CAPR: An improved spectral variable selection algorithm for cobalt ion detection in zinc hydrometallurgy

Abstract

In order to accurately detect the concentration of cobalt ions in zinc solution, we used a self-developed automatic detection system to complete the full spectrum collection of cobalt zinc solution with different cobalt ion concentrations, and established various full spectrum models. The test results show that the combination model of competitive adaptive reweighted sampling (CARS) and partial least squares regression (PLSR) has good comprehensive performance. However, in the process of theoretical analysis and experimentation, we found potential weaknesses of CARS algorithm and proposed an improved variable selection algorithm called competition area proportional reduction (CAPR). The test results show that the performance of the CAPR-PLSR model on the cobalt ion spectral dataset is superior to other ten regression models, including the CARS-PLSR model. In addition, we conducted sixteen additional comparative tests on four spectral datasets from different detection domains to demonstrate the effectiveness of CAPR.