A Class-Imbalanced Heterogeneous Federated Learning Model for Detecting Icing on Wind Turbine Blades

Cheng, Xu; Shi, Fan; Liu, Yongping; Zhou, Jiehan; Liu, Xiufeng; Huang, Linyan

doi:10.1109/tii.2022.3167467

Cited by 21 publications

(3 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…• Blade Icing Federated Learning (BiFL): BiFL proposed a heterogeneous FL model for blade icing detection. In this model, the exchange between clients and server is the encoded feature map rather than gradient [12]. • Federated Learning Batch Normalization (FedBN):…”

Section: Comparison With State-of-the-art Fl Methodsmentioning

confidence: 99%

“…FL has been applied to a range of tasks [23]. Cheng et al were the first to apply FL with a heterogeneous structure between the client and server for detecting wind turbine in icing conditions [12]. They later proposed an improved version that integrates Blockchain with FL for blade icing detection [24], and a version that emphasizes class imbalance learning [25].…”

Section: B Federated Learningmentioning

confidence: 99%

“…Traditional centralized data-driven approaches for model training rely on the consolidation of data in a single location, such as a cloud. However, this requires a significant amount of storage and computing resources and may also raise concerns about data privacy [12]. In addition, the network bandwidth between the server and each wind turbine must meet certain requirements.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

FedBIP: A Federated Learning-Based Model for Wind Turbine Blade Icing Prediction

Zhang

Tian

Cheng

et al. 2023

IEEE Trans. Instrum. Meas.

Self Cite

View full text Add to dashboard Cite

Prediction of icing on wind turbine blades is crucial, particularly in high-latitude areas where ice accumulation is a frequent occurrence. Traditional centralized data-driven approaches for predicting blade icing have demonstrated promising performance, but they require a large amount of storage and computational resources and may also raise concerns about data privacy. Federated Learning (FL) presents a potential solution to address these issues. These challenges include redundant features in the collected data, a highly imbalanced data distribution between normal and icing samples, and slow model convergence during FL training. To tackle these challenges, we proposed a novel FL model called FedBIP. FedBIP employs a feature selection approach enhanced with human knowledge to select relevant features, a segmentation-based oversampling method to alleviate class imbalance, and a new aggregation method that takes into account data size, timestamps, and offsets of each participating client. In addition, knowledge distillation is employed in the local model training to accelerate model convergence and speed up the overall training process. The results of comprehensive experiments demonstrate that FedBIP outperforms state-of-the-art FL methods, aggregation methods, and feature extractors. Ablation and sensitivity analysis were also conducted to validate the importance of each component and key parameters in FedBIP.

show abstract

Section: Comparison With State-of-the-art Fl Methodsmentioning

confidence: 99%