Consensus algorithms are key technologies in blockchains that ensure the stability and effectiveness of decentralized networks. The traditional DPoS consensus algorithm features a straightforward and rapid voting election approach; however, circumstances such as low enthusiasm in node voting and the monopolization of block-packing power by a minority of nodes might arise, thereby influencing the degree of decentralization of the network. To solve these problems, this paper presents a novel voting counting method based on complex networks (GR-DPoS). First, each vote cast by a node carries a weight indicating the strength of the node’s support, and the voting results among nodes can be abstracted as a weighted directed network. The combined strength value of nodes is computed on the basis of cross-strength and adjacency strength. A reputation value is subsequently introduced to quantify the node’s consensus behavior, which changes with the node’s voting and block-out behavior. Eventually, the agent node is determined by integrating the reputation value and the combined strength value. With respect to the constructed network, after deleting the selected nodes, the maximum connected subgraph scale ratio and the number of subgraphs of the remaining network are calculated to verify the effectiveness of the node selection method proposed in this paper. Simulation experiments are conducted for the proposed method and other methods, and the results indicate that this method can increase the likelihood of normal nodes becoming agent nodes, reduce the frequency of malicious nodes becoming agent nodes, increase the probability of new nodes becoming agent nodes, more effectively mitigate the risk of power concentration, and stimulate nodes to participate actively in consensus.