A Deep Learning-Based Method for Preventing Data Leakage in Electric Power Industrial Internet of Things Business...

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• The Power Industry Internet of Things (IIoT) faces significant data security challenges, particularly concerning sensitive data interactions.
• Current anti-leakage strategies, relying on regular expressions, are inadequate for unstructured data in power business data interactions.

Purpose of the Study:

• To propose a novel deep learning-based anti-leakage method for securing data interactions within the power business.
• To ensure the security of data exchange between the State Grid business platform and third-party platforms.

Main Methods:

• A hybrid approach combining named entity recognition, regular expressions, and a DeBERTa-BiLSTM-CRF model.
• Utilizing DeBERTa for pre-training feature extraction and BiLSTM for sequence context semantic feature extraction.
• Employing a CRF layer for optimal tag sequence identification of sensitive data.

Main Results:

• The proposed method achieved an F1 score of 81.26% in identifying sensitive data entities on the CLUENER 2020 dataset.
• Demonstrated effective identification of privacy-sensitive information in both structured and unstructured power business data.

Conclusions:

• The deep learning-based method significantly enhances the security of power business data interactions.
• Provides an innovative solution to mitigate the risk of data leakage in the power industry's IIoT ecosystem.