BP神经网络在电线积冰定量预报中的应用  

Application of BP Neural Network in Quantitative Prediction of Wire Icing

作　　者：肖雯 刘春 赵文灿 XIAO Wen;LIU Chun;ZHAO Wencan(Jiangxi Meteorological Service Center,330096,Nanchang,PRC;Jiangxi Meteorological Observatory,330096,Nanchang,PRC;Zhejiang Guangsheng Technology Co.,Ltd.,310016,Hangzhou,PRC)

机构地区：[1]江西省气象服务中心,南昌330096 [2]江西省气象台,南昌330096 [3]浙江广晟信息技术有限公司,杭州310016

出　　处：《江西科学》2022年第4期718-723,共6页Jiangxi Science

基　　金：2017年江西省科研项目“江西省电线积冰定量预报研究”(205001);2019年江西省预报员专项“2018年台风“山竹”作用下一次远距离飑线特征及机理分析”(193020100190)。

摘　　要：对1970—2016年江西省电线积冰数据进行统计分析,得到其时空分布特征,并利用BP神经网络方法对电线积冰直径进行定量预报,得到结论如下:1)江西省电线积冰集中发生在12月到2月,占84.4%;以雨凇型积冰为主,占67.5%;江西省电线积冰事件多发生在赣北、赣中,大值区位于赣西北;2)江西省积冰观测站可分为3类,第1类包括修水、宜春、吉安、宁冈、井冈山、广昌,第2类包括波阳、景德镇、南昌、樟树、贵溪、玉山、南城,第3类包括遂川、赣州、寻乌;分类因子按照重要程度排序为:平均风速、最大风速、纬度、降水量和平均气温;3)利用BP神经网络方法进行电线积冰直径定量预报发现,过滤后的预报效果有小幅提升,过滤前后平均绝对误差分别为2.41 mm和2.24 mm,均方根误差分别为3.17 mm和2.81 mm。BP神经网络方法在10 mm以下电线积冰定量预报中的效果较好,当积冰直径超过10 mm就会出现较大的误差。对于电线积冰的客观定量预报有明显参考价值。The characteristics of the annual variation of the wire icing in Jiangxi province from 1970 to 2016 are analyzed.The BP neural network method is used to quantitatively prediction of the wire icing diameter.The results showed that the wire icing occurred from December to February.And the main type of icing was glaze icing accounting for 67.5%.Most of the wire icing events in Jiangxi province occur in northern and central area.The large value area is located in the northwest of Jiangxi.The wire icing observation stations in Jiangxi province can be divided into three categories.The first category includes Xiushui、Yichun、Ji′an、Ninggang、Jinggangshan and Guangchang.The second category includes Boyang、Jingdezhen、Nanchang、Zhangshu、Guixi、Yushan and Nancheng.The third category includes Suichuan、Ganzhou and Xunwu.According to the importance the factors are sorted by average wind speed,maximum wind speed,latitude,precipitation and average temperature.It is found that the prediction effect after filtering is slightly improved.The MAE before and after filtration are 2.41 mm and 2.24 mm and the RMSE are 3.17 mm and 2.81 mm respectively.BP neural network method has a good effect in the quantitative prediction when the wire icing diameter is below 10 mm.If it exceeds 10 mm there will be a large error.

关 键 词：电线积冰 时空分布 聚类分析 BP神经网络 

分 类 号：P426[天文地球—大气科学及气象学]