Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{208473,
        author = {Priya Savaleram Borhade and Shivam Amrut Kale and Siddhi Ravindra Warade},
        title = {Cross-Protocol Domain Adaptation and Anomaly Generalization in Smart Grid Intrusion Detection Systems Using Tabular Transformers},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {4},
        pages = {1858-1862},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=208473},
        abstract = {Contemporary smart-grid infrastructures utilize a diverse range of industrial control and communication protocols, including Modbus, DNP3, and IEC 60870-5-104. Machine-learning-based Intrusion Detection Systems (IDSs) can operate effectively when the distributions of their training and deployment data are closely aligned. However, performance may degrade when a model trained for one protocol is applied to another, as differences in telemetry distributions, feature semantics, packet formats, and reporting behaviors can arise. This study presents a domain-adaptive tabular Transformer framework designed for anomaly detection across smart-grid protocols. The framework initially tokenizes both heterogeneous numerical and categorical attributes and subsequently employs a Transformer encoder to model the contextual dependencies among telemetry features. By leveraging labeled source data alongside unlabeled target data through a domain-adversarial component equipped with a Gradient Reversal Layer (GRL), the model fosters latent representations that are less reliant on the specific communication domain. The source domain anomaly classifier is trained using binary cross-entropy, while the domain discriminator is tasked with distinguishing between source and target representations. During cross-protocol transfer, the proposed framework is evaluated against XGBoost, a multilayer perceptron, and a non-adaptive FT-Transformer. During evaluation on the target domain, the model achieved a precision of 0. 941, recall 0. 923, F1-score of 0. 932, a false-positive rate of 2. 3%, and an ROC-AUC of 0. 968. The results highlight the promise of integrating tabular Transformers with adversarial domain adaptation to enhance the generalization of intrusion detection systems across diverse smart-grid protocols.},
        keywords = {Smart Grid Security, Intrusion Detection Systems, Tabular Transformers, Domain Adaptation, Cross-Protocol Generalization, Cyber-Physical Systems.},
        month = {September},
        }