This paper aims to study the vulnerability of the network to sequential cascading failures attacks where the attack strategy integrates network theory and discounted reward with Markov decision process (MDP) in the target selection process. A control strategy is designed to maximize the attack's long-term expected reward while reducing the attack sequence duration. The attack model identifies the most suitable targets by prediction through a Markov process for predicting the propagation and consequences of the failure. The state transition probabilities through a hidden failure model embedded in an independent edge-dependent network evolution model is estimated. Value iteration algorithms are used to identify targets at every attack stage. Target selection is updated depending on network changes. The results provide an optimal attack strategy based on network congestion with maximum damage, considering congestion as a cascade propagation mechanism. Reward functions based on increasing congestion and immediate power loss are compared. Strategies designed with network congestion as the attack reward function produce more vulnerability of the network to sequential attacks.