- Conference Article
- 10.1109/cespe68033.2025.00081
Research on Lightweight Electricity Abnormal Detection Model for Low Carbon Parks and Its Collaborative Optimization for Energy Storage Scheduling
- Nov 21, 2025
- Xiuliang Liu
In response to the increasingly prominent abnormal electricity risk in low-carbon parks due to complex loads and high proportion of new energy access, this paper proposes a lightweight time-series anomaly detection model and its collaborative optimization framework with energy storage systems. Traditional methods have response lag and fail to effectively link energy storage systems, resulting in operational risks and economic losses. This study first constructs a lightweight neural network that integrates multisource electricity consumption data and equipment status data to achieve accurate identification and minute level warning of power anomalies (such as sudden increases and overloads). Furthermore, by incorporating warning information as forward-looking constraints and embedding it into the energy storage optimization scheduling model, a closed-loop collaborative mechanism of “detection warning scheduling” is formed. Experiments on real park datasets show that our model significantly outperforms traditional baseline methods in both anomaly detection accuracy and early warning time, with an accuracy rate of $95.2 \%$ and an average early warning time of 35 minutes. The case analysis further confirms that the collaborative optimization framework can effectively utilize the warning window, proactively adjust the energy storage charging and discharging strategy, smooth out load fluctuations, and reduce the daily electricity consumption cost of the park by about $7.3 \%$. This study provides an effective solution to enhance the autonomy, safety, and economy of low-carbon park energy systems.
Read more