摘要
针对光伏发电系统预测精度不高等问题,建立以门控循环单元(GRU)为基础的预测模型。使用社交网络搜索算法(social network search)和注意力机制(attention)相结合对构建的门控循环单元进行参数优化,采用K-均值对天气类型进行划分,提出材料生成算法(material generation)对变分模态分解中的模态分解数量和惩罚因子进行寻优来确定最佳组合,实现对初始数据的分解操作。利用社交网络搜索算法超参数优化后的门控循环单元对时序特征进行提取,引入注意力机制对时序输入中重要信息的关注进行加强。选用新疆某光伏电站2021年运行数据进行分析,仿真结果表明:所提出的MGA-VMD-SNSAttention-GRU预测模型能有效提升光伏输出功率预测精度。与SVR、Elman模型相比,平均MAPE分别降低8.14%和8.59%。
A prediction model based on gated recurrent unit is developed to address the problem of low prediction accuracy of photovoltaic power generation systems.A social network algorithm and an attention mechanism are used to optimize the parameters of the gated recurrent unit,K-means is used to classify the weather types,and a material generation algorithm is proposed to find the best combination of the number of mode decompositions and penalty factors in the variational mode decomposition to realize the decomposition of the initial data.The gated recurrent unit after hyperparameter optimization of social network search algorithm is used to extract the temporal features,and an attention mechanism is introduced to enhance the attention to important information in the temporal input.The operation data of a PV plant in south Xinjiang in 2021 is selected for analysis.The simulation results show that the proposed MGA-VMD-SNS-Attention-GRU prediction model can effectively improve the PV output power prediction accuracy.The average MAPE is decreased by 8.14%and 8.59%compared with SVR and Elman models,respectively.
作者
李宏扬
高丙朋
Li Hongyang;Gao Bingpeng(School of Electrical Engineering,Xinjiang University,Urumqi 830049,China)
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2023年第8期292-300,共9页
Acta Energiae Solaris Sinica
基金
新疆维吾尔自治区自然科学基金(2019D01C079)。
关键词
光伏发电
功率预测
门控循环单元
变分模态分解
注意力机制
photovoltaic power generation
power forecasting
gated cycle unit
variational mode decomposition
attention mechanism
