With the transformation of power system to Cyber-Physical Systems (CPS), especially the threat of attack between network and physical domain is increasing. In this regard, the researchers use the attack prediction model to simulate network physical attacks. The existing attack prediction models are generally aimed at the network domain and seldom consider the physical domain, so they do not meet the attack and defense requirements of the power system. To solve this problem, a Weighted Colored Petri Net (WCPN) model is proposed to simulate the network and physical nodes in the Power Generation Side of Cyber-Physical Systems (PGSCPS) and the threat propagation between them. Through the WCPN model, the security threats and potential attack paths faced by power CPS are analyzed. The experimental results show that, compared with the traditional data analysis method and attack-defense tree model, WCPN model has advantages in attack resource loss and detection performance, which provides a certain reference for improving the security of power system.