Broad Learning System (BLS) has been widely applied in classification tasks due to its efficient training and low computational complexity. However, traditional BLS limits its performance by ignoring redundant features and sample importance differences when dealing with complex and high-dimensional data. To resolve this challenge, a trapezoidal fuzzy broad learning system based on distance correlation (DT-BLS) is proposed. The proposed model utilizes distance correlation to select key features, effectively reducing the interference of redundant information. Simultaneously, it employs a trapezoidal fuzzy membership function to dynamically allocate weights based on the importance of samples to classes, thereby enhancing the model's ability to handle imbalanced data and complex classification tasks. Experimental results confirm that the proposed DT-BLS model significantly outperforms traditional BLS and other mainstream methods in terms of classification accuracy and generalization in multiple publicly available datasets, validating its effectiveness. The research in this paper provides new ideas for the application of BLS in complex data scenarios.