Abstract: With the deepening of informationization, the traditional power system has been transformed into a typical cyber-physical system（CPS）. Considering the open cyber system environment, the security operation of the cyber-physical power system（CPPS） faces threats from various potential cyberattacks. In recent years, machine learning approaches have been developing rapidly and have been widely used in CPPS cyber security. On the one hand, the explosive growth of data in the CPPS and the improvement of hardware computing power create the right conditions for applying machine learning approaches. On the other hand, compared with the traditional model-based approaches, the data-based machine learning approaches has advantages in two aspects: modeling and real-time requirements. This paper summarizes the application of machine learning in CPPS cyber security from the perspectives of attack and defense, respectively. The perspective of attack mainly includes three aspects: topology inference, attacking resource optimization, and attack construction. The perspectives of defense mainly include three aspects:security protection, attack detection, and attack mitigation. Finally, the challenges and future research directions in the field of CPPS cyber security are proposed.