Soft Sensing For Ratio Of Soda To Alumina And Leaching Rate Based On Intelligent Integrated Model And Its Application

As an important working procedure of alumina manufacturing process with Bayer method, high pressure digestion (HPD) is a very complicated metallurgical process. During HPD process, ratio of soda to alumina (RSA) and leaching rate (LR) will affect the outputs, quality and consumption of soda. In order to control HPD process optimally, the key is to measure RSA and LR online. However, until now there is no any instrument can be used to measure RSA and LR directly. RSA and LR can only be measured through chemical analysis, so large lag exists and optimal control for HPD process is influenced severely. Because of many characters in HPD process such as complex mechanism, high nonlinearity, strong coupling, time-variability, large lag and disturbance, it is very difficult to establish precise model with any single method. On the basis of mechanism analyzing of HPD, soft sensing technique for RSA and LR is studied, and soft sensing scheme based on intelligent integrated model (IIM) is proposed in this dissertation. Then RSA and LR are measured online and mixing of the original mine slurry (MOMS) is optimized. The main work and contributions in this dissertation are as follows:(1) Based on analysis of characters of complex industry process and defects of common modeling methods, the mainframe of intelligent integrated soft sensing model (IISSM) is proposed. The universal definition of IISSM is described, and basic integration form of model structure and algorithm are summarized. At the same time, the formalized description, design principles and steps of soft sensing system based on IIM are given.(2) For the low speed and precision of RPCL clustering algorithm, a space distribution of samples based RPCL (SDS-RPCL) is proposed. In SDS-RPCL, data are selected based on the space distribution of samples while centers are modified. Then the probability of moving towards clusters' margin for centers is decreased, so the clustering speed andprecision is improved.(3) Based on elaborate analysis of mechanism of PHD and summary of expert knowledge, expert mechanism model (EMM) of RSA and LR is established. EMM can explains the effect of every factor to RSA and LR intuitively.(4) In order to modify the prediction error of EMM, for the character of large number of input variables and widely distribution of samples in RSA and LR soft sensing, distributed multiple neural networks (D-MNN) is proposed. In D-MNN, PCA is used to regrouped the input variables, then the regrouped principle components (PCs) are divided into several groups according to the original information they contained, and paratactic MNN is used to approach RSA and LR step by step. MNN not only can simplify the model, but also can depict actual object more rationally because that the input variables are grouped properly. On the other hand, SDS-RPCL clustering algorithm is used to cluster the training samples, and different MNNs are used to depict samples in different clusters respectively. Then a fuzzy classifier is used to calculate the membership degree of each MNN, and established the D-MNN. The simulation results with actual data show that: when the input is close to a certain cluster of the training samples space, the prediction precision is very high; on the other hand, when the input is far from any cluster, the prediction precision is very low.(5) Take into account the above flaw of D-MNN, gray model of RAS and LR is established. An intelligent coordinator is designed to coordinate the prediction of each sub-model, and the prediction precision is improved. Then the prediction of IIM is modified to eliminate large errors, and the robustness of IIM is improved. On line correction method of IIM is proposed to make sure that the prediction precision would not decreased along with the changes of working condition.(6) Optimal control system for MOMS based on soft sensing is exploited. In this system, based on the PlantScape distributive control system, industrial control computer, PLC, high speed Ethernet are used torealize high-lever online prediction of RSA and LR, expert optimization guidance for MOMS and low-lever automatic control for mixing. Application of this system improves the LR and qualification rate of MOMS, and stabilizes the alumina manufacturing process. Moreover, the system makes great contribution to informationization of the enterprise and achieves distinct economic benefit and social benefit. At the same time, the contribution of this dissertation is helpful to other complex industry process.