基于改进CNN-BiLSTM模型和地磁监测数据的多时间长度GIC预测  

Multiscale GIC Prediction Based on Improved CNNBiLSTM Model and Geomagnetic Monitoring Data

作　　者：蓝东亮 陈延云 吴影 赵淼 王亮 吴伟丽 黄冲 LAN Dongliang;CHEN Yanyun;WU Ying;ZHAO Miao;WANG Liang;WU Weili;HUANG Chong(East China Electric Power Research Institute of China Datang Group Science and Technology Research Institute Co.,Ltd.,Hefei 230088;Anhui ZGD Electric Power Technology Co.Ltd.,Hefei 231299;School of Electrical and Control Engineering,Xi’an University of Science and Technology,Xi’an 710054)

机构地区：[1]中国大唐集团科学技术研究总院有限公司华东电力试验研究院,合肥230088 [2]安徽正广电电力技术有限公司,合肥231299 [3]西安科技大学电气与控制工程学院,西安710054

出　　处：《空间科学学报》2024年第3期488-499,共12页Chinese Journal of Space Science

基　　金：国家电网科技项目(SGXJCJ00KJJS2100582);合肥市关键共性技术研发项目(2021GJ039)共同资助。

摘　　要：太阳风暴在电力系统网络中驱动产生的GIC会影响电力设备和系统的安全运行,严重时还会引发大面积停电事件.预测电网GIC水平能够为电力系统保护措施提供重要参考,然而对这方面的研究仍显不足.为了解决该问题,将卷积神经网络(CNN)与双向长短时记忆(BiLSTM)以及注意力机制相结合,利用空间天气的相关监测信息,提出了大规模电网GIC多时间长度的预测方法.本文在分析太阳风暴驱动产生电网地磁感应电流(Geomagnetically Induced Current, GIC)基础上,构建了GIC预测模型;提出了基于多头注意力机制的CNN-BiLSTM改进模型,对GIC进行预测,并给出了预测流程.采用CNN捕获地磁扰动局部信息,根据BiLSTM综合地磁暴扰动信息的全局特征,综合利用多头注意力机制评估对GIC关键作用的地磁信息片段,实现电网GIC的预测.利用2004年11月8日00:00 LT-20:00 LT巨型磁暴期间DED地磁台站和QGZH地磁台监测数据,应用所提方法对岭澳500 kV变电站GIC进行回归预测.经过训练后, GIC预测相对误差均在12%以内,精度高于其他模型的预测结果.The GIC generated by solar storms driving in power system networks can affect the safe operation of power equipment and systems,and even lead to major power outages.Predicting the level of GIC in power grids can provide an important reference for power system protection measures,but re-search in this area continues to be insufficient.In order to solve this problem,a multi-scale GIC predic-tion method for large-scale power grids is proposed by combining Convolutional Neural Networks(CNN),Bidirectional Long and Short Term Memory(BiLSTM),and attention mechanisms,using rele-vant monitoring information of spatial weather.Firstly,based on the analysis of the mechanism of GIC generated by solar storms,a GIC prediction model is constructed;Secondly,a dual-channel GIC predic-tion architecture based on CNN-BiLSTM is proposed:first,local geomagnetic disturbance information is captured using CNN,then the global characteristics of geomagnetic storm disturbance information are synthesized using BiLSTM,and finally,the geomagnetic information fragments that play a key role in GIC are comprehensively evaluated using the multi-head attention mechanism,achieving the prediction of the power grid GIC.Using monitoring data of the DED geomagnetic station and the QGZH geomag-netic station during the giant magnetic storm from 00:00 LT-20:00 LT on 8 November 2004,the pro-posed method was applied to regression prediction of the GIC of the 500 kV Ling’Ao substation.After 220 rounds of training,the relative error of GIC prediction is within 12%,the accuracy is higher than the prediction results of other models.

关 键 词：太阳风暴 GIC预测 卷积神经网络 地磁数据 

分 类 号：P353[天文地球—空间物理学]