Source author record

Jinyuan Stella Sun

Jinyuan Stella Sun appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Cryptography and Security eess.SP eess.SY Machine Learning Systems and Control

Catalog footprint

What is connected

2works

5topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

Exploring Physical-Based Constraints in Short-Term Load Forecasting: A Defense Mechanism Against Cyberattack

Short-term load forecasting is an essential task that supports utilities to schedule generating sufficient power for balancing supply and demand, and can become an attractive target for cyber attacks. It has been shown that the power system state estimation is vulnerable to false data injection attacks. Similarly, false data injection on input variables can result in large forecast errors. The load forecasting system should have a protective mechanism to mitigate such attacks. One approach is to model physical system constraints that would identify anomalies. This study investigates possible constraints associated with a load forecasting application. Looking at regional forecasted loads, we analyze the relation between each zone through similarity measures used in time series in order to identify constraints. Comprehensive results for historical ERCOT load data indicate variation in the measures recognizing the existence of malicious action. Still, these static measures can not be considered an efficient index across different scenarios.

preprint2020arXiv

SearchFromFree: Adversarial Measurements for Machine Learning-based Energy Theft Detection

Energy theft causes large economic losses to utility companies around the world. In recent years, energy theft detection approaches based on machine learning (ML) techniques, especially neural networks, become popular in the research literature and achieve state-of-the-art detection performance. However, in this work, we demonstrate that the well-perform ML models for energy theft detection are highly vulnerable to adversarial attacks. In particular, we design an adversarial measurement generation algorithm that enables the attacker to report extremely low power consumption measurements to the utilities while bypassing the ML energy theft detection. We evaluate our approach with three kinds of neural networks based on a real-world smart meter dataset. The evaluation result demonstrates that our approach can significantly decrease the ML models' detection accuracy, even for black-box attackers.