Abstract: To enhance the accuracy of short-term wind power prediction, a extend Long Short-Term Memory (xLSTM)-Transformer model based on Bayesian Optimization (BO) and Feature Fusion (FF) was proposed. The proposed model integrates the temporal processing capability of Long Short-Term Memory (LSTM) network with the dynamic feature fusion ability of Transformer’s self-attention mechanism. By employing BO method, the hyperparameters of the model were efficiently optimized in a fewer iteration, significantly reducing the computational resource requirements. Experimental results show that, compared with single LSTM, Transformer model, Gated Recurrent Unit (GRU) model, and xLSTM-Transformer model without FF and BO, the proposed model performs better in key predictive indicators such as coefficient of determination R², Mean Absolute Error (MAE), and Root Mean Square Error (RMSE).

Key words: wind power prediction, neural network model, Bayesian Optimization (BO), feature fusion, deep learning

摘要： 为提高短期风电功率预测的准确性，提出一种基于贝叶斯优化（BO）和特征融合（FF）的xLSTM（extend Long Short-Term Memory）-Transformer模型。该模型综合应用了长短期记忆（LSTM）网络的时序处理能力和Transformer的自注意力机制的特征动态融合能力。借助BO方法，可在较少的迭代次数条件下优化模型的超参数，显著降低了模型对计算资源的依赖。实验结果表明，与单一的LSTM、Transformer模型、门控循环单元（GRU）模型以及未采用FF和BO的xLSTM-Transformer模型相比，所提模型的决定系数R²、平均绝对误差（MAE）和均方根误差（RMSE）等关键预测指标表现更优。

关键词: 风电功率预测, 神经网络模型, 贝叶斯优化, 特征融合, 深度学习