基于STL分解和TPA机制的光伏功率区间预测  

PV POWER INTERVAL PREDICTION BASED ON STL DECOMPOSITION AND TPA MECHANISM

作　　者：李逸航 肖辉[1] 易纯 龙飞宇 Li Yihang;Xiao Hui;Yi Chun;Long Feiyu(National Key Laboratory of Disaster Prevention and Reduction for Power Grid,Changsha University of Science and Technology,Changsha 410114,China)

机构地区：[1]长沙理工大学电网防灾减灾全国重点实验室,长沙410114

出　　处：《太阳能学报》2024年第12期22-29,共8页Acta Energiae Solaris Sinica

基　　金：长沙市自然科学基金(kq2202213)。

摘　　要：针对光伏功率点预测包含的信息不足,无法对电网的调度提供充分依据的问题,提出一种基于STL分解和TPA机制的光伏功率预测方法。首先将原有光伏功率序列进行STL分解,得到趋势项、季节项以及残差项3类子序列。接着通过极限学习机(ELM)对趋势项进行预测;采用基于时间模式注意力机制(TPA)的双向门控循环单元(BiGRU)对季节项以及残差项进行预测;最后通过分位数回归获得区间预测结果,二者区间结果叠加获得光伏输出区间预测结果。在湖南某地光伏输出数据集上进行算例实测,通过点预测结果及区间预测结果验证了所提方法的有效性。Aiming at the problem that the PV power point prediction contains insufficient information to provide a sufficient basis for grid scheduling,a PV interval prediction method based on STL decomposition and temporal pattern attention mechanism is proposed.Firstly,the original PV power sequence is decomposed by STL to obtain subsequences:trend term,seasonal term and residual term.Then,the trend term is predicted by extreme learning machine(ELM);the seasonal term and the residual term are predicted by bi-directional gated recurrent unit(BiGRU)based on temporal pattern attention(TPA).Finally,the interval prediction is obtained by quantile regression,and the PV output interval prediction is obtained by superposing the two outputs.The proposed method is validated by the point prediction results and interval prediction results on the PV output dataset of a certain place in Hunan province.

关 键 词：光伏发电 功率预测 神经网络 分位数回归 双向门控循环单元网络 

分 类 号：TM615[电气工程—电力系统及自动化]