序列视角下的短期风功率渐进式预测方法  

ASYMPTOTIC PREDICTION METHOD FOR SHORT-TERM WIND POWER IN SEQUENTIAL PERSPECTIVE

作　　者：高宇航 姜爱华[1] 田君杨 甘定福 Gao Yuhang;Jiang Aihua;Tian Junyang;Gan Dingfu(Guangxi Key Laboratory of Power System Optimization and Energy Technology(Guangxi University),Nanning 530004,China;Power Dispatching Control Center of Guangxi Power Grid,Nanning 530023,China)

机构地区：[1]广西电力系统最优化与节能技术重点实验室(广西大学),南宁530004 [2]广西电网电力调度控制中心,南宁530023

出　　处：《太阳能学报》2025年第3期504-513,共10页Acta Energiae Solaris Sinica

摘　　要：针对短期风功率预测现有预测方法缺少对序列子过程时延相似性的考虑、预测结果受制于序列分解效果以及预测精度低等问题,提出基于小波软阈值去噪(WSTD)和改进Autoformer的组合预测方法。首先,使用小波软阈值去噪对原始数据进行预处理,减少噪声对预测精度的影响;其次,将具有自相关机制的Autoformer模型应用于短期风功率预测,在序列视角下挖掘周期依赖关系;最后,基于多级离散小波变换构建深度分解架构对Autoformer模型进行改进,提高Autoformer模型对复杂时间模式的分解能力。实验结果表明,所提组合模型预测精度优于单一的模型且具有良好的适应性,在4个季节的算例中,与Autoformer模型相比RMSE和MAE指标平均下降19.86%和19.07%,R2平均提高5.15%。Short-term wind power prediction plays a crucial role in formulating effective power system production scheduling plans,but the inherent stochastic nature and volatility of wind power output present challenges to achieving high prediction accuracy.This paper addresses several limitations prevalent in existing prediction methods,including the neglect of time delay similarity in sequential subprocesses,reliance on sequence decomposition effects,and the suboptimal accuracy of current approaches.In response to these challenges,the WSTD-improved Autoformer combination prediction method is proposed.Firstly,the preprocessing of raw data is performed using wavelet soft threshold denoising to mitigate the impact of noise on prediction accuracy.Secondly,the Autoformer model,characterized by its autocorrelation mechanism,is employed for short-term wind power prediction,exploring periodic dependencies from a sequential perspective.Additionally,a deep decomposition architecture based on multi-level discrete wavelet transforms is constructed to enhance the Autoformer model's capability to decompose intricate temporal patterns.Experimental results demonstrate that the proposed combination prediction model outperforms individual models and exhibits excellent adaptability.In the four seasonal scenarios,the RMSE and MAE metrics decreased by an average of 19.86%and 19.07%,and the R2 increased by an average of 5.15%compared to the Autoformer model.

关 键 词：风功率 小波变换 自相关 改进Autoformer模型 短期预测 

分 类 号：TK83[动力工程及工程热物理—流体机械及工程]