Modeling And Prediction Of Key Variables For Thickening Process In Hydrometallurgy

The problem of resources has already become a fundamental problem of our country and even the world sustainable development. The remarkable advantages of hydrometallurgy are high comprehensive recovery rates of valuable metals in complex and low-grade metal resources, and benefit environmental protection, therefore many new technologies of hydrometallurgy have appeared constantly and they have been widely used. Thickening process is an important unit operation that is widely used in hydrometallurgy. Thickener is the key equipment for the thickening process and the underflow concentration, overflow concentration, sludge concentration distribution and sludge blanket height of thickener are the key quality indices. However, at present, most thickening processes are still in manual operation state, making its concentration and flow fluctuate largely, and influencing subsequent production index. Due to the harsh working conditions and other factors, the large time-delay and the slow time-varying, the automatic control of thickening process has been a difficult problem. To realize the optimization control of the thickening process, the prediction model of key variables that related to optimization control is of great importance.This thesis aims at the difficulty of on-line measuring key variables in hydrometallurgy thickening process. Research on modeling for thickening process and prediction of key variables of thickening process in hydrometallurgy are carried out comprehensively and systematically, based on deeply analyzing the characteristics of thickening process, and using hybrid modeling method, which is composed of mechanism and data modeling methods. The main researches are summarized as follows:(1) Through analyzing the basic principle of sludge sedimentation and based on solid flux theory and mass conservation theory, the dynamic first principle model of thickening process is established. The prediction of underflow concentration, overflow concentration, sludge concentration distribution are achieved through the established model Through model simulation, the dynamic characteristics and steady characteristics of the thickening process are revealed to find the main factors of the thickening, to select the secondary variables of data modeling, and to establish the foundation of hybrid modeling and model correction.(2) To aim at the difficulty of applying the dynamic first principle model to the industrial field directly, prediction model of thickening process is established adopting parallel hybrid model. The model is composed of the simplified first principle model and compensated model in parallel, therefore the advantages of different modeling methods can be exerted. The purpose of simplification is to reduce the unmeasured variables in the first principle model, and improve the computational efficiency of the model. The compensated model, which employed the process data, is used to remedy the problem of the decline in prediction accuracy brought about by model simplifying.(3) To aim at the difficulty of slowly time-varying characteristics on-line tracking, a model correction strategy is proposed based on model evaluation. Thus a model correction strategy composed of long-term and short-term correction is presented in this thesis. The purpose of the description of model prediction error distribution characteristics via the Gaussian Mixture Model is to evaluate the model performance and to improve model prediction via the model output offset correction on-line. When the samples reaches a certain amount, meanwhile the prediction error is big enough or varies largely, the PLS model is corrected by the method of discount recursive PLS. Model output offset compensation and data model correction are short-term correction. When the model operates a period of time and enough new sample data are accumulated, the first principle model parameters correction can be achieved, it’s a long-term correction. By alternately using the two correction methods, the prediction accuracy and practicality are further enhanced. The effectiveness of the correction strategy is verified through the simulation results.