基于Inception-CNN-LSTM的光伏发电输出功率预测模型研究  

RESEARCH ON PREDICTION MODEL OF PV POWER GENERATION OUTPUT BASED ON INCEPTION-CNN-LSTM

作　　者：张芸芸 陈家乐[1] 李铮伟 Zhang Yunyun;Chen Jiale;Li Zhengwei(Shanghai Dongfang Yanhua Energy Saving Technology Service Co.,Ltd.,Shanghai 200333,China;School of Mechanical Engineering,Tongji University,Shanghai 200000,China)

机构地区：[1]上海东方延华节能技术服务股份有限公司,上海200333 [2]同济大学机械与能源工程学院,上海200000

出　　处：《太阳能》2025年第4期69-75,共7页Solar Energy

基　　金：住房和城乡建设部科技示范项目(S20200064)。

摘　　要：输出功率作为光伏发电系统运维时的重要指标,是了解光伏发电系统发电能力和运行情况的重要方式。结合Inception网络的多尺度特征提取能力、卷积神经网络(CNN)的局部特征捕捉能力和长短期记忆网络(LSTM)的时间序列建模能力,提出基于Inception-CNN-LSTM的光伏发电输出功率预测模型,并将其与其他3种模型的预测精度进行了对比。研究结果表明:Inception-CNN-LSTM模型在平均绝对百分比误差、均方根误差变异系数和模型拟合度指标方面均优于传统LSTM模型、CNN-LSTM模型和随机森林模型。该模型在电网电力调度、故障诊断和光伏组件维护方面具有广阔的应用前景,能够为光伏发电系统的高效运行提供有力支持。Output power,as an important indicator for the operation and maintenance of PV power generation systems,is an important way to understand the power generation capacity and operating conditions of PV power generation systems.This paper combines the multi-scale feature extraction ability of Inception network,the local feature capture ability of convolutional neural network(CNN),and the time series modeling ability of long short term memory network(LSTM)to propose a PV power generation system output power prediction model based on Inception CNN-LSTM,and compares its prediction accuracy with the other three models.The research results show that the Inception-CNN-LSTM model outperforms traditional LSTM models,CNN-LSTM models,and random forest models in terms of average absolute percentage error,root mean square error coefficient of variation,and model fitting index.This model has broad application prospects in power dispatching,fault diagnosis,and PV module operation and maintenance,and can provide strong support for the efficient operation of PV power generation systems.

关 键 词：光伏发电 输出功率预测 卷积神经网络 长短期记忆网络 神经网络 

分 类 号：TM615[电气工程—电力系统及自动化]