Due to limited energy storage units in microgrids, how to regulate peak demand is one of the main challenges. Thus, researchers propose different techniques to flatten peak demand in individual residential buildings. However, if each home in the grid flattens peak demand only with its own power consumption information, it is possible that peak demand of the microgrid would not be flattened but only shifted to another period. Therefore, it is critical for homes to cooperate with each other to flatten peak demand. In this paper, we utilize the power-voltage relationship in individual homes to enable that each home can infer the information of power consumption in the community by locally monitoring the voltage value on the common power line. The inferred information is then used for homes to flatten peak demand of the microgrids in a distributed manner. Furthermore, we leverage existing thermal appliances (e.g., water heaters) as thermal "batteries" in individual homes instead of purchasing batteries to flatten peak demand. We evaluate our system's performance by conducting experiments and extensive empirical data driven simulations. Evaluation results indicate that our design enables homes to effectively flatten peak demand by more than 29% without affecting homeowners' behaviors.