Research On Soft Sensor Tenchnique For Rare Earth Extraction

China is the first place in the rare earth production, export and consumption in the whole world, and has all kinds of the light, medium and heavy rare earth. Although rare earth extraction technology has reached advanced world level, the control of the production process is placed in a relatively low level.Its basic characteristics are off-line analysis, experience adjustments and manual control. Consequently, it results in low efficient of producting, waste of resources and unstable of product quality. It seriously restricts the development of rare earth industry. Industry of rare earth uses cascade extraction technology. It is very complex in the mechanism with characteristics of multiple variables, strong coupling, nonlinearity, time variability and time delay. To improve and implement the automation of industrial processes, there should be achieving on-line continuous testing of component content of rare earth elements in extraction tank. However, it is difficult to employ industrial instruments to measure component content of rare earth elements in reality. With deep analysis of the extracting process, this paper adopts the soft sensor technology to solve the technical problems.In this paper, least squares support vector machines (LSSVM) and intelligent optimization algorithms are used to solve the soft sensor problems of rare earth element component content. The main contents are as follows:Firstly, the status and development and methods of soft sensor technology is summarized. This paper designs the steps of soft sensor technology on basis of engineering design of soft sensor technology, analysis and research of modeling method.Secondly, based on analysis the basis theory, model structure, and algorithms of Radial Basis Function (RBF) neural network and LSSVM, the simulation results show that the generalization ability of soft sensor model based on RBF neural network is poor and it requires a lot of sample data. However, the LSSVM model can better address these issues, its performance is more superior.Thirdly, the performance of LSSVM modeling is directly affected by the model parameters. Therefore, intelligent optimization algorithms are used to solve this problem. Intelligent optimization algorithms include genetic algorithm, particle swarm algorithm, quantum particle swarm optimization and ant colony algorithm. the methods and steps of LSSVM based on intelligent optimization algorithms are designed by analysis of the intelligent optimization algorithms mechanism, the regularization parameter and kernel function parameters optimized. The simulation results of these algorithms show that the algorithms designed are feasible. Finally, soft sensor mode structure based on LSSVM of extraction section rare earth element component content are designed by analysis of countercurrent extraction process mechanism, and establishs intelligent optimization algorithms of soft sensor model based on LSSVM. From these studies, the model accuracy through optimization is improved. According to analysis of the complexity of the algorithm, training objectives, prediction ability and running time, ant colony optimization algorithm is more suitable to optimize the parameters of LSSVM.