基于扩散模型和爬坡趋势分类的风电功率自适应区间预测  

Adaptive Interval Forecasting of Wind Power Based on Diffusion Model and Ramping Trend Classification

作　　者：韩丽[1] 程颖洁 王施琪 陈硕[1] HAN Li;CHENG Yingjie;WANG Shiqi;CHEN Shuo(School of Electrical Engineering,China University of Mining and Technology,Xuzhou 221116,Jiangsu Province,China)

机构地区：[1]中国矿业大学电气工程学院,江苏省徐州市221116

出　　处：《电网技术》2024年第6期2448-2457,I0051-I0054,共14页Power System Technology

基　　金：国家自然科学基金项目(62076243);徐州市科技项目(KC22343)。

摘　　要：扩散模型基于马尔可夫链的概率性质,能够定量描述风电的随机性和不确定性。然而,传统基于扩散模型的时序预测方法以当前输入前一段样本的均值作为基准进行特征缩放,导致预测区间在高峰时段过大、低谷时段过小。因此,提出一种基于扩散模型和爬坡趋势分类的风电功率自适应区间预测方法。首先,利用基于扩散模型的区间预测框架获取初始预测区间。然后,将风电波动过程划分为6种模式,对不同模式下的预测区间采取自适应规整策略,进而获得初始改进区间。接着,针对高出力模式中非爬坡时段的区间带宽不匹配问题,建立爬坡趋势分类评估模型,并结合所属出力模式进行区间修正,获得最终的区间预测结果。最后,实验结果表明所提方法的区间预测效果更优。Diffusion models,based on the probabilistic properties of Markov chains,can quantitatively describe the stochasticity and uncertainty of wind power.However,the traditional diffusion model-based time-series forecasting method uses the average of the current input previous period samples as the benchmark for feature scaling,resulting in a forecasting interval that is too large during peak periods and too small during trough periods.Thus,an adaptive interval forecasting method for wind power based on diffusion model and ramping trend classification is proposed.First,the diffusion model-based interval forecasting framework obtains the initial forecasting intervals.Then,the wind power fluctuation process is divided into six patterns,and an adaptive regularization strategy is adopted for the forecasting intervals under different patterns,to obtain the initial improved intervals.Then,to address the problem of interval bandwidth mismatch during non-ramping periods in high output patterns,an evaluation model for ramping trend classification is established,and the intervals are corrected by combining them with the affiliated output patterns to obtain the final interval forecasting results.Finally,the experimental results show that the interval forecasting effect of the proposed method is better.

关 键 词：扩散模型 自适应规整 波动特征 爬坡趋势分类 区间预测 

分 类 号：TM614[电气工程—电力系统及自动化]