The invention relates to the technology of metering automation terminal abnormality diagnosis, and specifically relates to a fault diagnosis device and method for a portable metering automation terminal. The device specifically comprises a detection submodule, a power management module and a man-machine interaction module. Through the combination of the neural network technology and the support vector machine technology, the device carries out the classification of various types of data of the metering automation terminal through a trained Ababoost classifier and a trained neural network, extracts the features, finds an optimal classification plane for the extracted data through a support vector machine, carries out the precision positioning of a fault cause and a fault point of the terminal, and improves the diagnosis efficiency of the metering automation terminal equipment.