基于Spearman系数和TCN的光伏出力超短期多步预测  

ULTRA-SHORT-TERM PHOTOVOLTAIC POWER MULTI-STEP PREDICTION BASED ON SPEARMAN COEFFICIENT AND TCN

作　　者：吴珺玥 赵二刚[1,2,3] 郭增良 张亚萍[4] 张建军 Wu Junyue;Zhao Ergang;Guo Zengliang;Zhang Yaping;Zhang Jianjun(College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China;Engineering Research Center of Thin Film Opto Electronics Technology,Ministry of Education,Tianjin 300350,China;Key Laboratory of Photoelectronics Thin Film Devices and Technology,Tianjin 300350,China;Tianjin Green Future Energy Management Co.,Ltd.,Tianjin 300350,China)

机构地区：[1]南开大学电子信息与光学工程学院,天津300350 [2]薄膜光电子技术教育部工程研究中心,天津300350 [3]天津市光电子薄膜器件与技术重点实验室,天津300350 [4]天津绿动未来能源管理有限公司,天津300350

出　　处：《太阳能学报》2023年第9期180-186,共7页Acta Energiae Solaris Sinica

摘　　要：研究一种基于Spearman相关系数和改进时间卷积网络(TCN)的超短期多步光伏功率预测方法。首先,采用Spearman相关系数方法对输入的天气特征量进行筛选;然后,构建合适的时间卷积网络使其适配光伏功率预测问题。经过实际的光伏电站数据测试,单步预测模型拟合度为99.41%,预测平均绝对误差为61.04,均优于传统的长短期记忆神经网络(LSTM)。This paper developed a ultra-short-term photovoltaic power prediction model based on Spearman coefficient and improved TCN.First,we used Spearman coefficient to screen the input weather characteristics,and then we built a proper TCN to make it suitable for the photovoltaic power prediction problems.Through the actual data test of photovoltaic power station,the fitting degree of the model is 99.41%,and MAE of prediction is 61.04,which is better than the traditional time series problem model LSTM.

关 键 词：光伏发电 预测 神经网络 数据处理 时间序列 

分 类 号：TM615[电气工程—电力系统及自动化]