A cobalt ion concentration detection model with temperature interference Resistance via a novel contrastive neural network

Accurately detecting the concentration of cobalt ions in zinc solution is of great significance for improving the energy efficiency and stability of zinc hydrometallurgy. However, traditional cobalt ion detection methods have low efficiency and are susceptible to temperature interference. For this purpose, we collected spectral data of cobalt zinc solutions with different concentrations under multiple temperature conditions using a self-developed automatic detection system. Based on this dataset, we proposed a novel comparative neural network with self-attention layer, and optimized its parameters through comparative experiments. The test results indicate that the network can effectively eliminate temperature interference in spectral data. In addition, the problems of distortion and redundant dimension in the output features of comparative neural networks were discovered, and feature correction and screening were completed using isometric mapping and competitive adaptive reweighting sampling algorithms. Finally, the partial least squares regression was used to regress the obtained features, and the final regression accuracy reached 0.064 mg/L. The test results show that the accuracy of this model is 4.52 times that of the neural network model that directly maps spectral data to concentration.