A Blocking-State Prediction Method for MMC Oriented to Large-Step Electromagnetic Transient Real-Time Simulation
Real-time electromagnetic transient simulation of modular multilevel converters (MMCs) on multi-core CPU platforms at large time steps is constrained by the difficulty of predicting diode conduction states during blocking operation. Conventional direct prediction methods sample node voltages at the previous time step and extrapolate the next switching state; at 50 μs, however, multiple natural commutation events can occur within a single step, producing over-shoot, numerical oscillation, and elevated prediction failure rates. This work proposes a blocking-state prediction method grounded in the internal topological and electrical constraints of submodules. Diodes exhibiting identical behavior under blocking are aggregated into a unified equivalent circuit, and a constraint-based state prediction mechanism with enhanced robustness is constructed, eliminating reliance on high-speed FPGA timestamping or variable-step rollback. Offline and real-time simulations on the ADPSS platform demonstrate that at a 50 μs step the proposed method maintains computational error within 3‰–7‰, substantially reduces prediction failure probability relative to direct prediction, and completes a single simulation step in approximately 22.23 μs on a multi-core CPU architecture, satisfying the 50 μs real-time constraint. The method provides a viable pathway for efficient CPU-based real-time simulation of large-scale MMC systems.