The invention discloses an online transformer fault diagnosis system based on an artificial intelligence extreme learning machine. The system comprises data sample acquisition nodes, a field analysiscontrol unit, a communication unit and a local control center, wherein each data sample acquisition node comprises an oil sample acquisition module, an oil sample separation module, a gas detection module and a data acquisition card; and the oil sample acquisition module, the oil sample separation module, the gas detection module and the data acquisition card are sequentially connected. The systemrealizes data acquisition and data analysis of the nodes by using the Internet of things technology, and realizes the function of online monitoring through the display of the remote control center; and multiple groups of data sample acquisition nodes are arranged for diversified sample acquisition and learning training and verification are carried out through the extreme learning machine, therebyensuring the quantification of samples and ensuring the accuracy of monitoring.