基于BERT-CRF和数据加密的石油勘探数据脱敏与安全保护研究  

Research on desensitization and security protection of oil exploration data based on named entity identification and data encryption

作　　者：高博 常莉 张金波 孙超 刘锐 GAO Bo;CHANG Li;ZHANG Jinbo;SUN Chao;LIU Rui(CNOOC Research Institute,Beijng 100028)

机构地区：[1]中海油研究总院有限责任公司,北京100028

出　　处：《自动化与仪器仪表》2025年第3期20-24,共5页Automation & Instrumentation

摘　　要：为了避免石油勘探数据上传过程中发生泄漏,提出一种改进BERT-CRF的敏感数据智能识别方法。该方法一方面引入具有双向多层Transformer编码器的BERT进行敏感特征学习;另一方面通过鲸鱼优化算法对BERT-CRF进行超参数优化,提高模型对敏感数据的识别性能。结果表明,改进后的BERT模型在测试集上精确率、召回率和F1值达到97.53%、98.24%、98.01%,相较于原始BERT模型,训练效率和训练效果都有了显著提升,结果更稳定;相较于Transformer-CRF、LSRM-CRF、ELMo-CharCNN-BiLSTM-CRF等较为热门的命名实体识别模型,改进模型的识别精确率分别高出3.44%、1.30%、0.26%,对敏感实体的识别效果更准确;基于改进BERT-CRF敏感数据识别算法构建石油勘探脱敏与安全保护系统,随着脱敏数据的增加,在检测中丢包率始终不超过3.6%,证明了所使用的敏感识别方法可以较为准确地定位与标注出石油勘探数据中的敏感信息,具有一定的实用性。In order to avoid leakage during oil exploration data uploading,an improved BERT-CRF method for sensitive data intelligent identification is proposed.On the one hand,BERT with bi-directional multilayer Transformer encoder is introduced for sensitive feature learning.On the other hand,the hyperparameter optimization of BERT-CRF is carried out by whale optimization algorithm to improve the recognition performance of sensitive data.The results show that the accuracy rate,recall rate and F1 value of the improved BERT model on the test set reach 97.53%,98.24% and 98.01%.Compared with the original BERT model,the training efficiency and training effect have been significantly improved,and the results are more stable.Compared with popular named entity recognition models such as Transformer-CRF,LSRM-CRF and ELMo-CharCNN-BiLSTM-CRF,the improved model has a higher recognition accuracy rate of 3.44%,1.30% and 0.26%,respectively,and has a more accurate recognition effect on sensitive entities.The oil exploration desensitization and security protection system is built based on the improved BERT-CRF sensitive data recognition algorithm.With the increase of the desensitization data,the packet loss rate in the detection is always less than 3.6%,which proves that the sensitive identification method used can locate and label the sensitive information in the oil exploration data more accurately,and has certain practicability.

关 键 词：命名实体识别 数据脱敏 石油勘探 超参数优化 数据加密 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]