18 January 2026 in Results in Engineering
30 March 2026 in IEEE Transactions on Consumer Electronics
Rapid advancements in urbanization and industrialization have increased consumer-end applications in smart grids, where advanced data-driven methods have significantly improved consumer-level energy management using smart meters. However, these systems are highly vulnerable to cyberattacks, notably false data injection attacks (FDIA), which can compromise grid reliability and stability. Existing detection techniques often rely on centralized data collection and processing, which raises substantial privacy concerns and leads to overfitting due to limited data diversity. To address these issues, this paper presents a novel framework that integrates federated learning with an autoencoder-based 1D deep convolutional neural network (AE-1DdeepCNN) for FDIA in consumer applications. The proposed framework enables collaborative model training based on the IEEE-14 bus system dataset for anomaly detection while ensuring data privacy. Extensive experimental results with row-accuracy of 93.35%~99.40% in different smart meters demonstrate the effectiveness and robustness of the proposed method for FDIA detection in smart grid systems and consumer smart metering applications.
29 April 2026 in IEEE Transactions on Consumer Electronics
The rapid growth of intelligent sensor-based consumer applications has significantly increased the need for an accurate, reliable, and trustworthy load forecasting framework. These frameworks play an important role in addressing key challenges in real-time resource allocation, operational planning, and the control of complex industrial networks, particularly through the integration of sensing, computing, and artificial intelligence. To mitigate these challenges, this paper presents an AI-assisted framework based on the hybrid combination of i) variational mode decomposition to decompose complex and non-stationary load patterns into different intrinsic modes to improve the interpretability, ii) autoencoder to enable the extraction of salient latent feature representation and noise reduction from the decomposed signals, and iii) the bidirectional long-short term memory network to effectively capture the complex spatial and temporal dependencies in both forward and backward directions for accurate and efficient load forecasting. The robustness of the proposed frameworks is evaluated using publicly available power consumption datasets. The extensive experimental results demonstrate that the proposed model outperforms other baseline models, with overall improvements of 18.54%~81.95% in mean absolute error (MAE), 21.72%~82.51% in root mean square error (RMSE), and 17.03%~82.11% in mean absolute percentage error (MAPE). The proposed model is also compared with state-of-the-art methods and it shows the overall improvement of 8.67%~72.85% in RMSE, 3.83%~71.30% in MAE, and 2.59%~69.79% in MAPE. The results validate that the proposed model provides an effective, efficient, reliable, and robust predictive solution for intelligent management and optimization in consumer energy applications.
12 April 2026 in Energy Conversion and Management: X