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

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。

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.