基于Seq2Seq深度学习方法的气象预警纠错模型研究
Research on Meteorological Early Warning Error Correction Model Based on Seq2Seq Deep Learning Method
侯天宇 1张珊 2金峰 1苑超 1陈子煊1
作者信息
- 1. 天津市突发公共事件预警信息发布中心 天津 300300
- 2. 天津市气象信息中心 天津 300074
- 折叠
摘要
针对全国气象预警信息发布语义类错误,研发一种预警信息纠错模型.通过建立全国气象历史预警信息语料库,训练基于Seq2Seq深度学习方法的纠错模型,并与基于统计方法的规则模型相互验证,形成预警预报信息合法性监测质控平台,构建"智能语义分析+人工验证"的质控业务流程,实现敏感词的快速定位与提醒.预警质控平台业务应用后,信息内容错情率较上一年降低 70%,语义纠错效果显著.
Abstract
An error correction model of early warning information is developed for semantic errors in the release of national meteorological early warning information.By establishing a national meteorological historical warning information corpus,training an error correction model based on Seq2Seq deep learning method,and verifying it with a rule model based on statistical methods,a quality control platform for monitoring the legitimacy of warning and forecasting information was formed.A quality control business process of"intelligent semantic analysis+manual verification"was constructed to achieve rapid localization and reminder of sensitive words.Since the early warning quality control platform began to be applied,the error rate of information content decreased by 70%compared with the previous year,and the semantic error correction effect was remarkable.
关键词
预警发布/语义分析/Seq2Seq深度学习/预警合法性监测/质控模型Key words
early warning release/semantic analysis/Seq2Seq deep learning/early warning legality monitoring/quality control model引用本文复制引用
基金项目
安全天津、科技惠民与可持续发展实验区建设科技专项(17ZXCXSF00060)
出版年
2024
NETL
NSTL
万方数据