基于LSTM网络的光伏发电功率短期预测方法的研究  

Short-term forecasting method of photovoltaic power based on LSTM

作　　者：宋绍剑[1] 李博涵 Song Shaojian;Li Bohan(College of Electrical Engineering,Guangxi University,Nanning 530004,China)

机构地区：[1]广西大学电气工程学院,广西南宁530004

出　　处：《可再生能源》2021年第5期594-602,共9页Renewable Energy Resources

基　　金：国家自然科学基金(61863003);广西自然科学基金(2016GXNSFAA380327)。

摘　　要：提高光伏发电功率预测结果的精度对电网规划和调度具有重要意义。基于前向神经网络或回归分析法的传统预测模型因缺乏历史记忆能力而导致自身鲁棒性较差、适应能力较弱。为了解决上述问题,文章提出了一种基于LSTM网络的光伏发电功率短期预测方法。在预处理过程中,文章先将天气类型依据日照晴朗指数量化为具体数值;然后,利用主成分分析法将与光伏发电功率相关性较高的多元数据序列进行降维,得到主成分数据序列;最后,建立基于LSTM网络的光伏发电功率短期预测模型,并将该模型的预测结果与BP网络预测模型和RNN网络预测模型的预测结果进行对比。模拟结果表明,基于LSTM网络的光伏发电功率短期预测模型能较好地反映时序数据的动态特性,预测精度较高,预测结果能够为电力调度部门提供可靠的数据支持。It is vital for power grid planning and dispatching to improve the accuracy of photovoltaic power generation forecasting.Considering the poor robustness and adaptability of the traditional prediction models which based on feedforward neural network or regression analysis with the lack of historical memory ability,a short-term forecasting model of photovoltaic power based on long short-term memory(LSTM)network was proposed.The weather type was quantified into specific values according to sunshine sunny index in the preconditioning process.The principal component data sequence was obtained by reducing the dimension of the multivariate data sequence which had a high correlation with the photovoltaic power generation by principal component analysis(PCA).Establishing the prediction model of photovoltaic power generation based on LSTM network,and comparing with the prediction results of back propagation(BP)model and recurrent neural network(RNN)model.The model we proposed can reflect the dynamic characteristics of time series data more effectively with the higher prediction accuracy,that provids more reliable data support for power dispatching department.

关 键 词：光伏发电功率预测 主成分分析法 长短期记忆 神经网络 

分 类 号：TK51[动力工程及工程热物理—热能工程]