Aiming at the problem of topology tracking in grid digital twin applications, the paper proposes a new modeling approach based on a graphical scheme and for topology modeling and change detection. Two modeling algorithms are proposed along with conditions that allow for accurate modeling without ignoring spurious edges. The first algorithm relies on local neighborhood counts in the composite voltage graph model, while the second algorithm uses the algebraic sum of the entries in the voltage inverse covariance matrix. The proposed method does not require the acquisition of a priori knowledge of line impedance and node injection statistics. An algorithm for topology change detection by comparing the deviation of the inverse covariance matrix of voltage fluctuations before and after Topper changes is also proposed. The effectiveness of the method is verified by an extensive simulation analysis based on a test system.