A deep association discovery framework for the multidimensional data: Application to power grid analysis

The discovery of associative relationships is essential for analyzing economic, industrial, and societal issues. However, identifying such relationships in high-dimensional data poses significant challenges, including high computational complexity and limited interpretability, particularly in complex systems such as power grids. To address these issues, we propose CORD, a deep association discovery framework for multidimensional data, designed to uncover relationships between cross-sectional transmission capacity and power grid operational modes. Our approach begins with an initial feature screening based on expert knowledge, followed by a secondary screening using the DBSCAN algorithm enhanced with rank correlation coefficients. We then construct a decision tree model that incorporates an improved K-Means++ clustering algorithm and information gain based on entropy to extract coarse-grained relationships. Furthermore, we develop a deep convolutional neural network integrated with rank correlation coefficients to capture fine-grained associations. Experiments on real-world power grid data demonstrate that the proposed CORD framework effectively uncovers meaningful relationships between transmission capacity and operational modes, providing a robust and interpretable solution for complex system analysis.