摘要
为满足对非线性风速序列进行高精度建模的需求,提出了一种基于贝叶斯优化的神经基扩展时间序列分析(Neural Basis Expansion Analysis for Time Series,简称NBEATS)风速预测模型。其中,NBEATS模型是近年发展起来的一种可解释深度学习模型,用于高效提取时间序列中的趋势、周期性等可解释性成分;贝叶斯优化用于NBEATS模型的结构参数选择。实验结果表明,与主流的LSTM、GRU相比,该模型在预测精度上具有显著优势,在测试集上的MSE降低至0.2371、RMSE降低至0.4869。同时,与WOA-NBEATS、GA-NBEATS、QGA-NBEATS和PSO-NBEATS等四种优化模型进行了对比实验,该模型表现出较快的迭代速度和较高的预测精度。此外,该模型还可有效预测风速的趋势和周期性特征,因此兼具可解释性和有效性。
To meet the demand for high-precision modeling of nonlinear wind speed series,this paper proposes a Bayesian optimization-based Neural Basis Expansion Analysis for Time Series(NBEATS)model for wind speed prediction.The NBEATS model,a recently developed interpretable deep learning framework,efficiently extracts interpretable components such as trends and periodicities from time series to enable multi-modal forecasting.Bayesian optimization is employed to select structural parameters of the NBEATS model.Experimental results demonstrate that the proposed model significantly outperforms mainstream LSTM and GRU models in prediction accuracy,reducing the mean squared error(MSE)and root mean squared error(RMSE)on the test set to 0.2371 and 0.4869,respectively.Comparative experiments with four other optimization strategies further validate that Bayesian optimization achieves faster convergence and higher prediction accuracy.Additionally,the model effectively predicts trend and periodic characteristics of wind speed,demonstrating both interpretability and effectiveness.
作者
陈艳华
杨金戈
车金星
CHEN Yanhua;YANG Jinge;CHE Jinxing(School of Science,Jiangxi University of Water Resources and Electric Power,Nanchang 330099,China)
出处
《南昌工程学院学报》
2025年第6期88-94,112,共8页
Journal of Nanchang Institute of Technology
基金
国家自然科学基金资助项目(72471108)
江西省高等学校教学改革研究课题(JXJG-23-18-21)。
