基于改进STGCN与N-BEATS的风功率超短期预测  

Wind power ultra⁃short⁃term prediction based on improved STGCN with N⁃BEATS

作　　者：程旭初 刘景霞[1] 康荣凯 CHENG Xuchu;LIU Jingxia;KANG Rongkai(School of Automation and Electrical Engineering,Inner Mongolia University of Science and Technology,Baotou 014010,China)

机构地区：[1]内蒙古科技大学自动化与电气工程学院,内蒙古包头014010

出　　处：《现代电子技术》2025年第8期115-121,共7页Modern Electronics Technique

基　　金：内蒙古自治区自然科学基金(2020LH05019)。

摘　　要：精准的风功率预测对电网调度具有重大意义,针对现有预测方法中数据特征提取不充分、输入序列过长时产生梯度消失和预测精度低的问题,提出一种基于改进时空图卷积(STGCN)与神经基扩展分析(N-BEATS)模型的组合预测模型,该方法通过充分提取数据时空特征来提高预测精度。首先,利用STGCN对多元输入序列进行深度特征提取,充分挖掘风机SCADA数据中的时空潜在关系;同时,为了进一步提高预测精度,通过构建序列分解模块与多分辨率卷积对STGCN模型进行改进,使其能够更好地适应风电数据的复杂特性;然后,神经基扩展分析(N-BEATS)新型神经网络对STGCN提取的时空信息数据进行时序关系分析,得到最终预测结果;最后,以内蒙古某风场SCADA数据为例,通过多模型对比实验与自身消融实验验证了所提组合模型策略的有效性以及对STGCN的改进效果。实验结果表明,所设计模型在预测精度上取得了显著的提升,为风电功率预测领域的研究提供了新的思路和方法。Accurate wind power forecasting is of great significance for grid dispatching.In order to solve the problems of insufficient data feature extraction,gradient vanishing when the input sequence is too long,and low prediction accuracy in the existing prediction methods,a combined prediction model based on improved spatio-temporal graph convolution network(STGCN)and neural basis expansion analysis for interpretable time series(N-BEATS)model is proposed,which can improve the prediction accuracy by fully extracting the spatio-temporal feature of the data.The STGCN is used to conduct the deep feature extraction of multivariate input sequences,so as to fully exploit potential spatial-temporal relationships in the SCADA data of wind turbines.In order to further improve the prediction accuracy,the STGCN model is improved by constructing a sequence decomposition module and multi-resolution convolution,so as to make it better adapt to the complex characteristics of wind power data.The N-BEATS new neural network is used to analyze the temporal series relationship of the spatio-temporal information data extracted by STGCN,so as to obtain the final prediction results.Taking SCADA data of a wind farm in Inner Mongolia as an example,the effectiveness of the proposed combined model strategy and the improvement effect of STGCN are verified by means of the multi-model comparison experiments and self-ablation experiments.The experimental results show that the designed model can realize significant improvement in prediction accuracy,which provides new ideas and methods for the research in the field of wind power prediction.

关 键 词：超短期风功率预测 时空图卷积 神经基扩展分析 序列分解 深度特征提取 图卷积网络 

分 类 号：TN919-34[电子电信—通信与信息系统] TM614[电子电信—信息与通信工程]