Modern power system is a typical multi-level complex giant system consisting of physical infrastructures, human operators, and social resources, etc. The conventional analytical methods and simulation systems can't provide sufficient guidance for its operation and management, because they are mainly based on physical models, natural phenomenon, or other existing control methods which are based on reductionism. ACP approach, mainly consisting of artificial systems (A), computational experiments (C) and parallel execution (P), which is based on holism and complex system theory, has its specific advantages in the research on power systems. In this article, ACP approach is applied to build up artificial power system (APS) by using multi-agent complex networks. With the help of APS, actual power system's control, scheduling, optimization and management can be improved further via providing theoretical guidance and technical support for the rolling optimizations in normal situations and emergency management in abnormal situations. As a case study, an APS constructed with actual data from North China power grid is constructed and its vulnerability is simulated and analyzed under random, dynamic and static attacks.