Abstract
Near-infrared (NIR) spectroscopy has been widely applied for the real-time measurements of quality variables, which plays an important role in process control, monitoring and optimization. Since the prediction accuracy of NIR model strongly depends on the structure of training samples, it is important to optimize the process of training samples selection. Therefore, in the present work, a cross validation based approach which combined with kmeans++ algorithm is developed for this optimization. Based on the results, an efficient adaptive multi- model approach can be developed. During online application, according to the similarity distance between query sample and sub-models, the optimal sub-model can be selected and the high-performance predictions can be achieved. The usefulness and superiority of the proposed method is demonstrated and compared with other modeling algorithms in a real-world gasoline blending process in China.
