Research On Mechanism And Hybrid Modeling Method Of Underflow Concentration In Thick Washing Process

At present,it is a crucial period for China to achieve a comprehensive well-off society,with the continuous rapid growth of the social economy and the continuous improvement of the modern industrial level,the demand for mine resources is increasing,especially non-ferrous metal resources.And there is a serious shortage of resources,China's sustainable development strategy is facing serious challenges.Hydrometallurgical technology plays an important role in the treatment of mineral resources.It can not only deal with complex ore and low-grade minerals,but also emits less pollution to the environment.The thick washing process is an important part of hydrometallurgical technology,and the underflow concentration of the thickener not only affects the quality of the underflow products and the subsequent process products,but also is an important guarantee for the safe and stable operation of the enterprise.In actual production,the underflow concentration of the thickener is controlled and tested by the worker's operating experience,and automatic control has not been solved.In the new era,under the guidance of China's sustainable development strategy,how to achieve high efficiency and low cost development and utilization of metal mineral resources through optimization of control technology is a difficult problem in the development of hydrometallurgical technology in China.In this paper,the problem of difficult to accurately measure the underflow concentration in the hydrometallurgical thick washing process is analyzed.The in-depth analysis of the sedimentation principle and process characteristics of the slurry inside the thickener is combined with the successful application of the artificial intelligence algorithm in the actual production process.The mechanism model of the characteristic and the data-driven data model,and finally the hybrid model method combining the mechanism model and the data compensation model,the soft-measurement method of the underflow concentration in the dense washing process is studied.This paper mainly carries out research work from the following three aspects:(1)Based on the slurry settlement principle of thick washing process and on the basis of the analysis of traditional mechanism model,a model of pulp concentration distribution in thick washing process is established,and the parameters of the mechanism model are identified by Recursive Least Square(RLS).The transformation of data variables is realized by Bernoulli principle.Through the simulation of the established model,the characteristics of the thick washing process are deeply understood,the main factors affecting the thick washing process are found out,and the auxiliary variables in the data model and the soft sensing model are determined.(2)In view of the random initialization of the initial parameter value of the basic algorithm of Extreme Learning Machine(ELM),the network structure of the ELM is determined by the user according to his own practical experience.This paper will improve the structure of the traditional ELM.Based on the above,twice hidden layers are added to the network structure to form a network structure with one input layer,three hidden layers and one output layer.Based on the perfect Three hidden layers Extreme Learning Machine(TELM),the data model based on the TELM is established by preprocessing the field data,and the prediction effect is analyzed and studied.(3)In view of the difficult problem that dynamic mechanism model is difficult to be directly applied in industrial field,a hybrid model soft-measurement method based on mechanism modeling and Entire Distribution Optimization(EDO)improved error compensation model for the TELM was developed to accurately measure the underflow concentration.The hybrid model implements the improved EDO-TELM algorithm compensator,compensates the uncertainties and theoretical assumptions of the mechanism model in the modeling process,and gives a reasonable estimate.Comparison shows that the prediction accuracy of the hybrid model has been improved obviously,which can meet the needs of the industrial field measurement.