<p>The purpose of ensemble clustering is to produce more robust, higher quality clustering results by combining several different underlying clusterings. Removing low-quality base clusterings and assigning weights to clusters of different quality can obtain a more accurate ensemble solution. In order to solve this optimization problem, this paper proposes an ensemble strategy of constructing inter-cluster relations after screening the base clusterings. The algorithm consists of three main steps. The first step screens base clustering members by using group consistency measure. The second step evaluates the effectiveness of clusters by using granularity distance knowledge. The third step performs consistency ensemble clustering by two consensus functions. The proposed ensemble clustering algorithm changes the cluster evaluation metrics and removes the need for parameter adjustments during the construction of the metrics. It further compensates for the shortcomings of the single improvement algorithm. In addition, filtering base clustering by the concept of group consistency reduces redundant computation and provides a feasible solution for efficient ensemble clustering in large-scale data scenarios. Experiments on 10 real-world datasets and 2 synthetic datasets verify the accuracy and robustness of the algorithm.</p>