With the ongoing development of modern power systems, short-circuit current limit violations have become increasingly prominent, and single current-limiting measures are no longer sufficient to ensure safe system operation. To address this issue, this paper proposes a mechanism-data-driven power system optimal operation method based on short-circuit current constraint learning. First, to cope with complex operation modes in power systems, a set of combined current-limiting measures, including line switching, busbar sectionalizing, and unit start/stop, is proposed. Second, to overcome the limited ability of multi layer perceptron (MLP) models in learning topology variations, a topology feature enhancement method based on One-hot encoding is proposed to improve the model’s adaptability to current-limiting measures. Third, the concept of short-circuit current safety margin is introduced to quantify the degree of violation of short-circuit current constraints under different current limiting measures. The big-M method is then employed to process the forward propagation formula of the MLP, thereby establishing data-driven short-circuit current constraints. Finally, with the objective of minimizing the total cost of unit operation and network topology adjustments, and considering grid operating constraints, N-1 security constraints, and short-circuit current constraints, a mechanical-data-driven power system optimal operation model based on short-circuit current constraint learning is established. Case studies verify the effectiveness of the proposed model.