Decentralized energy management using consensusbased algorithm is a vibrant research field since it can promote the local accommodation of the renewable energy generation without raising privacy and scalability issues. Most of the existing methods assume that the communication links are reliable, which might not be true in real-world implementations. This paper focuses on tackling the problem of random packet drops. We first formulate the models of the energy management problem in the microgrid and the information packet drop in the communication network. Based on the models, we conclude that losing the information about incremental electricity cost estimation is tolerable, while losing the information about the power mismatch estimation is not. We propose a novel consensus algorithm that tracks and exchanges the accumulated value of the power mismatch estimation so that the information loss can be recovered. An equivalent form of the proposed method is established by augmenting the communication links with virtual buffering nodes. Based on the augmented communication topology, we theoretically prove the convergence of the proposed algorithm and the optimality of the solution. Several case studies are provided to validate the effectiveness of the proposed algorithm.