Chemometric Model Determination via Spectral Data Combination

The patent applies preliminary action by performing reactor runs only for a limited number of distinct conditions (e.g., 3-5 runs) rather than exhaustively testing all possible conditions. The training data is collected under these preliminary conditions and used to build a chemometric model that can predict outcomes for additional conditions without requiring actual experimental runs for each condition, thus reducing time while maintaining adequate model accuracy.

The patent uses copying by creating a chemometric model that replicates or predicts the results of multiple reactor runs based on data from a smaller number of actual runs. The model copies the relationship between input parameters and spectral outputs, allowing virtual reproduction of data for conditions that were not physically tested, thereby reducing the number of required reactor runs while maintaining predictive accuracy.

The patent applies preliminary action by limiting the number of actual reactor runs to a small set of representative conditions. By collecting training data from only these preliminary runs and using it to build a chemometric model, the system avoids the need to perform exhaustive computational simulations for all possible conditions, thereby reducing computational resource consumption while maintaining adequate model accuracy for prediction.

The patent replaces mechanical/experimental reactor runs with a computational chemometric model once trained. After the model is built from limited training data, it can predict spectral outputs for various conditions through computation rather than requiring physical reactor runs, substituting mechanical experimentation with computational prediction and significantly reducing ongoing resource requirements.