic characteristics under such conditions. This paper proposes a novel ultra-short-term prediction method for renewable energy that takes extreme weather factors into account.

The method starts by using the maximum information coefficient (MIC) to identify key climatic factors influencing renewable energy production. Subsequently, quadratic reconstruction decomposition and denoising techniques are applied to extract multi-band features, reduce input data dimensionality, and enhance the quality of the sequences.

To prevent the algorithm from getting stuck in local optima, the particle swarm optimization (PSO) algorithm is tailored and utilized to optimize the hyperparameters of the Bidirectional long short-term memory (BiLSTM) network. The BiLSTM model is further combined with Convolutional Neural Network (CNN) and Transformer architectures to improve the prediction of renewable energy output.

The proposed method was tested on extreme weather data collected from a specific region in Xinjiang and demonstrated superior accuracy in ultra-short-term renewable energy predictions under extreme weather conditions, showing great potential for practical engineering applications. Support for this research was provided by the Major Science and Technology Project of Gansu Province (24ZDGA003).

The authors thank State Grid Gansu Electric Power Company, Lanzhou, China, State Grid Zhangye Power Supply Company, Zhangye, China, and Shanghai Yanjing Kechuang Intelligent Technology Co., Ltd., Shanghai, China for their contributions. This article is licensed under a Creative Commons Attribution 4.0 International License, allowing for unrestricted use, sharing, adaptation, distribution, and reproduction in any medium or format, provided that appropriate credit is given to the original authors and the source.