Abstract: This article proposes a collaborative diagnosis system based on intelligent sensing technology and knowledge driven methods to address the problem of latent thermal and electrical faults in the long-term operation of main transformers in thermal power plants. By designing a layered distributed architecture, integrating photoacoustic spectroscopy sensors and ultra-high frequency sensors, real-time collection of gas components and transient electrical signals in oil is achieved. At the same time, a fault diagnosis knowledge graph with domain ontology as the core is constructed, and a multi-scale one-dimensional convolutional neural network (1D-CNN) is combined to extract features from multi-source temporal signals, establishing a collaborative diagnosis model of knowledge graph and deep learning. Experiments have shown that this method outperforms traditional methods in terms of accuracy, recall, and F1 score, and can effectively improve the ability to identify early faults and provide early warning.