Hybrid Modeling Study Of Copper Extraction Process Based On Denoising Neural Network And SCN

Hydrometallurgy is an important part of green production in the metallurgical industry,in which the chemical solvents are used to extract and isolate the metals from the raw ores.In the face of the severe worldwide competition for mineral resources and the demands of domestic sustainable development strategies,hydrometallurgical processes can not only achieve efficient metal recovery but also organic solvent recycling.Currently,the development in the hydrometallurgical industry is primarily concentrated on the chemical solvent innovation and little research is put effort into the key production index prediction for improving the automation level of production processes.As a result,establishing a key production index prediction model applied to the extractive metallurgy industry is an important guarantee to enhance the enterprise automation level.In this thesis,aiming at the problem that the unknown key production parameters in copper extraction process in hydrometallurgical industry cannot be directly measured online,the corresponding mechanism,data-driven and hybrid model are proposed after the in-depth investigation on the extraction process mechanism,which is of great help for the further optimization of the industrial production operation.The main research contents are as follows:(1)Based on the detailed analysis of the copper extraction process mechanism,the multistage countercurrent extraction process is taken as the main research object and its dynamic mathematical model are established according to the material conservation principle,the effectiveness of which is verified by simulation experiments.The above established model can be used to simulate the actual copper extraction process and determine the influence of several key factors in the extraction process on the copper ions concentration in the raffinate.(2)During the process of modeling the copper extraction mechanism,the kinetic reaction parameter(mass transfer rate)is unknown.In the actual industrial production process,the kinetic reaction parameters cannot be directly measured and must be estimated using the measurable but noisy data.In order to reduce the influence of measurement noise on the estimation results,an effective method based on the denoising neural network is proposed to estimate the unknown mass transfer rate in this thesis,which are compared with the classical existing methods such as finite difference,polynomial fitting and Tikhonov regularization.The simulation experiments show that the proposed estimation strategy based on the denoising neural network can both achieve satisfactory results under noise or no noise,which is of great help in suppressing the influence of the measuring noise on the estimation results.(3)Based on the above estimation results of mass transfer rate,a serial hybrid model for estimating the copper ions concentration in the raffinate of copper extraction process is proposed in this thesis,which is composed of the mechanism model with unknown parameters and a datadriven parameter estimation prediction model.The random vector function link network(RVFLNN)and random configuration network(SCN)are used to predict the unknown mass transfer rate as the data prediction model of unknown parameters,respectively.The data-driven prediction model is integrated into the mechanism model to form a serial hybrid model in order to improve the model prediction accuracy.It can be easily seen from the simulation experiments that compared with the traditional mechanism model and the pure data-driven model,the proposed serial hybrid model improves the prediction accuracy of the copper ions concentration in the raffinate effectively,which lays an important model foundation for the optimization and control of the whole hydrometallurgical process.