With the gradual expansion of the power grid, the digitalization of the distribution network has brought numerous benefits to the operation of the power grid. However, it has also increased the risks posed by serious cyber security threats. To detect potential attacks and vulnerabilities, and to meet the additional research needs for security and privacy mechanisms, an automated process is required to systematically process a large amount of cross-domain information and correlate various network intelligence to accurately assess the situation. For this purpose, we propose a power grid multi-stage attack perception risk quantification and defense technology. This technology employs an artificial neural network to detect attacks in the power grid and performs multi-stage context attack risk quantification on the system modeling attack defense tree. Furthermore, the proposed technology utilizes the killing chain to find the optimal path for defense, ultimately achieving a fully automated threat perception and defense decision system for the power grid.