An Improved Locally Weighted Regression for a Converter Re-Vanadium Prediction Modeling

Locally weighted regression (LWR) is a local memory learning strategy which performs regression around an interest point, which is very efficient for learning the modeling of nonlinear system. This paper researches the possibility of using locally weighted regression for prediction modeling of a nonlinear system for converter re-vanadium in metallurgical process and proposes some improved methods by finding the optimized regression coefficients by gradient descent and kernel function bandwidth by weighted distance. To overcome the computational difficulties of kernel functions, the complexity of LWR has been reduced by K-Medoids clustering. The experimental results show that improved locally weighted regression outperforms the BP method when significant amounts of noise are added and the computing time has been shortened. This proves the implementation of the proposed nonlinear prediction model to be effective and practicable for its industrial application.