The research on smart grid security has led to several important results. Nevertheless, previous works neglect to consider the time-varying properties of smart grid, which in many attack strategies play a pivotal role. Mainly because of the time-varying collaboration among communication agents and the time-varying confrontation between defenders and attackers, smart grid communication network should not be considered as a static entity. Understanding the smart grid resilience through time-varying analysis is thus crucial to both the grid protection and to the design of new countermeasures against cyber threats. In this work we attempt to bring time dimension into smart grid security analysis. In each time step, attackers determine the attack targets based on a 2-state Markov process. Defenders (smart meters) are empowered with two defense actions: either to monitor their neighbors which requires more resources but yields timely discovery of attacks, or not to monitor their neighbors which saves resources but leads to slow response to attacks. Defenders choose between the two by estimating the attack probability based on previous time slots. The objective of defenders is to minimize their defense cost. Simulation results are provided to demonstrate the superior performance of the proposed scheme over the conventional methods.