An efficient federated transfer learning framework for collaborative monitoring of wind turbines in IoE-enabled wind farms

Wind turbine (WT) condition monitoring has gained increasing interests in the era of the Internet of Energy (IoE), and existing monitoring approaches mainly focus on training a reliable model in a centralized manner. However, there are three main challenges: data islands due to data privacy and strict confidentiality of wind farm owners, domain shift due to distribution difference of different WTs, and computing and communication burden due to a large number of model parameters. To address these challenges, we propose an efficient federated transfer learning framework (EFTLWT) for collaborative monitoring of WTs in IoE-enabled wind farms, which integrates the adversarial domain adaptation into the federated framework to address the domain shift issue across multiple WTs from different wind farms. It is the first attempt to apply federated transfer learning into the field of WTs monitoring. Specifically, we design a lightweight multiscale neural network model to reduce the computation and communication cost between the client and the server. Furthermore, we propose a partial aggregation strategy to aggregate partial model parameters to reduce the model weight during the uploading and downloading, thus further reducing the burden of communication bandwidth and speeding up the response of the monitoring system. We carry out extensive experiments on real operating datasets of WTs and the results show that our EFTLWT can effectively reduce the domain shift and communication cost, decrease the response time, and greatly improve the performance of the local and federated model while maintaining the favorably comparable performance as the centralized model.