Abstract: In order to analyze the current risk status of the cyber physical system (CPS) in power system, this paper proposes a risk assessment method for cyber physical power system based on attack prediction, with consideration of the close coupling characteristics of cyber system and power system. Firstly, we use alert message to identify the possible attack scenarios based on the hidden Markov model (HMM), and speculate the attacker's attack intention and analyze the next attack target and attack probability. The results of attack prediction represent the current attack threat status of the system and are used as the input of the risk assessment process. Secondly, we use the traditional single-domain (cyber domain or physical domain) risk assessment method to calculate the single-domain risk, and then assess the cross-domain risk based on the complex network model of cyber physical power system. The final risk value is obtained through integrating the results of both domains. Based on the smart distribution network simulation platform of IEEE 33 BUS, the attack prediction method and risk assessment method are verified, and the results have proved the feasibility and rationality of the attack prediction-based risk assessment method.