The invention provides a real-time monitoring alarm system for power grid abnormity based on a regional recommendation convolutional network. The real-time monitoring alarm system comprises a fixed camera device, a regional recommendation convolutional neural network, a sample generation system and an alarm system. All the parts are matched with one another, and real-time monitoring work on the designated position of the power grid is automatically completed. The region recommendation convolutional neural network is composed of two networks, the region recommendation network predicts a regionwhich may be abnormal, and the convolutional neural network accurately positions the type, size and position of an abnormal object. And after an abnormal condition is found, an alarm is given immediately through mail, short messages and the like to inform an inspector. Meanwhile, the invention provides a real-time monitoring and alarming method of the real-time monitoring and alarming system for power grid abnormity. According to the invention, automatic real-time monitoring of the designated abnormal condition of the designated position of the power grid is realized, and a lot of manpower issaved.