基于IMVMD和BiLSTM-SARIMA组合模型的台区光伏短期发电功率预测  

SHORT TERM PHOTOVOLTAIC POWER GENERATION PREDICTION BASED ON IMVMD AND BILSTM-SARIMA COMBINATION MODEL IN STATION AREA

作　　者：李承皓 杨永标 宋嘉启 张翔颖 徐青山[1,2] Li Chenghao;Yang Yongbiao;Song Jiaqi;Zhang Xiangying;Xu Qingshan(School of Electrical Engineering in Southeast University,Nanjing 210096,China;Nanjing Center for Applied Mathematics,Nanjing 211135,China;NARI Group Corporation(State Grid Electric Power Research Institute),Nanjing 211106,China;Guodian NARI Energy Co.,Ltd.,Nanjing 211106,China)

机构地区：[1]东南大学电气工程学院,南京210096 [2]南京应用数学中心,南京211135 [3]南瑞集团有限公司(国网电力科学研究院有限公司),南京211106 [4]国电南瑞能源有限公司,南京211106

出　　处：《太阳能学报》2025年第2期433-440,共8页Acta Energiae Solaris Sinica

基　　金：国家重点研发计划(2022YFB2703500)。

摘　　要：针对台区分布式光伏短期发电功率预测精度低的难题,提出一种基于增强型鲸鱼优化算法的多元变分模态分解方法,并结合反向传播神经网络耦合双向长短期记忆网络和季节性差分自回归滑动平均的组合模型,实现台区分布式光伏短期发电功率预测。首先对鲸鱼优化算法的收敛因子、权重等进行改进,然后用它去优化多元变分模态分解方法中的通道数量和惩罚因子,得到最佳分解效果的参数值。再针对与外界气象等因素强相关的光伏发电功率时间序列数据,利用改进多元模态分解将序列最优分解。将分解后的各模态分量输入到单独构建的双向长短期记忆网络和季节性差分自回归滑动平均模型中,获取分量预测值,两个模型得到的分量预测值分别叠加得到各自的完整预测结果。将它们分别乘以权重后相加即为最终预测结果,权重通过反向传播神经网络进行修正。仿真结果说明相比于其他方法,所提模型能有效提高光伏短期发电的预测精度。Aiming at the difficulty of low accuracy of short-term power prediction of distributed PV in the station area,a multivariate variational modal decomposition approach based on the enhanced whale optimization algorithm is proposed and combined with a back propagation neural network coupled with bidirectional long and short-term memory network and a seasonal differential autoregressive sliding average is combined to achieve the prediction of short-term power generation of distributed PV in the station area.In the paper,the convergence factor and weight of the whale optimization algorithm are firstly improved,and then it is used to optimize the number of channels and the penalty factor in the multivariate variational modal decomposition method,to get the parameter values of the best decomposition effect.Then for the PV power time series data which is strongly correlated with external meteorological factors,the sequence is optimally decomposed using the improved multivariate modal decomposition.The decomposed modal components are input into the separately constructed bidirectional long-and short-term memory network and seasonal differential autoregressive sliding average model to obtain the component prediction values,and the component prediction values obtained from the two models are superimposed to obtain their respective complete prediction results.The final prediction result is obtained by multiplying them by the weights,which are corrected by the back-propagation neural network.The outcomes of the simulation indicate that compared to other methods,an effective enhancement in the predictive accuracy of short-term photovoltaic power generation with model proposed in the paper.

关 键 词：模态分解 神经网络 光伏发电 预测 BiLSTM SARIMA 

分 类 号：TM615[电气工程—电力系统及自动化]