多策略优化的长短时记忆网络短期多维时序光伏功率预测  

Short-term Multidimensional Time Series Photovoltaic Power Prediction Using Multi-strategy Optimized Long Short-term Memory Neural Network

作　　者：唐波 岳喜稳 朱瑞金 龚雪娇 TANG Bo;YUE Xiwen;ZHU Ruijin;GONG Xuejiao(College of Electrical Engineering,Xizang Agricultural and Animal Husbandry University,Nyingchi 860000,China;Electric Power Research Institute,State Grid Xizang Electric Power Co.,Ltd.,Nyingchi 860000,China)

机构地区：[1]西藏农牧学院电气工程学院,林芝860000 [2]国网西藏电力有限公司电力科学研究院,林芝860000

出　　处：《电力系统及其自动化学报》2024年第11期19-29,共11页Proceedings of the CSU-EPSA

基　　金：西藏自治区科技计划项目(XZ202201ZR0024G)。

摘　　要：光伏发电功率短时内的随机性和间歇性影响光伏并网的稳定性。为提高功率预测精度,提出一种基于互补集合经验模态分解、多策略混合改进松鼠搜索算法和长短时记忆网络的功率组合预测模型。首先,使用互补集合经验模态分解对光伏功率时间序列分解;其次,为获得长短时记忆网络的最佳学习参数,提出种群初始化、非线性捕食和反向学习的多策略混合改进方法,解决原始算法种群多样性差、收敛慢和易陷入局部最优的缺点,并通过CEC 2005基准测试函数验证,具有最优的寻优性能。经过2个实例验证,所提模型显著提高了预测精度。The stochastic and intermittent nature of photovoltaic(PV)power generation within a short period of time affects the stability of PV grid-connection.To improve the accuracy of power prediction,a combined power prediction model based on complementary ensemble empirical mode decomposition,multi-strategy hybrid improved squirrel search algorithm and long short-term memory neural network is proposed in this paper.First,the PV power time series is decomposed using complementary ensemble empirical mode decomposition.Second,a multi-strategy hybrid improved method combining population initialization,nonlinear predation and backward learning is put forward to obtain the optimal learning parameters of long short-term memory neural network,which solves the problems of the original algorithm such as poor population diversity,slow convergence and vulnerability to falling into local optimum.In addition,it is verified by the CEC 2005 benchmark test function,indicating that it has the best optimization performance.Through two practical examples,it is shown that the proposed model significantly improves the prediction accuracy.

关 键 词：自适应算法 光伏发电 功率预测 光伏并网 经验模态分解 

分 类 号：TM615[电气工程—电力系统及自动化]