运用Dropout-GRU模型的短期负荷预测  

Short-term load forecasting by using Dropout-GRU

作　　者：闫方 吕梦娜 杨文艺 张顺利[1] 王丹阳 YAN Fang;LYU Mengna;YANG Wenyi;ZHANG Shunli;WANG Danyang(College of Computer Science and Technology,Henan Institute of Science and Technology,Xinxiang 453000,China;College of Information Engineering,Henan Geology Mineral College,Zhengzhou 450000,China;Xinxiang Power Supply Company,State Grid Henan Electric Power Company,Xinxiang 453000,China;Second Clinical College,Xinxiang Medical University,Xinxiang 453000,China)

机构地区：[1]河南科技学院计算机科学与技术学院,河南新乡453000 [2]河南地矿职业学院信息工程学院,河南郑州450000 [3]国家电网河南省电力公司新乡供电公司,河南新乡453000 [4]新乡医学院第二临床学院,河南新乡453000

出　　处：《电子设计工程》2024年第24期124-128,共5页Electronic Design Engineering

基　　金：河南省科技攻关项目(222102210020)。

摘　　要：为提高短期电力负荷预测精度,提出采用Dropout-GRU模型的短期负荷预测方法。该方法基于Python爬虫获取对负荷预测有影响的多种气象因素,降低人为采集数据时由于主观因素导致数据错误的可能性;构建多层GRU网络以充分挖掘波动较大的负荷数据之间的非线性关系;在GRU网络的非循环部分加入Dropout技术,使神经元按照一定概率失活,有效解决了多层GRU网络易产生的过拟合问题,从而提高短期负荷预测精度。以某县负荷数据为例进行实验可知,该文方法的MAPE、RMSE和MAE相比单纯GRU网络分别降低58.90%、61.54%和58.17%,说明该文预测方法效果更佳。To improve the accuracy of short-term power load forecasting,the method of short-term load forecasting by using Dropout-GRU is proposed.The method is based on python crawlers to obtain various meteorological factors that have an impact on load forecasting,reducing the possibility of data errors caused by subjective factors during manual data collection.The multi-layer GRU is constructed to fully explore the nonlinear relationships between load data with large fluctuations.Dropout was added to the non recurrent part of the GRU to make neurons inactivate with a certain probability,effectively solving the overfitting problem that multi-layer GRU,thereby improving the accuracy of short-term load forecasting.Taking the load data of a county as an example for the experiment,the MAPE,RMSE and MAE of the method proposed were reduced by 58.90%,61.54%and 58.17%respectively compared to the simple GRU,indicating that the prediction method in this paper has better performance.

关 键 词：短期负荷预测 Dropout技术 GRU网络 气象因素 过拟合 

分 类 号：TN99[电子电信—信号与信息处理]