A recursive optimization method for data-driven operating trajectory

which was based on daily normal operation batches

was proposed for long cycle batch processes. First

original nonlinear optimization was simplified to a high dimensional linear optimization by segmented discretization and high dimensional segmented variables were transformed into lower dimensional one by PCA (principal component analysis) algorithm. Then

snapshot optimization for original operating trajectory was performed according to cosine similarity between time-segmented variables in dimensionality-reduced principal element plane and performance index of the final products. Finally

recursive algorithm for trajectory optimization was developed by consideration of changes in square errors and similarity of time-segmented variables between batches. Application to a batch crystallization process has illustrated effectiveness of this method.