Co-Simulation Interface Prediction for Accurate Parallel Simulation

The patent applies preliminary action by using data prediction models (polynomial functions, Hermite interpolation) to forecast interface variables before actual data exchange occurs. This allows receiving subsystems to proceed with simulation steps using predicted values, reducing the frequency of actual communications while maintaining accuracy. The prediction models are prepared in advance and continuously updated based on historical data patterns.

The patent introduces prediction models as intermediary elements between sending and receiving subsystems. These models act as mediators that translate and forecast interface variables, allowing subsystems to operate with predicted values instead of requiring continuous direct communication. This intermediary layer reduces communication overhead while preserving simulation accuracy.

The patent implements continuity of useful action by using continuous prediction models (polynomial functions and Hermite interpolation) to generate interface variables between discrete communication events. This creates a continuous approximation of the interface behavior, eliminating the discontinuities that cause numerical instabilities while maintaining the efficiency of discrete parallel simulation architecture.

The patent applies parameter changes by transforming discrete sampled data into continuous prediction models with adjustable parameters (polynomial coefficients, interpolation points). This allows the system to maintain discrete communication timing while generating continuous variable representations that prevent numerical artefacts and solver restarts.

The patent uses preliminary action by pre-computing prediction models based on historical data patterns before larger time steps are taken. This allows the system to skip intermediate communication events while having accurate predictive models ready, maintaining global solution accuracy even with higher time steps between actual data exchanges.

The patent implements feedback mechanisms where prediction models are continuously updated and refined based on actual measured data from previous simulation steps. This feedback loop ensures that even with larger time steps, the prediction accuracy is maintained by adapting models to actual system behavior patterns observed during simulation.