Abstract:
Catalytic reforming unit is one of the most important devices in petroleum processing. However, because of its high dimensionality, complex mechanism, and complex working conditions, catalytic reforming unit is time-consuming for the optimization via direct using the mechanism model. Aiming at these shortcoming, researchers have proposed the concept of surrogate model, which can effectively approximate mechanism models. The surrogate model usually includes two steps: firstly, a series of sampling points are generated in the input space and the true response values are obtained; secondly, the corresponding surrogate model is established through the sampling point and its real response value. Apparently, the sampling strategy has direct impact on the speed and accuracy of surrogate model. Hence, this paper proposes a new adaptive sampling algorithm, termed as adaptive sampling algorithm-based nearest neighbor and Mahalanobis distance (ASA-NNMD). Through global exploration and local exploitation of sampling strategy, the proposed algorithm can obtain the sample point in the key information area. It is shown via seven test functions that ASA-NNMD can select sampling points of greater impact on the accuracy of surrogate model so that it can effectively improve the accuracy of surrogate model. Finally, the corresponding surrogate models are established for these key quality indexes of catalytic reforming process via four inlet temperatures, feed load, and hydrogen / oil ratio.