The physical essence of the electro-dewetting phenomenon lies in the interfacial migration of ionic surfactants driven by an external electric field, a complex process that poses a significant challenge to theoretical modeling. Traditional molecular dynamics (MD) struggles to reach the macroscopic spatiotemporal scales of droplet evolution, whereas macroscopic electrohydrodynamic (EHD) models typically rely on empirical parameters, lacking atomistic physical accuracy. To overcome this limitation, a cross-scale co-simulation method is proposed. First, to address the lack of dedicated empirical force fields for complex large-molecule surfactants, a highly accurate machine learning potential is constructed by combining first-principles calculations with a deep learning workflow. Based on this potential, high-precision MD simulations are performed on the bulk solution under a uniform electric field. Simultaneously, the mean square displacement of the surfactant molecules is extracted, and a drift-diffusion model is introduced to obtain a more accurate microscopic diffusion coefficient of the surfactant. Subsequently, this coefficient is incorporated into the macroscopic EHD model to substitute traditional empirical assumptions, thereby solving the coupled governing equations of liquid film thickness and surfactant concentration. Finally, the proposed cross-scale model is comprehensively validated through droplet manipulation experiments. Numerical results demonstrate that, compared to conventional EHD models, the cross-scale method exhibits lower errors and higher morphological agreement when capturing the dynamic behaviors near the three-phase contact line. Quantitative analysis indicates that the prediction accuracies of the macroscopic contact angles during the electro-dewetting and re-wetting stages are improved by approximately 4.59% and 3.59%, respectively, and the accuracy of the contact angle difference between the two stages is enhanced by 26.67%.