Rectified Flow-Based Generative Transfer Learning for System Fault Diagnosis
Ruonan Liu, Shuai Wu, Qiming Liu, Dong-Sheng Guo, Weidong Zhang
Abstract
With the rapid development of new energy sources under the “dual carbon” background, the safety and stability of energy storage and power generation systems have attracted widespread attention. This paper investigates the challenges of insufficient training data and the limitations of existing generative transfer learning methods in the field of system fault detection. To address these issues, we propose a graph neural network model based on a generative transfer learning method utilizing the Rectified Flow algorithm for system fault detection. The Rectified Flow algorithm is applied to transfer learning to enhance the efficiency and effectiveness of system fault diagnosis, especially when training data is limited. The model leverages source domain data to train the target domain data, overcoming the data scarcity issue in the target domain. Our experimental results demonstrate that the proposed model outperforms traditional methods in terms of computational efficiency and model conciseness, showing remarkable performance in system fault detection. This study not only provides an effective solution to the data insufficiency problem in system fault detection but also offers valuable insights for further enhancing the performance of system fault diagnosis.