Abstract: Daily load curve clustering serves as the foundation for optimal scheduling of power systems, providing critical data support for formulating time-of-use electricity tariffs and enabling coordinated source-grid-load-storage regulation. To address the limitations of existing load clustering methods, namely, their inability to capture multi-scale spatiotemporal correlations and their lack of the capacity to effectively deal with high-dimensional nonlinear data, a collaborative clustering method that incorporates singular spectrum analysis (SSA), Transformer, and bidirectional long short-term memory (BiLSTM) is proposed. First, SSA is employed to filter out noise and extract low-frequency components. Secondly, a Transformer-BiLSTM fusion model is constructed to capture both global dependencies and local temporal features in the load data. Finally, clusters are generated using the K-means algorithm. Case study results show that the clustering metrics obtained using the proposed method, including Davies Bouldin index (DBI) , silhouette coefficient (SC), and sum of squared error (SSE), are 0.76, 0.5 and 0.34, respectively, thereby exhibiting the superiority of the proposed method. In addition, the methodology demonstrates robustness under 30% noise contamination, significantly enhancing clustering quality for high-dimensional, nonlinear load curves and providing an effective tool for refined load forecasting and regulation.