Deadline-Deterministic Sparse Matrix Updates with Versioned Factor Banks for Real-Time Distribution Network EMT Simulation

Real-time electromagnetic-transient (EMT) simulation must complete every integration step on time even when a fixed-pattern topology change—an N-type numerical state change within an initialization-time superset sparsity pattern—alters registered matrix entries. We present deadline-deterministic sparse execution (DDSE). Here, “deadline-deterministic” means that the foreground transaction is preallocated and capacity-bounded for a declared hardware, software, matrix, workload, and capacity envelope; it is not a platform-independent worst-case execution-time (WCET) guarantee. DDSE makes an accepted change visible to the current solve through a compact Sherman–Morrison–Woodbury (SMW) overlay, while a background worker absorbs the target state into an inactive factor bank and publishes it only after a sequence check at a quiescent step boundary. This separation preserves same-step numerical visibility without placing matrix reconstruction in the event-control transaction. In controlled comparisons, DDSE met the configured deadline when in-step absorption and complete fixed-pattern KLU refactorization did not. Sustained converter-rich and feeder-oriented hardware-in-the-loop tests recorded no deadline misses or dropped timing samples, and all same-step waveform-equivalence tests passed. The results qualify DDSE for the evaluated deployment envelope.