Energy saving has become an important issue in wireless communications from both environment and economic considerations. In this paper, an energy minimization algorithm based on distributed dynamic clustering (DDCEM) with lower complexity, higher performance, and better adaptability for heterogeneous network (HetNet) is proposed. A HetNet could be divided into several clusters, which is defined as one group of a network node and users served by the node. A energy efficient user association, which dynamically changes according to real-time energy efficiency (EE) evaluation with traffic load and location distribution of each cluster, can be employed to save consumed energy. Then, the optimal sleeping relay is found as follows. First, the sleeping probability cost of each relay station (RS) is computed and ranked based on the user traffic and the position distribution of each cluster, and the relay with minimum sleeping probability is selected to be switched off. Hence, the sleep node is selected taking into account the traffic load and location of the Evolved Node B (eNB) and all the RSs. The complexity of the algorithm is greatly reduced because the user association operation and network load evaluation are fulfilled cluster by cluster. Simulation results show that the proposed DDCEM strategy offers EE gain with low system complexity.