부산대학교 도서관

To realize the online status monitoring of power transformers, a type of integrated sensing terminal has been developed, and the corresponding fault identification algorithm is also proposed. Benefiting from the well-designed hardware and software, this smart device is able to sample, process, store and transmit multi-status data such as partial discharge signals and infrared images. Based on this kind of data stream, an intelligent algorithm composed of two data processing branches is able to fuse different types of features and to improve fault detection accuracy. To be specific, Transformer Network and similarity calculation are combined for infrared image processing, and Multi-layer Perceptron Models are utilized for feature vector embedding. The final step is to classify samples with K-means clustering algorithm combined with slide windows. Experiments on field data show that this method is much better than other data fusion strategies, providing a practical solution to multi-status fault detection problem of power transformers.