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

Wind farms are usually located at high latitudes, leading to an increased risk of blade icing. Data-driven approaches offer promising solutions for blade icing detection but rely on a considerable amount of data. Data exchange between several wind farms would improve the performance of detection models, due to spatio-temporal dependencies capable of reflecting different weather conditions. However, due to commercial competition, data owners may be reluctant to share their data. To address the privacy issue, this research proposes a heterogeneous federated learning (FL) model. Unlike traditional FL, the structures of the server and client models are different and a privacy-preserving approach is incorporated to solve the class-imbalance problem in the sensor data. Using real data from 20 turbines at two wind farms, this model was evaluated and compared to two state-of-the-art FL models and five well-known class-imbalance methods. The experiment results verify the effectiveness and superiority of the proposed model.