Time series deep clustering methods can effectively learn discriminative features and clustering structures automatically from unlabeled temporal data. To further enhance the model"s robustness and generalization capability, clustering ensemble techniques have been introduced. However, existing deep clustering ensemble methods often require training multiple models independently to generate diverse base clusterings, resulting in high computational costs, significant resource consumption, and difficulty in collaboratively optimizing quality and diversity within a single model. To address these issues, this paper proposes a time series deep clustering ensemble algorithm based on contrastive learning and dynamic diversity. First, a temporal encoder is designed to extract multi-level temporal features, and the discriminative power of the features is enhanced through temporal-level and instance-level contrastive learning. Second, a multi-projection-head clustering generation mechanism is constructed, which enforces dynamic diversity constraints based on mutual information, thereby collaboratively optimizing the quality and diversity of base clusterings. Finally, both clustering objectives and diversity constraints are integrated into an end-to-end training framework to achieve robust and accurate partitioning of temporal data. Experimental results on ten public UCR datasets demonstrate that the proposed algorithm significantly outperforms existing baseline methods in clustering performance, while effectively reducing computational costs while maintaining high performance.