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{191604,
        author = {ESRAT SHARMIN ALIZA},
        title = {Anomaly Detection in Industrial Energy Usage for Sustainability Compliance},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {8},
        pages = {7206-7215},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=191604},
        abstract = {The growing infiltration of renewable resources, modern metering systems and distributed energy sources have not only contributed greatly to the observability of the modern power systems but have also led to the growing complexity and vulnerability of operations. The subsequent developments require effective anomaly detection controls that can detect abnormal load behavior in the non-stationary operating regime of considerable imbalances. The paper introduces analytical information of real-time smart grid load anomaly detection using unsupervised machine learning and deep learning. The suggested model combines preprocessing of systematic data, feature engineering and time series modeling to identify anomalous consumption behavior based on multivariate smart grid load data. Three demonstrative models (Isolation Forest, Dense Autoencoder, and Transformer Autoencoder) are applied and tested in a single experimental framework using an open-source dataset on monitoring smart grid loads. Accuracy, precision, recall, F1-score and complementary visual analysis is used to determine model performance in terms of accuracy, precision, recall, F1-score, and confusion matrix, receiver operating characteristic, and precision-recall curves. The findings suggest that conventional unsupervised and dense reconstruction-based models, however, capture normal operating behavior well, but have low sensitivity to anomalous events because no explicit temporal dependency models are present. Conversely, the Transformer Autoencoder is a better performer with a total accuracy of 85 and better metrics based on the anomalies since it uses the self-attention mechanisms to identify long-range temporal relationships between load profiles. Also, the threshold sensitivity analysis helps to evidenced that the proposed approach is practical in terms of its flexibility of balancing between false-alarm rates and detection reliability. The results verify that deep learning models based on the attention can provide a solid and scalable solution to the real-time anomaly detection in contemporary smart grid operations.},
        keywords = {Smart grid, Anomaly detection, Transformer autoencoder, Time-series analysis, Load monitoring},
        month = {January},
        }