Abstract: Modular multilevel converter (MMC) is widely applied in high power motor drive, energy storage and flexible DC power transmission & distribution system, and parameter identification for MMC is essentially important for its reliable operation. The concept of digital twins (DT) enables mapping of real-time state information from physical entity onto a digital twin model, thus facilitating device state visualization. In this paper, we propose a DT based method for model construction and parameter identification of the MMC half-bridge sub-module (HBSM). The DT model of the HBSM is built, consisting of a simplified MMC sub-module (SM) equivalent model and an arm current simulator based full-bridge converter. A modified particle swarm optimization (MPSO) algorithm is proposed to achieve the online identification of internal parameters in the HBSM, thereby enabling the real-virtual interaction feedback for real-time sensing of the operation state of the HBSM. On this basis, a non-intrusive DT simulation platform is constructed for MMC HBSM. The result demonstrate that the average errors between the actual and model-calculated capacitor voltage values are within 0.1%, while the parameter identification results combined error is less than 1%.