In order to make the deep clustering algorithm achieve robust clustering results for different types of complex data, the strategy of selective ensemble was integrated with deep clustering to propose a deep clustering algorithm with fusion selective clustering ensemble. It effectively improved the robustness and clustering performance of deep clustering algorithm. The algorithm utilizes an autoencoder-based deep clustering approach with different initialization parameters to generate multiple distinct base clustering results. And it constructed ensemble quality and diversity evaluation metrics of base clusterings. A certain number of base clusterings with higher quality and rich diversity were selected as clustering ensemble candidates. The reliability of the cluster was considered in the clustering ensemble strategy to construct the weighted graph partition consensus function. Experiment results demonstrate that the robustness of deep clustering with fusion selective clustering ensemble has been enhanced. Comparing with many existing clustering ensemble methods, it could obtain better clustering results for multiple different types of data.