This article aims to develop a lightweight, decentralized federated learning (FL)-based strategy for electricity theft detection (ETD). Different from most of the existing ETD solutions, which typically deploy centralized deep learning models, our proposed method utilizes well-pruned lightweight networks and operates in a completely decentralized manner while maintaining the performance of the ETD model. Specifically, to protect data privacy, a novel sequential decentralized FL (SDFL) framework was designed, eliminating the centralized parameter aggregation node in traditional FL. Each client communicates model parameters only with its neighbors and trains its model locally. In addition, to facilitate deployment on edge devices, model pruning techniques are integrated with the sequential transmission characteristics of the SDFL framework. A progressive channel pruning technique is proposed, gradually reducing the number of model channels during training to promote model compression and simplify field deployment. Experiments demonstrate that our strategy compressed the model floating point operations from 18.32 to 3.60M and reduced the number of parameters from 8.61 to 3.47M, while protecting user privacy, and maintaining good performance. Deployment results on the edge devices, i.e., Raspberry Pi, indicate that our proposed strategy reduces the model inference time from 329.35 to 141.50 s, enhancing the detection efficiency by 57.04%.

Recently, data-driven methods have demonstrated excellent performance in diagnosing faults in power transformers. However, concerns about data privacy and computational resources arise due to the requirement of collecting and training large numbers of labeled samples in a centralized manner. Although centralized federated learning (CFL) provides optimal solutions, the dependence on central servers engenders issues associated with reliability and increased communication expenditures. Meanwhile, existing decentralized federated learning (DFL) frameworks are not specifically designed for industrial applications with unbalanced data and high communication costs. To tackle these issues, this paper proposes a novel parallel sequential DFL (PS-DFL) framework that exchanges information directly between client pairs, thus overcoming the limitation caused by the central server in CFL. In particular, by combining parallel training with sequential communication techniques, the impact of unbalanced data on model performance is mitigated, as well as communication costs are reduced. The experiment conducted on IEC TC 10 and other public datasets demonstrate that, despite handling tasks involving unbalanced data, the diagnostic accuracy is maintained at 97.09%. Furthermore, the communication overhead is significantly lower than that of existing methods, thereby proving its feasibility for industrial deployment within the power grid.

Recently, partial discharge pattern recognition (PDPR) for transmission cables has garnered increasing attention due to the severe power outages, equipment damage, and even major safety incidents resulting from the failure of partial discharge (PD) detection. However, existing PD data samples usually suffer from highly similar features and unbalanced distribution. Determining how to precisely realize the PDPR has become a challenge. In this study, an effective PDPR approach is proposed based on a newly designed deconstructed PD (DecPD) model and a customized loss function for PDPR. Notably, the refined deep learning network captures the discriminative features in both temporal and spatial dimensions through a dual-channel learning architecture. Additionally, an adaptive focal loss function is designed, which introduces a peak factor to establish focusing parameters for PDPR, thereby addressing the class imbalance issues. A comprehensive experimental evaluation using real datasets generated on a physical platform is conducted to verify our proposed method. Compared to other existing methods, our DecPD approach demonstrates superior performance, achieving an overall accuracy of 96.65% in the presence of environment noise.

Malicious clients participating in data collection and interaction may launch attacks such as model and data poisoning to degrade the performance of the global model and conceal their electricity theft behaviors. Although existing studies have introduced blockchain technology to achieve decentralization, they still suffer from limited pre-aggregation validation dimensions. To address these issues, this paper proposes a blockchain-based federated learning approach with dual-verification (BFL-DV) for electricity theft detection. In the pre-aggregation stage, a multi-metric reputation-based consensus committee verification strategy is designed, which effectively mitigates the impact of malicious participants. In the post-aggregation stage, a dynamic threshold-based blockchain verification strategy is developed to counter security risks during the transmission process, which can refuse malicious global updates adaptively. Experimental results demonstrate that BFL-DV can accurately reduce the impact of all malicious clients under the data poisoning attack. Notably, across various proportions of malicious clients, the proposed framework achieves an average AUC improvement of 32.68% compared with SOTA methods, demonstrating its consistent performance advantage.