Just-in-time learning algorithm using the improved similarity index

Just-in-time learning (JIT) is a kind of effective local modeling methods. The accuracy of JIT depends on the quality of the selected samples. How to choose the modeling sample is the key issue in the JIT algorithm. Using the information entropy, an improved similarity index integrated the mutual information and the weighted samples is proposed in this paper. Firstly, the correlation degree between each input variable and output variable is estimated by using the mutual information algorithm, and the mutual information is set as weight factors to discount the corresponding variables. Then, the weighted input variables are used in the sample selection stage. At last, the traditional JIT algorithm is used for local modelling. The effectiveness of the improved similarity index algorithm is validated on a numerical example and Penicillin concentration soft sensor.